Identifying Apoptosis-Related Transcriptomic Aberrations and Revealing Clinical Relevance as Diagnostic and Prognostic Biomarker in Hepatocellular Carcinoma

Abstract

In view of the unsatisfactory treatment outcome of liver cancer under current treatment, where the mortality rate is high and the survival rate is poor, in this study we aimed to use RNA sequencing data to explore potential molecular markers that can be more effective in predicting diagnosis and prognosis of hepatocellular carcinoma. RNA sequencing data and corresponding clinical information were obtained from multiple databases. After matching with the apoptotic genes from the Deathbase database, 14 differentially expressed human apoptosis genes were obtained. Using univariate and multivariate Cox regression analyses, two apoptosis genes (BAK1 and CSE1L) were determined to be closely associated with overall survival (OS) in HCC patients. And subsequently experiments also validated that knockdown of BAK1 and CSE1L significantly inhibited cell proliferation and promoted apoptosis in the HCC. Then the two genes were used to construct a prognostic signature and diagnostic models. The high-risk group showed lower OS time compared to low-risk group in the TCGA cohort (P < 0.001, HR = 2.11), GSE14520 cohort (P = 0.003, HR = 1.85), and ICGC cohort (P < 0.001, HR = 4). And the advanced HCC patients showed higher risk score and worse prognosis compared to early-stage HCC patients. Moreover, the prognostic signature was validated to be an independent prognostic factor. The diagnostic models accurately predicted HCC from normal tissues and dysplastic nodules in the training and validation cohort. These results indicated that the two apoptosis-related signature effectively predicted diagnosis and prognosis of HCC and may serve as a potential biomarker and therapeutic target for HCC.

Keywords: apoptosis; diagnosis; hepatocellular carcinoma; overall survival; prognosis; relapse-free survival.

Figure 1

Differential expression analysis and identification of key apoptosis genes related to prognosis in HCC. (A) The heat map showed the expression levels of 14 apoptotic genes in HCC tissues and noncancerous tissues. (B) The histogram indicated the expression patterns of 14 apoptotic genes in HCC tissues and noncancerous tissues. (C) The volcano plot indicates that the 14 differentially expressed genes satisfy the condition of P < 0.05 and log FC > 1. (D) Univariate and multivariate Cox regression analysis performed to identify the genes closely associated with the prognosis of HCC patients. (E, F) qRCR was used to measure the mRNA level of CSE1L (E) and BAK1 (F) in human L02 hepatocytes and HCC cell lines. *P < 0.05; **P < 0.01; ****P < 0.0001.

Figure 2

Measure the oncogenic effect of BAK1 and CSE1L in the HCC. (A–D) Western blot determined that expression of BAK1 and CSE1L was inhibited with BAK1 and CSE1L siRNA administration. (E–H) CCK8 assay showed BAK1 and CSE1L inhibition significantly suppressed HCC cells proliferation. (I, J) Edu assay indicated knock down BAK1 and CSE1L inhibited HCC cells proliferation, respectively. (K, L) Flow cytometry confirmed decreased BAK1 and CSE1L expression promoted HCC cells apoptosis in the HCC. (M–P) Quantitative statistical results of the effects of BAK1 and CSE1L expression on cell apoptosis. Data are shown as the mean ± SD of at least three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 3

Assessment of the predictive performance of the apoptosis-related prognostic signature. Kaplan-Meier survival analysis, heat map and time-dependent ROC curves of the prognostic signature in the TCGA HCC cohort, GEO HCC cohort, and ICGC cohort. (A, E, I) The survival status of the HCC patients with different risk scores. (B, F, J) K-M curves indicate that OS in high-risk patients is significantly lower than in low-risk patients (P < 0.05). (C, G, K) Time-dependent ROC curve analysis of the apoptosis-related prognostic signature. (D, H) K-M curves showed the difference of recurrence rates between high-risk patients and low-risk patients. (L) The difference in predictive performance of the two prognostic models.

