Background: Atopic dermatitis (AD) is a highly heterogenous inflammatory skin disease with a diverse clinical manifestations and treatment responses. Therefore, patient stratification based on the underlying molecular mechanisms/endotypes is critical for the development of targeted therapeutic approaches.
Aim: We sought to identify robust and reproducible AD endotypes using lesional skin transcriptomic data.
Methods: We carried out an integrated unsupervised clustering analysis using both K-means and NMF (non-negative matrix factorization) algorithms with lesional skin transcriptomic data from a collection of 8 independent cohorts respectively (total AD patients=427). The subtypes from each of the 8 cohorts were systematically compared with each other by looking at the correlation of cluster-specifically expressed genes. The endotypes that are consistently identified from all cohorts were further characterized by clinical information and disease-relevant cell and pathway signatures.
Results: Two main disease subtypes were consistently identified across all the 8 AD cohorts. The first subtype was characterized as inflammatory with relative higher disease severity (e.g. EASI/SCORAD/POEM), and higher activity of immune cells and pathways (such as T, B, Th1 and Th2). The second subtype was characterized as metabolic with relative lower disease severity, but higher activity for melanocyte and fatty acid metabolic pathways. The integrated results of K-means and NMF clustering indicate distinct molecular mechanisms underlying these two subtypes.
Conclusion: Two distinct and reproducible AD endotypes were identified and further characterization of these endotypes would help generate hypotheses of patient stratification strategy for targeted therapies.
