Prognostic Significance and Immune Environment Analysis Using PANoptosis Molecular Clustering in Gastric Cancer

Introduction

Globally, gastric cancer ranks as the fifth most common malignant neoplasm and is the fourth leading cause of death related to cancer [1]. Patients diagnosed with early-stage gastric cancer and undergoing treatment can achieve a survival rate surpassing 90% over a period of 5 years [2]. The detection rates for early gastric cancer in China currently remain below 10%; however, the rates in Japan (70%) and Korea (50%) are significantly higher, which indicates a gap in early diagnostic capabilities [3,4]. Thus, the 5-year overall survival rate for patients with gastric cancer in China is less than 30% [5]. To enhance management of patients with gastric cancer in China, understanding the underlying mechanisms of a disease and pinpointing precise diagnostic and therapeutic targets are essential for enhancing early detection and improving treatment efficacy.

To preserve normal physiological homeostasis in organisms, cells undergo various death processes. Among the recognized pathways of cell death, apoptosis, pyroptosis, and necrosis are typical examples of programmed cell death (PCD). Earlier research on cell death did not consider these biochemical functions collectively; recent studies have shown that the biological roles of the 3 primary programmed cell death pathways – pyroptosis, apoptosis, and necroptosis – are interconnected. PANoptosis, recognized as a novel form of cell death, is governed by the activity of the PANoptosome complex, which incorporates key molecules from all 3 PCD mechanisms. These molecules trigger the activation of 3 cell death pathways, culminating in a pro-inflammatory response and PANoptosis. Notably, PANoptosis cannot be completely described through any individual cell death pathway; instead, it involves a multifaceted interaction of all 3 pathways [6]. Cysteine aspartase-8 (CASP8) functions as a critical molecular regulator that governs the mechanisms of pyroptosis, apoptosis, and necrosis [7]. Jiang et al demonstrated that CASP8 plays a critical role in the intricate signaling interactions of PANoptosis within cancerous cells [8]. Jin’s research team reported that depletion of NFS1 significantly increases the vulnerability of colorectal cancer cells to the chemotherapeutic agent oxaliplatin. Additionally, their experiments revealed that the deletion of NFS1 works in synergy with oxaliplatin to trigger widespread apoptosis in cells by increasing the intracellular levels of reactive oxygen species (ROS) [9]. Therefore, exploring the significance of PANoptosis in cancer may improve the probability of early detection of tumors and facilitate the development of innovative treatment strategies. A number of studies have recently implicated PANoptosis in tumor progression. For instance, Zhang et al linked PANoptosis to prognosis and immune response in hepatocellular carcinoma [10], while Karki et al identified ADAR1 as a checkpoint limiting anti-tumor immunity by suppressing ZBP1-mediated PANoptosis [11].Wang et al discovered a correlation between PANoptosis molecular clustering and survival prognosis and tumor microenvironment characteristics in colon cancer [12]. However, the potential link between gastric cancer and PANoptosis remains unexplored in detail.

Our recent research has demonstrated that the identification and analysis of molecular clusters and prognostic markers associated with PANoptosis significantly increase the precision in predicting clinical outcomes of gastric cancer patients. The research classified 871 patients into 2 distinct groups on the basis of their PRG expression levels. Subsequent to the initial analysis, patients were categorized into 2 distinct groups on the basis of their gene expression profiles, emphasizing the differentially expressed genes (DEGs). Moreover, we assessed prognostic indicators and risk scores to estimate overall survival (OS) and treatment response.

Material and Methods

DATA COLLECTION:

Clinical data and gene expression profiles were obtained from individuals diagnosed with gastric cancer via the TCGA (https://portal.gdc.cancer.gov) and GEO (https://www.ncbi.nlm.nih.gov/geo/, ID: GSE84437) databases. On the basis of previous studies, a total of 19 PRGs were identified [8,13–20]. Data were merged using R software (version 4.3.1), resulting in final inclusion of 871 patients.

