Constitutive patterns of gene expression regulated by RNA-binding proteins

Cirillo et al. Genome Biology Cirillo et al. - Constitutive patterns of gene expression regulated by RNA-binding proteins Open Access Constitutive patterns of gene expression regulated by RNA-binding proteins Davide Cirillo1,2, Domenica Marchese1,2, Federico Agostini1,2, Carmen Maria Livi1,2, Teresa Botta-Orfila1,2 and Gian Gaetano Tartaglia1,2* Background: RNA-binding proteins regulate a number of cellular processes, including synthesis, folding, translocation, assembly and clearance of RNAs. Recent studies have reported that an unexpectedly large number of proteins are able to interact with RNA, but the partners of many RNA-binding proteins are still uncharacterized. Results: We combined prediction of ribonucleoprotein interactions, based on catRAPID calculations, with analysis of protein and RNA expression profiles from human tissues. We found strong interaction propensities for both positively and negatively correlated expression patterns. Our integration of in silico and ex vivo data unraveled two major types of proteinRNA interactions, with positively correlated patterns related to cell cycle control and negatively correlated patterns related to survival, growth and differentiation. To facilitate the investigation of proteinRNA interactions and expression networks, we developed the catRAPID express web server. Conclusions: Our analysis sheds light on the role of RNA-binding proteins in regulating proliferation and differentiation processes, and we provide a data exploration tool to aid future experimental studies. Background With the advent of high-throughput proteomic and transcriptomic methods, genome-wide data are giving previously unprecedented views of entire collections of gene products and their regulation. Recently, approaches based on nucleotide-enhanced UV cross-linking and oligo(dT) purification have shown that a number of proteins are able to bind to RNA [1,2]. RNA-binding proteins (RBPs) are key regulators of post-transcriptional events [3] and influence gene expression by acting at various steps in RNA metabolism, including stabilization, processing, storing, transport and translation. RBP-mediated events have been described using recognition and regulatory elements in RNA sequences [4,5] as well as expression profiles [6] that are tissue specific and conserved across species [7-9]. Although heterogeneity in gene regulation is responsible for phenotypic variation and evolution [10], very little is known about constitutive expression patterns controlled by RBPs [11,12], which are the subject of this work. Data from recent transcriptomic and proteomic studies [13,14] are becoming attractive for studying mechanisms of gene regulation [15,16]. Despite the increasing amount of genomic data, the development of computational methods for integrating, interpreting and understanding molecular networks remains challenging [17,18]. Here we combine our predictions of proteinRNA interactions, based on catRAPID calculations [19,20], with the information obtained from expression data to investigate constitutive regulatory mechanisms. The catRAPID approach has been previously employed to predict protein associations with non-coding RNAs [21,22] as well as ribonucleoprotein interactions linked to neurodegenerative diseases [23,24]. Our theoretical framework has been used to unravel self-regulatory pathways controlling gene expression [25]. The catRAPID omics algorithm, validated using photoactivatable-ribonucleoside-enhanced cross-linking and immunoprecipitation (PAR-CLIP) data, has been recently developed to predict proteinRNA associations at the transcriptomic and proteomic levels [26]. Using comprehensive and manually annotated databases of expression profiles in human tissues, at both protein and RNA levels, we investigated the correlation between RBP activity and regulation. The link between interaction propensity and expression levels was exploited to reveal the fine-tuned functional sub-networks responsible for regulatory control. To explore the results further, we developed the catRAPID express web server [27]. Results In this study, we focused on the mRNA interactomes of RBPs detected through nucleotide-enhanced UV cross-linking and oligo(dT) purification approaches [1,2]. Exploiting gene ontology (GO) annotations [28] for protein-coding genes, we systematically analyzed protein RNA interactions and expression data for human tissues. At present, few studies have investigated how altering protein expression affects the abundance of RNA targets. Interrogating the Gene Expression Omnibus (GEO) [29] and ArrayExpress databases [30], we found two human proteins, ELAV-like protein 1 (or human antigen R, HuR) [31] and Protein lin-28 homolog B (LIN28B) [32,33], whose knock-down has been shown to alter the expression of target genes identified by PAR-CLIP (see Materials and methods). Our predictions, made using the catRAPID algorithm [26], identified experimentally validated interactions with high significance (HuR: P = 10-8; LIN28B: P = 10-3; Fishers exact test; see Materials and methods). The interactions were effectively discriminated from non-interacting pairs using score distributions (LIN28B: P = 10-4; HuR: P = 10-16; Students t-test; see Materials and methods). Hence, catRAPID is very good at predicting physical interactions between a protein and RNA partners (other statistical tests are given in Materials and methods and Additional file 1). To understand the regulation of HuR and LIN28B targets better, we studied the relation between interaction propensities and expression levels. We found that the expression of predicted HuR targets is altered (log-fold change, LFC) when HuR is knocked down (P < 10-5; KolmogorovSmirnov test; Figure 1A), which is in agreement with experimental data [31]. Similarly, predicted LIN28B targets are downregulated upon protein depletion (P < 10-2; KolmogorovSmirnov test; Figure 1B), as shown in a previous study [33]. Moreover, we compared the top 1% of predicted associations with the top 1% of experimental interactions and found the same enrichments for transcripts changing in expression levels upon protein depletion. Specifically, 62% of HuR experimental interactions and 63% of HuR predicted associations had LFC > 0. Similarly for LIN28B, 57% of experimental interactions and 56% of predicted associations had LFC > 0. These HuR and LIN28B examples indicate that changes in protein expression influence the abundance of RNA targets, suggesting that a large-scale analysis of co-expression and interaction propensities could improve understanding of RBP-mediated regulatory mechanisms. RNA-binding proteinmRNA interactions and relative expression profiles Our predictions indicate that interacting molecules have both more correlated and anti-correlated expression patterns (see Materials and methods and Figure 2). By contrast, non-correlated expression is not associated with any enrichment in interaction propensity (Additi (...truncated)