Analysis of residue interaction networks to improve prediction of protein succination

Introduction: Protein succination results from a specific adduction of fumarate to certain cysteine (Cys) residues in proteins. In the last years, the interest toward this type of post-translational Cys modification is markedly increased following the observation of an aberrant level of modified proteins in tumours found in patients affected by hereditary leiomyomatosis and renal cell cancer1, in tissues of obese and diabetic mice2, as well as in cells exposed to dimethyl fumarate3, which is a drug used to treat patients with multiple sclerosis or psoriasis). Despite the compelling findings, however, further efforts remain to be accomplished to obtain a comprehensive examination and characterization of this type of Cys modification. The aim of this study was to elucidate the role of microenvironment-related factors in governing the specificity of protein succination. Methods: Taking advantage of a growing number of fumarate-sensitive proteins and sites4, a computational analysis was carried out. First, the 3D structure of the included proteins (n=41) was retrieved from the PDB repository and used to generate the corresponding 2D representation as residue interacting networks using RINalyzer (http://www.rinalyzer.de), as previously described5. Then, a dataset was built by collecting for each site [51 modifiable Cys (MC) and 227 non-modifiable Cys (NMC)] biochemical and topological data. Finally, a library of predictive models, in the realm of supervised learning (including both regression models and tree-based models) and implemented in R (https://www.r-project.org/), were used to quantify the descriptive/predictive value of the collected data. Results: Significant differences between MC and NMC sites were determined when the frequency distributions for number (p=5.76×10-3, Pearson’s χ2 test) and type (p=2.72×10-4, Pearson’s χ2 test) of Cys-interacting amino acids, accessible surface area of the sulphur atom (p=8.84×10-4, Wilcoxon rank sum test), and rates of the secondary structure type of the Cys-flanking peptides (p=6.51×10-4, Pearson’s χ2 test) were compared. Moreover, the reactivity of a Cys site toward fumarate was accurately predicted when the data for topological features were analysed by using a library of classification models. Conclusion: The propensity of a Cys site in protein to be modified by fumarate should be considered as an emergent property, which depends upon a joint action and interaction of different factors. Some of these factors can be studied by using integrative tools and methods. The analysis of topological data provides helpful strategies for evaluating the likelihood of a Cys site to be modified by fumarate. References: 1. Yang et al. (2014). Metabolites 4: 640-54. 2. Thomas et al. (2012). Obesity (Silver Spring) 20: 263-9. 3. Piroli et al. (2016). Biochem J 462: 231-45. 4. Miglio et al. (2016) Biochim Biophys Acta 1864: 211-8. 5. Doncheva et al. (2012) Nat Protoc 7: 670-85.