Cross-species scRNA-seq finds ideal mouse models for aortic dissection mechanisms

Summary

Aortic dissection is a life-threatening cardiovascular disease whose complex cellular pathophysiology is studied using various mouse models. To systematically evaluate their fidelity, we performed cross-species single-cell RNA sequencing, integrating data from human aortic dissection with five mouse models (BAPN, Ang-II, Ang-II apoE^−/−, elastase, and CaCl₂). Comparative analysis across four key cell types revealed model-specific transcriptional parallels to human disease. The BAPN model recapitulated human pro-inflammatory and lipid-associated macrophage states, as well as pro-inflammatory fibroblast signatures. The Ang-II apoE^−/− model mimicked pathogenic vascular smooth muscle cell phenotypes, particularly lipid-associated and oxidative stress states. Endothelial cell dysfunction was a conserved feature across all models. Our study provides a single-cell-based comparative framework highlighting model-specific transcriptional similarities to human aortic dissection.

Graphical abstract

Highlights

• •Single-cell atlas compares human aortic dissection with five mouse AD models
• •BAPN model best mirrors inflammatory macrophage and fibroblast states
• •Ang-II apoE^−/− model most closely recapitulates pathogenic VSMC phenotypes
• •Endothelial dysfunction signatures are conserved across all AD models

Biological sciences; Bioinformatics; Expression study

Introduction

Aortic dissection (AD) is a life-threatening pathology with high morbidity and mortality. In recent years, important discoveries about the mechanisms of AD have relied on mouse models that mimic the anatomical features and pathogenesis of human AD. Commonly used mouse models for AD include the β-aminopropionitrile (BAPN)-induced mouse model,¹ the angiotensin II (Ang Ⅱ)-induced mouse model,² the AngⅡ-induced apolipoprotein E-deficient (apoE−/−) mouse model,³ the elastase-induced mouse model,⁴ the CaCl2-induced mouse model,⁵ and so forth. Each of these models can mimic the anatomical features of the AD. However, whether these models can fully mimic the realistic pathophysiologic process of human AD remains unknown.

Aortic cells contribute most to the structural integrity of the aortic wall, principally endothelial cells, VSMCs, and fibroblasts. A healthy physiologic state of these cells is essential for maintaining the balance of the aortic microenvironment. However, the development of AD involves the disruption of the aortic microenvironment, particularly the transcriptional reprogramming of vascular cells. Macrophages,⁶ VSMCs,⁷ and fibroblasts⁸ reprogrammed to pathogenic phenotypes in response to cellular stress. Therefore, there is a necessity to investigate the conservation of critical cellular phenotypes in human AD and mouse AD models.

Previous evaluations and comparisons of AD models have focused on pathological features and biomechanical characteristics,⁹ which lack a comprehensive evaluation of the model’s pathogenic mechanisms. The advent of single-cell RNA sequencing (scRNA-seq) represents a significant advancement in investigating the transcriptional heterogeneity at a resolution of individual cells. For instance, studies by Zernecke et al.¹⁰ and Frank et al.¹¹ have utilized scRNA-seq to uncover transcriptional heterogeneity of immune cells within mouse models of atherosclerosis and to explore cellular diversity in abdominal aortic aneurysms, respectively. Preliminary studies employing scRNA-seq to analyze mouse AD models and human AD have begun to shed light on the AD mechanism. Moreover, these studies offer an avenue to directly compare the heterogeneity of cellular phenotype between mouse models and human AD.

In the present article, we revealed the transcriptional heterogeneity of macrophages, fibroblasts, VSMCs, and ECs. Subsequently, the four cell types underwent cross-species integration with their counterparts from mouse AD models, respectively (Figure 1A). Through assessing the subpopulation heterogeneity of these four cell types, we compared the mechanism differences among human AD, the BAPN model, Ang-II model, Ang-II apoE−/− model, elastase model, and CaCl2 model.

Figure 1.

Profile of human and mouse scRNA-seq datasets of aortic dissection

(A) Overview of workflow.

(B) Human and mouse aortic dissection scRNA-seq datasets used in this work.

(C) UMAP representation of integrated human scRNA-seq expression data from aortic dissection (AD) and normal aorta (NA). UMAP representation of mouse scRNA-seq expression data from BAPN-induced AD model (D), Ang-Ⅱ-induced apoE−/− mouse model (E), Ang-Ⅱ-induced model (F), elastase-induced model (G), and CaCl2-induced model (H).

Results

Single-cell RNA sequencing mapping the cell type landscape of human and mouse aortic dissection

For human aortic dissection, 6 samples from our center and 9 samples from GSE213740 were integrated and filtered.¹² Following rigorous quality control protocols, a comprehensive dataset comprising 158,440 cells, corresponding to 70,264 unique genes, was amassed. Utilizing unsupervised clustering with uniform manifold approximation and projection (UMAP) analysis, we identified 28 distinct cellular clusters. These clusters were systematically classified into nine principal cell types based on the analysis of genes exhibiting high variability in expression across the clusters. The identified cell types, along with their canonical markers, include: B cells (marked by CD19, CD79A, and CD79B), endothelial cells (VWF, CDH5, and PECAM1), fibroblasts (LOX, LUM, and DCN), macrophages (CD14, CD68, and FCGR3A), mast cells (TPSB2, TPSAB1, MS4A2, and KIT), neutrophils (FCGR3B, CSF3R, and G0S2), NK cells (KLRB1, KLRD1, and KLRF1), vascular smooth muscle cells (VSMCs) (MYH11, ACTA2, and CNN1), and T cells (CD3D, CD3E, and CD8A) as detailed in Figure 1C. Notably, in the context of human AD, there is a prominent elevation in the fraction of macrophages, although not statistically significant (Figure S1A).

Regarding the scRNA-seq dataset for the BAPN-induced AD model, CRA003013, we processed and filtered the data to yield expression values for 80,947 genes across 31,947 cells (Figure 1D).¹³ Following data filtration, these cells were stratified into 20 clusters, which were further categorized into seven predominant cellular phenotypes, delineated as follows: B cells (identified by Cd79a, Cd19), endothelial cells (Cdh5 and Tie1), macrophages (Itgam and Cd14), VSMCs (Acta2, Myh11, and Cnn1), fibroblasts (marked by Sca1/Ly6a, Pdgfra, and Pdgfrb), PCDs, and T cells (Cd3d and Cd3g) (Figure S1B).

