Novel prognostic biomarkers of gastric cancer based on gene expression microarray: COL12A1, GSTA3, FGA and FGG.

Gastric cancer (GC) is the fifth most common malignancy and the third leading cause of cancer‑associated mortality in the world. However, its mechanisms of occurrence and development have not been clearly elucidated. Furthermore, there is no effective tumor marker for GC. Using DNA microarray analysis, the present study revealed genetic alterations, screened out core genes as novel markers and discovered pathways for potential therapeutic targets. Differentially expressed genes (DEGs) between GC and adjacent normal tissues were identified, followed by pathway enrichment analysis of DEGs. Next, the protein‑protein interaction (PPI) network of DEGs was built and visualized. Analyses of modules in the PPI network were then performed to identify the functional core genes. Finally, survival analysis of core genes was conducted. A total of 256 genes were identified as DEGs between the GC samples and normal samples, including 169 downregulated and 87 upregulated genes. Through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis, the present study identified a total of 143 GO terms and 21 pathways. Six clusters of functional modules were identified, and the genes associated with these modules were screened out as the functional core genes. Certain core genes, including collagen type 12 α1 chain (COL12A1), glutathione S‑transferase α3 (GSTA3), fibrinogen α chain (FGA) and fibrinogen γ chain (FGG), were the first reported to be associated with GC. Survival analysis suggested that these four genes, COL12A1 (P=0.002), GSTA3 (P=3.4x10‑6), FGA (P=0.00075) and FGG (P=1.4x10‑5), were significant poor prognostic factors and therefore, potential targets to improve diagnosis, optimize chemotherapy and predict prognostic outcomes.

Introduction

Gastric cancer (GC) is the fifth most common malignancy in the world, especially in developing countries (1). GC has been the leading cause of cancer death worldwide until the mid-1990s, after which the occurrence has been substantially declining. However, certain countries in Eastern Asia, such as China, Korea and Japan, remain as highly endemic areas (2). In China, the 5-year survival rate of GC has improved 16% for patients diagnosed during 1995–2009 (3). However, GC is still the third leading cause of cancer death in the world and its prognosis is relatively poor (1).

In recent years, a growing number of researchers devoted to GC research and achieved considerable achievements. In the previous studies, various molecular mechanisms and biomarkers related to GC have been identified. Through a retrospective study, Pectasides et al (4) found that using carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA 19-9) and carbohydrate antigen 50 (CA-50) as GC biomarkers had significant value in early detection of recurrence and monitoring progression. Ara et al (5), discovered that non-canonical Wnt signaling pathway contributed to GC progression, which may serve as a new potential therapeutic target for GC. Despite of these advances, the mechanisms of GC occurrence and development are yet to be clearly elaborated. Furthermore, the biomarkers commonly used for GC diagnosis and treatment, including carbohydrate antigen 125 (CA125), CA19-9 and CEA, have significant sensitivity and specificity issues (6). As such, further research on revealing genetic alterations, identification of new biomarkers and exploration of pathways associated with GC is critically needed.

DNA microarray analysis is a rather powerful method to screen out cancer-related genes as novel diagnostic and prognostic markers and disclose genetic alterations of cancer evolution and progression. For example, Yang et al (7) found that Chromogranin A (CHGA) and Thy-1 cell surface antigen (THY1) were novel biomarkers in the diagnosis of cancer via DNA microarray analysis. Similarly, Sun et al (8) identified the COL family as promising prognostic markers for GC. Often used in combination with DNA microarray analysis, gene ontology (GO) enrichment analysis is a strategy to characterize the function categories affected by cancer and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis is an approach used to find out biologic pathways enriched in cancer. For example, with the above analysis methods, Hu et al (9) confirmed that several pathways, including focal adhesion, ECM-receptor interactions and the metabolism of xenobiotics by cytochrome P450, were associated with the progression of GC. Another important technique, modules analysis, is more crucial to study the specific behavior of modules and identify functional genes as cancer biomarkers compared with genes with straightforward high interaction degree (10).

