Comprehensive proling of aberrant prognostic alternative splicing signatures in clear cell renal cell carcinoma: a retrospective study based on large-scale sequencing data

Backgroud: Clear cell renal cell carcinoma(ccRCC) is the most common type with poor prognosis in kidney tumor. Growing evidence has indicated that aberrant alternative splicing (AS) events are ecacious signatures for tumor prognosis predicting and therapeutic targets. Systematic and comprehensive analysis of AS in ccRCC is in urgent need. Methods: Level 3 RNA-seq data were acquired from TCGA data portal and the AS proles were performed with assistance of SpliceSeq software. Univariate cox regression analysis was applied for screening prognosis-related AS events. Gene functional enrichment analysis revealed the pathways enriched by prognosis-related AS. The nal AS panel was developed by LASSO-penalized method for predicting prognosis and compared with traditional clinical factors. The potential regulatory network was analyzed via Spearman correlation between splicing factors (SFs). Results: A total of 2100 survival-associated AS events were ltered from 1666 parent genes. Gene functional enrichment analysis suggested that the regulation of autophagy could be a potential mechanism of splicing regulatory in ccRCC. 17 aberrant AS events formed the nal AS panel which can estimate OS probability in ccRCC patients. The AUC values of ROC curves for the nal AS panel can keep above 0.7 spanning 1 year to 5 years. Conclusion: We developed a robust and individualized predictive model based on large-scale sequencing data. The identied vital AS events and splicing networks may be valuable in deciphering the potential mechanisms AS analyses of nal AS signature in ccRCC cohort. Kaplan-Meier of overall for patients in subgroup stratied by showing that the survival of patients was signicantly poorer in patients with high risk

exclusion of different exons or parts of exons in pre-mRNA. This process has a signi cant impact on the diversity of both transcriptome and proteome of human [8; 9]. AS plays a vital role in increasing the complexity of functional proteins and regulating of cell metabolism and cell-speci c functions, which provides opportunity for tumorigenesis and progression [10]. Over the last two decades, molecular tools have been developed to correct or redirect AS events. It has been observed that a switch on particular AS event could occur in cancer related genes [11]. Aberrant AS can lead to loss-of-function in tumor suppressors or activation of oncogenes [12]. Moreover, researches show that splicing factors can not only regulate AS events directly, but also can be regulated by various signaling pathways, making AS sensitive to tumor microenvironment [13; 14]. Accordingly, deciphering valuable AS events may help to elucidate underlying mechanisms of oncogenesis in ccRCC and provide new therapeutic approaches.
Most current researches of AS in tumor have focused on nding cancer-speci c AS events by distinguishing tumor tissues from normal controls [15; 16]. As far as we know, the prognostic value of abnormal AS events in ccRCC has not been comprehensively studied. In our study, we systematically analyzed genome-wide AS events in ccRCC cohort by mining The Cancer Genome Atlas (TCGA) database and further constructed a powerful predictive model based on a series of survival-related AS events. This will provide a novel insight on exploring the pathogenesis of ccRCC and make a signi cant impact on personalized precision-based care in patients.

Data acquisition and pre-processing
Level 3 RNA-seq data and the corresponding clinical datasheets for ccRCC samples were downloaded from GDC data portal of the Cancer Genome Atlas (TCGA) website (https://portal.gdc.cancer.gov/) [17; 18]. Analysis of AS pro les was performed with assistance of SpliceSeq software and TCGA SpliceSeq database, which evaluated the mRNA splicing pattern and quanti ed AS events with the Percent Spliced In (PSI) score calculated, ranging between 0 and 1 [19; 20]. To generate a reliable set of AS events, we implemented the following lter criteria: percentage of samples with PSI score ≥ 75, PSI range across samples ≥ 0.05 and standard deviation of PSI score ≥ 0.02. And the remaining AS events with not available (Null) were imputed via the k-Nearest Neighbor (kNN) imputation algorithm [21]. Besides, our subsequent analyses were based on ccRCC cohort comprised of patients that meet the inclusion criteria: (1) a histological diagnosis of ccRCC; (2) patients who did not receive prior treatment, especially neoadjuvant chemotherapy; (3) patients with relatively complete clinical features including gender, age, tumor site, Furhman grade, pathological stage, T stage, M stage and overall survival (OS) information; (4) the follow-up time was no less than 30 days; (5) patients with corresponding mRNA splicing pro les. The present study fully satis es the TCGA publication guidelines.