The GSE203594 dataset, derived from angiotensin II (AngⅡ)-infused Apolipoprotein E (apoE) deficient mice, was meticulously processed to obtain a refined gene-cell matrix comprising 18,775 cells and 32,285 genes, as shown in Figure 1E.¹⁴ Through UMAP analysis and variable expressed genes, all cells were clustered into 18 populations and annotated to 10 major cell types as follows: fibroblasts (Pdgfra, Lum, Dcn, and Col1a1), B cells (Cd79a and Cd19), T cells (Cd3d and Cd3g), macrophages (Itgam, Cd68, Cd14, and Csf1r), VSMCs (Acta2, Myh11, and Cnn1), Monocytes, endothelial cells (Cdh5, Tie1, Pecam1, and Vwf), cycling cells (Smoc1, Stmn1, Mki67, and Top2a), adipocytes (Adipoq, Apoc1, Plin1, and Adig), PCDs (Upk1b and Upk3b), stromal cells (Krt8, Krt15, Krt19) (Figure S1C).

In the GSE213735 dataset, deriving from AngⅡ-infused wild-type mice, a gene-cell matrix encompassing 18,705 cells and 14,898 genes was obtained. This culminated in a UMAP analysis (Figure 1F).¹⁵ A total of 17 cell populations were clustered and were annotated as 7 major cell types. Cycling cells (Stmn1, Top2a, and Cks2), endothelial cells (Cdh5, Tie1, and Vwf), fibroblasts (Pdgfra, Dcn, and Lum), macrophages (Itgam, Cd14, and Csf1r), pericytes (Lamc3, Prrx2, Smoc1, and Osr1), T cells (Cd3d and Cd3g), VSMCs (Acta2, Myh11, and Cnn1) (Figure S1D). There was a reduction in the proportion of VSMCs and fibroblasts. Conversely, there was an increase in the populations of macrophages, T cells, and cycling cells.

In the GSE152583 and GSE164678 datasets, 4,881 and 4,689 were respectively derived from the elastase model and the CaCl2 model. A total of 7 major cell types were annotated in the two datasets: Cycling cells (Stmn1, Top2a, and Cks2), endothelial cells (Cdh5, Tie1, and Vwf), Fibroblasts (Pdgfra, Dcn, and Lum), macrophages (Itgam, Cd14, and Csf1r), B cell (Cd79a and Cd79b), T cells (Cd3d and Cd3g), VSMCs (Acta2, Myh11, and Cnn1) (Figures 1G and 1H) (Figures S1E and S1F).

Drawing comparisons with human AD, it was discerned that the cellular compositions identified in each mouse AD model did not completely align with those observed in human AD. This discrepancy might stem from various factors, including differences in the sampling sites, sampling methods, and potential batch effects in sequencing. Notably, an increase in the proportion of macrophages was a consistent feature across both human AD and all mouse AD models examined, indicating a common inflammatory response in AD pathology. However, the variations in the proportions of other cell types, such as VSMCs and fibroblasts, across the different models suggested that these changes could be model-specific or influenced by the unique pathology presented by each model. This highlighted the complex nature of AD and underscores the necessity of considering model-specific dynamics when extrapolating findings to human disease.

β-aminopropionitrile model and elastase model produced macrophage subtype profiles similar to human aortic dissection

Macrophages are a cell type with a large heterogeneity. To explore the changes in macrophage condition in the normal aorta and AD microenvironments, we redefined macrophage subpopulations and their markers. Macrophages from human and 5 mouse models were retrieved and re-clustered into 6 subpopulations using the top 10 dimensions (Figures 2A and 2B). The 6 macrophage subpopulations were manually annotated based on typically assessing the significant DEGs across subpopulations. Next, we identified 1 predominant macrophage subpopulation in the normal aorta: resident macrophages (FLOR2, MRC1, F13A1, and LYVE1) (Figure 2A). Another 2 major macrophage subpopulations were enriched in AD tissue: lipid-associated macrophages were identified since the high expression level of lipid transport-related genes (FABP5 and CD36); pro-inflammatory macrophages were named based on elevated inflammatory genes (IL1B, NLRP3, and S100A8) (Figure 2A). In addition, we defined 3 macrophages types, which explained the sources of macrophages. Proliferative Macrophages were identified for the expression of proliferative/cycling markers (Mki67, STMN1, and TOP2A); VSMCs−derived macrophages and fibroblasts−derived macrophages were characterized by their co-expression of VSMCs markers (TAGLN, ACTA2, and MYH11) and fibroblasts markers (PDGFRA, CD34, COL1A1), respectively.

Figure 2.

Characterization of macrophage subpopulations in human AD and multiple mouse AD models

(A) Dotplot shows the marker genes of each macrophage subtypes.

(B) UMAP plot shows all macrophage subclusters with cells colored according to the macrophage subtypes.

(C) Trajectory analysis of total macrophages.

(D) The macrophages M1/M2 polarization scores in human AD and each mouse AD model.

(E and F) Proportion analysis and UMAP represented the conserved macrophage subpopulations between human AD and each mouse AD model.

(G) Staining of lipid-associated macrophages in human AD and NA.

To elucidate the diverse origins and differentiation trajectory of macrophages implicated in AD, we utilized Monocle2 to analyze the developmental trajectories of various macrophage subpopulations (as depicted in Figure 2C). Our analysis discerned that pro-inflammatory and lipid-associated macrophages culminated at the terminal stages of these trajectories, suggesting these macrophages represent mature, functionally specialized forms in the AD microenvironment. Further examination of the trajectory bifurcation revealed two principal sources for these macrophages: one being a lineage derived from vascular smooth muscle cells (VSMCs-derived macrophages) and fibroblasts (fibroblasts-derived macrophages), and the other, a lineage characterized by a proliferative phenotype (proliferative macrophages). Gene set variation analysis (GSVA) provided insights into the functional profiles of these macrophage subsets. Specifically, pro-inflammatory macrophages were distinguished by their engagement in apoptosis signaling pathways, as well as in IL2-STAT5 and TNFα-NFκB signaling processes (illustrated in Figure S2A). This functional characterization suggests a role for pro-inflammatory macrophages in orchestrating the vascular inflammatory response, potentially contributing to the exacerbation of disease pathology through the promotion of cellular apoptosis and inflammation. Conversely, lipid-associated macrophages were notable for their capacity to synthesize and secrete a range of proteins (Figure S2A), including the proteinases (MMP19, ADAM9, ADAM10, ADAM17) and chemokines (CCL2, CCL3, CCL4, and so forth) (highlighted in Figures S2B and S2C). The secretion of these proteinases implicates lipid-associated macrophages in the remodeling of the extracellular matrix, a critical process in the progression of AD.

We assessed macrophage polarization status via scoring polarization markers using the “addmodulescore” function, and the violin plot indicated that macrophages in all mouse models lost M2 phenotype and gained M1 phenotype except for AngⅡ-induced apoE−/− mouse model (Figure 2D). Further, our integration analysis showed that the BAPN model, elastase model, and Ang-Ⅱ apoE−/− model produced similar macrophage subpopulations to human AD (Figures 2E and 2F). They all showed increased lipid-associated and pro-inflammatory macrophages. In contrast, the AngⅡ-induced model and CaCl2 model did not present the prevalence of lipid-associated macrophages (Figures 2E and 2F). To further clarify the role of LAM in AD, we co-stained CD68 and FABP5 on human AD and NA samples to label LAM, and determined that the amount of LAM increased in human AD (Figure 2G). We also co-stained CD68 and FABP5 on 5 mouse AD models to compare LAM differences in different AD models. We found that the BAPN model, elastase model, and Ang-Ⅱ apoE−/− model indicated enriched LAM in AD tissue, while the CaCl2 model showed moderate LAM, and the AngⅡ model showed rare LAM (Figure 3).