CONSENSUS CLUSTERING ANALYSIS OF PRGS:

We used the ConsensusClusterPlusR software package to identify the classifications exhibiting the highest and lowest intraclass correlations between the PRG and gastric cancer subtypes. To assess the prognostic variables among identified clusters, we employed the Kaplan-Meier approach complemented by the log-rank test for statistical analysis. We analyzed the clinical features of these clusters via the Wilcoxon test. To pinpoint DEGs, we utilized the limma package, selecting those with a log fold change (FC) exceeding 1 and a P value below 0.05. Gene set variation analysis (GSVA) was performed using the gsva R package, while assessment of immune function activities and calculation of immune cell infiltration scores were performed through single-sample gene set enrichment analysis (ssGSEA).

METABOLIC PATHWAYS AND GENE FUNCTION:

The investigation of gene functions and metabolic pathways was conducted through the use of Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG). For data analysis, the study employed several R packages, including ggplot2, Bioconductor Hs.eg.db, and org. To ensure statistical significance, a threshold was set, where P values needed to be less than 0.05.

CONSTRUCTION OF PROGNOSTIC MARKERS ASSOCIATED WITH PANOPTOSIS:

We utilized univariate Cox regression analysis to identify genes whose differential expression was linked to prognosis, termed prognosis-related differentially expressed genes (PRDEGs). We subsequently classified patients into 2 distinct categories, Groups A and B, on the basis of the observed expression patterns of these PRDEGs, enabling a comparative analysis of clinical characteristics, PRDEG expression, and survival times between these gene clusters. Next, we utilized both least absolute shrinkage and selection operator (LASSO) regression and multivariate Cox regression methods to pinpoint 8 pivotal genes for the development of prognostic markers using survival and survminer R packages. We calculated risk scores from the gene expression profiles. To examine survival disparities among groups classified by these scores, Kaplan-Meier (KM) survival analyses were performed. Using the timeROC R package, the effectiveness of the risk scores in predicting survival was evaluated via receiver operating characteristic (ROC) curves and the associated area under the curve (AUC) values. We also developed prognostic frameworks that integrate these risk scores with clinical features. By employusing calibration plots, we are able to compare the predicted outcomes with the observed survival data.

EVALUATION OF THE TUMOR MICROENVIRONMENT:

We conducted an in-depth examination of tumor expression profiles to investigate the link between adverse outcomes and genes related to the tumor microenvironment (TME). We used the CIBERSORT tool to analyze the infiltration of immune cells and explored the associations between these cells and 8 specific genes. Additionally, we applied the Wilcoxon signed-rank test to evaluate differences in TME scores among various groups.

MUTATION AND IMMUNE LANDSCAPE ANALYSIS:

To examine gene mutations across the 2 groups, we calculated tumor mutational burden (TMB) scores using the maftool R package and utilized Spearman’s correlation method to assess the relationship between risk scores and TMB. In addition, we explored the associations of risk classification with microsatellite instability (MSI) and cancer stem cell (CSC) indices.

SENSITIVITY ANALYSIS OF CHEMOTHERAPEUTIC AGENTS:

We utilized the “PRROPHIC” package to determine the IC50 values for chemotherapeutic agents. Next, we assessed the sensitivity of both groups of patients to these anti-neoplastic agents.

RT-QPCR:

Normal gastric epithelial cells (GES-1) and gastric cancer cells (MGC-803) were acquired from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). These cells were cultured in RPMI 1640 medium (11875101, Gibco, USA) enriched with 10% fetal bovine serum (FBS; 16140071, Gibco), along with 100 IU/mL penicillin and 100 μg/ml streptomycin. RNA was isolated via TRIzol reagent (Invitrogen, Waltham, MA, USA). The isolated RNA was then converted to cDNA via cDNA synthesis kits from Thermo Fisher Scientific (Waltham, MA, USA). Cancer and normal tissue expression levels were compared via unpaired t-test. GraphPad Prism (version 9.5) was used for figure generation.