In the present study, we selected expression profile (GSE79973) to perform DNA microarray analysis to reveal genetic alterations, identify new biomarkers and explore novel pathways associated with GC. In addition to biomarkers previously reported, we identified four novel biomarkers, COL12A1, GSTA3, FGA, and FGG, as significant poor prognostic factors. Most of the enriched GO terms and KEGG pathways of those genes are related to GC, such as chemical carcinogenesis, metabolism of xenobiotics by cytochrome P450, ECM-receptor interaction, focal adhesion, and platelet activation. These findings may provide insights on GC occurrence and development, as well as potential therapeutic targets for future research.

Materials and methods

Microarray data

To account for tumor heterogeneity, we conducted DNA microarray analysis with the new gene profiles, GSE79973, to identify genes as novel biomarkers. In the present study, the expression profiles associated with GC, GSE79973, were downloaded from GEO (www.ncbi.nlm.nih.gov/geo/), which is a public functional genomic data repository. GSE79973 containing 20 tissues (10 pairs of GC tissues and adjacent non-tumor tissues) were obtained from Zhejiang Provincial People's Hospital, Zhejiang, China. The platform was based on GPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0.

Preprocess microarray data

The Microarray Data was preprocessed by affy package in R (11). The purpose of this step was to filter out unwanted noise of the raw microarray and ensure background and data were standardized (12). Furthermore, for these genes that corresponded with multiple probes, we used the average expression values of those probes as the expression value of each gene. Furthermore, to visualize the difference before and after normalization, we constructed box plots of raw and normalization data.

Identification of the DEGs

Using LIMMA package in R, we collected a data list from GSE79973 (13). We then screened out these DEGs by means of t-test with P-value <0.05 and |log2(fold change) |>2. The heatmap and volcano map of these DEGs were constructed using pheatmap package in R and gplots package in R, respectively.

GO terms and KEGG pathways of DEGs

To elaborate how DEGs affected the GC cells, GO and KEGG pathway enrichment analyses of DEGs were conducted using DAVID (david.abcc.ncifcrf.gov/). The DAVID database includes a total of four modules (functional annotation, gene functional classification, gene ID conversion and gene name batch viewer). The database for annotation was used in the presnt study. The GO terms and pathways of DEGs with P<0.05 and at least five genes were screened out as significant function annotation of DEGs.

PPI network analysis of the DEGs

To study the molecular interactions between DEGs, we built the PPI network using the STRING (www.string-db.org/) database (14). The PPI network includes direct (physical) and indirect (functional) associations and stem from computational prediction, knowledge transfer between organisms, and interactions aggregated from other (primary) databases. Using the STRING database, we could obtain certain integrated scores of interactions among DEGs and select out the genes whose integrated scores were bigger than 0.4 (the default threshold in the STRING database). The PPI networks were then visualized by the Cytoscape software (15).

Sub-network modeling analysis of PPI networks

In PPI networks, genes in the same module typically show the same or similar function and work together to implement their biological function. To visualize the network and identify the modules in the network, MCODE plug-in on the Cytospace software (www.cytoscape.org/) was used. The parameters were set as follows: Degree cutoff ≥2 (degrees of each node in module were at least larger than 2), K-core ≥2 (subgraphs of each node in module were at least more than 2). After that, the GO enrichment analysis was performed using DAVID and the functional core genes of subsets were selected in each module.

Survival analysis of core genes

The survival analysis of core genes was conducted by Kaplan-Meier plotter (KM plotter, www.kmplot.com), whose gene expression data as well as relapse free and overall survival information are downloaded from the well-known public database, including GEO (Affymetrix microarrays only), EGA and TCGA. The Kaplan Meier (KM) plotter is capable of assessing the effect of 54,675 genes on survival using 10,461 cancer samples, 1,065 of which are GC patients with a mean follow-up of 69/40/49/33 months. Based on the KM plotter on the webpage, the hazard ratio (HR) with 95% confidence intervals and log rank P-values were calculated and the curves were generated.

