View Abstract

A7434 - Tensor Decomposition of RNA-Sequencing Data Across Three Tissues Identifies Potential COPD Genes and Networks
Author Block: J. Morrow1, R. Chase1, M. J. Divo2, C. A. Owen3, P. Castaldi1, D. L. DeMeo3, E. K. Silverman3, C. P. Hersh1; 1Channing Division of Network Medicine, Boston, MA, United States, 2Pulm and Crit Care Div, Brigham and Women's Hospital, Boston, MA, United States, 3Brigham and Women's Hospital, Boston, MA, United States.
Rationale: Multiple gene expression studies have been performed separately in peripheral blood, lung, and airway tissues from different sets of COPD patients. We performed RNA-sequencing gene expression profiling of large-airway, alveolar macrophage and peripheral blood samples from the same set of COPD cases and controls from the COPDGene study who underwent bronchoscopy at a single center. Using integrative statistical and network medicine approaches, we sought to improve the understanding of COPD by studying interacting gene sets and pathways across these tissues, instead of individual genomic determinants.
Methods: We performed differential expression analysis using RNA-seq data obtained from 94 samples across 39 COPD cases and controls (includes six non-smokers) via the R package DESeq. We tested associations between gene expression and variables related to lung function, smoking history, and CT scan measures of emphysema and airway disease. We performed tensor decomposition using the Bayesian method SDA (Sparse Decomposition of Arrays) to observe preservation of gene networks across the tissues. This decomposition of the 3-dimensional data (subjects, genes, tissues) produces a set of latent components that we tested for association with several COPD phenotypes. We hypothesized that this integration would reveal shared and private gene expression signatures across the tissues.
Results: The known smoking-related genes CYP1B1, AHRR, and GPR15 were among the top differential expression results for smoking in the large-airway data; GPR15 was also found in the blood analysis. We identified pathways related to mineral absorption and response to metal ions in the differential expression profiling for emphysema (15th percentile of the lung density histogram) in the alveolar macrophage data. These pathways include several metallothionein genes. MAPK signaling was identified in the FEV1/FVC results for peripheral blood. Neutrophil degranulation was highlighted in the pathway analysis of the higher scoring genes for a latent component associated with COPD and airway disease from the tensor decomposition analysis.
Conclusions: Our statistical and network integration across relevant tissues reveals shared and tissue-specific disease biology. These replicated and novel findings in the airway and peripheral blood have highlighted candidate genes and gene networks for COPD pathogenesis.