Differentially spliced AS events analysis
To identify differentially expressed AS events within each AS type between ccRCC and normal tissues, student's t-test, followed by an adjustment of p-value using the Benjamini-Hochberg correlation for multiple comparison, was applied. The |log2FC| > 1 and adjusted p-value < 0.05 indicated the AS events were signi cant differences between tumor and adjacent normal tissues.

Identi cation of survival-associated AS events and functional enrichment analyses
After rigorous screening, a total of 455 ccRCC patients with differentially expressed AS pro les and survival information were subjected to subsequent analyses. For each speci c AS event, patients were divided into two group based on median cut. A univariate Cox proportional hazard regression analysis was used to de ne the prognostic value of each AS event. In this analysis, AS events were regarded as signi cant at p-value < 0.05. And Upset plot, generated by UpsetR, was visualized to quantitatively analyze the gene interactions among the seven types of prognostic AS events [22]. Then to further shed light on the potential modifying mechanism of aberrant AS events on corresponding protein, the parent genes of these survival-associated AS events were sent for functional enrichment analyses via clusterPro ler R package, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis [23]. The GO terms and KEGG pathways were considered signi cant with a threshold of false discovery rate (FDR) < 0.05.

Dimension reduction and generation of AS signature
For avoiding multicollinearity of highly correlated variables, the Cox regression model, with least absolute shrinkage and selection operator (LASSO) penalty, was implemented to reduce dimension [24]. Firstly, among the top 20 most signi cant AS events in univariate analysis within seven AS types, the key prognostic biomarkers were selected to further develop prognostic signatures via LASSO-penalized method. Then we utilized the regression coe cients derived from tting multivariate Cox model to multiply the PSI scores of prognostic indicators for constructing AS signature in each AS type. Furthermore, the candidate AS events from seven signatures were combined together to build the nal AS panel for ccRCC cohort. Then we divided patients into low-risk and high-risk subgroups based on median value of each model, the Kaplan-Meier survival analysis and log-rank test were applied to verify the prognostic ability of prognostic models. Moreover, the predictive e ciency of each model was assessed by time-dependent receiver operator characteristic (ROC) curves. Therein, the LASSO Cox model was constructed with the "glmnet" R package, using the 10-fold cross-validation and "lambda.min" criteria such that the tuning parameters (lambda) were at the minimum [25]. The area under curve (AUC) of timedependent ROC curve was calculated with "timeROC" R package [26].

Independence of nal AS panel from clinicopathological features
To investigate the independent prognostic value of nal AS panel from conventional clinicopathological characteristics (including age, gender, Furhman grade, pathological stage, T stage, M stage) in ccRCC patients, univariate followed by multivariate Cox regression analyses were conducted. Then to further con rm whether the nal AS panel was of high applicability and robust in various subgroups, strati cation Cox analyses were also performed.

Construction and evaluation of the nomogram
In order to provide surgeons with a clinically relevant quantitative approach for predicting the short-term and long-term survival probability of ccRCC patients individually, we assembled a nomogram that integrated the nal AS panel and independent clinicopathological risk factors based on the above results of multivariate analysis. Then the calibration curves were graphically assessed to determine whether the derived nomogram performed well compared to ideal model.Therein, the nomogram and calibration plot were depicted using rms package [27; 28]. And Harrell's concordance index (C-index) was calculated to estimate the discrimination ability of the nomogram with the help of Hmisc package. Moreover, the clinical usefulness of the nomogram, the AS panel and clinical risk factors were compared with the aid of decision curve analysis (DCA)[29; 30; 31; 32].