Results

Data preprocessing

Genes with systematic bias among original data were removed after preprocessing using the Affy package in R software. The expression data of genes before and after normalization are shown in Fig. 1. The black lines in each of the boxes represent the medians of each dataset. The black lines are shown at almost the same level in the box plots, indicating a significant effect of standardization.

Figure 1.

Normalized expression value data. The black line in each box represents the median of each set of data, which determined the degree of standardization of data through its distribution. (A) Box plots of expressed value data prior to normalization. (B) Box plots of expressed value data following normalization. The gene expression value is presented on the y-axis and the samples on the x-axis.

Get the DEGs

Using the R package LIMMA, we identified the DEGs between the GC and normal samples. A total of 256 DEGs in the data list were identified from GSE79973, including 169 down-regulated and 87 up-regulated genes. We ranked these DEGs by P-value and constructed the heatmap (Fig. 2) as well as volcano map (Fig. 3). Based on the heatmap and volcano map, the gene expressions of these DEGs exhibit significant differences.

Figure 2.

Heatmap of the DEGs. Overview of the associations between DEGs from normal and GC tissues. Each row represents the tissues (10 GC and 10 normal). Red coloration indicated 10 GC tissues. Blue coloration indicated 10 adjacent normal tissues. Each column represents the gene IDs. Log ratio scale bar for the Treeview color change was also presented. The red in the log ratio scale bar indicates highly expressed genes in GC tissues when compared with the normal tissues, while the blue represents the low expression of genes. The number in each block with its color in the log ratio scale bar represents the expression level of each gene in different samples. GC, gastric cancer; DEGs, differentially expressed genes.

Figure 3.

Volcano plot overview of the distribution of DEGs. Red spots indicate upregulated genes, while green spots indicate downregulated genes. Black spots represent genes that were not DEGs. The-log10 (adj. P-value) on the x-axis is the P-value following correction and the log2FC value on the y-axis represents the fold change of genes. DEGs, differentially expressed genes.

GO term enrichment analysis of the DEGs

Using the DAVID and pathway enrichment analysis of the DEGs, we identified a total of 100 GO terms and 16 pathways. The top 10 enriched GO terms of the DEGs based on the P-values are listed (Table I). As shown in Table I, DEGs were significantly enriched in biological processes (BP), including extracellular matrix organization, skeletal system development, cell adhesion and xenobiotic metabolic process. For molecular function (MF), DEGs were significantly enriched in extracellular matrix structural constituent and extracellular matrix binding. In addition, GO cell component (CC) analysis indicated that the DEGs were significantly enriched in extracellular space, extracellular region, proteinaceous extracellular matrix and collagen trimer.

Table I.

Top 10 enriched GO terms, which were sorted by P-value in ascending order.

Category	GO ID	GO name	P-value	Gene number
CC	GO:0005615	Extracellular space	1.07×10−15	55
CC	GO:0005576	Extracellular region	4.88×10−15	59
BP	GO:0030198	Extracellular matrix organization	1.99×10−11	19
CC	GO:0005578	Proteinaceous extracellular matrix	3.87×10−9	19
CC	GO:0005581	Collagen trimer	1.12×10−8	12
BP	GO:0001501	Skeletal system development	9.95×10−9	14
BP	GO:0007155	Cell adhesion	2.95×10−8	23
MF	GO:0005201	Extracellular matrix structural constituent	9.78×10−7	9
BP	GO:0006805	Xenobiotic metabolic process	3.84×10−5	8
MF	GO:0050840	Extracellular matrix binding	2.09×10−4	5

GO, Gene Ontology; BP, biological process; CC, cellular component; MF, molecular function.

