www.nature.com/scientificreports OPEN Transcriptomic insight into the translational value of two murine models in human atopic dermatitis Young‑Won Kim1,5, Eun‑A Ko2,5, Sung‑Cherl Jung2, Donghee Lee1, Yelim Seo1, Seongtae Kim1, Jung‑Ha Kim3, Hyoweon Bang1, Tong Zhou4* & Jae‑Hong Ko1* This study sought to develop a novel diagnostic tool for atopic dermatitis (AD). Mouse transcriptome data were obtained via RNA‑sequencing of dorsal skin tissues of CBA/J mice afected with contact hypersensitivity (induced by treatment with 1‑chloro‑2,4‑dinitrobenzene) or brush stimulation‑ induced AD‑like skin condition. Human transcriptome data were collected from German, Swedish, and American cohorts of AD patients from the Gene Expression Omnibus database. edgeR and SAM algorithms were used to analyze diferentially expressed murine and human genes, respectively. The FAIME algorithm was then employed to assign pathway scores based on KEGG pathway database annotations. Numerous genes and pathways demonstrated similar dysregulation patterns in both the murine models and human AD. Upon integrating transcriptome information from both murine and human data, we identifed 36 commonly dysregulated diferentially expressed genes, which were designated as a 36‑gene signature. A severity score (AD index) was applied to each human sample to assess the predictive power of the 36‑gene AD signature. The diagnostic power and predictive accuracy of this signature were demonstrated for both AD severity and treatment outcomes in patients with AD. This genetic signature is expected to improve both AD diagnosis and targeted preclinical research. Patients with atopic dermatitis (AD) ofen exhibit an itchy rash, xerosis, skin barrier defects, chronic relapses, and emotional distress, which reduces their quality of life1. Te diagnosis of AD is generally based on visible clinical symptoms, with limited therapeutic options available for this condition. Te most common diagnostic criteria and severity scoring tools are the Hanifn and Rajka criteria2 and the SCORing AD (SCORAD) index3, respectively. Various murine models have been developed for studying AD; however, their ability to recapitulate the pathophysiological features and complex clinical manifestations of human AD is limited. In the contact hyper- sensitivity (CHS) model, hapten 1-chloro-2,4-dinitrobenzene is applied to the skin to stimulate keratinocytes, which produce various biochemical mediators, such as interleukin (IL)-1β and tumor necrosis factor (TNF)-α4. Tese responses promote the migration and maturation of dermal dendritic cells, which then migrate to drain- ing lymph nodes, presenting contact allergens to naïve T cells. In the skin-scratching stimulation (SSS) model, mice exhibit a temporary self-scratching behavior within a few minutes of brush stimulation. Tis leads to the physiological stimulation of the skin via activation of the substance P signaling pathway, following the binding of tachykinin receptor 1 5,6. Using these murine models, we previously suggested—from a molecular genetic per- spective—that itching is caused by induction of damage to the chemical/physical skin barrier, which is related to the rate of wound healing, particularly in the case of infammatory reactions, and pain signal intensity7. We also noted that pruritus and a skin barrier disorder were representative symptoms of AD. Terefore, we employed these two pruritus murine models, which demonstrate early stages of skin reactions, rather than the chronic AD NC/Nga mouse model8. In this study, we developed objective AD criteria based on molecular signatures that can be applied as poten- tial tools for improving the accuracy of AD diagnosis and evaluating AD treatment outcomes 9. 1Department of Physiology, College of Medicine, Chung-Ang University, Seoul 06974, Korea. 2Department of Physiology, School of Medicine, Jeju National University, Jeju 63243, Korea. 3Department of Family Medicine, College of Medicine, Chung-Ang University Hospital, Seoul 06973, Korea. 4Department of Physiology and Cell Biology, University of Nevada, Reno School of Medicine, Reno, NV 89557, USA. 5These authors contributed equally: Young-Won Kim and Eun-A Ko. *email: [email protected]; [email protected] Scientifc Reports | (2021) 11:6616 | https://doi.org/10.1038/s41598-021-86049-w 1 Vol.:(0123456789) www.nature.com/scientificreports/ Cxcl1 Cxcl2 2 0 3 * * VT CHS Cxcl3 01 ) ) 2 NT SSS 2 Csf3 S100a8 81 81 Cxcl2 * 012 Il1b xpression (log Lcn2 S100a9 xpression (log e PC2 (21.1%) Uox e Cxcl1 Pla2g4d ve −1 Upp1 0246 ) Relativ Relati 02468 −2 −2 . NT Mmp13 −2 −1 0123 Hist1h2ah S S VT VT PC1 (24.1%) NT NT SSS SSS Itgam CH CH (SSS vs Gzmb FC Cxcl3 Csf3 2 Pla2g4b 2 g Pla2g4e * * lo Fgf7 4 0 ) ) 2 2 81 . NT) 81 0246 pression (log pression (log 04 (SSS vs ex ex Sfrp2 FC Bcl2a1b Entpd3 Lif Ngf 2 ve ve −2 g Bcl3 Fosl1 Mapk6 Ripk3 −4 lo r = 0.281 Capn2 Il33 Mlkl Sele 0123 −10 Creb3l3 Isg15 Myd88 Tuba1c Relati Relati P < 10 Wnt10b 0246 −8 −1 −8 −4 04812 −2 02468 S S VT VT log2FC (CHS vs. VT) log2FC (CHS vs. VT) NT NT SS SS CHS CHS Figure 1. Diferential gene expression in contact hypersensitivity (CHS) and skin-scratching stimulation (SSS) murine models. (a) Principal component analysis of whole-genome expression. (b) Correlation of log2- transformed gene expression fold-changes (log2FC) between the CHS and vehicle control (VT) groups (X-axis), and SSS and non-treated control (NT) groups (Y-axis). Pink/blue dots denote genes commonly upregulated/ downregulated in CHS and SSS samples relative to their expression in the VT and NT samples, respectively. (c) Top dysregulated genes within prioritized Kyoto Encyclopedia of Genes and Genomes pathways. X-axis: log 2FC values between the CHS and VT groups; Y-axis: log2FC values between the SSS and NT groups. Red/blue dots denote genes commonly upregulated/downregulated in both murine models, and the orange dots denote genes dysregulated in opposite directions. (d) qPCR validation of selected genes. *Indicates the statistical signifcance at P < 0.05. Results Gene dysregulation patterns in CHS and SSS murine models. We prepared 16 polyA-enriched RNA-seq libraries of mouse skin samples with four biological replicates per group (vehicle control, VT; non- treated control, NT; CHS model, and SSS model). In total, 13,259 genes were identifed with average expression levels > 1 transcript per million. To assess transcriptome heterogeneity within the diferent murine models, we conducted principal component analysis of the whole-genome gene expression data, which revealed distinct transcriptome patterns between the VT and CHS and NT and SSS samples (Fig. 1a). However, the frst and second principal components of the VT and NT samples did not difer signifcantly (Fig. 1a), suggesting simi- lar transcriptomic landscapes for these two controls. To identify diferentially expressed genes (DEGs) in both murine models, we compared gene expression patterns between the VT and CHS samples, and between the NT and SSS samples. Using the following cut-ofs: a false discovery rate (FDR) < 5% and fold-change (FC) > 2, 993 upregulated and 1,214 downregulated DEGs were detected in the CHS samples, relative to the VT samples (Supplementary Tables S1 and S2). Comparatively, 1,608 and 999 DEGs were upregulated and downregulated, respectively, in the SSS samples, relative to the NT samples (Supplementary Tables S3 and S4). Interestingly, FC gene expression values in the VT and CHS groups were positively correlated with those observed between the NT and SSS groups (Pearson’s correlation (r) = 0.281, P < 10−10). In both models, 292 and 293 DEGs were com- monly upregulated and downregulated, respectively (Fig. 1b), suggesting that a considerable number of DEGs shared similar dysregulation patterns in the two murine models. Dysregulated pathways in the murine models. To investigate transcriptomic alterations in the two murine models, we examined dysregulated pathways based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database annotations 10–12. For each pathway, we obtained a pathway score using the FAIME algorithm, with a higher pathway score indicating higher overall expression. In total, 39 upregulated and 49 downregulated KEGG pathways (t-test: corrected P < 0.05) were detected in CHS samples compared with VT samples (Supplementary Tables S5 and S6). Additionally, 46 and 23 pathways were found to be upregulated and downregulated, respectively, in SSS samples, relative to the NT samples (Supplementary Tables S7 and S8). We then investigated the dysregulated pathways shared between the two murine models. Eleven KEGG pathways were commonly upregulated in both models, including TNF signaling pathway, IL-17 signaling pathway, RIG- I-like receptor signaling pathway, apoptosis, and necroptosis (Supplementary Fig. S1). Lipoic acid metabolism and Wnt signaling pathway, were commonly downregulated in the two murine models (Supplementary Fig. S1). Interestingly, the Rap1 signaling pathway was dysregulated in a contradictory manner in the two murine mod- els (downregulated in the CHS model and upregulated in the SSS model) (Supplementary Fig. S1). We further Scientifc Reports | (2021) 11:6616 | https://doi.org/10.1038/s41598-021-86049-w 2 Vol:.(1234567890) www.nature.com/scientificreports/ DE SE DE SE 6 6 r = 0.338 r = 0.544 P = 6.91× 0−8 P < 10−10 03 03 036 036 −3 −3 statistic (human) −statistic (human) − −statistic (human) −statistic (human) −3 −3 6 t t t t −6 r = 0.158 r = 0.429 − −2 −10 P = 1.41× 0 P < 10 6 −6 − −9 −9 −10−50 510 −10−50 510 −20−10