Single-cell profiling reveals comparatively inflammatory polarization of human carotid versus femoral plaque leukocytes

45 46 Rationale: Femoral atherosclerotic plaques are less inflammatory than carotid plaques 47 histologically, but limited cell-level data exist regarding comparative immune landscapes 48 and polarization at these sites. 49 50 Objectives: We investigated intraplaque leukocyte phenotypes and transcriptional 51 polarization in 49 total patients undergoing femoral (N=23) or carotid (N=26) 52 endarterectomy using single-cell ribonucleic acid sequencing (scRNA-seq; N=13), flow 53 cytometry (N=24), and immunohistochemistry (N=12). 54 55 Findings: Comparative scRNA-seq of CD45 positive-selected leukocytes from femoral 56 (N=9; 35265 cells) and carotid (N=4; 30655 cells) plaque revealed distinct 57 transcriptional profiles. Inflammatory foam cell-like macrophages and monocytes 58 comprised 2.5to 4-fold higher proportions of myeloid cells in carotid plaques, whereas 59 non-inflammatory foam cell-like macrophages and LYVE1-overexpressing resident-like 60 macrophages comprised 3.5to 9-fold higher proportions of myeloid cells in femoral 61 plaque (p<0.001 for all). A significant comparative excess of CCR2 macrophages in 62 carotid versus femoral plaque was observed by flow cytometry in a separate validation cohort. B 63 cells were more prevalent and exhibited a comparatively anti-inflammatory profile in 64 femoral plaque, whereas cytotoxic CD8+ T cells were more prevalent in carotid plaque. 65


Vascular inflammation is a hallmark of atherosclerosis and is broadly
To determine whether the general differences in monocyte/macrophage and T cell 243 transcriptional phenotypes corresponded to changes in cell types based on canonical 244 protein markers, we prospectively enrolled an additional 24 patients undergoing femoral 245 (N=9) or carotid (N=15) endarterectomy for flow cytometry analysis of plaque and paired 246 blood (see Table 2 and Methods). Patient clinical and demographic characteristics were 247 similar between those undergoing femoral vs. carotid endarterectomy -including similar 248 age and sex distribution as well as prevalence of diabetes, hypertension, and statin use 249 between femoral and carotid endarterectomy groups. The exception was that current 250 smoking was more common in femoral endarterectomy patients, consistent with the 251 higher prevalence of smoking among patients with PAD versus CAD or stroke.(50, 51) 252 Plaque specimens were digested into single-cell suspensions using the same digestion 253 methods as for scRNA-seq analyses, gated on macrophages 254 (CD11b + CD14 + CD64 + HLA-DR hi cells), and further sub-classified based on expression of 255 CCR2 ( Figure 4D-F), a chemokine receptor expressed on monocyte-derived 256 inflammatory macrophages (52)(53)(54). Consistent with our overall comparisons of myeloid 257 phenotypes in the scRNA-seq data (with carotid plaques demonstrating a comparatively 258 inflammatory myeloid signature), we observed a significantly higher proportion of 259 CCR2 + macrophages in carotid (71.4% of total plaque macrophages) versus femoral 260