Gene set enrichment analysis (GSEA) of nal AS panel
To further identify crucial biological processes and cancer-speci c pathways related to the nal AS panel, we performed a GSEA using the adjusted expression pro les for all transcripts. The annotated gene set les of "c2.cp.v6.2.symbols" and "c5.bp.v6.2.symbols" download from the "Molecular Signatures Databases (http://software.broadinstitute.org/gsea/msigdb)" were employed for running GSEA using the java software (http://software.broadinstitute.org/gsea/downloads.jsp) [33; 34]. Enrichment p-values were based on 1000 permutations and the signi cance threshold was set at nominal p-value < 0.05.

Construction of splicing regulatory network
A list of 88 human splicing factors (SFs) was created by integrating SpliceAid 2 database (www.introni.it/spliceaid.html) and the work from Xiong et al, which collected the experimentally validated SFs through hand-curated screening of literature and databases [35; 36]. First, we compared ccRCC samples and adjacent normal samples to identify differentially expressed SFs using student's ttest. Then we assessed the correlation between the normalized expression value (variance stabilizing transformation via DESeq2 package) of SFs and OS through tting univariate regression analysis in the entire cohort, where the SFs with p-value < 0.05 were selected as prognostic ones for further analysis [37].
The "surv_cutpoint" function of the "survminer" R package was used to iteratively determine the optimal cutpoints of prognostic SFs achieving the maximally selected rank statistics. Then Spearman correlation analyses were performed between expression value of prognostic SFs and PSI scores of the most signi cant AS events in each AS type. P-values were adjusted by Benjamini-Hochberg (BH) procedure and the signi cance threshold was set at an adjusted p-value < 0.05.

Overview of integrated AS events pro ling in TCGA ccRCC cohort
After a series of stringent ltering and screening, a total of 455 ccRCC patients from the TCGA with integrated mRNA splicing variants pro ling and counterpart clinical information were retrospectively analyzed in depth. The included cohort was comprised of 155 female (34.1%) and 300 male (65.9%) patients, among which 145 (31.9%) died and the median survival time was 79.5 months. The median follow-up time of these patients with ccRCC was 37.0 months (range, 1.3-112.6 months). For the entire cohort, the 1-, 3-, 5-year OS rates were 89.9%, 74.7% and 61.0%, respectively. Besides, we detected and quanti es the AS events of transcripts by using SpliceSeq software. According to the exclusive and distinct splicing pattern, these AS events can be roughly divided into seven particular types, including Exon Skip (ES), Alternate Terminator (AT), Alternate Promoter (AP), Retained Intron (RI), Mutually Exclusive Exons (ME), Alternate Donor site (AD) and Alternate Acceptor site (AA). As results, we obtained 32487 mRNA splicing events from 9286 parent genes, which contains 2634 AAs in 1964 genes, 2366 ADs in 1751 genes, 6155 ATs in 2806 genes, 173 MEs in 167 genes, 11371 ESs in 5230 genes, 2031 RIs in 1404 genes and 7757 APs in 3259 genes. And ES (35%), as the predominant type, accounts for more than one-third of all events, followed by AP and AT type in ccRCC and paracancerous tissues. We also noticed that one gene might possess two or more events. Considering this, Upset plot was generated to quantitatively visualize the intersecting sets among these seven AS types ( Supplementary Fig. S1). Intriguingly, the vast majority of splicing variants were from the same genes while one gene could have up to ve types of AS events, revealing the fact that AS hold largest potential on expanding transcript diversity and increasing proteome complexity.

Identi cation of aberrant cancer-speci c AS events in ccRCC
Considering the signi cantly spliced AS events between tumor and adjacent normal samples could play crucial roles in the tumorigenesis and development of ccRCC, we performed differential expression analyses with 72 normal tissues and 533 TCGA ccRCC tissues summarized. Using the preliminarily screening criteria mentioned above, we derived a list of 4660 aberrant splicing events, including 1069 ESs in 823 genes, 543 RIs in 438 genes, 1777 APs in 1470 genes, 719 ATs in 676 genes, 231 ADs in 204 genes, 304 AAs in 280 genes and 17 MEs in 17 genes. Interestingly, ES only contributes about 22.9% aberrant events even though it constitutes the highest proportion of AS events. As is shown in Fig. 1, most differential expressed AS events tend to be up-regulated in ccRCC tissues, suggesting the dysregulated splicing variants prefer to generate oncogenic transcripts rather than silence tumor suppressor transcripts.

