Development and validation of a kidney renal clear cell carcinoma prognostic model relying on pyroptosis-related LncRNAs-A multidimensional comprehensive bioinformatics exploration

We used API v3.0.0 to download KIRC patient TCGA database mutation information, corresponding clinical data, and transcriptome RNA-seq data available at https://​portal.​gdc.​cancer.​gov (Release date: October 29, 2021). We used expression profile information from 91 RCC samples in the ICGC database (https://​dcc.​icgc.​org/​) as our validation cohort.

According to Additional file 1: Table S1, 52 genes relevant to pyroptosis were gathered from earlier research and papers. Using a screening process based on gene annotation, 2876 lncRNAs were ultimately found in the TCGA cohort. The association study between 52 genes and lncRNAs associated with pyroptosis was conducted using the Pearson correlation coefficient. We identified lncRNAs associated with pyroptosis based on their absolute correlation coefficient (> 0.4) and P-value (< 0.001). As a consequence, 576 lncRNAs associated with pyroptosis were examined. After that, we performed univariate Cox regression analysis for OS, choosing lncRNAs with and without value in the process. P-values < 0.05 were used to determine whether lncRNAs are related to prognosis. Overall, 295 lncRNAs of prognostic relevance were found to be linked to pyroptosis.

The training set served as the basis for construction of the prognostic model, and the test set and the whole TCGA set were used to evaluate the model's prediction performance. The training set was analyzed using multivariate Cox regression, revealing six pyroptosis-related lncRNAs. The model, named "PLnRM", uses the following formula to calculate the risk score: risk score = coef (lncRNA1) × expr (lncRNA1) + coef (lncRNA2) × expr (lncRNA2) + …… + coef (lncRNAn) × expr (lncRNAn). In this equation, "coef" represents coefficients. Specifically, "coef (lncRNAn)" refers to the survival-related coefficient of lncRNAs, and "expr (lncRNAn)" refers to expression of lncRNAs.

We evaluated and calculated abrupt change data using maftools of the R package, and we estimated the tumour mutation burden (TMB) by tumour-specific mutation genes. In addition, we used tumour immune dysfunction and exclusion (TIDE) computation to estimate immunotherapy response probability [27].

Six pyrogenic lncRNAs, as well as genes and lncRNAs associated with pyroptosis, were studied by examining expression patterns [28]. In addition, differences in OS between the two models were assessed using Kaplan‒Meier survival analysis.

Using the GDSC website, we were able to identify the semimaximum inhibitory concentration (IC50) of medications to locate therapeutic chemicals for KIRC patients and enhance clinical effectiveness. To calculate the IC50 for KIRC patients, we employed a pRophetic formula found in the R package.

Cox regression models, including single- and multiple-variable models, were utilized to show that individuals with KIRC do not have a prognostic pattern that is distinct from or unrelated to other clinical characteristics [29].

We created a graph throughout the procedure to predict OS at 1, 3, and 5 years for KIRC patients to improve practicability and make the findings clear. We use the Hosmer‒Lemeshow test to demonstrate the consistency of the corrective curve.

During signalling pathway analysis, we first analyzed differential expression of gene samples with R's limma package to assess samples with high (n = 285) and low (n = 245) PLnRM scores. Enrichment analysis was conducted to identify signalling pathways in which differentially expressed genes are involved using the gene set enrichment analysis (GSEA) approach. The KEGG and HALLMARK gene sets were used, along with the cluster profile package of R. A significance threshold of P < 0.05 and FDR < 0.25 was applied. For a few common gene sets, we next conducted single-sample GSEA (ssGSEA). Differences in patient survival were then explored using Kaplan‒Meier survival curves. The cBioPortal database provides information on gene changes for gene mutation research. Using the R package Maftools, we examined the quantity and quality of mutations in two PLnRM risk groups. Correlation analysis was performed on expression of PD1, PDL1, CTLA4, BTLA, and CD24 between the two PLnRM risk groups. We imported expression data from 530 KIRC samples into CIBERSORT (https://​cibersort.​stanford.​edu/​). After 1000 iterations, we estimated the relative percentage of 22 categories of immune cells to identify their immunological features. Additionally, the matching proportions of 22 different kinds of immune cells as well as clinicopathological variables were compared between the two PLnRM risk groups, and the findings are shown in a landscape map. ssGSEA was conducted on specific gene signatures to better elucidate the immunological and molecular functions between the two PLnRM risk groups, and the resulting scores were also compared [30‐33].

