BACKGROUND: Asthma is a heterogeneous disease characterized by distinct phenotypes with associated microbial dysbiosis. OBJECTIVES: To identify severe asthma phenotypes based on sputum microbiome profiles and assess their stability after 12-18 months. Furthermore, to evaluate clusters' robustness after inclusion of an independent mild-to-moderate asthmatics. METHODS: In this longitudinal multicenter cohort study, sputum samples were collected for microbiome profiling from a subset of the U-BIOPRED adult patient cohort at baseline and after 12-18 months of follow-up. Unsupervised hierarchical clustering was performed using the Bray-Curtis β-diversity measure of microbial profiles. For internal validation, partitioning around medoids, consensus cluster distribution, bootstrapping and topological data analysis were applied. Follow-up samples were studied to evaluate within-patient clustering stability in severe asthmatics. Cluster robustness was evaluated by an independent mild-moderate asthma cohort. RESULTS: Data were available for 100 severe asthma subjects (median age: 55 yrs, 42% males). Two microbiome-driven clusters were identified, characterized by differences in asthma onset, smoking status, residential locations, percentage of blood and/or sputum neutrophils and macrophages, lung spirometry, and concurrent asthma medications (all p-values <.05). Cluster 2 patients displayed a commensal-deficient bacterial profile which was associated with worse asthma outcomes compared to cluster 1. Longitudinal clusters revealed high relative stability after 12-18 months in the severe asthmatics. Further inclusion of 24 independent mild-to-moderate asthmatics was consistent with the clustering assignments. CONCLUSION: Unbiased microbiome-driven clustering revealed two distinct robust severe asthma phenotypes, which exhibited relative overtime stability. This suggests that the sputum microbiome may serve as a biomarker for better characterizing asthma phenotypes.