NETWORK-BASED MODEL FOR STUDYING CHRONIC OBSTRUCTIVE PULMONARY DISEASE

Recently, we developed SWIM, a wizard-like software that integrates gene expression data with network topological properties for identifying a small pool of genes (i.e., switch genes) critically associated with drastic changes in cell phenotype. SWIM was amenable to detect switch genes in different organisms and cell conditions, with the potential to uncover important players in biologically relevant scenarios, including but not limited to human diseases. In this work, we present an application of SWIM to the microarray gene expression profiling of a large sample of resected lung tissues from subjects with severe chronic obstructive pulmonary disease (COPD). COPD is a progressive and obstructive lung disease characterized by an airflow obstruction that leads to a chronic inflammation and for which no cure is known. We aimed to find switch genes in the comparison between 111 severe COPD cases and 40 control smokers, freely available from GEO dataset GSE76925. We found 397 differentially expressed genes (DEGs) at a 10% FDR; and four of them - DLG2, ELMO1, NNT, SPAG16 - were at significant GWAS loci. From DEGs, SWIM built the COPD correlation network, in which two nodes are connected if the absolute value of the Pearson correlation coefficient for their expression profiles is greater than 0.5. This network encompasses 355 DEGs. Partitioning the COPD correlation network in communities, we found 3 modules, ranging in size from 408 genes in module 1, 387 genes in module 2, and 126 genes in module 3. In particular, module 2 was found enriched for B cell pathways, and included SERPINE2, CD79A, BCL2, POU2AF1, BCL11A that were previously considered as putative interactors of genes at COPD GWAS loci [3]. Then, SWIM identified 33 switch genes in COPD correlation network; all switch genes except one (E2F6) resulted up-regulated in COPD case with respect to control smokers; 29 switch genes are protein coding, including 2 transcription factors, ZNF143 and E2F6. The top differentially expressed switch gene was ZNF143 which negatively interacts in the network (i.e. highly negatively correlated) with NNT, a known COPD GWAS gene (default p-value < 10-5) from the NHGRI-EBI Catalog (www.ebi.ac.uk/gwas/), and positively interacts with BCL2 (i.e. highly positively correlated).