Fig 1.
Flowchart of microarray data selection.
Data series collected from GEO and ArrayExpress were subjected to a selection process resulting in 14 data series from 12 studies. Samples of patients with sepsis and healthy controls were further assessed to meet various standards for analysis.
Table 1.
Overview of included studies with indication of available metadata and number of samples (both retrievable and published).
BS: Bloodstream, CAP: community-acquired pneumonia, FP: fecal peritonitis.
Fig 2.
Hierarchical cluster analysis of microarray expression data provided for the 5,000 most-variable gene symbols in 949 patients of the septic group.
(A) Dendrogram illustrating the arrangement of the clusters produced: Cluster 1 (gray) comprises 655 subjects, while cluster 2 (cyan) includes 294 individuals. (B) Scatterplot showing the amount of data variance explained by the first three principal components. Subjects are colored according to their respective cluster assignment. (C) Volcano plot showing the gene symbols differentially expressed (solid black color highlights results with absolute logFC ≥ 1, adjusted p-value < 0.05). Resulting number of significant genes above the defined absolute logFC threshold are indicated by numbers (orange: up-regulated, blue: down-regulated). (D) Heatmap depicting processed expression values of the 33 differentially regulated genes for all cluster-assigned individuals.
Table 2.
Top five up- and down-regulated genes according to the different comparisons.
Log2FC: Log2 fold change; adj.: adjusted; HPGD: 15-hydroxyprostaglandin dehydrogenase; IL18R1: Interleukin-18 Receptor I; OLAH: Oleoyl-ACP Hydrolase; CD: Cluster of Differentiation; MMP8: Matrix metalloproteinase 8; GZMH: Granzyme H; GZMK: Granzyme K; HP: Haptoglobin; RETN: Resistin; GNLY: Granulysin; IL7R: Interleukin-7 Receptor; IL2RB: Interleukin 2 Receptor Subunit Beta; HLA-DRB1: Human Leucocyte Antigen DRB1; IFI27: Interferon Alpha Inducible Protein 27; CHI3L1: Chitinase 3-Like 1; TRBC1: T Cell Receptor Beta Constant 1; CCR3: C-C Motif Chemokine Receptor 3; TRDV2: T Cell Receptor Delta Variable 2.
Fig 3.
Differential expression analysis between “Cluster 1” and individuals from the healthy control group.
(A) Heatmap of processed expression values for 368 dysregulated genes showing absolute logFC ≥ 1 (adj. p-value < 0.05). (B) Volcano plot of differentially expressed genes (solid black color indicates absolute logFC ≥ 1.0, adj. p-value < 0.05; numbers indicate up- (orange) or down-regulated (blue) genes). Results of GO-term analysis for enriched biological processes separately for up-regulated (top panel) and down-regulated genes (bottom panel) above defined threshold.
Fig 4.
Differential expression analysis between “Cluster 2” and individuals from the healthy control group.
(A) Volcano plot of differentially expressed genes (solid black color indicates absolute logFC ≥ 1.0, adj. p-value < 0.05; numbers indicate up- (orange) or down-regulated (blue) genes) and results of GO-term analysis for enriched biological processes separately for down-regulated (left panel) and up-regulated genes (right panel) above defined threshold (B).
Fig 5.
Networks of differentially expressed genes between defined clusters and healthy controls.
(A) Top network of differentially regulated genes between patients of “Cluster 1” and healthy individuals. (B) Manually selected network consisting of differentially regulated immune-related genes. Nodes showing an orange color implicate up-regulation for the conditions in contrast, while blue elements represent down-regulation. (C) and D) Overlays of the respective expression data for “Cluster 2” subjects in comparison to individuals from control group. (E) Heatmap showing the results of data deconvolution to identify cell origin of signals. Orange color represents an up-regulated “cell abundance”, representing more signals deriving from this cell type compared to healthy controls, blue color vice versa. Only informative cell types were visualized. Mega: megakaryocyte; Ery: erythroid; HSC: hematopoetic stem cell.
Fig 6.
Hierarchical cluster analysis of microarray expression data provided for the 5,000 most-variable gene symbols in the full dataset of 1084 subjects.
Dendrogram and color track (origin) illustrate the sample re-arrangement of the clusters produced: New cluster 1 (blue) consists of 839 subjects, while 245 individuals are attributed to cluster 2 (green). In comparison to Fig 2A, new cluster 1 identity is unchanged in most cases (99.4% retention rate), while adding new subjects from original cluster 2. The 245 individuals assigned to new cluster 2 cover both original cluster 2 samples as well as the vast majority of healthy controls (96.3% assignment rate).
Fig 7.
Pathophysiological model of sepsis genomic response.
The early blindspot of sepsis (blue box) spans the highly individual timeframe from infection to clinical manifestation of symptoms. The quantitative and qualitative kinetic of response depends on both host and pathogen attributes. Our results originating from samples of patients early after ICU admission for sepsis prove the presence of (at least) two molecular signatures of sepsis (Cluster 1 and Cluster 2), with Cluster 1 implicating a higher degree of dysregulation towards immunosuppression than Cluster 2. Within the clusters, different cell types are likely to have contradictory or even ambivalent activation states, e.g. monocytes (Mo) with impaired cytokine production but with maintained migratory function. Neut: neutrophilic granulocytes; NK: natural killer cells; T: T cells.