Signed Differential Co‐Expression Network Analysis Suggests Differential Regulation of SP/KLF Family of Transcription Factors in Dilated Cardiomyopathy

INTRODUCTION Co‐expression analyses utilize correlation to identify co‐regulated gene groups. Differential co‐expression (DC) networks asses the differential co‐regulation of genes across conditions. Differences in properties of networks with unsigned and signed (with a sign on the magnitude) edges have been previously documented (e.g. in social networks). However, research in biological DC networks, have rarely focused on the influence of signed edges prior to network clustering and their biological interpretation. To address this, we developed an approach to DC network analysis from signed , weighted co‐expression data and apply it to understand dilated cardiomyopathy. METHODS A signed scalar matrix of difference (δ) was first calculated using absolute Fisher r to z‐scores of correlations from each condition (1 and 2). Taking a difference in absolutes allowed us to capture the change in magnitudes of correlations between conditions 1 and 2. For e.g., positive δ indicated higher magnitudes of correlation in condition 1. Adjacency matrix was then calculated from δ and hierarchically clustered. Cut height was dynamically determined to identify differentially co‐expressed modules (M). We defined signed node connectivity to measure “hubness” of a gene as conn + (u)‐ conn − (u); where, conn +/− (u) is the sum of all +/− edge weights incident at node u from all nodes in M respectively. RESULTS We used the above approach to understand mechanisms underlying ischemic (ICM) and idiopathic dilated (IDCM) cardiomyopathy‐ two progressive, largely irreversible complex heart diseases. We obtained publicly available cardiac muscle expression data (from Gene Expression Omnibus), for ICM and IDCM. Using our approach on a list of 4363 genes across 166 ICM and 149 IDCM samples, we identified 11/28 modules as significantly differentially co‐expressed (p<0.05, via permutation testing). Functional analysis corresponded well with current understanding of aberrant processes underlying dilated cardiomyopathy (Tbl. 1). Interestingly, all 11 modules were significantly associated with protein‐protein interactions (p<0.05, extracted from STRING database) (Tbl. 1). 152 hub genes were identified in the 11 significant modules with 100/152 in top 5 modules (identified via module ranking metrics). Topological assessment of 100 hubs exhibited strong associations with known markers of ICM and IDCM, such as BMP6, ACVRL1, VCAM1 and IRS2. Hubs were also enriched for variants in cis‐eQTL within heart left ventricle (p<0.05, from the GTeX consortium's cis‐eQTL data within the heart left ventricle) (Tbl. 2). SP/KLF family of transcriptional factors and its targets (Fig. 2) were over‐represented in significantly differentially co‐regulated modules, suggesting their differential role in ICM and IDCM. CONCLUSION We provide a scalable and unsupervised approach to identifying signed DC from two condition studies. Our results not only corroborate the extant knowledge on dilated cardiomyopathy but provide novel insights for further scientific research in identifying mechanistic differences between ICM and IDCM. Support or Funding Information Supported by NIH Grants R01 HL108735, R01 HL106579, U01 CA200147, U01 CA198941, and NSF Grant CCF‐0939370. This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal .