Compared to other areas of computational annotation and prediction, it is very difficult to predict reliably what protein-protein interactions (PPI) an individual protein engages in (e.g. in disease development) and what the characteristics of the interaction are. This is often the case even if knowledge of PPI is available from homologous proteins;while the inference seems true more often than not, instances of differing behavior of homologous proteins in model organisms have often led to surprises when inferences to humans turned out invalid (e.g. in the case of leptin (Gaucher et al 2003)). There are indications both from theory (grounding in redundancy and robustness of pathways) and practice that some of these instances may be reflected in detectable signal of adaptive (or positive) selection pressure. In this project we will systematically screen chordate protein families using their protein sequences;the encoding gene sequences;and modelled structures of hypothetical PPI observed between homologous partners. We will use previously developed data bases and software developed in our groups, most importantly The Adaptive Evolution Database TAED (Liberles et al 2001) and the Binary SubComplex Database BISC (Juettemann &Gerloff 2011). Signals of evolutionary adaptation will be sought in the protein surface properties and the encoding genes, in context of the 3D-structure of the target proteins. Specifically we will screen for gene families where we find indications for possible changes among orthologs to human proteins closely related species, and closely similar paralogous domains e.g. in surface receptors. Using high-end computational biology methods we will produce refined complex models and attempt to validate the predicted not-conserved PPI computationally. Due to currently largely lacking benchmarks we will also validate a subset in the laboratory and further characterize the nature of the change in PPI (e.g. whether a new interaction has evolved;binding specificity has changed, or an the interaction has been lost altogether). Our approach integrates (data-driven) discovery and (hypothesis-driven) validation of examples of evolutionary change within protein families. We surmise that (i) non-conserved/individual PPI characteristics are more common than is currently assumed in PPI prediction and annotation, even among orthologs;and that (ii) if we can devise strategies to identify instances that challenge the paradigm, this indicates an angle for further improving PPI prediction for individual proteins. Perhaps most excitingly if we are correct, this should be considered in the modelling and analysis of PPI networks, ultimately for applications in personalized medicine and the design of potentially interaction-disrupting drugs. Similarly important this approach may help flag instances where model organism-based inferences to human diseases are unlikely to hold true.
This project aims to shed light on how valid a common assumption is, that underpins most inference of the function of e.g. human proteins, from knowledge regarding the related proteins in other species. Are interactions with other directly related proteins, really nearly always conserved? Computational methodology aiming to detect signals of selection pressure in genes and protein surface properties will be used on computational models of the target proteins, and predictions will be validated by laboratory experiments, to generate new insight in this area.
| Sharman, Joanna L; Gerloff, Dietlind L (2013) MaGnET: Malaria Genome Exploration Tool. Bioinformatics 29:2350-2 |
