Gastric cancer has the fifth highest incidence rate among all cancers, and it has a poor prognosis, with a low 5-year survival rate 36. Patients with early-stage gastric cancer are asymptomatic and thus difficult to diagnose, and screening for significant genes that can effectively distinguish normal and tumor samples can provide very important auxiliary evidence in gastric cancer diagnosis37.
In this study, the WGCNA approach was proposed to screen for the hub genes, which can effectively differentiate the normal gastric samples from tumor samples. WGCNA is an important approach to systematically describe the correlation patterns among genes and identifies modules of highly-correlated genes, followed by candidate biomarker screening. Nodes and edges in this co-expression network represent genes and the correlations between gene pairs, respectively. By using this systematic method, the connectivity in each module can be detected while also taking clinical traits into account 22.
Eventually, blue and black modules were identified as clinically significant modules. The GO and KEGG analyses revealed that the genes in these two modules were significantly enriched in the biological processes of the cell cycle, cell division, and stomach-related functions. All these biological functions are closely related to gastric cancer 38, 39. ACADL, ADIPOQ, ARHGAP39, ATAD3A, C1orf95, CCKBR, GRIK3, SCNN1G, SIGLEC11, and TXLNB were identified as the hub genes from the screening the black and blue modules.
ACADL is a mitochondrial enzyme catalyzing the initial step of fatty acid oxidation and has been reported to correlate with esophageal 40, breast 41, hepatocellular 42, and gastric 43 cancers. ACADL is upregulated in esophageal squamous cell carcinoma cell lines and specimens, and its upregulation is associated with disease progression 40. However, another study reported the opposite result, in which ACADL was downregulated and significantly correlated with poor prognosis in hepatocellular carcinoma 42. ACADL is also among the genes that are significantly differentially methylated between the ER + and ER- breast cancer tumors 44. By comprehensive whole-genome and transcriptome sequencing analyses, a study has revealed that ACADL is one of the mutated genes in gastric cancer 43.
ADIPOQ is one of the most important adipocytokines secreted by adipocytes 45, and the polymorphisms of ADIPOQ have been reported to correlate with several types of cancer, including colorectal 46, 47 and breast 48 cancers. A study focusing on the molecular mechanisms ADIPOQ participated in has revealed that ADIPOQ induces cytotoxic autophagy in breast cancer cells through STK11/LKB1-mediated activation of the AMPK-ULK1 axis 49. Another study has reported that miR-370 inhibits the proliferation, invasion, and epithelial-mesenchymal transition of gastric cells by directly downregulating receptor 4 of ADIPOQ 50. Overexpression of another microRNA, miR-15b-5p, promotes the metastasis of gastric cancer by regulating ADIPOQ receptor 3 51.
ATAD3A is a nuclear-encoded mitochondrial enzyme, involving in mitochondrial dynamics, cell death, and cholesterol metabolism 52. It has been reported to correlate with hepatocellular carcinoma 53 and breast cancer 54, and it might be an effective therapeutic target in cancer treatment 55. ATAD3A is differentially expressed between paclitaxel-resistant and -sensitive MCF7 breast cancer cells 54. A study has revealed that ATAD3A is upregulated in hepatocellular carcinoma and ATAD3A upregulation is correlated with poor prognosis 53.
Cholecystokinin is a well-known trophic factor for the gastrointestinal tract 56, and CCKBR is correlated with pancreatic 56–59, breast 60, and gastric 61 cancers. Gastrin upregulates CCKBR in gastric cancer cell lines 62, and thus it serves as a biomarker in gastric cancer treatment 63. A study has shown that miR-148a has anti-cancer effects on gastric cancer through the inhibition of STAT3 and Akt activation by targeting CCKBR 64. Another research has revealed that trastuzumab inhibits the growth of HER2-negative gastric cancer cells by regulating the CCKBR signaling pathway 61.
GRIK3 mainly participates in the neuroactive ligand-receptor interaction pathway, and GRIK3 upregulation is associated with poor survival in gastric cancer 65. GRIK3 promotes epithelial-mesenchymal transition by regulating the SPDEF/CDH1 signaling in breast cancer cells 66.
