Rewired type I IFN signaling is linked to age-dependent differences in COVID-19

Abstract

Advanced age is the most important risk factor for severe disease or death from COVID-19, but a thorough mechanistic understanding of the molecular and cellular underpinnings is lacking. Multi-omics analysis of 164 samples from SARS-CoV-2-infected persons aged 1 to 84 years reveals a rewiring of type I interferon (IFN) signaling with a gradual shift from signal transducer and activator of transcription 1 (STAT1) to STAT3 activation in monocytes, CD4⁺ T cells, and B cells with increasing age. Diversion of IFN signaling is associated with increased expression of inflammatory markers, enhanced release of inflammatory cytokines, and delayed contraction of infection-induced CD4⁺ T cells. A shift from IFN-responsive germinal center B (GCB) cells toward CD69^high GCB and atypical B cells during aging correlates with immunoglobulin (Ig)A production in children, whereas complement-fixing IgG predominates in adults. Our data provide a mechanistic basis for inflammation-prone responses to infections and associated pathology during aging.

Keywords: B cells; COVID-19; SARS-CoV-2; STAT1; STAT3; T cells; age; antibodies; children; immune response; monocytes; signaling; type I IFN.

Graphical abstract

Figure 1

Distinct age-dependent patterns of monocyte activation are characterized by opposite HLA-DR and CD11c versus CCR6 expression (A) Overview of the study cohort and methodological pipeline. Blood samples were collected from SARS-CoV-2-infected children and adults as well as controls spanning an age range from 1 to 86 years. Whole-blood mass cytometry (CyTOF) and scRNA-seq combined with VDJ (Variable, Diversity, and Joining)-seq-based clonotype identification were used to determine age-specific alterations in the monocyte, T cell, and B cell compartment. The obtained results together with serum antibody profiles were used to develop hypotheses on their functional properties, inducing mechanisms and transcriptional control, which were tested in ex vivo cultures. Detailed information on samples included in all reported assays can be found in Table S1. Additional cohort summary is included in Table S2. Multiple icons are used throughout the paper to identify data from different experiments (Methods legend in the lower right corner of the figure). (B) UMAP (Uniform Manifold Approximation and Projection), showing pre-gated monocyte and dendritic cell (DC) populations from the CyTOF dataset. 18 clusters have been produced using a semi-supervised approach with FlowSOM algorithm. (C) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (D) Boxplots, showing relative abundance of infection-induced clusters resulting from the FlowSOM algorithm, calculated per sample within all monocytes and DCs from the CyTOF data. Kruskal-Wallis + Wilcoxon p values. (E) Scatterplots, showing mean Z score (calculated by subtracting population mean from value divided by standard deviation) normalized HLA-DR and CCR6 expression in relationship to patient age for infected patients (linear model fitted to data and Spearman’s rank correlation coefficient in black) within monocyte and DC CyTOF data.

Figure 2

Gradual age-dependent change in the phenotype of infection-induced CD4⁺ T cells and B cells (A) UMAP, showing pre-gated CD4⁺ T cells from the CyTOF dataset. 19 clusters have been produced using a semi-supervised approach with FlowSOM algorithm. UMAP presents all the patients that were part of the dataset and used for clustering, including follow-up measurements of some patients done approximately 2 weeks and 6 months after the first, acute infection phase measurement (details in Table S1). (B) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (C) Boxplots, showing relative abundance of infection-induced clusters resulting from the FlowSOM algorithm, calculated per sample within all CD4⁺ T cells from the CyTOF data. Kruskal-Wallis + Wilcoxon p value. (D) Scatterplots, showing mean Z score normalized CD38 and CCR6 expression in relationship to patient age for infected patients (linear model fitted to data and Spearman’s rank correlation coefficient in black) within CD4⁺ T cell CyTOF data. (E) UMAP, showing pre-gated B cells from the CyTOF dataset. 15 clusters have been produced using a semi-supervised approach with FlowSOM algorithm. (F) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (G) Boxplots, showing relative abundance of infection-induced clusters resulting from the FlowSOM algorithm, calculated per sample within all B cells from the CyTOF data. Kruskal-Wallis + Wilcoxon p values. (H) Scatterplots, showing mean Z score normalized CXCR5 and CD69 expression in relationship to patient age for infected patients (linear model fitted to data and Spearman’s rank correlation coefficient in black) within B cell CyTOF data. (I) Time-dependent stacked line graphs, displaying the relative mean cluster abundance of all activated CD4⁺ T cell clusters determined by CyTOF. Activated clusters were defined as clusters having above-average Z-scored expression of activation markers (CD25, HLA-DR, CD38, CD137, CD69, and Ki67) compared to other clusters. Patient group color-coded figurines on the right-hand side point out cluster accumulation patterns. (J) Scatterplot, showing the sum of the relative abundance of all activated CD4⁺ T cell clusters determined by CyTOF.