Figure 1 illustrates the precise procedure for building the risk model and subsequent analyses. The TCGA database was used to derive matrix expression of 52 pyroptosis genes and 2,876 lncRNAs. We classified pyroptosis-related lncRNAs as those strongly linked to at least one of the 52 pyroptosis genes (|Pearson R|> 0.4 and P < 0.001). The pyroptosis-lncRNA coexpression network is depicted using a Sankey diagram (Fig. 2A). The results revealed 576 lncRNAs to be associated with pyroptosis. Figure 2B shows the relationship between pyroptosis genes and pyroptosis-related lncRNAs in the whole TCGA dataset.

We conducted univariate Cox regression analysis on 2876 pyroptosis-related lncRNAs in the KIRC dataset included in the TCGA database to identify prognostic lncRNAs. The TCGA database contains 295 pyroptosis-related lncRNAs with a strong correlation with the OS of KIRC patients (Fig. 3A). Multiple regression analysis often uses LASSO-penalized Cox analysis. It may concurrently perform selection and regularization of variables in addition to improving the statistical model's capacity for and accuracy at making predictions. This technique has been used widely to choose the best features in data in high dimensions with little connection and a strong projected value to prevent overfitting. As a result, this technique can efficiently identify the best predictive markers and provide a prognostic indicator to predict clinical outcomes. The value of log l that ranks first is shown on the dashed perpendicular line with the least bias in segment probability. Hence, Fig. 3B and C shows that 10 lncRNAs associated with pyroptosis were chosen for further multivariate analysis. Based on this, autocephalous prognostic proteins were identified using multivariate Cox ratio hazard regression analysis. To create a risk model for assessing the prognosis of KIRC patients, we utilized 6 pyroptosis-related lncRNAs (listed in Additional file 2: Table S2) that were independently associated with OS in the training set (as shown in Additional file 3: Table S3).

We conducted Kaplan‒Meier survival analysis on KIRC samples, dividing them into low-risk and high-risk groups using median prognostic risk scores. Figure 3D–F displays the survival status of patients in both groups across the full TCGA set, training set, and test set. The results showed a significant (P < 0.001) difference in outcomes between the high-risk and low-risk groups. Figure 4 A1 and 4A2 displays the patient survival status, survival time, and risk level distribution for both groups. The six lncRNAs associated with pyroptosis and their relative expression standards in each patient are shown in Fig. 4A3.

We used the same formula to calculate each patient's risk score in both the training and test sets. This allowed us to assess how effective our model is at predicting prognosis. Figures 4B, C depict expression of pyroptosis-related lncRNAs in the training set (Fig. 4B1–B3) and test set (Fig. 4 C1–C3). Additionally, they illustrate the distribution of risk grades, survival status patterns, and survival times (Fig. 4C1–C3).

To confirm the precision and applicability of the model, we verified expression profile data for 91 RCCs in the ICGC database. The results demonstrate the model's continued effectiveness in predicting survival time, particularly long-term survival (Fig. 5).

We compared differences in OS between low- and high-risk groups across the entire TCGA dataset, stratifying by common clinicopathologic characteristics. The low-risk group's OS remained superior to that of the high-risk group across subgroups categorized by age, sex, stage, and grade (see Additional file 4: Figure S1).

Comprehensive gene expression profiles, 52 pyroptosis genes, lncRNAs related to pyroptosis, and a risk model separated by lncRNA expression profiles all contributed to this model, which was then subjected to principal component analysis (PCA) to validate the distinction between low-risk and high-risk groups (Fig. 6A–D). As shown in Fig. 6A–C, there was absolutely no uniformity in the distribution of the high-risk and low-risk groups. Nonetheless, the findings from this model showed that the two groups’ distributions were significantly different (Fig. 6D). Based on these findings, it seems that there is a difference in the prognostic signature between the two groups.

PLnRM was utilized to analyze the enrichment and activity levels of various immune cells, pathways, and activities in 530 KIRC patients. The findings showed that most immunological indicator expression levels between the two groups were significantly different (Fig. 7B). A GO enrichment study was performed to explore the putative molecular mechanisms related to PLnRM. Regarding BP, the model genes are related to a variety of biological functions relevant to the immune system (Fig. 7A). The relationship between PLnRM and the effectiveness of immunotherapy was, therefore, investigated. PLnRM can predict TIDE by indicating that the high-risk group will be more responsive to immunotherapies than the low-risk group, as hypothesized (Fig. 7H). In addition, maftools of the R package were used to investigate and assemble the data associated with mutations, and the consequences of mutation were used for stratification. Figure 7C and D displays the top 20 driver genes that show the greatest frequency change across all categories. TMB scores were then computed on the basis of the TGCA somatic mutation data. The findings indicate no significant difference in tumour mutation between the high-risk and low-risk groups, as shown in Fig. 7E. Kaplan‒Meier survival analysis of TMB was conducted using tumour tissues. The results (Fig. 7F) indicated that the low-mutation group had a higher chance of surviving than the high-mutation group. Furthermore, it was found that the prognosis for those in the high-mutation and high-risk groups was worse than that for those in the low-mutation and low-risk groups. Even when there were two groups with either a high or low mutation risk (Fig. 7G), individuals in the high-risk group still had a poorer prognosis than those in the low-risk group. This risk model seems to be reliable and stable based on these findings and our earlier findings.

