Modeling circRNAs expression pattern with integrated sequence and epigenetic features identifies H3K79me2 as regulators for circRNAs expression

Abstract

47Circular RNAs (circRNAs) are an abundant class of noncoding RNAs with widespread, 48 cell/tissue specific pattern. Because of their involvement in the pathogenesis of multiple disease, 49 they are receiving increasing attention. Previous work suggested that epigenetic features might 50 be related to circRNA expression. However, current algorithms for circRNAs prediction neglect 51 these features, leading to constant results across different cells. 52 53Here we built a machine learning framework named CIRCScan, to predict expression status 54 and expression levels of circRNAs in various cell lines based on sequence and epigenetic 55 features. Both expression status and expression levels can be accurately predicted by different 56 groups of features. For expression status, the top features were similar in different cells. 57 However, the top features for predicting expression levels were different in different cells. 58 Noteworthy, the importance of H3K79me2 ranked high in predicting both circRNAs expression 59 status and levels across different cells, indicating its important role in regulating circRNAs 60 expression. Further validation experiment in K562 confirmed that knock down of H3K79me2 61 did result in reduction of circRNA production. 62 63Our study offers new insights into the regulation of circRNAs by incorporating epigenetic 64 features in prediction models in different cellular contexts. 65 66 which are believed to be necessary for circularization (Dubin et al. 1995; Zhang et al. 2014). 87 Specifically, Alu repeats were found to be enriched in the flanking intron regions of circRNA-88 forming exons with high conservation and were correlated with the human circRNAs formation. 89The competition between these inverted repeated Alu pairs can promote and regulate alternative 90 circularization, resulting in multiple circular RNA transcripts derived from one gene (Jeck et al. 91 2013; Liang and Wilusz 2014; Zhang et al. 2014). According to the "exon skipping" (Vicens 92 and Westhof 2014; Barrett et al. 2015; Starke et al. 2015) model of exon circularization, a 93 5 method (Ivanov et al. 2015) was developed to predict circRNAs according to sequence features 94 in the intron region. Besides, some tools also tried to distinguish circRNAs from other lncRNAs 95 based on conformational and conservation features, sequence compositions, Alu, SNP densities, 96 and thermodynamic dinucleotide properties (Pan and Xiong 2015; Liu et al. 2016). However, 97 these methods based on genomic sequence features generate indiscriminate predicted results 98 across different tissue/types, which is unable to find the tissue/cell type circRNAs. More 99 importantly, these tools only focus on indicating the circRNAs expression status, and are not 100 capable of predicting circRNAs expression values. 101 102 Several studies have used epigenetic or chromatin features to predict gene expression, for 103 example, Karlić et al. applied a linear regression model using histone modifications to predict 104 gene expression on human T-cell ...

Exploring the interplay between PARP1 and circRNA biogenesis and function

“…For instance, Su et al (2014) showed genome‐wide that Alu elements are usually bound by two well‐phased nucleosomes containing the active histone marks H3K4me2 and H3K4me1 (Barski et al, 2007) to the exclusion of repressive histone marks (J.‐B. Chen, Dong, et al, 2018; Su et al, 2014). Another study using predictive models found H3k36me3, H3k79me2, and H4k20me1 correlated with low circRNA expression (Zhang et al, 2020).…”