Single cell RNA sequencing identifies unique inflammatory airspace macrophage subsets

Abstract

Macrophages are well recognized for their dual roles in orchestrating inflammatory responses and regulating tissue repair. In almost all acutely inflamed tissues, 2 main subclasses of macrophages coexist. These include embryonically derived resident tissue macrophages and BM-derived recruited macrophages. While it is clear that macrophage subsets categorized in this fashion display distinct transcriptional and functional profiles, whether all cells within these categories and in the same inflammatory microenvironment share similar functions or whether further specialization exists has not been determined. To investigate inflammatory macrophage heterogeneity on a more granular level, we induced acute lung inflammation in mice and performed single cell RNA sequencing of macrophages isolated from the airspaces during health, peak inflammation, and resolution of inflammation. In doing so, we confirm that cell origin is the major determinant of alveolar macrophage (AM) programing, and, to our knowledge, we describe 2 previously uncharacterized, transcriptionally distinct subdivisions of AMs based on proliferative capacity and inflammatory programing.

Keywords: Immunology; Macrophages; Pulmonology.

Figure 1. Single cell transcriptional profiling identifies 5 discrete AM populations across homeostasis, acute inflammation, and resolving inflammation.

Mice were treated with intratracheal LPS and macrophages were isolated from lavage at days 0 (homeostasis), 3 (peak neutrophil inflammation), and 6 (resolution of lung inflammation). (A) T-distributed stochastic neighbor embedding (tSNE) plot shows clustering of 902 cells based on gene expression. Point coordinates are based on tSNE dimensionality reduction of the top 6 principal components calculated from the 5,784 most informative genes. Cell color specifies assignment of cells to 1 of 5 clusters (c1–5) inferred using shared nearest neighbor clustering. (B) Normalized expression of macrophage markers overlaid on tSNE plot. (C) Time course information overlaid on tSNE plot. (D) Relative proportion of cells in each cluster versus time.

Figure 2. AM populations revealed by single cell RNA-seq reflect cell origin.

(A) Relative expression of Mrc1 and CD14 overlaid on tSNE plot. Cells that express both markers are turquoise. High versus low expression is defined relative to the 85th percentile. (B) Bubble plot comparing expression of resident (blue) and recruited (red) biomarkers across the 5 macrophage clusters. Bubble size is proportional to percentage of cells in a cluster expressing a gene, and color intensity is proportional to average scaled gene expression within a cluster. (C) Summary expression of 4 resident biomarkers (Mrc1, Itgax, Siglecf, and Siglec1) and 6 recruited biomarkers (Cd14, Ly6c1, Apoe, Ccr5, Mafb, and Sell) overlaid on tSNE plot.

Figure 3. Macrophage polarization states are not mutually exclusive and reflect cell origin.

(A) Cells with high expression of Arg1 and/or Nos2 overlaid on tSNE plot (high versus low expression defined relative to the 85th percentile). (B) Bubble plot shows relative expression of M2 (blue) and M1 (red) markers across clusters. Bubble size is proportional to percentage of cells expressing a gene, and color intensity is proportional to average scaled gene expression within a cluster. (C) Coexpression of Arg1 and Nos2 shown as square root–transformed expression of one gene against the other. Sample color denotes cluster from Figure 1A. (D) Summary expression of genes from B overlaid on tSNE plot. (E and F) Mean normalized expression of M2 (E) and M1 (F) markers relative to homeostatic RAMs. Data are shown across days for RAMs and clusters for RecAMs. Relative-to-baseline expression obtained by subtracting median value observed in clusters 1 and 2 at day 0.

Figure 4. Resident AMs contain a proliferative subpopulation present during health and inflammation.

(A and B) Mean normalized expression of genes annotated for enriched pathways indicated that they were also upregulated in cluster 1 (A) or cluster 2 (B) when compared with RecAM clusters. (C–E) Normalized expression of 3 classic markers of proliferation overlaid onto tSNE plot. (F–H) Characterization of proliferating RAMs in cluster 2 after statistically removing expression heterogeneity related to cell cycle from dataset. (F) t-SNE plot–based scaled residual expression obtained from modeling relationship between gene expression and estimated cell cycle score. Cells colored based on original clusters in Figure 1A. (G) Time course information overlaid on new t-SNE plot. (H) Distribution of cells from original proliferation cluster across new clusters inferred using cell cycle–corrected dataset. New clusters were broadly similar to original clusters and were, thus, named and colored to match cluster analogs in Figure 1A.

Figure 5. Macrophage subsets are identified during inflammation.

(A–C) Bar graphs comparing mean normalized expression. Mean normalized expression of genes annotated for enriched pathways upregulated in clusters 3 and 4 together when compared with all other clusters (A), or cluster 3 (B) or cluster 4 (C) when compared with each other. (D–F) Normalized expression of 3 inflammatory cytokines overlaid on tSNE plot. (G) Bar plots comparing mean normalized expression of genes annotated for enriched pathways upregulated in cluster 5 when compared with all other clusters.

Figure 6. Expression of identified gene sets distinguishes clusters.

Scaled expression of top 30 differentially expressed genes from each cluster. Individual cells are represented on the horizontal axis and grouped by cluster and day of inflammation.

Figure 7. Gene networks from bulk RNA-seq are coexpressed within single cell RNA-seq–based clusters.

(A–G) Seven coexpressed gene networks derived from bulk RNA-seq data exhibit peak expression within single cell–derived resident (A–C) or recruited (D–G) macrophage clusters (see also Supplemental Figure 3). For each network, mean scaled eigengene expression is shown across bulk RNA-seq resident and recruited AMs from each day (heatmaps). For comparison with the single cell dataset, relative eigengene expression of each network is overlaid onto the single cell tSNE plot. Select results from pathway analysis of each network are shown in Table 1.