Identification of Prognostic Pyroptosis-Related lncRNAs in LGG Patients
As the methodology is shown in Figure 1A, in this study, we initially matched the ENSEMBL IDs with the lncRNA annotation file, identifying 12,083 lncRNAs in the TCGA and 12,091 lncRNAs in the CGGA. Additionally, we compiled a pyroptosis-related gene set comprising 45 genes from the Molecular Signatures Database. A lncRNA was classified as pyroptosis-related if its expression showed a notable correlation with one or more pyroptosis-related genes (R > 0.4 or R < -0.4, P < 0.05). In the TCGA cohort, 396 lncRNAs were significantly correlated with pyroptosis-related genes. Utilizing univariate Cox regression analysis for prognosis, we selected 48 lncRNAs from this group. Similarly, in the CGGA dataset, we identified 73 lncRNAs following the same criteria. Eventually, 18 lncRNAs common to both datasets were recognized as key pyroptosis-related lncRNAs. Figure 1B illustrates the correlations between these 18 lncRNAs and the pyroptosis-related genes in the TCGA.
Construction of the PLPS
Here, the PLPS was established to predict OS in LGG patients. This was achieved by incorporating 18 pyroptosis-related prognostic lncRNAs into a LASSO regression model, leading to the selection of 8 pivotal lncRNAs for the PLPS construction. These lncRNAs included AL359643.3, AC025171.3, AC010319.4, CYTOR, NEAT1, LINC02381, LIN00641, and LINC00672 (Figure 2A-B). The coefficient values related to this analysis are displayed in Figure 2C. Further analysis entailed categorizing LGG patients into high and low-expression groups. This division was based on the median expression values of the 8 identified lncRNAs within the TCGA. Survival analysis presented that high AL359643.3, AC025171.3, AC010319.4, CYTOR, NEAT1, and LINC02381 expression levels were meaningfully associated with shorter OS (Figure 2D-I). In contrast, high expression of LIN00641 and LINC00672 was linked to a better prognosis (Figure 2J-K). These findings were corroborated in the CGGA cohort (Figure S1). The PLPS for each LGG patient was then calculated according to the coefficients and expression levels of these 8 lncRNAs in the TCGA cohort. The results provide a nuanced understanding of how individual lncRNAs within the PLPS contribute to patient prognosis. The distinct survival outcomes associated with high expression levels of specific lncRNAs highlight their potential as biomarkers for stratifying patients into different risk categories, thereby aiding in personalized treatment planning and prognosis estimation in LGG.
The association between PLPS and clinicopathological features in LGG patients
Patients were separated into a high-risk group (n=238 in TCGA, n=80 in CGGA) and a low-risk (n=238 in TCGA, n=81 in CGGA) group with the median risk value. To identify the difference between two groups, we performed PCA based on the genome expression data in both TCGA and CGGA datasets. The analysis revealed that despite some overlap, patients classified as high or low risk tended to cluster in separate directions (Figure 3A-B). Furthermore, the heat map with lncRNAs expression and other clinical features (MGMT status, 1p/19q codeletion, IDH1 status, age, gender, subtype) of the TCGA dataset revealed the expression of AL359643.3, AC025171.3, AC010319.4, CYTOR, NEAT1, LINC02381 upregulated with increasing risk score while the expression of LIN00641, LINC00672 downregulated with increasing risk score. In addition, we found that MGMT unmethylated status, 1p/19q non-codeletion, IDH1 wildtype, advanced age, mesenchymal subtype, and grade 3 patients were greatly enriched in the high-risk class by using the chi-square test (Figure 3C). Additionally, the association between the PLPS and various clinicopathological factors was examined using a t-test. The findings indicated that the risk score was notably higher in patients aged over 45, those with WHO grade 3 gliomas, IDH1 wildtype, MGMT promoter methylation, 1p19q non-codeletion, and in those with gliomas of the mesenchymal and classical subtypes (Figure 3D-I). In contrast, the risk score was unrelated to gender (Figure S2). Similarly, the risk score showed consistent trends in the CCGA dataset, but no significant elevations were observed in MGMT unmethylated and classical subtype gliomas (Figure 3J-O). This comprehensive analysis underscores the potential of PLPS as a tool for assessing the prognosis of LGG patients, correlating genomic data with clinicopathological features to facilitate more tailored therapeutic approaches.
Prognostic Validity of the PLPS for LGG
The prognostic significance of the PLPS was additionally assessed by employing the log-rank test and Kaplan-Meier analysis in both the TCGA and CGGA datasets, comparing the low-risk and high-risk groups. The findings revealed that LGG patients with lower risk scores had notably better prognoses than those with higher risk scores (Figure 4A-B, P<0.05). The distribution of risk scores and survival status is illustrated in Figures 4C-D, indicating a concentration of living patients in the low-risk group. Furthermore, the ROC curve analysis demonstrated that PLPS had a significant potential to predict overall survival in both the TCGA cohort (1-year AUC = 0.846, 2-year AUC = 0.841, 3-year OS = 0.769, Figure 4E) and CGGA cohort (1-year AUC = 0.811, 2-year AUC = 0.828, 3-year OS = 0.844, Figure 4F). Stratification analysis within the TCGA dataset revealed that PLPS maintained its predictive ability across various subgroups. For grade 2 and 3 gliomas, a higher risk score was correlated with a poorer prognosis (Figure 4G-H, P<0.05). When dividing patients into younger (age <45 years) and older (age ≥45 years) groups, the prognostic value of the risk score remained consistent (Figure 4I-J, P<0.05). Classification based on three important molecular markers – IDH1 mutation, MGMT promoter status, and 1p/19q codeletion – showed that a lower risk score correlated with longer OS in all subgroups except for the 1p/19q non-codeletion cohort, where the trend was similar despite the P value being 0.062 (Figure 4K-P). In patients who received radiotherapy and chemotherapy, the high-risk group exhibited reduced OS compared to the low-risk group (Figure 4Q-R), a finding consistent with results from the CGGA dataset (Figure S3). These findings suggest that PLPS is a robust tool for accurately identifying LGG patients with unfavorable prognoses, regardless of their clinical, pathological, molecular, and treatment characteristics.
