Details of AS events
A total of 544 OV patients were included in the present study, and 47,922 AS events in 21,794 gene symbols were identified (Supplementary Table 1). The AS events consisted of 3,995 AAs in 2,770 genes, 3,494 ADs in 2,386 genes, 9,652 APs in 3,889 genes, 8,438 ATs in 3,685 genes, 19,197 ESs in 6,916 genes, 207 MEs in 201 genes, and 2,939 RIs in 1,947 genes (Figure 1A). The results showed that ES was the main splicing pattern, while ME was the least frequent event among the seven types of AS events in OV patients. It was important to note that the number of AS events far exceeded the number of genes. Figure 1B shows that one gene could undergo up to five types of AS events (Figure 1B).
Survival-associated AS events
We identified 1,472 AS events using the AS event profiles in the OV cohort, which were significantly associated with overall survival (OS) of OV patients by univariate Cox regression analysis (P< 0.05). Figure 2A lists the number of each type of AS event. To better visualize the intersection, an UpSet plot was created as shown in Figure 2B, and We found that up to three survival-associated AS events could occur in the same gene (Figure 2B). Specifically, ES, AP, AT, AA, AD, and RI were all significantly linked to the OS of patients. Figure 2C indicates the AS events that were associated with survival of patients (red dots) and not associated with survival of patients (blue dots), showing that most AS events were significantly associated with patients’ survival. Figure 2D–J show the top 20 most significant survival-associated AS events of each type. For ME events, only eight AS events were related to survival (Figure 2I). The results were shown in Supplementary Table 1.
Prognostic model selection and survival analysis
Lasso regression analysis was performed to avoid over-fitting and exclude the co-expressed AS events, which were selected by Univariate Cox analysis. Figure 3 presents the results of Lasso regression analysis including seven types of AS events and all AS events, and we selected the most highly correlated AS events. Multivariate Cox analysis was then used to construct predictive models based on particular AS events and calculate the risk scores. Table 1 lists the prognostic models of seven types of AS events and total AS events. Patients were then divided into high-risk and low-risk subgroups according to the median risk score of each model. There were 192 cases in each subgroup. The median risk scores of AA, AD, AP, AT, ES, ME, RI, and the whole cohort were 0.9788, 0.9437, 0.9235, 0.9892, 0.9083, 1.006, 0.9735, and 0.9137, respectively. According to K-M survival analysis, we found that eight types of prognostic models played significant roles in distinguishing good or poor outcomes of patients (Figure 4A-H). We plotted the ROC curves and calculated the AUC to compare the efficiency of these predictors. The results revealed that the risk score of AD reflected the greatest prognostic power with an AUC of 0.768, followed by AP with an AUC of 0.758, ME with an AUC of 0.756, and AT with an AUC of 0.755 (Figure 5A-H). Figure 6 illustrates the distribution of patients’ survival status, risk score, and the expression heatmap of prognostic models. The risk curve showed the result of patient ranking based on the risk scores. There was a difference between the high-risk group and the low-risk group in risk score. The survival status of patients disclosed that there were higher mortality rates in the high-risk group (green dots represent survival, and red dots represent death). The color transition from green to red in the heat map indicated that the PSI score of the AS events was increased from low to high.
Estimation of independent prognostic value
We used Univariate and Multivariate Cox regression analyses to estimate the independent prognostic value of age, grade, and risk score of each prognostic model. Univariate Cox regression analysis indicated that both age and risk score could predict survival of OV patients (Table 2). However, multivariate Cox regression showed that the risk score of each prognostic model was the independent prognostic indicator of OV survival. The age of predictive models, such as AA, AD, AP, ES, ME, RI, and the whole cohort, was the independent prognostic indicator of OV patients, except for AT events (Figure 7).
Correlation network of SFs
To analyze the correlation between survival-associated AS events and SFs, an AS-SF network was constructed based on the result of Pearson’s correlation test. Figure 8A showed that the network contained 56 SFs (blue triangles) and 104 survival-associated AS events, including 45 down-regulated AS events and 59 up-regulated AS events (red and green dots). The green lines represented AS events, which were positively correlated with the expressions of SFs, while red lines indicated negative correlations.
Interestingly, we found that MSI1 could positively regulate TACC2-13336-AP and negatively regulate TACC2-13333-AP. A total of 359 OV tumor tissues were used to show the correlation between expression of MSI1 and PSI value of TACC2-13336-AP (r=0.6579, P<0.0001), TACC2-13333-AP (r=-0.6554, P<0.0001).
GO functional and KEGG pathway enrichment analyses of OV patients
GO analysis demonstrated that “mRNA splicing via spliceosome”, “regulation of RNA splicing”, “mRNA processing”, “RNA processing”, and “regulation of alternative mRNA splicing via splicesome” were the most significant biological process terms. Moreover, “nucleoplasm”, “membrane”, “catalytic step 2 splicesome”, and “nuclear speck” were the most three significant cellular component terms. Besides, “poly (A) RNA binding”, “nucleotide binding”, and “ATP binding” were the most three significant molecular function terms (Figure 9A). KEGG analysis revealed four remarkably enriched pathways, including “spliceosome”, “RNA transport”, “mRNA surveillance pathway”, and “RNA degradation”. It also revealed that these genes were mainly involved in the “spliceosome” pathways (Figure 9B).