Fig 1: External and internal validation of the three-gene prognostic model. Two independent expression microarray HCC data sets GSE54236 (A) and GSE14520 (B) were downloaded from the GEO and were used as the signature validation cohort. The risk scores were calculated as Prognostic Index (PI) = 0.384 × RTN3-0.561 × SOCS2-0.434 × UPB1. (C) Kaplan-Meier curves of overall survival in the TCGA cohort stratified by the three-gene prognostic signature. We used the time-dependent ROC curve analysis to assessed the prognostic accuracy of the three-gene prognostic signature of the microarray data and LIHC dataset (Fig. 3A lower panel, Fig. 3B lower panel, Fig. 3C lower panel). ROC: receiver operator characteristic. AUC: area under the curve. (D) Kaplan–Meier curves of HCC patients based on the SOCS2 (left panel), RTN3 (middle panel), or UPB1 (right panel) expression level. 360 patients were sorted by the RSEM values of a gene, and the lower third of the patients was defined as the low mRNA expression, the upper third as high mRNA expression, thus divided patients into 3 groups. (E) The survival curve of HCC patients stratified by T stage and gene expression. Firstly, 360 patients were divided into two groups based on SOCS2 (left panel), RTN3 (middle panel), or UPB1 (right panel) expression. The cutoff value of high and low expression was set as the median. Then, patients were classified into three subgroups according to the T stage. (F) Survival curves of 360 HCC patients based on the combination of RTN3 and UPB1 (left panel), RTN3 and SOCS2 (middle panel), SOCS2 and UPB1 (right panel) expression. The cutoff value of high and low expression was set as the median.
Fig 2: Flowchart describing the process used to generate and validate the prognostic signature in the analysis. The level 3 RNAseqv2 RSEM_genes_normalized files and clinical datasets were downloaded from TCGA. 10 non-HCC patients were ruled out. The gene abundance was filtered by requiring the RSEM normalized count to be =3 in =60% of 49 paired tumor tissues or nontumor tissues. The remaining 15353 genes were recruited for the differential expression analysis with paired t-test or Wilcoxon signed-rank test. The resulting 9932 differentially expressed genes were candidates for prognostic analysis. Clinical characteristics were selected as variables for univariate and cox regression analysis. T stage and tumor weight were found to be statistically significant in both univariate and multivariate cox regression analysis. 360 patients were stratified by T stage, divided into 3 groups. Kaplan-Meier survival analysis was performed in every group with the median as the cutoff value. We obtained 15 common genes in three groups, and further narrowed this gene list with univariate and cox regression analysis and built a prognostic model that included 3 genes: UPB1, SOCS2 and RTN3. This model was used to calculate risk scores for two independent expression microarrays retrieved from GEO. Finally, we validated the expression pattern of UPB1, SOCS2 and RTN3 by western blot and immunohistochemistry with our HCC samples.
Fig 3: Identification of prognostic genes with T stage stratification. (A) Kaplan-Meier curve for overall survival in HCC patients with different T stages from TCGA LIHC dataset. Patients were divided into 3 groups according to the T stages (T1, n = 176; T2, n = 89; T3 and T4, n = 92). Advanced T stage is a significant factor for poor prognosis (p < 0.0001, log-rank test). (B) All HCC patients were divided into three groups based on the T stages, and Kaplan-Meier survival analysis was performed in each group separately. Two subgroup were separated on the basis of the median expression level of each gene. 617 survival-related genes were obtained from T1 group, and 672 genes from T2 group, 1590 genes from T3/4 group. 15 common genes were found in different groups. (C) Cluster analysis of the 13 survival related genes based on the spearman correlation coefficient. The color in the heatmap represents spearman correlation coefficient. Red indicates positive correlation, and blue indicates negative correlation. 13 genes were clustered into 4 clusters, and RTN3, SOCS2, UPB1 (marked in red) were in different groups.
Fig 4: Validation the expression pattern of prognostic genes in HCC samples. (A) Western blot analysis of UPB1 (upper panel), SOCS2 (middle panel), and RTN3 (lower panel) in 7 paired normal and tumor tissue samples. (B) Western blot quantitative densitometry of the UPB1, SOCS2, and RTN3 relative expressions. Proteins was quantified and normalized to ß-actin. Statistical analysis was done using Wilcoxon rank test. Data are expressed as mean values ± SD. *p < 0.05, **p < 0.01. (C) Representative images from immunohistochemistry (IHC) staining of UPB1 (upper panel), SOCS2 (middle panel), and RTN3 (lower panel) in 82 normal tissues and paired HCC tissues. Left panel: 100 × magnifications; right panel: 200 × magnifications. Bars = 100 µm. (D) Statistical analyses of the average IHC scores of UPB1 (upper panel), SOCS2 (middle panel), and RTN3 (lower panel). Statistical analysis was done using Wilcoxon rank test. Data are expressed as mean values ± SD.
Fig 5: Functional analysis of prognostic genes UPB1, SOCS2, and RTN3 using TCGA dataset. (A) The expression pattern of UPB1 (left panel), SOCS2 (middle panel), and RTN3 (right panel) in 49 HCC tissues and paired adjacent non-tumor tissues from TCGA dataset. The Shapiro-Wilk test was applied to determine whether data followed a normal distribution. The paired t-test was applied to normally distributed data otherwise the Wilcoxon rank test for paired data was applied to assess the expression pattern in HCC tissues. UPB1 and SOCS2 were downregulated in HCC tissues, while RTN3 was upregulated. (B) Association of UPB1 (left panel), SOCS2 (middle panel), and RTN3 (right panel) expression with vascular invasion. Mann-Whitney-Wilcoxon test showed significant differences between vascular invasion group (n = 200) and no vascular invasion group (n = 105) of UPB1 and SOCS2 expression, but not RTN3. (C) Association of UPB1 (left panel), SOCS2 (middle panel), and RTN3 (right panel) expression with the histologic grade. 356 patients with histologic grade information (grade 1, n = 52; grade 2, n = 171; grade 3/4, n = 133) were recruited for the analysis. The Kruskal-Wallis test revealed a negative correlation between UPB1 and SOCS2 expression and histologic grade, while no significant differences was found in RTN3 expression in different groups. (D) Association of UPB1 (left panel), SOCS2 (middle panel), and RTN3 (right panel) expression with pathologic stage. 337 patients with pathologic grade information (stage 1, n = 167; stage 2, n = 82; stage 3/4, n = 88) were recruited for the analysis. The Kruskal-Wallis test was performed. UPB1 and SOCS2 were found to be associated with pathologic stage, while RTN3 was not.
Supplier Page from Abcam for Anti-UPB1 antibody [EPR9132]