Quantitative methods for reconstructing protein-protein interaction histories

Abstract

Protein-protein interactions (PPIs) are vital for the function of a cell and the evolution of these interactions produce much of the evolution of phenotype of an organism. However, as the evolutionary process cannot be observed, methods are required to infer evolution from existing data. An understanding of the resulting evolutionary relationships between species can then provide information for PPI prediction and function assignment. This thesis further develops and applies the interaction tree method for modelling PPI evolution within and between protein families. In this approach, a phylogeny of the protein family/ies of interest is used to explicitly construct a history of duplication and specification events. Given a model relating sequence change in this phylogeny to the probability of a rewiring event occurring, this method can then infer probabilities of interaction between the ancestral proteins described in the phylogeny. It is shown that the method can be adapted to infer the evolution of PPIs within obligate protein complexes, using a large set of such complexes to validate this application. This approach is then applied to reconstruct the history of the proteasome complex, using x-ray crystallography structures of the complex as input, with validation to show its utility in predicting present day complexes for which we have no structural data. The methodology is then adapted for application to transient PPIs. It is shown that the approach used in the previous chapter is inadequate here and a new scoring system is described based on a likelihood score of interaction. The predictive ability of this score is shown in predicting known two component systems in bacteria and its use in an interaction tree setting is demonstrated through inference of the interaction history between the histidine kinase and response regulator proteins responsible for sporulation onset in a set of bacteria. This thesis demonstrates that with suitable modifications the interaction tree approach is widely applicable to modelling PPI evolution and also, importantly, predicting existing PPIs. This demonstrates the need to incorporate phylogenetic data in to methods of predicting PPIs and gives some measure of the benefit in doing so.