PLPS was an independent prognostic indicator for LGG patients
To determine whether the PLPS acts as an independent prognostic factor for LGG patients, univariate and multivariate Cox regression analyses were performed. In the TCGA dataset, the univariate Cox analysis revealed significant associations of age, tumor grade, radiotherapy, IDH1 mutation, and MGMT promoter status with OS. The multivariate Cox regression further confirmed that a high-risk score (multivariate: HR: 1.702, CI: 1.117–2.594, P=0.013) independently predicted poorer prognosis in LGG patients (Figure 5A). This finding was corroborated by the CGGA dataset, which also identified PLPS as an independent risk factor for OS in LGGs (multivariate: HR: 1.713, CI: 1.275–2.301, P<0.001, Figure 5B). Time-dependent ROC curves were then employed to compare the prognostic predictive ability of PLPS against other independent predictors such as age, WHO grade, and IDH1 status within both TCGA and CGGA. The 1-, 2-, and 3-year ROC curves indicated that the risk score based on pyroptosis-related lncRNAs had higher prediction accuracy than age, WHO grade, and IDH1 status (Figure 5C-H). These results suggest that PLPS is an independent indicator and could be valuable in clinical prognosis evaluation of LGG patients.
Correlation of the PLPS With the Immune Landscape of LGG Microenvironment
To elucidate the biological functions and signaling pathways associated with PLPS in LGG, GO analysis was conducted using DAVID online tools, focusing on the top 400 genes most correlated with PLPS in the TCGA dataset. The analysis revealed that these genes predominantly participate in immune-related biological processes, like immune response, interferon-gamma-mediated signaling pathway, antigen processing and presentation (Figure 6A). Similarly, GSEA indicated crucial enrichment of immune-related biological processes in the high-risk group, including activation of immune response, T cell receptor signaling pathway, type I interferon production, and B cell activation (Figure 6B). Further exploration of the correlation between PLPS and the immune landscape of the LGG microenvironment was conducted using the TCGA dataset. The risk score proved a significant positive correlation with the immune score (R=0.553, P<0.001), stromal score (R=0.603, P<0.001), ESTIMATE score (R=0.592, P<0.001), and a negative correlation with tumor purity (R=-0.605, P<0.001) (Figures 6C-F). The analysis of immune cell infiltration, conducted through ssGSEA and TIMER algorithms, indicated a significant enrichment of immune cell types, including macrophages, activated CD4 and CD8 T cells, myeloid dendritic cells, and activated B cells, within the high-risk group (Figures 6G-H). This group also said great correlations with most immune-related functions (P<0.001, Figure 6I). The expression of several immune checkpoints, including CD274 (PD-L1), CD80, CD44, CD48, CTLA4, LAG3, PDCD1, NRP1, CD276, and BTLA, was notably higher in the high-risk group compared to the low-risk group in the TCGA (P<0.001, Figure 6J). The aforementioned findings were further validated in the CGGA dataset (Figure S4). Overall, these results indicate a strong association between PLPS and immune infiltration, with the high-risk group demonstrating highly activated immune characteristics. This could potentially make PLPS a useful biomarker for predicting the response to immune checkpoint inhibitor therapies.
Genomic profiles in different PLPS groups of LGG patients
To delve deeper into the molecular differences between high-risk and low-risk LGG groups, analyses of somatic mutations, copy number alterations (CNA), and TMB were conducted using the TCGA database. The analysis began with identifying the top 20 genes with the highest mutation rates in LGG patients. It was found that IDH1 mutations were the most frequent in both high-risk and low-risk groups. However, IDH1, CIC, and NOTCH1 mutations were significantly more prevalent in low-risk patients, whereas TP53, ATRX, TTN, and EGFR mutations were observed more frequently in high-risk gliomas (Figure 7A). The study then explored somatic CNAs, revealing different chromosomal alteration patterns between low and high-risk LGGs. Amplification of Chr 7 and deletion of Chr 10 were notably more common in high-risk LGGs. Conversely, the incidence of 1p/19q codeletion, a hallmark of oligodendroglioma, decreased with higher risk scores (Figure 7B). GISTIC 2.0 analysis, comparing the lower and upper quartile groups, identified focal amplifications and deletions. High-risk cases showed focal amplification peaks at regions like PIK3C2B (1q32.1), PDGFRA (4q12), EGFR (7p11.2), CDK4 (12q14.1), and a focal deletion peak at 9p21.3 (CDKN2A, CDKN2B). Significant amplification peaks were also observed at 2p24.2, 7q34, 8q24.13, 11q23.3, and 19p13.3, with frequent deletions at 2q37.3, 4q34.3, 10q26.3, and 19q13.42 (Figure 7C, Supplementary Table S4). TMB analysis revealed that high-risk patients had significantly higher TMB than low-risk patients (P<0.001, Figure 7D), and a positive correlation was observed between TMB and PLPS (Figure 7E). Kaplan–Meier survival analysis showed that patients with high TMB and high-risk scores had the worst prognosis, while those with low TMB and low-risk scores had the highest survival rate (Figure 7F). These findings suggest that TMB, along with PLPS, could be an important factor in understanding the prognosis of LGG patients.