One approach to drug design is the screening of libraries of compounds for a desired effect on a target (i.e. inhibition of an enzyme) using high throughput screening assays. Large numbers of compounds are screened rapidly with the expectation that this process will identify a high-affinity ligand for the target. More often than not, these screenings produce leads of low to moderate binding affinity which require further refinement to obtain a high-affinity lead. Shuker et al. [1] have reported a nuclear magnetic resonance (NMR)-based method to guide the synthesis of high-affinity inhibitors by linking two or more low-affinity inhibitors which bind in proximal sites of an enzyme. This methodology assumes that the free energy of binding can be approximated as an additive function; the free energy of this linked inhibitor is

* To whom correspondence should be addressed.

approximately the sum of the free energies of binding of each initial molecule and the linker groups. Hence linking together a micromolar inhibitor with a millimolar inhibitor could possibly create a nanomolar inhibitor. In this NMR-based methodology, the NMR serves as the high throughput screening assay and determines both the orientation of the ligands in their binding sites and their affinities. Both the orientation and binding affinities of the ligands are crucial pieces of information to design a composite inhibitor from the ligands. The orientations of the ligands within the enzyme and the enzyme structure guide the identity of the linker groups, the appropriate length of the linkers used to join the lead molecules, and the points of attachment to the lead molecules. The affinities of the ligands determine which compound in a series of analogues should be chosen as the optimized lead prior to linking them together so that the tightest-binding inhibitor can be achieved.

Figure 1 is a schematic diagram, which describes this novel experimental approach. A set of small organic compounds is screened and the binding constants of these ligands and the location of their binding sites are determined. A lead molecule is identified within this set, and analogs of this lead are then screened to optimize the binding to this primary site. In the presence of saturating concentrations of the ligand optimized for the primary site, small organic ligands are screened to determine their affinity for a site proximal to the primary site. A lead for this secondary site is identified and analogs are tested to optimize binding to the second site. Once the binding sites and affinities are determined for each small molecule lead, they are linked together to form a composite ligand. This method was successful for designing high-affinity inhibitors for the FK506 binding protein [13, FKBP, and stromelysin [2].

Computational docking methods [3] have been developed to rapidly screen databases and identify putative drug candidates. These screenings usually identify only low- to moderate-binding ligands that require further optimization. Here, we describe a computational docking method, 'virtual NMR screening', that is similar in spirit to an NMR screening approach, but is performed using computer models of binding rather than using NMR. As with the NMR technique, our approach relies on predicting both the structures of the leads and their affinities correctly. Figure 1 shows a schematic of the process, with the computational steps shown in boxes. The first step is to screen for ligands that bind in a primary pocket. Molecules are docked from a database of compounds and scored using an empirical free energy function. The best scoring (lowest energy) small molecule is selected as the initial lead, and can be optimized using several methods. If the empirical free energy function is accurate enough to distinguish between very similar molecules, optimization can be accomplished by docking and scoring molecules

Figure 1. A comparison of the experimental NMR-based approach to ligand design and 'virtual NMR screening'. The computational procedures that can be performed at each step of the process are outlined in boxes.

that closely resemble the primary compound. At this stage, more costly free energy simulations, such as free energy perturbation (FEP) [4], could also be used to optimize the ligand, by 'growing' in or 'removing' functional groups computationally. Once a ligand has been optimized, its coordinates are frozen to its lowest energy docked structure. Next, molecules from a database are docked in the presence of the optimized primary lead. These docked molecules interact with the protein and the primary lead, which allows the molecule docked in the primary site to help guide the positioning of the molecules. The secondary lead is then optimized by docking and scoring or using more elaborate free energy techniques. Once the positions and the affinities of the primary and secondary compounds have been determined, visual examination of the proximity of the compounds to one another can be used to estimate the length and type of linkers needed to connect them. Several linker compositions and lengths can be tested by docking the composite ligands that are generated from the two low-affinity leads. The composite ligands can then be ranked and their synthesis prioritized.

Many de novo computational drug-design methods rely on the docking or the joining of fragments or small molecules to create high-affinity ligands [5]. Our computational method differs from other methods; secondary small molecules are docked in the presence of a bound ligand, which influences the docking of the secondary ligands. This approach can reduce intramolecular repulsions when the compounds are assembled into an inhibitor.

Both the experimental NMR-based approach and the computational approach have advantages and disadvantages when compared to one another. The biggest advantage of the 'virtual NMR screening' is that we are not restricted by experimental difficulties, such as proteolysis of proteins, the size of the enzyme, or solubility difficulties that can occur in the actual experiments. We can dock a small molecule with very low solubility, which could become an integral and essential part of a large soluble inhibitor. However, such a compound may not easily be screened experimentally. We can screen compounds from a database much more quickly than an NMR experiment, and can perform multiple runs in parallel on several computer processors. However, 'virtual NMR screening' is prone to problems that are common to many other computer-modeling approaches: the quality of the energy function and adequate sampling of conformations. Given our success here, we have addressed these potential problems in an appropriate manner by using high-quality potential energy functions and sampling many ligand conformations. The major advantage of experimental NMR screening over the present version of its computational counterpart is the ability to measure the structure and affinity of the ligands in the presence of a mobile protein. The fixed protein structure used in computational docking imparts the high speed; only

Figure 2. FKBP inhibitors used in the docking study.

conformations of the ligand are sampled. If a ligand causes major structural changes within a protein upon binding that affect the position of the binding site or the affinity of the ligand, 'virtual NMR screening' will miss these results. Hence, the 'virtual NMR screening' approach is useful for systems where these large structural changes do not occur.

To validate our computational approach, we have tested this procedure on both the FKJBP and stromelysin results described for the experimental NMR-based screening systems [2]. These experimental results provide crucial experimental data for which to validate our computational approach. Not only do they possess data for single compounds bound to enzymes, but they provide results for compounds bound in the presence of saturating concentrations of a primary ligand and for the full-size inhibitors. We present our computational results for FKBP, which examine the binding of compounds F2-F14, shown in Figure 2. The numbering of the compounds is the same as in Shuker et al. [1], which reports the experimental method and results, but they are prefixed with 'F' to indicate these compounds were used in the FKBP study. From Shuker's study, compound F2 binds in the primary site with modest affinity (K = 2 ^M). In the saturating presence of F2, a benzanilide derivative binds weakly to a proximal site. Several benzanilide derivatives (F3-F9) were examined, and compound F9 exhibited the highest affinity in the presence of F2. Thus compounds F10-F14 were synthesized, which linked groups F2 and F9. We also describe the results for stromelysin, which examines the binding of the compounds in Figure 3. The numbering of the compounds is the same as in the experimental paper by Hadjuk et al. [2], but they are prefixed with 'S' to indicate these compounds were involved in the stromelysin study. We present the results of these validation studies and show that the docking approach presented here shows promise as a computational drug discovery tool.

0 0

Post a comment