Differential expression of PRG in transcription level of PAAD samples
We have made a rigorous differential analysis of the expression of 46 PRGs in pancreatic cancer. Containing 350 samples, and the results indicated that there were 46 PRGs with significant differences in transcription level between tumor and normal tissues. Moreover, more interestingly, compared with normal tissues, the transcription level of all PRGs is much higher in tumor tissues. In general, mRNA level of a gene is affected by a variety of environmental variables, such as DNA methylation and transcription factors, and mRNA stability is also a major factor. Gene such as GSDMC are molecules of that Gasdermin family, which is Pore-forming proteins that cause membrane permeabilization and pyroptosis (Ding, Wang et al. 2016). Significant differences in the genetic landscape and expression levels of PRGs between PAAD and normal samples indicated a potential role for PRGs in the development of PAAD.
Identification of pyroptosis subtypes in PAAD
PRG interactions, regulatory ligation, and their prognostic value in patients with pancreatic cancer are demonstrated in a Pyroptosis network. In order to further explore the expression characteristics of PRGs in PAAD, the consensus clustering algorithm had been established on the expression profiles of the 46 PRGs. Our results show that k = 3 appears to be the best choice for categorizing the entire cohort into three subtypes (Figure 2A). Principal component analysis revealed significant differences in sample profiles for the three subtypes, with subsequent analyses showing significant statistical significance (Figure 2B). The Kaplan-Meier curve showed that the OS of patients with subtype B and C was shorter than that of patients with subtype A (log-rank test, p = 0.014; Figure 2C）.Differences in clinicopathologic features and expression levels of PRGs between the three distinct subtypes was shown by heat map（Figure 2d. In Figure 2D, we can see that the expression levels of most PRG in B PRGcluster are much higher than those of A and C, which indicates that the prognosis of our patient was significantly correlated with the expression levels of PRG.
Characteristics of the TME in distinct subtypes
Gene set enrichment analysis is used to assess changes in activity in the pathway/function in which the gene set is located. A comparative analysis of three different PRGcluster (A/B/C) was performed here. It can be seen that B_PRGcluster is always enriched in the immune pathway. In contrast to A_PRGcluster, B_PRGcluster is found in the KEGG _ primary _ immune, KEGG _ CYTOKINE _ CYTOKINE _ RECEPTOR _ INTERACTION, KEGG _ INTESTINAL _ IMMUNE _ NETWORK _ FOR _ IGA _ PRODUCTION, KEGG _ TOLL _ LIKE _ RECEPTOR _ SIGNALING _ PATHWAY, KEGG _ T _ CELL _ RECEPTOR _ SIGNALING _ PATHWAY, and KEGG _ B _ CELL _ RECEPTOR _ SIGNALING _ PATHWAY. In contrast to B_PRGcluster and C_PRGcluster, the pathways through which B_PRGcluster is significantly enriched include KEGG _ NATURAL _ KILLER _ CELL _ MEDIATED _ CYTOTOXICITY, KEGG _ LEUKOCYTE _ TRAN ENDOTHYAL _ MIGRATION, KEGG _ CHEMOKINE _ SIGNAGING _ PATHWAY, KEGG_PATHWAYS_IN_CANCER.(Figure. 3A-C). For the purpose of studying the role of PRG in the TME of PAAD, we evaluated the association between the three subtypes and 22 human immunocyte subsets in each PAAD sample with the CIBERSORT algorithm. We observed a significant difference in the infection of most cells. We found that the infiltration level of most immune cells in B_PRGcluster was much higher than that of the other two PRGclusters, for example, Activated. B. cell, Activated. CD4. T. Cell, Activated. CD8. T. Cell, Immunoture. B. Cell, Eosinophil, Gamma.delta.T.cell, and so on. There were also cases in which the infiltration level of individual immune cells in B_PRGcluster was the lowest, for example, Cd56Dim. Natural. Killer. Cell (Figure. 3D).
Identification of gene subtypes based on DEGs
To explore the potential biological behavior of each of the pyroptosis patterns, 631 DEG associated with the pyroptosis subtypes were identified and analyzed for functional enrichment using the R-package "limma" (Figure 4A). These genes related to pyroptosis subtypes are significantly enriched in immune-related biological processes (Figure 4B), such as immune-related pathways as T cell activation, Leucocyte migration, Leucocyte mediated immunity, and Leucocyte cell adhesion. The KEGG analysis revealed enrichment of immune-and cancer-related pathways (Figure 4C), such as the Cytokine Cytokine Receptors Interaction, Cell adhesion molecules. All these indicate that pyroptosis plays a vital role in the tumor immune microenvironment of the patients. univariate Cox regression analysis was then performed to determine the prognostic value of a large number of subtype-related genes, and several genes related to the time of RFS were screened for subsequent analysis (p<0.05). To further verify this regulatory mechanism, patients were divided into two genomic subtypes based on prognostic genes by consistent clustering algorithm; Namely, gene subtype A–B. The Kaplan-Meier curve showed that patients with genotype A had the worst RFS, whereas patients with genotype B had better RFS (Figure 4F). In addition, the pyroptosis gene subtype A pattern was associated with advanced TNM stage (Figure 4E). The two pyroptosis gene subtypes showed significant differences in PRG expression, consistent with the expected results of the pyroptosis patterns（Figure 4G）.
