Spatially distinct molecular patterns of gene expression in idiopathic pulmonary fibrosis

Abstract

Background: Idiopathic pulmonary fibrosis (IPF) is a heterogeneous disease that is pathologically characterized by areas of normal-appearing lung parenchyma, active fibrosis (transition zones including fibroblastic foci) and dense fibrosis. Defining transcriptional differences between these pathologically heterogeneous regions of the IPF lung is critical to understanding the distribution and extent of fibrotic lung disease and identifying potential therapeutic targets. Application of a spatial transcriptomics platform would provide more detailed spatial resolution of transcriptional signals compared to previous single cell or bulk RNA-Seq studies.

Methods: We performed spatial transcriptomics using GeoMx Nanostring Digital Spatial Profiling on formalin-fixed paraffin-embedded (FFPE) tissue from 32 IPF and 12 control subjects and identified 231 regions of interest (ROIs). We compared normal-appearing lung parenchyma and airways between IPF and controls with histologically normal lung tissue, as well as histologically distinct regions within IPF (normal-appearing lung parenchyma, transition zones containing fibroblastic foci, areas of dense fibrosis, and honeycomb epithelium metaplasia).

Results: We identified 254 differentially expressed genes (DEGs) between IPF and controls in histologically normal-appearing regions of lung parenchyma; pathway analysis identified disease processes such as EIF2 signaling (important for cap-dependent mRNA translation), epithelial adherens junction signaling, HIF1α signaling, and integrin signaling. Within IPF, we identified 173 DEGs between transition and normal-appearing lung parenchyma and 198 DEGs between dense fibrosis and normal lung parenchyma; pathways dysregulated in both transition and dense fibrotic areas include EIF2 signaling pathway activation (upstream of endoplasmic reticulum (ER) stress proteins ATF4 and CHOP) and wound healing signaling pathway deactivation. Through cell deconvolution of transcriptome data and immunofluorescence staining, we confirmed loss of alveolar parenchymal signals (AGER, SFTPB, SFTPC), gain of secretory cell markers (SCGB3A2, MUC5B) as well as dysregulation of the upstream regulator ATF4, in histologically normal-appearing tissue in IPF.

Conclusions: Our findings demonstrate that histologically normal-appearing regions from the IPF lung are transcriptionally distinct when compared to similar lung tissue from controls with histologically normal lung tissue, and that transition zones and areas of dense fibrosis within the IPF lung demonstrate activation of ER stress and deactivation of wound healing pathways.

Fig. 1

Overview of the GeoMx platform (A). PCA plot by disease and region (B). Example ROI selection of tissue from an IPF subject, H&E (C) with corresponding immunofluorescence (D) showing regions of normal appearing lung parenchyma (a,b), normal bronchiolar epithelium (c), transition (d), dense fibrosis (e,f) and honeycomb epithelium (g). H&E showing regions of normal appearing lung parenchyma at higher magnification in a control with histologically normal lung tissue (E, F) and IPF (G)

Fig. 2

Mean cell proportions estimated with SpatialDecon based on IPF cell reference (A). Histology image showing changes in alveolar cell composition (B). Immunofluorescence changes in SFTPC and AGER between IPF and controls with histologically normal lung tissue in normal parenchyma regions (C). FOVs were chosen based on presence of features of interest (e.g. parenchyma, bronchiolar, etc.) in each section and 9 20x images centered randomly within the feature of interest were assembled into a composite. Pairwise comparisons are Mann-Whitney U test with significance set to (*) p < 0.05, (**) p < 0.01, (***) p < 0.001, (****) p < 0.0001

Fig. 3

Differential expression between IPF and controls with histologically normal lung tissue in normal appearing lung parenchyma (A). Top 15 Ingenuity canonical pathways in the IPF vs. control normal lung parenchyma comparison (B). The activation z-score is a statistical measure based on the directional relationships between genes and their biological function. Orange indicates increased predictions over that of the null (positive z score), blue indicates decreased predictions (negative z score) and white z score of zero. Differential expression between IPF and controls with histologically normal lung tissue in normal bronchiolar regions (C) adjusting for batch and estimated cell proportions of AEC1, AEC2, fibroblasts and ciliated cells

Fig. 4

Differential expression between transition and normal appearing lung parenchyma regions (A) and dense fibrosis and normal appearing lung parenchyma (B) in IPF subjects adjusting for batch and estimated cell proportions of AEC1, AEC2, fibroblasts and ciliated cells. Venn diagram showing the overlap in DEGs between comparisons with normal lung parenchyma (C). Ingenuity canonical pathway (D) and upstream regulator analysis (E) of DEGs.  The activation z-score is a statistical measure based on the directional relationships between genes and their biological function. Orange indicates increased predictions over that of the null (positive z score), blue indicates decreased predictions (negative z score) and white z score of zero. Dots indicate that z score did not reach significance (1.645). Histology image showing changes in MUC5B and ATF4 (F). Immunofluorescence changes in MUC5B, ATF4 and ATF4 in KRT8 + cells between IPF and control in normal parenchyma regions (G). FOVs were chosen based on presence of features of interest (e.g. parenchyma, bronchiolar, etc.) in each section and 9 20x images centered randomly within the feature of interest were assembled into a composite. Pairwise comparisons are Mann-Whitney U test with significance set to (*) p < 0.05, (**) p < 0.01, (***) p < 0.001, (****) p < 0.0001