KEGG pathway analysis of the DEGs

A total of 16 enriched pathways with P<0.05 were shown in Table II. The first enriched pathway, chemical carcinogenesis, was directly related to cancer and all the others have been reported to play an important role in cancer progression via certain biological processes, such as protein digestion and absorption, metabolism of xenobiotics by cytochrome P450, ECM-receptor interaction, focal adhesion, metabolic pathways, and platelet activation.

Table II.

KEGG pathways enriched in DEGs.

Category	Pathway name	Gene number	P-value
KEGG_PATHWAY	hsa05204:Chemical carcinogenesis	11	1.65×10−7
KEGG_PATHWAY	hsa04974:Protein digestion and absorption	11	4.12×10−7
KEGG_PATHWAY	hsa00980:Metabolism of xenobiotics by cytochrome P450	10	9.02×10−7
KEGG_PATHWAY	hsa04512:ECM-receptor interaction	10	3.59×10−6
KEGG_PATHWAY	hsa00982:Drug metabolism-cytochrome P450	9	4.86×10−6
KEGG_PATHWAY	hsa00830:Retinol metabolism	7	3.10×10−4
KEGG_PATHWAY	hsa04971:Gastric acid secretion	7	5.83×10−4
KEGG_PATHWAY	hsa00010:Glycolysis/Gluconeogenesis	6	0.0027
KEGG_PATHWAY	hsa04510:Focal adhesion	10	0.0027
KEGG_PATHWAY	hsa04978:Mineral absorption	5	0.0041
KEGG_PATHWAY	hsa00350:Tyrosine metabolism	4	0.0134
KEGG_PATHWAY	hsa05146:Amoebiasis	6	0.0177
KEGG_PATHWAY	hsa00071:Fatty acid degradation	4	0.0247
KEGG_PATHWAY	hsa01100:Metabolic pathways	26	0.0379
KEGG_PATHWAY	hsa04611:Platelet activation	6	0.0383
KEGG_PATHWAY	hsa04972:Pancreatic secretion	5	0.0440

DEGs, differentially expressed genes; KEGG, Kyoto Encyclopedia of Genes and Genomes.

PPI network of DEGs

The whole PPI network shown in Fig. 4 contained 116 nodes and 197 edges. The nodes represent DEGs with higher integrated scores of interactions and the edges represent interactions among the DEGs. The genes with interaction degree ≥5 were shown in Table III. COL1A2, COL2A1, COL11A1 and SPARC, whose corresponding degree was beyond 10, were hub genes in the PPI network and closely related to cancer.

Figure 4.

PPI network of DEGs. The network contained 116 nodes and 197 edges. The 39 red nodes were the upregulated genes and the 76 green nodes were the downregulated genes. The one pink node indicates a gene that was automatically generated and was potential associated with DEGs. The edges represented the correlation between two genes. PPI, protein-protein interaction; DEGs, differentially expressed genes.

Table III.

Core genes with corresponding degree ≥5.

Gene	Degree
COL1A2	15
COL2A1	15
COL11A1	12
SPARC	11
COL5A2	9
FN1	9
IL8	9
TIMP1	9
ALDH3A1	8
ATP4A	8
COL6A3	8
GSTA1	8
GSTA3	8
THBS2	8
ADH1A	7
ADH1B	7
ADH7	7
ADIPOQ	7
ALDOB	7
COL10A1	7
COL12A1	7
COL8A1	7
TNFRSF11B	7
BGN	6
CYP2C19	6
CYP2C9	6
PLA2G1B	6
AKR1C1	5
ASPN	5
CHGA	5
COMP	5

Modules in PPI network

Six modules were identified in this study, including 8, 6, 3, 3, 3 and 3 genes, respectively (Fig. 5). The top three GO terms of each modules are shown in Table IV. Interestingly, the GO terms of these modules were mainly associated with collagen trimer, (S)-limonene 7-monooxygenase activity, extracellular matrix organization, multicellular organism development, glutathione derivative biosynthetic process and fibrinogen complex, respectively. Among the core genes involved in these modules, COL12A1, GSTA3, FGA and FGG were firstly reported to be associated with GC.

