Lipid-associated macrophages transition to an inflammatory state in human atherosclerosis increasing the risk of cerebrovascular complications

Abstract

The immune system is integral to cardiovascular health and disease. Targeting inflammation ameliorates adverse cardiovascular outcomes. Atherosclerosis, a major underlying cause of cardiovascular disease (CVD), is conceptualised as a lipid-driven inflammation where macrophages play a non-redundant role. However, evidence emerging so far from single cell atlases suggests a dichotomy between lipid associated and inflammatory macrophage states. Here, we present an inclusive reference atlas of human intraplaque immune cell communities. Combining scRNASeq of human surgical carotid endarterectomies in a discovery cohort with bulk RNASeq and immunohistochemistry in a validation cohort (the Carotid Plaque Imaging Project-CPIP), we reveal the existence of PLIN2^hi/TREM1^hi macrophages as a toll-like receptor-dependent inflammatory lipid-associated macrophage state linked to cerebrovascular events. Our study shifts the current paradigm of lipid-driven inflammation by providing biological evidence for a pathogenic macrophage transition to an inflammatory lipid-associated phenotype and for its targeting as a new treatment strategy for CVD.

Keywords: Atherosclerosis; Cardiovascular Disease; Inflammation; Lipid-associated macrophages; Mechanisms.

Fig. 1. Identification of plaque T and NK cell populations with cytotoxic and activation signatures.

a–d, Cells identified as CD4 T, CD8 T, NK and proliferating cells in the overall analysis (n = 15,052 cells) were extracted and analyzed separately. a, The UMAP shows the 11 identified lymphocyte subpopulations. b,c, The expression of selected automatically discovered cluster marker genes (BH-adjusted P < 0.05, two-sided Wilcoxon tests) and known cell-type marker genes is shown on the UMAP (b) and summarized in the dot plot, where the color of the dots represents average expression and size represents the percentage of cells within the cluster that express it (c). Additional cluster markers are shown in Extended Data Fig. 2a,b. d, Selected KEGG pathways and GO biological processes (BP) and molecular functions (MF) that showed significant overrepresentation in the cluster marker genes (color of the dots represents odds ratio from one-sided Fisher exact tests, and size of the dots represents the number of genes enriched in category or cell type; BH-adjusted P < 0.1). P values for individual marker genes and pathways are provided in Source Data Fig. 1. Source data

Fig. 2. Plaque myeloid cells harbor diverse subsets of macrophages with distinct gene signatures of functional association.

a–d, Cells identified as macrophages or cDCs in the overall analysis (n = 4,533 cells; average of n = 747 cells per patient) were extracted and analyzed separately. a, The UMAP shows the 12 identified myeloid subpopulations. b,c, The expression of selected automatically discovered cluster marker genes (BH-adjusted P < 0.05, two-sided Wilcoxon tests; Extended Data Fig. 3c) and known cell-type marker genes is shown on the UMAP (b) and summarized in the dot plot, where the color of the dots represents average expression and size represents the percentage of cells within the cluster that express it (c). Additional cluster markers are shown in Extended Data Fig. 3a–c. d, Selected KEGG pathways and GO biological processes (BP) and molecular functions (MF) that showed significant overrepresentation in the cluster marker genes (color of the dots represents odds ratio from one-sided Fisher exact tests, and size of the dots represents the number of genes enriched in category or cell type; BH-adjusted P < 0.1). P values for individual marker genes and pathways are provided in Source Data Fig. 2. Source data

Fig. 3. Trajectory analysis of plaque macrophage populations.

a–j, The plaque macrophages (n = 3,628 cells; see Fig. 2) were extracted, and RNA-velocity analysis was performed on the relationship between the eight macrophage clusters. a, The arrows on the UMAP indicate the directions of the predicted future transcriptional states of the cells. b–d, Per-cell scores for lipid metabolism (b), apoptosis (c) and inflammation (d) were computed with AUCell using custom gene lists (see Methods, Supplementary Table 1) and visualized on the UMAP. e,f, Targeted RNA-velocity analysis (e) and CytoTRACE random walk analysis (see also Extended Data Fig. 8e) (f) of the TREM2^hi and PLIN2^hi/TREM1^hi populations. g, The volcano plot shows genes differentially expressed between the TREM2^hi and PLIN2^hi/TREM1^hi populations. Significantly differentially expressed genes are in red (DESeq2 patient-level pseudobulk analysis, paired Wald tests, BH-adjusted P < 0.1). Additional pairwise analysis between PLIN2^hi/TREM1^hi and all other macrophage clusters is shown in Extended Data Fig. 8. h, Selected GO and KEGG pathways associated with gene expression differences between the TREM2^hi and PLIN2^hi/TREM1^hi populations (FGSEA analysis, genes ranked by the DESeq2 test statistic, BH-adjusted P values < 0.05). i,j, PLIN2⁺/TREM1⁺ (red arrows), PLIN2^+/TREM2⁺ (blue arrows) and TREM1⁺/TREM2⁺/PLIN2⁺ (black arrows) plaque areas shown by immunostaining on human carotid plaque specimens. Scale bars, 100 μm (staining was performed on n = 5 plaques). Source data

