Serum-Proteomic Profiling Reveals Distinct Atopic Dermatitis Severity-Linked Signatures

Atopic Dermatitis (AD) is a chronic inflammatory skin disease characterized by complex pathogenesis, variable clinical phenotypes, and broad severity spectrum. We utilized a serum-proteomic approach integrated with machine learning (ML) to identify novel biomarkers that distinguish mild from severe AD.

Serum from sixty-seven AD adults, stratified by Eczema Area and Severity Index/EASI and Rajka-Langeland/RJL into mild (≤ 7 and ≤ 4, respectively; n = 33) or severe (≥ 20 and ≥ 8, respectively; n = 34), was analyzed with Olink Explore 3072. Differentially expressed proteins (DEPs) were identified using t-tests and False Discovery Rate correction (FDR ≤ 0.05). Tissue enrichment analysis was conducted using HPAStainR. Pearson correlations were performed between DEPs and clinical variables, serum lactate dehydrogenase/LDH and biomarkers for Th-pathways. ML (TMLE/SuperLearner, Boruta, MUVR) was used to identify top severity biomarkers.

469 DEPs distinguished severe vs. mild AD. DEPs were significantly enriched for epithelial proteins (e.g., skin, tonsil, and esophagus epithelium). Subsets correlated strongly with LDH (68 DEPs) and/or Th2/Th22 markers (90 DEPs; r ≥ 0.6, FDR ≤ 0.05). Nine serum proteins (CCL17, CCL22, DEFB4A/B, EZR, GPR15L, IL22, PRSS53, SERPINB8, SETMAR) overlapped across three ML analyses as severity discriminating biomarkers (cross-validated AUC = 0.989; 95% CI: 0.974-1.00).

Severe AD exhibits a proteomic footprint enriched in epithelial-associated proteins that correlate with LDH-associated tissue injury and/or Th2/Th22 pathways. In this cross-sectional cohort, ML identified biomarkers that discriminated mild/severe AD groups with high cross-validated accuracy. Longitudinal and external validation studies, including healthy controls, are needed to determine specificity, generalizability, and prognostic utility.