Lymphoid re-clustering suggests comparatively inflammatory and cytotoxic T cell 281 bias in carotid plaque versus B cell bias in femoral plaque 282
Lymphoid cell-specific re-clustering revealed 9 distinct lymphoid populations (Ly. 0-8, 283 Figure 6). Clusters were determined using a combination of previously defined cluster-284 specific genes, reference datasets of combined human protein and transcript single cell 285 transcriptomes (58) (21), and gene set enrichment analyses. Clusters Ly.0-2, 4, and 7 286 all highly expressed CD3D and represent T cell clusters, with Ly.0 and Ly.7 expressing 287 CD4, representing CD4+ T cells (and Ly.7 expressing TIGIT and FOXP3 as a T reg 288 cluster). Ly.1 and 2 highly expressed CD8 as CD8+ T cells, most highly expressing 289 cytotoxicity markers GZMK and GZMH, respectively. Gene set enrichment analyses 290 suggested high composition of T helper type 1 (T h 1) cells in Ly.0, which also had high 291 IL7R expression (previously observed in T h 1 effector cells (59)). Conversely, Ly.7 highly 292 expressed FOXP3 and other T reg -related genes and without inflammatory or cytotoxicity-293 associated gene expression profiles. Cluster 4 represented a group of CD3-expressing 294 T cells that did not have clear CD4 or CD8 expression but had comparatively high 295 expression of CXCL8 (60, 61). Cluster 3 highly expressed NKG7 and did not express 296 CD3D (while also having minimal CD4 and CD8 expression), thus representing natural 297 killer (NK) cells. Clusters 5 and 6 were CD79-expressing B cells, with distinctions in 298 gene expression between these two B cell clusters suggesting Ly.5 is type 1 B cells 299 (B1) and Ly.6 is type 2 B cells (B2). B1-b cells have been implicated in the secretion of 300  Figure 7C, Supplemental Figure S8 for T cell gating); interestingly, the 315 higher CD8+ T cell proportion was also observed in the blood of carotid patients. 316 Meanwhile, femoral plaques were highly enriched in B cells compared with carotid 317 plaque (9.0% versus 0.02% for Ly.5 and 7.9% versus 1.1% for Ly.6; p<0.001). To 318 explore lymphoid cell clustering in situ, we quantified T cell and B cell aggregates in 12 319 separate patients who underwent carotid (N=7) or femoral (N=5) endarterectomy and 320 observed that T cells were the most populous cell type in these aggregates in femoral inflammatory markers that may differ between groups. A separate concern relates to 448 potential batch effects from samples being harvested and undergoing sequencing 449 reactions at different times, with potential differential effects on early response genes. 450 Although we aimed to correct for these with Harmony, a software package used for 451 batch-effect correction of scRNA-seq data (90), and observed overall good sample-and 452 plaque site-level integration, residual confounding related to processing and batch 453 effects remains possible. 454

455
As with many single-cell analyses of relatively rare plaque specimens obtained from 456 humans in vivo, (18,19,91) our sample size was limited, with the potential to adversely 457 affect generalizability of our plaque site-specific conclusions. We profiled 35265 femoral 458 plaque CD45+ cells derived from 9 patient specimens and 30655 carotid CD45+ cells 459 derived from 4 patient specimens. However, our number of individual cells analyzed is 460 larger than those from recent single-cell analyses of carotid plaque (scRNA-seq on 461 3282 and 7169 cells (18,19)) and the only prior scRNA-seq study of femoral plaque to 462 our knowledge, which compared scRNA-seq data from a single femoral plaque 463 specimen to data from an existing carotid plaque scRNA-seq dataset. (20) The potential 464 generalizability of our findings is further supported by (1)  For analyses, femoral plaque sample matrices were imported into the Seurat v4 R 571 package (93-96) and combined into a "femoral plaque" Seurat object. Likewise, carotid 572 plaque sample matrices were imported and combined into a "carotid plaque" Seurat 573 object. For both Seurat objects, cells were filtered for mitochondrial reads <10%, 200 < 574 nCount_RNA < 10 000, and 200 < nFeature_RNA < 10 000. Each Seurat object was 575 then filtered to remove mitochondrial (MT-) and ribosomal (RP-) genes. Each Seurat 576 object was also normalized and scaled and filtered to keep the top variable features 577 (greatest standardized variance; n = 3000) across the datasets. The objects were then 578 merged using the Seurat merge command and integrated using the R package 579 Harmony.(97) The RunHarmony command was used to calculate harmonized 580 dimension reduction components using the samples as the grouping variable, and 581 doublet discrimination was performed. Principle components were then calculated, and 582 an elbow plot was generated to select principal components to use for downstream analysis; here, 30 principal components explained most of the variation. UMAP 584 dimensional reduction was then computed followed by unsupervised clustering using 585 the FindNeighbors and FindClusters Seurat functions, using the number of 586 principle components mentioned above and a resolution of 0.3 (FindClusters), 587 which captured distinct cell types empirically. 588 589

