Identification of an RNA-Binding-Protein-Based Prognostic Model for Ewing Sarcoma

Abstract

RNA-binding proteins (RBPs) are important transcriptomic regulators and may be important in tumorigenesis. Here, we sought to investigate the clinical impact of RBPs for patients with Ewing sarcoma (ES). ES transcriptome signatures were characterized from four previously published cohorts and grouped into new training and validation cohorts. A total of three distinct subtypes were identified and compared for differences in patient prognosis and RBP signatures. Next, univariate Cox and Lasso regression models were used to identify hub prognosis-related RBPs and construct a prognostic risk model, and prediction capacity was assessed through time-dependent receiver operating characteristics (ROCs), Kaplan-Meier curves, and nomograms. Across the three RBP subtypes, 29 significant prognostic-associated RBP genes were identified, of which 10 were used to build and validate an RBP-associated prognostic risk model (RPRM) that had a stable predictive value and could be considered valuable for clinical risk-stratification of ES. A comparison with immunohistochemistry validation showed a significant association between overall survival and NSUN7 immunoreactivity, which was an independent favorable prognostic marker. The association of RBP signatures with ES clinical prognosis provides a strong rationale for further investigation into RBPs molecular mechanisms.

Keywords: Ewing sarcoma; RNA-binding proteins; prognosis prediction; regulation network; risk model.

Figure 1

DERBP identification and the transcriptional subtypes of ES (A). Consensus matrices of identified clusters (k = 3). (B). Kaplan-Meier curves show the overall survival in ES patients among the three subtypes (p = 0.016, p = 0.045, p = 0.003 for RS1 vs. RS2 vs. RS3, RS1 vs. RS2 and RS1 vs. RS3, respectively). (C). Abundances of 22 DERBPs (identified in dead ES patients) in the three RSs. (D). Expression profile of RBPs between RS1 versus RS2 and RS1 versus RS3, respectively. Red and blue dots represent upregulated and downregulated RBPs in RS2 or RS3, respectively (FDR < 0.05). (E). Overlap up-regulated and down-regulated DEREPs between RS1 versus RS2 and RS1 versus RS3.

Figure 2

Functional enrichment analysis of DERBPs and PPI Network (A). GO terms (biological process) grouped network (kappa score levels ≥ 0.45, p < 0.01). Ellipse represents the GO terms. The node size represents the significance of the term enrichment (p-value), and the colors represent different functional groups. (B). The REACTOME pathway grouped network (kappa score levels ≥ 0.45, p < 0.01). Ellipse represents the GO terms. The node size represents the significance of the term enrichment (p-value), and the colors represent different functional groups. (C). The MCC score of top 50 genes in the PPI network of DERBPs (combined score ≥ 0.7) is represented by a red to yellow gradient.

Figure 3

RBP regulatory network and prognostic risk model (A). Univariate analyses of 29 significant DERBPs with overall survival (p < 0.05). (B) Abundances of 18 DETFs (identified in dead ES patients) in the three RSs. (C) Overlap of DETFs between RS1 versus RS2 and RS1 versus RS3. (D). DERBPs network in ES. Circles represents all genes. Red indicates TF; blue indicates high risk RBP; and green indicates low risk RBP. The size of each circle indicates the degree of correlation. Grey lines indicate positive correlations, and purple lines indicate negative correlations (r > 0.4, p < 0.01). (E). Partial likelihood deviance under each log (lambda) was drawn in a LASSO Cox regression model. (F). The change trajectory of each independent variable in the model. (G). ROC curve of the prognostic values of the RPRM risk model in training group in 1-, 3- and 5-year OS.

Figure 4

Validation and assessment of RPRM (A). Kaplan–Meier curves show overall survival between high-risk and low-risk patients in the training group (p < 0.001). (B). Distribution of risk score and survival time of patients in the training group. The patients were divided into high-risk and low-risk subgroups based on the median value of the risk score. Blue and yellow dots represent high- and low-risk patients, respectively. In the plot below, red and green dots indicate dead and live patients, respectively. (C). Abundances of 10 significant RBPs (involved in RPRM) in the training group. (D). ROC curve of the prognostic values of RPRM risk model in 1-, 3- and 5-year validation groups. (E). Kaplan–Meier curves show overall survival in validation groups between high- and low-risk patients (p < 0.001). (F). Nomogram for predicting 1-, 3-, and 5-year survival probability of ES patients in the training group. The total score of these genes for each patient is on the total points axis, which corresponds to the survival probabilities plotted on the three axes below. (G). Abundances of 23 DEHallmarks between high- and low-risk patients in the three RSs. (H). The correlation between LDA score and enrichment scores of 23 DEHallmarks. * p < 0.05; ** p < 0.01, *** p < 0.001.

Figure 5

Validation of the prognostic and expression value of the RBPs involved in RPRM in training cohort (A). Kaplan–Meier curves show the overall survival in training group with high-risk and low-risk subgroups by DDX23, NSUN7, and RPL6, based on the median value of these genes, respectively. (p < 0.05). (B). The expression levels of three significant genes in the training group by survival status. The Wilcoxon rank sum test was used to compare the differences between groups. * p < 0.05. (C). Kaplan–Meier curves show overall survival in validation group with high-risk and low-risk subgroups by DDX23, NSUN7, and RPL6 based on the median value of these genes, respectively. (D). The expression levels of DDX23, RPL6, NSUN7 in validation group by survival status. The Wilcoxon rank sum test was used to compare the differences between groups. ** p < 0.01, ns: no significant.

Figure 6

Examples of histology imaging and follow-up in patients with ES. (A). Comparison of the survival status and the NSUN7 IHC scores in 24 ES patients (Fisher’s exact test). (B). Kaplan–Meier curve shows the overall survival of NSUN7 positive and negative in the subgroup of IHC staining. (C). Comparison between NSUN7 immunoreactivity in negative, weak, and moderate scoring at 100× and 400× magnification.