Figure 5.

Functional modules involved in the PPI network. The red nodes represent the upregulated genes and the green nodes represent the downregulated genes. The edges indicate the correlation between two genes. (A) Module A was comprised of 8 nodes and 28 edges. (B) Module B was comprised of 6 nodes and 9 edges. (C) Module C was comprised of 3 nodes and 3 edges. (D) Module D was comprised of 3 nodes and 3 edges. (E) Module E was comprised of 3 nodes and 3 edges. (F) Module F was comprised of 3 nodes and 3 edges. PPI, protein-protein interaction.

Table IV.

Top three GO terms in each module.

GOID	GO name	P-value	Genes ID
Module A
0005581	Collagen trimer	2.49×10−9	COL6A3, COL1A2, COL2A1, COL8A1, COL10A1
0030199	Collagen fibril organization	1.17×10−7	COL1A2, COL2A1, COL5A2, COL11A1
0005201	Extracellular matrix structural constituent	1.19×10−7	COL1A2, COL2A1, COL5A2, COL11A1
Module B
0018676	(S)-limonene 7-monooxygenase activity	4.74×10−4	CYP2C19, CYP2C9
0018675	Limonene 6-monooxygenase activity	4.74×10−4	CYP2C19, CYP2C9
0052741	(R)-limonene 6-monooxygenase activity	4.74×10−4	CYP2C19, CYP2C9
Module C
0030198	Extracellular matrix organization	1.36×10−4	TNFRSF11B, SPARC, SPP1
0050840	Extracellular matrix binding	0.0031	SPARC, SPP1
0005615	Extracellular space	0.0055	TNFRSF11B, SPARC, SPP1
Module D
0007275	Multicellular organism development	9.61×10−4	WNT2, SFRP4, WIF1
0017147	Wnt-protein binding	0.0037	SFRP4, WIF1
0030178	Negative regulation of Wnt signaling pathway	0.0062	SFRP4, WIF1
Module E
1901687	Glutathione derivative biosynthetic process	0.0026	GSTA1, GSTA3
0004364	Glutathione transferase activity	0.0041	GSTA1, GSTA3
0006749	Glutathione metabolic process	0.0067	GSTA1, GSTA3
Module F
0005577	Fibrinogen complex	2.17×10−7	FGG, FGA, FN1
0031093	Platelet α granule lumen	8.94×10−6	FGG, FGA, FN1
0002576	Platelet degranulation	3.73×10−5	FGG, FGA, FN1

GO, Gene Ontology.

Kaplan-Meier survival curves

The Kaplan-Meier (KM) survival curves are shown in Fig. 6. Using the KM plotter platform, the GC patients were divided into two groups, with low and high expression of the four core genes, respectively. We found that each of the four core genes, COL12A1 [HR 1.4 (1.13–1.74) P=0.002], GSTA3 [HR 1.64 (1.33–2.02) P=3.4e-06], FGA [HR 1.36 (1.14–1.63) P=0.00075] and FGG [HR 1.45 (1.23–1.72) P=1.4e-05], was a poor prognostic factor of overall survival.

Figure 6.

Kaplan-Meier survival curves of four genes generated from Kaplan-Meier plotter. Curves of (A) COL12A1, (B) GSTA3, (C) FGA and (D) FGG. COL12A1, collagen type 12 α1 chain; GSTA3, glutathione S-transferase α3; FGA, fibrinogen α chain; FGG, fibrinogen γ chain; HR, hazard ratio; CI, confidence interval.

