CRK: An evolutionary approach for distinguishing biologically relevant interfaces from crystal contacts

Abstract

Protein crystals contain two different types of interfaces: biologically relevant ones, observed in protein-protein complexes and oligomeric proteins, and nonspecific ones, corresponding to crystal lattice contacts. Because of the increasing complexity of the objects being tackled in structural biology, distinguishing biological contacts from crystal contacts is not always a trivial task and can lead to wrong interpretation of macromolecular structures. We devised an approach (CRK, core-rim K(a)/K(s) ratio) for distinguishing biologically relevant interfaces from nonspecific ones. Given a protein-protein interface, CRK finds a set of homologs to the sequences of the proteins involved in the interface, retrieves and aligns the corresponding coding sequences, on which it carries out a residue-by-residue K(a)/K(s) ratio (omega) calculation. It divides interface residues into a "rim" and a "core" set and analyzes the selection pressure on the residues belonging to the two sets. We developed and tested CRK on different datasets and test cases, consisting of biologically relevant contacts, nonspecific ones or of both types. The method proves very effective in distinguishing the two categories of interfaces, with an overall accuracy rate of 84%. As it relies on different principles when compared with existing tools, CRK is optimally suited to be used in combination with them. In addition, CRK has potential applications in the validation of structures of oligomeric proteins and protein complexes.

An analysis of oligomerization interfaces in transmembrane proteins

“…First of all we calculated the buried surfaces and number of interface core residues, which, as shown before for soluble proteins [6,32] are a strong indication of an interface to be biological. Additional file 1 presents the data for all interfaces.…”

“…Additional file 1 presents the data for all interfaces. We compared the values for the TM interfaces with those of a composite dataset of soluble protein interfaces, obtained by merging the DCbio [6], PLP [32], Ponstingl dimer [33] and Bahadur dimer [34] sets. Overall the geometry is quite similar to that of soluble proteins with large interfaces (only 7 interfaces below 900 Å 2 ) and many core residues (only 30 interfaces with 5 or fewer core residues, 15 with 4 or fewer).…”

“…Interface core residues are considered those that bury more than 95% of their ASAs upon interface formation [32]. For the evolutionary predictions the version 2013_08 of the UniProt database was used.…”

“…Statistics were gathered for both our newly compiled biological TM interfaces dataset and a soluble interface dataset composed of several published datasets: DCbio [6], PLP [32], Ponstingl dimers [33] and Bahadur dimers [34]. The enrichments are defined as the log-odds ratios of frequencies in interface core residues (at 95% burial cut-off) with respect to the frequencies of all residues in the full proteins.…”

ProteinsPlus: a web portal for structure analysis of macromolecules

“…Diverse methods exist which try to predict PPIs based on the computation of free energies or classification models based on physico-chemical and geometrical descriptors, e.g. PQS (48), NOXclass (49), EPIC (50), PISA (51), DiMoVo (52), CRK (53), OringPV (54), IPAC (55) or IChemPIC (56). Most of those methods achieve high accuracies of 85–97%.…”