Mild Cystic Fibrosis Lung Disease Is Associated with Bacterial Community Stability

Microbial communities in the airways of persons with CF (pwCF) are variable, may include genera that are not typically associated with CF, and their composition can be difficult to correlate with long-term disease outcomes. Leveraging two large data sets characterizing sputum communities of 167 pwCF and associated metadata, we identified five bacterial community types. These communities explain 24% of the variability in lung function in this cohort, far more than single factors like Simpson diversity, which explains only 4%. Subjects with Pseudomonas-dominated communities tended to be older and have reduced percent predicted FEV₁ (ppFEV₁) compared to subjects with Streptococcus-dominated communities, consistent with previous findings. To assess the predictive power of these five communities in a longitudinal setting, we used random forests to classify 346 additional samples from 24 subjects observed 8 years on average in a range of clinical states. Subjects with mild disease were more likely to be observed at baseline, that is, not in the context of a pulmonary exacerbation, and community structure in these subjects was more self-similar over time, as measured by Bray-Curtis distance. Interestingly, we found that subjects with mild disease were more likely to remain in a mixed Pseudomonas community, providing some support for the climax-attack model of the CF airway. In contrast, patients with worse outcomes were more likely to show shifts among community types. Our results suggest that bacterial community instability may be a risk factor for lung function decline and indicates the need to understand factors that drive shifts in community composition. IMPORTANCE While much research supports a polymicrobial view of the CF airway, one in which the community is seen as the pathogenic unit, only controlled experiments using model bacterial communities can unravel the mechanistic role played by different communities. This report uses a large data set to identify a small number of communities as a starting point in the development of tractable model systems. We describe a set of five communities that explain 24% of the variability in lung function in our data set, far more than single factors like Simpson diversity, which explained only 4%. In addition, we report that patients with severe disease experienced more shifts among community types, suggesting that bacterial community instability may be a risk factor for lung function decline. Together, these findings provide a proof of principle for selecting bacterial community model systems.