General information of ACC patients in the TCGA cohort
As shown in Table 1, 92 tumors with clinical information was downloaded from TCGA. In our study cohort, the median age at diagnosis was 47.16 years old, patient age ranged from 14 year to 83 years. The overall female-to-male ratio was 3:1.6. Most patients were white (84.8%). Stage I disease was found in 9 patients (9.8%), stage II in 44 (47.8%), stage III in 19 (20.7%), stage IV in 18 (19.6%). ACC was the first diagnosed malignancy in 86 patients.
Overview of AS events and survival associated AS events in TCGA ACC cohort
Comprehensive AS events in seven splicing types, including AA, AD, AP, AT, ES, ME, and RI were summarized for ACC. A total of 34,420 AS events in 8994 genes were detected from TCGA SpliceSeq dataset, indicating that one gene may undergo multiple AS events simultaneously. In detail, 2707 AAs in 1960 genes, 2382 ADs in 1688 genes, 6342 APs in 2575 genes, 8201 ATs in 3575 genes, 12,269 ESs in 5337 genes, 124 MEs in 122 genes and 2395 RIs in 1605 genes (Fig. 1A).
Univariate Cox analysis revealed that 3781 AS events in 2366 genes were associated with ACC OS rates (P<0.05). In detail, as shown in Fig. 1B, 199 AAs in 186 genes, 248 ADs in 221 genes, 679 APs in 423 genes, 1224 ATs in 724 genes, 1184 ESs in 937 genes, 8 MEs in 8 genes and 239 RIs in 209 genes were identified as survival-associated AS events. The UpSet plot (Fig. 1C) vividly revealed that one gene could undergo multiple AS events simultaneously.
Molecular characteristics of survival-associated AS events
The distributions of AS events significantly associated with patient survival are shown in Fig. 2A. We displayed the top 20 (if available) most significant survival-associated AS events for each AS type in Fig. 2B-H. To explore the molecular characteristics of genes with the top 50 most significant survival-associated AS events (if available), we performed several bioinformatics analyses. Firstly, we established a PPI network to reveal the relationships among these genes. As shown in Fig. 3A-B, POLR2H, TCEB2, PSMA1, PSMD11 and SKP2 ranked at the core in the network. According to the functional enrichments of these genes, “intracellular membrane-bounded organelle”, “mitochondrion”, “membrane-bounded organelle”, “organelle part” and “intracellular organelle part” were the five most significant cellular component terms (GO) (Fig. 4A). For biological process terms (GO), “metabolic process”, “cellular metabolic process”, “nitrogen compound metabolic process”, “intracellular transport” and “organic substance metabolic process” were the five most significant enrichments (Fig. 4B). There were no significant pathway enrichments observed in molecular function (GO). Finally, we observed that “Thermogenesis” was the only significant pathway (FDR=0.032) correlated with these genes in KEGG pathway analysis.
Prognostic predictors for ACC patients
We used the Lasso regression and multivariate Cox regression analysis to generate prognostic models (PMs) for seven AS types and for all types: PM-AA, PM-AD, PM-AP, PM-AT, PM-ES, PM-ME, PM-RI, and PM-ALL (Fig. 5 and Table 2) following univariate Cox. Then, we divided ACC patients into low and high risk groups based on median values to analyze the efficacy of prognostic models by using Kaplan-Meier (K-M) method. As shown in Fig. 6A-H, all the prognostic models could predict good and poor outcomes of ACC patients. ROC curves validated the efficiency of these prognostic models (Fig. 6I). To further elucidate the independent prognostic significance of PM-ALL, univariate and multivariate Cox regression analyses were performed. After adjusting for the clinical factors, the PM-ALL remained an independent prognostic factor for ACC patients, with an HR of 1.012 (95%CI: 1.003-1.020, P=0.007) (Table 3).
Network of survival-associated AS genes and SFs expression
SFs are RNA-binding proteins that recognize cis-regulatory elements within the pre-mRNA to influence exon selection and splice site choice (38). SF alternations are a hallmark of cancer. Therefore, we explored the interaction networks of survival-associated AS genes and SFs. Firstly, we found 20 SFs related to survival that could be used as independent prognostic factors by using K-M method and Cox regression analysis (Table 4). Next, correlation analyses between the expression of these 20 SFs and the PSI values of survival-associated AS events were performed by using Pearson’s correlation analysis (cor > 0.6, P < 0.001). Correlation plots were then generated using Cytoscape 3.7.1. The results showed that the expression of 19 survival-related SFs (triangular nodes) were correlated with 206 survival-associated AS events (Fig. 7). Overall, 97 AS events were correlated with favorable OS (rea ovals) and 109 AS events were correlated with poor OS (green ovals).