3.1 DEGs search and WGCNA analysis
The above experimental ideas steps are all in Supplementary Fig. 1. GSE57691 samples were extracted from the database and divided into Control and AAA groups, and the samples were analyzed differentially by R package “limma” for RNA-seq expression profiles, by restricting the DEGs to satisfy the conditions of | log2FC |> 1.0 and P adj < 0.05, we obtained 1249 up-regulated differential genes and 3653 down-regulated differential genes (Fig. 1A, B). We extracted the expression profile matrix of 1249 up-regulated DEGs from the experimental cohort to construct the up-regulated gene matrix and performed WGCNA on it. To make the network scale-free, we chose the soft threshold (β = To make the network scale-free, we chose the soft threshold (β = 10) and R2 = 0.87 (Fig. 1C, D) and the adjacency relation was transformed into a topological overlap matrix (TOM). Cluster analysis was performed on the gene expression profiles of AAA group and Control group (Fig. 1E). By determining the minimum size (genome) of the gene tree to be 30 and the sensitivity to be 3, combined with a distance of less than 0.25, we used hierarchical clustering to divide genes with similar expression profiles into 5 gene modules through average hierarchical clustering and dynamic tree clipping (Fig. 1F). We performed the screening of modular genes by delineating the MM threshold of 0.8, GS threshold of 0.1, and weight threshold of 0.1, and finally screened 44 blue modular genes, 93 brown modular genes, 18 green modular genes, 14 yellow modular genes and 1 gray modular gene (Supplementary Table 3). In addition, we plotted a heat map of module feature vector clustering and investigated the relationship between modules and clinical information (Fig. 1G). We found that blue (correlation score(cor) = 0.45 and P = 3.5e-4), yellow (cor = 0.47 and P = 1.7e-16), and green modules (cor = 0.45 and P = 3e-4) were highly correlated with AAA (Fig. 1H). In order to select the final required module, the correlation between module membership (MM) and gene significance (GS) of the five modules was further studied, and the blue module (r = 0.07, p = 0.46), yellow module (r=-0.22, p = 0.46), grey module (r = 0.31, p = 0.02), green module (r = 0.13, p = 0.36) and brown module (r=-0.01, p = 0.88 had the correlation with AAA (Fig. 1I-M).
3.2 Functional enrichment analysis
We performed a KEGG enrichment analysis on each module's genes in order to further narrow down the relevant modules. We discovered that the genes in the blue module were primarily enriched for genes associated with the hematopoietic cell lineage, T cell receptor signaling pathway, Cytokine-cytokine receptor interaction, Viral protein interaction with cytokine and cytokine receptor, Measles, Cell Adhesion Molecules (CAMs), Chemokine Signaling Pathway, Epstein-Barr virus infection, Human T-cell Leukemia Virus 1 infection, B cell The findings of the KEGG Enrichment Analysis's green, brown, and yellow modules are shown in Supplementary Fig. 2A–C. We chose the blue module as the most appropriate module by combining the WGCNA and KEGG results. Next, we ran a GO enrichment analysis on the genes in the blue module. According to BP, the immune system process, immune response, leukocyte activation, cell activation, regulation of immune response, lymphocyte activation, T cell activation, and antigen receptor-mediated signaling pathway were the main areas of enrichment in the blue module (Fig. 2B). In terms of CC, the blue module was primarily enriched with T cell receptor complex, immunological synapse, plasma membrane part, cell surface, cytoplasmic vesicle membrane, vesicle membrane, secretory granule, receptor complex, and secretory granule membrane (Fig. 2C). Molecular transducer activity, transmembrane signaling receptor activity, protein tyrosine kinase binding, G protein-coupled purinergic nucleotide receptor activity, G protein-coupled nucleotide receptor activity, purinergic nucleotide receptor activity, nucleotide receptor activity, interleukin-2 receptor activity, and interleukin-15 receptor activity are the main enriched components of the blue module in MF (Fig. 2D). Additionally, we ran a GSEA analysis. The primary four pathways that were enriched in the data were cardiac muscle contraction, oxidative phosphorylation, ubiquitin-mediated proteolysis, and vascular smooth muscle contraction (Fig. 2E–H).
