Data Collection and Pre-processing
The overall design and procedures are described in a flow chart (Fig. 1). Publicly available mRNA-seq data in ESCC cancer tissue and adjacent noncancerous tissue samples were directly downloaded from the TCGA data portal (http://cancergenome.nih.gov/). We obtained the miRNA profiles of 81 ESCC cancer tissue samples and 1 adjacent noncancerous tissue samples together with the clinical information (level 3) of the corresponding patients. The DEGs were identified by calculating the FC (|log2(FC)| >1 and adjusted P-value < 0.05) with the R package edgeR.
Construction of Weighted Co-expression Network and Division of Co-expression Modules
Firstly, we constructed a Pearson’s correlation matrix of all pairwise genes. Secondly, we converted the Pearson’s correlation matrix into an adjacency matrix (scale-free network) by the soft-thresholding value. To decide the most appropriate the soft-thresholding value, we calculated the scale-free fit index and mean connectivity. Then, we transformed the adjacency matrix into a topological overlap matrix (TOM) by calculating the topological overlap between pairwise genes, by which we could take indirect correlations into consideration as well as reduce noise and spurious correlations. Finally, we used the average linkage hierarchical clustering based on the TOM-based dissimilarity measure to divide genes into several co-expression modules, so that genes with co-expression relationships were gathered in the same module and genes expressed separately were divided.
Identification of Clinical Significant Modules
Gene significance (GS) and module significance (MS) were used to identify clinical significant modules. The module with the largest absolute MS was generally considered to be a module related to clinical characteristics. Finally, select modules that were highly relevant to certain clinical features for further analysis.
Identification of hub Genes
The hub genes were defined as genes with high module membership (MM) (cor. Weighted > 0.8). Then, the protein–protein interaction (PPI) network was also constructed based on the STRING database (https://string-db.org/). In the PPI network, genes with Top 30 hubba nodes ranked by Maximal Clique Centrality (MCC) were also defined as hub genes. The common hub genes in both co-expression networks and PPI networks were regarded as “real” hub genes for further analyses.
Construction of a prognosis genes model
After filtration of hub Genes through WCGNA and PPI, candidate prognostic genes were selected via integrated analysis of two algorithms consisting of the LASSO algorithm with penalty parameter tuning conducted, and the SVM-RFE algorithm searching for lambda with the smallest classification error to determine the variable. According to the optimal cut-off value of prognostic genes, patients were divided into high-risk group and low-risk group. A multivariate Cox regression model was finally used to construct a prognostic signature based on the candidate genes generated from the above filtration. A receiver operating characteristic (ROC) curve was used to estimate the accuracy and efficiency of the signature in a time‐dependent manner.
Predictive Nomogram Construction
In order to determine the independent prognostic value of genes and clinic pathological parameters (including age, gender, TNM staging) in the TCGA dataset. P < 0.05 was considered statistically significant. After the collinearity test, all independent prognostic parameters and important clinical parameters were included in the prognostic nomogram constructed by the Cox regression model. The nomogram was used to evaluate the prognostic significance of these parameters in ESCC. The nomogram calibration curve was used to compare predicted and observed overall survival rates. According to the total score of the nomogram, the best cut-off value of RiskScore was calculated. According to the RiskScore, and the patients were divided into high or low group. The survfit function was further evaluated to analyze the prognostic by logrank. The 1-year, and 3-year AUC were analyzed by pROC package.
ESCC cell lines and siRNA infection
ESCC cell lines including TE-1, KYSE150 and KYSE520 were cultured in DMEM medium (Gibco, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) (Gibco, Carlsbad, CA, USA). To establish transfectants with ENO1 knockdown, TE-1 and KYSE150 were transfected with siRNAs (target sequence for siENO1-1#: 5'-GCCAUGCCAGGGAGAUCUUUTT-3', siENO1-2#:5'-GCUGGCAACUCUGAAGUCATT-3', siENO1-3#: 5'- CCCAGUGGUGUCUAUCGAATT-3'). The transfection was performed using jetPEI (Dakewei, China).
Quantitative real-time PCR assay (qRT-PCR)
Total RNA from cell lines was isolated by a trizol agent and reverse transcribed to synthesize complementary DNA (cDNA) by the RevertAid First Strand cDNA Synthesis Kit. We performed the qRT-PCR with SYBR Green Mix in the RT-PCR detection system (Bio-Rad, USA) based on the manufacturer’s protocols. The relative mRNA expression of ENO1 was measured by 2-∆∆Ct methods and the GAPDH was served as an internal control. The primers involved in our study are as follows:
ENO1 forward:5’- AGTCTACGGGACCGAAAGACA − 3’
ENO1 reverse: 5’- CAGACCTTGCAGTTCGTTCAG − 3’
GAPDH forward: 5’- GGACCTGACCTGCCGTCTAG − 3’
GAPDH reverse: 5’- GTAGCCCAGGATGCCCTTGA − 3’
Cell proliferation assay
The transfected cells were plated in 96-well plates at 104 cells per well with 100 µl of culture medium in a humidified condition at 37 ℃. Then, 10 µl CCK-8 solutions were added into each well and further incubated in humidified condition at 37 ℃ for 1 h. Finally, the cells were measured at the absorbance of 450 nm according to the indicated time point. Each experiment was repeated at least three times.
Cell migration and invasion assay
As for the cell migration experiment by wound healing assay, the cells were sowed in 6-well plates with serum-free culture medium (106 cell/well). The scratch was formed by the tip of a 200 µl plastic pipette. After 48 hours, the migrated cells were washed with phosphate buffer solution (PBS) and fixed with 4% paraformaldehyde. The migration distance was calculated under an optical microscope.
As for invasion assay through the 8- um chamber. Firstly, the chamber was pretreated with matrigel. Then, 5×104 cells were seeded into the serum-free upper chamber, while medium with 10% FBS was added to the lower chamber. After 24 hours, the subsurface migrating cells will be washed with PBS, fixed with 4% paraformaldehyde and stained with crystal violet solution. The invasion capability was determined by calculating the migrating cells on the sub-surface through an optical microscope.
Western blot assay
Total protein was extracted with RIPA buffer containing protease and phosphatase inhibitor. After being blocked, the membranes were incubated with the primary antibody for 1 h at room temperature. Then, the membranes were incubated with secondary antibodies at room temperature for 1 h. The protein bands were visualized using enhanced chemiluminescence chromogenic substrate with horseradish peroxidase.
Statistical analysis
All statistical analysis was conducted with SPSS 21.0 and the results in our study were expressed as mean ± SD. The significance of the changes between the two groups was determined by Student’s t-test, and the data were considered significant when P < 0.05.