Fig. 4. Cell–cell interaction analysis suggests a central role for macrophages in the immune cell communication network of the human atherosclerotic plaque.

a–o, Cell–cell interactions between major immune cell types were investigated using NATMI to determine the number of ligand–receptor pairs that connected each pair of cell types. a, The overall cell-connectivity-summary network is summarized in the heat map (cells expressing ligands are shown in rows, and cells expressing receptors are shown in columns). The number of significant ligand–receptor pairs is indicated for each interaction. The heat map is colored according to the NATMI specificity score (product of ligand specificity × receptor specificity). b, Selected examples of predicted ligand–receptor interactions (y axis; ordered by specificity) between the given source and target cell populations (x axis) are shown in a heat map. c–j, Gene expression of TREM1 (c), TREM2 (d), PLIN2 (e), CCL2 (f), IL6 (g), IL1B (h), TLR2 (i) and TLR4 (j) was measured in hMDMs treated with TLR4 ligand (LPS, 1 ng ml⁻¹), TLR2 ligand (FSL-1, 100 ng ml⁻¹), oxLDL (25 μg ml⁻¹) or ACCM for 24 h. Data are reported as relative gene expression compared with housekeeping gene (n = 17 biologically independent donors from seven independent experiments for all genes, with the exception of IL6 where n = 13 biologically independent donors were analyzed from five independent experiments; values reported as mean ± s.e.m., one-way analysis of variance (ANOVA), mixed-effect analysis). Blocking TLR2 in hMDMs abrogated the effect of ACCM. hMDMs were treated with LNA-ASOs targeting TLR2 (Methods) for 3 days prior to a 24-h treatment with TLR2 ligand (FSL-1, 100 ng ml⁻¹) or ACCM. k–o, Gene expression of TLR2 (k), TREM1 (l), PLIN2 (m), CCL2 (n) and IL1B (o) was measured. Data are reported as relative gene expression compared with housekeeping gene (n = 3 biologically independent donors from two independent experiments; values reported as mean ± s.e.m., two-way ANOVA). Source data

Fig. 5. Symptomatic plaques have more cells associated with the inflammatory TREM1 LAM signature.

a, The CPIP biobank samples were used as validation cohort following the illustrated diagram (diagram created by BioRender). b, Plaque areas positive for CD68, PLIN2, TREM1 and Oil Red O were shown to stain the same areas (marked by rectangles) by immunostaining on human carotid plaque specimens. Scale bars, 1 mm (whole carotid section image) and 100 μm (in the magnified images; representative image from n = 37 stained plaques). An additional staining example with the corresponding antibody controls is shown in Extended Data Fig. 10a. c, The plaque area that stained positive for TREM1 correlated positively with the PLIN2 plaque area (Spearman test was used for the correlation analysis; n = 37). d, TREM1 and PLIN2 gene expression, assessed by bulk RNA-seq in carotid plaque samples collected from the CPIP biobank, were strongly correlated (n = 78; Spearman rank correlation test). e,f, Plaque areas stained positive (% of total plaque area) for PLIN2 and TREM1 (n = 19 asymptomatic and n = 18 symptomatic; Mann–Whitney U-test was used for group comparisons; e), as well as gene expression levels of PLIN2 and TREM1 (n = 51 symptomatic and n = 27 asymptomatic; Student’s t-test was used for group comparisons; f), were significantly higher in symptomatic plaques than in asymptomatic plaques. g, Human carotid plaque gene expression levels of TLR2, TLR4, CCL2, CXCL2, CXCL3 and CXCL8 comparing TREM1/PLIN2 high-expressing and TREM1/PLIN2 low-expressing plaques from the CPIP cohort (see Methods). Data are presented as log₂(CPM) compared between the two groups using two-sided Student’s t-test. h, Distribution of symptomatic and asymptomatic patients in TREM1/PLIN2 high-expressing versus TREM1/PLIN2 low-expressing plaques from the CPIP cohort (n = 32 high and n = 46 low, OR = 7, P = 6.6 × 10⁻⁴, Fisher exact test). Boxes indicate interquartile range (IQR; 25th and 75th percentiles); center line indicates median (50th percentile); whiskers indicate minimum (within lower quartile − 1.5 × IQR) to maximum (within upper quartile + 1.5 × IQR).