Figure 3

Age-dependent shift from type I IFN responsiveness to increased inflammatory potential (A) UMAP, showing monocyte cells, subset from the PBMC scRNA-seq data. 11 clusters have been produced using a graph-based approach as implemented in Seurat package (k-nearest-neighbors [KNN] graph with Louvain community detection). Some clusters share annotation due to phenotypical similarity and accumulation pattern and are treated as one population for abundance testing in S3E. Clusters annotated as “dropped” were of low quality (as concluded from inspecting number of features, no. of counts, and percentage of mitochondrial genes as well as other population-specific genes). (B) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (C) Dotplot, showing scaled average expression of genes in monocyte clusters, increased with infection (1, 4, and 7) in cells belonging to infected patients. A horizontal line splits the dotplot in two parts; genes above the line were curated based on the presence of clusters with pronounced ISG signature and include other genes useful for annotation; genes below the line were found to be differentially expressed between the clusters (FindMarkers Seurat function). (D) Volcano plot, showing the results of differential expression (DE) analysis of monocytes from clusters expanded with infection (1, 4, and 7) using DESeq2. Significantly enriched genes (baseMean counts over 50 and adjusted p value < 0.05, a total of 189 genes) are colored green, and 20 significant genes with the highest absolute log2 fold change are labeled. Additional significant genes, tangential to other findings, were labeled ad hoc.

Figure 4

Molecular characterization of infection-associated TCRab⁺ T cells and B cells indicates increased STAT3 involvement and altered differentiation during aging (A) UMAP, showing TCRab⁺ T cells, subset from the PBMC scRNA-seq data. 21 clusters have been produced using a graph-based approach as implemented in Seurat package (KNN graph with Louvain community detection). Clusters annotated as “dropped” were of low quality (as concluded from inspecting number of features, no. of counts, and percentage of mitochondrial genes as well as other population-specific genes) (details in Table S1). (B) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (C) Dotplot, showing scaled average expression of genes in TCRab⁺ T cell clusters, increased with infection (2, 7, 8, 12, 15, 16, 17, and 18) for cells of infected patients. A horizontal line splits the dotplot in two parts; genes above the line were curated based on the presence of clusters with pronounced ISG signature and include other genes useful for annotation; genes below the line were found to be differentially expressed between the clusters (FindMarkers Seurat function). (D) Volcano plot, showing the results of differential expression (DE) analysis of TCRab⁺ T cells from clusters expanded with infection (2, 7, 8, 12, 15, 16, 17, and 18) using DESeq2. Significantly enriched genes (baseMean counts over 50 and adjusted p value < 0.05, a total of 464 genes) are colored green, and 20 significant genes with the highest absolute log2 fold change are labeled. Additional significant genes, tangential to other findings, were labeled ad hoc. (E) Gene set enrichment analysis (GSEA) applied to TCRab⁺ T cells from the PBMC scRNA-seq experiment and clusters, expanded with infection (2, 7, 8, 12, 15, 16, 17, and 18). GSEA was done using genes from the infected adults—infected children comparison with baseMean counts > 50 (total of 5,181 genes). R package fgsea was used with Reactome gene list R-HSA-1169410 (Table S4). Genes are shown as ticks on a bold black line and ranked by log2 fold change. (F) UMAP, showing B cells, subset from the PBMC scRNA-seq data. 13 clusters have been produced using a graph-based approach as implemented in Seurat package (KNN graph with Louvain community detection). Some clusters share annotation due to phenotypical similarity and accumulation pattern and are treated as one population for abundance testing in S4E. Clusters annotated as “dropped” were of low quality (as concluded from inspecting number of features, no. of counts, and percentage of mitochondrial genes as well as other population-specific genes). (G) UMAPs, showing the location of cells belonging to the respective group (colored in, whereas gray identifies cells in other groups). Gray outlines indicate cluster regions enriched in infected children or adults. (H) Dotplot, showing scaled average expression of genes in B cell clusters, increased with infection (4, 5, 6, 8, and 12) for cells of infected patients. A horizontal line splits the dotplot in two parts; genes above the line were curated based on the presence of clusters with pronounced ISG signature and include other genes useful for annotation; genes below the line were found to be differentially expressed between the clusters (FindMarkers Seurat function). (I) Volcano plot, showing the results of differential expression (DE) analysis of B cells from clusters expanded with infection (4, 5, 6, 8, and 12) using DESeq2. Significantly enriched genes (baseMean counts over 50 and adjusted p value < 0.05, a total of 482 genes) are colored green, and 20 significant genes with highest absolute log2 fold change are labeled. Additional significant genes, tangential to other findings, were labeled ad hoc. (J) Stacked bar chart, showing proportions of cluster labels, given to the cells with callable B cell receptor (BCR) clonal identity that overlaps with at least one of the cells in the plasmablast population (cluster 12), calculated over all expanded B cell clusters.

