The transcriptional and phenotypic characteristics that define alveolar macrophage subsets in acute hypoxemic respiratory failure

Abstract

The transcriptional and phenotypic characteristics that define alveolar monocyte and macrophage subsets in acute hypoxemic respiratory failure (AHRF) are poorly understood. Here, we apply CITE-seq (single-cell RNA-sequencing and cell-surface protein quantification) to bronchoalveolar lavage and blood specimens longitudinally collected from participants with AHRF to identify alveolar myeloid subsets, and then validate their identity in an external cohort using flow cytometry. We identify alveolar myeloid subsets with transcriptional profiles that differ from other lung diseases as well as several subsets with similar transcriptional profiles as reported in healthy participants (Metallothionein) or patients with COVID-19 (CD163/LGMN). We use information from CITE-seq to determine cell-surface proteins that distinguish transcriptional subsets (CD14, CD163, CD123, CD71, CD48, CD86 and CD44). In the external cohort, we find a higher proportion of CD163/LGMN alveolar macrophages are associated with mortality in AHRF. We report a parsimonious set of cell-surface proteins that distinguish alveolar myeloid subsets using scalable approaches that can be applied to clinical cohorts.

Fig. 1. Alveolar monocyte and macrophage subsets are highly diverse in acute hypoxemic respiratory failure.

We performed bronchoscopy with bronchoalveolar lavage (BAL) on 8 individuals with acute hypoxemic respiratory failure supported with invasive mechanical ventilation (Bronchoscopy 1 = B1). Four of the participants were sampled again 4 days later (Bronchoscopy 2 = B2). We isolated single cells and assessed them with CITE-seq. A Uniform manifold approximation and projection (UMAP) plot displaying clustering of 64,317 cells based on gene expression. We annotated the clusters by mapping them to published datasets to identify B cell, T cell, myeloid (alveolar macrophages = AM, alveolar monocytes = mono, classical dendritic cells = cDCs), and other cell-types (designated by color). B Cells identified as myeloid (including macrophages, monocytes, and cDCs) in panel A were re-clustered. Color designates assignment of cells to one of the 9 clusters identified by Seurat. C Dot plot comparing the expression of marker genes (x-axis) across nine alveolar myeloid cell clusters (y-axis). Each cluster is annotated based on the marker genes (Table S5). The proportion of each cluster as a percentage of all alveolar myeloid cells is displayed. The dot size is proportional the percentage of cells expressing the gene in each color. The color intensity is proportional to the average scaled log-normalized expression within a cluster. D Bar plot displaying the individual percentages, median, and interquartile range of each subset as a proportion of all alveolar myeloid cells at B1 and B2.

Fig. 2. Correlations between alveolar myeloid subsets, biomarker profiles, and clinical severity.

A Heatmap of the correlation coefficients between alveolar myeloid subset proportions (y-axis) and log₂ alveolar biomarker levels (x-axis). Colors represent the correlation with scale indicating value of Pearson’s r correlation. Axes are ordered by clustering based on Pearson correlation-distances using pheatmap. B Associations between the proportion of alveolar myeloid subsets as a percentage of all alveolar myeloid cells (y-axis) and oxygenation index (OI) (x-axis). OI is a measure of respiratory failure severity that accounts for both oxygenation and mean airway pressure being delivered by mechanical ventilation. Higher values indicate more severe respiratory failure. Depicted are the individual values, linear regression best-fit line, and 95% confidence intervals (n = 8 unique participants). P-values test whether the slope (β-coefficient) is significantly non-zero and are nominal. Bonferroni p-values are adjusted for 9 statistical tests (multiple hypothesis testing for an association between each of the nine subsets and the clinical outcome). C The percentage of each alveolar myeloid subset as a proportion of all alveolar myeloid cells in participants with or without ARDS. Depicted are the individual values, median, and interquartile range of each subset as a proportion of all alveolar myeloid cells (n = 8 unique participants). P-values were generated with two-sided Mann-Whitney tests and are nominal. Bonferroni p-values are adjusted for 9 statistical tests (multiple hypothesis testing for an association between each of the nine subsets and the clinical outcome).

Fig. 3. Intermediate monocyte-macrophage subsets are present in the lung.