STATISTICAL ANALYSIS:

Statistical analysis was performed using R version 4.3.1 software and | log fold change (FC) | >1 and P value <0.05 were considered statistically significant.

Results

GENETIC DIVERSITY OF GENES:

Expression data for 447 patients were obtained from the TCGA database, which includes 36 normal samples and 411 tumor samples. As shown in Figure 1A, the occurrence of somatic mutations in these 19 PRGs among gastric cancer patients revealed that 75 out of 431 samples (17.4%) presented mutations, and the gene exhibiting the highest mutation frequency among those studied was CASP8. Figure 1B depicts the positions of copy number variation (CNV) alterations across the chromosomes of the PRGs, indicating a significant increase in CNV for FADD, GSDMD, and NLRP3, whereas a decrease was observed for CASP7, IRF1, RIPK1, PSTPIP2, and CASP1 (Figure 1C). Additionally, 18 PRGs were differentially expressed between gastric cancer samples and adjacent normal tissue samples (Figure 1D).

BUILDING THE PANOPTOSIS-RELATED GENE CLUSTER:

To explore the relationships and prognostic relevance of 19 PANoptosis-related genes (PRGs), we developed a network (Figure 2A). We subsequently employed the consensus clustering method to introduce a clustering variable (k), and upon setting k=2, we identified the optimal classification (Figure 2B). According to the PRG expression, we constructed 2 independent PANclusters in the gastric cancer patient population. We found that patients in group A had considerably higher survival rates than those in group B, with a significant P value of less than 0.001 (Figure 2C). Additionally, principal component analysis (PCA) depicted a distinct division between these groups (Figure 2D). To further elucidate the relationship between PRG expression levels and clinical outcomes in patients with gastric cancer, additional analyses were performed, and the results are presented in Figure 2E. Moreover, significant differences were observed in the depth of tumor invasion across different PANclusters, with a statistical significance of P<0.05. We utilized GSVA and ssGSEA to examine 2 distinct clusters, aiming to discover the connections between PRGs and both prognosis and immune responses in patients. Our findings indicated a significant enrichment of immune-related pathways in PRG cluster A (P<0.05) (Figure 2F), and this cluster also demonstrated increased levels of immune cell infiltration (Figure 2G). These results indicate that PRGs influence the prognosis and immunity of patients with STAD.

IDENTIFICATION OF DEGS IN THE PRG CLUSTER:

After completing the preliminary assessment, our research progressed to an in-depth comparison of DEGs across the identified clusters through a GO enrichment analysis. Our findings indicated that these DEGs are involved in a range of cytokine-mediated signaling pathways (Figure 3A). The KEGG analysis indicated that the DEGs were linked to pathways involving the interaction of viral proteins with cytokines, as well as interactions between cytokines and their receptors (Figure 3B). We subsequently employed consensus cluster analysis to classify patients into 2 distinct groups, with KM curves revealing significant differences in survival times between these clusters (Figure 3C). Notably, differential expression of PRGs was observed in the 2 DEG clusters (Figure 3D). Finally, we developed a heatmap to illustrate these relationships (Figure 3E).

PANOPTOSIS PROGNOSTIC ANALYSIS:

Multivariate Cox and LASSO regression were used to create the risk score (Figure 4A, 4B). The boxplots demonstrate that both clusters, labeled A, have low risk scores (Figure 4C, 4D). The Sankey diagram illustrates our process of constructing a prognostic model (Figure 4E). Among the 19 PRGs, 13 had differential expression (Figure 4F). We screened 8 genes to construct prognostic markers.

