Development of novel gene signatures for the risk stratification of prognosis and diagnostic prediction of osteosarcoma patients using bioinformatics analysis

Abstract

Background: Osteosarcoma (OS) is a common malignant bone cancer in children and teenagers that originates from osteoblast cells. Although many biomarkers have been reported in OS, they have not improved the prognosis of this disease. This study sought to identify effective biomarkers for the early diagnosis and prognosis of OS using a comprehensive bioinformatics analysis.

Methods: OS-associated microRNAs (miRNAs) were screened in the Human microRNA Disease Database (HMDD). The differentially expressed genes (DEGs) related to OS were screened using 3 data sets (GSE16088, GSE36001, and GSE56001) from the Gene Expression Omnibus (GEO) database. By comparing the targets of these miRNAs with DEGs in response to OS, we identified OS-associated candidate genes. The gene expression and clinical data of 96 OS samples with complete clinical information was downloaded from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) database. Comprehensive bioinformatics analyses, including univariate, multivariate Cox, and Kaplan-Meier (KM) analyses were conducted based on these data to identify the prognostic genes and construct prognostic signature for OS survival and recurrence. Logistic regression analysis was performed based on the GSE42352 data set (including 103 OS and 15 normal samples) to develop a diagnostic model for OS.

Results: By comparing the DEGs and predicted targets of the 28 OS survival-associated miRNAs, we identified 267 OS-associated candidate genes. Additionally, 14 genes were found to be significantly associated with the survival of OS patients. Finally, 3 genes [i.e., signal transducer and activators of transcription factor 4 (STAT4), heat shock protein family E member 1 (HSPE1), and actin-related protein 2/3 complex subunit 5 (ARPC5)] were integrated into a prognostic index. The 3-gene signature was an independent factor for OS survival [hazard ratio (HR) =1.699; P<0.001] and recurrence (HR =2.532; P=0.004) and was found to have an excellent predictive performance [area under the receiver operating characteristic (ROC) curve (AUC) >0.7]. Additionally, 2 genes (i.e., STAT4 and HSPE1) were identified to be associated with OS diagnosis (P<0.05). This 2-gene diagnostic signature for OS presented a good discriminative power (AUC =0.981) and the error between the predicted and actual value was 0.029.

Conclusions: We constructed a 3-gene prognostic signature and a 2-gene diagnostic signature that have the potential to assist in prognosis predicting and diagnosis of OS in clinic.

Keywords: Prognostic signature; biomarker; diagnosis; osteosarcoma (OS); recurrence.

Figure 1

Identifying OS-associated genes. (A) Twenty-eight shared miRNAs were identified by comparing 180 miRNAs related to OS and 67 miRNAs related to survival in the HMDD database. (B) DEGs were identified between the OS and normal samples in GSE16088, GSE36001, and GSE56001 data sets (|log₂FC| ≥0.5 and P value <0.05). (C) Target genes of the 28 miRNAs were predicted by miRDB and TargetScan. A total of 267 OS-associated candidate genes were identified by comparing the target genes and DEGs. DEGs, differentially expressed genes; OS, osteosarcoma; miRNAs, microRNAs; HMDD, Human microRNA Disease Database; FC, fold change.

Figure 2

Evaluation of our 3-gene signature for OS survival. (A) X-tile was used to identify the optimal cutoff value for the risk score. (B) Risk score and survival time of each patient. (C) Survival curves for high- and low-risk groups according to X-tile stratification. (D) Survival curves for high- and low-risk groups according to the median risk score. (E) ROC curves of 1-, 3-, 5-, 10-, and 15-year survival for the prognostic signature. ROC, receiver operating characteristic; AUC, area under the ROC curve; OS, osteosarcoma.

Figure 3

Examination of the predictive performance of our 3-gene signature. (A) Univariate Cox analysis of the 4 factors of gender, metastasis status, risk score, and risk stratification by X-tile. (B) Multivariate Cox analysis of the 3 factors of gender, metastasis status, and risk stratification by X-tile. (C) Multivariate Cox analysis of the 3 factors of gender, metastasis status, and risk score. (D) A nomogram was developed by integrating the expression of 3 prognosis-related genes of STAT4, HSPE1, and ARPC5 to predict the 3-, 5-, 10-, and 15-year survival. (E) The calibration curves were plotted for 3-, 5-, 10-, and 15-year survival. HR, hazard ratio; CI, confidence interval; STAT4, signal transducer and activators of transcription factor 4; HSPE1, heat shock protein family E member 1; ARPC5, actin-related protein 2/3 complex subunit 5.

Figure 4

Validation of our 3-gene signature in the GSE21257 data set. (A) Risk score and survival time of OS patients. (B) Survival curves for high- and low-risk groups stratified based on risk score. (C) ROC curves of the 3-gene signature for the prediction of 1-, 3-, and 5-year survival. (D) The nomogram was developed to predict 1-, 3-, and 5-year survival. (E) Calibration curves of the 3-gene signature for the prediction of 1-, 3-, and 5-year survival. ROC, receiver operating characteristic; AUC, area under the ROC curve; STAT4, signal transducer and activators of transcription factor 4; HSPE1, heat shock protein family E member 1; ARPC5, actin-related protein 2/3 complex subunit 5; OS, osteosarcoma.

Figure 5

Construction of the 3-gene signature for recurrence prediction. (A) Risk score and relapse rate of OS patients in the TARGET database. (B) Relapse curves for high- and low-risk groups stratified based on risk score. (C) ROC curves of our 3-gene signature for the prediction of 1-, 3-, 5-, 8-, and 10-year relapse. (D) Univariate Cox analysis of the 3 factors of gender, metastasis status and risk stratification by X-tile. (E) Multivariate Cox analysis of the 3 factors of gender, metastasis status, and risk stratification by X-tile. (F) A nomogram was developed to predict 3- and 10-year relapse. (G) Calibration curves of the 3-gene signature for the prediction of 3- and 10-year relapse. ROC, receiver operating characteristic; AUC, area under the ROC curve; HR, hazard ratio; CI, confidence interval; STAT4, signal transducer and activators of transcription factor 4; HSPE1, heat shock protein family E member 1; ARPC5, actin-related protein 2/3 complex subunit 5; OS, osteosarcoma; TARGET, Therapeutically Applicable Research to Generate Effective Treatments.

Figure 6

Construction of our 2-gene signature for OS diagnosis. (A) ROC curve of our 2-gene signature for OS diagnosis in the GSE42352 data set. (B) A diagnosis nomogram model was established based on the expression of STAT4 and HSPE1. (C) The calibration plot for the nomogram showed that the error between the predicted and actual value was 0.029. (D) The confidence ellipse based on the PCA was plotted to evaluate the effectiveness of the diagnosis model. (E,F) ROC curve of the 2-gene signature for OS diagnosis in validation sets, including the GSE19276 (E) and GSE36001 (F) data sets. ROC, receiver operating characteristic; AUC, area under the ROC curve; HSPE1, heat shock protein family E member 1; STAT4, signal transducer and activators of transcription factor 4; PC, principal component; OS, osteosarcoma.