Construction and validation of the prognostic PRG_score
PRG_score was established based on subtype-related DEG. The distribution of patients for the three PRGcluster, the two genecluster, the high-low risk, and the two Fustat types is illustrated in Figure 5A. From the observation of risk score in TCGA, we can comprehend that in the case of two subgroups of genecluster, the risk score of Gene Cluster A was much higher than that of Gene Cluster B (Figure 5B). Additionally, A PRGcluster had the highest risk score, whereas C_PRGcluster had the lowest risk score (Figure 5C). At the same time, PRGs were grouped and compared according to the level of risk score, and we observed that the expression level of most PRGs was in direct proportion to risk score, such as CHMP2B, NOD2, HMGB1, CASP6; There were also cases where PRG expression levels were negatively correlated with risk score, such as IL6, NLRP1, GSDMA, and GZMA (Figure 5D). Next, we plotted the survival curve by grouping according to the level of risk score and found that the high-risk group had a much worse prognosis outcome (Figure E5E-G). Meanwhile, the time-dependent ROC curve was also drawn. The prediction performances at the three time points of 1 year, 3 years and 5 years were all excellent (Figure 5H-J; AUC>0.5).
Development of a nomogram to predict survival
Considering the inconvenience and clinical applicability of PRG_score in predicting the survival of patients with PAAD, a nomogram containing PRG_score and clinical pathology parameters was established to predict the incidence in 1, 2, and 3 years. Predictors included PRG score and patient stage. Our AUC experimental results on the nomogram model showed that the RFS of 1, 2, and 3 years had higher accuracy (Figure 6A) in the training set, the test set, and two external verification sets. Calibration curves also showed the accuracy of the prediction model. We judged the prediction effect of the model on the actual results by observing whether the fold lines were fitted on the diagonal (Figure 6B). Patients with PRG scores lower than the median risk score were classified as low-risk groups, while patients with PRG scores higher than the median risk score were classified as high-risk groups. The risk profile of PRG score showed that with the increase of PRG score, the survival time was decreased, and the recurrence rate was increased. We can learn that with the rise of Risk Score, the number of patients in Dead state was significantly increased, and the gene expression showed a positive correlation with Risk Score (Figure 6C-K).
Evaluation of TME and checkpoints between the high- and low-risk groups
To explore the relationship between PRG_score and the immune microenvironment, we have performed the CIBERSORT algorithm to evaluate the association between PRG_score and immune cells abundance. The results as shown in the scatter diagram exhibited that PRG_score had a negative correlation with four types of immune cells: B cells naïve, Monocytes, T cells CD8, and T cells gamma delta, and a positive correlation with Macrophages M0 and T cells regulatory (Tregs) (Figure 7A-F). At the same time, low PRG_score was accompanied by higher immune score, whereas high PRG_score was accompanied by low immune score (Figure 7G). At the same time, we also showed the correlation of several PRG with 22 immune cells in the form of heat map. As in the previous results, a positive correlation could be observed between PRG and Macrophages M0, T cells regulatory (Tregs); Moreover, a negative correlation could be observed with B cells naïve, Monocytes, T cells CD8 (Figure 7H).
Mutation and drug susceptibility analysis
We then analyzed the distribution of somatic mutations between the two PRG_score groups in the TCGA-PAAD cohort. The first ten mutated genes in the high-risk group and the low-risk group were KRAS, TP53, CDKN2A, SMAD4, TNN, RNF43, MUC16, PCDH15, RRIR1, and DAMS12 (Figure 7I-J). At the same time, the correlation between Tumor Burden Mutation and risk score was also revealed, and a significant positive correlation could be observed (Figure 7K). We next selected chemotherapeutic agents currently used for the treatment of PAAD to assess the susceptibility of patients in the low-risk and high-risk groups to these agents. Notably, ABT.263, ABT.888, AG.014699, AMG.706, Axitib had higher susceptibility in the High risk group and higher susceptibility to A.443654, A.770041, AUY922, AKT.inhibitor.VII,AZD.0530 in the low risk group, all of which indicated that PRGs was associated with drug susceptibility (Figure 8).