Significant differences in the expression levels of the 8 evaluated genes were observed between the 2 defined risk groups (Figure 5A). Figure 5B illustrates the individual patient risk scores and corresponding survival outcomes. Analysis via the KM curve indicated a notably greater survival probability for individuals classified within the low-risk category, with statistical significance (P<0.001) (Figure 5C). Prognostic accuracy was assessed ROC curves, which yielded AUC values of 0.615, 0.637, and 0.657 for 1-, 3-, and 5-year survival prediction, respectively (Figure 5D). The nomogram verified that the observed and predicted survival probabilities were similar (Figure 5E). In addition, line graph models containing various clinical characteristics versus risk scores were developed (Figure 5F).

ASSESSMENT OF THE TUMOR MICROENVIRONMENT:

We explored the link between risk score evaluations and the prevalence of various types of immune cells. The results indicated a positive correlation of risk scores with populations of naïve B cells, M2 macrophages, resting mast cells, monocytes, neutrophils, and regulatory T cells. Conversely, populations such as CD8+ T cells, resting NK cells, follicular helper T cells, activated mast cells, activated memory CD4+ T cells, and plasma cells demonstrated inverse relationships with risk scores, as illustrated in Figure 6A. This analysis revealed significant interactions between risk scores and specific immune cell types in our investigated cohort. Additionally, we investigated the associations between 8 prognostic marker genes and immune cell abundance (Figure 6B). Our analysis revealed that the group with lower risk scores had a lower tumor microenvironment (TME) score, as illustrated in Figure 6C.

ANALYSIS OF MUTATIONS AND THE IMMUNE LANDSCAPE:

Our investigation of somatic mutations revealed marked discrepancies in mutation frequencies across 2 distinct groups. The genes TTN, TP53, MUC16, LRP1B, and ARID1A had the highest rates of mutation (see Figure 7A, 7B). Importantly, the tumor mutational burden (TMB) was considerably lower in the high-risk group (P<0.01), as shown in Figure 7C. Similarly, microsatellite instability (MSI) was more prevalent within the low-risk group (P<0.01) (Figure 7D). Additionally, an inverse relationship was identified between cancer stem cell (CSC) levels and risk scores (Figure 7E).

RESPONSE TO NONSURGICAL TREATMENT:

We detected a reduction in the TIDE score for group A, which was correlated with a stronger positive response to immunotherapy (Figure 8A). Further investigation of drug responsiveness revealed that low-risk patients responded more favorably to treatment with paclitaxel, rapamycin, and sorafenib. Conversely, patients classified as high-risk were more responsive to dasatinib, lapatinib, and imatinib (Figure 8B).

RT-QPCR RESULTS:

Using the GEPIA database, we identified 2 critical prognostic genes, GPA33 and APOD, whose expression levels are significantly higher in gastric cancer samples. Further analysis via PCR confirmed the upregulation of these genes in gastric cancer cells compared with normal controls (Figure 9A). A complementary quantitative assessment of immunohistochemical data from the HPA database (https://www.proteinatlas.org) revealed distinct expression levels between normal and cancerous gastric tissues (Figure 9B). Consequently, our research indicates that GPA33 and APOD can serve as fundamental biomarkers for the diagnosis and treatment of gastric tumors.

Discussion

There is growing evidence that pyroptosis, apoptosis, and necroptosis do not function independently. Instead, there are significant interactions among these processes, allowing them to regulate one another and contribute to anti-cancer effects. Both pyroptosis and necroptosis play vital roles in anti-cancer immunity [21–23]. PANoptosis is a hybrid form of cell death. Recent research has explored the role of dysregulated cell death pathways in cancer genesis, leading to the development of prognostic models based on genes associated with these cell death mechanisms. Despite existing studies, comprehensive research elucidating the crucial role of PANoptosis in the initiation and development of gastric cancer remains insufficient. Among the 19 PRGs analyzed in this study, most have been previously associated with the induction of gastric cancer. For example, GSDMD has been found to prevent and inhibit the progression of gastric cancer by modulating regulatory cell death (RCD) [24]. AIM2 was shown to promote epithelial carcinogenesis by connecting cytokine-STAT3 signaling, innate immunity, and the migration of epithelial cells [25]. We stratified gastric cancer cases into 2 unique PRG clusters to analyze the enrichment of immune-related pathways and evaluate immune cell infiltration. Our findings indicated that PRG cluster B was correlated with worse prognostic outcomes. Research has indicated that the spatial distribution of tumor-infiltrating immune cells (TIICs) can predict whether gastric cancer patients can receive effective immunotherapy [26]. DEGs enriched in tumor immune-related pathways were identified. On the basis of these DEGs, patients were classified into 2 distinct groups, where the DEG clusters exhibited notable differences in survival rates and PRG expression. Further analysis through GO and KEGG confirmed the potential association of these DEGs with cancer progression.

