Hallmarks of PANoptosis analysis
Our purpose was to understand the molecular diversity of AD and identify discernible PANoptosis gene patterns. We utilized consensus clustering to sort the AD cohort into three distinct groups based on their PANoptosis gene expression (k = 3). This resulted in the identification of cluster 1 (n = 43), cluster 2 (n = 44), and cluster 3 (n = 38), with small correlation between the groups (Fig. 2A-D). Further, PCA analysis was performed to demonstrate differences in transcriptome profiles between the three clusters (Fig. 2E). The majority of the PANoptosis genes, with the exception of nine genes (AKT3, BNIP3L, BMF, E2F1, FASLG, HMGB1, HMGB2, TNF, and IRF2), showed substantial variations in expression in the three clusters (Fig. 2F, Figure S1).
We also analyzed the expression of AD-related genes, including the amyloid precursor protein catabolic process, the amyloid precursor protein metabolic process, the Aβ metabolic process, the S100 binding protein, and the Tau protein binding. The analysis indicates that there are notable variations in the expression of AD-associated genes across distinct clusters. For example, in the APP catabolic process, the expression of ADAM10, BACE1, and PSEN1 was decreased in cluster 1, but increased in cluster 2 (Fig. 3A); In the APP metabolic process, the expression of ACHE was increased in cluster 2, while the expression of APOE was decreased in cluster 2 (Fig. 3B); In the Aβ metabolic process, the expression of BECN1 and MGAT3 increased in cluster 2, but the mRNA expression of BACE2, MME, and REN decreased in cluster 2 (Fig. 3C). In Tau protein binding, the expression of S100A1, S100A11, and S100A6 was decreased in cluster 2 (Fig. 3D). The mRNA expression of MAP2, PIN1, CDK5, and SNCA were increased in cluster 2 (Fig. 3E).
PANscore development
Describing PANoptosis patterns may be limited for genetic profiling of large cohorts, and it may not allow for quantitative prediction of risk assessment in individual AD patients. Thus, we constructed a PANscore according to PANoptosis gene expression to quantify PANoptosis patterns in individual AD patients.
Initially, we performed WGCNA analysis (Fig. 4A-C) and identified correlation between 13 modules and AD. The lightcyan module (R2 = 0.31, p = 5.3 10 − 7) and the brown module (R2 = -0.46, p = 1.7 10–14) exhibited the most positive and negative associations with AD, respectively (Fig. 4D). With the greatest association coefficient, the brown module (consisting of 4787 genes) was chosen for additional investigation. These genes were shown to have a strong relationship with modules and phenotypes in the scatterplots (R2 = 0.72, p < 0.001; Fig. 4E).
We found 17 key genes associated with AD by obtaining the intersection of the 4787 genes in the brown module of WGCNA and the 41 PANoptosis genes (Figure S2, Fig. 5A). Subsequently, we downscaled these 17 PANoptosis genes with LASSO regression analysis and constructed PANscore models. We identified five PANoptosis hub genes and estimated the coefficient of each gene(Fig. 5B and C). Finally, we obtained the PANscore using the following formula: [expression of DFFA × 0.275] + [expression of GSDMD × 0.012] + [expression of IRF1 × 0.142] + [expression of MLKL × 0.138] + [expression of UNC5B × 0.219]. The predictive model indicated that it performed diagnostically well (AUC 0.84, Fig. 5D). The model was examined in AD tissues and blood samples. The AUCs were comparatively high at 0.87, 0.65, and 0.66 (GSE8521, GSE63061, and GSE63060, respectively, Fig. 5E-G).
Clinical pathological and molecular characteristics
The Clinical pathological characteristics of the PANscore group in the expression data were analyzed. The PANscore was significantly higher in AD (p < 0.001, Fig. 5H) and female (p < 0.05, Fig. 5I) patients compared to ND and male samples, respectively. There was no significant difference between different age groups (Fig. 5J).
