Modeling protein-peptide interactions using protein fragments: fitting the pieces?

An estimated 15-40% of all interactions in the cell are mediated through protein-peptide interactions [1, 2] meaning that, at the most extreme, nearly every protein is affected either directly or indirectly by peptide-binding events.

We compared the modes of interaction between protein-peptide interfaces and those observed within monomeric proteins and found surprisingly little differences [3]. Over 65% of 731 protein-peptide interfaces could be reconstructed within 1 Å RMSD using solely fragment interactions occurring in monomeric proteins, using our fragment database BriX containing over 1000 non-redundant protein structures [4]. Interestingly, more than 80% of interacting fragments used in reconstructing a protein-peptide binding site were obtained from monomeric proteins of an entirely different structural classification, with an average sequence identity below 15%. Nevertheless, geometric properties perfectly match the interaction patterns observed within monomeric proteins (see Figure 1), suggesting that our fragment interaction approach might provide an alternative to homology modelling.

Figure 1

Relation between intermolecular protein-peptide interface architectures (blue for receptor, green for peptide ligand) and intramolecular protein architectures from our database of monomeric proteins, BriX (red) [4].

We show the usefulness of our method by redesigning the interaction scaffold of nine protein-peptide complexes, for which five of the peptides can be modelled to within 1 Å RMSD of the original peptide position.

These data suggest that the wealth of structural data on monomeric proteins could be harvested to model protein-peptide interactions and, more importantly, that sequence homology is no prerequisite. In addition, we have made our dataset of 505 non-redundant protein-peptide complexes from 1431 entries in the PDB available at http://pepx.switchlab.org[5] and the BriX database at http://brix.crg.es[6].