3.3 The immune microenvironment of abdominal aortic aneurysms
Our KEGG enrichment analysis revealed that the blue module's gene enrichment findings were strongly correlated with immunity. In the AAA data set, the immunological microenvironment was examined using ESTIMATE, MCPcounter, CIBERSORTs, and ssGSEA, respectively. The ImmuneScore, StromalScore, and ESTIMATEScore were all higher in the AAA group, even though the ESTIMATE findings were not statistically significant (p > 0.05) (Fig. 3A). Second, we used MCPcounter to determine the abundance of 10 immune-related cells. When compared to the Control group, the AAA group had higher levels of T cells, cytotoxic lymphocytes, B lineage, myeloid dendritic cells, and neutrophils (Fig. 3B). In the AAA group, there were more T follicular helper cells, T regulatory cells, monocytes, and dendritic activated cells (Fig. 3C). In order to further investigate the degree of immune infiltration in the samples, we employed ssGSEA. The AAA group had higher concentrations of activated CD4 T cells, central memory CD4 T cells, gamma delta T cells, type 17 T helper cells, type 2 T helper cells, immature dendritic cells, eosinophils, and neutrophils (Fig. 3D, E).
3.4 Machine learning and PPI networks together screen for the most characteristic genes
We used two plug-ins from PPI network and Cytoscape to screen the characteristic genes in the blue module and three different learning machine algorithms to determine the most characteristic genes in the blue module. We then obtained the characteristic genes from the aforementioned five methods to establish them as Hub genes. IncMS is used in the random forest method. IncMSE and IncNodePurity are used by the random forest method to evaluate the genes. We screened the top ten crucial genes: ITGAL, PRKCQ.AS1, SELL, PLAC8, NIBAN3, RASGRP1, STAP1, CR2, NUP210, and LRMP because IncNodePurity was more beneficial (Fig. 4A, B). regression results were presented by rainbow plots and Binomial Deviance (Fig. 4C, D). Lasso regression screened out five signature genes (SELL, IL2RB, ITGAL, RCSD1 and PRKCQ-AS1). At the same time, we used the SVM-REF algorithm and repeated the 10-fold cross-validation with 5 repetitions, choosing the "random" repeated cross-validation method, SIZE = 1:10, to extract the genes with higher variable importance and plot the histogram (Fig. 4E). We then evaluated the accuracy of the regression model by Root Mean Square Error (RMSE) and found the point with the lowest RMSE (RMSE = 5) and the best subset of genes: PRKCQ.AS1, ITGAL, SELL, NUP210 and RASGRP1 (Fig. 4F). The model functions best when K is equal to 1 or 3. We calculate the K value of the blue module using Mean Absolute Error, as indicated in the picture (Fig. 4G). Then, using XGBoost, we determined the blue module genes' significance scores, displayed a histogram, and selected the top 10 genes for SHAP analysis (Fig. 4H, I). In addition, we imported the genes from the blue module into the STRING database and visualized them by Cytoscape (Fig. 4J). We screened the signature genes by two plugins, Cytohubba and MCODE, respectively (Fig. 4K, L). Finally, the genes screened by these five methods were intersected to obtain the most characteristic genes: ITGAL and SELL (Fig. 4M). The ROC curves were used to assess the diagnostic value of the most characteristic genes (Supplementary Fig. 3A, B)
3.5 Single-cell analysis located the expression of signature genes in cells
We divided the six samples of the dataset into two groups into AAA1 (GSM5077727, GSM5077728, GSM5077730 and GSM5077731), AAA2 (GSM5077729 and GSM5077732) respectively for QC evaluation and data filtering to prevent mitochondrial and nuclear RNA contamination (Fig. 5A-C). We then performed data cleaning and removed batch effects using the harmony method based on a subset of features for highly variable features (HVGs) (Fig. 5D, Supplementary Fig. 4A). We then searched for 2000 HVGs for cell clustering and identification and showed the top 5 genes: PRSS1, CLPS, PNLIP, AMY3A, CPA1 (Fig. 5E). We performed principal component analysis (PCA) and identified several PC gene populations with high variability, and used the Umap method to downscale the PC gene populations into 18 cell subgroups (Fig. 5F, Supplementary Fig. 4B, C). We screened the top 10 marker genes of each cell subgroup to identify the type of this cell subgroup, and finally re-marked the type of each cell subgroup (Fig. 5G). We then examined the expression of ITGAL and SELL across samples and cell populations. We found that ITGAL showed higher expression mainly in NK cells and Lymphoid cells, while SELL showed higher expression mainly in T regulatory cells, Lymphoid cells, B cells and CD1C+_B dendritic cells (Fig. 5H, I). Combined with the results of the previous immuno-infiltration analysis, we conclude that ITGAL and SELL may affect T regulatory cells, NK cells, B lineage and lymphocytes and thus the development of AAA.
