Lung tissue multi-omics integration in copd

Background: COPD is a heterogeneous disease. We hypothesized that profiling and integrating three different omic levels (mRNA, miRNA and methylome) in the same lung tissue samples using biased (FEV1 driven) or unbiased approaches can identify novel endotypes. Methods: We quantified mRNAs (Affymetrix array), microRNAs (Illumina small RNA-Seq) and methylation profile (EPIC array) in lung tissue of former-smokers with COPD (n=60) and varying degrees of airflow limitation (GOLD grades 1-4), as well as from never-smoker controls with normal spirometry (n=17). Results were integrated with MixOmics-DIABLO using biased (i.e., FEV1 driven) and unbiased (Similarity Network Fusion -SNF) approaches. Results: We found that: (1) the DIABLO multi-omic integration using biased FEV1-driven groups constructed a predictive model able to discriminate controls from GOLD 3-4, but not GOLD 1-2 patients; (2) an unbiased SNF-driven approach identified three groups, and the DIABLO predictive multi-omics model showed consistent discrimination for SNF#3 (which included 57% of controls) and SNF#2 (67% GOLD 3-4), but not for SNF#1. Interestingly, although the mean FEV1 of patients included in SNF#1 and SNF#2 were different, GOLD 1-2 and 3-4 patients were included in both groups; and, finally, (3) the biological processes (endotypes) underlying the groups of COPD patients identified by biased and unbiased approaches varied in gene ontologies (epithelial apoptotic process vs. abnormal axoneme assembly) and lung cell lineage composition estimated by gene set variation analysis. Conclusions: Our observations highlight the very significant biological heterogeneity underlying what we call COPD today and the need for a new definition and taxonomy of the disease.