Abstract
Alternative splicing (AS) plays critical roles in acute myeloid leukemia (AML), but the prognostic values of AS events are rarely studied. In this study, we performed comprehensive analysis in AS events of 126 AML patients from the Cancer Genome Altas (TCGA) database by using the TCGA Splice Seqdataset. Univariate Cox analysis was performed to identify prognosis-associated (PA) AS events. Then LASSO regression analysis was conducted to obtain appropriate PAAS events and multivariate Cox analysis was used to build the risk models of all PAAS events and seven individual PAAS events, which were verified by Kaplan-Meier plot and AUC curve. Later, the correlation between PAAS and splicing factors (SFs) was analyzed by Spearman's correlation analysis. Gene function analysis was used to explore the role of SFs in AML development. A total of 1,847 AS events were related to overall survival of AML patients. All risk score models that were constructed based on prognosis-associated (PA) AS events of different AS types showed superior accuracy in predicting 5-year survival, especially the model of alternative acceptor (one subtype of AS) with an area under the receiver operating characteristic (ROC) curve of 0.953. And the risk score turned out to be an independent prognostic factor in AML. Fifty-five differentially expressed splicing factors (SFs) were found and four (JUN, YBX3, HSPA1B and RNU5A-1) were correlated with PAASs as regulators. And YBX3 was considered the core SF as its prognostic value in patients with AML. Function prediction suggested YBX3 played key roles in AML differentiation. Notably, 75.7% of prognostic AS events showed differential splicing between AML and normal controls, supporting their biological relevance. Our findings revealed that AS events were excellent outcome predictors for AML patients and they provided clues of potential mechanisms and therapeutic targets of AML.