All experiments involving human patients were conducted in accordance with the ethical policies and procedures of the ethics committee of West China Hospital, Sichuan University (No. 2019-428, 2019-06-15) and in accordance with Good Clinical Practice guidelines and the World Medical Association Declaration of Helsinki. The collection, preprocessing, sequencing, and data management procedures of unruptured and ruptured CA domes have been described in detail previously (Ji et al., 2024a). The cell-to-gene expression matrix, feature and barcode files of unruptured and ruptured murine CA tissues were retrieved from the GEO data portal (GSE193533) (Martinez et al., 2022). Briefly, fifty-eight-week-old male C57/BL6 mice were pharmacologically induced hypertension through ligation of the left renal artery and the right internal carotid artery and stereotactically injected of elastase into the basal cistern 1 week after surgery. For the comparison of human and murine CA, we constructed standard unruptured and ruptured cerebral aneurysm atlases to avoid potential bias of cell counts.

We constructed a human Standardized Unruptured CA Atlas (SUCA) for the comparison with unruptured CA from murine model. The SUCA consists of a total of four unruptured CA of human adult origin (Supplementary Table S1). We conducted stringent quality control for each sample separately, with the criteria being set as 350 < nfeature <4,500, and the proportion of mitochondrial genes <35%. Doublets were excluded using the R package “DoubletFinder” (McGinnis et al., 2019). The samples are integrated by utilizing the “HarmonyIntegration” method of the “IntegrateLayers” function of the R package “Seurat” (v5.0) (Hao et al., 2023). We initially set the resolution to 0.35 for identifying cell communities. After assignment of cell types, we randomly selected a quarter of each type of cell for the composition of the SUCA. Similarly, a human Standardized Ruptured CA Atlas (SRCA) using two ruptured IA domes following the same criteria and approach (Supplementary Table S1).

The cell-by-gene expression matrix of sham (psudo-operation), formed (CA formation), and ruptured CA developed by Martinez AN et al. were retrieved from the Gene Expression Omnibus (GSE193533) (Martinez et al., 2022). The conversion of homologous genes of mus musculus and homosapiens were conducted using the R package ‘biomart’ (Durinck et al., 2009). Data integration was conducted using the ‘harmony’ algorithm implemented in the framework of ‘Seurat’ (v5.0) (Hao et al., 2023).

The differential expressed genes (DEGs) were calculated using the “FindAllMarkers” function of the “Seurat” R package with default parameter. The average log2 fold change >1, adjusted p value <0.05 and pct.1 > pct.2 was set as the cutoff. Functional enrichment analysis was conducted based on the Gene Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) using the webtool Metascape (https://metascape.org/) (Zhou et al., 2019; Thomas et al., 2021; Goad and Kanehisa, 1982).

The principal component analysis (PCA) was performed to explore the similarity of cells using the R function “prcomp.” The count matrix was log transformed and scaled. Top 8,000 variable genes were filtered through calculating the mean absolute deviation (MAD). The cell type average expression matrix was used as the input. The coefficient matrix and eigen vectors was calculated to avoid overweighting of an outlier.

Multiple trajectory analyses were conducted for dissecting the internal association between cells. Diffusion map, implemented in the R package “destiny,” leverages local similarity and nonlinear characteristics of data to preserve the intrinsic structure and continuity of the dataset, thereby effectively capturing the features inherent in its continuous distribution of cells (Angerer et al., 2016; Haghverdi et al., 2015). The monocle algorithm was additionally employed to further identify branches of the trajectory and generate pseudotime (Trapnell et al., 2014). Its main principle is to project high-dimensional gene expression data into a lower-dimensional space while preserving the complete structure of the data using the DDRTree algorithm.

The interaction between cell types were inferred using the R package ‘CellChat’ (Luo et al., 2023). We extracted the count data and do normalization and log transformation as the input of CellChat. The CellChat database is a manually curated literature validated database of ligand receptor interactions in humans and mice, including autocrine/paracrine signaling interactions, extracellular matrix (ECM) receptor interactions, and cell-cell contact interactions. To ensure the reliability of the “computCommunProb” function, we set the truncation value of ‘trimean’ to 0.25, meaning that if the percentage of a group of expressing cells is less than 25%, the average gene expression is defined as 0. The p value of cell-cell communication was adjusted using the false discovery rate (FDR).

The transcriptional gene regulatory network (GRN), the transcription factors (TFs) and their target genes, was inferred using the R package “SCENIC” (Bravo González-Blas et al., 2023). Genome reference files of human (hg38) and mice (mm10) were retrieved from the CisTarget database (https://resources.aertslab.org/cistarget/). The filtering criterion is set as the sum of gene expression levels >3% of the total number of cells, and it should be expressed in over 1% of the cell population. The co-expression modules were constructed by grouping genes with weight >0.001 (w001) around each TF.

Multiplex immunofluorescence staining (mIHC) was performed on sections of a giant unruptured IA sample. Fresh IA domes were fixed in 4% paraformaldehyde and dehydrated in graded methanol overnight at −20°C. Samples were rehydrated, washed, and blocked with SEA BLOCK Blocking Buffer sequentially, followed by incubation with primary antibodies overnight at 4°C. Anti-CD31 (1:1000; HUABIO, China), anti-αSMA (1:2000; HUABIO, China), anti-CD68 (1:1000; HUABIO, China), anti-CD11b (1:1000; HUABIO, China), and anti-DCN (1:500; HUABIO, China) were employed to determine the expression of proteins of interest. The second antibodies used were rabbit poly-HRPs labeled with cyclic-480/550/630 dyes using IRISKit HyperView mIF kit (LUMINIRIS, China). DAPI was used as the nuclear counterstain. The slices were scanned using the Olympus VS200 scanner. The images were adjusted using ImageJ followed by analyzing using the QuPath software version 0.4.4.

All statistics was performed using the R software (v4.3.3). Mann-Whiney U rank sum test was conducted for the calculation of DEGs, and p values were adjusted through applying false discovery rate. An adjusted p value less than 0.05 was defined as statistically significant.

The SUCA is constructed using four unruptured CA domes and comprises of a total of 4,600 quantified cells, including endothelial (n = 531, 11.54%), vSMC (n = 600, 13.04%), fibroblast (n = 26, 0.57%), lymphocytes (T cell, B cell, plasma cell, CD56⁺ NK cell and CD56⁻ NK cell, n = 1,307, 28.41%), neutrophil (n = 1,272, 27.65%), monocyte (n = 241, 5.24%), macrophage (n = 574, 12.48%), and mast cell (n = 49, 1.07%). The proportion of each type of cell is exhibited (Figure 1A). No dendritic cell (DC) is found in the unruptured human CA. In turn, the SRCA comprises of 5,406 quantified cells, including endothelial (n = 83, 1.54%), vSMC (n = 172, 3.18%), lymphocytes (T cell, B cell, and NK cell, n = 1,321, 24.44%), neutrophil (n = 2,620, 48.46%), monocyte (n = 289, 5.35%), macrophage (n = 714, 13.21%), DC (n = 189, 3.50%), and mast cell (n = 18, 0.33%) (Figure 1B). No endothelial cell and fibroblast are captured in the ruptured human CA, which is consistent with the severe damage to the artery wall caused by an aneurysm rupture (Etminan and Rinkel, 2016). Overall, the proportion of vascular cells, including endothelial, vSMC, and fibroblast, decreased markedly in rupture CA (unruptured/ruptured: 25.15% vs. 4.72%), while the proportion of neutrophil is remarkably increased (unruptured/ruptured: 27.65% vs. 48.46%).

As for CA tissue from murine model, a total of 4,783 quantified cells are captured from the unruptured CA, encompassing endothelial (n = 249, 5.21%), vSMC (n = 2,249, 47.02%), fibroblast (n = 544, 11.37%), lymphocytes (B cell and T cell, n = 121, 2.53%), neutrophil (n = 49, 1.02%), monocyte (n = 96, 2.01%), macrophage (n = 1,393, 29.12%), DC (n = 67, 1.40%) and mast cell (n = 15, 0.31%) (Figure 1C). The ruptured murine CA contains 3,438 quantified cells, including endothelial (n = 69, 2.01%), vSMC (n = 478, 13.90%), fibroblast (n = 47, 1.37%), lymphocytes (B cell, T cell, n = 251, 7.30%), neutrophil (n = 1,096, 31.88%), monocyte (n = 215, 6.25%), macrophage (n = 1,138, 33.10%), DC (n = 121, 3.52%), and mast cell (n = 23, 0.67%) (Figure 1D). A remarkable decreasing of the vascular cells (endothelial, vSMC, and fibroblast) in the ruptured murine CA is also observed (63.6% vs. 17.28%). Notably, the proportion of vascular cells of unruptured CA of human patients were lower than that of murine model (25.15% vs. 63.6%), which may be ascribed to over 2,000 vSMCs (47.02%) are captured in the unruptured CA from murine model. The most significant decrease is observed in the proportion of vSMC (human, 9.86% reduction; murine model, 33.12% reduction). Additionally, the unruptured CA of human patients contained remarkably higher proportion of neutrophil than that of murine model (27.65% vs. 1.02%), which plays a crucial role in facilitating CA rupture (Korai et al., 2021). Therefore, these results demonstrate preliminary discrepancies between human and murine model CAs.

