2.5. Statistical Analysis

We analyzed the association of gene expression or other clinical parameters (sex, age, tumor location, tumor stage) with survival using a Cox proportional hazard model. In the multivariable regression model, only the factors that were statistically significant in univariable analyses were included. After determining the cut-off value that allows most significant split with log-rank test in the TCGA dataset, we used the same cut-off value to generate Kaplan-Meier plots for all cohorts. The Spearman's rank correlation was applied for the strength of association between tested genes. The differences between groups were analyzed using 2-way ANOVA analysis, t-test, or nonparametric test, with the GraphPad software. Graphics represent the mean ± standard deviation from at least two independent experiments repeated in triplicate, unless otherwise stated. Statistical significance was considered if P < 0.05.

Additional methods, including Cell Culture, RNA interference experiments, Proliferation Assay, Cell Scratch Assay, Cell-Cycle Assay, Click-iT EdU Assay, Reverse Transcription Quantitative RT-PCR Analysis, Western Blot, Plasmid and Virus Generation, and Northern Blot are available in Supplemental Experimental Procedures. Reagent, primer, and antibody information is available in Table S3–5.