Integrative clustering analysis to discover novel IPF disease subtypes in the IPF-PRO Registry

Abstract

Background: Limited data are available on phenotypes of idiopathic pulmonary fibrosis (IPF). Aim: To identify molecular-based IPF subtypes using integrative clustering models based on proteomics and transcriptomics data. Methods: Proteomics, miRNA and total RNAseq data from blood samples taken from 243 patients in the multicenter observational IPF-PRO Registry were analyzed. iClusterPlus, iClusterBayes and Similarity Network Fusion (SNF) models were applied for integrative analyses. Composite physiologic index (CPI); GAP stage; diagnostic criteria; DLco, FEV1 and FVC % predicted at baseline; and time to death, lung transplant or FVC decline >10% predicted were compared between the clusters. Pathway analyses were performed using Enrichr. Results: There was large overlap in cluster memberships across the models. There were significant differences in mean CPI, DLco % predicted and FVC % predicted at baseline, and in time to death, lung transplant or FVC decline >10% predicted between clusters identified using the SNF model (Figure). Pathways analysis of the most discriminative genes and proteins revealed enrichment for terms related to the immune system. Conclusion: Integrative multi-omics clustering analysis identified distinct molecular-based IPF phenotypes that differed in measures of disease severity and disease progression.