Figure 3.

Lipid-associated macrophages in the normal mouse aorta and five mouse models of aortic dissection

(A) Representative immunofluorescence images of aortic sections from control mice and five AD mouse models. Scale bars, 200 μm (low magnification) and 20 μm (high magnification).

(B) Quantification of FABP5⁺CD68⁺ macrophages normalized to total CD68⁺ macrophages across different models. Data are presented as mean ± SD. Statistical significance was assessed using one-way ANOVA with multiple comparisons. ∗∗∗∗p < 0.0001 and ∗∗p < 0.01; ns, not significant.

The angiotensin II apolipoprotein E−/− model has the most conserved vascular smooth muscle cell phenotype

To uncover transcriptome changes in VSMC subset, we integrated VSMCs from all human and mouse datasets and yielded 8 VSMCs subclusters. We identified myofibroblasts with different transcription patterns: SMOC2, PLN, VCAN, and OGN (Figures 4A and 4B). The contractile VSMCs were characterized by high expression level of ACTA2, MYH11, CNN1, MYL9, MYL6, ACTN4, and TAGLN. Oxidative stress-VSMCs possessed high transcription level of redox reaction genes (MT1X, ENO1, GLRX, SLC39A14, and SOD2). Lipid−associated VSMCs were identified by lipid transport-associated genes (APOE, FABP4, FABP5, and STEAP4).

Figure 4.

Characterization of VSMC subpopulations in human AD and multiple mouse AD models

(A) Dotplot shows the marker genes of each VSMC subtype.

(B) UMAP represented the distribution of all 4 VSMC subclusters.

(C) Trajectory analysis of total VSMCs.

(D) The contractile function score and synthetic function score of total VSMCs in human AD and each mouse AD model.

(E and F) Proportion analysis and UMAP represented the conserved VSMCs subpopulations between human AD and each mouse AD model.

(G) Staining of lipid-associated VSMCs in human AD and NA.

Previous research has illuminated that VSMCs, which exhibit a contractile phenotype in normal aorta, are capable of undergoing phenotypic switching to adopt alternative cellular identities, concomitantly losing their contractile capacities in diseased states. Therefore, in order to investigate the phenotypic switching direction of VSMCs in AD, the developmental trajectories of VSMCs subtypes were examined. Additional trajectory analysis showed a similar developmental trajectory for AD-associated smooth muscle cell subtypes, which fell into 2 pathways: from the fibroblasts or contractile VSMCs to LUM+myofibroblasts, oxidative stress-VSMCs, and SMOC2+myofibroblasts (Figure 4C). GSVA analysis was employed to decode the functional distinctions between these subgroups, revealing that oxidative stress-VSMCs exhibit reduced oxidative phosphorylation alongside enhanced glycolytic activity. Additionally, the activation of p53, TNFα, and apoptosis signaling pathways was observed in these subtypes (Figure S3A). Lipid−associated VSMCs represented a high level of chemokines (CCL2, CCL19, CCL21, and CXCL12) (Figure S3B). Contrastingly, SMOC2+myofibroblasts displayed pronounced myogenic functionality, albeit with lower glycolysis and autophagy levels. Intriguingly, our findings highlighted that both myofibroblasts and oxidative stress-VSMCs were subject to positive regulation by the GATA6 and ATF4 regulons, indicating a regulatory influence on their phenotypic states (Figure S3C).

To ascertain whether phenotypic switching of VSMCs manifested similarly across human and murine models of AD, we calculated the contractile score and synthetic score of all VSMCs. As a result, the AngⅡ model, the BAPN model, and the CaCl2 model all demonstrated elevated synthetic score and decreased contractile score (Figure 4D). However, the Ang-II apoE−/− model only observed a significant decrease in contractile score, but no increase in synthetic score, and the elastase model showed no significant change in both contractile score and synthetic score.

We compared the proportion of VSMCs subpopulations among human AD and AD models. The proportion of contractile VSMCs decreased in all AD models. The proportion of myofibroblasts increased in all models except for the Ang-II apoE−/− model. Only the Ang-II apoE−/− model showed transformation into lipid-associated VSMCs, which is consistent with lipid metabolism disorder caused by apoE deficiency (Figures 4E and 4F). Another pathogenic phenotype, oxidative stress-VSMCs, was also present in Ang-II infused apoE−/− mouse model and CaCl2 model (Figures 4E and 4F).

To further clarify the role of lipid-associated VSMCs in AD, we co-stained αSMA and FABP5 on human AD and NA samples, and found that a little of lipid-associated VSMCs existed in human AD but no lipid-associated VSMCs in NA (Figure 4G). We further co-stained αSMA and FABP5 on 5 mouse AD models to determine which mouse AD model could generate lipid-associated VSMCs (Figure 5). Consistent with the results of scRNAseq, only the Ang-Ⅱ apoE−/− model indicated enriched lipid-associated VSMCs in AD models. Thus, the Ang-II apoE−/− model is more comprehensive in mimicking VSMCs lesions, whereas other models are more focused on the pathogenic mechanisms of myofibroblasts.

Figure 5.

Lipid-associated VSMCs in the normal mouse aorta and five mouse models of aortic dissection

(A) Representative immunofluorescence images of aortic sections from control mice and five AD mouse models. Scale bars, 200 μm (low magnification) and 20 μm (high magnification).

(B) Quantification of FABP5⁺ VSMCs normalized to total αSMA⁺ VSMCs across different models. Data are presented as mean ± SD. Statistical significance was assessed using one-way ANOVA with multiple comparisons. ∗∗∗∗p < 0.0001; ns, not significant.

β-aminopropionitrile model best conserved fibroblasts in human aortic dissection

The integration of all fibroblasts revealed 7 subclusters, which were subsequently defined as 5 fibroblast subpopulations (Figures 6A and 6B). CD34+PDGFRA+Fibroblasts were annotated by the high expression of CD248, CD34, PDGFRA, and PDGFRB, which have powerful differentiation potential (Figure 6A). A subpopulation of fibroblasts was annotated as pro-inflammatory fibroblasts through high expression of inflammatory genes (CCL2, C1QBP, ADAMTS4, and ENO1). Lipid−associated fibroblast was characterized by genes involved in lipid metabolism (EBF2, ANGPTL7, and APOD). Similarly, myofibroblasts also constituted a subset of fibroblasts with specific transcriptome signatures (FAP, MYL9, TAGLN, and ACTA2). Of note, Neuron was identified via specific markers of neurons (NRXN1, S100B, PLP1, and MPZ). CD34+PDGFRA+ Fibroblasts account for 80% of total fibroblasts in the normal aorta. In aortic dissection, this percentage decreased, with pro-inflammatory fibroblasts, pro-inflammatory Fibroblasts, and myofibroblasts occupying a greater percentage (Figure 6E).

