First, the Gene Expression Omnibus (GEO) database was searched for IA-related bulk and single-cell RNA-seq datasets. To investigate the possible role of hypertension in IA development, the keyword “Hypertension” was retrieved from the GEO dataset to obtain HT-related datasets. The filtering criteria for the datasets were as follows: First, bulk RNA-seq datasets were derived from human samples, and each dataset contained at least 10 samples. Second, single-cell RNA-seq datasets were assigned as the control design. Transcriptomic data from the peripheral blood mononuclear cells (PBMCs) of 11 patients with hypertension and 10 healthy controls (HCs) were collected from the GSE75360 dataset to analyze hypertension-related differentially expressed genes (DEGs) according to the set conditions (30). The mRNA transcriptomic data from 44 IA samples and 16 HC samples of the intracranial cortical artery from the GSE122897 dataset were used to analyze IA-related DEGs. The GSE13353 dataset contains 11 ruptured and 8 unruptured intracranial aneurysm samples, and this dataset was used as an external validation cohort in this study (31). To assess the transcription of AI-related genes at the single-cell level, we included a single-cell RNA-seq dataset (Gene Expression Omnibus accession no. GSE166676) with four cases of abdominal aortic aneurysm (AAA) and two cases of non-aneurysmal control (NAC) among patients with atherosclerotic occlusive disease after a cautious examination (32). The single-cell RNA sequencing (scRNA-seq) dataset contains transcriptomic data from 14,088 cells. All analyses and plots were created using R (version 4.0.2). Differences between groups were assessed using the Wilcoxon rank-sum test.

First, the R package Seurat (version 4.0) was used for the quality control of scRNA-seq data from AAA and NAC samples (33). The metrics used for quality control included the number of gene signatures detected in each cell, total RNA count, and proportion of mitochondrial and hemoglobin genes expressed. High-quality data from the scRNA-seq dataset were screened by removing cells containing <200 or > 2,500 genes as well as cells with > 10% of mitochondrial genes. Finally, we obtained 9,796 cells for subsequent analysis. The “SelectIntegrationFeatures” function was then used to identify the top 2,000 highly variable genes shared among the six samples. Further, the “FindIntegrationAnchors” function was then used to find the anchors from the highly variable genes. Finally, by applying the “IntegrateData” function, the scRNA-seq data from the six samples were combined for subsequent dimension reduction and clustering analysis.

The uniform manifold approximation and projection (UMAP) algorithm was used to analyze and visualize cell clustering (34). The top 30 principal components (PCs) were selected to further perform UMAP and clustering analysis. We first used SingleR (version 1.4) to predict the cell types of individual cell clusters (35). The Database of Immune Cell Expression (DICE) and Monaco Immune Cell Data were selected as the reference datasets to predict the types of immune cells in the IA single-cell dataset. Then, the cell type predictions were manually validated against the marker molecules of immune cells from the CellMarker (http://biocc.hrbmu.edu.cn/CellMarker/index.jsp) website to complete the final cell annotation.

According to a literature review, monocytes/macrophages play an essential role in IA progression (36). Therefore, we further included monocytes/macrophages for subsequent analysis. First, the RNA transcripts of all genes in the monocytes/macrophages between the AAA and NAC groups were subjected to differential expression analysis. Two approaches were used to identify DEGs between the two groups. The first approach applied the “FindMarker” function to identify DEGs between the two cell populations. The other approach used DESeq2 to the constructed pseudo-bulk RNA data after establishing the pseudo-bulk RNA-seq data for differential expression analysis. DEGs identified with these two approaches were combined as monocyte/macrophage-related DEGs.

Based on the type of bulk RNA-seq data, the limma or edgeR package was used for differential expression analysis. We applied the limma package for the differential analysis of FPKM-type RNA data and obtained IA-related DEGs (37). The R package “edgeR” was used to identify hypertension-related DEGs and to perform the differential analysis of count-type RNA data (38). DEGs were screened with a threshold of P < 0.05. Subsequently, we performed intersection analysis of IA-related DEGs, hypertension-related DEGs, and monocyte/macrophage-related DEGs to obtain shared DEGs, which were considered as genes of interest (GOIs).

Intersection analysis of IA-related DEGs and monocyte/macrophage-related DEGs revealed 57 shared DEGs. Then, using the bulk RNA-seq data for IA, we incorporated these 57 genes into the Support Vector Machines–Recursive Feature Elimination (SVM–RFE) model to predict the incidence of IA. This model can be applied to select the optimal number of variables that must be included in the model and can yield the most relevant variables for classification prediction (39).

To explore the significant TF regulatory network in monocytes/macrophages, we applied the TRRUST (version 2.0) database to predict the TF–gene pairs corresponding to monocyte/macrophage-related DEGs (40). Cytoscape (version 3.9) was used to visualize this TF–gene network, and the Molecular Complex Detection plug-in was further used to identify the core sub-networks in this network (degree cutoff = 2, node score cutoff = 0.2, and max depth = 100).

To explore the potential pathways involved in DEGs, we used the clusterProfiler R package for Kyoto Encyclopedia of Genes and Genomes and Gene Ontology (GO) analysis as previous research (41, 42). Gene Set Enrichment Analysis (GSEA) was applied to analyze the functional pathways of enrichment of core genes at the individual and cellular levels. The “c2.cp.v7.2.symbols.gmt [Curated]” gene set downloaded from MSigDB Collections (https://www.gsea-msigdb.org/gsea/msigdb/) was used as the reference gene set. The filtering threshold for differences in functional pathways was set at P < 0.05 (43).

