Rationale: Chronic obstructive pulmonary disease (COPD) is a heterogeneous condition. Objectives: We hypothesized that the unbiased integration of different COPD lung omics using a novel multilayer approach might unravel mechanisms associated with clinical characteristics. Methods: We profiled mRNA, microRNA and methylome in lung tissue samples from 135 former smokers with COPD. For each omic (layer), we built a patient network on the basis of molecular similarity. The three networks were used to build a multilayer network, and optimization of multiplex modularity was used to identify patient communities across the three distinct layers. Uncovered communities were related to clinical features. Measurements and Main Results: We identified five patient communities in the multilayer network that were molecularly distinct and related to clinical characteristics, such as FEV1 and blood eosinophils. Two communities (C#3 and C#4) had both similarly low FEV1 values and emphysema but were molecularly different: C#3, but not C#4, presented B- and T-cell signatures and a downregulation of secretory (SCGB1A1/SCGB3A1) and ciliated cells. A machine learning model was set up to discriminate C#3 and C#4 in our cohort and to validate them in an independent cohort. Finally, using spatial transcriptomics, we characterized the small airway differences between C#3 and C#4, identifying an upregulation of T-/B-cell homing chemokines and bacterial response genes in C#3. Conclusions: A novel multilayer network analysis is able to identify clinically relevant COPD patient communities. Patients with similarly low FEV1 and emphysema can have molecularly distinct small airways and immune response patterns, indicating that different endotypes can lead to similar clinical presentation.
Read full abstract