Figure 6.

Characterization of fibroblast subpopulations in human AD and multiple mouse AD models

(A) Dotplot shows the marker genes of each fibroblast subtype.

(B) UMAP plot shows all 4 fibroblast subtypes and neurons.

(C) Trajectory analysis of the total fibroblast.

(D) The inflammatory score and ECM synthetic score of fibroblasts in human AD and each mouse AD model.

(H–I) Proportion analysis and UMAP represented the conserved fibroblasts subpopulations between human AD and each mouse AD model.

(G) Staining of PDGFRA+αSMA+myofibroblasts in human AD and NA.

The trajectory analysis elucidated that CD34+PDGFRA+Fibroblasts constituted the origin, diverging across developmental pathways toward distinct terminal stages (Figure 6C). Within this developmental trajectory, Lipid-associated fibroblasts were positioned at an intermediate stage, whereas myofibroblasts and pro-inflammatory fibroblasts were delineated as termini. Furthermore, gene set variation analysis (GSVA) provides insights into the functional dynamics of cell populations, revealing that the differentiation of cells into myofibroblasts and pro-inflammatory fibroblasts is concomitant with the process of epithelial-mesenchymal transition (EMT) (Figure S4A). Pro-inflammatory fibroblasts potentially intensify the inflammatory milieu of aortic dissection (AD) through the secretion of a specific array of chemokines, including CCL2, CCL19, CCL26, CXCL2, CXCL12, and CXCL19 (Figure S4B). These molecules may play a pivotal role in recruiting immune cells to the inflammatory site, thereby exacerbating the disease process. This association underscores the phenotypic plasticity of these cells and their likely contribution to the remodeling of the aortic wall. Additionally, the Single-cell regulatory network inference and clustering (SCENIC) analysis shed light on the transcriptional regulation underlying the generation of myofibroblasts and pro-inflammatory fibroblasts (Figure S4C). Notably, regulons such as ENO1, FOSL1, and HIF1A were identified as key regulatory elements.

Regarding the inflammatory state of fibroblasts, all AD models showed elevated inflammatory scores, among which the elastase model and the BAPN model presented the most significant increase. In terms of ECM remodeling, only the BAPN model and the elastase model demonstrated reduced ECM synthesis capacity (Figure 6D). Across all models examined, PDGFRA+CD34⁺ fibroblasts were ubiquitously present (Figures 6E and 6F). Notably, Lipid-associated fibroblasts were infrequently identified across all models, denoting their rare occurrence. Regarding pro-inflammatory fibroblasts, their representation was distributed in all models (Figures 6E and 6F). On the other hand, no model significantly stimulated fibroblasts transformed into myofibroblasts (Figures 6E and 6F). However, the staining in human AD samples also did not observe αSMA+PDGFRA+ myofibroblasts (Figure 6G). Overall, we believe that BAPN best conforms to the characteristics of fibroblasts in human AD.

All models demonstrated a similar pathogenic mechanism in endothelial cells

The role of ECs in AD is being gradually uncovered, particularly the disruption of endothelial cell tight junctions, but it is not clear in which states ECs are present in AD. We re-clustered ECs in human AD into 12 subclusters, and annotated each cluster into 7 EC subpopulations in accordance with their marker genes (Figures 7A and 7B), referring to published literature.^16,17 ECs highly expressing ACKR1was named as venous ECs (Ven ECs); ECs highly expressing PROX1 and LYVE1 were named as lymphatic ECs (LymECs); ECs highly expressing SEMA3G, HEY1, and IGFBP3 were named as artery ECs (ArtECs); ECs highly expressing CA4, CD36, and RGCC were named as capillary ECs (CapECs) (Figure 7A). Similarly, pro-inflammatory venous ECs (pro-inflam VenECs) were named as high expression of IL1B, CCL3, S100A9 (Figure 7A). Hypoxia venous EC was characterized by hypoxia-related genes, including HIF1A, LDHB, LDHA, and so forth (Figure 7A). The ECM-secreting venous EC subpopulation shows high levels of collagen gene (COL1A1, COL1A2, COL3A1) (Figure 7A).

Figure 7.

Characterization of ECs subpopulations in human AD and multiple mouse AD models

(A) Dotplot shows the marker genes of each fibroblast subtype.

(B) UMAP projection of 5 EC subpopulation.

(C) Tight junction score of total EC in human AD and each mouse AD model.

(D and E) UMAP represented the conserved EC subpopulations between human AD and each mouse AD model.

(F) Staining of lipid-associated ECs in human AD and NA.

We further calculated the tight junction scores of each EC, and found that the tight junction scores in the human, Ang-II, elastase, and BAPN models were reduced compared to those in the normal aorta, whereas the score was increased in the CaCl2 model and Ang-II apoE−/− model (Figure 7C). In human AD, hypoxia VenECs were more produced in human AD, but ArtECs and LymECs were reduced (Figure 7D). Thus, reduced EC adhesion function and the genesis of hypoxia in venous ECs may be responsible for the onset and progression of AD.

We subsequently compared the differences in the distribution of each EC subgroup in AD and the normal aorta. Hypoxia VenECs were upregulated in all mouse models (Figures 7D and 7E). Of note, we observe an increase of pro-inflammatory VenEC in the BAPN model and the Ang-II apoE−/− model (Figures 7D and 7E). Moreover, we noticed that CapECs highly express lipid-associated genes such as FABP5. The IF staining revealed no difference in FABP5+EC, which is mainly located in the adventitia of the aorta (Figure 7F). Therefore, we speculate that CapECs do not participate in the occurrence and development of AD. All the specific markers for the above-mentioned cell subpopulations are summarized in Table S4. The strength of each mouse models were summarized in Table 1.

Table 1.

