BAL cell gene expression is predictive of Mortality in Idiopathic Pulmonary Fibrosis and enriched for Genes of Airway Basal Cells (II)

Background: Idiopathic pulmonary fibrosis (IPF) is a fatal disease with variable outcome. Currently, there is little information whether changes in molecular pathways in the alveolar compartment of patients with IPF are indicative of disease progression. To address this question we analyzed gene expression signatures in cells obtained from bronchoalveolar lavage (BAL) of patients with IPF. Methods: BAL cells were harvested from 212 IPF patients and 20 healthy donors at the time of diagnosis. RNA was extracted, labeled and hybridized to Agilent gene expression arrays. Our study included a discovery cohort of 62 patients from Freiburg, Germany, and two independent validation cohorts, Siena, Italy (50 patients) and Leuven, Belgium (64 patients). Survival analysis was performed by applying Cox models and component-wise boosting. Results: We found 1582 genes predictive of mortality in the discovery cohort in univariate analyses adjusted for age and gender at FDR<0.05. A nine gene signature, developed by component-wise boosting in the Freiburg array dataset, performed well in both validation cohorts, Siena (p<0.0032) and Leuven (p=0.0033). The genes associated with mortality in BAL cells were significantly enriched for genes expressed in airway basal epithelial cells, airway progenitor cells recently implicated in fibrosis. Conclusions: Our results identify and validate a BAL cell gene expression signature that predicts mortality in IPF. This signature unmasks a potential novel role for airway basal epithelial cells in IPF. BAL cells were harvested from 212 IPF patients and 20 healthy donors at the time of diagnosis. RNA was extracted, labeled and hybridized to Agilent gene expression arrays. Our study included a discovery cohort of 62 patients from Freiburg, Germany, and two independent validation cohorts, Siena, Italy (50 patients) and Leuven, Belgium (64 patients). Survival analysis was performed by applying Cox models and component-wise boosting.