The GSE38129 gene expression dataset was submitted by Hu et al (10) and can be obtained from the publicly accessible Gene Expression Omnibus (GEO) database. The dataset was downloaded and analyzed from the GEO at the National Center for Biotechnology Information website (https://www.ncbi.nlm.nih.gov/geo/). The study was based on the GPL571 platform (Affymetrix Human Genome U133A 2.0 Array, Affymetrix; Thermo Fisher Scientific, Inc.). The samples used for gene profile analysis were obtained from 30 patients with ESCC and paired adjacent normal tissues, the patients were from high-risk areas of China, and the most recent update was in April 2017.

GEO2R online software was used for GSE38129 analysis to detect the DEGs between the tumor and normal tissues. GEO2R is an interactive networking tool that helps users to compare various groups of samples in the GEO series and identify DEGs under specific experimental conditions. The adjusted P<0.01 and |log fold change (FC)|>1 values were used as the cut-off criteria for DEG identification. Subsequently, 928 DEGs were identified following GSE38129 analysis. Among these DEGs, 498 and 430 were upregulated and downregulated, respectively.

GO analysis (https://david.ncifcrf.gov/) (11), which is a bioinformatics tool that can be used to annotate genes and gene products and determine the biological characteristics of high-throughput genome or transcriptome data, includes three categories, namely, biological process (BP), cellular component (CC) and molecular function (MF). The KEGG knowledge database (12) is a group of databases used for all types of biological data and can be used to determine functional and metabolic pathways. P<0.05 was set as the cut-off criterion and considered to indicate a statistically significant difference (Fig. S1). The Database for Annotation, Visualization and Integrated Discovery (DAVID) (13) is a web-based online bioinformatics resource and a functional interpretation tool with a large scale gene or protein dataset that can provide comprehensive functional annotation for genes.

The Search Tool for the Retrieval of Interacting Genes (STRING; http://string-db.org/) (14) database is an online tool that contains comprehensive information of various proteins and detects potential associations among the DEGs. The results were input into Cytoscape to visualize the PPI networks of the DEGs. A high combined score indicated reliable PPIs. In the present study, interactions with a combined confidence of >0.7 were considered significant. The PPI network was constructed using Cytoscape software. The Molecular Complex Detection plug-in of Cytoscape (15) further indicated the essential modules in the PPI networks (degree node score cut-off=0.2, K-Core=2, degree cut-off=2).

Gene Expression Profiling Interactive Analysis (GEPIA) (16) is a web-based server for cancer and normal gene expression analyses and interactive analysis on the basis of The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) data. Multiple types of analyses can be performed, including differential expression analysis, profiling plotting, correlation analysis and patient survival analysis. Through GEPIA analysis, ephrin-A1 (EFNA1) and collagen type IV α1 (COL4A1) were expressed at high levels in ESCC and were associated with a poor prognosis. The low expression of C-X-C chemokine receptor 2 (CXCR2) was not statistically significant.

A total of 36 ESCC tissue samples and 35 normal esophageal tissue samples were collected for the present study, which had been surgically removed from Kazakh patients at the First Affiliated Hospital of Shihezi University (Xinjiang, China) between June 2018 to March 2019. The research protocol was approved by the Medical Ethics and Human Clinical Trial Committee of Shihezi University School of Medicine (Xinjiang, China) and all recruited subjects were enrolled following the provision of written informed consent. All surgical samples were used as residual specimens following diagnostic sampling.

A total of 36 esophageal cancer tissue samples and 35 normal samples from Kazakh patients were selected from formalin-fixed and paraffin-embedded tissue chips. The sample tissue chips, with a diameter of 0.6 mm, were obtained using ALPHELYS. The tissue microarrays were heated in an oven at 65°C for 30 min, rehydrated with graded alcohols, immersed in ethylenediaminetetraacetic acid buffer (pH 9.0) at 130°C, and autoclaved in a microwave oven for 10 min. Following cooling to 30°C, the tissues were incubated at room temperature with 3% H₂O₂ solution for 10 min. The tissue sections were then incubated at 4°C with anti-securin antibody [also termed anti-pituitary tumor transforming gene 1 (PTTG1) antibody, Bioss antibodies, rabbit polyclonal, cat. no. bs-1881R, dilution 1:400] overnight. The tissue sections were organized and washed in PBS three times for 5 min each and then incubated with secondary antibody [universal kit (mouse/rabbit polymer method detection system), cat. no. PV6000, ZSGB, Ready-to-use antibody] at 37°C for 30 min. Diaminobenzidine (DAB) solution was used for 5 min at room temperature and hematoxylin was used to counterstain the sections. The IHC score was performed independently by two pathologists using a light microscope (BX51; Olympus, Tokyo, Japan; magnification, ×400) according to the color intensity as either negative (score 0), weak (score 1), moderate (score 2) or strong (score 3), and coloring area as negative (score 0), ≤10-25% (score 2), 25–50% (score 3) or >50% (score 4). The final score was determined as the coloring intensity multiplied by the coloring area. Scores 0–4 and 5–12 indicated low and high expression groups, respectively.