Summary of characterization of human AD and mouse AD models

Cell subpopulations	Human AD	AngII model	AngII+ ApoE−/− mice	Mouse AD model
				BAPN+ WT mice	Elastase model	CaCl2 model
Mφ
M1 score	up	up	down	up	up	up
M2 score	down	down	down	down	down	down
Pro−inflam Mφ	✓✓	✓✓	✓	✓✓	✓✓	✓✓
Lipid−associated Mφ	✓✓	✓	✓	✓	✓	✓
Fibroblast
Lipid−associated Fibro	✓
Pro−inflam Fibro	✓✓	✓	✓	✓	✓	✓✓
VSMC
contractile score	down	down	down	down	down	up
synthetic score	slightly up	up	down	up	up	up
Myofibro	✓✓	✓✓	✓✓	✓✓	✓✓	✓✓
Oxidative_stress_VSMC	✓✓		✓
Lipid−associated VSMC	✓✓		✓✓
EC
Tight junction score	down	down	Up	down	down	down
Hypoxia VenEC	✓✓	✓✓	✓✓	✓✓	✓✓	✓✓
Pro-inflam VenEC	✓	✓	✓	✓	✓
Programmed death
Necroptosis (Mφ)	up	Up	up	up	up	up
Pyroptosis (Mφ)	up	Up	down	up	up	up
Apoptosis (Mφ)	up	Up	down	up	up	up
Ferroptosis (Mφ, VSMC)	up	up	up	down	up	up

The proportions of each subtype of macrophages, fibroblasts, VSMC, and EC in each model were compared. ✓✓ > 10%, ✓1%–10%.

Programmed death and cell-cell communication analysis

Progressive VSMC loss is a key feature of aortic dissection, and programmed VSMC death may be a major contributor to VSMC loss. Similarly, endothelial cell injury and death may promote aortic dissection. Published literature reported that necroptosis, pyroptosis, apoptosis, and ferroptosis are associated with aortic disease, so we assessed whether mouse models would also induce these programmed cell deaths (Figure S5). Necroptosis was significantly elevated in monocytes/macrophages in human AD, and we also observed this alteration in macrophages across all models. Pyroptosis was significantly elevated in macrophages in human AD, consistent with all mouse models. Specifically, the pyroptosis score of cycling cells was significantly elevated in the elastase model, CaCl2 model, and Ang-II apoE−/− model. Apoptosis was also significantly elevated in macrophages in human A, and all models mimicked this mechanism. Specifically, the Ang-II model and CaCl2 model enhanced apoptosis scores in T cells. Ferroptosis in human AD was markedly elevated in macrophages and VSMCs. All models except BAPN elevated the ferroptosis score in macrophages and VSMCs. Other models also showed up-regulation of ferroptosis score in ECs, fibroblasts, and adipocytes.

We subsequently identified disease-associated subpopulations from macrophages, VSMCs, ECs, and fibroblasts, and performed cell-cell communication analysis using the CellChat package (Figure S6). In human AD, all subpopulations except myofibroblasts exhibited strong intercellular communication, with proinflammatory ECs showing the highest activity. In the Ang-II model, proinflammatory fibroblasts and myofibroblasts were the dominant mediators of intercellular signaling, whereas proinflammatory ECs and lipid-associated VSMCs displayed minimal communication activity. In both the Ang-II apoE−/− model and the BAPN model, proinflammatory fibroblasts again mediated robust communication. Notably, lipid-associated VSMCs in the Ang-II apoE−/− model showed significantly stronger communication capacity compared with those in the Ang-II and BAPN models. In the CaCl₂ model, lipid-associated macrophages were the most active communicating subpopulation, whereas in the elastase model, hypoxia-associated venous ECs represented the major signaling hub.

Comparison between angiotensin II model and angiotensin II apolipoprotein E−/− models

The distinction between the Ang-II model and the Ang-II apoE^−/− model is particularly noteworthy. The Ang-II model reflects AD driven primarily by hypertension, whereas the Ang-II apoE^−/− model simulates AD induced by combined hypertension and hyperlipidemia/atherosclerosis. Given the observed differences in the abundance of lipid-associated VSMCs and lipid-associated macrophages between these two models, we further analyzed transcriptional and functional differences in macrophages and VSMCs. GO enrichment analysis of differentially expressed genes revealed that macrophages in the Ang-II model were predominantly enriched in extracellular matrix (ECM) remodeling pathways, with high expression of ECM-related genes such as MMP2 and COL6A1. In contrast, macrophages in the Ang-II apoE^−/− model were strongly associated with leukocyte migration and exhibited elevated expression of inflammatory genes (IL1A and IL6) and lipid metabolism genes such as LDLR (Figures S7A–S7C). For VSMCs, the Ang-II model showed upregulation of smooth muscle development pathways and increased expression of ECM-remodeling genes such as FBLN2 and COL1A2. Conversely, VSMCs in the Ang-II apoE^−/− model displayed enhanced activation of lipid metabolic pathways and high expression of ABCC9, FABP4, and related genes (Figures S7D–S7F).

Discussion

Multiple basic studies have demonstrated that AD is associated with a variety of cells in a pathogenic state; however, these studies are based on diverse AD mouse models. Mechanisms found in BAPN-induced AD models may not be present in Ang II-induced models. Macrophage, VSMCs, Fibroblast, and EC are the major components of the aorta and are also closely associated with the pathogenesis of AD. Therefore, from the perspective of heterogeneity of these four cell types, we revealed the mechanism differences between human AD and various mouse AD models to provide a reference for model selection.

In the present study, macrophages were categorized into six subpopulations. Macrophages used to be categorized as either classically activated macrophages (M1) or alternatively activated macrophages (M2), associated with pro- and anti-inflammatory, respectively. However, the realistic phenotype of macrophages in the Human body well surpasses the simple categorization of M1/M2.¹⁸ Tissue resident macrophages (TRMs) originates from the primary hematopoiesis and transient definitive hematopoiesis of the endothelium of the yolk sac and the subsequent definitive hematopoiesis of the endothelium of the aorta-gonad-mesonephros (AGM) region.¹⁹ TRM acts as a receptor and effector of tissue homeostasis and is responsible for the recycling of extracellular matrix to promote tissue remodeling and regeneration.²⁰ The Identification of resident macrophages in scRNA-seq typically requires markers FOLR2, PF4, F13A1, LYVE1, and MRC1.^10,21 TRM located in the intima prevents intravascular coagulation by effectively removing fibrinogen and regulating thrombin activity in areas with disturbed blood flow.²² We found that TRM is conserved in the aorta across species as a protector of the aortic extracellular matrix. TRM accounted for a higher proportion in the normal aorta and a decreased proportion in AD, which may be attributed to the infiltration of a large number of bone marrow-derived macrophages into the aortic wall. Pro-inflammatory macrophages are characterized by the inflammatory factors IL1B, NLRP, S100A8, S100A9, which are similar to the markers named inflammatory macrophages in Frank et al. and Cheng et al.^11,23 Pro-inflammatory macrophages were present in all mouse AD models, but only a low count in the Ang-Ⅱ model. This may be attributed to Ang-Ⅱ predominantly inducing VSMCs lesions. In addition to this, we identified two macrophage subpopulations expressing fibroblast markers and VSMC markers, respectively, suggesting that VSMCs and fibroblasts possess the potential to differentiate into macrophage-like cells in AD in addition to resident macrophages and monocyte-derived macrophages. MKI67 and STMN1 are markers for cells with high proliferative capacity,^24,25 and we found that proliferative macrophages are present in all mouse models of AD, and it may be an essential source of macrophages in AD. LAM was present in human AD, BAPN model, Ang-II apoE−/− model, and elastase model. This suggests the presence of lipid metabolism disorders in these three models.

