Three-Dimensional Quantitative Structure-Activity Relationship and Comparative Molecular Field Analysis of Dipeptide Hydroxamic Acid Helicobacter pylori Urease Inhibitors

Data set.A group of 24 dipeptide hydroxamic acid derivatives that were assayed in one laboratory under the same assay conditions was selected for use as the primary set of compounds for which data were obtained. The 50% inhibitory concentrations (IC₅₀s) of the dipeptide derivatives were previously determined by Odake et al. (23), and these data are reported in Table 1. The primary structural variation among these compounds was the amino acid side chain.

Computational approaches. (i) Homology modeling.The amino acid sequence for H. pylori urease was retrieved from SWISS-PROT data bank entry URE2_HELPY (5). The X-ray crystal structure of the urease of K. aerogenes, entry 2KAU (13), was obtained via the Protein Data Bank. The construction of the protein model was based on the homologous structure of the K. aerogenes urease, which was used as a template. Amino acid sequence alignment indicated a 61.4% residue identity between the primary structures of the urease enzymes. The three-dimensional model was constructed by copying aligned coordinates of identical residues, building loops, and structural refinement (10). The protein modeling tools available in the computer software package MOE (2000; Chemical Computing Group Inc. Montreal, Quebec, Canada) were used for protein modeling. The partial equalization of orbital electronegativity (PEOE) (7) charges were calculated, and the force field was set to AMBER94 (28). This was followed by geometric optimization of the protein structure, minimizing the structure to a root mean square deviation (RMSD) gradient of 0.01 kcal/mol · Å.

(ii) Docking.The geometrically optimized protein structure was used as a starting point for docking experiments, and PEOE partial charges were calculated. The acetohydroxamic acid was docked into the active pocket of the protein by using the docking protocol of the MOE software package. Specific information identifying the amino acids involved in the formation of the urease active pocket was obtained from the literature, and this information was used to define the docking box or volume to be explored during docking (6). The MOE software package uses grid-based potential fields to calculate interaction energies between flexible ligands and rigid targets. Twenty-five putative complex geometries were generated and optimized during the docking process by an annealing procedure with Monte Carlo simulation (9); however, the final configuration was selected on the basis of similarity to the crystal structure of acetohydroxamic acid as bound to the K. aerogenes urease (24). Energy minimization of the enzyme-inhibitor complex, which was selected from the docking results, was accomplished by using the energy minimization protocol of MOE with chirality constraints to a gradient of 2.0 kcal/mol · Å. The PEOE charges were calculated prior to minimization of the structure using the AMBER94 force field, so that the PEOE charge calculation is independent of the force field.

(iii) Molecular dynamics.The molecular dynamics study was carried out by using a constant temperature and a constant volume of the proposed enzyme-inhibitor complex obtained from docking studies. The enzyme-inhibitor complex was heated from 1 to 300 K over 10 ps with a time step of 1 fs. Data were collected during the following 100-ps equilibrium phase, while the temperature response was fixed at 25. Snapshots were collected every 1 ps throughout the equilibrium phase. Of the 100 conformations obtained in this manner, the 10 conformations showing the lowest potential energy were selected for further minimization to a gradient of 0.01 kcal/mol · A^o, as described above.

Structure alignment.Each of the dipeptide hydroxamic acids was constructed by using the acetohydroxamic acid-H. pylori urease bound conformation as a template. As expected, at physiological pH the amino groups are ionized and were modeled as such. All structures were subsequently imported to Sybyl software (version 6.7, 1995; Tripos Associates, St. Louis, Mo.), with which Gasteiger-Huckel charges were calculated. The biologically active portions of the structures, as illustrated with asterisks in Fig. 1, were overlaid by using the fit atoms protocol available in Sybyl (version 6.7) software (3, 4). The resulting structure alignment of the compounds (RMSD, 0.077 for the fitted atoms) is shown in Fig. 2.

• Open in new tab
• Download powerpoint

FIG. 1.