Construction of AS panel and evaluation of its predictive ability in ccRCC cohort
Moreover, we attempted to assess the prognostic ability of these identi ed aberrant AS events. By using univariate Cox regression analyses, we identi ed a total of 2100 survival-associated AS events from 1666 parent genes (p-value < 0.05), including 1975 unfavorable prognostic splicing events (hazard ratio (HR) > 1, p-value < 0.05) and 125 favorable prognostic ones (HR < 1, p-value < 0.05). The exact number of aberrant events indicating poor prognosis far exceeded the number of those indicating good prognosis, which were consistent with the results from differential analysis and appear to con rm a major potential oncogenic role of aberrant AS in initiation and progression of ccRCC. To further quantify the interactions among these aberrant survival-associated events, we applied Upset plot and revealed that some genes clearly had up to ve types of AS events which were both cancer-speci c and prognosis-related ( Fig. 2A). For example, AD, AP, ES, RI and AA events of DMKN (red dotted line) were all signi cantly associated with patient OS in ccRCC. NDRG2 and POFUT2 (green dotted lines) could possess four types of prognostic AS events. In addition, we presented the top 20 most signi cantly survival-associated events in each AS type (only 5 ME events available) in Fig. 2B-H, respectively. Furthermore, these top most signi cant ones were chosen as candidates for predictive model construction. To discover the speci c AS events with the greatest prognostic value, we applied LASSOpenalized regression model with 10-fold cross validation to the candidate AS events in seven types, separately (Supplementary Figs. S2 and S3). In our data analysis, based on OS information, we have generated seven AS signatures, which were comprised of 4 AA events, 6 AD events, 10 AP events, 7 AT events, 11 ES events, 4 RI events and 3 ME events, respectively. Then patients were strati ed into low-risk and high-risk subgroups based on the corresponding risk score, utilizing the median cutoff calculated in the entire cohort, as demonstrated in Supplementary Fig. S4. The distribution of patients' survival time and status, heatmaps of selected splicing events were also displayed. The Kaplan-Meier survival curves also indicated that patients of high-risk were with signi cantly poor OS compared than those of low-risk ( Supplementary Fig. S5, p-value < 0.0001). Eventually, with the combination of all available types of AS events that formed signatures, we screened 17 particular AS events (see detailed information in Supplementary Figure S6) via LASSO-penalized Cox method ( Fig. 3A and Fig. 3B) and further constructed a nal AS-based panel by tting multivariate Cox regression model (Fig. 3C). In addition, the median risk score also served as the cutoff point for assigning ccRCC patients into low-and high-risk groups. As shown in Fig. 3D, the high-risk patients showed a 3.13-fold higher risk (95% con dence intervals (CI): 2.18-4.50, p-value < 0.0001) than low-risk patients. Encouragingly, time-dependent ROC curves ( Fig. 4A-C) also revealed that the AUC values of established AS panel were 0.73 at 1 year, 0.724 at 3 years and 0.75 at 5 years, con rming the predictive potential of our AS panel for both short-term and long-term prognosis with AUCs being robust above 0.7 across time (Fig. 4D). The detailed formula of each AS signature was listed in Supplementary Table S1.

