Prediction of indirect interactions in proteins

Abstract

Background: Both direct and indirect interactions determine molecular recognition of ligands by proteins. Indirect interactions can be defined as effects on recognition controlled from distant sites in the proteins, e.g. by changes in protein conformation and mobility, whereas direct interactions occur in close proximity of the protein's amino acids and the ligand. Molecular recognition is traditionally studied using three-dimensional methods, but with such techniques it is difficult to predict the effects caused by mutational changes of amino acids located far away from the ligand-binding site. We recently developed an approach, proteochemometrics, to the study of molecular recognition that models the chemical effects involved in the recognition of ligands by proteins using statistical sampling and mathematical modelling.

Results: A proteochemometric model was built, based on a statistically designed protein library's (melanocortin receptors') interaction with three peptides and used to predict which amino acids and sequence fragments that are involved in direct and indirect ligand interactions. The model predictions were confirmed by directed mutagenesis. The predicted presumed direct interactions were in good agreement with previous three-dimensional studies of ligand recognition. However, in addition the model could also correctly predict the location of indirect effects on ligand recognition arising from distant sites in the receptors, something that three-dimensional modelling could not afford.

Conclusion: We demonstrate experimentally that proteochemometric modelling can be used with high accuracy to predict the site of origin of direct and indirect effects on ligand recognitions by proteins.

Figure 1

Increase in affinity of α-MSH predicted from the binary proteochemometric model. Shown is the increase in affinity S^U (log(M)) for α-MSH afforded by swapping each of segments S1 – S5 in the MC₄ receptor with the corresponding segment in the MC₁ receptor as predicted from the binary proteochemometrics model.

Figure 2

Predicted and experimentally determined increase in α-MSH affinity for site-directed mutants. Predicted and experimentally determined increase in affinity (i.e., computed SU and measured pK increase, log(M), respectively) for α-MSH afforded by point mutations in the MC₄ receptor. Shown by red bars is the change in affinity predicted by the model utilizing physicochemical descriptions of amino acids of the receptors' transmembrane regions and facing potential ligand binding clefts to be afforded by the indicated point mutations (i.e., the predicted increase in affinity that should be gained by exchanging the indicated amino acids in the MC₄ receptor with the corresponding ones in the MC₁ receptor). Shown in blue bars is the experimentally determined change in affinity for these mutations vs. the wild-type MC₄ receptor. Significance (nonparametric Wilcoxon Rank Sum statistical test [29]) denoted as follows: * p <0.05; ** p < 0.005; *** p < 0.0005.

Figure 3

Experimentally determined increase in α-MSH affinity by mutations in segment S4. Experimentally determined increase in affinity (pK) afforded by mutations N240G/M241L/I245V (indicated in green), and V253I/V255F/V256L (indicated in blue) (i.e., exchanging amino acid residues in the MC₄ receptor with the corresponding residues in the MC₁ receptor), and swapping intracellular loop 3 (IL3, indicated in dark yellow) in the MC₄ receptor with the corresponding in the MC₁ receptor. Also shown is the increase in affinity for the whole S4 segment (S4, indicated in pale yellow). Significance (nonparametric Wilcoxon Rank Sum statistical test [29]) denoted as indicated in the legend to Fig. 2.

Figure 4

Outline of the mapping of direct and indirect interactions in molecular recognition using proteochemometrics. Outline of a general procedure for mapping molecular recognition by biomacromolecules. Initially, sets of wild-type macromolecules are identified. Using statistical molecular design a library is then created from the wild-type macromolecules. The library can be selected from the wild-type molecules if the initial collection contains sufficient chemical variation. Chemical variation may also be introduced artificially by mutagensis. Shuffling sequence fragments can then be used to create a library, where three or more segments and three or more macromolecules are used as the starting point. After evaluating the interaction of the library with a suitable library of ligands of interest, a proteochemometric model can be created. This model may be used to localize the regions in each macromolecule that contribute to the selectivity of each particular ligand evaluated. It may happen that it is not possible to unambiguously localize individual amino acids within a particular region, due to co-varying amino acid positions in the macromolecular library. In that case, an extension of the library can be made in order to resolve the ambiguity.