Based on TCGA (The Cancer Genome Atlas) and GTEx (the Genotype-Tissue Expression) databases, we used the GEPIA (http://gepia2.cancer-pku.cn/#degenes) web server to screen DEGs associated with esophageal cancer by the LIMMA method (Tang et al., 2019). DEGs satisfying the criteria with adjusted p < 0.01 and |log2 fold change (FC)|>2 were designated as statistically significant. The volcano map of DEGs was drawn using the SangerBox tools, a free online platform for data analysis (http://www.sangerbox.com/tool).

STRING (Search Tool for the Retrieval of Interacting Genes, https://cn.string-db.org/) was adopted to construct the protein–protein interaction (PPI) based on highly DEGs (Franceschini et al., 2013). The cytoHubba plug-in Cytoscape (version 3.8.2) was used to calculate hub nodes to select the top 20 DEGs (Shannon et al., 2003). The intersection genes calculated based on degree and MCC algorithms were used as the candidate hub genes.

Through bioinformatics methods, we know that POSTN and TIMP1 were highly expressed in patients with esophageal cancer. Therefore, we further detect the expression levels of autoantibodies corresponding to these two proteins in the serum of patients with ESCC by ELISA. Purified recombinant proteins POSTN and TIMP1 were purchased from the CLOUD-CLONE CORP (Wuhan, China). Horseradish peroxidase (HRP)-conjugated mouse anti-human IgG (Wuhan Aoko Biotechnology Co. Ltd.) was used as the secondary antibody. The coated concentrations of POSTN and TIMP1 proteins were both 0.5 ng/ml. Each ELISA plate included six repeated serum samples and two blank controls. The repeated serum samples were used to normalize the difference between plates, and the blank controls were consulted for quality control. The detailed steps of ELISA were described in our previous study (Wang et al., 2018). The microplate reader was performed to measure the optical density (OD) of wells at 450 and 620 nm. The absorbance difference between 450 and 620 nm was used for the subsequent analysis.

The serum samples of 185 patients with ESCC included in this study came from a third class hospital in Henan Province. They were all patients with ESCC diagnosed by histopathology and without any treatment and did not suffer from other cancer diseases. The serum samples of 185 normal controls were received from the specimen bank of Henan Key Laboratory of Tumor Epidemiology. The normal controls excluded autoimmune diseases, esophageal cancer, and related diseases.

IBM SPSS 22.0 and GraphPad Prism 9.1.1 were used in the study. All statistical analyses were based on the two-tailed test, and p < 0.05 was considered to be statistically significant. The non-parametric test was adopted to compare the expression levels of autoantibodies between ESCC patients and normal controls. The OD value corresponding to the maximum Youden’s index when the specificity is greater than 90% was determined as the cutoff value. The Chi-square test was employed to analyze the positive rates of autoantibodies in different clinical subgroups in all ESCC patients. The receiver operating characteristic curve (ROC) was used to evaluate the diagnostic value of the autoantibodies in different groups. The positive likelihood ratio (PLR), negative likelihood ratio (NLR), positive predictive value (PPV), negative predictive value (NPV), accuracy, and Youden’s index were calculated to estimate the diagnostic value of the two autoantibodies.

By analyzing the gene expression profiles of esophageal cancer tissues and adjacent tissues, the differentially expressed genes in esophageal cancer tissues were significantly higher or lower than those in normal esophageal tissues. After differential analysis, a total of 375 DEGs with high expression and 449 DEGs with low expression were obtained in esophageal cancer, and the results of the visualization are in a volcano plot (Figure 1). The PPI network generated by STRING is presented in Figure 2 for highly DEGs. The 20 hub genes calculated using the cytoHubba plugin were IL1B, MMP9, CXCL8, COL1A1, SPP1, TIMP1, CXCL10, STAT1, ICAM1, SERPINE1, COL1A2, POSTN, SOX9, MMP1, MMP3, COL3A1, BGN, CXCL9, THY1, and COL4A1 by degree algorithm (Figure 3A), and the 20 hub genes calculated using the cytoHubba plugin were CXCL10, STAT1, IFIT1, RSAD2, ISG15, IFIT3, OASL, OAS2, DDX60, IFI44L, IFI27, IFI6, CMPK2, EPSTI1, TIMP1, POSTN, COL1A1, COL1A2, COL3A1, and BGN by MCC algorithm (Figure 3B).