Discussion

GC, as one of the most malignant tumor, was a typical heterogeneous cancer with molecular complexity and heterogeneity (16,17). Recently, a substantial number of effort have been paid in exploring the underwent molecular mechanism of GC and several genetic and epigenetic alterations have been identified as biomarkers with effect on diagnosis, treatment, stratification and prognosis (18–20). However, with the technological improvement, its heterogeneity was found to be more complex than previous imagined, regarding genomic instability, differentially expressed genes (DEGs), genetic variations, epigenetic heterogeneity, protein heterogeneity and so on (21,22). The heterogeneity of GC presented some complex biological characteristics, such as recurrence and metastasis of cancer, sensitivity to adjuvant therapy, and so on (23–25). Therefore, these identified biomarkers still were ineffective and inconstant, which slow their clinical application (17,26). Even so, it still should not be neglected that exploring novel biomarkers was helpful to further construct the molecular network of GC, improve our understanding on heterogeneity, and identify more meaningful genetic subtypes related to individual characteristics, such as prognosis, sensitivity of adjuvant therapy and so on (21).

In our study, a total of six modules were identified. These modules were analyzed with these DEGs that met with the default parameters. And the expression data of these DEGs were identified based on the gene profiles (GSE79973). Therefore, as a result of the tumor heterogeneity, these modules were associated with this gene profiles. Interestingly, from the function annotation of these modules, on the one hand, these modules were enriched in some common cancer signaling pathway such as PI3K-Akt signaling pathway, Wnt signaling pathway and so on (27,28). On the other hand, these modules also were enriched in the processes, such as gastric acid secretion, protein digestion and absorption and so on, which were specific in the GC.

We also found that certain well-known gastric prognosis factors, such as some non-canonical Wnt signaling genes and PICT1, were not included in the identified modules (29). In fact, these well-known gastric prognosis factors were identified as DEGs in our study. But they were not included in further module analysis for their expression level without meeting with the default parameter. This issue resulted from the tumor heterogeneity. It is the heterogeneity that restricts these well-known prognostic factors as sites of targeted therapy, for example trastuzumab for the GC patients with HER-2 positive (30). The trastuzumab has been a standard treatment strategy for the GC patients with HER-2 positive (31). However, only a small number of patients can benefit from it, which mainly results from HER-2 in a fraction of GC patients (~7–34%) being overexpression due to the heterogeneity (32). Even so, the tumor heterogeneity makes it possible to explore some novel genes related to prognosis of patients with GC as the candidates of biomarkers. These novel genes may not be as important as these well-known prognostic genes. But It is helpful to construct tumor molecular network and provide more accurate individualized treatment plan. Remarkably, we identified four GC-associated functional core genes, COL12A1 in Module A, GSTA3 in Module E and FGA, FGG in Module F, and the level of these genes correlate with poor prognosis.

COL12A1 was reported to be related to several types of cancers, such as subungual exostosis, ovarian, breast and colon cancer, which suggest that COL12A1 may be a new potential cancer biomarker (33–36). As a member of the FACIT (fibril associated collagens with interrupted triple helices) collagen family, COL12A1, together with COL6A3, COL8A1, COL1A2, COL5A2, COL10A1, COL11A1 and COL2A1, has the molecular function of extracellular matrix structural constituent. These FACIT members take part in the biological processes of collagen fibril organization and construct the cellular component of collagen trimer. Although COL12A is expressed in a variety of tumor tissues, its exact function remains poorly understood. Januchowski et al (34) found that COL12A1 played a role in the drug resistance of cancer cells and tumor progression. Based on our observed GO terms and pathways in module A (Fig. 5), COL12A1 appear to participate in pathways of protein digestion and absorption as well as focal adsorption, which may allow tumor cells to invade into the surrounding microenvironment. In addition, high expression of COL12A1 is associated with poor prognosis. Thus, COL12A1 may be a new potential marker and prognostic factor of GC and its pathways may be potential treatment targets.