First, we performed quality control on the scRNA-seq data of arterial tissue. The Figure 1A shows the distribution of total NRA counts, the number of gene signatures, and the proportion of mitochondrial gene expression and hemoglobin-related gene expression in cells before filtration. Following the pre-defined filtering conditions, 9,796 cells were included in the subsequent analysis (Figure 1B). UMAP showed that each cell cluster overlapped between the two groups and between the six samples. Therefore, data integration eliminated the batch effect (Figures 1C,D). Data were reduced in dimension and clustered into 21 cell clusters (Figure 2A). Figures 2B,C show four and eight cell clusters annotated using the DICE and Monaco reference datasets, respectively. Among them, monocytes had a relatively large overlap between the two annotation methods. Subsequently, we applied cellular annotation with nine cellular marker genes from CellMarker, which were consistently expressed between the two groups (Figure 2D). Figures 3A,B illustrate the expression of these nine cell markers within each cluster of cells. Among them, CSF1R was used to annotate monocytes/macrophages, CD14 to annotate monocytes, CD68 to annotate macrophages, CD3D and CD2 to annotate T cells, PRF1, and KLRF1 to annotate NK cells, ITGAX to annotate dendritic cells, and CD19 to annotate B cells. Finally, four cell types were annotated by combining the cell annotation results of SingleR and CellMarker, including T cells, B cells, NK cells, and monocytes/macrophages (Figure 3C). A significant difference in the percentage of T cells was observed between the AAA and NAC groups (p 0.05) (Figure 3D). In addition, the proportion of monocytes/macrophages differed significantly between samples in the AAA group. Thus, monocytes/macrophages might be involved in the different stages of IA progression.

In total, 2,102 monocytes/macrophages were extracted and used for differential analysis between the AAA and NAC groups. Further, DESeq2 and FindMarker identified 419 and 539 DEGs, respectively, with a total of 869 genes after removal of intersections. Then, 2,550 and 3,079 DEGs were identified from the bulk RNA-seq data of HT and IA, respectively. These genes of interest (GOIs) were annotated in the volcano map (Figures 4A,B). The top 20 upregulated and downregulated DEGs in AAA identified by FindMarker and DESeq2, respectively, are shown in Figures 4C,D. To further explore the functional pathways that are jointly dysregulated in hypertension and IA, GO function enrichment analysis of the 95 DEGs common to HT and IA was performed. Figures 4E,F show the pathways that are jointly upregulated and downregulated in hypertension and IA, respectively. Finally, cross-tabulation analysis revealed that five GOIs, including IFI30, SERPINE1, HMOX1, IL24, and RUNX1, were found to be associated with hypertension, IA, and monocytes/macrophages (Figure 4G). And the differential expression of IFI30, SERPINE1, HMOX1, IL24, and RUNX1 were shown in the volcano plot (Figures 4A,B).

In total, 57 genes were identified as shared DEGs in monocytes/macrophages from AAA and IA (Figure 4G). These DEGs were then further screened by applying SVM–RFE to the bulk RNA-seq dataset from IA. The variable screening results of the SVM–RFE algorithm showed that the IA prediction model constructed from seven gene features (RPS4Y1, DDX3Y, RUNX1, CLEC10A, PLAC8, SLA, and LILRB3) had the highest accuracy (Figure 5A), with an area under the curve (AUC) of 0.862 (Figure 5B). This seven-gene SVM model was validated by ROC analysis with the GSE13353 dataset as an external cohort (AUC: 0.812) (Figure 5C). Figures 5D,E depict the ROC curves for each of the seven genes (AUC: 0.601–0.817). The above studies suggest that these genes (RPS4Y1, DDX3Y, RUNX1, CLEC10A, PLAC8, SLA, and LILRB3) can predict the prevalence of IA. Figure 5F depicts the expression distribution of these seven genes in the bulk RNA of IA. By analyzing the heat map, RUNX1, CLEC10A, PLAC8, SLA, and LILRB3 were found to be highly expressed in IA, while RPS4Y1 and DDX3Y were lowly expressed in IA.

We analyzed and constructed the TF regulatory network in monocytes and macrophages (Figure 6A). In addition, the intersection analysis of the seven genes and five GOIs used to build the IA prediction model revealed that RUNX1 was the hub gene (Figure 6B). In the PPI network, RUNX1 was responsible for regulating four downstream molecules. According to the core sub-network of this PPI network, NFKB1, regulated by RUNX1, is the hub gene of the entire network (Figure 6C). Therefore, RUNX1 was identified as a hub gene in the IA process by network analysis and crossover analysis.

We performed GSEA analysis of monocytes/macrophages in the single cell dataset. GSEA showed phospholipase C-mediated cascade FGFR4, dopaminergic neurogenesis, and RHO GTPases activate PAKs were downregulated in monocytes/macrophages upregulated with RUNX1 (Figure 6D). Based on GSE122897, GSEA was also further performed to explore the functional pathways enriched for RUNX1 at the individual level (Figure 7). Moreover, it revealed that in samples with upregulated RUNX1 expression, nonalcoholic fatty liver disease, oxidative phosphorylation, RUNX1 regulates the transcription of genes involved in the differentiation of HSCs and those involved in megakaryocyte differentiation, and platelet function pathway RUNX1 regulates genes involved in megakaryocyte differentiation and platelet function. Therefore, RUNX1 plays a complex role in IA progression. Further, it might be involved in IA progression via the regulation of HSC differentiation and platelet production and might play a potential regulatory role in oxidative phosphorylation.