Consensus clustering of gene expression profiles in peripheral blood of acute ischemic stroke patients

Abstract

Acute ischemic stroke (AIS) is a primary cause of mortality and morbidity worldwide. Currently, no clinically approved immune intervention is available for AIS treatment, partly due to the lack of relevant patient classification based on the peripheral immunity status of patients with AIS. In this study, we adopted the consensus clustering approach to classify patients with AIS into molecular subgroups based on the transcriptomic profiles of peripheral blood, and we identified three distinct AIS molecular subgroups and 8 modules in each subgroup by the weighted gene co-expression network analysis. Remarkably, the pre-ranked gene set enrichment analysis revealed that the co-expression modules with subgroup I-specific signature genes significantly overlapped with the differentially expressed genes in AIS patients with hemorrhagic transformation (HT). With respect to subgroup II, exclusively male patients with decreased proteasome activity were identified. Intriguingly, the majority of subgroup III was composed of female patients who showed a comparatively lower level of AIS-induced immunosuppression (AIIS). In addition, we discovered a non-linear relationship between female age and subgroup-specific gene expression, suggesting a gender- and age-dependent alteration of peripheral immunity. Taken together, our novel AIS classification approach could facilitate immunomodulatory therapies, including the administration of gender-specific therapeutics, and attenuation of the risk of HT and AIIS after ischemic stroke.

Keywords: acute ischemic stroke; consensus clustering; immunosuppression; molecular subgroups; peripheral immunity; restricted cubic spline functions.

Figure 1

Consensus clustering analysis of gene expression profiles of peripheral blood samples in patients with AIS. (A) Box plots showing the normalized relative expression (y-axis) of data from genesets GSE16561 and GSE37587. (B) The heatmap representing the consensus matrix with a cluster count of 3, which was determined by the minimal consensus scores of >0.8. (C) Bar-plots showing the consensus scores (y-axis) and the corresponding numbers of the subgroup. The maximum number of subgroups was set to 10.

Figure 2

Identification of subgroup-specific differentially expressed genes and comparison with control. (A–C) GSEA enrichment plots (red lines) showing the subgroup-specific upregulated gene (Specific-up) and subgroup-specific downregulated gene (Specific-down) in subgroup I (A), subgroup II (B), and subgroup III (C). Green lines represent the differentially expressed genes between the corresponding subgroup and the normal controls. NES denotes the normalized enrichment score with FDR (** <0.01, and *** <0.001).

Figure 3

Clinical characteristics within subgroups. (A) Barplot showing the proportion of male patients in each subgroup. (B,C) Barplot showing the age distribution in each subgroup in male (B) and female patients (C). Values are means ± SD. NS indicates no significant difference between groups (p > 0.05). *P < 0.05, **P < 0.01, and ***P < 0.001. (D) Non-linear relationship between subgroup and the age of female patients analyzed by restricted cubic spline model. The black line represents the pooled odds ratio (OR), and the red lines indicate a 95% confidence interval (CI). (E,F) Box-plots showing the proportion of sampling time (within 24 h and between 24 and 8 h) after AIS in each subgroup of female (E) and male (F) patients. NS indicates no significant difference between groups (p > 0.05).

Figure 4

GSEA analysis of subgroup-specific differentially expressed genes. (A) The enrichment plot illustrates that most subgroup-specific genes had a significantly non-linear relationship with female age. The normalized enrichment score (NES) and FDR (*** <0.001) in the female group. (B) The enrichment plot illustrates that most subgroup-specific genes had a significantly non-linear relationship with male age. The normalized enrichment score (NES) is shown without significant difference (FDR > 0.05).

Figure 5

Identification and characterization of 8 modules by weighted gene co-expression analysis (WGCNA). (A) Module-trait relationship of gender and age in male and female patients. The positive and negative correlation coefficients of WGCNA modules and clinical characteristics are colored from red to blue. Each row corresponds to a module eigengene, each column corresponds to a trait. Each cell depicts the corresponding Pearson correlation r-values and p values in the bracket. (B) The scaled expression values of genes in each of the 8 WGCNA modules are displayed in the heatmap. (C) Alluvial diagram showing the inherent relationship between modules and subgroups.

Figure 6

(A) Heatmap showing association of gender-specific distribution of age with the expression of subgroup-specific genes in each module. Red and blue indicate upregulation and downregulation, respectively. Samples were ordered by increasing age in each gender. (B,C) The enrichment plots of the restricted cubic spline-GSEA showing the significance of the non-linear relationship between all normalized expression of genes in each module and log-transfer age in female patients (B) and male patients (C).

Figure 7

Module-based pathway analysis. Visualization of pathways identified by the Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis.

Figure 8

Heatmap showing the scaled expression values of genes in representative subgroup-specific KEGG pathways related to inflammation and corresponding signature of each co-expression module among subgroups. (A) Chemokine signaling pathway and cytokine-cytokine receptor interation. (B) C-type lectin receptor signaling pathway, TNF signaling pathway, and complement and coagulation cascades. Heatmap colors correspond to the level of mRNA expression as indicated in the color rang.

Figure 9

Module-based pathway analysis. Visualization of pathways identified by the Reactome.

Figure 10

Heatmap showing the expression level of genes in representative subgroup-specific Reactome pathways related to inflammation as indicated in Signaling by Interleukins (A) and TNFs bind their physiological receptors, Chemokine receptors bind chemokines, Peptide ligand-binding receptors, Phosphorylation of CD3 and TCR zeta chains, Formation of Fibrin Clot Formation, and Relation of gene expression by Hypoxia-inducible Factor (B) and corresponding signature of each co-expression module among subgroups, heatmap colors correspond to the level of mRNA expression as indicated in the color rang.

Figure 11

Genes in modules in subgroup I overlaps with hemorrhage transformation (HT)-specific differentially expressed genes (DEG). (A) Pre-ranked gene set enrichment analysis showing the enrichment of DEG related to HT in the co-expression modules with subgroup I-specific signature genes. (B) Venn diagram showing the number of overlapped genes (267) between HT-specific DEG and subgroup I-specific genes. (C) Heatmap representing the overlap genes between DEGs related to HT and subgroup I-specific genes. (D) Diagram showing pathways identified by Reactome with the negative log₁₀ p-values of the overlapped genes shown in B and C.