Simões Inês C M, Costa Inês P D, Coimbra João T S, Ramos Maria J, Fernandes Pedro A
UCIBIO, REQUIMTE, Departamento de Química e Bioquímica, Faculdade de Ciências, Universidade do Porto , Rua do Campo Alegre, s/n, 4169-007 Porto, Portugal.
J Chem Inf Model. 2017 Jan 23;57(1):60-72. doi: 10.1021/acs.jcim.6b00378. Epub 2016 Dec 22.
Knowing how proteins make stable complexes enables the development of inhibitors to preclude protein-protein (P:P) binding. The identification of the specific interfacial residues that mostly contribute to protein binding, denominated as hot spots, is thus critical. Here, we refine an in silico alanine scanning mutagenesis protocol, based on a residue-dependent dielectric constant version of the Molecular Mechanics/Poisson-Boltzmann Surface Area method. We have used a large data set of structurally diverse P:P complexes to redefine the residue-dependent dielectric constants used in the determination of binding free energies. The accuracy of the method was validated through comparison with experimental data, considering the per-residue P:P binding free energy (ΔΔG) differences upon alanine mutation. Different protocols were tested, i.e., a geometry optimization protocol and three molecular dynamics (MD) protocols: (1) one using explicit water molecules, (2) another with an implicit solvation model, and (3) a third where we have carried out an accelerated MD with explicit water molecules. Using a set of protein dielectric constants (within the range from 1 to 20) we showed that the dielectric constants of 7 for nonpolar and polar residues and 11 for charged residues (and histidine) provide optimal ΔΔG predictions. An overall mean unsigned error (MUE) of 1.4 kcal mol relative to the experiment was achieved in 210 mutations only with geometry optimization, which was further reduced with MD simulations (MUE of 1.1 kcal mol for the MD employing explicit solvent). This recalibrated method allows for a better computational identification of hot spots, avoiding expensive and time-consuming experiments or thermodynamic integration/ free energy perturbation/ uBAR calculations, and will hopefully help new drug discovery campaigns in their quest of searching spots of interest for binding small drug-like molecules at P:P interfaces.
了解蛋白质如何形成稳定的复合物有助于开发抑制剂以阻止蛋白质 - 蛋白质（P:P）结合。因此，识别对蛋白质结合起主要作用的特定界面残基（即热点）至关重要。在这里，我们基于分子力学/泊松 - 玻尔兹曼表面积方法的残基依赖性介电常数版本，改进了一种计算机模拟丙氨酸扫描诱变方案。我们使用了大量结构多样的P:P复合物数据集来重新定义用于确定结合自由能的残基依赖性介电常数。通过与实验数据进行比较验证了该方法的准确性，考虑了丙氨酸突变后每个残基的P:P结合自由能（ΔΔG）差异。测试了不同的方案，即几何优化方案和三种分子动力学（MD）方案：（1）一种使用显式水分子，（2）另一种使用隐式溶剂化模型，（3）第三种我们使用显式水分子进行加速MD。使用一组蛋白质介电常数（范围从1到20），我们表明非极性和极性残基的介电常数为7，带电残基（和组氨酸）的介电常数为11时可提供最佳的ΔΔG预测。仅通过几何优化在210次突变中相对于实验实现了1.4 kcal mol的总体平均无符号误差（MUE），通过MD模拟进一步降低（使用显式溶剂的MD的MUE为1.1 kcal mol）。这种重新校准的方法能够更好地通过计算识别热点，避免昂贵且耗时的实验或热力学积分/自由能扰动/uBAR计算，并有望帮助新药发现活动寻找在P:P界面结合类药物小分子的感兴趣位点。
