Predicting nonspecific ion binding using DelPhi

Abstract

Ions are an important component of the cell and affect the corresponding biological macromolecules either via direct binding or as a screening ion cloud. Although some ion binding is highly specific and frequently associated with the function of the macromolecule, other ions bind to the protein surface nonspecifically, presumably because the electrostatic attraction is strong enough to immobilize them. Here, we test such a scenario and demonstrate that experimentally identified surface-bound ions are located at a potential that facilitates binding, which indicates that the major driving force is the electrostatics. Without taking into consideration geometrical factors and structural fluctuations, we show that ions tend to be bound onto the protein surface at positions with strong potential but with polarity opposite to that of the ion. This observation is used to develop a method that uses a DelPhi-calculated potential map in conjunction with an in-house-developed clustering algorithm to predict nonspecific ion-binding sites. Although this approach distinguishes only the polarity of the ions, and not their chemical nature, it can predict nonspecific binding of positively or negatively charged ions with acceptable accuracy. One can use the predictions in the Poisson-Boltzmann approach by placing explicit ions in the predicted positions, which in turn will reduce the magnitude of the local potential and extend the limits of the Poisson-Boltzmann equation. In addition, one can use this approach to place the desired number of ions before conducting molecular-dynamics simulations to neutralize the net charge of the protein, because it was shown to perform better than standard screened Coulomb canned routines, or to predict ion-binding sites in proteins. This latter is especially true for proteins that are involved in ion transport, because such ions are loosely bound and very difficult to detect experimentally.

Figure 1

Schematic representation of the clustering algorithm. The left upper panel shows a protein mapped onto a grid. A small region (shown with dashed square) is zoomed and shown in panel A. Large circles symbolize the border of clusters, small open circles represent all points in a cluster, and solid dark circles represent points with the highest absolute potential. In panel A the radius of each cluster was 5 Å (the dashed line shows a cutoff distance of 5 Å away from the protein surface). (B) A more rigorous condition for cluster determination was applied (radius of clustering = 10 Å, and distance between the geometric average of all points in the cluster and farthest to its point in the cluster ≤5 Å. (C) The final step of clustering is to search for all resultant points <5 Å from each other and leave only those with greater absolute potential.

Figure 2

Distribution of the electrostatic potential at experimental ion positions grouped with respect to ion type.

Figure 3

Distribution of all representative grid points found by the clustering method with Rank = 1 (dark bars) and the Rank of the closest representative grid point with respect to the original ion's position (light bars).

Figure 4

ROC curves for Ca, Mg, Zn, and Cl ions containing proteins data set, calculated with respect to different parameters. The first number corresponds to the dielectric constant of the solution, and the second one corresponds to the ionic strength in moles/l. The x axis represents the Rank of the closest representative grid point to the experimentally determined ion position. The y axis is the number of successful predictions (true predictions) in percentage of all predictions. A prediction is considered to be true if the representative point situated at the shortest distance from the ion experimental position (Dmin) is predicted.