Yue-Min Zou, Man-Ning Wu, Dong-Mei Zhou, Wen-Bo Jiang, Yan-Ping Bai
Background/Objectives: Chronic pruritus of unknown origin (CPUO) is mechanistically poorly resolved, and its relationship to atopic dermatitis (AD) is unclear. We asked how much apparent separation survives removal of leakage. Methods: Public bulk skin data (GSE237920; four healthy, four AD, four CPUO) were analyzed by differential expression and an XOmiVAE-inspired latent model, with public single-cell and atlas resources for context. Every label-dependent step was then re-executed inside each cross-validation fold, with label permutation and 1000 bootstraps. A frozen composite niche score was applied unchanged to prurigo nodularis (GSE210854) and AD (GSE174582) cohorts. Results: Fully nested, the latent classifier separated CPUO from healthy skin (AUC 1.000; permutation p = 0.020) but not from AD (0.741; p = 0.140); randomized labels also reached 1.00. The composite score reached 0.919 against healthy skin but did not exceed its null (p = 0.086). Six genes recurred in ≥80% of bootstraps, five within the signature. Cell-type-restricted blood analysis recovered CPUO-associated monocyte CCL3 upregulation. In both validation cohorts, the frozen score was reduced, not elevated, in lesional skin. Conclusions: CPUO skin differs from healthy skin and occupies the low-inflammatory, epidermal-stress-dominated pole opposite lesional inflammatory skin, but cannot be robustly separated from AD at twelve samples.