A novel signature derived from immunoregulatory and hypoxia genes predicts prognosis in liver and five other cancers

Abstract

Background: Despite much progress in cancer research, its incidence and mortality continue to rise. A robust biomarker that would predict tumor behavior is highly desirable and could improve patient treatment and prognosis.

Methods: In a retrospective bioinformatics analysis involving patients with liver cancer (n = 839), we developed a prognostic signature consisting of 45 genes associated with tumor-infiltrating lymphocytes and cellular responses to hypoxia. From this gene set, we were able to identify a second prognostic signature comprised of 8 genes. Its performance was further validated in five other cancers: head and neck (n = 520), renal papillary cell (n = 290), lung (n = 515), pancreas (n = 178) and endometrial (n = 370).

Results: The 45-gene signature predicted overall survival in three liver cancer cohorts: hazard ratio (HR) = 1.82, P = 0.006; HR = 1.84, P = 0.008 and HR = 2.67, P = 0.003. Additionally, the reduced 8-gene signature was sufficient and effective in predicting survival in liver and five other cancers: liver (HR = 2.36, P = 0.0003; HR = 2.43, P = 0.0002 and HR = 3.45, P = 0.0007), head and neck (HR = 1.64, P = 0.004), renal papillary cell (HR = 2.31, P = 0.04), lung (HR = 1.45, P = 0.03), pancreas (HR = 1.96, P = 0.006) and endometrial (HR = 2.33, P = 0.003). Receiver operating characteristic analyses demonstrated both signatures superior performance over current tumor staging parameters. Multivariate Cox regression analyses revealed that both 45-gene and 8-gene signatures were independent of other clinicopathological features in these cancers. Combining the gene signatures with somatic mutation profiles increased their prognostic ability.

Conclusions: This study, to our knowledge, is the first to identify a gene signature uniting both tumor hypoxia and lymphocytic infiltration as a prognostic determinant in six cancer types (n = 2712). The 8-gene signature can be used for patient risk stratification by incorporating hypoxia information to aid clinical decision making.

Keywords: Gene signature; Hepatocellular carcinoma; Hypoxia; Pan-cancer; T cells; Tumor-infiltrating lymphocytes.

Fig. 1

Schematic diagram of the study design and development of gene signatures. A liver cancer cohort (GSE14520) was used to define the first 45-gene signature. Briefly, 79 tumor-infiltrating T-cell genes were identified as HIF targets using a HIF-1α/2α ChIP-seq dataset. Of these 79 genes, 26 genes were > 1.5-fold upregulated in GSE14520. Independently, 23 hypoxia genes were identified as HIF targets. Uniting the 23 hypoxia-HIF genes and 26 T-cell-HIF genes resulted in 45 unique genes representing the first gene signature. Cox regression analyses of individual 45 genes in each of the three liver cancer cohorts (GSE14520, TCGA-LIHC and LIRI-JP) revealed a common prognostic set consisting of 8 genes that represent the second signature. This 8-gene signature is further validated in liver and five other cancers using Kaplan–Meier, Cox regression and receiver operating characteristic analyses

Fig. 2

Patient stratification using the 45-gene signature in HCC cohorts. a Kaplan–Meier plots of overall survival in HCC patients across three cohorts stratified into low- and high-risk groups using the 45-gene signature. P-values were calculated from the log-rank test. b Kaplan–Meier plots show independence of the signature over current staging systems in HCC cohorts. Patients were sub-grouped according to TNM stages and further stratified using the 45-gene signature. The signature successfully identified high-risk patients in different TNM stages. P-values were calculated from the log-rank test. c Analysis of specificity and sensitivity of the signature in HCC cohorts with receiver operating characteristic (ROC). Plots depict comparison of ROC curves of signature and clinical tumor staging parameters. The signature demonstrates an incremental value over current staging systems. AUC: area under the curve. TNM tumor, node, metastasis staging. d Kaplan–Meier plots depicting combined relation of TP53 or CTNNB1 mutation status with the signature on overall survival in HCC

Fig. 3

Minimal prognostic 8-gene signature in HCC. a Forest plots depict Cox proportional hazards analysis on 45 signature genes in three HCC cohorts. Hazard ratios (HR) were denoted as dark blue circles and light blue bars represent 95% confidence interval. Eight genes are consistently prognostic across all three cohorts, thereby constituting the minimal prognostic signature. Significant Wald test P values were indicated in red. Signature genes were highlighted in red. b The 8-gene signature successfully identified high-risk patients in different TNM stages. Kaplan–Meier plots of overall survival in HCC patients across three cohorts stratified by 8-gene signature into low and high-risk groups. Patients were stratified by the signature as a full cohort, or sub-grouped according to TNM stages. P-values were calculated from the log-rank test. Plots show independence of signature over current staging systems. c Analysis of specificity and sensitivity of the signature in HCC cohorts with ROC. Plots depict comparison of ROC curves of signature and clinical tumor staging parameters. AUC: area under the curve. TNM tumor, node, metastasis staging. d Kaplan–Meier plots depicting combined relation of TP53 or CTNNB1 mutation status with the 8-gene signature on overall survival in HCC

Fig. 4

Prognosis of the 8-gene signature in 5 other non-HCC cancers. a Kaplan–Meier plots of overall survival in patients across multiple cancers stratified into low and high-risk groups using the prognostic 8-gene signature. P-values were obtained from the log-rank test. b Analysis of specificity and sensitivity of the 8-gene signature in multiple cancers. Plots depict comparison of ROC curves of the 8-gene signature and clinical tumor staging parameters. AUC: area under the curve. TNM tumor, node, metastasis staging