The link between immune- and stromal-scores and the rate of Prognosis in LUAD Patients
The transcriptome outcomes of 490 LUAD cases were taken from the TCGA databank, and the ESTIMATE algorithm was employed to identify their scores (immune/stromal). Then we separated the patients with LUAD into higher and lower score groups (stromal/immune) and examined the variations in prognosis between the two groups using the mean of stromal and immunological scores. Based on the data obtained from K-M curves, lower scores (immune and stromal) were substantially linked with the lower OS in cases with LUAD (Fig. 1A,1B). Although no considerable link was detected between immunological and stromal scores with PFS in patients with LUAD (Supplement Fig. 1A, 1B). In addition, the differences in TNM staging between the high and low groups were also compared, and it was found that the T and M stages were statistically significant between the high and low immune groups (Fig. 1C,1D), and the M staging was statistically significant between the high and low stromal groups (Fig. 1E).
DEAS Events between Higher and Lower Score Groups (Immune/stromal)
RNA-Seq data was utilized to create integrated AS event profiling, which led to 43948 AS events being identified. To guarantee high stringency, the AS events were evaluated using several filters (Standard Deviation ≥ 0.01, average PSI value ≥ 0.05). As a result, 30569 AS events were discovered. The AS events in LUAD patients were then sorted, and it has been revealed that ES (Exon Skip) was the most common AS event, succeeded by AT (Alternate Terminator), and AP (alternate promoter). The intersections between AS events and the accompanying gene intersections were visualized using the UpSet plot (Fig. 2A). The DEAS events were then evaluated by the comparison of higher and lower score groups (immune or stromal), which were presented using heatmaps (Fig. 2B,2D) and volcano plots (Fig. 2C,2E). Thus, in the immune groups, 147 DEAS events were highly expressed while the expression of 131 DEAS events was decreased (Supplement Table 1). In the stromal groups, 199 DEAS events were highly expressed while the expression of 179 DEAS events was decreased in the underlined group (Supplement Table 2). The UpSet plots of DEAS events and the related gene intersections of the immune as well as stromal groups are depicted in Fig. 2F,2G, respectively. The Venn diagram program was employed to evaluate 96 DEAS events that were upregulated and 71 DEAS events that were downregulated (Fig. 3A,3B). The upset plot corresponding to the intersection DEAS events was also shown in the Supplement Fig. 1C.
The biological events and pathways were then evaluated by the GO and KEGG pathway analysis on the parent genes of DEAS events (Fig. 3C,3D). In the BP category, the most enriched pathways were immune response-activating cell surface receptor cascade of signaling and immune response-regulating cell surface receptor cascade of signaling; in the CC category, the most enriched GO terms were adherens junction and cell leading edge; in the MF category, cell adhesion molecule binding and Ras guanyl-nucleotide binding were the primary functions of these genes. For the KEGG pathway enrichment study, the major pathways were Axon guidance and T cell receptor signaling. The enrichment analysis results show that the parental genes of these AS events are closely related to the immune response.
AS-based clusters are considerably linked with prognosis, and also molecular, and immune characteristics
The AS profiling revealed that the AS events among LUAD patients were incredibly heterogeneous. This finding makes us wonder whether we can predict the clinical outcome of each patient through the analysis of changes in the expression of AS events. Furthermore, unsupervised consensus analysis was carried out to explore whether AS had any discernible patterns (Fig. 4A). We found three distinct AS-based molecular clusters based on the consensus matrix heatmap: C1 (n = 128, 26.3%), C2 (n = 174, 35.7%), and C3 (n = 185, 38.0%). To better understand the clinical effects of the evaluated AS clusters, we looked at the correlations between clusters and clinicopathological features, as depicted in Fig. 4B. The stage of TNM and survival status (OS and DFS) were not distributed in a random manner across clusters. Furthermore, the link of K–M analysis of cluster with prognosis yielded distinct survival patterns. Notably, C3 was linked to a positive OS, whereas C1 and C2 were related to a negative OS (Fig. 4C). DFS, on the other hand, did not differ considerably among the clusters (Supplement Fig. 1D).
In addition, differences in the immunological microenvironment amongst AS clusters were investigated. The scores (immune and stromal) were determined using the ESTIMATE method to measure the existence of immune and stromal cell infiltration in samples of the tumor. We found that the immune and stromal score were all highest in C3 (Fig. 4D,4E). Furthermore, using ssGSEA, a heatmap was created to examine the relative abundances of the 23 immune infiltrating cell subpopulations among different AS clusters (Fig. 4F). Higher scores (immune and stromal) were related to the C3 cluster, while lower scores (immune and stromal) were connected with the C1 and C2 clusters. The biological behaviors among these different AS patterns were then investigated using a GSVA enrichment analysis. B cell receptor signaling pathway, cytokine-cytokine receptor interaction, natural killer cell mediated cytotoxicity, leukocyte transendothelial migration, T cell receptor signaling pathway and Toll like receptor signaling pathways were among the enrichment cascades presented by the C2 and C3, as depicted in Fig. 4G,4H. The results show that patients with LUAD have a better prognosis when their immunological and stromal scores are greater.