Data were assessed using the SPSS (version 17.0; SPSS, Inc.) statistical software package, and GraphPad Prism 5.0 (GraphPad Software, Inc.) was used to describe data. Comparisons of the expression levels of proteins between the ESCC (n=36) and normal tissues (n=35) were performed using the independent-samples t-test and χ² test. All data are presented as the mean ± SD.

The DEGs were detected using the GEO2R online analytical tool with adjusted P<0.01 and |logFC|>1 as cut-off criteria. A total of 928 DEGs were obtained between the ESCC and normal samples, including 498 upregulated and 430 downregulated genes. Eight core genes were selected on the basis of the degree of connectivity and adjusted P-value (Table I), including checkpoint kinase 1 (CHEK1), BUB1B, PTTG1, COL4A1, CXCR2, adrenoreceptor β2 (ADRB2), acyl-CoA oxidase 2 (ACOX2) and EFNA1. The top 50 DEGs are shown in Table II (19 upregulated and 31 downregulated genes). Additionally, by setting |logFC|>2.5 and adjusted P<0.01, 52 DEGs were selected, of which 30 and 22 were upregulated and downregulated, respectively. The heat maps and volcano plots show the different DEG samples (Figs. 1 and 2). These volcano plots and heat maps indicate all genes, and the top 52 DEGs, respectively.

To appreciate the functions of the DEGs further, DAVID (https://david.ncifcrf.gov/) was used to apply the GO function and KEGG pathway for enrichment analysis. The BPs, CCs and MFs of the DEGs were annotated and classified by GO analysis. The present study identified 39 GO terms on the basis of the DEGs of modules with a false discovery rate (FDR) <0.05 and count of >2 as thresholds and these terms were then sorted by the P-value. The top five enriched GO terms for the BPs, CCs and MFs were selected from the GO terms (Fig. 3 and Table III). The signaling pathways were obtained through the KEGG database, and the major signaling pathways included ‘cell cycle’, ‘ECM-receptor interaction’, ‘p53 signaling pathway’, ‘protein digestion and uptake’, ‘small cell lung cancer’ and ‘proteoglycans in cancer’ (Table IV).

The STRING database was used to predict the interaction between 928 DEGs (minimum required interaction score of >0.7). To select important modules in the PPI network, the MCODE plug-in was used and 25 modules were found. The top five modules were also selected for further analysis, which included 60, 15, 13, 32 and 7 genes (Table V). The DEGs in the top five modules were also enriched in important pathways (Fig. 4A-J). Module A had 60 nodes and 1,643 interactions, and all of the DEGs were upregulated in this module. The genes in this module, including CHEK1, cyclin A2 (CCNA2) and TTK, were considerably enriched in the cell cycle and p53 signaling pathway-related functions (Fig. 4B).

Gene expression and survival analyses were performed by GEPIA in the TCGA database. The resulting box plots (Fig. 5A-H) showed that EFNA1 and COL4A1 were expressed at a high level in ESCC (Fig. 5B and C), whereas the expression of CXCR2 was low in ESCC (Fig. 5E). Survival analysis (Fig. 6A-H) further showed that EFNA1 and COL4A1 were associated with poor prognosis and exhibited statistically significant differences (Fig. 6B and C).

According to the degree of connectivity and adjusted P-value, eight core genes were selected (Table I), among which the most strongly correlated genes were CHEK1 (degree of connectivity=88, adjusted P=1.64E-08), BUB1B (degree of connectivity=84, adjusted P=2.10E-08), PTTG1 (degree of connectivity=64, adjusted P=1.62E-04), COL4A1 (degree of connectivity=16, adjusted P=2.26E-04), and CXCR2 (degree of connectivity=15, adjusted P=1.15E-08). No significant prognostic significance was found for CHEK1 or BUB1B (Fig. 6G and H). Relevant references were also reviewed and it was found that PTTG1 is an oncogene that is overexpressed in several tumors. The high expression of PTTG1 also exhibited a relatively poor prognosis through GEPIA survival analysis (Fig. 6A). Therefore, PTTG1 was selected for IHC analysis. IHC was used to detect the expression of PTTG1 in 36 ESCC tissue samples and 35 normal tissue samples of the Kazakh patients. The results showed that the expression of PTTG1 (Fig. 7A and B) in esophageal cancer tissues was significantly higher than that in normal tissues, and the difference was statistically significant (P=0.002). In addition, the PTTG1 IHC staining scores in the ESCC and normal tissues were compared using independent-samples t-test analysis, and the difference was statistically significant (P<0.001; Fig. 7C).