Extended Data Fig. 1: The overall workflow integrating human multi-omics, in vivo murine model experiments and in vitro cellular assays to elucidate airway microbe-host interactions in COPD.

a) Induced sputum was collected from COPD patients and healthy controls from Guangzhou (discovery cohort) and Shenzhen (validation cohort), China. Demographic and clinical metadata including age, gender, smoking history, medication history, spirometry, sputum and blood differential cell counts were obtained. b) Induced sputum was subject to simultaneous characterization of metagenome (n = 135), metabolome (n = 129), host transcriptome (n = 130) and proteome (n = 59). The multi-omic profiles were processed by a combination of knowledge-driven and data-driven dimensionality reduction to generate KEGG and co-abundance modules, which were then filtered by COPD-association testing. To identify microbial-metabolite-host interaction links, a sequential mediation analysis was performed along the metagenome-metabolome-transcriptome-proteome axis for modules associated with neutrophilic or eosinophilic inflammation, respectively. Cross-omic biological links were then identified from pairs of modules with significant mediation effects, utilizing microbial genetic information, metabolite-target pairs and canonical pathways from existing databases. c) The selected microbiome-metabolite-host interaction hypothesis was tested using a murine model of emphysema and in vitro cell assays to elucidate molecular and cellular mechanisms. Lung function, airway inflammation, tissue destruction, and apoptosis were measured as key endpoints in the in vivo and in vitro experiments.