Atoms of dipeptide hydroxamic acid used for the structure alignment. Asterisks, biologically active portions of the molecules.

• Open in new tab
• Download powerpoint

FIG. 2.

Structure alignment of 24 dipeptide hydroxamic acids. RMSD, 0.077 Å for the atoms marked with asterisks in Fig. 1.

Generation and analysis of data by CoMFA.The CoMFA tools available in the Sybyl (version 6.7) software were used to calculate steric and electrostatic fields at grid points by using the Lennard-Jones and the Coulomb potential functions of the Tripos Standard field class. An sp³ carbon probe with a charge of +1 was used for the grid calculations. The software program was prompted to automatically generate a single grid, overlapping all entered molecules and extending past them by at least 4 Å along all axes. Steric and electrostatic cutoffs were set to 30 kcal/mol and proceeded with smooth transition, except that the electrostatic contributions were dropped for each row where the steric cutoff was reached. Both steric and electrostatic contributions were calculated by use of a distance-dependent dielectric constant of 4.0 with a smoothing function. The grid spacing was set to 2 Å. Standard CoMFA scaling was applied to give each individual CoMFA field the same potential influence on the resulting QSAR.

Model validation.The partial-least-squares (PLS) method (3) was used to derive a linear relationship between the biological activities and the molecular fields. Leave-one-out cross validation with a 2-kcal/mol column filter was performed by omitting from the analysis those columns (lattice points) with an energy variance less than 2.0 kcal/mol. This allowed determination of the optimum number of components associated with the lowest standard error of prediction.

Four components provided the highest cross-validated r² (q²) with the lowest standard error of prediction. The cross validation was performed by dividing the total number of compounds into 10 groups. Only models with q² values over 0.5 and a fraction of variance greater than 0.85 for the optimum number of components were considered further. The model found to be the best by cross validation was used to perform the nonvalidation run.

Homology modeling.Our homology modeling yielded a Z score of 66.7. This is an estimate of the statistical significance of the alignment score (10). The homology obtained between the sequences was 61.4%. An overlay of the K. aerogenes urease and the H. pylori urease is illustrated in Fig. 3A.

• Open in new tab
• Download powerpoint

FIG. 3.

(A) Overlay of the urease of K. aerogenes (white; entry 2KAU in the Protein Data Bank) with the homology model H. pylori urease (purple) (Z score, 61.7%); (B) comparison of active pockets (green, H. pylori urease; white, K. aerogenese urease).

Docking and molecular dynamics.Docking, minimization, and molecular dynamics produce an orientation of acetohydroxamic acid in the H. pylori urease model that is very similar to that in the K. aerogenes urease crystal structure. The hydroxamic acid binds to the nickel ions and is very important for activity (6). The overlap of the modeled H. pylori urease-acetohydroxamic acid on the K. aerogenes urease complex obtained from the crystallographic data is shown in Fig. 3B.

CoMFA.The results of our CoMFAs by PLS analysis are as follows: q² was 0.610, r² was 0.988, the standard error of prediction was 0.131, and F (the Fisher indication of statistical significance) was 294.55 in tests with four components. With the 24 dipeptides, a q²_L00 of 0.609 was obtained with the single optimal principal component. The correlation coefficient obtained by the nonvalidation analysis was 0.988, and the q² found was 0.610. It is widely accepted that a correlation with a q² value greater than 0.5 to 0.6 is useful for the prediction of new biologically active molecules (2). The predicted and actual IC₅₀s of the compounds in the test set are provided in Table 2.

