In silico screening of mutational effects on enzyme-proteic inhibitor affinity: a docking-based approach

Abstract

Background: Molecular recognition between enzymes and proteic inhibitors is crucial for normal functioning of many biological pathways. Mutations in either the enzyme or the inhibitor protein often lead to a modulation of the binding affinity with no major alterations in the 3D structure of the complex.

Results: In this study, a rigid body docking-based approach has been successfully probed in its ability to predict the effects of single and multiple point mutations on the binding energetics in three enzyme-proteic inhibitor systems. The only requirement of the approach is an accurate structural model of the complex between the wild type forms of the interacting proteins, with the assumption that the architecture of the mutated complexes is almost the same as that of the wild type and no major conformational changes occur upon binding. The method was applied to 23 variants of the ribonuclease inhibitor-angiogenin complex, to 15 variants of the barnase-barstar complex, and to 8 variants of the bovine pancreatic trypsin inhibitor-beta Trypsin system, leading to thermodynamic and kinetic estimates consistent with in vitro data. Furthermore, simulations with and without explicit water molecules at the protein-protein interface suggested that they should be included in the simulations only when their positions are well defined both in the wild type and in the mutants and they result to be relevant for the modulation of mutational effects on the association process.

Conclusion: The correlative models built in this study allow for predictions of mutational effects on the thermodynamics and kinetics of association of three substantially different systems, and represent important extensions of our computational approach to cases in which it is not possible to estimate the absolute free energies. Moreover, this study is the first example in the literature of an extensive evaluation of the correlative weights of the single components of the ZDOCK score on the thermodynamics and kinetics of binding of protein mutants compared to the native state.Finally, the results of this study corroborate and extend a previously developed quantitative model for in silico predictions of absolute protein-protein binding affinities spanning a wide range of values, i.e. from -10 up to -21 kcal/mol. The computational approach is simple and fast and can be used for structure-based design of protein-protein complexes and for in silico screening of mutational effects on protein-protein recognition.

Figure 1

Cartoon representation of the 3D structure of hRI-Ang complex. Residues target of mutagenesis are represented in sticks. Here, as in the following drawings, the protein that was kept fixed in docking simulations (i.e. the target) is coloured in blue, whereas the one sampling the rotational and translational space (i.e. the probe) is coloured in green. Drawings were prepared with the software Pymol [52].

Figure 2

Cartoon representation of the 3D structure of Bn-Bs complex. Residues target of mutagenesis are represented in sticks. Here, as in the following drawings water molecules explicitly included in docking simulations are represented by red spheres.

Figure 3

Cartoon representation of the 3D structure of β-Tryp-BPTI complex. Lysine 15 is the residue target of multiple mutagenesis, and is represented in sticks.

Figure 4

Experimental relative affinities (ΔΔG°) versus relative ZD-s for hRI-Ang interaction. The fitting line equation is ΔΔG° = -0.37 + 0.89ΔZD-s, the correlation coefficient is R = 0.92 and its probability p(R) < 0.0001. The number of experimental points is N = 23.

Figure 5

Experimental relative affinities (ΔΔG°) versus relative ZD-s for Bn-Bs interaction. (A) Plot referred to data reported in Table 2, i.e. 16 variants of Bn-Bs without water molecules at the interface and protonated form of H102A^Bn. The fitting line equation is ΔΔG° = 1.57 + 0.73ΔZD-s, R = 0.77, p(R) = 0.00046, N = 16, where R is the correlation coefficient, p(R) is the probability of such coefficient and N is the number of points. (B) Correlative model derived from the one at point (A) by leaving out four points. The dataset in this plot is limited to the 11 variants of Bn-Bs, for which water positions could be defined at acceptably high resolution (Table 3). The correlation equation and its parameters are: ΔΔG° = 0.35 + 0.95ΔZD-s, R = 0.79, p(R) = 0.0041, N = 11. (C) Same data set as in B, but with ZD-s derived by docking simulations with explicit interfacial water molecules and H102^Bn in its protonated state. The correlation equation and its parameters are, respectively: ΔΔG° = 0.06 + 1.22ΔZD-s, R = 0.90, p(R) = 0.00017, N = 11.

Figure 6

Experimental (ΔΔG°_exp) versus predicted (ΔΔG°_pred) relative affinities for hRI-Ang and Bn-Bs data, analyzed as a unique set. The predicted values refer to a leave-one-out test. The fitting line equation is ΔΔG°_exp = 0.04 + 0.98ΔΔG°_pred, R = 0.87, p(R) < 0.0001 and N = 34.

Figure 7

Plot of experimental ΔG° versus ZD-s for the β-Tryp-BPTI complex, where K15^BPTI was substituted in eight different amino acids named by their one-letter code. (A) Results of docking simulations starting from the X-ray structures of the eight mutants, including explicit water molecules. The fitting line equation is ΔG° = 17.9 - 0.57ZD-s, R = 0.86, p(R) = 0.006 and N = 8. (B) Correlation model derived by docking simulations on the same molecular models as in (A) but without explicit water molecules. The fitting line equation is ΔG° = 14.8 - 0.50ZD-s, R = 0.82, p(R) = 0.012 and N = 8. (C) Correlation model derived from docking simulations starting from the in silico-modelled structures of the eight mutants, with no interface water molecules. The fitting line equation is ΔG° = 24.2 - 0.72ZD-s, R = 0.79, p(R) = 0.019 and N = 8.

Figure 8

A general quantitative model for docking score-based free energy predictions in protein-protein interactions. (A) Linear correlation between average ZD-s and in vitro-determined standard free energy of association for a set of ten protein-protein complexes. Each dot is labelled according to the PDB code of the complex. Experimental and computational data are reported from Ref [4], except for 1BRS and 1A4Y, which both refer to this study. The linear correlation equation is ΔG° = 3.13 -0.37ZD-s (R = 0.97, p(R)< 0.0001, N = 10). (B) Predicted versus in vitro-determined free energy of association of the same ten complexes. The predicted values refer to a leave-one-out test. The fitting equation is ΔG°_exp = -0.51 + 0.96ΔG°_pred (R = 0.96, p(R) < 0.0001, N = 10).