Sie sind auf Seite 1von 8

Current Pharmaceutical Design, 2002, 8, 1571-1578

1571

Protein Flexibility is an Important Component of Structure-Based Drug Discovery


Heather A. Carlson*
Department of Medicinal Chemistry, College of Pharmacy, University of Michigan, 428 Church St., Ann Arbor, MI 48109-1065, USA
Abstract: Receptor-based drug discovery can increase the novelty of a hit list over ligandbased models that are dependent on known inhibitors. It is important to explore new conformational and chemical space, but it is difficult to predict the plasticity of the binding site. Receptor-based methods are usually based on crystal structures of ligand-protein complexes, and hit lists can be restricted to the size and shape of the receptor model. Many improvements that accommodate protein flexibility in computer-aided drug design are being developed. These methods are reviewed with the focus being techniques that move beyond the rotation of side chains.The use of multiple protein structures is emerging as the best choice for including more realistic changes in protein conformation, but the optimal way to using these structures is still unclear.

INTRODUCTION Only recently has the computer power become available to address the issue of protein flexibility in drug discovery, but these same advances are also allowing large increases in the number of compounds available in chemical databases [1-3]. The Available Chemical Directory (ACD) has well over 250,000 compounds, and it is straightforward to create new combinatorial libraries of that same size. Existing algorithms require ~100 seconds to dock a flexible ligand to the surface of a rigid protein [4], so screening a library like the ACD could take 41 weeks. Adding protein flexibility makes the task even slower; some docking techniques require several minutes per ligand. The calculations are much faster if receptor-based pharmacophore models are used. These models are simplified maps that describe the spatial arrangement of functional groups that complement a receptor. They are used to generate inhibitors from database mining or de novo design [5]. Fitting a flexible molecule into these simple maps is faster than docking to a protein surface. By incorporating both protein flexibility and speed into the calculations, improved pharmacophore models can reduce two sizable limitations in structure-based drug discovery (SBDD). Many proteins are flexible and cannot be adequately described with a single, rigid structure. Unfortunately, this is the standard protocol when using SBDD for the design of complementary inhibitors [6-8]. This works well when exploring a bound crystal structure to identify small modifications that improve the specificity and affinity of the

inhibitor, but predicting induced fit and the plasticity of a target receptor is highly speculative. Studies have appeared demonstrating the need for protein flexibility in ligand docking in order to achieve proper results [1,9,10]. It is important to note that the binding of a ligand within an enclosed binding site inherently requires that part of the receptor be flexible in order to allow access. Consequently, it has been observed that many active sites contain large regions of low structural stability [11,12]. Furthermore, membrane-bound receptors are very flexible [13], and their conformational changes related to binding agonists versus antagonists are the subject of intense crystallographic and SAR studies. As the body of structural data grows for these important pharmaceutical targets, it will be essential to use more accurate methods of modeling that incorporate protein flexibility.

STUDIES THAT ILLUSTRATE THE NEED TO INCLUDE PROTEIN FLEXIBILITY In terms of SBDD, dihydrofolate reductase (DHFR) and HIV-1 protease (HIVp) are possibly the two best-studied protein systems [14]. It is rather remarkable that these two canonical examples used to develop many of the SBDD methods that employ one rigid protein conformation are both poorly described by the use of a single structure. Kinetic studies and NMR studies of DHFR have shown that the protein exists in two conformational states when unbound [15,16], and undergoes ligand-induced conformational changes [17]. Computer simulations by Radkiewicz and Brooks have established that protein flexibility and dynamics are essential to ligand recognition, catalysis, and product release in DHFR [18]. Pan et al. have further studied the bound and unbound forms of DHFR, using an ensemble-based approach to their analysis of several crystal structures [19]. A fascinating cooperativity is seen for perturbations to the structure whether those perturbations
2002 Bentham Science Publishers Ltd.

*Address correspondence to this author at the Department of Medicinal Chemistry, College of Pharmacy, University of Michigan, 428 Church St., Ann Arbor, MI 48109-1065, USA; Tel: (734) 615-6841; Fax: (734) 7632022; E-mail: carlsonh@umich.edu 1381-6128/02 $35.00+.00