Extended Data Fig. 1. Single-cell RNA sequencing of human atherosclerotic plaque.

Atherosclerotic plaques were collected from n = 6 patients undergoing CEA, enzymatically digested, sorted for CD45⁺ cells and subjected to scRNA-seq analysis using 10x Chromium platform (Diagram created by Biorender) (a). The barplots show the frequencies of the clusters distribution in the different patients (b). Unsupervised Leiden clustering of the integrated dataset identified n = 10 distinct clusters encompassing all major immune cell populations (c). The clusters were annotated by classical marker gene expression and by their top marker genes (BH adjusted p < 0.05; Wilcoxon tests) (d,e). Automatic cell type predictions computed using Azimuth to map cells to the Lung v2 (HLCA)^– reference data set (f). Gene set over-representation analysis of cluster markers using sets of known cell type marker genes from xCell(g).

Extended Data Fig. 2. T and NK cell cluster markers and per-patient frequency.

Supplementary marker genes for the T and NK populations are shown in the dot plot as (see also Fig. 1c) (a). The heatmaps shows the top marker genes for the T and NK cell populations (b). The barplots show the frequencies of the T and NK cell clusters in the different patients (c). Gene set over-representation analysis of cluster markers using sets of known cell type marker genes from xCell(d). Automatic cell type predictions computed using Azimuth to map the data to the Lung v2 (HLCA) (e) and PBMC(f) reference datasets.

Extended Data Fig. 3. Myeloid cell clusters markers and per-patient frequency.

Supplementary marker genes for the myeloid clusters are shown in the dot plots (see also Fig. 2C) (a, b). The heatmaps shows the top marker genes for the myeloid clusters (c). The barplots show the frequencies of the myeloid clusters in the different patients (D). The asterisks in (c) and (d) mark the position of the PLIN2^hi/TREM1^hi cluster. Automatic cell type predictions performed using Azimuth to map cells to the Lung v2 (HLCA) reference dataset (e). Gene set over-representation analysis of cluster markers using sets of known cell type marker genes from xCell(f).

Extended Data Fig. 4. Confirmation of dendritic cell sub-population annotations using known marker genes from the literature.

To confirm our annotation of the individual DC subpopulations we examined the expression of known marker genes in our clusters. The dotplot shows the expression of the mReg DC population markers reported in ref. (their Fig. 4b) in our myeloid clusters (a). The dotplot shows the expression of the DC populations markers reported in ref. (their Fig. 1d) in our myeloid clusters (b).

Extended Data Fig. 5. Alternative clustering analysis of myeloid data using a Seurat-based workflow.

For this analysis, data normalisation was performed using the sctransform algorithm and integration was performed using Seurat as described in the methods. The UMAP shows the identified clusters (Louvain clustering algorithm) (a). The expression of selected markers of each cluster are shown on UMAP (b). The dotplot shows the expression of the marker genes from Fig. 2c in the Seurat clusters (cluster 3 comprises of PLIN2^hi/TREM1^hi macrophages) (c). The Alluvial plot shows the mapping between the clusters from Fig. 2 (cluster_id.main) and the clusters identified using the Seurat based workflow (Cluster_id.sct) (nearly all of the cells in the PLIN2^hi/TREM1^hi cluster from Fig. 2 map to cluster 3 from the Seurat-based workflow) (d).

Extended Data Fig. 6. Expression of PLIN2^hi/TREM1^hi signature genes in all identified immune cell populations.

