Principles of self-organization in biological pathways: a hypothesis on the autogenous association of alpha-synuclein

Andreas Zanzoni 0 1 2 Domenica Marchese 0 1 2 Federico Agostini 0 1 2 Benedetta Bolognesi 0 1 2 Davide Cirillo 0 1 2 Maria Botta-Orfila 0 1 2 Carmen Maria Livi 0 1 2 Silvia Rodriguez-Mulero 0 1 2 Gian Gaetano Tartaglia 0 1 2 0 Universitat Pompeu Fabra (UPF) , 08003 Barcelona, Spain 1 Present address: Andreas Zanzoni, Inserm TAGC U1090, Aix-Marseille Universite , Parc Scientifique de Luminy, 13288 Marseille, France 2 Gene Function and Evolution, Bioinformatics and Genomics, Centre for Genomic Regulation (CRG) , 08003 Barcelona, Spain Previous evidence indicates that a number of proteins are able to interact with cognate mRNAs. These autogenous associations represent important regulatory mechanisms that control gene expression at the translational level. Using the catRAPID approach to predict the propensity of proteins to bind to RNA, we investigated the occurrence of autogenous associations in the human proteome. Our algorithm correctly identified binding sites in wellknown cases such as thymidylate synthase, tumor suppressor P53, synaptotagmin-1, serine/arigininerich splicing factor 2, heat shock 70 kDa, ribonucleic particle-specific U1A and ribosomal protein S13. In addition, we found that several other proteins are able to bind to their own mRNAs. A large-scale analysis of biological pathways revealed that aggregation-prone and structurally disordered proteins have the highest propensity to interact with cognate RNAs. These findings are substantiated by experimental evidence on amyloidogenic proteins such as TAR DNA-binding protein 43 and fragile X mental retardation protein. Among the amyloidogenic proteins, we predicted that Parkinson's disease-related a-synuclein is highly prone to interact with cognate transcripts, which suggests the existence of RNA-dependent factors in its function and dysfunction. Indeed, as aggregation is intrinsically concentration dependent, it is possible that autogenous interactions play a crucial role in controlling protein homeostasis. - Although proteins are involved in almost every cellular process, increasing evidence indicates that coding and non-coding RNAs play fundamental roles in gene regulation (1,2) and disease (3,4). Recent studies showed that establishment of aberrant associations or disruption of functional proteinRNA interactions occurs in neurological disorders (5,6). For instance, interaction with RNA favors conversion of alpha-helix rich prion protein PrPC into the pathogenic beta-structure-rich insoluble conformer PrPSc that propagates in CreutzfeldtJakob disease (7). In Alzheimers disease, the association between Amyloid Precursor Protein mRNA and iron regulatory protein 1 is disrupted, resulting in compromised translation efEciency and elevated cytotoxicity (8). ProteinRNA associations regulate several processes such as synthesis, folding, translocation, assembly and clearance of molecules. Previous studies suggested that ribonucleoprotein interactions might be able to facilitate protein and RNA folding (9,10). As a matter of fact, it has been observed that there is strong affinity between amino acids and their corresponding codons (11,12), which could imply a direct interaction between proteins and their own mRNAs (13,14). Indeed, TAR DNA-binding protein 43 (TDP-43) and Fragile X Mental Retardation protein (FMRP) have been found to interact with their own mRNAs (15,16). In these cases, expression is regulated by a negative feedback loop involving the 30 untranslated region (UTR). Other autogenous associations have been observed in proteins associated with cell proliferation and gene expression (17,18). Also structurally disordered proteins such as Serine/Arginine-rich splicing factor 2 (SRSF2) (19) as well as heterogeneous ribonucleoprotein members (20,21) are able to inhibit their translation by associating with their own mRNAs. How often do autogenous associations occur in the human proteome? Recent technological advances revealed that a large number of proteins have RNAbinding abilities (22), which suggests that interaction with cognate mRNAs could be more frequent than previously thought. Are autogenous associations linked to specific functions? It is possible that autoregulatory mechanisms are involved in controlling protein production. For instance, in the case of TDP-43 and FMRP, inhibition of expression via autogenous interaction is a way to preserve protein functionality (15,16). Overexpression leads to high protein production and enhanced amyloidogenicity, resulting in harmful gain- or loss-of-function effects on cellular metabolism (23). In this work, we focused on the ability of proteins to establish autogenous associations. Using our computational approach catRAPID (24), we studied the occurrence of these interactions in the human proteome. A large-scale analysis was performed to identify the role of autogenous associations in biological pathways and characterize their properties. MATERIALS AND METHODS Biological pathway annotations We downloaded (September 2012) pathway data from two manually curated and high-quality resources: Reactome (25) and the NCI Pathway Interaction Database (NCIPID) (26). The Reactome annotations (version 41) were gathered via the BioMart query interface returning a list of 167 canonical pathways containing 5375 unique protein coding genes, whereas the NCI-PID pathways were fetched directly from the database website (241 pathways, 2053 unique protein coding genes). In both cases, UniprotKB (27) accession numbers were converted to Ensembl (version 68) gene identifiers using the UniprotKB id-mapping file (version 2012_08). Subsequently, the gene pathway annotations were transferred to the corresponding polypeptides and coding/non-coding transcripts. ProteinRNA interaction prediction We used the catRAPID algorithm (24) to predict interaction propensity among all peptides and transcripts belonging to Reactome and NCI-PID pathways. catRAPID was trained on a large set of proteinRNA pairs available in the Protein Data Bank to discriminate interacting and non-interacting molecules using secondary structure propensities, hydrogen bonding and van der Waals contributions (28). The method was tested on the non-nucleic acid-binding database (NNBP; area under the receiver operating characteristic curve of 0.92), the NPInter database (area under the receiver operating characteristic curve of 0.88) and a number of individual interactions (e.g. RNAse mitochondrial RNA MRP and X-inactive specific transcript XIST networks; average accuracy of 78%). Owing to CPU limitations in the calculation (29), we restricted the predictions to RNA sequences with a length between 50 and 1500 nt as well as to polypeptides with a length between 50 and 750 amino acids. The fragment and strength algorithms were used to identify regions involved in the binding and compute the specificity with respect to random proteinRNA associations (5,29). For each proteinRNA pair under investigation, a reference set of (...truncated)