Membrane protein topology : prediction, experimental mapping and genome-wide analysis

Abstract

Membrane proteins constitute a large and important class of proteins. They are responsible for many crucial processes in living cells, and about 25% of all proteins encoded by a typical genome are predicted to be membrane proteins. Despite their importance, very few structures have been determined at atomic resolution. As an alternative, the topology of a membrane protein may provide valuable information on e.g. the location of ligand binding sites on receptors and transporters.

In this thesis, bioinformatics and experimental approaches have been applied to study membrane protein topologies. A method to identify reliably predicted topologies or partial topologies, using a consensus of five different prediction algorithms, has been developed. The approach has been used in combination with limited fusion protein analysis to rapidly produce topology models of E. coli inner-membrane proteins. In addition, a comprehensive statistical analysis has been performed on predicted membrane protein topologies from the genomes of 108 organisms. The correlation between the number of independent prediction methods that agree on a topology model, and the reliability of the predicted topology was investigated. Analysis on a data set of E. coli inner-membrane proteins with experimentally determined topologies revealed that predicted topologies were very reliable (>90% correct) when all methods or all but one agreed on the prediction. When the consensus approach was applied to all putative membrane proteins in a number of completely sequenced genomes, we found that highly reliable topology predictions could be made for nearly half of all membrane proteins in most prokaryotic organisms. The method was subsequently extended to predict also parts of a global topology with an equally high reliability. With the partial consensus topology predictions, the average fraction of prokaryotic membrane proteins for which either a complete or a partial topology of high reliability could be predicted was increased to approximately 70%. Fusion protein analysis has been used in combination with the consensus topology predictions to reduce the amount of experimental efforts required to establish a topology model. The topologies of 43 E. coli inner-membrane proteins have been mapped using the combined approach.

Comparative studies of amino acid distributions in predicted membrane proteins from a total of 108 genomes were performed. The main objective of the study was to establish if the so called positive-inside rule (the asymmetric distribution of Lys and Arg residues between polar loops on opposite sides of a membrane) is valid for all organisms where the genome has been sequenced. An algorithm for identification of membrane spanning helices was optimised to yield a minimum number of incorrect predictions. The optimised method was then applied to all putative membrane proteins in each genome, and the average frequency bias of each amino acid residue type between polar segments on opposite sides of the membrane was estimated. The study strongly suggested that the positive-inside rule is valid in most, if not all, organisms investigated, although it is less pronounced in eukaryotes than in eubacteria and archaea.