Final Report Summary - EEPPIBM (Exploring the evolution of protein-protein interactions and their networks using biophysical models)
The initial project consisted of two parts. Firstly, to develop and benchmark fast empirical methods of determining the binding energy of protein-protein interactions and how they change upon mutation, and then to apply these methods to investigate how protein-protein interactions evolve. Excellent progress has been made in both parts, with a number of publications resulting from the earlier stages of the project, and with manuscripts in preparation, and investigations continuing, for the second part.
One of the first tasks was to develop and implement fast empirical energy functions for protein-protein interactions suitable for the investigation of how energy changes upon mutation in the context of evolution. Towards this goal, we developed the first energy functions to be trained from experimental changes in binding free energy upon mutation. These functions were based on the idea that, upon mutation, there are changes in the intermolecular contacts between proteins, such as the addition and removal of stabilising interactions, or the addition and removal of destabilising interactions. Our functions were based on the idea that the change in binding free energy upon mutation could be related to these changes in contacts. Thus, using a data set of 1949 mutations for which a crystal structure of the wild-type complex is known, we modelled the mutant interactions and evaluated their changes in intermolecular contacts. From this, we trained a potential of pairwise additive contact energies using the experimental changes in binding free energy. The method was validated using absolute binding free energies and its ability to rank protein-protein docking poses. This was published in the prestigious Journal of Chemical Theory and Computation. More recently, the method has performed very well in a new benchmark of binding energies curated from the literature with collaborators in Utrech, Boston and London (manuscript in preparation). It was also extended to investigate the surface energies of protein-protein interactions, which was published in the journal Proteins: Structure, Function and Bioinformatics.
Also in the earlier stages of the project, we evaluated over 100 scoring functions for their applicability to protein-protein interactions, including their ability to predict changes in binding affinity upon mutation and at being able to rank docking poses. These functions include many molecular mechanics methods and statistical potentials, as well as terms relating to electrostatics, desolvation, hydrogen bonding, interface packing, surface complementarity and so on. Some of these investigations were published in BMC Bioinformatics, and indicated promising strategies to combine the scoring functions. This publication garnered a number of inquiries about the implementation of the functions, as well as data requests. As as result of this interest, we created the CCharPPI web server in order to make these functions available to the wider scientific community (http://life.bsc.es/pid/ccharppi/). This server was published in the journal Bioinformatics, and is highly accessed. These investigations also inspired us to combine the functions using methods developed for performing web searches. This proved very effective at ranking docked poses, and the method has now been implimented in the SwarmDock server (http://bmm.cancerresearchuk.org/~SwarmDock/) with a manuscript in preparations. More closely aligned with the initial objectives of the project, the ability of the functions to evaluate the effect of mutations was then used in the design of protein inhibitors of a metalloprotease with no known natural inhibitors, with the potential for commercialisation for the use in mass spectroscopy. The designs are currently being evaluated by collaborators at the Molecular Biology Institute of Barcelona.
Finally, with regards to the application of the developed methods to the investigation of the evolution of protein-protein interactions, progress has also been made despite some of the diversions described above. Our scoring functions were applied to a large set of structural interologs derived from the protein databank. Exhaustive computational alanine-scanning of the interologs revealed patterns in the distribution of binding hotspots, and how these have changed over evolutionary timescales. Following up from this, we are currently investigating representative sequences within the lineages that are intermediate between the interologs, and are beginning to garner new insights into how nature shapes the spatial organisation of energy imbuing contacts across the binding interface. At the same time, ongoing investigations into the tracking of absolute binding energies through evolutionary lineages are ongoing, and some preliminary ideas on this topic have been published in the journal Current Opinion in Structural Biology.
One of the first tasks was to develop and implement fast empirical energy functions for protein-protein interactions suitable for the investigation of how energy changes upon mutation in the context of evolution. Towards this goal, we developed the first energy functions to be trained from experimental changes in binding free energy upon mutation. These functions were based on the idea that, upon mutation, there are changes in the intermolecular contacts between proteins, such as the addition and removal of stabilising interactions, or the addition and removal of destabilising interactions. Our functions were based on the idea that the change in binding free energy upon mutation could be related to these changes in contacts. Thus, using a data set of 1949 mutations for which a crystal structure of the wild-type complex is known, we modelled the mutant interactions and evaluated their changes in intermolecular contacts. From this, we trained a potential of pairwise additive contact energies using the experimental changes in binding free energy. The method was validated using absolute binding free energies and its ability to rank protein-protein docking poses. This was published in the prestigious Journal of Chemical Theory and Computation. More recently, the method has performed very well in a new benchmark of binding energies curated from the literature with collaborators in Utrech, Boston and London (manuscript in preparation). It was also extended to investigate the surface energies of protein-protein interactions, which was published in the journal Proteins: Structure, Function and Bioinformatics.
Also in the earlier stages of the project, we evaluated over 100 scoring functions for their applicability to protein-protein interactions, including their ability to predict changes in binding affinity upon mutation and at being able to rank docking poses. These functions include many molecular mechanics methods and statistical potentials, as well as terms relating to electrostatics, desolvation, hydrogen bonding, interface packing, surface complementarity and so on. Some of these investigations were published in BMC Bioinformatics, and indicated promising strategies to combine the scoring functions. This publication garnered a number of inquiries about the implementation of the functions, as well as data requests. As as result of this interest, we created the CCharPPI web server in order to make these functions available to the wider scientific community (http://life.bsc.es/pid/ccharppi/). This server was published in the journal Bioinformatics, and is highly accessed. These investigations also inspired us to combine the functions using methods developed for performing web searches. This proved very effective at ranking docked poses, and the method has now been implimented in the SwarmDock server (http://bmm.cancerresearchuk.org/~SwarmDock/) with a manuscript in preparations. More closely aligned with the initial objectives of the project, the ability of the functions to evaluate the effect of mutations was then used in the design of protein inhibitors of a metalloprotease with no known natural inhibitors, with the potential for commercialisation for the use in mass spectroscopy. The designs are currently being evaluated by collaborators at the Molecular Biology Institute of Barcelona.
Finally, with regards to the application of the developed methods to the investigation of the evolution of protein-protein interactions, progress has also been made despite some of the diversions described above. Our scoring functions were applied to a large set of structural interologs derived from the protein databank. Exhaustive computational alanine-scanning of the interologs revealed patterns in the distribution of binding hotspots, and how these have changed over evolutionary timescales. Following up from this, we are currently investigating representative sequences within the lineages that are intermediate between the interologs, and are beginning to garner new insights into how nature shapes the spatial organisation of energy imbuing contacts across the binding interface. At the same time, ongoing investigations into the tracking of absolute binding energies through evolutionary lineages are ongoing, and some preliminary ideas on this topic have been published in the journal Current Opinion in Structural Biology.