Distinct pulmonary and systemic effects of dexamethasone in severe COVID-19

Abstract

Dexamethasone is the standard of care for critically ill patients with COVID-19, but the mechanisms by which it decreases mortality and its immunological effects in this setting are not understood. We performed bulk and single-cell RNA sequencing of the lower respiratory tract and blood, and plasma cytokine profiling to study the effect of dexamethasone on systemic and pulmonary immune cells. We find decreased signatures of antigen presentation, T cell recruitment, and viral injury in patients treated with dexamethasone. We identify compartment- and cell- specific differences in the effect of dexamethasone in patients with severe COVID-19 that are reproducible in publicly available datasets. Our results highlight the importance of studying compartmentalized inflammation in critically ill patients.

Extended Data Figure 1 |

Consort chart.

Extended Data Figure 2 |. Differences in cytokine expression.

a, Volcano plot of cytokines comparing Dex (right > 0) and NoDex (left < 0), colored by significance (red; Wilcoxon test, adjusted p-value < 0.1). N = 23 Dex, N = 15 NoDex, for 18 cytokines; day 0 of hospitalization. b, Cytokine differences, stratified by time between first dexamethasone dose and sample collection.

Extended Data Figure 3 |. Gene set enrichment of bulk RNA-seq from PBMC.

Gene set enrichment plot of 19 most significant – top 10 for Dex (orange) and top 9 for NoDex (blue) – Reactome terms, based on differential gene expression results.

Extended Data Figure 4 |. Cross tissue differential gene expression.

log₂ fold-difference in gene expression of Dex and NoDex in TA (y-axis) v. blood (x-axis) plotted for additional cell types not shown in Figure 3. Significant genes in TA only (blue), blood only (brown), both compartments (red) are shown (adj. p-value < 0.1 & | log₂ fold-difference| > 0.5). Spearman’s correlation R value shown between the two compartments.

Extended Data Figure 5 |. Immune cell frequencies quantified and compared between Dex and NoDex samples.

X-axis shows log₂ fold-difference of Dex compared to NoDex in whole blood (purple circle); TA (orange circle); a blood validation set (Sinha et al, purple diamond); a lung validation set (bronchial alveolar lavage; Liao et al, orange triangle). Significance shown by boxes. The size of each shape corresponds to −log₁₀ p-value calculated using the Wilcoxon test.

Extended Data Figure 6 |. Gene set enrichment of Tregs in blood and lung.

Net enrichment scores from gene set enrichment analysis in blood and lung shown for Tregs (remaining cell types shown in Figure 4). Fold differences are shown for dexamethasone-treated samples (Dex), or healthy control samples, all relative to the NoDex samples within that dataset. Orange shows up in Dex or healthy relative to NoDex COVID-19 samples, bule shows down in Dex or healthy. Datasets represented are from COMET (whole blood, TA), Sinha et al (blood) and Liao et al (BAL).

Extended Data Figure 7 |. Whole blood cell interactions using CellChat.

CellChat interaction networks for COLLAGEN, ANNEXIN, ICAM and ITGB2 interactions shown comparing NoDex (left), and Dex (middle) patients, and healthy controls (right) for COMET whole blood dataset. Line thickness represents predicted strength of the interaction.

Figure 1:. Dexamethasone modulates cytokine and immune cell gene expression in the blood of patients with COVID-19

a, The introduction of dexamethasone (Dex) as standard of care for critically ill patients hospitalized with COVID-19 based on the results of the RECOVERY trial. Blood and tracheal aspirate (TA) samples were collected from intubated patients enrolled either before or after this change. b, Included patients and time points per analysis. A single sample was used per patient. Each patient was either treated with Dex (orange) or not (blue). Samples used in DIABLO analysis (Figure 2) are the overlap in PBMC bulk RNA sequencing and plasma cytokine rows. c, Individual plots of log-transformed significant cytokines IL-6, IL-10, and interferon gamma (IFN-gamma) (Wilcoxon test, adjusted p-value < .1). N = 23 Dex, N = 15 NoDex. d, Volcano plot of differential gene expression of PBMC RNA-seq data. N = 10 Dex, N = 11 NoDex.