1572

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

Heather A. Carlson

are small (such as point mutations) or large (ligand binding). Through energetic couplings between different regions of the protein, these perturbations can shift the weighting within the ensemble of conformational states. The energetic couplings can even be traced to residues with no visible connectivity pathway, which undermines the mechanical view that conformational changes arise from propagating distortions through a series of internal coordinates. The flaps region of HIVp is known to close upon binding inhibitors, forming a bioactive conformation [20], and the reduced binding affinity observed for escape mutants of HIVp has been linked to changes in the flexibility of the system [21]. Using HIVp as a test case, Bouzida et al. docked two flexible ligands, sb203386 and skf107457, into 10 rigid protein conformations obtained from crystal structures of HIVp-inhibitor complexes [22]. Out of 10 conformations of the protein, sb203386 was only identified using its own crystal structure. Only four of the 10 protein conformations allowed skf107457 to be docked within 2.0 RMSD of its position in the crystal structure. The majority of the protein structures would not have identified either inhibitor. This illustrates the inherent bias of using only a single, rigid protein structure. The importance of protein flexibility and induced fit is not limited to DHFR and HIVp. Murray et al. have presented a study of the effects of induced fit for thrombin, thermolysin, and neuraminidase [23]. Specifically, the authors were seeking to establish the extent to which a rigid receptor can compromise the accuracy of docking flexible ligands. Six crystal structures of ligand-protein complexes were available for thrombin and neuraminidase, and nine were used for thermolysin. In 76% of the cases, a ligand could be correctly docked when using the protein conformation derived from the crystal structure of that ligand bound to the target protein. Only 49% of the cases were successful when docking a ligand against a structure of the protein bound to a different inhibitor. The most notable problems were related to the motion of the backbone. Any recognition via coordination to the backbone was altered, but also, the backbone motion effected the orientation of the side chains. The authors note that treatments of protein flexibility that focus only on side chain rotations will be inadequate for the three test systems. Brevity limits the number of protein systems that can be discussed, but the point is clear. Methods for SBDD are needed that include flexibility of the target receptor. Furthermore, these methods must move beyond the technique of side-chain rotations to include the motion of the backbone, loops, and domains.

GENERATING A COMPLEMENTARY MAP OF THE RECEPTOR SBDD focuses on the receptor and designing inhibitors to complement the surface and shape of the binding site [24]. Several techniques can be used to determine the key features of a potential inhibitor, see Fig. (1).

Fig. (1). Given the structure of a binding site (solid black), several techniques are available for describing the complementary space. (A) Pre-calculated grids sum the interaction with all atoms of the target protein. (B) Many probe molecules can be energy minimized to map the most favorable positions for certain functional groups to complement the surface of the receptor. Different probes (gray versus black) map out different types of interactions. (C) Template points are often assigned with geometry-based rules. Different interactions (hydrophobic, hydrogen-bond donor, etc.) are described with a variety of points.

Protein Flexibility is an Important Component

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

1573

Standard molecular mechanics uses two-body potentials to evaluate the interaction between the potential inhibitor and the biopolymer. This can lead to slow calculations where thousands of atom-atom potentials must be summed for each potential orientation of the ligands. This practice was revolutionized in 1985 when Goodford introduced GRID [25]. As the name implies and as shown in Fig. (1a), the technique uses a pre-calculated grid to represent the interaction with the receptor, eliminating the need to sum over all the atoms of the biopolymer. Grids are employed in many popular docking programs such as DOCK [26] and Autodock [27]. Multiple copy minimizations, such as Multiple Copy Simultaneous Search (MCSS) [28] and Multi-Unit Search for Interacting Conformers (MUSIC) [29], can be used to dock small molecule probes to the surface of a receptor, see Fig. (1b). These methods use two-body potentials, but only the probe-receptor interaction is calculated while the probeprobe interaction is ignored. This allows the probes to overlap and cluster, defining optimal positions for various functional groups. Location, orientation, variability (spread of probes in a cluster), and favorability (interaction energy) are important factors to consider when using this information in SBDD. Multiple copy methods have a long history of success [30-32]. One consideration when using small molecule probes is how well the appropriate chemical space can be covered. Recently, Bemis and Murko [33] have surveyed the Comprehensive Medicinal Chemistry (CMC) database to identify the most common side chains of drug molecules. It was determined that most chains are rather small, with very few having more than five non-hydrogen atoms. The majority of the 46 most common side chains can be represented by, or built from, the following 15 small molecule probes: CH3COCH3, CH3CH3, CH3OH, CH3OCH3, CH3Cl, CH3NH2, CH3NH3+ , CH3CO2- , CH3COOCH3, CH3NO2, CH3CONH2, CH3SO2NH2, CH3CN, CH3SO2OH, CH3SO3- . Benzene and other rings were not considered side chains in the analysis of the CMC, but they are important components of many inhibitors. It is unlikely that all probes would be necessary for most systems, but they would be very useful for de novo design. The most common probes will likely be ethane, benzene, methanol, methylammonium ions, acetone, acetate ions, and acetamide. The positions found with these probes can indicate the need to use other probes listed above (for example, relative positions of acetone and methanol probes in a receptor could suggest using methyl acetate probes). Another common method to determine the complementary interactions necessary to interact with a receptor is the use of template points [34,35] or pharmacophore models, see Fig. (1c). These can be derived from geometric rules or multiple copy minimizations [29,36]. Rather than calculating the interaction energy as with grids and minimizations, this last method is faster. Inhibitors are evaluated based on whether each can adopt a conformation where the appropriate functional groups can be arranged into the proper geometry described by the model. The gain in speed is at the expense of information that is more complete. It is possible that pharmacophore models

will be more limited than grids because of the loss of information about the receptor, located between the sites of the pharmacophore model.

