Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2013 Dec;27(12):1051-65.
doi: 10.1007/s10822-013-9696-9. Epub 2013 Dec 5.

Simultaneous prediction of binding free energy and specificity for PDZ domain-peptide interactions

Joseph J Crivelli  1 , Gordon Lemmon, Kristian W Kaufmann, Jens Meiler
Affiliations

Simultaneous prediction of binding free energy and specificity for PDZ domain-peptide interactions

Joseph J Crivelli et al. J Comput Aided Mol Des. 2013 Dec.

Abstract

Interactions between protein domains and linear peptides underlie many biological processes. Among these interactions, the recognition of C-terminal peptides by PDZ domains is one of the most ubiquitous. In this work, we present a mathematical model for PDZ domain-peptide interactions capable of predicting both affinity and specificity of binding based on X-ray crystal structures and comparative modeling with ROSETTA. We developed our mathematical model using a large phage display dataset describing binding specificity for a wild type PDZ domain and 91 single mutants, as well as binding affinity data for a wild type PDZ domain binding to 28 different peptides. Structural refinement was carried out through several ROSETTA protocols, the most accurate of which included flexible peptide docking and several iterations of side chain repacking and backbone minimization. Our findings emphasize the importance of backbone flexibility and the energetic contributions of side chain-side chain hydrogen bonds in accurately predicting interactions. We also determined that predicting PDZ domain-peptide interactions became increasingly challenging as the length of the peptide increased in the N-terminal direction. In the training dataset, predicted binding energies correlated with those derived through calorimetry and specificity switches introduced through single mutations at interface positions were recapitulated. In independent tests, our best performing protocol was capable of predicting dissociation constants well within one order of magnitude of the experimental values and specificity profiles at the level of accuracy of previous studies. To our knowledge, this approach represents the first integrated protocol for predicting both affinity and specificity for PDZ domain-peptide interactions.

