You are here:
Integrating gene expression data with demographic, clinical, and environmental exposure information to reveal endotypes of childhood asthma
Citation:
REIF, D., J. GALLAGHER, E. A. COHEN-HUBAL, L. M. NEAS, E. E. HUDGENS, B. HEIDENFELDER, AND S. W. EDWARDS. Integrating gene expression data with demographic, clinical, and environmental exposure information to reveal endotypes of childhood asthma . Presented at American Thoracic Society Annual Meeting, San Diego, CA, May 15 - 20, 2009.
Impact/Purpose:
RESULTS. Our final model divides study subjects into asthmatic subsets based upon gene expression patterns evident in blood. Thanks to the internal annotation of gene function(s) provided by covariate-genomic associations, we infer that each asthmatic subset represents endotypes having distinct etiological mechanisms, separating eosinophilia, nitric oxide dysregulation, obesity, and inappropriate innate immune responses. CONCLUSIONS. It is hoped that a contextual analysis strategy such as this contributes to a more comprehensive understanding of the mechanistic underpinnings of varied asthma subtypes may lead to more personalized diagnosis, management, and treatment of the disease.
Description:
RATIONALE. Childhood asthma is a multifactorial disease whose pathogenesis involves complex interplay between genetic susceptibility and modulating external factors. Therefore, effectively characterizing these multiple etiological pathways, or “endotypes”, requires an integrative approach that includes multiple data types. The Mechanistic Indicators of Childhood Asthma (MICA) study has collected multiple types of clinical, demographic, and exposure information, as well as gene expression data in an asthma case/control cohort of children (aged 9-12 years) from Detroit, MI. We hypothesize that asthmatic endotypes will be evident in blood gene expression data analyzed in the context of a diverse collection of covariates (i.e. all data except gene expression measurements). METHODS. Oligonucleotide microarrays were used to measure gene expression of 54,675 probesets from MICA blood samples. First, we performed an unbiased (without knowledge of asthma status) assessment of the association between each type of covariate data and observed gene expression values for all 193 subjects. For subsequent analyses, we select only gene expression probesets that are significantly correlated with at least one covariate. This filtering method prevents us from biasing our selecting to only genes whose expression is associated with broadly-defined, imperfect asthma diagnoses. Next, we use unsupervised clustering of the covariate-genomic correlation matrix to reveal groups of functionally-related genes. Then, the clustering results are used to build meta-genes that summarize expression across multiple genes within each functional group. Finally, recursive partitioning trees are constructed from the meta-genes to segregate individual subjects according to asthma status.