Exploring the cross-phenotype network region of disease modules reveals concordant and discordant pathways between chronic obstructive pulmonary disease and idiopathic pulmonary fibrosis

Abstract

Chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF) are two pathologically distinct chronic lung diseases that are associated with cigarette smoking. Genetic studies have identified shared loci for COPD and IPF, including several loci with opposite directions of effect. The existence of additional shared genetic loci, as well as potential shared pathobiological mechanisms between the two diseases at the molecular level, remains to be explored. Taking a network-based approach, we built disease modules for COPD and IPF using genome-wide association studies-implicated genes. The two disease modules displayed strong disease signals in an independent gene expression data set of COPD and IPF lung tissue and showed statistically significant overlap and network proximity, sharing 19 genes, including ARHGAP12 and BCHE. To uncover pathways at the intersection of COPD and IPF, we developed a metric, NetPathScore, which prioritizes the pathways of a disease by their network overlap with another disease. Applying NetPathScore to the COPD and IPF disease modules enabled the determination of concordant and discordant pathways between these diseases. Concordant pathways between COPD and IPF included extracellular matrix remodeling, Mitogen-activated protein kinase (MAPK) signaling and ALK pathways, whereas discordant pathways included advanced glycosylation end product receptor signaling and telomere maintenance and extension pathways. Overall, our findings reveal shared molecular interaction regions between COPD and IPF and shed light on the congruent and incongruent biological processes lying at the intersection of these two complex diseases.

Figure 1

Determining disease modules. (A) Left: the LCC size z-score, plotted against the top N genes with respect to their DADA prioritization rank, for the top 500 ranked genes for IPF. Right: the proportion of seed genes in the LCC, plotted against the IPF module size. The resulting IPF disease module has a significantly large LCC size compared to random expectation (z-score = 41) and consists of 109 genes. (B) Left: the LCC size z-score, plotted against the top N genes with respect to their DADA prioritization rank, for the top 500 ranked genes for COPD. Right: the proportion of seed genes in the LCC, plotted against the COPD module size. The resulting COPD disease module has a significantly large LCC size compared to random expectation (z-score = 12) and consists of 425 genes.

Figure 2

Search for disease-specific signals within disease modules using gene expression data. The logarithm of the gene expression FC for differentially expressed (FDR-adjusted P < 0.05) module genes (blue) versus all other genes in the interactome (orange). FC for IPF and COPD calculated for diseased samples compared to healthy samples. Seed genes are excluded in the analysis. Group comparisons are made using the Mann–Whitney U test. Subscripts ‘1’ and ‘2’ are used to denote the Agilent GPL14550 platform and Agilent GPL6480 platform, respectively.

Figure 3

The overlap of the COPD and IPF disease modules. (A) The COPD and IPF disease modules mapped onto the interactome. Node size is proportional to the number of edges. Red and cyan nodes represent IPF and COPD seed genes, respectively. Purple and green nodes represent IPF and COPD module genes. Orange nodes in the middle represent the overlapping genes shared between the two modules. Orange nodes with red borders represent the shared seed genes. Edges connecting the two disease modules are omitted from the visualization for clarity. (B) The overlap of the two modules (red arrow), measured by the Jaccard index, compared to those of random modules of the same size (red bars). A higher Jaccard index indicates a higher level of overlap between two modules. (C) The closeness of the two modules (red arrow), measured by the average shortest distance, compared to those of random modules of the same size (red bars). A lower average shortest distance indicates a higher degree of closeness between two modules.

Figure 4

Overview of the NetPathScore approach. (A) Enriched pathways of Disease A (e.g. COPD) are determined by ORA. (B) For each enriched pathway of Disease A, the extended pathway network is formed by including the first neighbors, i.e. direct interactors, of pathway genes. The overlap of the extended pathway with the disease module of Disease B is calculated. This overlap is compared against random expectation, and the z-score obtained from this comparison is used to calculate the NetPathScore of the Disease A pathway with respect to Disease B. (C) In the case of directional pathway analysis, an additional step is included where the directional pathways of Disease A (e.g. COPD up pathways) are prioritized using NetPathScore with respect to the up- and down-regulated portions of Disease B module (e.g. IPF up and IPF down). The resulting rankings are then subtracted from each other. The pathways with a positive rank difference correspond to the pathways highly ranked with respect to the up-regulated module, and the pathways with a negative rank difference correspond to the pathways highly ranked with respect to the down-regulated module. These pathways can be organized in a 2 × 2 table where the diagonal entries are the concordant pathways (same direction in both diseases) and off-diagonal entries are the discordant pathways (opposite directions in the two diseases).

Figure 5

Example of concordant pathway between COPD and IPF. The extracellular matrix organization pathway, one of the COPD-up pathways highly ranked by NetPathScore according to its network overlap with the IPF up-regulated module. Green and gray nodes represent the pathway genes and their first neighbors, respectively. The purple nodes represent the IPF disease module, with the darker and lighter shades of purple indicating the up- and down-regulated portions of the module.