Immune network dysregulation precedes clinical diagnosis of asthma

Abstract

Allergic asthma is a chronic disease beginning in childhood that is characterized by dominant T-helper 2 cell activation without adequate counter-regulation by T-helper 1 cell and regulatory T cell activity. Prior transcriptomic studies of childhood asthma have primarily investigated subjects who already have a disease diagnosis, and have generally taken an approach of differential gene expression as opposed to differential gene interactions. The immune states that predispose towards allergic sensitization and disease development remain ill defined. We thus characterize immune networks of asthmatic predisposition in children at the age of 2, prior to the diagnosis of allergic asthma, who are subsequently diagnosed with asthma at the age of 7. We show extensive differences of gene expression networks and gene regulatory networks in children who develop asthma versus those who do not using transcriptomic data from stimulated peripheral blood mononuclear cells. Moreover, transcription factors that bind proximally to one another share patterns of dysregulation, suggesting that network differences prior to asthma diagnosis result from altered accessibility of gene targets. In summary, we demonstrate non-allergen-specific immune network dysregulation in individuals long before clinical asthma diagnosis.

Figure 1

Stimulation of peripheral blood mononuclear cells (PBMCs) with tetanus toxoid (TT) perturbs expression of thousands of genes both in controls and asthma. The number of genes that increase expression (upper Venn diagram) and decrease expression (lower Venn diagram) with TT stimulation are shown for the controls (n = 30) and asthma (n = 19). DESeq2 was used to perform differential gene expression analysis with FDR-corrected p < 0.05.

Figure 2

Module eigengenes change significantly with tetanus toxoid (TT) stimulation of peripheral blood mononuclear cells, but demonstrate no group effects. Eigengene expression is displayed for each WGCNA module, separately for controls and asthma, with no stimulation (NS) and TT stimulation of PBMCs. Every module demonstrates a significant effect of stimulation (p < 0.01), but none demonstrate a significant group x stimulation effect (p < 0.05).

Figure 3

Gene module network differences in asthma are characterized primarily by aberrant negative co-regulation. All significantly positively co-regulated modules (q < 0.05, FDR corrected) are connected with blue edges, while negatively co-regulated modules are connected with red edges. Edge thickness corresponds to strength of correlation.

Figure 4

Asthma regulatory networks demonstrate extensive alterations, with both increased and decreased transcription factor (TF) regulation strength. Regulatory strengths outputted from PANDA are displayed for several example TFs in two representative WGCNA modules (MHCI_1 and IL1). The y axis represents regulatory strength in the asthma network while the x axis represents regulatory strength in the control network. Each individual point represents the regulatory strength of the given TF on each of its gene targets in the representative modules. Points above and below the y = x line respectively indicate stronger regulation in asthma relative to controls, and controls relative to asthma.

Figure 5

Transcription factors (TFs) cluster into groups based upon the pattern of their regulatory alteration across gene expression modules. A heatmap of the median regulatory shift of targets within each gene expression module (for a given TF: median value of z_Asthma–z_Control across targets within a module). The x axis represents different WGCNA modules, while TFs are represented on the y axis, with branches colored by TF communities from hierarchical clustering. Blue/positive values represent stronger regulatory control in asthma, while red/negative values represent strong regulatory control in controls.

Figure 6

Transcription factors (TFs) within the same regulatory communities bind in similar locations on differentially methylated regions (DMRs). Binding locations of TFs with binding motifs (based on 80% of maximum confidence) in a DMR in TNFSF13B. The TFs are separated based upon their regulatory cluster memberships as defined in Fig. 4.

Figure 7

Regulatory communities of transcription factors (TFs) exhibit significant clustering based on binding distances within differentially methylated regions (DMRs). The genes on the X axis represent all DMRs from the Reese et al. meta-analysis of CBMCs, which demonstrated perturbation of expression from tetanus toxoid (TT) stimulation. The green dot represents the mean silhouette score (SS) calculated from the binding location distance matrix with TF regulatory clusters. The gold points and density plots represent the null distribution of mean SS from each of 10,000 permutations of the regulatory cluster labels. Asterisks represent statistical significance *p < 0.05; **p < 0.01; ***p < 0.001.