The Wilcoxon test was used to examine the immune cell distribution in various PLnRM subgroups with the intention of examining the immune cell composition in various PLnRM subgroups. According to the results, the group with PLnRM-high risk had a higher prevalence of regulatory T cells (Tregs), activated memory CD4 T cells, CD8 T cells, memory B cells (MBCs), and macrophages (M0). Resting dendritic cells, macrophages (M2), resting NK cells, gamma delta (γδ) T cells, T follicular helper (Tfh) cells, resting memory CD4 T cells, plasma cells, naïve B cells, and activated dendritic cells were more prevalent in the PLnRM-low risk group, as shown in Fig. 8A, B. Figure 8C displays traits connected to the immunological landscape.

The gene sets enriched in various PLnRM subgroups were determined using GSEA. The P53 signalling pathway, ribosome, cytokine receptor interaction, systemic lupus erythematosus, and taste transduction pathways were enriched in the gene sets for samples with significant levels of PLnRM (Fig. 8D). However, according to Fig. 8E, the PLnRM-low samples showed gene sets that were enriched in pathways related to tumour metastasis (P < 0.05, FDR < 0.25). To further investigate how the two groups' responses to immunotherapy differed, variations in immune checkpoint expression and correlation were evaluated. As shown in Fig. 8F–J, CD47, B- and T-lymphocyte attenuator (BTLA), cytotoxic T lymphocyte-associated antigen (CTLA), PD ligand 1 (PD-L1), and programmed cell death (PD-1) were all higher in the high-risk group.

This investigation utilized univariate and multivariate Cox regression analyses to determine whether the risk model incorporates independent prognostic factors for KIRC. Multivariate Cox regression analysis showed a hazard ratio (HR) of 1.067 and a 95% confidence interval (CI) of 1.050–1.085, with a significance level of P < 0.001 (Fig. 9A). In univariate Cox regression analysis, the odds ratios (HRs) were found to be 1.034, with a corresponding CI of 1.014–1.054, and significant results at P < 0.001 (Fig. 9B). The findings support the notion that the risk model can predict prognosis regardless of other clinical variables. We evaluated the distinctiveness and sensitivity of risk scores in predicting KIRC patient outcomes using the concordance index and area under the receiver operating characteristic (ROC) curve (AUC), as shown in Fig. 9D, E. Longer follow-up periods resulted in a greater concordance index of risk ratings, which ultimately trumped other clinical criteria and increased risk severity. Thus, the prognosis of KIRC patients may be accurately predicted using this model's risk score (Fig. 9C). Moreover, the AUC representing the risk level rose to a point where it exceeded that of the vast majority of other clinicopathological characteristics. Our results tended to support the idea that PLnRM is a viable option for incorporation into the predictive risk model that is currently being used for KIRC patients (Fig. 9C).

The 1-, 2-, and 3-year OS of KIRC patients can be predicted using a nomogram that takes into account both risk levels and clinical risk factors. The nomogram revealed that the prediction model's risk level has high predictive power when compared with clinical parameters (Fig. 10A). The 1-, 2-, and 3-year OS observation and prediction rates were all shown to be in good agreement by means of appropriate diagrams (Fig. 10B).

We used the prediction approach to evaluate treatment response for KIRC patients by analyzing the IC50 of each sample in Genomics of Drug Sensitivity in Cancer (GDSC). Our goal was to identify potential medications for PLnRM. There were 110 discovered compounds in all, and the predicted IC50 values between the two groups varied significantly. Partially sensitive chemicals are shown in Additional file 5: Figure S2.

Several possible therapeutic medications for KIRC were evaluated using the connection map (CMAP) database to elucidate a new therapy plan for the disease. The top 10 closely related molecules of medications were chosen to depict the three-dimensional structure (Additional file 6: Figure S3).