Figure 4

Prognostic analysis of the apoptosis-related prognostic signature according to clinical TNM stage. (A) Violin plot revealing significantly different risk scores between patients in stages I and II and patients in stages III and IV (**P < 0.01). (B) Violin plot revealing significantly different risk scores between alive patients and dead patients (**P < 0.01). (C, D) Kaplan-Meier (K-M) survival analysis of different clinical TNM stages.

Figure 5

Cox regression analysis and nomogram incorporating three independent prognostic factors was constructed to predict OS in HCC patients. (A) Univariate and multivariate Cox regression analysis of the apoptosis-related prognostic signature and clinical features with overall survival (OS) in HCC patients (P < 0.05). (B) The total points were obtained by cumulatively adding the corresponding points of the independent prognostic factors on the point scale to predict the OS of the HCC patients at 1 year, 3 years, and 5 years in the nomogram. (C–E) The calibration curve to assess the predictive value of the nomogram in OS at 1 year, 3 years, and 5 years. The Y-axis represents the actual OS measured by K-M curve analysis, and the X-axis represents the OS predicted by the nomogram.

Figure 6

The time-dependent ROC and DCA curves were drawn to compare the predictive accuracy for prognosis between different nomograms. (A–C) Time-dependent ROC curve analysis validated the predictive accuracy of the nomograms constructed from different independent prognostic factors at 1 year, 3 years, and 5 years. (D–F) DCA visually assesses the clinical benefits of nomograms and the range of applications in which the nomogram can yield clinical benefit at 1 year, 3 years, and 5 years.

Figure 7

Construction and validation of diagnostic model to distinguish HCC from normal samples. (A, B) Confusion matrices of binary data showed high specificity and sensitivity of diagnostic model in the training cohort (TCGA cohort) and validation cohort (ICGC cohort). (C, D) Cox analysis and ROC curve predicted the good efficiency of diagnostic model both in TCGA cohort and ICGC cohort. (E, F) Unsupervised hierarchical clustering showed the expression of BAK1 and CSE1L in HCC and normal samples.

Figure 8

Establishment and validation of diagnostic model to distinguish HCC from dysplastic nodules. (A, B) Confusion matrices of binary data indicated high specificity and sensitivity of model to identify HCC and dysplastic nodules in training cohort (GSE6764 cohort) and validation cohort (GSE98620 cohort). (C, D) Cox analysis and ROC curve predicted the good efficiency of diagnostic model both in GSE98620 cohort and GSE6764 cohort. (E, F) Unsupervised hierarchical clustering indicated the level of BAK1 and CSE1L in HCC and dysplastic nodules.

Figure 9

The expression patterns of BAK1 and CSE1L and their predictive power for OS and RFS. (A, B) Box plot indicating higher expression of BAK1 and CSE1L in HCC tissues (n = 243) than those in normal liver tissues (n = 241) in GSE14520. (C, D) Box plot indicating that expression of BAK1 and CSE1L in HCC tissues (n = 243) were significantly higher than those in normal liver tissues (n = 202) in ICGC. (E–H) Survival analysis of BAK1 and CSE1L in Kaplan-Meier Plotte database. (I, J) K-M curves indicating the predictive power of different expression of BAK1 for OS and RFS. (K, L) K-M curves indicating the predictive power of different expression of CSE1L for OS and RFS. (M, N) Time-dependent ROC curve analysis of the two apoptosis genes in the GSE14520 cohort. (O, P) Time-dependent ROC curve analysis of the two apoptosis genes in the TCGA cohort.

Figure 10

The potential regulatory mechanism of BAK1 and CES1L in the HCC. (A, B) Dependence Analysis showed BAK1 expression had positive correlation with CSE1L in the HCC in TCGA cohort and ICGC cohort. (C, D) PPI network showed the proteins interacted with BAK1 and CSE1L. (E) GSEA analysis indicated BAK1 and CSE1L regulated the underlying signaling pathways in the HCC and the top 10 signaling pathways were showed by Sankey diagram.