In response to injury or stress, vascular adventitial fibroblasts undergo phenotypic switching into myofibroblasts, which co-express the smooth muscle marker αSMA as well as fibroblast markers. Similarly, VSMCs in media undergo phenotypic switching to form myofibroblasts. Myofibroblast development has been identified as a key pathological mechanism in aortic aneurysms.²⁶ However, we did not observe the transformation of fibroblasts to myofibroblasts in AD, and myofibroblasts are more likely to originate from VSMCs. The principal functions of myofibroblasts are the synthesis of extracellular matrix (ECM), facilitating angiogenesis and the secretion of pro-inflammatory factors, as well as providing some contractile force.²⁷ We found that myofibroblasts synthesize type I collagen more weakly than fibroblasts, but synthesize CCL2 and CCL19 more strongly. When compared with contractile VSMCs, myofibroblasts had an enhanced expression level of MMP2, CCL2, and CXCL12. Thus, myofibroblasts lead to AD progression through ECM remodeling and mediating inflammatory responses. Myofibroblasts have the highest proportion in the Ang-Ⅱ model and elastase model, which is related to the fact that Ang-Ⅱand elastase induce VSMCs phenotypic switching.

Both aortic aneurysm and AD are associated with ECs dysfunction.^28,29,30 EC dysfunction contributes to the disruption of inter-endothelial cell junctions and therefore promotes the infiltration of immune cells.³¹ We named ECs with high expression of adhesion molecules as adhesion ECs, and it is apparently the most important EC subpopulation for maintaining the normal function of the aortic endothelial barrier. The Ang II model and the BAPN model had the lowest tight junction score of ECs, suggesting adequate disruption of the endothelial barrier. We found that hypoxia ECs accounted for a large proportion in human AD and a high proportion in all models, suggesting a common mechanism.

The controversy between the use of wild-type mice or apoE−/− mice for AD models arises from the fact that it remains unclear whether lipid metabolism is involved in AD pathogenesis. Although we defined lipid-associated subpopulations in macrophages, fibroblasts, and VSMCs, the causal relationship between them and AD needs to be further investigated. Lipid-associated macrophages constitute a high proportion of total macrophages in human AD, and they are present in all five mouse models but are underrepresented in Ang II-induced entrapment models. Macrophages associated with lipid metabolism in atherosclerosis are named as foam cells (expressing ABCA1, ABCG1, APOC1, and FABP5).^32,33 Our identified lipid-associated macrophages expressing FABP5 and CD36, which may be associated with atherosclerosis on the aorta. Lipid-associated Fibro expresses EBF2, APOD, and ANGPTL7, whereas this subpopulation is absent in all mouse models. EBF2, a transcription factor mediating terminal maturation of brown adipocytes, may mark the differentiation of fibroblasts to adipocytes.³⁴ Lipid-associated VSMCs were present in both normal aorta and AD with no significant difference in proportion; among the five models, only the Ang-II apoE−/− model displayed a higher proportion of lipid-associated VSMCs, which is consistent with the susceptibility of the apoE−/− mouse to atherosclerosis. We found that lipid-associated VSMCs highly expressed CCL2, CXCL12, CCL19, and CCL21, suggesting that it may promote AD progression through an inflammatory response. A Mendelian randomization study noted that high plasma lipid levels were positively associated with AA risk, while there was no causal relationship between elevated lipid levels and AD risk.³⁵ Similarly, a 26-year observational community-based study found that HDL cholesterol, LDL cholesterol, and plasma triglycerides were associated with mortality in patients with AA, but not in patients with AD.³⁶ Since AD precedes atherosclerotic lesion formation in apoE-deficient mice infused with angiotensin II,³⁷ the role of lipid-related pathogenesis in AD needs to be further elucidated, even though the Ang-II apoE−/− model more readily mimics lipid-related pathogenesis.

Surprisingly, lipid-associated macrophages were also present in the BAPN model. Since BAPN acts as a LOX inhibitor, it disrupts the structural integrity of the arterial wall by inhibiting the cross-linking of collagen and elastin, leading to AD development. The earliest studies found that reduced LOX activity in an obese mouse model resulted in aortic ECM remodeling. BAPN has been reported to stimulate the trans-differentiation of white adipocytes into beige adipocytes by inhibiting LOX.³⁸ Thus, BAPN may lead to the generation of lipid-associated macrophages by exerting an influence on lipid metabolism.

Programmed cell death is central to the pathogenesis of aortic dissection. Excess apoptosis and ferroptosis of vascular smooth muscle cells weaken the medial layer, disrupt extracellular matrix homeostasis, and favor intimal tearing and false lumen formation.³⁹ Clarifying the differences in programmed cell death across different AD models can assist researchers in selecting the appropriate model for their specific field of study. We compared the programmed death in five models. Necroptosis, pyroptosis, and apoptosis mainly occurred in macrophages, and ferroptosis was involved in macrophages and VSMCs. Interestingly, we found that pyroptosis, apoptosis, and ferroptosis decreased in the macrophages of the Ang-II apoE−/− mouse model. In summary, we identified key subpopulations of fibroblasts, VSMCs, and macrophages and deduced their potential functions. We used these subpopulations as a basis for comparing the mechanisms of human AD and the five mouse AD models.

Altogether, this study provides a single-cell-based comparative reference for examining transcriptional similarities between human aortic dissection and commonly used mouse models. Rather than establishing mechanistic equivalence, our analyses highlight that different models preferentially capture distinct pathogenic transcriptional states across major vascular cell types. Specifically, macrophage- and fibroblast-associated transcriptional programs show the greatest similarity in the BAPN model, pathogenic VSMC states are most prominently represented in the Ang-II apoE^−/− model, and endothelial dysfunction-related signatures are observed across all models. Importantly, these similarities are identified at the transcriptomic level and may reflect shared stress or injury responses induced by distinct experimental triggers. Beyond model comparison, our study delineates key disease-associated subpopulations of VSMCs, macrophages, and fibroblasts, providing a resource for future mechanistic and functional investigations of aortic dissection.

