This year we head to Boston for 3Dsig and ISMB 2014. The outcome was excellent with James Dunbar giving a talk at the 3Dsig on “Examining variable domain orientations in antigen receptors gives insight into TCR-like antibody design” and Alistair Martin oral poster presentation at ISMB on “Unearthing the structural information contained within mRNA ”. The Deane group received the most votes by different judges for the poster competition at 3Dsig with James Dunbar poster winning the best poster prize and Jinwoo Leem and Reyhaneh Esmaielbeiki posters receiving the honorary mentioned (all presented posters are Here).
This blog post contains a brief description of what we found most interesting at the conference.
N-terminal domains in two-domain proteins are biased to be shorter and predicted to fold faster than their C-terminal counterparts.
Authors: Jacob, Etai; Unger, Ron; Horovitz, Amnon
Chosen by: Alistair Martin
Is is not surprising that protein misfolding is selected against in the genome, with aggregation of misfolded proteins being associated to a an array of diseases. It is suggested that multi-domain proteins are more prone to misfolding and aggregation due to an effective higher local protein concentration. Jacob et al. investigated what mechanisms have developed to avoid this, focussing on ~3000 two domain proteins contained within Swiss-Prot.
They found that there are notable differences between the N- and C-terminal domains. Firstly, there exists a large propensity for the C-terminal domain to be longer than the N-terminal domain (1.6 times more likely). Secondly, the absolute contact order (ACO) is greater in the C-terminal domain when compared to the N-terminal domain. Both length and ACO inversely correlate to folding speed and thus they draw the conclusion that the first domain produced by the ribosome is under pressure to fold faster than the latter domain. These observations are enhanced in prokaryotes and lessened in eukaryotes, thereby suggesting a relationship to translational speed and cotranslational folding.
A novel tool to identify molecular interaction field similarities
Authors: Matthieu Chartier and Rafael Najmanovich
Chosen by: Claire Marks
One of the prize-winning talks at 3Dsig this year was “A novel tool to identify molecular interaction field similarities”, presented by Matthieu Chartier from the Université de Sherbrooke. The talk introduced IsoMIF – a new method for finding proteins that have similar binding pockets to one another. Unlike previous methods, IsoMIF does not require the superposition of the two proteins, and therefore the binding sites of proteins with very different folds can be compared.
IsoMIF works by placing probes at various positions on a grid in the binding pocket and calculating the interaction potential. Six probe types are used: hydrogen bond donor and acceptor, cation, anion, aromatic group, and hydrophobic group. A graph-matching algorithm then finds similar pockets to a query using the interaction potentials for each probe type at each grid point. On a dataset of nearly 3000 proteins, IsoMIF achieved good AUC values of 0.74-0.92.
The promise of evolutionary coupling for 3D biology
Presenter: Debora Marks
Chosen by: Reyhaneh Esmaielbeiki
I was impressed by the keynote talk given by Debora Marks at the 3Dsig. She gave an overall talk on how their group have worked on detecting evolutionary couplings (EC) between residues in proteins and how they use this information in predicting folds. In general, looking at the interacting residues in a 3D structure and comparing these position in a MSA displays co-evolving relationship. But the challenge is to solve the inverse, from sequence to structure, since not necessary all co-evolving residues are close in the 3D space (this relationship is shown in the figure below).
Debora showed that previous studies for detecting co-evolving residues used Mutual Information (MI). But, comparing the prediction out of MI with contacts maps shows that these methods perform poorly. This is because MI looks at the statistics of pair of residue at a time while residues in proteins are highly coupled and pairs are not independent from other pairs. Therefore, MI works good for RNA but not for proteins. Debora’s group have used mean field and pseudo likelihood maximization to overcome the limitation of MI and introduced the EVcoupling tool for predicting EC (Marks et al. PLoS One, 6(12), 2011). They have used the predicted EC as a distance restraint to predict the 3D structure of proteins using EVfold. Using EVfold they have managed to build structure with 2-5Å accuracy.