Functional enrichment results of identi ed aberrant survival-associated AS events
It was evident that AS has a profound effect in altering transcript architecture and further modifying corresponding protein via deleting or adding functional domains. Thus, analyzing the in uencing of aberrant AS on their encoding proteins might give us a better understanding and direction of the roles of aberrant events in the splicing machinery of ccRCC. In our research, 1666 genes from 2100 both ccRCCspeci c and OS-related splicing events were sent for bioinformatics analysis, including GO and KEGG pathway enrichment analyses. As shown in Supplementary Table S2, a total of 48 GO-CC categories, 60 GO-BP and 25 GO-MF terms were identi ed as signi cantly enriched ones (FDR < 0.05), most of which were closely related to mitochondrial and energy metabolic activities, such as mitochondrial matrix (FDR < 0.0001), generation of precursor metabolites and energy (FDR = 0.003), and mitochondrial inner membrane (FDR = 0.002). Fig. 5 also shows that speci c biological processes closely associated with ccRCC, such as focal adhesion (FDR = 0.002), RNA splicing (FDR = 0.015), and GTPase regulatory activities, were signi cantly enriched in these parent genes occurred aberrant AS events. Besides, several KEGG pathways (Fig. 5D) that were experimentally validated participating in ccRCC progression or metastasis, including central carbon metabolism in cancer (FDR = 0.011), ubiquitin mediated proteolysis (FDR = 0.011), base excision repair (FDR = 0.043) and apoptosis (FDR = 0.016), were also revealed. Of note, both KEGG and GO enrichment analyses suggested the regulation of autophagy could be a potential mechanism of splicing regulatory in ccRCC. Taken together, above functional enrichment results not only provided novel clues for further exploring the underlying mechanisms upon aberrant prognosis-related events, but also con rmed the reliability of our screening procedure in turn.
3.5. The independent predictive power of nal AS panel for ccRCC patients We performed univariate followed by multivariate Cox hazard regression analyses of data in TCGA ccRCC cohort in order to further investigate whether the nal 17 AS based panel was an independent prognostic factor, where the AS panel was treated as a binary variable. The univariate analysis results suggested that age, Fuhrman grade, pathological stage, T stage, M stage and nal AS panel were all remarkably correlated with the OS of ccRCC patients (Fig. 6A). Therefore, those signi cant factor were included in a multivariate analysis, which showed that the AS panel (HR = 2.487; 95% CI: 1.717-3.605; p-value = 1.47e-06), age, Fuhrman grade, pathological stage were four independent prognostic factors when adjusted by those risk factors (Fig. 6B).
Moreover, in order to investigate the prognostic value of our AS panel in strati ed cohort, we classi ed patients into various subgroups based on age, gender, pathological stage, Fuhrman grade, T stage and M stage then we performed strati cation analysis. As shown in Fig. 7, the AS panel identi ed patients with distinct prognoses in all cohorts analyzed, thus con rming its robustness for independently predicting ccRCC prognosis.

Altered biological pathways between the high-and low-risk ccRCC patients
The strong risk strati cation ability of our 17 AS-based panel for ccRCC patients could be attributed to the oncogenic pathways and biological process that participate in ccRCC regulation during the tumor development or metastasis. Hence, we performed GSEA to elucidate potential biological functions of the nal AS panel (Fig. 3E and Fig. 3F). In the GSEA enrichment results, we observed that genes highly expressed in the high-risk group showed signi cantly enriched in "P53 signaling transduction" and "MAPK signaling pathway", which have been widely acknowledged being associated with the progression of ccRCC. Furthermore, several cancer-related pathways, such as the cell cycle regulation, response to DNA damage and neutrophil chemotaxis, were also up-regulated in high-risk patients, while the high-risk correlated genes were revealed to be associated with down-regulation of autophagy, which was in accordance with the KEGG enrichment results above. In summary, our GSEA analysis implied that the nal 17 AS-based signature might be involved in crucial ccRCC-related pathways and their functional dysregulation subtly in uences the OS of ccRCC patients.
3.7. Development and apparent performance of a nomogram based on the nal AS panel A second multivariate Cox regression model that incorporated the above independent predictors, including the nal AS panel, age, Furhman grade and pathological stage, was developed and presented as the nomogram (Fig. 8A). It substantiated that the AS panel comprised of 17 AS events contributes the most risk points (ranging from 0 to 100), whereas the other clinicopathological characteristics contribute much less. The C-index for OS prediction was 0.816 with 1000 bootstrap replicates (95% CI: 0.784-0.848), suggesting derived nomogram showed good predictive discrimination ability. And the calibration plot showed that the bias-corrected lines were close to the ideal line (45° line), indicating an optimal agreement between the survival prediction by our nomogram and actual observation (Fig. 8B). Moreover, we conducted DCA analysis to determine the real-world clinical usefulness of the AS-clinicopathological nomogram by quantifying the net bene ts against a range of threshold probabilities in ccRCC cohort. And the DCA results are presented in Fig. 8C, which showed that using our AS-based nomogram would add more bene ts in prognosis prediction than either treat-all-patients or treat-none scheme if the threshold probability of a patients or clinician is more than 10%. Within this range, the net bene ts were comparable, with several overlaps, based on the nal AS panel and the model with clinicopathological factors integrated only, further convincing the AS panel is non-inferior to clinical features. Besides, the clinical usefulness of the nomogram signi cantly overwhelmed the clinical risk factors. In a word, these ndings suggested that the derived nomogram was a better prognostic model for predicting short-term and long-term survival probability in ccRCC patients than individual clinical indicators.