We conducted a comparison of gene expression involved in the regulation of AD progression, with respect to high and low PANscore. In the APP catabolic process, ABCA7, ABCG1, ADAM17, APH1A, CLN3, and NCSTN were up-regulated in the high PANscore, however, ADAM19, DHCR24, DYRK1A, and PSEN2 were down-regulated (Fig. 6A). In the APP metabolic process, ACHE was down-regulated in high PANscore, while DLG1, KLK6, LDLRAP1, and TMCC2 were up-regulated in high PANscore (Fig. 6B). For the Aβ metabolic process, APEH and BECN1 were down-regulated in high PANscore, while, BACE2, MME, NAT8B, and REN were up-regulated (Fig. 6C). In S100 protein binding, AGER, AHNAK, ANXA2, EZR, FGF1, IQGAP1, S100A1, S100A11, and S100A6 were significantly up-regulated in high PANscore, while САСYВР was down-regulated (p < 0.001, Fig. 6D). For Tau protein binding, AATF, BIN1, BRSK1, BRSK2, FYN, GSK3B, HDAC6, HSPA2, ROCK1, ROCK2, and TAOK1 were up-regulated in high PANscore, meanwhile, ACTB, CDK5, GSK3A, HSP90AA1, HSP90AB1, LGMN, MAP1A, MAP2, MARK1, MARK3, PIN1, PPP2CA, PPP2CB, and SNCA were down-regulated (Fig. 6E). As a whole, our analysis revealed significant differences in the expression of multiple genes involved in AD development between high and low PANscore.
Enrichment Analysis
The potential biological function of five PANoptosis hub gene were investigated using enrichment analysis. The results exhibited that these genes are mainly involved in programmed cell death, cell death, and regulation of myd88 dependent toll-like receptor signaling pathway (Figure S3). We analyzed the high and low PANscore groups and five PANoptosis hub gene using GSEA. The presence of these genes has been linked to several neurodegenerative diseases (AD, Parkinson’s disease, Huntington’s disease), tumors, and other pathways (Fig. 7).
Immune infiltration cells analysis
Clinical therapeutic sensitivity and diagnostics are significantly impacted by immune infiltration cells. The CIBERSORT algorithm was used in the current investigation to assess the immune infiltration cells in AD (n = 125) and ND (n = 120) samples (Fig. 8A). We were comparing immune infiltration cells between AD and ND samples as well as between groups with high and low PANscore (Fig. 8B). The proportion of plasma cells, follicular helper T cells (Tfh), NK cells activated, macrophages M1 was significantly lower in the AD samples, while γδ T cells and T CD4+ memory activated. In high PANscore, T cells CD8+, Tfh, NK cells activated, eosinophils were significantly lower levels of infiltration, while T cells CD4+ memory resting cells, NK cells resting, and macrophages M1. The investigation evaluated the correlation between immune infiltration cells and the PANoptosis hub gene (Fig. 8C). Findings revealed that DFFA was positively correlated with plasma cells (R2 = 0.3, p < 0.001), IRF1, and macrophages M1 (R2 = 0.38, p < 0.001), while GSDMD (R2 = -0.37, p < 0.001), and MLKL (R2 = -0.37, p < 0.001) with Tfh were significantly negatively correlated. The results indicate that hub genes could potentially have a significant impact on the immune microenvironment.
Correlation between PANoptosis and different apoptosis
We calculated the enrichment scores in the AD cohort by GSVA method. The finding revealed a positive correlation between PANoptosis and apoptosis (R2 = 0.59, p = 2.6×10–24, Fig. 9A), and a negative correlation with cuproptosis (R2 = -0.54, p = 7.4×10–20, Fig. 9B), and ferroptosis (R2 = -0.37, p = 2.2×10− 9, Fig. 9C). This suggests that PANoptosis and apoptosis, cuproptosis, and ferroptosis may be related modes of cell death.
Additionally, we investigated the connection between AD-related metabolic processes and PANoptosis. We obtained AD, neuroinflammation, neurofibrillary tangles, regulation of Tau protein kinase activity and response to oxidative stress related genes and calculated enrichment scores. The results showed a strong negative correlation between PANoptosis and AD (R2 = -0.7, p = 2.8×10− 37, Fig. 9D); a weak negative correlation with neuroinflammation (R2 = 0.23, p = 3.1×10− 4, Fig. 9E), neurofibrillary tangles (R2 = -0.25, p = 1.0×10− 4, Fig. 9F), and regulation of Tau protein kinase activity (R2 = -0.23, p = 3.1×10− 4, Fig. 9G); not significant correlation with response to oxidative stress (R2 = -0.02, p = 0.74, Fig. 9H). This result still needs further experimental verification.