PubMed Disclaimer

Conflict of interest statement

Conflicts of interest: none

Figures

Figure 1
Figure 1
Stereoview of the DLG4-3 PDZ domain with bound KKETWV peptide (PDB ID: 1TP5). The canonical hydrophobic C-terminal side chain (ΦCOO−) of the peptide inserts into a hole on the domain surface. The interaction is further stabilized by backbone-backbone hydrogen bonds between the C-terminal carboxylate moiety of the peptide and the “carboxylate binding loop” (CBL) of the domain. Backbone-backbone hydrogen bonds also allow the peptide to participate in antiparallel strand pairing interactions with β-strand 2 (β2), which adjoins β-strand 3 (β3). Side chain-side chain hydrogen bonds are present throughout the PDZ domain-peptide interface and those involving α-helix 2 (α2) are particularly important for interaction specificity.
Figure 2
Figure 2
Schematic overview of the PDZ domain-peptide interaction prediction protocols tested. All possible combinations of domain and peptide sequences were threaded onto a structure determined through x-ray crystallography. Structural refinement was carried out through one of five schemes involving rigid body docking, flexible docking, iterative repacking and minimization, or combinations of these. For protocols with iterative repacking and minimization, there were eight alternating cycles of repacking of side chains at the interface and gradient-based minimization of the entire structure. Throughout the eight iterations, the repulsive weight (wrep) was ramped up. In the final step of each protocol, the domain and peptide were separated and the binding energy (ΔΔE) was calculated as the difference between the total energies of the bound and unbound complexes.
Figure 3
Figure 3
Impact of linear reweighting on prediction of PDZ domain-peptide interactions. Charts or graphics on the left hand side correspond to prediction using default Rosetta Score12 weights and those on the right hand side correspond to prediction using optimized weights. a: Through FPD + IRM, area under the receiver operating characteristic curve for prediction of frequent/infrequent amino acids improved by 0.04. b: Correlation of computational binding energies predicted through FPD + IRM and experimental binding energies (ΔΔEtotal and ΔΔG, respectively) for the DLG4-3 PDZ domain and 28 hexapeptides improved by 0.18. c: Sequence logos depicting computationally predicted and phage-derived specificity profiles for interactions between the wild type Erbin PDZ and hexapeptides. Following weight optimization, the average distance between profiles predicted through FPD + IRM and profiles predicted through phage display (dAvg) decreased by 0.20 (i.e. predictive accuracy increased). d: Profiles predicted through RBD + IRM and profiles predicted through phage display for interactions between the wild type Erbin PDZ and tetrapeptides. After weight optimization, dAvg increased by 0.53 (i.e. predictive accuracy decreased).
Figure 4
Figure 4
Specificity prediction performance using flexible peptide docking followed by iterative repacking and minimization (FPD + IRM) on the training set of Erbin PDZ single mutants. Distance values (dk) are represented for all 92 domains at each peptide position k in the heat map. Domains and peptide positions marked with an asterisk performed worse (i.e. had a greater dk) with optimized weights than with default Rosetta Score12 weights. Example sequence logo comparisons are provided, which have numerous notable features. Q51M: Prediction of preferred amino acids and degree of specificity was close to ideal. H79R: Preference for an aromatic residue at P−2 and aspartate at P−3 was detected. V83K: The preferred aspartate residues at P−2 and P−3 were not predicted; however, dAvg was relatively low for this mutant because predicted amino acid frequencies were accurate even though rank ordering was not. R49A: P−3—P0 were well predicted but the slight preference for phenylalanine at P−4 was missed. T48K: Aspartate/glutamate specificity at P−3 was detected albeit in reverse order. S26N: Glycine at P−3 was not predicted and overall specificity at P−4 and P−5 was under-predicted although preference for an aromatic residue at P−5 was detected. S28K: Glycine at P−3 was predicted but specificity for phenylalanine at P−4 was dramatically over-predicted. L23V: This was among the worst performing mutants due to lack of detection of isoleucine preference at P0, under-prediction of specificity for threonine at P−2, and failure to predict the strong preference for acidic residues at P−3.
Figure 5
Figure 5
Prediction of affinity and specificity using flexible peptide docking followed by iterative repacking and minimization (FPD + IRM) for an independent set of five PDZ domains. Comparison of predicted and experimental ΔΔG values was performed using the dΔΔG metric; comparison of predicted and experimental specificity profiles was performed using the dk metric at each peptide position k. Both are illustrated in the heat map at the center of the figure. The ratio of the average dk in the independent set to that of the training set is provided in the last row of the heat map. The correlation of dPep (a measure of the extent to which a peptide is represented in the experimentally determined specificity profile) and dΔΔG for five peptides interacting with the SNTA1-1 PDZ domain is shown at left. Two example sequence logo comparisons are provided at right, which have several notable features. DLG1-1: This was the best performing domain due to nearly ideal predictions at P−5—P−2, though the preference for a hydrophobic residue at P−1 and the strong preference for valine at P0 were not captured. MPDZ-10: This was the worst performing domain because strong specificities for arginine, serine, and aspartate were missed at P−4, P−2, and P−1, respectively; however, another basic residue (lysine) was predicted in place of arginine, another polar residue (threonine) was predicted in place of serine, and the prediction at P−5 was very accurate.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

References

    1. Pawson T, Nash P. Assembly of cell regulatory systems through protein interaction domains. Science. 2003;300:445. - PubMed
    1. Socolich M, Lockless SW, Russ WP, Lee H, Gardner KH, Ranganathan R. Evolutionary information for specifying a protein fold. Nature. 2005;437:512–518. - PubMed
    1. Russ WP, Lowery DM, Mishra P, Yaffe MB, Ranganathan R. Natural-like function in artificial WW domains. Nature. 2005;437:579–583. - PubMed
    1. Grigoryan G, Reinke AW, Keating AE. Design of protein-interaction specificity gives selective bZIP-binding peptides. Nature. 2009;458:859–864. - PMC - PubMed
    1. Tonikian R, Xin X, Toret CP, Gfeller D, Landgraf C, Panni S, Paoluzi S, Castagnoli L, Currell B, Seshagiri S, Yu H, Winsor B, Vidal M, Gerstein MB, Bader GD, Volkmer R, Cesareni G, Drubin DG, Kim PM, Sidhu SS, Boone C. Bayesian modeling of the yeast SH3 domain interactome predicts spatiotemporal dynamics of endocytosis proteins. PLoS Biol. 2009;7:e1000218. - PMC - PubMed

Publication types

MeSH terms

Substances

LinkOut - more resources