Limitations of the study

Admittedly, our study has some weaknesses. That is, our integrated data does not originate from a single platform. Thus, the sampling methods, time points, and batch effects may bias the results. Second, our conclusion is that the constrained number of biological replicates is due to high sequencing costs. This limits the robust statistical power for differential abundance testing of cell populations across conditions. Additionally, we have verified that the pathogenic subpopulations, such as lipid-associated macrophage and pro-inflammatory fibroblasts, were consistently present across the integrated dataset that includes both our in-house samples and the independent public cohort (GSE213740). While our study provides a detailed transcriptional map and infers functional potentials (e.g., inflammatory cytokine secretion and ECM remodeling) of pathogenic cell states, it is limited by the lack of direct functional validation for these specific pathways. Additionally, the variability in disease stage and patient comorbidities (such as hypertension and diabetes) can not be considered in our study. Furthermore, the time and speed required to induce AD in different mouse models vary, which may lead to inconsistencies in the onset stage of AD and thereby mislead the results. Of course, there are established standards for the intervention time for each model, and these differences also need to be taken into account by every researcher. Our pseudotime analysis suggests VSMCs and fibroblasts may give rise to specific macrophage subpopulations. However, this computational inference, while plausible given known cellular plasticity, requires validation by future lineage-tracing studies. Importantly, our cross-species comparison underscores that the five commonly used AD mouse models should not be viewed as competing or complete replicas of the human condition. Instead, each model serves as a powerful tool that illuminates a distinct facet of AD pathophysiology.

A key methodological limitation of this study lies in the identification of DEGs from integrated scRNA-seq datasets. Although we applied Seurat’s standard integration workflow to mitigate batch effects at the embedding and clustering levels, DEGs were identified using the FindAllMarkers function on the integrated expression matrix. This approach, while widely adopted in the field, does not explicitly adjust for dataset-of-origin or inter-study technical variation, and may therefore inflate false-positive DEGs in multi-source datasets.^40,41 Accordingly, DEG results in this study should be interpreted with caution and primarily viewed as indicative markers supporting cell-state characterization rather than definitive gene-level discoveries.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Lixin Wang (wang.lixin@zs-hospital.sh.cn).

Materials availability

All materials will be shared by the lead contact upon request.

Data and code availability

• •All data reported in this article will be shared by the lead contact upon request
• •This article does not report original code.
• •Other items will be shared by the lead contact upon request.

Acknowledgments

We gratefully acknowledge the Gene Expression Omnibus (GEO) database for providing access to the single-cell RNA sequencing datasets used in this study. We also thank the researchers who shared these datasets, whose contributions made this work possible.

The authors acknowledge funding from the National Science Foundation of China (grant number: 82270415), the Shanghai Municipal Science and Technology Commission Fund (grant numbers: 20234Z00120 and 22S31904800), and the Shanghai Municipal Health Commission (NO: 20214Y0474).

Author contributions

G.C., S.Q., L.W., and W.F. designed the study. G.C., S.L., Z.L., and C.H. performed data analysis. Data interpretation: G.C., Y.H., S.L., Q.F., Z.L., and L.W. conducted data analysis. Article writing: G.C., S.Q., C.H., and S.L. wrote the article. G.C., S.Q., Z.L., C.H., Y.H., W.F., and L.W. are responsible for article preparation and data verification.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
α-SMA	Abcam	RRID: AB_11129103
CD68	Servicebio	RRID: AB_2935658
FABP5	Protein-tech	RRID: AB_2100341
CD31	Abcam	RRID: AB_2920881
PDGFRA	Abcam	RRID: AB_2892065
Biological samples
Aorta samples	Zhongshan hospital of Fudan university	N/A
scRNA-seq data
GSE254132	GEO database	https://www.ncbi.nlm.nih.gov/geo/
GSE213740	GEO database	https://www.ncbi.nlm.nih.gov/geo/
GSE213735	GEO database	https://www.ncbi.nlm.nih.gov/geo/
GSE203594	GEO database	https://www.ncbi.nlm.nih.gov/geo/
GSE152583	GEO database	https://www.ncbi.nlm.nih.gov/geo/
GSE164678	GEO database	https://www.ncbi.nlm.nih.gov/geo/
CRA003013	Genome sequence archive database	https://ngdc.cncb.ac.cn/gsa/

Experimental model and study participant details

Study approval

All human research was conducted in accordance with official ethical guidelines and approved by the Ethics Committee of the Zhong Shan Hospital of Fudan University (Approval NO: B2019-231R). The investigation conforms to the principles outlined in the Declaration of Helsinki. The patients provided their written informed consent before they participate in this study. Three patients with AD and 3 patients received heart transplant donated their AD samples or normal thoracic aorta samples. The animal experiment has been approved by animal ethics committee of the Zhong Shan Hospital of Fudan University (No.2024-035).

Human participants

Patient characteristics of sequencing data included in this study were summarized in Table S1. Patient characteristics of aorta samples (1 normal aorta and 1 aortic dissection) used for IF staining were summarized in Table S2.

Animals

BAPN models

The BAPN (β-aminopropionitrile)-induced AD model was established by administering 3-Aminopropionitrile fumarate salt (A3134, Sigma-Aldrich) to 3-week-old male C57BL/6J mice (SM-001) purchased from Shanghai Model Organisms Center, Inc. BAPN was dissolved in drinking water at a concentration of 0.5 g/L for 4 weeks.

Ang-II model and Ang-II apoE−/− model

The Ang-II-induced AD model and Ang-II-induced apoE−/− mice AD model were created using 12-week-old male C57BL/6 mice (SM-001) purchased from Shanghai Model Organisms Center, Inc., or 12-week-old male apoE−/− mice (NM-KO-190565) purchased from Shanghai Model Organisms Center, Inc. Ang-II (A9525, Sigma-Aldrich) was delivered via osmotic minipumps (1004, Alzet) implanted subcutaneously. Ang-II was infused at a rate of 1500 ng/kg/min for 28 days. The minipumps were primed according to the manufacturer’s instructions and surgically implanted under anesthesia.

Elastase model and CaCl2 model

8-week-old C57BL/6 male mice (SM-001) purchased from Shanghai Model Organisms Center, Inc., were anesthetized by ether inhalation. A ventral midline incision was made through the abdominal skin and muscles. The colon and intestine were exposed and gently moved to the right abdomen region and covered with 0.9% saline-soaked gauze. The abdominal aorta was subjected to adventitial elastase infiltration (10 mg/mL) or adventitial CaCl2 (0.5 mol/L) infiltration. We established these AD mouse model for experimental verification.

Method details

Single-cell RNA sequencing

For human AD scRNA-seq datasets, we performed scRNA-seq in aorta samples from AD 3 patients and 3 from heart transplant donors in our center, which has been deposited in Gene Expression Omnibus database (GSE254132); another scRNA-seq dataset of human AD GSE213740 (comprising 6 samples from AD patients and 3 from heart transplant donors) were obtained from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo).

For mouse AD scRNA-seq datasets, ang-Ⅱ induced mouse model GSE213735, ang-Ⅱ induced apoE−/− mouse model GSE203594, elastase induced mouse model GSE152583, and CaCl2 induced mouse model GSE164678 were obtained from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo); mouse AD model induced by BAPN CRA003013 were obtained from Genome sequence archive database (https://ngdc.cncb.ac.cn/gsa/) (Figure 1B).

