Atopic Dermatitis Is an IL-13-Dominant Disease with Greater Molecular Heterogeneity Compared to Psoriasis

Abstract

Atopic dermatitis (AD) affects up to 20% of children and adults worldwide. To gain a deeper understanding of the pathophysiology of AD, we conducted a large-scale transcriptomic study of AD with deeply sequenced RNA-sequencing samples using long (126-bp) paired-end reads. In addition to the comparisons against previous transcriptomic studies, we conducted in-depth analysis to obtain a high-resolution view of the global architecture of the AD transcriptome and contrasted it with that of psoriasis from the same cohort. By using 147 RNA samples in total, we found striking correlation between dysregulated genes in lesional psoriasis and lesional AD skin with 81% of AD dysregulated genes being shared with psoriasis. However, we described disease-specific molecular and cellular features, with AD skin showing dominance of IL-13 pathways, but with near undetectable IL-4 expression. We also demonstrated greater disease heterogeneity and larger proportion of dysregulated long noncoding RNAs in AD, and illustrated the translational impact, including skin-type classification and drug-target prediction. This study is by far the largest study comparing the AD and psoriasis transcriptomes using RNA sequencing and demonstrating the shared inflammatory components, as well as specific discordant cytokine signatures of these two skin diseases.

Figure 1.. Transcriptomes of different skin types.

a) the top three principal components for the samples in the cohort; b) euclidean distance (in logarithmic scale) distributions between all pairwise samples within the AD and Psoriasis skin conditions. The distance was computed using the top 3 principal components; c) Venn Diagram showing the overlap between the differentially expressed genes in lesional skin; d) number of differentially expressed genes in lesional skin; e-f) scatter plots illustrating the concordance between lesional (e) and non-lesional (f) skin for psoriasis and AD, coloring genes differentially expressed in both axis (red), only in the x-axis (orange), or only in the y-axis (blue); g) proportion of differentially expressed genes under each category for different differential expression conditions.

Figure 2.. Functional analysis for differentially expressed genes.

a) Biological functions with corresponding genes significantly overlapped (-log(p-value) for each function is shown) against genes commonly differentially expressed in AD and psoriatic lesional skin; b) functional enrichment results between the psoriasis-only (x-axis) versus AD-only (y-axis) DEGs, each point represents one function; c) top significant functions enriched among genes differentially expressed only in psoriatic lesional skin but not in AD lesional skin; d) significant functions encompassing genes differentiating AD and Psoriasis, and the genes are differentially expressed AD lesional skin but not in psoriatic skin.

Figure 3.. Gene expressions for cytokines.

a) Heatmap illustrates the expression levels of different cytokines (rows; stratified by their families) across different samples (columns; stratified by different skin types); b) boxplots to illustrate the expression distributions of six cytokines across the different skin conditions.

Figure 4.. Cell type signatures in different skin types.

a) Heatmap illustrates the enrichment (negative logarithm of enrichment p-value) of differentially expressed genes (columns for different comparisons) against nearby H3k27ac marks specific in different immune cells (rows). * indicates FDR≤5%; b) proportions (%) of CD8 cells under different skin types (in logarithmic scale).

Figure 5.. Cytokine-induced signatures.

a) Heatmap shows the correlation between fold changes of dysregulated genes under cytokine-stimulated keratinocytes (rows) and fold changes of the DE analysis conducted in this study (columns). *indicates FDR≤5% and correlation coefficient ≥0.25; b) the scatter plot projects how each sample responds to IL-17A, IL-13, and IL-4 stimulations.

Figure 6.. Translational implications of this transcriptomic study.

a) Area under ROC to evaluate the performance of classifying different skin types using machine learning approach; b) & c) for each gene, the effect size in control versus lesional skin comparison (x-axis) versus its Spearman correlation between the expression level and severity index (y-axis); d) for each gene, the comparison between the PASI correlation (x-axis) versus SCORAD correlation (y-axis). Black line indicates lowess fit using all genes (grey); red line represents lowess fit using only common DEGs (blue) in both AD and psoriatic lesional skin).