Myeloid and Lymphoid Sub-clustering and Batch Correction 590
Sub-clustering of myeloid cells was completed by first extracting the raw expression 591 matrix from all myeloid cells for each sample using the GetAssayData function from 592 the Seurat package. All carotid myeloid matrices were then combined into a carotid 593 myeloid Seurat object, whereas all femoral myeloid matrices were combined into a 594 femoral myeloid Seurat object. The carotid and femoral myeloid objects were then 595 merged and, to correct for batch effects, integrated with Harmony (90) using sample as 596 the grouping variable. UMAP dimensional reduction, downstream differentially 597 expressed genes (DEG) and pathway analyses was then performed in the same 598 manner as described above. This process was repeated with lymphoid cell data to 599 create sub-clusters for lymphoid cells. 600 601

Detection of differentially expressed genes 602
Detection of DEGs between clusters was performed using the FindAllMarkers 603 Seurat function, specifying return of significantly (Bonferroni p adjusted < 0.05) 604 upregulated genes with a log 2 fold change (log 2 FC) threshold of 0.25. Cell types were 605 assigned to clusters by evaluating gene expression of individual clusters using differential gene expression. For individual clusters, detection of DEGs between carotid 607 and femoral plaque location was performed using the FindAllMarkers command 608 specifying to return both positively and negatively changed genes, and no log 2 FC or P 609 value cutoffs. Genes with positive and negative log 2 FC values were used to identify 610 upregulated genes in the carotid or femoral plaque location, respectively. For all DEG 611 calculations, the "RNA" assay and data slot were used and performed using the default 612 Wilcoxon rank-sum method. To identify subtypes of CD4+ T cells in our dataset, we performed gene set enrichment 628 analysis (GSEA) on clusters that were identified as CD4+ T cell clusters using DEG.  Table 2). To determine differences by 667 plaque site, we performed pairwise t-tests at an alpha of 0.05 comparing patients 668 undergoing carotid vs. femoral endarterectomy regarding: (1) proportions of 669 macrophages in plaque that were CCR2+ for carotid versus femoral plaque, (2) blood 670 monocyte CCR2 expression (as mean fluorescent intensity), and (3) proportions of T 671 cells in plaque and blood that were CD4+ and CD8+ (as a proportion of total T cells in 672 that specimen). Flow cytometry reagents are included in Supplemental Table T1. For immunohistochemistry analyses, post-surgical femoral and carotid plaque tissue 686 from patients was prepared through the Pathology Core Facility at Northwestern 687 University. Each tissue was fixed in formalin and embedded in paraffin, and 5μm-thick 688 slices were cut from each paraffin block and stained with hematoxylin and eosin (H&E). 689 Histology was reviewed by a trained pathologist blinded to plaque location. The 690 pathologist screened whole H&E-stained slides cut from these formalin-fixed, paraffin 691 embedded blocks of atheroma from 42 additional patients who underwent carotid or 692 femoral endarterectomy, for the purpose of determining whether 5 or more cells with 693 lymphoid appearance were present in any high-powered field; 12 out of 42 samples (7 694 carotid, 5 femoral) met these criteria. Notably, none of the 42 samples contained 695 adventitial tissue per the reviewing pathologist, consistent with endarterectomy 696 technique of avoiding adventitial tissue.(102) Contiguous slides from the same blocks of 697 these 12 plaque samples subsequently underwent immunohistochemistry. This included 698 processing for 3, 3-diaminobenzidine (DAB-HRP) staining, counterstaining with 699 hematoxylin, then immunostaining for four separate surface markers (CD45, CD20, and 700 CD3). Antigen retrieval and antibody staining was optimized at the Northwestern 701 University Pathology Core Facility. The stained sections were imaged utilizing 4x and 702 40x objectives on the brightfield mode via the Vectra 3 Automated Quantitative 703 Pathology Imaging System (PerkinElmer) at the Immunotherapy Assessment Core at 704