The construction of an integrated CA atlas of human and murine model is carried out according to stringent criteria. We performed independent quality control for each dataset. To maximize the retention of vSMCs, we limited the mitochondrial proportion to ≤35%. To minimize the presence of deceased cells or cell debris, we employed different resolutions (0.5, 1.0, and 1.5) for cell community detection, resulting in a final identification of 9, 14, and 20 clusters. We excluded cell clusters exhibiting high expression of ribosomal transcripts, such as RPL32, RPL13, and RPL18A, which accounted for 0.31% of all cells. As a result, a total of 18,170 cells were assigned, including endothelial (n = 849), vSMC (n = 3,499), fibroblast (n = 617), lymphocyte (T cell, B cell/plasma cell, NK cell, n = 2,965), mono/macro (monocyte, macrophage, DC, n = 5,027), neutrophil (n = 5,107), and mast cell (n = 106) (Figure 2A). There are cell types in humans and murine model that can be projected into a common community in the umap embedding, including endothelial cells, lymphocytes (T cell, B cell and NK cell), and mast cells, indicating conserved gene expression patterns. Intriguingly, there are also cell types classified into distinct communities with a proximal distance in the umap embedding, encompassing vSMC, neutrophil, and mono/macro (monocyte, macrophage, DC) that possess modifiable phenotypes (Alencar et al., 2020; Salcher et al., 2022; Xue et al., 2014) (Figure 2B). Collectively, these findings suggest that the potential discrepancies between human and murine models of CA may be attributed, at least in part, to cells of modifiable phenotype.

The endothelial cell dysfunction plays a critical role in the formation and progression of CA (Etminan and Rinkel, 2016; Turjman et al., 2014; Frösen et al., 2012), we therefore characterized the transcriptomic features of human and murine model endothelial cells. As a result, a total of 531 endothelial cells were identified in human unruptured CA and 318 from murine model, accounting for 5.34% and 3.87% of the total cell population, respectively. These cells can be projected into a common umap community (Figure 3A). Consistent with these findings, the PCA analysis shows the spatial proximity of endothelial cells, further indicating conserved transcriptomic features (Figure 3B). In addition, multiple trajectory analyses were performed for further dissecting the potential disparities, including the monocle that is sensitive to branches of trajectory and the diffusion map that better reflecting the continuity of single cells. Despite the observed disparities between human and murine model CA endothelial cells along the pseudotime in monocle, the findings of diffusion map do not provide support for such discrepancies (Figures 3C, D). Therefore, these findings may suggest that the transcriptomes of endothelial cells in human and murine model of CA are conserved.

Recent studies highlight the modifiable phenotype of vSMCs that are involved in various cardiovascular and cerebral vascular diseases (Yap et al., 2021; Sorokin et al., 2020; Lei et al., 2023). In our study, a total of 772 vSMCs from human CA samples and 2,727 from murine model are designated as distinct communities in the umap embedding, with the majority of the cells originating from unruptured CA (81.42%) (Figures 4A, B). The umap embedding shows a closer proximity between human vSMCs and fibroblasts (Figure 2A), which is corroborated by the observation that human vSMC express higher levels of fibroblast marker genes, including COL1A1/2, FBN1, and DCN (Figure 4C; Supplementary Figure S1). To further understand the disparities between human and murine model vSMCs, the trajectory analyses were performed. As a result, both monocle and diffusion map analyses reveal a significant uneven distribution of human and murine model vSMCs along the trajectory, with a notable enrichment of human vSMCs toward the terminal stage of pseudotime (Figures 4D, E). To unbiasedly understand the gene expression features associated with vSMC heterogeneity, we combined the calculation of DEGs and their correlation with our constructed pseudotime. As a result, the well-defined vSMC gene marker gene MYH11 is highly expressed by murine model vSMCs, as well as other myosin encoding genes such as MYL6 and MYL9 (Figure 4F). Instead, the vSMC cells from human CA highly express genes associated with extracellular matrix (COL6A2, COL1A2, LGALS3), cytoskeleton remodeling (TMSB10, DCN), and HLA class I molecule coding genes that are involved antigen presentation (HLA-A, HLA-B, HLA-C). Moreover, the purine nucleoside triphosphate metabolic process is enriched in human vSMCs (Figure 4F), which is associated with cell proliferation upon pathological stimulus (Ma et al., 2023; Ma et al., 2022). Given the alternative phenotypes of vSMCs, these results demonstrate a deviation of human vSMC phenotype from normal contractile vSMCs.

