3.1 Subgroup analysis of m6A regulators
As a result, a total of thirteen m6A RNA methylation regulators changed mRNAs expression values and clinicopathological characteristics of \UM were obtained from TCGA and GEO. Based on “ClassDiscovery” algorithm, 80 UM patients from TCGA and 28 UM patients from GEO can be identified two clusters of groups, respectively. (Figure 1A and B). Then, we contrasted the clinical features of these two subgroups, namely, C1 and C2. The subgroups analysis of clinical characteristics showed that only time and Chromosome.3.status have a signifcant difference (Table 1 ). The others clinical characteristics like stage, gender and age have no statistical significance. To find out the potential correlation of overall survival with C1 and C2. Kaplan-Meier survival analysis was performed and the curves showed that overall survival of samples in C2 is longer than the samples in the C1 group (Figure 1C ,D ). Then, expression levels of thirteen m6A RNA methylation regulators in UM patients with different C1/2 groups were shown in Figure 1E ,F.
3.2 Gene mutation and m6A regulators
Then, we assessed the relationship between gene mutation and m6A regulatorys. we firstly evaluated the gene mutation in 80 UM samples at TCGA database and found 20 highly variant mutated genes. (Figure 2A ).The heatmap of m6A regulatorys expression and 20 highly variant mutated genes indicated that SF3B1, CYSLTR2 and ADAMTSL1 were the most significantly regulated the expression of m6A regulatorys. (Figure 2B ). Kaplan-Meier analysis of these 3 mutant genes showed that only SF3B1 have a significant difference with overall survival. And then studied the relationship between SF3B1, CYSLTR2 and ADAMTSL1 mutation status and expression levels of each m6A RNA methylation regulator in TCGA database, respectively. The results showed that there are significant differences between with mutant-SF3B1 and wildtype-SF3B1 for the expression levels of ALKBH5, FTO, WTAP, YTHDF1, YTHDF2, YTHDC2 and KIAA1429 respectively(Figure 2C ).Compared with mutant-CYSLTR2 and wildtype-CYSLTR2, the expression levels of ALKBH5, FTO, METTL14, WTAP, YTHDF2, YTHDC2, ZC3H13, KIAA1429 and RBM15 are significantly different (Figure 2D). The subgroup analysis of mutant-ADAMTSL1 and wildtype-ADAMTSL1 also showed that m6A regulators of ALKBH5, METTL14, WTAP, YTHDF2, YTHDC1, YTHDC2, KIAA1429 and RBM15 are also significantly different (Figure 2E).
3.4 Clustered molecular subtype of uveal melanoma
The above results revealed that the clustered molecular subtype was intimately related to the prognosis of uveal melanoma. For better understanding of the interrelations among the thirteen m6A regulators, we also analyzed the interrelation (Figure. 3A) and correlation (Figure. 3C) among these regulators. ALKBH5 seems to be the hub gene of the ‘Eraser’, and correlated or co-expressed with METTL3, WTAP, YTHDF2, M ETTL14, YTHDF1, YTHDC1, YTHDC2, RBM15, KIAA1429. The correlation analysis of these regulators showed that ALKBH5 was also significantly negatively correlated with METTL3, RBM15, KIAA1429, YTHDC1, YTHDC2 and HNRNPC. Principal components analysis showed that C1 samples and C2 samples in TCGA datasets could be well differentiated based on the expression of m6A regulators. (Figure. 3B). To investigate biologic pathways shared by the different C1/2 subtype, we performed GSEA analysis. According to the following criteria: p value<0.05 and | NES | ≥1. 49 BP terms were differentially enriched in C2 expression phenotype. The top 5 BP terms indicated that pathways are commonly enriched T cell mediated pathways , including positive regulation of T cell mediated cytotoxicity, antigen processing and presentation of endogenous antigen, regulation of T cell mediated cytotoxicity, positive regulation of T cell mediated immunity and regulation of T cell mediated immunity. (Figure 3D). What’s more, The GSEA analysis of malignant hallmarks of tumors showed that 9 terms including mTORC1 signaling, oxidative phosphorylation, interferon-a response and apoptosis signaling were significantly associated with the C1 subgroup expression phenotype. (Figure 3E).
3.5 Identification and confirmation of m6A regulators signature
For better predict the clinical and pathologic outcomes of UM with m6A regulators. Then we used LASSO modelling to evaluate associations between generally changed thirteen m6A regulators and overall survival in TCGA dataset. Totally, a 2-m6A regulators signature was screened out of thirteen m6A regulators to build the risk signature based on the minimum criteria.( Figure 4A, B)The risk score formula for OS was calculated as follows: risk score = 0.02 × (expression value of ALKBH5) + -0.01 × (expression value of YTHDC2). The risk system reckons a risk score for each patient. Applying the cut-off value (0.664) of the risk scores. 80 UM patients were divided into high-risk and low-risk groups (Figure 4C). The life status and 2 m6A regulators expression value of each patient are showed in Figure 4C as well. Kaplan-Meier curve indicated that there is a significant difference between high-risk and low-risk group with log-rank test of p=0.0052 (Figure 4D). To verify the predictive ability of the 2 m6A regulators, validation analysis was performed in GEO dataset. The curve of Kaplan-Meier revealed that the low-risk group have a significantly better survival than the patients in high-risk group with log-rank p=0.047. The subgroups analysis of clinical characteristics between low- and high- risk groups showed that only time in TCGA and GEO have a significant difference (Table 2 ).Combination group analysis of 2 m6A regulators (ALKBH5 and YTHDC2) signature showed that patients with high expression of ALKBH5 and Low expression of YTHDC2 markers have the worse overall survival (p<0.0429) of all four groups. (Figure 5)