Identification and validation of a costimulatory molecule-related signature to predict the prognosis for uveal melanoma patients

Abstract

Uveal melanoma (UVM) is the most common primary tumor in adult human eyes. Costimulatory molecules (CMs) are important in maintaining T cell biological functions and regulating immune responses. To investigate the role of CMs in UVM and exploit prognostic signature by bioinformatics analysis. This study aimed to identify and validate a CMs associated signature and investigate its role in the progression and prognosis of UVM. The expression profile data of training cohort and validation cohort were downloaded from The Cancer Genome Atlas (TCGA) dataset and the Gene Expression Omnibus (GEO) dataset. 60 CM genes were identified, and 34 genes were associated with prognosis by univariate Cox regression. A prognostic signature was established with six CM genes. Further, high- and low-risk groups were divided by the median, and Kaplan-Meier (K-M) curves indicated that high-risk patients presented a poorer prognosis. We analyzed the correlation of gender, age, stage, and risk score on prognosis by univariate and multivariate regression analysis. We found that risk score was the only risk factor for prognosis. Through the integration of the tumor immune microenvironment (TIME), it was found that the high-risk group presented more immune cell infiltration and expression of immune checkpoints and obtained higher immune scores. Enrichment analysis of the biological functions of the two groups revealed that the differential parts were mainly related to cell-cell adhesion, regulation of T-cell activation, and cytokine-cytokine receptor interaction. No differences in tumor mutation burden (TMB) were found between the two groups. GNA11 and BAP1 have higher mutation frequencies in high-risk patients. Finally, based on the Genomics of Drug Sensitivity in Cancer 2 (GDSC2) dataset, drug sensitivity analysis found that high-risk patients may be potential beneficiaries of the treatment of crizotinib or temozolomide. Taken together, our CM-related prognostic signature is a reliable biomarker that may provide ideas for future treatments for the disease.

Keywords: Biomarker; Costimulatory molecular; Prognosis signature; Tumor immune microenvironment; Uveal melanoma.

Figure 1

The flowchart of the study.

Figure 2

Identification of costimulatory molecular genes associated with UVM prognosis. (A) 34 CM genes were identified through univariate Cox regression analysis. (B) 10-time cross-validation for tuning parameter selection in the Lasso model. (C) LASSO regression of the 34 OS-related genes. The Kaplan–Meier curves of TNFRSF19 (D), TNFRSF18 (E), LTBR (F), RELT (G), LTB (H), and TNFSF13 (I) from the TCGA dataset.

Figure 3

Consensus clustering of costimulatory molecule-related prognostic genes. Clustering analysis based on the expression profile of six CM genes (A,B). TCGA UVM cohorts were grouped into two (C) and three (D) clusters according to the consensus clustering matrix. The optimal value for consensus clustering was observed to be k = 2 by principal component analysis (PCA) (E,F). (G) Overall survival demonstrated a poorer prognosis for UVM patients in the cluster 1 cohort compared to cluster 2. (H) The heat map of two cohorts along with the expression of CM genes.

Figure 4

Evaluating the predictive efficacy of the risk score model. Kaplan–Meier curves indicated the difference between high- and low-risk group overall survival (A) and progression free interval (D) in training cohort. (G) Progression free survival analysis for high- and low-risk patients in validate cohort. The ROC curve of measuring the predictive value in training cohort (B,E) and validate cohort (H). (C,F,I) Scatterplot and heatmap demonstrating the distribution of high- and low-risk groups and the expression levels of prognosis-related genes.

Figure 5

The risk score correlated with clinicopathological features and in UVM. Risk scores were independent of age (A) and sex (B), but higher in patients with progressive tumors (Stage III, IV) (C). Univariate (D) and multivariate (E) Cox regression analysis of the relationship between each risk factor and prognosis. (F) Construction of predictive nomogram to predict the 1-, 2- and 3 year overall survival of UVM patients. *P < 0.05; **P < 0.01; ***P < 0.001.

Figure 6

The risk score correlated with immune cell infiltration. (A) The bar chart demonstrated the immune cell infiltration in different patients. (B) The heat map illustrates the correlation of CM-related prognostic genes with immune cell infiltration. (C) The infiltrating levels of 28 immune cell types in high/low-risk groups in UVM. (D) The expression of immune checkpoints in UVM. The stromal scores (E), immune score (F) and estimate scores (G) differed between the high- and low-risk patients. *P < 0.05; **P < 0.01; ***P < 0.001.

Figure 7

Enrichment analysis of differentially expressed genes (DEGs) between high- and low-risk patients. (A) High infiltration of immune cells positively correlated with high risk scores, including activated CD4 T cell, central memory CD4 T cell, gamma delta T cell, activated CD8 T cell, CD56bright natural killer cell, and central memory CD8 T cell. The representative results of GO enrichment (B) and KEGG pathways (https://www.kegg.jp/kegg/kegg1.html) analysis of DEGs (C). BP biological process, CC cellular component, MF molecular function.

Figure 8

Pathway enrichment analysis and mutation landscape between high- and low-risk patients. Gene Set Enrichment Analysis (GSEA) revealed differentially expressed genes between the two groups of patients, mainly associated with immune response (A), immune system process (B), immune cell activation (C), and cytokine production (D). (E) The waterfall plot demonstrates the differences in somatic cell mutations between high- and low-risk patients. (F) Tumor mutation burden profiles of patients in the TCGA-UVM cohort, with a median of 0.2/MB. (G) No statistically significant differences in tumor mutation burden were identified in the high- and low-risk groups.

Figure 9

Prediction of chemotherapeutic drug sensitivity. Selecting suitable drugs for patients with OS via OncoPredict. The IC50 value of crizotinib (A), selumetinib (B) and temozolomide (C) in high-risk and low-risk groups of TCGA-UVM. ****P < 0.0001.