Single-cell data quality control, processing and analysis
The Seurat package (v.3.2.2)(70, 71) was applied for quality control, processing, and analysis. Each Seurat object was created with genes that were expressed in more than three cells. Quality control conditions were set as: (1) only genes within 200 and 3000; (2) the percentage of mitochondrial genes less than 20% were included for downstream analysis(Figure S5a) . After QC, the remaining cells were used for further analysis, as shown in Table S2 . A total of 29942 cells were included in the computational analysis. Among them, the case group (TA) had 8966 cells and the healthy control group (HC) had 20976 cells.
The CellcycleScoring function was applied to mitigate the possible influence of cell cycles. SCTransform was adopted by default to reduce potential batch effects or technical variations. PCA was set at 1:20, and then unsupervised cell clustering was performed (Figure S5b) .
Find All Markers by default methods was used to obtain the DEGs specifically in each cluster, and the representative markers (genes with the highest avg_logFC and adjusted p value < 0.05) were then chosen for cluster labeling. DEGs between TA samples and healthy controls in each cluster were identified with function Find Markers.
In order to count the differences in cell composition between the TA group and the HC group, we used the χ2 analysis method in SPSS 20.0 to analyze the differences in the composition ratios of various cell types, and cells with higher composition ratios in the TA group were used.