ES is the main selective splicing event in patients with cervical cancer
We counted all AS events in 253 cancer samples, as shown in Additional file 1: Table S1; the seven AS events are shown in Fig. 1a. We detected 41,766 AS events in 9961 genes, of which the seven types of AS events were distributed as shown in Fig. 1b. A single gene might display several types of mRNA AS events, and we found that nearly one-third of all AS events were ES events.
Pronounced consistency between the genes involved in overall survival and the genes involved in recurrence
To observe the relationship between AS events and prognosis, we integrated clinical follow-up data as shown in Additional file 2: Table S2. A total of 41,766 AS events were analyzed by univariate survival analysis to observe the relationship between these AS events and the prognosis of patients with cervical cancer. By selecting p < 0.05, we obtained a total of 3306 AS events that were significantly related to survival and 2077 genes. There were 2174 AS events significantly associated with the recurrence of cervical cancer, including 1443 genes (Additional file 3: Table S3). There were 524 intersections of AS events that were significantly related to overall survival and recurrence (Fig. 2a) that contain a total of 589 overlapping genes (Fig. 2b). This shows that there is a pronounced consistency between the genes involved in overall survival and the genes associated with recurrence. The statistics of the AS events significantly related to overall survival are shown in Fig. 2, which shows that the largest proportion of ES events are reduced in relation to recurrence. In contrast, AP and AT events increased, and among the AS events associated with recurrence, AT events occurred the most (Fig. 2d), indicating that most ES events were not associated with prognosis, while approximately 10% of AT events were significantly associated with recurrence of cervical cancer.
A gene may have multiple AS events of different types
Because AS events affect gene translation and subsequent protein diversity, we selected AS events related to prognosis to analyze the distributions of the genes involved in these events. As shown in Fig. 3a, b, a gene can be seen in several different types of AS events, and each of these AS events may be related to prognosis.
AS events can be used as a new prognostic classification for cervical cancer
To observe whether an AS event can be used as an independent prognostic factor for overall survival, the multivariate Cox regression model was applied to the seven types of survival-related AS events. According to the results of the univariate Cox regression model, the first 15 genes with the most significant AS event p value per type were selected. Multivariate Cox analysis models were carried out according to the percent spliced in (PSI) value of the AS events of these 15 genes.
The results are presented in Fig. 4a–g, which shows that the seven predictors built with different types of AS events have considerable power in distinguishing favorable or poor outcomes for cervical cancer patients. As shown in Fig. 4h, the type of AS event can classify patient overall survival; the performance of the AP and AT models are optimal for overall survival (area under the curve (AUC) = 0.89), followed by the ES model (AUC = 0.86).
Then, we used multivariate Cox regression models to determine whether the seven types of AS events could be used as independent prognostic factors for predicting recurrence in patients with cervical cancer. As presented in Fig. 4i–o, the seven predictors built with different types of AS events showed considerable power in distinguishing favorable or poor outcomes for cervical cancer patients.
Regarding AS events associated with recurrence (Fig. 4p), the AT, AA, AD, AP and RI models performed optimally, with AUCs close to 1, suggesting that AS events could be used as a new method of prognostic classification.
Forest map analysis
The most significant top 15 genes in each type of AS event were selected for forest map analysis (Fig. 5). Most of the survival-associated AA events were favorable for prognosis (HR < 1) (Fig. 5a, h), which is consistent with recurrence risk, and the AP events were almost identical in recurrence and overall survival.
Gene interaction networks and functional analysis of the different types of AS events
As mentioned previously, one gene might have more than one AS event. To observe the relationship between different types of genes in AS events that are significantly related to prognosis, we mapped the genes to the STRING database to obtain the interaction of these genes using a score greater than 0.4. Cytoscape was used to visualize the results shown in Fig. 6a, b. AD and RI show most of the interactions, and most of the genes in AS events related to prognosis have protein interactions. To observe the function of genes in different types of AS events that were significantly associated with prognosis, KEGG enrichment analysis was performed for each type of gene significantly related to prognosis. The results are shown in Fig. 6c, d. We found that the enrichment of these genes is related to multiple disease pathways, suggesting that the genes are involved in many biological functions.
Relationship between gene expression profile and prognosis in AS events
To observe the relationship between gene expression and prognosis in AS events, we used TCGA RNA-seq expression profile data to analyze the survival rate related to each gene and found that 241 AS genes were associated with overall survival. The expression of 159 AS genes was significantly correlated with recurrence. Furthermore, we analyzed the correlation between the 241 genes associated with overall survival and the occurrence of AS events and found that 115 genes were significantly correlated with AS (Pearson p < 0.05). Similarly, 71 of the 159 genes associated with recurrence were significantly associated with AS events, indicating that the AS events of most genes were significantly associated with their expression.
Characteristic gene selection and construction of a prognostic model
We chose to study those genes with Pearson correlation coefficients greater than 0.2 or less than − 0.2 for gene expression and AS events. There were 46 genes related to the overall survival of patients with cervical cancer and 82 related to recurrence. Four genes were associated with both overall survival and recurrence: HSPA14, SDHAF2, CAMKK2 and TM9SF1 (correlations with transcriptome levels are shown in Fig. 7). The graph shows that all of these genes have negative correlations between AS and transcription levels.
The four characteristic genes were selected for the construction of a multivariate survival model to classify prognosis regarding AS events and gene expression profiles. The map shows that these four genes have a high prognostic classification effect in both datasets (Fig. 8b, d, f, h), and the AUC value for 5 years is greater than 0.7, with high accuracy (Fig. 8a, c, e, g). These results suggest that these four genes can be used as prognostic markers of cervical cancer. The flow chart of the whole data analysis is shown in Fig. 9.