A novel immune-related gene signature predicting survival in sarcoma patients

Abstract

Sarcomas are a heterogeneous group of rare mesenchymal tumors. The migration of immune cells into these tumors and the prognostic impact of tumor-specific factors determining their interaction with these tumors remain poorly understood. The current risk stratification system is insufficient to provide a precise survival prediction and treatment response. Thus, valid prognostic models are needed to guide treatment. This study analyzed the gene expression and outcome of 980 sarcoma patients from seven public datasets. The abundance of immune cells and the response to immunotherapy was calculated. Immune-related genes (IRGs) were screened through a weighted gene co-expression network analysis (WGCNA). A least absolute shrinkage and selection operator (LASSO) Cox regression was used to establish a powerful IRG signature predicting prognosis. The identified IRG signature incorporated 14 genes and identified high-risk patients in sarcoma cohorts. The 14-IRG signature was identified as an independent risk factor for overall and disease-free survival. Moreover, the IRG signature acted as a potential indicator for immunotherapy. The nomogram based on the risk score was built to provide a more accurate survival prediction. The decision tree with IRG risk score discriminated risk subgroups powerfully. This proposed IRG signature is a robust biomarker to predict outcomes and treatment responses in sarcoma patients.

Keywords: gene signature; immune infiltration; immunotherapy; prognosis; sarcoma.

Graphical abstract

Figure 1

Flow chart of the study design (A) Identification of the immune cell subtypes with prognostic relevance. (B) The approaches used to establish an immune-related gene signature for prognosis. (C) The prognostic value of the gene signature was investigated in different cohorts. (D) This application of personalized medicine might be used in daily routine. WGCNA, weighted gene co-expression network analysis; tROC, time-dependent receiver operating characteristic.

Figure 2

Immune cells were identified as the primary protective factors for survival (A) Univariate Cox regression analysis showed that 7 T cell types, NK cells, and the infiltration score were protective factors among the scores of various immune cell subsets. (B–H) Kaplan–Meier analysis indicated that patients with higher scores of NK (B), CD4+ T (C), Tfh (D), Tcm (E), Th1 (F), NKT (G), CD8+ T (I), and Tc (J) cells and infiltration score (H) had better OS (p < 0.05). NK, natural killer cells; Tfh, follicular helper T cells; Tcm, central memory T cells; Th1, T helper type 1 cells; NKT, natural killer T cells; Tc, cytotoxic T cells.

Figure 3

The construction of co-expression modules and module-trait relationships of sarcomas (A) Visualizing the gene network using a heatmap plot. (B) Correlation of module eigengenes with all the protective immune cell scores. Each unit contains the corresponding correlation coefficient and p value.

Figure 4

Establishment of the immune-related gene signature (A) A total of 220 prognosis-related candidates were identified among 573 genes extracted from the red module (“MEred”). (B) Gene symbols and the corresponding LASSO coefficients of the IRG signature. (C) Kaplan–Meier curve revealed that patients with higher IRG risk scores exhibited worse OS. (D) The time-dependent ROC curves of the prognostic signature for OS in the training cohort. IRG, immune-related gene; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic. (E) Kaplan–Meier analysis indicated that patients with higher IRG risk scores showed shorter DFS. (F) The time-dependent ROC curves of the prognostic signature for DFS in the training cohort.

Figure 5

Independent prognostic factors for OS and DFS in the TCGA-SARC cohort (A and B) Multivariate Cox regression analysis of the relationship between clinicopathological features (including the risk score) and OS (A)/DFS (B) of patients in the TCGA-SARC dataset.

Figure 6

Validation of the OS-related prognostic relevance of the IRG signature with three external datasets (A) Kaplan–Meier analysis of OS in the TARGET-Osteosarcoma cohort (validation cohort I). (B) The AUC of the risk score predicting the 1-, 3-, and 5-year OS in the TARGET-Osteosarcoma cohort. (C) Kaplan–Meier analysis of OS in the GEO: GSE17674 cohort (validation cohort II). (D) The AUC of the risk score predicting the 1-, 3-, and 5-year OS in the GEO: GSE17674 cohort. (E) Kaplan–Meier analysis of OS in the GEO: GSE119041 cohort (validation cohort III). (F) The AUC of the risk score predicting the 1-, 3-, and 5-year OS in the GEO: GSE119041 cohort.

Figure 7

Validation of the DFS-related prognostic relevance of the IRG signature with three external datasets (A) Kaplan–Meier analysis of DFS in the GEO: GSE71118 cohort (validation cohort IV). (B) The AUC of the risk score predicting the 1-, 3-, and 5-year OS in the GEO: GSE71118 cohort. (C) Kaplan–Meier analysis of OS in the GEO: GSE30929 cohort (validation cohort V). (D) The AUC of the risk score predicting the 1-, 3-, and 5-year DFS in the GEO: GSE30929 cohort (E) Kaplan–Meier analysis of DFS in the GEO: GSE40025 cohort (validation cohort VI). (F) The AUC of the risk score predicting the 1-, 3-, and 5-year DFS in the GEO: GSE40025 cohort.

Figure 8

Immune microenvironment analysis and immunotherapy response prediction (A–C) The correlation analysis indicated that the immune score (R = −0.48; p < 0.001) (A), the stromal score (R = −0.22; p < 0.001) (B), and the ESTIMATE score (R = −0.42; p < 0.001) (C) were significantly negatively correlated with the Z score of the IRG signature risk score. (D) 29 ssGSEA enrichment levels of the immune signatures in the high- and low-risk groups. (E) The correlation analysis bubble diagram showed that 46 immune checkpoint genes were negatively correlated with the risk score. (F) The ratio of immunotherapy response is significantly increased in the low-risk group compared with the high-risk group. ssGSEA, single sample gene set enrichment analysis.

Figure 9

Combination of the IRG signature risk score and clinicopathological features improves prognosis prediction and risk stratification (A) A nomogram was established to predict 3- and 5-year OS in individual sarcoma patients. (B) The AUC of the nomogram to predict the 3- and 5-year OS in the TCGA-SARC cohort. (C) A decision tree was constructed to categorize patients into three different risk levels. (D) Kaplan–Meier analysis of OS of the three different subgroups.