Acute Respiratory Distress Syndrome Molecular Phenotypes Have Distinct Lower Respiratory Tract Transcriptomes

Abstract

Rationale: Two molecular phenotypes of the acute respiratory distress syndrome (ARDS) with divergent clinical trajectories and responses to therapy have been identified. Classification as "hyperinflammatory" or "hypoinflammatory" depends on plasma biomarker profiling. Limited data are available about the differences in the pulmonary biology of the molecular phenotypes. Objectives: To identify differences in the pulmonary biology of ARDS molecular phenotypes Methods: We compared tracheal aspirate gene expression between hyperinflammatory and hypoinflammatory phenotypes in bulk RNA sequencing (RNASeq) from coronavirus disease (COVID-19) and non-COVID-19 ARDS and single-cell RNASeq from non-COVID-19 ARDS. In a subset of subjects, we also compared plasma proteomic data. Measurements and Main Results: In bulk RNASeq analyses, 1,157 genes were differentially expressed (false discovery rate < 0.1) between phenotypes in non-COVID-19 ARDS, and 85 genes were differentially expressed between phenotypes in COVID-19 ARDS. Eighteen genes were reproducibly differentially expressed between phenotypes in both cohorts, including greater expression of IL32, HSPA8, and PPP3CC in hyperinflammatory ARDS. A total of 195 pathways were reproducibly enriched across the two cohorts by gene set enrichment analysis, including greater expression of granulopoiesis, T-cell and IFN signaling, and integrated stress response pathways in hyperinflammatory ARDS. Network analysis of single-cell RNASeq in a third group of patients identified greater T-cell signaling to other immune cells in hyperinflammatory ARDS. Conclusions: Hyperinflammatory and hypoinflammatory ARDS molecular phenotypes have distinct respiratory biology. Hyperinflammatory ARDS is characterized by an increased IFN-stimulated gene expression and T-cell activation in the lungs.

Keywords: ARDS; RNA sequencing; molecular phenotypes; precision medicine.

Figure 1.

Differential expression analyses in tracheal aspirate bulk RNA sequencing. (A) Volcano plots of differentially expressed genes in the (A) ALI (Acute Lung Injury in Critical Illness) cohort of patients with non–coronavirus disease (non–COVID-19) acute respiratory distress syndrome (ARDS) (n = 41 patients), and COMET (COVID-19 Multiphenotyping for Effective Therapies) cohort of patients with COVID-19 ARDS (n = 72 samples from 39 patients). A positive log-fold difference indicates the gene is more highly expressed in hyperinflammatory ARDS. Genes with an adjusted P value <0.1 are identified in blue and orange. A complete table of differential expression results is in Data E1. (B) Heatmaps of the 18 genes that are reproducibly differentially expressed between molecular phenotypes in the ALI and COMET cohorts (false discovery rate <0.1 in independent analyses of both cohorts). Descriptions of gene functions are in Data E2. Gene expression was voom-transformed, centered, and z-scaled for plotting. Each row represents a different gene, and each column represents a different tracheal aspirate sample. Samples are clustered by Euclidean distance.

Figure 2.

Gene set variation and enrichment analyses. (A) Gene set variation analysis (GSVA) scores for the set of genes that were significantly higher in hyperinflammatory acute respiratory distress syndrome (ARDS) in the ALI (Acute Lung Injury in Critical Illness) cohort (false discovery rate [FDR] < 0.1). Each point represents an individual tracheal aspirate transcriptome. GSVA scores in each phenotype were compared using linear models adjusted for age and sex; models in the COMET (COVID-19 Multiphenotyping for Effective Therapies) cohort included a random effect term for subjects to account for repeated measures (*P < 0.05 and ***P < 0.001). GSVA scores for each sample are in Data E3. (B) Selected gene set enrichment analysis scores (GSEA) for pathways in the Reactome database that were significantly enriched (FDR < 0.1) in either phenotype in the ALI and COMET cohorts. A complete list of the 195 pathways that were reproducibly enriched in both pathways is in Data E4. GSEA net enrichment score (NES) is shown on the color bar. The area of the circles is proportional to the absolute NES. A positive NES indicates the Reactome gene set is more highly expressed in hyperinflammatory ARDS. COVID-19 = coronavirus disease.

Figure 3.

Tracheal aspirate (TA) single-cell RNA sequencing. (A) Seurat UMAP projection of 26,429 TA cell transcriptomes from four participants with hyperinflammatory acute respiratory distress syndrome (ARDS) and five participants with hypoinflammatory ARDS, annotated with cell type as predicted by SingleR. (B) UMAP projection of TA cell transcriptomes separated by ARDS phenotype. (C) Differential interaction between cell types predicted by CellChat. Red arrows identify cell pairs with greater strength of interaction in hyperinflammatory ARDS. (D) Differences in strength of ligand–receptor interaction for pathways in the CellChat database. Interaction scores from CellChat are in Data E6. UMAP = uniform manifold approximation and projection.

Figure 4.

Volcano plot comparing OLink normalized protein expression (NPX) for plasma biomarkers from 5 participants with hyperinflammatory acute respiratory distress syndrome (ARDS) and 20 participants with hypoinflammatory ARDS. The log₂-fold difference in mean NPX is shown on the x-axis, and a positive difference indicates higher NPX in hyperinflammatory ARDS. The log₁₀ Benjamini-Hochberg adjusted P value is shown on the y-axis. Significantly different biomarkers are shown in orange and are labeled in the plot. Results for all 73 biomarkers passing quality control are in Data E7.