All samples were prepared as single-cell suspensions using digest solution. The digest solution was mixed with DMEM+1 mg/ml collagenase Ⅰ+1 mg/ml hyaluronidase+20U/ml DNAse Ⅰ. Aorta tissue samples were rinsed one or two times with pre-cooled PBS, then minced. The minced tissue was transferred to the digestion solution and incubated in a 37°C water bath for 30–90 min of digestion. Cell viability and counting under microscope with trypan blue detection, samples with >85% viability were sequenced.

Single-cell RNA sequencing libraries were prepared using the 10X Genomics Chromium Single Cell 3′ GEM, Library & Gel Bead Kit v3.1 (Catalog No. 1000121). Single-cell suspensions, 10X barcode gel beads, and oil were loaded into a Chromium Chip G and processed on the 10X Genomics Chromium Controller to generate Gel Bead-in-Emulsions (GEMs). Within GEMs, poly-adenylated RNA from individual cells was reverse-transcribed into barcoded cDNA. Purified cDNA was amplified via PCR. The quality and concentration of the amplified cDNA were assessed using an Agilent 2100 Bioanalyzer (peak length: ∼1057–1079 bp) and a Qubit 4.0 Fluorometer (concentration: 3.96–6.34 ng/μL).The cDNA was fragmented, end-repaired, A-tailed, and ligated with adapters and sample indexes (Chromium i7 Multiplex Kit) to construct the final sequencing libraries. The libraries were purified using SPRIselect beads. The finished libraries were quantified using a Qubit fluorometer (concentration: 27–37.8 ng/μL) and their size distribution was confirmed on an Agilent 2100 Bioanalyzer (peak length: ∼453–455 bp). The libraries were sequenced on an Illumina platform using paired-end 150 bp (PE150) sequencing mode. The protocol aimed for a target cell recovery of 10,000 cells per sample during the GEM generation step on the Chromium Controller.

Cell type annotation and sub-clustering analysis

Differentially Expressed Genes (DEGs) were calculated for each cell population using the FindAllMarkers function with the Wilcoxon test (min.pct = 0.25, logfc.threshold = 0.25). The annotation of each cell subpopulation was performed according to DEGs and classical major vascular cell markers in the literature. Dotplot with colors representing expression levels and sizes representing percentage of expression was used to indicate the accuracy of cell annotation. For sub-clustering, cells from the same major cell type were retrieved. Following Data normalization, data scaling, and principal component analysis, cells from the same major cell type were re-clustered using FindClusters function. DEGs among subclusters were calculated for annotation (Data S2).

Trajectory analysis and cell-cell communication analysis for single-cell RNA-Seq

For revealing the developmental trajectory of macrophages, VSMCs, and fibroblasts in AD, R package Monocle2 (v2.24.1) was applied to conduct single-cell trajectory analysis.⁴² Official analysis pipeline was implemented. Gene-cell matrix of each major cell type and its annotation information was used to construct dataset. Top 2000 most variable genes were used for cell ordering. When “reduceDimension” function is used, the reduction method is “DDRTree” and the value of “max_components” is 2. Gene changed with pseudo-time were calculated by the “differentialGeneTest” function. the branch expression analysis modeling (BEAM) analysis were performed for screening genes which determine cells differentiate into distinct branch.

Cell–cell communication was inferred using CellChat (v2.0.0) in R. After preprocessing the single-cell expression matrix, we created a CellChat object and identified overexpressed genes and ligand–receptor interactions based on the built-in human/mouse ligand–receptor database. We then computed communication probabilities using the default mass-action model and filtered interactions by minimum cell proportion.

Integration of human datasets and mouse datasets

As a strategy to compare macrophages, VSMCs, fibroblasts, and ECs in mouse and human AD, cross-species integration of scRNA-seq data was performed. Major cell types were annotated for all single-cell datasets and major cell type of macrophages, VSMCs, and fibroblasts were extracted. Gene symbols from the mouse single-cell dataset were converted to corresponding human homologs using BioMart-Ensembl database.⁴³ Since not all mouse genes have human homologs, this leads to a reduction in the number of genes after conversion of mouse single-cell data. However, most marker genes of subtypes of macrophages, VSMCs, and fibroblasts were conserved between human and mouse. The number of genes before and after moue-to-human Homologous gene conversion using BioMart was shown in Table S3. The two human AD scRNA-seq datasets (our in-house data and GSE213740) were integrated using the standard Seurat integration workflow. Briefly, datasets were normalized independently using the SCTransform function with regression of mitochondrial percentage. The top 3000 integration anchors were identified using the SelectIntegrationFeatures and FindIntegrationAnchors functions. The datasets were then integrated using the IntegrateData function, which creates a combined assay where the expression values have been corrected for technical batch effects, allowing the cells to be co-embedded based on biological similarity rather than dataset origin. All downstream analysis, including clustering and UMAP visualization, was performed on this integrated and batch-corrected data.

Histological analysis

Immunofluorescence staining was performed to detect the specific markers of defined cell subpopulations in the human/mouse aorta. Briefly, arterial sections were dewaxed and hydrated for antigen repair and 3% BSA closure, and then incubated with primary antibodies against α-SMA (1:1000, ab124964, Abcam), CD68 (1:500, GB113109, Servicebio), FABP5 (1:500, 12348-1-AP, Protein-tech), CD31 (1:2000, ab182981, Abcam), PDGFRA (1:500, ab203491, Abcam) overnight at 4°C. The sections were rinsed with PBS and incubated with fluorescent dye-coupled secondary antibody (G1231, Servicebio) for 2 h at 37°C. The cell nuclei were counterstained with DAPI. Finally, the sections were sealed using an anti-fluorescence quenching sealer.

Analysis of programmed cell death pathways

Gene signature scores for four major programmed cell death pathways—ferroptosis, apoptosis, pyroptosis, and necroptosis—were calculated to assess the relative activation of these molecular programs across different cell types. Gene lists of programmed cell death pathways were obtained from molecular signatures database (https://www.gsea-msigdb.org/gsea/msigdb/) (Data S1). The activity score for each cell death pathway in individual cells was computed using the AddModuleScore function in Seurat.

Quantification and statistical analysis

R package Seurat (version 4.3.0) was used for general analysis for all scRNA-seq dataset with default parameters. Datasets were imported into Seurat objects using Read10X function and CreateSeuratObject function. Unqualified cells were filtered using following parameters: (<200 transcripts/cell, >5000 transcripts/cell, >10% mitochondrial genes, >5% hemoglobin genes). Cells with potential doublets were identified and removed using ‘DoubletFinder’ package. The top 2000 highly variable genes (HVGs) were generated from the normalized expression matrix and principal component analysis (PCA) was performed based on these HVGs. Through visualizing the standard deviations of the top 40 principle components using Jackstraw analysis, components 1 to 20 were employed to cluster distinct group of cells. Results of cell clustering are presented using uniform manifold approximation and projection (UMAP).

One-way ANOVA was performed for multiple comparisons. ∗∗∗∗p < 0.0001, ∗∗p < 0.01; ns, not significant.

Published: February 25, 2026