CoMFA fields.In order to visualize the CoMFA results, the CoMFA contour maps were created by using the data from PLS analysis. These maps help to explain the steric and electrostatic features of the compounds included in our analysis. The electrostatic contour maps are shown as red and blue polyhedra in Fig. 4A, with the red indicating regions in which the electronegative groups in the ligands are associated with increased biological activities. These can be envisioned as electropositive groups within the active pocket of the receptor. In contrast, the blue contours can be visualized as the electronegative groups in the active pocket and indicate where positively charged groups in the ligand have improved biological activity and thus lower the IC₅₀s. The steric contour maps (Fig. 4B) also illustrate the areas in which steric bulk on an inhibitor is favored (green contours). The yellow polyhedra represent regions in which steric bulk is detrimental to binding and hence increases the IC₅₀s by decreasing the overall biological activity. The green polyhedra may be envisioned as hydrophobic cavities in the receptor, which can accommodate hydrophobic groups. In contrast, the yellow polyhedra represent areas already occupied by the receptor, which would thus prohibit any effective binding.

• Open in new tab
• Download powerpoint

FIG. 4.

(A and B) electrostatic and steric maps derived by CoMFA overlaid on the H. pylori urease active pocket (acetohydroxamic acid is shown in orange). (A) Blue contours indicate where electropositive groups improve activity, and red contours show where electron-rich groups increase activity. (B) Yellow contours show the sterically disfavored areas of the pocket, and green contours are areas where steric bulk is predicted to have a favorable impact on the biological activity.

After the development of this enzyme-acetohydroxamic acid complex of the H. pylori urease active pocket, we compared the maps prepared by use of the coefficients from PLS analysis of data obtained by CoMFA with the geometric and chemical properties of the binding site. Since CoMFA studies are based solely on a set of active compounds, these maps cannot provide a picture of the binding pocket. Nevertheless, CoMFA does reflect the regions in space where differences in the ligand-probe interaction energy can be correlated with the variance associated with biological activity. Therefore, CoMFA data can be extremely useful in the design of new chemical moieties based on a prediction of their biological activities (2). Figure 4 illustrates the contour maps prepared by use of the coefficients from PLS analysis, in which the contour maps have been overlapped on the active pocket of the H. pylori urease. Figures 4A and B indicate that there exists considerable agreement between the contour maps and the observed positions of the amino acids. Interestingly, a red negative field was observed near the nickel atoms; this is an indication that only electronegative groups that can interact favorably with the positive charge on the nickel ions will be capable of effectively binding to the active pocket. The presence of a large positive field occupying the Asp223, Glu222, and His221 residues, as indicated by a blue contour, demonstrates that a requirement for a positive charge is most consistent with the increased biological activities. We believe that this explains why the NH₂ is crucial for biological activity and why either blockage or removal of this group would result in compounds with increased IC₅₀s. The presence of Asp223 in the active pocket also supports our decision to model the NH₂ group in its ionized state.

In Fig. 4B, the presence of yellow polyhedra surrounding the NH₂ group indicates an area of the pocket which is sterically hindered and which most likely prohibits any effective binding where large groups are present on the ligand. The green contour covering the hydrophobic pocket is created by amino acid residues Met366, Met317, and Ala365; and the presence of these residues helps explain why ligands with polar groups that fit this position correlate with improved biological activities. Indeed, the most active ligands possess hydrophobic substituents that interact in this position. A similar interpretation was obtained from previous QSAR studies with hydroxamic acid inhibitors of urease (21).

Our model, along with previously reported studies of urease inhibitors, indicates that only the inclusion of electronegative moieties like hydroxamic acids, thiols, and phosphoramides that can efficiently complex with nickel ions likely results in biologically active compounds (25). In addition, in illustrating the requirement for an ionizable NH₂ group, our model provides essential information for the design of novel H. pylori urease inhibitors. Finally, our model identifies a hydrophobic region whose identification should provide a significant advantage in the design of inhibitors.

Overall, we have developed a robust three-dimensional QSAR model that not only explains the variance in biological activities of a set of specific dipeptide hydroxamic inhibitors of H. pylori urease but also correlates with the homology model of this urease. In addition, we have incorporated active site-based information in our CoMFA studies to aid in the prediction and design of structural properties for novel hydroxamic acid molecules. We are synthesizing several hydroxamic acid derivatives to be evaluated against H. pylori urease in vitro and to help further refine our model.