Neutrophil infiltration is destructive to the arterial wall and is closely associated with CA rupture (Ji et al., 2024a; Korai et al., 2021; Ji et al., 2024b). The presence of neutrophils in unruptured CA is observed in both human (27.65% of total cell) and murine model (1.02%), with a significant increase upon CA rupture (48.68% in humans and 31.88% in murine model) (Supplementary Figure S1). Neutrophils of human and murine model origin are presented as two related communities in the umap embedding (Figure 5A). The functional enrichment analysis shows that the most enriched term in neutrophils is degranulation in both groups (Figure 5B). Intriguingly, the pseudotime analyses suggest discrepancies in human and murine model neutrophils. The diffusion map suggests a biased distribution of cells along the trajectory, with human neutrophils enriched in one end of the trajectory (Figure 5C). Subsequently, the monocle algorithm detected branches of the trajectory, which further characterizes the disparities (Figure 5C). Genes that are highly associated with the pseudotime and differentially expressed in human and murine model are calculated (Figure 5D). For instance, FTL1, involved in iron storage; ACOD1, involved in metabolic regulation; and TMSB4X and CCL3, involved in neutrophil migration and chemotaxis, are significantly enriched in murine model neutrophils. Instead, human CA neutrophils highly express CXCL8 and DOCK4 that regulating chemotaxis and migration. These findings may also suggest distinct patterns of neutrophil infiltration in human and murine model, prompting us to conduct a cell-cell communication analysis to gain further insights into the ligand-receptor-based interactions involved in neutrophil recruitment. Generally, monocyte/macrophage and neutrophil exhibit potent interactions with neutrophils, especially through chemokines and IL1β (Figure 5E). In human CA, the recruitment and chemotactic action of neutrophils towards themselves is remarkably CXCL8-dependent, while in the murine model, neutrophil recruitment is primarily mediated by CCLs (Figure 5F). Thus, the pattern of neutrophil recruitment in human and murine model may be different.

Macrophages play a paramount role in CA inflammation (Etminan and Rinkel, 2016). The proportion of macrophages in the SUCA was 12.48%, which increased to 13.21% in the SRCA (Supplementary Figure S1). In the murine model, there are higher proportion of macrophages, with proportions of 29.12% and 33.10% being observed in the unruptured and ruptured CA tissues, respectively. The functional enrichment analysis of DEGs indicates that terms associated with antigen presentation, immune response, and regulation of immune cells are significantly enriched in macrophages of human unruptured CA (Figure 6A), which are similar to that of macrophages from unruptured murine model. Nevertheless, macrophages from human unruptured CA express higher levels of M2 marker genes (CD36, CLEC7A, CD163 and MRC1) and lower levels of M1 marker genes (CD86, CCL2, CXCL1) than that of murine mode (Figure 6B), indicating a preference of M2 polarization of human CA macrophages. Further, we investigated the gene regulatory network underline these disparities. As a result, macrophages from unruptured human CA are characterized by the transcriptional activation of NF-κB, FOS and JUN, and their high-confidential target genes are enriched in pathways including TNF signaling pathway (Figure 6C). In turn, the macrophages from murine model are governed by TFs including Cebpb, Junb, and Nfe2l2. Particularly, the irf7 is associated with M1 polarization of macrophage (Ling et al., 2023). Moreover, functional enrichment analysis suggests pathways enriched in murine CA macrophages, including lipid and atherosclerosis and FoxO (Figure 6D), where the FoxO pathway is involved in the M1 towards M2 phenotype polarization (Chung et al., 2015).

Given the crucial role of inflammation in CA rupture and our finding that immune cells such as neutrophils and monocyte/macrophages in human and murine model CA remain distinct, we next characterized the overall disparities of the local inflammation status. To rigorously filter genes for downstream analysis, we eliminated transcripts expressed in fewer than 10 cells. Next, we selected genes highly variable between cells through retain those transcripts with a residual variance over 0.5, filtered down to 6,374 genes for further analysis. The DEGs were calculated and functional enrichment analysis suggests differences in immune-associated pathways. Despite that neutrophil degranulation and allograft rejection are enriched in both groups, terms including adaptive immune system, neutrophil extracellular trap formation, and IRF3-mediated induction of IFN I are enriched in human CA tissue (Figure 7A), indicating disparities in immune status. Further, we calculated the ssgsea score of several inflammation-related pathways, including acute and chronic inflammation response, cGAS sting pathway, interferon alpha and gamma, and TNF alpha signaling pathway. The cGAS-STING pathway is triggered by the release of free DNAs from damaged cells (Ablasser and Chen, 2019), a phenomenon commonly observed in CA tissue. The TNF alpha, interferon alpha and gamma are important pro-inflammatory cytokines that are involved in the activation of multiple immune cells (Zhang et al., 2016; Sathyan et al., 2015). As a result, the human CAs score significantly higher in cells and molecules involved in local acute inflammatory response (Figure 7B), indicating a feature of acute inflammation. Notably, the ssgsea scores of cGAS-STING pathway, interferon alpha and gamma responses, and TNF alpha signaling via NF-κB are all higher in the human CA tissue, indicating a more intense inflammation in human CA tissue.