There have been few studies focusing on the relationship between gastric cancer and SCNN1G, ARHGAP39, Clorf95, SIGLEC11, or TXLNB. SCNN1G is one of the genes significantly upregulated in Ewing’s sarcoma and fibromatosis samples 67. ARHGAP39 mutations or variations in copy number or expression level are found in several types of tumor-like tissues from the central nervous system, skin, prostate, and gastrointestinal tract 68. ARHGAP39 interacts with p53 and BAX, and decreased expression of ARHGAP39 increases cell proliferation, leading to tumorigenesis 69. Clorf95 is one of the uncharacterized proteins correlated with diverse human cancers 70. Another study focusing on scleroderma patients demonstrated the involvement of Clorf95 in cancer incidence 71. Sialic acid-binding immunoglobulin-like lectin-11 (SIGLEC11) is a primate-lineage–specific receptor of human tissue macrophages, and it is also expressed in brain microglia 72, 73. A missense mutation of SIGLEC11 has been detected in pancreatic cancer patients 74, and SIGLEC11 is significantly upregulated in the poor prognostic group of pancreatic cancer patients 75. Taxilin alpha (TXLNA), which is a binding partner of the syntaxin family, has been identified as a key factor in the coordination of intracellular vesicle trafficking, and it is upregulated in pancreatic adenocarcinoma patients 76.
In our study, several hub genes were implicated in the metabolic processes such as fatty acid oxidation and cholesterol metabolism. A previous study has demonstrated the association between metabolic syndrome and gastric cancer 77. Low total serum cholesterol levels are correlated with an increased risk of gastric cancer in Chinese Han population 78. A study has detected increased fatty acid oxidation in gastric cancer 79, and adipocytes fuel gastric cancer by mediating fatty acid metabolism 80.
All these results from the previous studies demonstrate that the hub genes identified in our study are closely correlated with gastric cancer and play important roles in cancer development, progression, or proliferation.
The significant module and hub genes identified in this study are biologically rational. First, the clinically significant module identified in our study bears strong preservation, implying that this clinically significant module is conservative and could also be reproduced in other datasets. Further, it suggests that that modules constructed by WGCNA are reliable. Second, most of the genes in the significant module were enriched for specific GO terms and KEGG pathways closely relating to stomach or cancer physiology. For instance, GO analysis demonstrated that most of the genes in the clinically significant modules were closely related to digestion, carbohydrate metabolic process, and gastric acid secretion, as well as cell division and cell cycle. KEGG enrichment analysis also indicated that most of the genes in the clinically significant module were implicated in gastric acid secretion, protein digestion and absorption, as well as glycerolipid metabolism and the p53 signaling pathway. Third, all the hub genes identified in our study had previously been reported to relate to cancer. Most of these hub genes are implicated in metabolic processes, influencing the development and progression of gastric cancer. It may thus be inferred that these genes are genuinely the hub genes in charge of the key processes in gastric cancer, and they deserve a deeper analysis and validation. Finally, by using machine learning methods, the hub genes were demonstrated to effectively discriminate the gastric tumor samples from normal samples. In our study, the predictive effects of different machine learning methods were evaluated by AUC values 81. Herein, the AUC values of both logistic regression was > 0.8, indicating the excellent predictive results. All the results indicated that the expression profiles of these ten hub genes have excellent predictive effects when discriminating gastric cancer samples from normal samples.
However, our study has limitations. First, all the hub genes were identified and validated only through bioinformatics, and further exploration of the biological functions and molecular mechanisms of these hub genes both in vitro and in vivo is required. Second, due to the limited availability of the data, we did not differentiate between intestinal-type and diffuse-type gastric cancers. More data are needed to analyze and identify the hub genes between these two types of gastric cancer and normal samples.
In summary, through WGCNA, we identified ten hub genes, which might serve as potential diagnostic and/or therapeutic biomarkers for gastric cancer. Profile data mining by bioinformatics analysis is an available method to find potential diagnostic or therapeutic biomarkers systematically. Nevertheless, further investigations about the molecular mechanisms in which these hub genes are involved are still needed to verify the involvement of these genes in gastric cancer. Our findings provide a better understanding of the molecular mechanisms and putative diagnostic or therapeutic biomarkers for gastric cancer.