Abstract
<title>Abstract</title> <p>Background Asthma airway inflammation is not fully captured by the conventional binary T2/non-T2 framework. We aimed to determine whether cross-cohort airway transcriptomics could define a more reproducible inflammatory structure, including both a continuous T2-related axis and heterogeneity within T2-low disease. Methods Six GEO-derived human airway epithelial or sputum transcriptomic cohorts (n = 454) were analyzed, including three discovery cohorts and three independent validation cohorts. T2 inflammation was quantified using a core three-gene signature (POSTN, CLCA1, and SERPINB2). A T2-associated epithelial axis was derived by integrating cross-cohort differential expression, pathway activity, and co-expression network analysis. Within T2-low samples, non-T2 endotypes were defined using axis-based clustering and evaluated by predictive modeling, differential expression, pathway analysis, and external validation. Results Core T2 signals showed stable directional concordance across cohorts. The core3-based T2 score robustly stratified samples across discovery and validation cohorts. A leakage-controlled elastic-net model discriminated T2-high from T2-low samples with strong performance and retained good discrimination in independent validation cohorts. The T2 epithelial axis score correlated strongly with the T2 score and remained positively associated with eosinophilic phenotype in external cohorts (pooled OR = 1.77, 95% CI 1.18–2.65; P = 0.0059). Within T2-low asthma, two reproducible non-T2 endotypes, E1 and E2, were identified. E1 was characterized by higher IFN_Antiviral, IL1_Inflammasome, and Th17/Neutrophil program scores, and a minimal E1/E2 classifier retained good discrimination (pooled AUROC = 0.899). Conclusions These findings support a layered model of asthma inflammation comprising a continuous T2 epithelial axis together with reproducible internal heterogeneity within T2-low disease.</p>