In a more recent work, they have built EVfold-membrane which is specific for membrane proteins (Hopf et al. Cell, 149(7), 1607-1621, 2012) and they tried modeling membrane proteins with unknown experimental structures. Recently close homologues to these structures were released and comparisons show that EVfold-membrane structures have accuracy of 3 to 4Å.
She also discussed the usage of detecting EC in identifying functional residues involved in ligand binding and conformational changes with an interesting example of two GPCRs, adrenergic beta-2 receptor and an opioid receptor (paper link).
She concluded her talk by talking about her recent work EVcomplex (paperlink). The aim is to use the detected EC between two different protein chains and use this information in the docking software as a distance restraint. Although, this method has provided models of the E.coli ATP synthase but there are currently several limitation (all mentioned in the Discussion of the paper) for using this work in large scale.
in general, EC was a popular topic at the conference with interesting posters from the UCL Bioinformatics group.
Characterizing changes in the rate of protein-protein dissociation upon interface mutation using hotspot energy and organization
Authors: Rudi Agius, Mieczyslaw Torchala, Iain H. Moal, Juan Fernández-Recio, Paul A. Bates
Chosen by:Jinwoo Leem
This was a paper by Agius et al. published in 2013 in PLOS Comp Biol. Essentially, the work was centralised around finding a set of novel descriptors to characterise mutant proteins and ultimately predict the koff (dissociation rate constant) of mutants. Their approach is quite unique; they perform two rounds of alanine-scanning mutagenesis, one on the wild-type (WT) and one on the mutant protein. They identify ‘hotspot’ residues as those that have a change of G of 2kcal/mol (or more) from alanine scanning, and the descriptor is formed from the summation of the energy changes of the hotspots.
The results were very encouraging, and from their random forest-trained model with hotspot descriptors, they see correlations to koff up to 0.79. However, the authors show that traditional ‘molecular’ descriptors (e.g. statistical potentials) perform just as well, with a correlation of 0.77. The exact contribution of their ‘hotspot’ descriptors to the prediction of koff seems unclear, especially considering how well molecular descriptors perform. Having said this, the paper shows a very unique way to approach the issue of predicting the effects of mutations on proteins, and on a larger dataset with a more diverse range of proteins (not necessarily mutants, but different proteins altogether!) these ‘hotspot’-specific methods may prove to be much more predictive.
On the Origin and Completeness of Ligand Binding Pockets with applications to drug discovery
Authors: Mu Gao & Jeffrey Skolnick.
Presented by Jeffrey Skolnick
Chosen by:Nicholas Pearce
The prevalence of ligand-binding pockets in proteins enables a wide range of biochemical reactions to be catalysed in the cell. Jeffrey Skolnick presented research which proposes that ligand-binding pockets are inherent in proteins. One mechanism that he hypothesised for the creation of these pockets is the mis-stacking of secondary structure elements, leading to imperfections in their surfaces – pockets. Using their method for calculating pocket similarity – APoc – Gao & Skolnick characterised the space of ligand-binding pockets, using artificial structures and structures from the PDB, and find it to be ~400-1000 pockets in size.
They suggest that the relatively small size of pocket-space could be one of the reasons that such a large amount of off-target promiscuity is seen in drug design attempts.
From this result, Skolnick went on to discuss several interesting possibilities for the evolutionary history of ligand-binding pockets. One of the interesting hypotheses is that many, if not all, proteins could have inherent low-level catalytic ability across a wide range of biochemical reactions. Motivated by the small size of pocket-space, it is conceivable that one protein would be able to catalyse many different reactions – this could give an insight into the evolutionary history of protein catalysis.
If primordial proteins could catalyse many different reactions, albeit inefficiently, this would give a possibility for how the first lifeforms developed. Nature need only then work on increasing specificity and efficiency from a background of weak catalytic ability. However, even through the course of evolution to produce more specific proteins, this background of activity could remain, and explain the background of ‘biochemical noise’ that is seen in biological systems.
Drug promiscuity and inherent reactivity may not only be present due to the small size of pocket-space – they may be a direct consequence of evolution and the fundamental properties of proteins.