The prognostic SFs and construction of splicing regulatory network
It has been proved that the global dysregulation of AS events could be orchestrated by a limited number of SFs, especially in ccRCC. The normalized level 3 RNA-seq pro les of TCGA ccRCC samples identi ed 53 SFs whose expression levels differed signi cantly between tumor tissues and adjacent normal tissues (Fig. 9A). According to the thresholds we set, 14 SFs, including 5 risky ones and 9 SFs indicating favorable prognosis, meeting the cancer-speci c and prognosis-related properties were further screened as candidates ( Fig. 9B and Supplementary Fig. S7). Next, to systematically analyze the cancer-speci c splicing regulatory connections between SFs and AS events in ccRCC, a splicing regulatory network (Fig.  9C), which enrolled the signi cant relationships (adjusted.p.value < 0.05) among these 14 SFs (blue and yellow dots) and 110 most signi cantly prognostic AS events comprised of 105 unfavorable ones (red dots) and 5 favorable ones (green dots), was nally built. From the regulatory network, we can see that the majority of genes could be synergistically or competitively regulated by different SFs via binding the same transcript region (AS event), further revealing the molecular complexity and tumor heterogeneity in ccRCC. Indeed, the entire SF-AS regulatory network was quite intricate, and altered SFs played vital driving roles in modifying pathological splicing events. Representative scatter plots presenting the signi cant correlations between expression levels of particular SFs and PSI scores of speci c AS events were displayed in Fig. 9D. For instance, the expression of ESRP2 was positively correlated with AD event of ZFP36, while the ES of SLC25A29 could be negatively regulated by SRP54.