GSTA3 was identified to be a down-regulated gene in our study. Together with GSTA1 and CYP2C18, GSTA3 constructed the Module E (Fig. 5). They interact with each other and participate in chemical carcinogenesis, glutathione metabolism and drug metabolism-cytochrome P450. GSTA3 and GSTA1 encode a superfamily of glutathione S-transferases (GST), which are involved in biotransformation of toxic xenobiotics and endobiotics, as well as arachidonic acid metabolism via conjugation with reduced glutathione (GSH) (37). The expression level of GST are associated with the resistance of cells to toxic chemicals, such as carcinogens, antitumor drugs, environmental pollutants, and products of oxidative stress (38). Therefore, GSTA3 may be a potential marker in selecting chemotherapy drug (39). Further research on the association of these molecules is needed to help us better understand the exact role of GSTA3 in GC regulation.

FGA and FGG encode the alpha and gamma components of fibrinogen, which are constituent parts of blood-borne glycoprotein (40). They were reported to differentially express in many types of cancer, such as prostate cancer, lung cancer, hepatocellular carcinoma, and pancreatic cancer (40–43). Plasma fibrinogen levels before surgery together with histological grade and lymph node involvement have been defined as independent prognostic factors in patients with colorectal cancer (44). Ghezzi et al (45), reported that plasma fibrinogen level might be a potential index to predict prognosis, improve the early diagnosis of endometrial cancer and optimize the treatment schedule. The possible mechanisms behind the association of plasma fibrinogen level with cancer include (1) the soluble form of fibrinogen may serve as the bridging molecule between tumor cells and host cells; (2) tumor cells and platelets may form large aggregates, which prevent tumor cells from being attacked by the innate immune system and (3) fibrinogen is essential to the sustained adherence of tumor cells to the endothelia of target tissues (46–48). In GC, fibrinogen plays a key role in hematogenous and lymphatic metastasis of cancer cells through spontaneous metastasis, facilitating the stable adhesion and/or survival of metastatic emboli after tumor cell intravasation (49). The plasma fibrinogen level on the prognosis was based on staging of tumor with worse prognosis in T2 GC as well as lymphatic and hematogenous metastasis, however, not in T3/T4 GC (50). However, they were directly reported to be associated with GC for the first time in this study. There still existed debate in expression level about these genes in tissues of cancer (51–53). In our study, FGA and FGG were identified to be down-regulated genes, while their high expression indicated poor prognosis for patients with GC. Besides tumor, many other factors may influence the alteration of expression level of FGG and FGA, such as acute-phase reactant and binding sites in tissues (40,54). Furthermore, levels of fibrinogen genes (FGA and FGG) changed following a circadian rhythms pattern and this additional level of fibrinogen transcriptional regulation has not yet been characterized (55). Considering pulmonary infarction as one of the most serious complications after GC surgery and the relationships between plasma fibrinogen level and GC, we should take it into consideration whether postoperative patients need a routine prophylactic anticoagulation. As evaluation indexes of blood coagulation state, FGG and FAG are potential biomarkers to improve diagnosis, optimize treatment and predict prognosis.

We have a lot of limitations in the universality and applicability of these novel biomarkers for the observational nature of this article. Our study was conducted based on the gene profiles, GSE79973, which contained 10 pairs of GC tissues and adjacent non-tumor tissues. The association of the identified genes and GC, their value to predict prognosis and molecular interactions among these genes were not verified in clinical samples by experiments. For above reasons, the further experiments and studies are needed to measure the identified genes in clinical practice and investigate the biological mechanisms of the interactions among these identified genes.

In conclusion, COL12A1, GSTA3, FGA and FGG have been identified to be associated with GC in this study and were selected as the core genes. Through GO and pathway enrichment of modules, we identified the functions and pathways of these core genes. Furthermore, based on survival analysis, the four genes were found to be significant poor prognostic factors. These core genes might be potential markers to improve diagnosis, optimize chemotherapy plan and predict prognosis. In addition, the pathways related to these genes may be potential therapeutic targets for GC. We plan to embark on validating the potential functions and pathways of these genes in our future studies.