Establishment and assessment of the prognostic signature for LUAD patients
Biomarkers for early illness identification and potential therapeutic targets are still a hot topic in medicine. Previous research has found that abnormal AS events in the initial phases of cancer and are used as prognostic markers in a variety of cancers (33, 34). We generated signatures based on the DEASs to extract the underlying predictive value of individual DEASs, followed by a univariate CRA. The findings demonstrated that in LUAD patients, 39 and 12 intersecting AS events were strongly linked with OS and PFS, accordingly (fig 5A,6A). To avoid model overfitting, the LASSO regression was adopted to select the optimal OS-(fig 5B,5C) and PFS-(fig 6B,6C) related DEASs to construct the prediction models. After performing multivariate CRA, 11 DEASs were used to create two prognostic signatures, i.e., 5 DEASs for the OS signature, 6 DEASs for the PFS signature, and one overlapping DEAS. We estimated each LUAD patient's risk score using the formula in the methods, and on the basis of the median of the risk score, we separated the cases into two groups: HRG and LRG. The AUCs of the OS signature to predict 1, 2, and 3 years OS were 0.709, 0.656, and 0.669, accordingly, according to time-dependent ROC curves (fig 5D). Patients in the HRG had a lower OS than those in the LRG, according to the K-M curves (fig 5E).
The PFS signature's time-dependent ROC curves were also developed in the same way. A median was used to determine the risk score cutoff, and 208 and 209 patients were categorized into two groups: HRG and LRG, respectively. The signature had AUCs of 0.723, 0.688, and 0.702 for predicting PFS after 1, 2, and 3 years, accordingly (fig 6D). The PFS was better in the patients with LR scores (fig 6E). The underlined data revealed that both signatures were considered to predict the survival rate of LUAD patients. Additionally, heatmaps (fig 5F,6F), survival status plots (fig 5G,6G), and risk score plots (fig 5H,6H) were created to clearly illustrate variations in prognosis and AS patterns.
Construction of nomograms based on DEAS signature as well as clinical parameters.
To enhance the clinical use of the underlined prognostic signatures, two complete nomograms integrating independent clinical characteristics were created. To investigate independent OS and PFS prognostic variables, we first used univariate and multivariate CRA. The obtained data revealed that Risk scores, T, and N were three independent variables associated with OS and PFS (Supplement Table 3,4). These findings showed that both DEASs-based signatures might be employed to predict the prognosis of LUAD cases independently.
Following that, on the basis of the independent prognostic factors, two novel nomograms were created for predicting OS (fig 7A) and PFS (fig 7E). The C-indices for the OS and PFS nomograms were 0.704 (95% CI 0.661–0.747) and 0.69 (95% CI 0.647–0.733), accordingly. The nomograms' significant calibration suggested that the outcomes predicted by the nomograms showed consistency with the observed data (fig 7B-D.7F-H).
Potential Regulatory Network Between SFs and AS Events
SFs are protein factors involved in the splicing process of RNA precursors and are closely related to AS events. We retrieved 71 SFs from the SpliceAid2 databank to investigate the fundamental regulatory network between SFs and AS events in LUAD cases. We identified 42 SFs (blue) that were substantially connected to 39 AS events linked with survival, comprised of 9 favorable AS events (green) and 30 adverse AS events (red) with OS signature using Spearman correlation analyses (fig 8A). The PFS signature was screened out of 35 SFs (blue) that were substantially connected to 11 survival-associated AS events, with 6 adverse AS events (red) and 5 favorable AS events (green) (fig 8B). Many SFs were found to be associated with numerous AS events and to play opposing roles in regulating various AS events. Different SFs may also regulate a certain AS event. The underlined phenomenon elucidates that a single transcript can result in many splicing events.
Analyzes of SOD2-78301-AT Regulating Relationship
In the multivariate CRA, SOD2-78301-AT was evaluated to be an independent OS- and PFS-associated AS and was added into both signatures as a DEAS. The expression of SOD2-78301-A T and the prognosis of LUAD patients were also investigated. A low level of SOD2-78301-AT was associated with considerably superior OS (fig 9A) and PFS (fig 9B), according to K-M survival curves. The link between SOD2|78301|AT and its only SF (ESRP2) was also investigated, and the findings verified their negative regulatory interaction (fig 9C). Finally, in LUAD, we investigated the link between SOD2|78301|AT's parental gene SOD2 and immune cells. There was a statistically considerable association between SOD2 expression in LUAD and the presence of immunological infiltrates (CD4 + T cells, B cells, macrophages, CD8 + T cells, neutrophils, and dendritic cells) (P < 0.05) (fig 9D). GISTIC 2.0 defines somatic copy number alterations (SCANs) as arm-level deletion, deep deletion, arm-level gain, diploid/normal, and strong amplification. In LUAD, box plots were used to indicate the distributions of all immune subcategories at each copy number status with SOD2 (fig 9E).