Discussion
As changeable and possibly heritable post-transcriptional process, AS mechanisms offer promising clues for the therapeutic strategies in various disease, including many cancers [11]. To data, accumulated evidence indicates the AS process could make crucial contribution to the initiation, progression and metastasis of cancer, especially in renal cell carcinoma. For instance, DCLK1-ASVs, as an alternate splice variant of DCLK1, plays a stem cell supportive role in kidney cancer via driving self-renewal and drug resistance to chemotherapy [38]. VEGF is widely regarded the most potent growth factor for tumor neovasculature, while in-depth study has revealed that VEGF 165 b, a differential splicing isoform from exon 7 to 3' untranslated region, is down-regulated in renal cell carcinoma and contributes to antiangiogenesis [39]. Similarly, PKM2, rather than PKM1, has been reported to promote different aspects of ccRCC progression, including cell proliferation, invasion and migration in vitro [40]. Meanwhile, the involvement of SFs is also re ected in the investigation of aberrant AS regulation in ccRCC. The protumorigenic function of SF3B3 can be partly mediated by up-regulation of EZH2Δ14, since the EZH2Δ14 acts as a suppressor for the expression of EZH2, which is commonly recognized as a tumor-suppressive gene [41]. Accordingly, AS events have been considered as multifaceted carcinogenesis hallmark of ccRCC, convincing that splicing events could be preeminent biomarkers for further investigation. Even so, painstaking exploration of therapies concerning aberrant AS, including clinical trials, are nevertheless under way [14; 42]. To our best knowledge, our present study is a relatively comprehensive attempt to decipher the untapped mechanisms of AS in ccRCC so far.
The prevalence of ccRCC, accounting for approximately 75% of primary kidney cancer, is increasing annually with recent implementation of screening program. Despite the advance of surgical procedures, such as partial or total nephrectomy, more than 20% postoperative patients suffering from ccRCC will develop distant metastasis [3; 43]. What's worse, adjuvant systematic therapies, including chemotherapy, radiotherapy and immunotherapy, have failed to show an obvious survival bene t for postoperative patients, highlighting the drug resistance and nancial burden as problems worth pondering [1; 44]. In this aspect, novel molecular biomarkers that could reliably evaluate disease progression and patient prognosis would have tremendous potential in guiding therapeutic strategies and clinical management of ccRCC patients. Taking the potential in uence of aberrant splicing events in renal carcinogenesis into account, it is reasonable to consider the AS events in to increase the effectiveness of adjuvant therapy. In this study, using the well-established public, large-scale multi-omics sequencing cohort, we proposed a robust, individualized AS panel that can estimate OS probability in ccRCC patients based on 17 aberrant splicing events. Attributed to the combination of all the seven types of AS events, the AUC values of ROC curves for this model can keep above 0.7 spanning 1 year to 5 years, suggesting enhanced power and great potential in short-term and long-term prognosis prediction for ccRCC patients. Moreover, our prognostic AS panel could further stratify clinically de ned groups of patients (e.g. age, TNM stage and Fuhrman grade) into subgroups with distinct prognostic outcomes.
Actually, to illustrate the clinical value of AS events in cancer, several investigators have performed genome-wide analyses of splicing events in non-small cell lung cancer, gastrointestinal adenocarcinoma and ovarian cancer [45; 46; 47]. With the rapid development of high-throughput sequencing technologies over the last decades, preliminary success has also been gained in insight into the involvement of splicing patterns in ccRCC. Recently, by integrating whole-exome sequencing, RNA-seq and copy number variation analyses, Lehmann et al analyzed 282 ccRCC patients and revealed that 915 probable splicing quantitative trait loci with tumor-speci c splicing pattern were identi ed, and that are involved in processes relevant for tumor growth and cancer progression [48]. However, few studies have broken the gap between aberrant AS events and clinical meaning in a systematic way. And it still remains to be unsolved problem that what extent alteration of AS pattern could change the clinical outcome among ccRCC patients. Despite a series of strict criteria were conducted in the process of sample screening and biomarker ltration, more AS events, consisting of a total of 2100 aberrant splicing events occurred in 1666 parent genes, were nally identi ed in present study, meeting the cancer-speci c and prognosisrelated properties. Moreover, our exploration of AS patterns broadens the understanding of traditionally conceptual prognostic biomarker, which serves as a powerful complementary to verify the link between AS and ccRCC.
Furthermore, the functional enrichment analyses, including GO, KEGG and GSEA analysis, provided more opportunities for deciphering the largely untapped mechanisms which our identi ed AS events might participate in. Intriguingly, there were several intersections among our enrichment results and preliminary investigation. The majority of enriched biological behaviors were metabolism-related and the autophagyrelated pathways were revealed as the signi cant overlap among GO, KEGG and GSEA results. Autophagy, as an intracellular self-degradative process for capturing and recycling degraded components with the assistance of autophagosome and lysosome, plays crucial roles in maintaining metabolism and homeostasis [49]. More importantly, the modulation of autophagy involves in dual-sided roles within cancer microenvironment, which could facilitate tumorigenesis via promoting proliferation and mediate tumor suppression via inducing cell death and apoptosis [50]. invasion [53].
With the advance of the times and the increasing demand for precision medicine, the limitations of traditional predictive factors have gradually emerged. They may be insu cient for proper tumor characterization and classi cation, leading to di culties for the prediction of disease progression, therapeutic response and prognosis. Nowadays, powerful and advanced technologies including highthroughput, genome-wide pro ling methodologies has led to signi cant progress in the biological understanding and prognosis of ccRCC [54]. Here, we demonstrated that our AS panel could serve as an independent prognostic factor for OS prediction in ccRCC patients after the modi cation of traditional risk factors. Besides, we further leveraged the complementary value of AS panel and other independent clinic pathological features (age, Furhrman grade and pathologic stage). Consequently, our nomogram could be a valuable novel prognostic scoring system for clinical decision-making of practitioners.
Even though current comprehensive pro ling of aberrant prognostic AS signatures provided us an ASclinicopathologic nomogram, which represented as the robust and reliable tool for prognosis prediction in ccRCC, several limitations were still required to be clari ed. First, a major limitation of our work is due to