Identifying ceRNA Networks Associated With the Susceptibility and Persistence of Atrial Fibrillation Through Weighted Gene Co-Expression Network Analysis

Background: Atrial fibrillation (AF) is the most common arrhythmia. We aimed to construct competing endogenous RNA (ceRNA) networks associated with the susceptibility and persistence of AF by applying the weighted gene co-expression network analysis (WGCNA) and prioritize key genes using the random walk with restart on multiplex networks (RWR-M) algorithm. Methods: RNA sequencing results from 235 left atrial appendage samples were downloaded from the GEO database. The top 5,000 lncRNAs/mRNAs with the highest variance were used to construct a gene co-expression network using the WGCNA method. AF susceptibility- or persistence-associated modules were identified by correlating the module eigengene with the atrial rhythm phenotype. Using a module-specific manner, ceRNA pairs of lncRNA–mRNA were predicted. The RWR-M algorithm was applied to calculate the proximity between lncRNAs and known AF protein-coding genes. Random forest classifiers, based on the expression value of key lncRNA-associated ceRNA pairs, were constructed and validated against an independent data set. Results: From the 21 identified modules, magenta and tan modules were associated with AF susceptibility, whereas turquoise and yellow modules were associated with AF persistence. ceRNA networks in magenta and tan modules were primarily involved in the inflammatory process, whereas ceRNA networks in turquoise and yellow modules were primarily associated with electrical remodeling. A total of 106 previously identified AF-associated protein-coding genes were found in the ceRNA networks, including 16 that were previously implicated in the genome-wide association study. Myocardial infarction–associated transcript (MIAT) and LINC00964 were prioritized as key lncRNAs through RWR-M. The classifiers based on their associated ceRNA pairs were able to distinguish AF from sinus rhythm with respective AUC values of 0.810 and 0.940 in the training set and 0.870 and 0.922 in the independent test set. The AF-related single-nucleotide polymorphism rs35006907 was found in the intronic region of LINC00964 and negatively regulated the LINC00964 expression. Conclusion: Our study constructed AF susceptibility- and persistence-associated ceRNA networks, linked genetics with epigenetics, identified MIAT and LINC00964 as key lncRNAs, and constructed random forest classifiers based on their associated ceRNA pairs. These results will help us to better understand the mechanisms underlying AF from the ceRNA perspective and provide candidate therapeutic and diagnostic tools.