Figure 5

Consequences for local T cell responses and generated antibody profiles (A) Dotplot, showing scaled average expression of genes in TCRab⁺ T cells, subset from the nasal swab scRNA-seq data. Clusters, increased with infection (0, 4, 5, 6, and 7). A total of 8 clusters have been produced using a graph-based approach as implemented in Seurat package (KNN graph with Louvain community detection). A horizontal line splits the dotplot in two parts; genes above the line were curated based on the presence of clusters with pronounced ISG signature and include other genes useful for annotation; genes below the line were found to be differentially expressed between the clusters (FindMarkers Seurat function). (B) Scatterplots showing CD38 and TNF genes transcription (average scaled expression in clusters 0, 4, 5, 6, and 7, expanded with infection) for each donor, plotted against donor’s age, using TCRab⁺ T cells, subset from nasal swab scRNA-seq data. Linear models fitted to the data points and Spearman’s rank correlation coefficients. (C) Stacked bar chart showing relative expression strength of heavy-chain genes encoding for the different IgG and IgA isotypes in plasmablasts (B cell cluster 12, PBMC scRNA-seq experiment). Plasmablasts, expressing either of the heavy-chain genes, were pre-selected. Expression values for each gene were calculated and normalized to the total expression of all heavy-chain genes. (D) Boxplots of S1-specific IgG (left) and IgA (right) antibody titers for the acute infection phase and follow-up measurements done approximately 2 weeks and 6 months later. Titers for second and third time points are normalized to the first time point for each patient (fold change and ratio). Wilcoxon p values.

Figure 6

Mechanistic in vitro studies link age-dependent rewiring of type I IFN responsiveness with in vivo-detected opposite activation profiles (A) Overview of the workflow used to study the responsiveness to type I IFN and IL-1b. PBMCs from uninfected children and adults were stimulated with either SEB or a combination of SEB, ODN CpG2216, B18R, and recombinant IFNa. In a parallel experiment series, different concentrations of IFNa as well as combinations of SEB, ODN CpG2216, IL-1b, and IL-1b inhibitor anakinra were tested. After 4 days of incubation, phenotypic differences in activation marker expression were determined by flow cytometry, while cell culture supernatants were used for cytokine and chemokine quantification. Experiments focused on IL-1b and anakinra influence were only measured in cytokine proteomics. (B) Boxplot of arcsinh-transformed median CD38 fluorescence intensity in proliferating CD4⁺ T cells, showing influence of CpG2216-mediated activation on CD38 expression. Wilcoxon test p values. Dropout in uninfected children group SEB condition is due to low cell number. (C) Boxplot of arcsinh-transformed median CD38 fluorescence intensity in proliferating CD4⁺ T cells, showing influence of CpG2216-mediated activation and IFNa (30 ng/mL) on CD38 expression in children. Wilcoxon test p values. Dropouts in SEB and SEB+IFNa perturbations are due to low cell number. (D) Boxplots of CD38 median signal intensity in proliferating CD4⁺ T cells, separated into CD45RA⁻ (violet filling) memory and CD45RA⁺ naive subpopulations, showing the difference in CD38 upregulation in response to CpG2216-mediated activation and IFNa release between memory and naive CD4⁺ T cells. Wilcoxon test p values. Dropout in uninfected children, SEB perturbation is due to low cell number. (E) Boxplot of IFNa concentration measured in cell culture supernatant and normalized to values detected in SEB condition for each patient, showing the effectiveness of CpG2216 in provoking IFNa release as well as of B18R in reducing the concentration of soluble IFNa. Dropout in uninfected children is due to low cell number. (F) Heatmap, showing scaled average log concentration of the 18 cytokines measured in co-culturing experiments for different perturbations using PBMC. (G) Line plots, showing the dependence of IFNg, IL-21, and IL-1b concentrations on the IFNa concentration. Wilcoxon p values. (H) Scatterplot, illustrating the correlation between donor age and IL-1b concentration in supernatant when PBMCs are stimulated with SEB and CpG.

Figure 7

Altered type I IFN signaling and gradual involvement of STAT3 activation in stimulated T and B cells of older individuals (A) Overview of the workflow studying the phosphorylation dynamics of STAT1 and STAT3 in T cells and B cells from children and adults. PBMCs from control children and adults were stimulated with either SEB or a combination of SEB and recombinant IFNa. After 25 min of incubation, levels of phosphorylated STAT1 and STAT3 were determined by flow cytometry for pre-gated populations. (B) Boxplots, showing the ratio of pSTAT1 to pSTAT3 (based on measured median fluorescence intensity) for CD4⁺ T cells and B cells of control children and adults following 25-min incubation with SEB or SEB and IFNa. Wilcoxon p value. (C) Scatterplots of STAT1/STAT3 transcription ratio and SOCS3 transcription (average scaled expression in clusters expanded with infection) for each donor, plotted against donor’s age, using PBMC scRNA-seq experiment data (details in Table S1). CD4⁺ T cells were a subset of the total TCRab⁺ pool using CD8A and CD8B genes (assigned CD4 FLAG, if cell is negative for both). Spearman’s rank coefficients. (D) Boxplots showing relative abundance of infection-induced clusters 11 and 15 from the FlowSOM algorithm, calculated per sample within all CD4⁺ T cells from the CyTOF data. Wilcoxon p values. (E) Boxplots, showing relative abundance of infection-induced clusters 8 and 11 resulting from the FlowSOM algorithm, calculated per sample within all B cells from the CyTOF data. Wilcoxon p values.