The boxplots show the expression of CD68, PLIN2, TREM1, TREM2, IL1B and CCL2 in all the immune cell clusters identified. The clusters correspond to those shown in Suppl. Figure 1 (for the B, Plasma, Mast and pDC populations), Fig. 2 (for the T/NK cell clusters) and Fig. 3 (for the myeloid clusters). Normalised per-patient pseudobulk expression values were computed with DESeq2 (Variance Stabilising Transformation). The lower and upper bounds of the boxes mark the first and third quartiles. Line within boxes represent median values. Whiskers extend to the smallest and largest values no further than 1.5 * inter-quartile range from the bounds of the box. Outlying points beyond the whiskers are plotted individually.

Extended Data Fig. 7. Pairwise differential expression analyses between the PLIN^hi/TREM1^hi macrophages and all the other macrophage populations.

The volcano plot shows genes differentially expressed between the PLIN2^hi/TREM1^hi and S100A8/IL1B⁺(a), S100A8/IL1B⁻(b), C1Q(c), HMOX1⁺(d), IL10⁺/TNFAIP3⁺(e) and IFNresp (f) macrophage populations. Significantly differentially expressed genes colored red (DESeq2 patient-level pseudobulk analysis, BH adjusted p < 0.1, list of DEG genes provided in Source Data Fig. 3). PLIN2, TREM1 and CCL2 (in bold) are consistently significantly more expressed in PLIN2^hi/TREM1^hi cells compared to all other clusters.

Extended Data Fig. 8. Re-analysis of published human carotid atherosclerosis scRNA seq data (GSE159677).

Unsupervised Leiden clustering of the dataset identified n = 13 distinct clusters (a) distributed in both atherosclerotic core (AC) and proximal adjacent (PA) portions of carotid artery tissue (b) and in all 3 studied patients (c). Identification of the main immune and non-immune clusters was performed using markers used in Alsaigh et al. Figure 1d (d). Cells identified as myeloid were extracted and analysed separately. The UMAP shows the 10 identified myeloid sub-populations (e). The expression of selected cluster marker genes is shown on the UMAP (f) and summarised in the dot plot (g). Distribution of myeloid clusters in the 3 patient samples (h).

Extended Data Fig. 9. LAM genset scores and macrophage trajectory analysis.

Boxplots for the distributions of the lipid, inflammation and apoptosis scores for each of the MNP clusters (n = 3628 cells in total) (a). The connectivity between MNP clusters was assessed by partition-based graph abstraction (PAGA) analysis (nodes size and edge width proportional to cluster cell number and degree of connectivity respectively) (b) and PAGA on velocity (c). The dendrogram shows the distance between the expression profiles of the clusters (d). CytoTRACE random walk analysis where PLIN2^hi/TREM1^hi was assigned as start cluster (e). For the boxplots, lower and upper bounds of boxes mark first and third quartiles. Lines within boxes represent median values. Whiskers extend to the smallest and largest values no further than 1.5 * inter-quartile range from the box. Outlying points beyond whiskers are plotted individually.

Extended Data Fig. 10

TREM1 protein expression in all treatment groups was measured using flow cytometry (representative of n = 3 biologically independent represented (a)). Quantification of mean MFI from n = 3 biologically independent samples (values reported as mean ± SEM, One way ANOVA, Dunnett’s multiple comparison) (b). Box plots for TLR2 and TLR4 differential expression in TREM2^hi and PLIN^hi/TREM1^hi cluster. TLR2 gene expression is higher in PLIN^hi/TREM1^hi (DESeq2’s Wald test BH adjusted p-value = 0.08) while no significant difference was found for TLR4 (BH adjusted p-value = 0.56) (DESeq2 analysis; two-sided, paired Wald tests) (c). The Carotid Plaque Imaging Project (CPIP) biobank samples were stained for CD68, PLIN2 and TREM1 and their respective controls as well as oil red o (ORO) stain. Scale bars 1 mm (far left and far right) and 100um (in the amplified images) (staining for a total of n = 37 plaques was performed)(d). The heatmap shows the correlation between TREM1 and PLIN2 expression from CPIP bulk transcriptomic data with markers of different cell types present in plaque. Both TREM1 and PLIN2 only showed high correlations with the macrophage markers. Spearman correlation coefficient was used, n = 60 patients (e). For the boxplots, lower and upper bounds of boxes mark first and third quartiles. Lines within boxes represent median values. Whiskers extend to the smallest and largest values no further than 1.5 * inter-quartile range from the box. Outlying points beyond whiskers are plotted individually. Source data