Host and Microbe Blood Metagenomics Reveals Key Pathways Characterizing Critical Illness Phenotypes

Abstract

Rationale: Two molecular phenotypes of sepsis and acute respiratory distress syndrome, termed hyperinflammatory and hypoinflammatory, have been consistently identified by latent class analysis in numerous cohorts, with widely divergent clinical outcomes and differential responses to some treatments; however, the key biological differences between these phenotypes remain poorly understood.Objectives: We used host and microbe metagenomic sequencing data from blood to deepen our understanding of biological differences between latent class analysis-derived phenotypes and to assess concordance between the latent class analysis-derived phenotypes and phenotypes reported by other investigative groups (e.g., Sepsis Response Signature [SRS1-2], molecular diagnosis and risk stratification of sepsis [MARS1-4], reactive and uninflamed).Methods: We analyzed data from 113 patients with hypoinflammatory sepsis and 76 patients with hyperinflammatory sepsis enrolled in a two-hospital prospective cohort study. Molecular phenotypes had been previously assigned using latent class analysis.Measurements and Main Results: The hyperinflammatory and hypoinflammatory phenotypes of sepsis had distinct gene expression signatures, with 5,755 genes (31%) differentially expressed. The hyperinflammatory phenotype was associated with elevated expression of innate immune response genes, whereas the hypoinflammatory phenotype was associated with elevated expression of adaptive immune response genes and, notably, T cell response genes. Plasma metagenomic analysis identified differences in prevalence of bacteremia, bacterial DNA abundance, and composition between the phenotypes, with an increased presence and abundance of Enterobacteriaceae in the hyperinflammatory phenotype. Significant overlap was observed between these phenotypes and previously identified transcriptional subtypes of acute respiratory distress syndrome (reactive and uninflamed) and sepsis (SRS1-2). Analysis of data from the VANISH trial indicated that corticosteroids might have a detrimental effect in patients with the hypoinflammatory phenotype.Conclusions: The hyperinflammatory and hypoinflammatory phenotypes have distinct transcriptional and metagenomic features that could be leveraged for precision treatment strategies.

Keywords: pathogens; phenotypes; sepsis; transcriptomics.

Figure 1.

Gene expression differences between phenotypes identified by transcriptional profiling. (A) Volcano plot highlighting differentially expressed genes between hyperinflammatory and hypoinflammatory phenotypes. Log₂ fold change values are reported on the x-axis. −Log₁₀ P values are represented on the y-axis. Each circle marker corresponds to a gene. Red dots indicate genes significantly differentially expressed. The top 25 (ordered by adjusted P value) differentially expressed genes are highlighted. (B) Enriched pathways based on differentially expressed genes between hyperinflammatory and hypoinflammatory phenotypes. Pathways were selected on the basis of Z score and P value and ordered given their Z score, which is represented on the x-axis. All significant pathways are represented. Z-score values quantify the overlap between differentially expressed genes and a given gene set. Z-score values also incorporate the direction of expression. Pathway terms are listed on the y-axis. The color of each dot represents the direction of the enrichment; a positive value represents a term enriched in the hyperinflammatory phenotype, and a negative value represents a term enriched in the hypoinflammatory phenotype. All reported terms were significantly enriched (absolute Z score > 2). (C) In silico deconvolution on the basis of the LM22 signature. Cell type proportions between the phenotypes for 11 types of cells are represented (x-axis). The different types of cells are listed on the y-axis. Adjusted P values are reported. Differences were tested for using Wilcoxon tests and Benjamini-Hochberg adjustments for multiple comparisons. gd = gamma delta; Hyper = hyperinflammatory phenotype; Hypo = hypoinflammatory phenotype; LCA = latent class analysis; NFAT = nuclear factor of activated T cells; NK = natural killer; PKC = protein kinase C.

Figure 2.

Microbial differences between sepsis phenotypes identified by plasma metagenomics. (A) Nonmetric multidimensional scaling-reduced representation of Bray-Curtis distances demonstrating potential plasma bacterial compositional differences between the hyperinflammatory (Hyper; orange; n = 74) and hypoinflammatory (Hypo; blue; n = 111) phenotypes. (B) The relative dominance of the most abundant pathogen detected by plasma metagenomics in hyperinflammatory versus hypoinflammatory phenotypes. Relative dominance represents the proportion of sequencing reads mapping to the most abundant pathogen in each sample relative to all other bacterial genera. The Wilcoxon-derived P value is reported. (C) Differences in the relative abundance of bacterial pathogens (30, 31) between the hyperinflammatory and hypoinflammatory phenotypes measured in sequencing RPM. The Wilcoxon-derived P value is reported. (D) Differential abundance analysis of sepsis-associated bacterial pathogens (genus level) detected in plasma from patients with hyperinflammatory versus hypoinflammatory sepsis phenotypes. Log₂ fold-change values for hyperinflammatory vs. hypoinflammatory phenotypes are reported on the x-axis. Each one of the top five genera per fold-change direction is a bar on the y-axis. Lighter colored bars indicate genera that did not meet the significance threshold of adjusted P  < 0.05. (E) Proportion of total bacterial reads matching to the Enterobacteriaceae family per phenotype. For each patient, represented by a dot, the proportion of total bacterial reads mapping to the Enterobacteriaceae family is reported on the y-axis. The Wilcoxon P value for the comparison of Hypo to Hyper is reported. LCA = latent class analysis. RPM = reads per million.

Figure 3.

Transcriptional comparison of the Early Assessment of Renal and Lung Injury (EARLI) sepsis phenotypes and other sepsis and/or acute respiratory distress syndrome (ARDS) phenotypes. (A) Summary of datasets comparatively evaluated. Each row corresponds to a dataset with described phenotypes of sepsis and/or ARDS. The names of the cohorts, corresponding condition, external phenotype names, and number of subjects with available gene expression data are listed. (B) Comparison between EARLI hyperinflammatory and hypoinflammatory phenotypes and the SRS1 and SRS2 phenotypes. Each dot represents a gene measured in GAinS and EARLI cohorts. The x-axis coordinates represent the log₂ fold-change values for the comparison of hyperinflammatory with hypoinflammatory phenotypes in EARLI. The y-axis coordinates represent the log₂ fold-change values for SRS1 versus SRS2. On the y-axis, positive values represent genes that are overexpressed in SRS1 when compared with SRS2. Spearman’s correlation coefficients (R) are reported. Pie charts represent a comparison of DE genes. (C) Comparison of EARLI hyperinflammatory and hypoinflammatory phenotypes with the MARS4 and MARS2 phenotypes. (D) Comparison between EARLI hyperinflammatory and hypoinflammatory phenotypes and the reactive and uninflamed phenotypes. DE = differentially expressed.