3.6 Gross morphological changes in aaa and conversion of smooth muscle phenotypic proteins
Mice were anaesthetized after 28 days. The gross morphology of the aortic aneurysm was visualized in the model group compared to the ApoE -/- saline group. The aneurysm diameter of the model group was measured and counted as more than 1.5 times the normal diameter (Fig. 6A, B). After HE and EVG staining, it was seen that the abdominal aortic wall was thickened, the mesenteric elastic fiber layer was reduced, smooth muscle cells were reduced, and EVG showed that the number of mesenteric elastic fiber layers was reduced and broken (Fig. 6C). Secondly, we examined the mRNA levels of arterial smooth muscle related proteins, we found that Sm22α, SmMHC, MMP1 and α-SMA expression levels were reduced in the aorta of mice after angiotensin stimulation (Fig. 6D-G). Finally, we detected the protein levels of α -SMA and Calponin in the AAA group, and the expression of both Calponin and α -SMA was decreased in the AAA group (Fig. 6H-J).
3.7 Inflammation and immune in abdominal aortic aneurysms
Increased secretion of pro-inflammatory factors is a key feature in the mechanism of abdominal aortic aneurysms. We measured serum levels of IL-1β and IL-6 in mice by ELISA kits. We found elevated levels of both IL-1β and IL-6 in the angiotensin-treated group compared to the saline group (Fig. 7A, B). In addition, we extracted RNA from the abdominal aorta of mice and performed in vivo RT-PCR analysis to detect changes in their inflammatory factors. Compared with the saline group, the mRNA expression of TNF-α, IL-1β, IL-4, IL-6, IL-8 and IL-17 was significantly higher in the abdominal aortic aneurysm model group, while the mRNA expression of the inflammatory factor IL-10 was lower (Fig. 7C-I).
3.8 Expression of signature genes in mouse abdominal aortic aneurysm model and MOVAS
To verify whether the expression of our signature genes in the model was consistent with the database results, we examined the mRNA levels of ITGAL and SELL in the tissues and found that the expression of SELL and ITGAL increased in the AAA group (Fig. 8A, B). In addition, we have verified this in in vitro experiments. We stimulated MOVAS cells with TNF-α and then extracted samples and examined the mRNA levels of the cells. As in the animal model, the mRNA levels of ITGAL and SELL were upregulated in MOVAS under TNF-α stimulation conditions (Fig. 8C, D).
3.9 Prediction of targeting drugs for AAA by cmap
We screened the top 10 compounds with the highest standardized connectivity score (CS) and false discovery rate (FDR (nlog10)). Clobetasol, Adapalene, Everolimus, Bifonazole, Vorinostat, PSB-06126, Tiotropium, Tasquinimod, Ecopipam and Marimastat) were visualized (Fig. 9A, B). These drugs suggest possible immunological guidance for the treatment of AAA.