A total of eight genes were selected by both algorithms, including CXCL10, STAT1, POSTN, TIMP1, COL1A1, COL1A2, COL3A1, and BGN (Figure 3C). According to the importance ranking of these eight genes in the two algorithms and the query of the relevant literature, we finally determined POSTN and TIMP1 as the target genes of this study. In terms of importance ranking, according to the comprehensive importance ranking calculated by the two algorithms, the order of the eight genes was CXCL10, STAT1, TIMP1, COL1A1, POSTN, COL1A2, BGN, and COL3A1. In addition, POSTN functioned as a cell adhesion molecule and participated in many biological processes, including cell adhesion, invasion, metastasis, and tumor angiogenesis (Bao et al., 2004; Kudo et al., 2006; Siriwardena et al., 2006). TIMP1 promotes the growth of human keratinocytes and several other cell types, inhibits apoptosis, and promotes growth (Gasson et al., 1985; Bertaux et al., 1991; Guedez et al., 1998). A high expression of TIMP1 has a significant correlation with a poor prognosis of cancer (Jackson et al., 2017). Therefore, the proteins encoded by POSTN and TIMP1 have the possibility of being potential tumor-associated antigens, and we further detected the level of anti-TAA autoantibodies in the subjects’ serum by ELISA experimental.

The expression levels of anti-POSTN and anti-TIMP1 autoantibodies obtained by bioinformatics methods in ESCC patients and normal controls were verified by ELISA. Sera of 185 ESCC patients and 185 normal controls were used in ELISA. The study was approved by the Medical Ethics Committee of Zhengzhou University, and informed consent was obtained from all participants. The detailed clinical information of 370 participants is described in Table 1. There was no significant difference in gender (p = 0.658) and age (p = 0.223) between ESCC patients and normal controls.

The expression levels of anti-POSTN and anti-TIMP1 autoantibodies in ESCC patients and normal controls were detected by ELISA. The expression level of anti-POSTN autoantibody in ESCC patients was distinctly higher than that in the normal controls (mean ± SD: 0.292 ± 0.149 vs. 0.224 ± 0.087) (Figure 4A). Anti-POSTN autoantibody can diagnose ESCC patients with an AUC of 0.638 at the specificity of 90.27% and sensitivity of 27.57% (Figure 4B). The expression level of anti-TIMP1 autoantibody in ESCC patients was elevated compared with that of normal controls (mean ± SD: 0.274 ± 0.106 vs. 0.240 ± 0.085) (Figure 4C). Anti-TIMP1 autoantibody can diagnose ESCC patients with an AUC of 0.585 at the specificity of 90.27% and sensitivity of 20.54% (Figure 4D). In order to further evaluate the diagnostic values of anti-POSTN and anti-TIMP1 autoantibodies in ESCC, other diagnostic parameters were also calculated, as shown in Table 2. Although the specificities of the two biomarkers were same, the sensitivity and accuracy of anti-POSTN were higher than those of anti-TIMP1. In terms of the diagnostic test, when the positive likelihood ratio was larger and the negative likelihood ratio was smaller, the diagnostic effect of biomarkers was better. Table 2 indicates that the diagnostic value of anti-POSTN autoantibody was more excellent than that of anti-TIMP1 autoantibody. All in all, the diagnosis effect of anti-POSTN for ESCC was better.

The positive rate of autoantibodies in patients was calculated by taking the mean value of autoantibodies in normal controls plus the expression level of standard deviation as the cutoff value. The diagnostic values of anti-POSTN and anti-TIMP1 autoantibodies in ESCC patients of different clinicopathological characteristics, including lymphatic metastasis, distance metastasis, differentiation, TNM stage, family tumor history, gender, and age, were further explored.

Anti-POSTN autoantibody can diagnose ESCC patients from normal controls in most subgroups except for patients with family tumor history and moderate and high differentiation, and it showed marginal difference in diagnosing ESCC patients younger than 60 years old (p = 0.0491) (Figures 5A–N). The AUCs of anti-POSTN autoantibody in diagnosing ESCC patients with different clinical characteristics ranged from 0.612 to 0.753 (p<0.05). The minimum AUC of 0.612 was observed in male ESCC patients (Figure 5K), and the maximum AUC of 0.753 was presented in patients with distance metastasis (Figure 5C). Anti-TIMP1 autoantibody can diagnose ESCC patients from normal controls in most subgroups (p < 0.05), but it failed to distinguish ESCC patients from normal controls in patients with lymphatic metastasis, a history of family tumor, different degrees of differentiation, and patients younger than 60 years old as well as male patients (p > 0.05) (Figures 6A–N). The AUC values of anti-TIMP1 autoantibody in different clinical subgroups ranged from 0.585 to 0.679(p<0.05). The minimum AUC of 0.585 was observed in both patients older than 60 and patients with no distance metastasis (Figures 6D,M), and the maximum AUC of 0.679 was detected in patients with distance metastasis (Figure 6C).

The positive rates of anti-POSTN and anti-TIMP1 autoantibodies were of no difference in clinical subgroups (p > 0.05) (Table 3). Furthermore, the AUC values of the two autoantibodies were not statistically different in all clinical subgroups (p > 0.05).