Figure 2:. Supervised integrative analysis of blood transcriptomic and plasma cytokine data captures co-varying effects of dexamethasone on immune cell pathways and modulators

a, Integrative analysis of plasma cytokines (17 cytokine variables) and bulk PBMC RNA-seq (500 gene variables) data (paired) from patients comparing Dex and NoDex using DIABLO and highlighting shared contributions from individual data modalities. N = 10 Dex, N = 11 NoDex; day 0 of hospitalization. First two variates from DIABLO run comparing Dex (orange) vs. NoDex (blue) samples. A parameter value of 0.5 was chosen to model the strength of the relationship between the data and the treatment status. b, Cytokine contribution (loadings) to DIABLO variate 1. The color indicates the treatment group in which the median value was the highest (orange for Dex and blue for NoDex). c, Gene set enrichment analysis of PBMC RNA-seq contribution to DIABLO variate 1 (loadings) using REACTOME gene sets (methods). 20 most significant terms represented: top 10 for Dex (orange) and top 10 for NoDex (blue).

Figure 3:. Single-cell analysis of lung and peripheral blood samples from patients treated with or without dexamethasone

a,b, Plot per patient showing the collection of blood (a; N = 7 Dex, 3 NoDex) or TA samples (b; N = 10 Dex, 7 NoDex) overlaid on hospitalization (gray bars) and dexamethasone treatment (pink bars). X-axis shows days of hospitalization (day 0 = admission to UCSF hospital). Dots show the day when sample was collected, colored by Study Day (methods). c,d, UMAP plots of single-cell RNA-seq data from blood (c) or TA (d) samples, clustered and annotated by major immune cell types. e,f, UMAP plots of single-cell RNA-seq data from blood (e) or TA (f) samples, colored by Dex (blue) or NoDex (pink) samples. g,h, log₂ fold difference of gene expression of Dex and NoDex in TA (y-axis) v. blood (x-axis) plotted for Neutrophils (g) and Tregs (h). Significant genes in TA only (blue), blood only (brown), both compartments (red) are shown (adj. p-value < 0.1 & |log₂ fold-difference| > 0.5). Spearman’s correlation R value shown between the two compartments.

Figure 4:. Dexamethasone has discordant effects on cell type specific gene expression in lung and peripheral blood that are reproducible in external datasets

a,b, Net enrichment scores from gene set enrichment analysis in blood (a) and lung (b), faceted by cell type. Orange circles have a positive net enrichment score (NES), indicating the pathway is more highly expressed in dexamethasone-treated COVID-19 patients (Dex) or healthy controls relative to NoDex subjects. Solid circles identify pathways where GSEA FDR < 0.1, empty circles identify pathways with GSEA FDR ≥ 0.1, and blank spaces indicate no GSEA NES score was calculated for that pathway. Datasets represented are from COMET (whole blood, TA), Sinha et al (blood) and Liao et al (BAL).

Figure 5:. Receptor ligand inference from single-cell sequencing data reveals decrease in inflammation, antigen presentation, and T cell recruitment in blood and lung in response to dexamethasone

a, Clustered heatmap of CellChat results of TA samples from Dex as compared to NoDex patients with significant receptor-ligand pairs shown (p-value < 0.05 and |log₂ fold-difference| > 1). b,c, Cell type interaction networks for MHC-II (b) and SELPLG interactions (c) shown comparing NoDex (left) and Dex (right) patients of TA samples. Line thickness represents predicted strength of the interaction. d, Clustered heatmap of CellChat results of blood samples from Dex (COMET), Dex (Sinha et al), NoDex (COMET), NoDex (Sinha et al), and healthy controls (COMET) with receptor-ligand pairs that are significant between at least one pair of patient groups are shown (p-value < 0.05 and | log₂ fold-difference| > 1). e, Comparisons for the COMET dataset shown between Dex, NoDex, and healthy controls for a subset of significantly detected receptor-ligand interactions (*adj. p<0.1, **adj. p<0.001, ***adj. p<0.0001, ****adj. p<0.00001).