In this study, we identified PRDEGs and then applied LASSO and Cox multivariate regression analyses to develop prognostic models and calculate risk scores [27]. Ultimately, we selected the following genes: IFNG, IL2RB, CXCL11, CKB, CRABP2, GPA33, HSPB6, and APOD. Research has indicated that some of these 8 genes are linked to various malignancies. According to WANG et al, when combined with an oncolytic adenovirus (oAds), CXCL11 has a significant anti-tumor effect, thereby enhancing the efficacy of glioblastoma CAR-T cell therapy [28]. Egan et al reported that elevated CRABP2 expression is correlated with advanced disease stage and an unfavorable prognosis in patients with high-risk endometrial cancer [29]. Lopes et al reported that GPA33 can identify gastric cancer patients with a favorable prognosis [30]. These findings indicate that these genes may play critical roles as diagnostic and therapeutic biomarkers in oncology. Patients were segregated into 2 categories on the basis of their calculated risk scores. Notably, individuals in the low-risk group had a superior prognosis. To increase the accuracy of these risk scores in predicting patient outcomes, a nomogram was developed to estimate patient survival times [31]. Furthermore, the analysis of risk scores revealed correlations with immune cells, revealing 6 that were positively correlated and 6 that were negatively correlated. Moreover, the high-risk group had elevated stromal scores and tumor purity, along with a reduced tumor mutational burden (TMB). Cancer cells often evade death by inhibiting programmed cell death (PCD) pathways. However, induction of PANoptosis can circumvent cancer resistance to individual PCD pathways, thereby effectively triggering cell death and inhibiting tumor progression. Therefore, our study may more accurately predict the survival rate of patients compared to previous prediction models for gastric cancer. Our subsequent validation through RT-qPCR confirmed the significant upregulation of GPA33 and APOD in gastric cancer tissues compared with normal controls. These findings suggest that both GPA33 and APOD could serve as valuable biomarkers for diagnosing and treating gastric cancer.

The TIDE score indicates a greater likelihood of immune escape in individuals in the high-risk group. Additionally, sensitivity analysis of chemotherapeutic agents revealed that these patients exhibited greater sensitivity to dasatinib, lapatinib, and imatinib. Consequently, PANoptosis aids in the development of personalized treatment options, indicating that these markers could serve as biological indicators for the diagnosis and treatment of gastric cancer.

However, this study is not without its limitations. The data used mainly came from public repositories and were compiled in a retrospective manner, so selection bias may have been introduced due to the lack of randomized, prospective samples. In addition, some important clinical variables such as tumor specificity, blood counts, surgical treatment, and neoadjuvant chemotherapy were not added to our analysis. Moreover, this study was limited to initial cell experiments to confirm our hypothesis. Regulation of upstream expression of prognostic genes, and how downstream impacts patient outcomes, need further exploration.

Conclusions

We developed a gastric cancer prediction model based on PANoptosis, which enhances our understanding of the pathogenesis of gastric cancer and helps to diagnose gastric cancer early and predict the survival rate of gastric cancer patients, as well as assessing their immune landscape. In addition, our results may assist in investigating the tumor microenvironment of gastric cancer and develop more reasonable and effective clinical treatment measures.