A Box-plots of median (center line), interquartile range (edge of box), 1.5x interquartile range (whiskers), and individual outliers (dots) of RNA velocity for each alveolar myeloid subset. B Partition-based graph abstraction (PAGA) of RNA velocity field projected on the alveolar myeloid UMAP (Fig. 1B). Gray dotted lines represent topologic connectivity of subsets. Arrows represent RNA velocity trajectory-inference (alveolar macrophage = AM). Dendric cells were excluded from RNA velocity analysis. C We collected paired peripheral blood mononuclear cells (PBMC) from participants who underwent research bronchoalveolar lavage (BAL). We isolated single cells and assessed them with CITE-Seq. We selected cells that mapped to blood myeloid lineage markers (monocytes, macrophages, and DCs) and then projected them into the BAL UMAP space. Blood monocytes clustered in the upper right of the BAL UMAP (occupying the same BAL UMAP space as FCN1 Alveolar Monocytes and Inflammatory Alveolar Monocytes). Blood DCs occupied the same BAL UMAP space as alveolar DCs.

Fig. 4. Cell-surface protein markers distinguish alveolar monocyte and macrophage subsets.

We used feature barcodes to identify the cell-surface proteins that best discriminated each alveolar myeloid transcriptional subset. A Table displaying the 9 most differentially expressed cell-surface proteins (y-axis) across nine alveolar myeloid cell transcriptional subsets (x-axis). Data on CD206 and CD14 are included at the bottom as a reference. The color intensity is proportional to the average scaled log-normalized expression for each cell-surface protein. Supplementary Data 1 shows the cell-surface protein intensities for each subset. B The normalized expression for each cell-surface protein (y-axis) for each transcriptional subset (x-axis). Depicted are violin plots (including median, interquartile range, and 1.5x interquartile range). The p-value was generated with a two-sided T-test of the pair-wise comparison between the two subsets with the largest difference in CD206 expression. C The normalized expression for each cell-surface protein (y-axis) for each transcriptional subset (x-axis). Depicted are violin plots (including median, interquartile range, and 1.5x interquartile range). P-values were generated with two-sided T-tests. The tables on the right summarize the relative cell-surface protein expression levels for each alveolar macrophage transcriptional subset.

Fig. 5. CD163/LGMN Macrophages are Associated with Mortality in Acute Respiratory Failure.

We collected bronchoalveolar lavage (BAL) fluid from intubated and mechanically ventilated participants (HMC Clinical Cohort) (n = 51). Table 2 shows the participant characteristics. We analyzed alveolar cells from the BAL fluid using flow cytometry. A Representative gating for identifying alveolar monocytes (Monos) (green box) and CD206⁺ alveolar macrophages (AMs) (orange box). We classified AMs into CD71^HICD163^HI (yellow box – Mature AMs) or CD71^LOCD163^HI (purple box – CD163/LGMN AMs) subsets based on our CITE-seq data (Fig. 4C). B The percentage of CD206⁺ AMs (orange box), airway monocytes (CD206⁺CD14⁺), CD71^HICD163^HI (yellow box – Mature AMs), and CD71^LOCD163^HI (purple box – CD163/LGMN) as a proportion of all alveolar myeloid cells between participants based on hospital mortality. Depicted are the individual values, median, and interquartile range of each subset as a proportion of all alveolar myeloid cells. P-values were generated with two-sided Mann-Whitney tests. CD71^LOCD163^HI (purple box – CD163/LGMN AMs) as a proportion of all alveolar myeloid cells between participants based on ventilator-free days (VFDs). Participants intubated > 7 days prior to bronchoscopy were excluded from this analysis. VFDs were defined as the number of days alive and free of invasive mechanical ventilation in the 21 days following bronchoscopy. VFDs were binned into tertiles. P-value was generated with Kruskal-Wallis test. Association between soluble CD163 BAL levels and the percentage of CD71^LOCD163^HI (CD163/LGMN AMs) as a proportion of all alveolar myeloid cells. Depicted are the individual values, linear regression line, and 95% confidence interval. P-values test whether the slope (β-coefficient) is significantly non-zero. r Pearson Correlation Coefficient.

Fig. 6. The proportions of alveolar myeloid subsets evolve during acute hypoxemic respiratory failure.

A Heat map showing differentially expressed genes in bulk alveolar myeloid cells between Bronchoscopy 1 (B1) and Bronchoscopy 2 (B2). B The percentage of each alveolar myeloid subset as a proportion of all alveolar myeloid cells at B1 and B2. Depicted are the individual values and lines connecting paired samples (n = 4 unique participants). P-values were generated with paired two-sided T-tests. Participants who met criteria for ARDS are shown in red. C UMAP plot incorporating our dataset with single-cell gene expression data from healthy human participants (HP) who underwent BAL. Color designates assignment of cells to either B1, B2, or HP. D Normalized gene expression of select genes between participants in B1 (n = 8), B2 (n = 4), and HP (n = 9). Depicted are the individual values, mean, and standard deviation of each subset as a proportion of all alveolar myeloid cells. P-values were generated with a two-sided T-test.