Development of alternative splicing signature in lung squamous cell carcinoma

SpliceSeq data of TCGA-LUSC and corresponding clinical information were downloaded from the TCGA data portal (https://​portal.​gdc.​cancer.​gov/​) [24]. SpliceSeq tool, a java-based application, was usually used to unambiguously quantify the mRNA splicing levels of samples in TCGA. A novel value could be calculated by SpliceSeq based on seven types of AS events about each protein-coding gene provided from Ensemble gene database [16, 25]. For the following seven kinds of AS events, the Percent Spliced In (PSI) value was calculated, quantifying splicing event levels are ranging from 0 to 1.

A total of 487 LUSC patients were included in this study (Table S1). The PSI value of AS events in samples were collected and subjected to univariate Cox analysis. All the AS events were screened out having P value < 0.05, and these events was considered as a candidate prognosis-related events. We performed the least absolute shrinkage and selection operator (LASSO) Cox regression model using “glmnet R” package to select the most valuable and concise AS events in all AS events filtered in univariate Cox analysis (P < 0.05) and then constructed predictive models based on survival-related AS events by multivariate Cox analysis. Then, Based on the coefficient of each above AS event, each patients' risk score could be calculated by the signature, respectively. Meanwhile, all patients were divided into distinct subgroups based on the median value of risk scores.

The Kaplan–Meier curve was performed to analyze differential survival. The receiver operating characteristic curves (ROC) were conducted to explore sensitivity and specificity of prognostic signatures using “survivalROC” R package (https://​www.​r-project.​org/​, v.3.5.3). Univariate and multivariate Cox regression analysis were performed to investigate the prognostic independence of AS signature and clinical characteristics using forest plot R package.

The Upset intersective plot, a more scalable visualizing diagrams to Venn, was used to explore the interactive sets of those AS events and “UpSet” R package was used to visualize their potential interrelationship. Expression data of the Splicing factors (SFs) were extracted from TCGA-LUSC mRNA-seq data. All SF genes were subjected to univariate Cox analysis and when their P < 0.05, these SFs were considered as the survival-associated splicing factors. The relationship closeness between SFs expression value and AS’s PSI value was calculated by spearman test. Meanwhile, a network of the interaction between these SFs and prognostic AS events was illustrated by Cytoscape 3.7.0 (https://​cytoscape.​org/​).

Integrated AS events were analyzed in 487 TCGA-LUSC data set (Fig. 1a). A total of 31,345 AS events from 9633 genes were detected including 12,416 ES in 5554 genes, 6073 AP in 3369 genes, 5579 AT in 3167 genes, 2675 AA in 1996 genes, 2323 AD in 1722 genes, 2129 RI in 1474 genes, and 150 ME in 144 genes (Fig. 1b). Among them, ES events were the most common type, accounting for more than one-third of all events, followed by AP and AT events, while ME was the least. Notably, the amount of AS events far exceeded the number of genes. Furthermore, a subset of overlapping AS events among the seven types of AS in LUSC was illustrated by UpSet plot diagram (Fig. 1c).

First of all, we conducted a univariate Cox analysis of the 31,345 AS events in 487 LUSC patients to evaluate the relationship between AS events and overall survival (OS) status in LUSC. Consequently, 1996 AS events within 1409 genes were obviously related to overall survival of LUSC patients (Fig. 2a). As shown in Fig. 2b–h, the top 20 AS events were significantly related to OS among seven types of AS events. Interestingly, some of survival-associated AS genes underwent multiple types of AS events. For example, AA, AD, RI, and ES of ATXN2L and AA, AP, and RI of NPIPB4 were conspicuously related to OS of LUSC patients.

The important prognostic-related AS events in all AS events in univariate Cox analysis were selected as candidates to select the most significant AS events by LASSO Cox regression model analysis (Fig. 3). Further, several prediction signatures based on these prognostic-associated AS events were constructed by multivariate Cox analyses. Eventually, a combined prognostic model was built integrated from different types of AS events (Table S2). As shown in Fig. 4a–h, Kaplan–Meier curves showed that LUSC patients in high-risk group had appreciably shorter OS than patients in low-risk group, demonstrating that these AS signatures could be powerful biomarkers to distinguish patients’ prognosis. Obviously, the combined prognostic indicator showed better performance than single type of AS events (Fig. 4h). Then, ROC curve was performed to appraise the prognostic efficiency of prognostic AS models. The results show that all signatures had a robust predictive property with AUC values from 0.837 to 0.978 (Fig. 4i). Conceivably, the combined model with all types of AS events had highest efficiency with 0.978 (AUC) than single prognostic models. Furthermore, we also explored the predictive effect of this combined prognostic model for RFS and found it had same powerful prognostic value and high accuracy (AUC = 0.793) in LUSC (Figure S1A-B). The distribution of patients’ risk score, survival status, and expression profiles of all AS models are shown in Fig. 5.

We compared the predictive ability between our combined prognostic model and other model to verify the superiority of our model [26]. The Kaplan–Meier curve and ROC results showed that other mRNA had significantly predictive ability in LUSC, but our model had higher sensitivity and specificity than other mRNA model (Figure S2). Taken together, our combined prognostic model had superior prognostic value in LUSC patients.

More importantly, extensive dysregulated AS events in many types of cancers are easily programmed by some specific SFs. Hence, an important issue is that whether several key SFs could potentially regulate these prognosis-associated AS events in LUSC. To determine those specific SFs which had close connection with the prognosis-associated AS events in LUSC, univariate Cox analysis of SFs were implemented according to gene expression level of LUSC patients. Consequently, there were 25 SFs obviously related to OS of LUSC patients shown in Table S3. Furthermore, correlations between SFs and prognostic AS events were tested in LUSC using Spearman’s test (Fig. 6a). In correlation networks, 22 SFs (purple dots) were obviously related to 546 prognosis-related AS events, including 202 favorable AS events (green dots) and 344 adverse AS events (red dots). Interestingly, there was a positive correlation (red lines) between majority of poor prognostic AS events (red dots) and SFs (purple dots), while there was a negative correlation (green lines) between majority of favorable prognostic AS events (green dots) and SFs. For example, splicing factors SNRNP48 and DDX39B had adverse survival for LUSC patients (Figs. 6b–c). ES of NADSYN1 and AP of TMEM25 were adverse factors, whereas AT of TNFRSF1A and AT of FBXL12 were related to favorable prognosis. Correlation between SNRNP48 and AT of TNFRSF1A or AP of TMEM25 were shown in dot plots, suggesting high expression of SNRNP48 had positive association with poor overall survival (Fig. 6d–e). Similarly, correlation between DDX39B and ES of NADSYN1 or AT of FBXL12 were shown in dot plots, implicating high expression of DDX39B had negative association with favorable prognosis (Fig. 6f–g).