INCORPORATING PROTEIN FLEXIBILITY BY COMBINING MULTIPLE PROTEIN STRUCTURES The earliest methods for accommodating small changes in the receptor were soft-docking techniques. This allows for a small degree of overlap between the potential ligand and the walls of the cavity. This still restricts the successful ligands so that they frequently will be limited to a certain size and conformation, but soft-docking has the benefit of being computationally efficient and easy to implement. The first attempt to explicitly include protein flexibility made use of a library of discrete rotameric states for each type of side chain [37]. Since then, improved techniques have been proposed that optimize the orientation of the side chains [3841]. However, minimizing the energy of the side chains in the gas phase can lead to inappropriate conformations that would not be observed in the presence of solvent, and focusing on the side chains neglects any possible changes in the backbone of the protein. As previously mentioned, the most modern techniques for incorporating protein flexibility have to move beyond simple side-chain reorientation. The use of multiple protein structures (MPS), an ensemble of conformational states, is emerging as the best option to take advantage of the full flexibility of the receptor [1]. At this point, the difficult task is determining how to combine the MPS in the most useful way. Many questions remain unanswered. What is the best source of MPS: crystal structures, NMR structures, or calculations? How many structures are needed? How do we know if we have included enough flexibility? How much is too much? Verkhivker et al. used a Monte Carlo method for simulated annealing, applied to 50 copies of an HIVp complex [42]. With repeated simulations at temperatures ranging 300 to 5000 K, they surprisingly found that the system could become trapped in local minima even at high temperatures. Despite that limitation, they found that using the large number of conformations, from a wide range of temperatures, produced more accurate predictions of the protein-ligand complex. Philippopoulos and Lim found that a set of 15 NMR structures of E. coli ribonuclease HI sampled more conformational space than a molecular dynamics (MD) simulation [43]. They found that the NMR and MD were sampling similar structures, but NMR sampled a wider range of the conformational well, showing more flexibility in both the side chains and backbone atoms. The thermal factors of two crystal structures of the system were correlated with the conformational sampling in the MD and NMR structures. In general, multiple crystal structures will cover more conformational space than is sampled in an MD simulation [44], but whether NMR or crystal structures sample more conformational space is most likely dependent on the system. Perhaps the most important question is how to combine the information from MPS. This is dependent on which

1574

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

Heather A. Carlson

Fig. (2). Two basic techniques for combining MPS. The first column shows the combination of the structural data. This composite is then used to generate a complementary map of the receptor using grids, probes, or template points. In the second, third, and fourth column, the complementary maps are calculated for each conformation and then combined (protein surface is shown in gray at the bottom to show that the emphasis is on the grids, probes, or points).

method is used for generating the complementary map of the receptor. The methods for combining grids are not the same as combining probes or template points. Fig. (2) shows schematically four possible means of generating a flexible model of the receptor from MPS. The techniques can be broken down into two categories: combining the protein structures as seen in the first column of Fig. (2) or combining the complementary maps as seen in the second, third, and fourth columns of Fig. (2). Both techniques are used in the examples described above. An important consideration when combining the protein structures is the dual character seen for the binding pockets of many proteins; they are part rigid and part plastic. Freire and coworkers [11,12] have found that binding sites are composed of regions of very high structural stability and regions with very low stability. Sinha and Nussinov [21] have found that the rigid and flexible regions of a protein remain the same regardless of the perturbation to the system; point mutations at many different locations lead to the same conformational changes. Therefore, it is fitting to combine protein structures via the most structurally similar regions. By using MPS, a weighted average of the proteins dynamic behavior is obtained. As Fig. (3) shows, the maximum and minimum dimension of the site could be identified, but this

simple scheme ignores the possibility of cooperative motion. Treatment of the flexible regions would depend upon the application and the desired breadth of the hit list.

Fig. (3). A schematic representation of combining three receptor structures (as seen in the lower left of Fig. (2). The regions of consensus are in black, and the areas where the structures are different are highlighted with cross-hash marks. Lighter regions (more hash marks) have less agreement and are flexible. The minimum and maximum size of the binding pocket can be inferred from including or excluding the flexible regions, respectively.

Protein Flexibility is an Important Component

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

1575

An interesting method recently described with the introduction of FlexE [45] creates an average structure for parts of the protein that are in good agreement (the backbone and some side chains in the test systems), but treats the flexible side chains like a rotamer library, see Fig. (4). During docking, different combinations of the flexible side chains are chosen, mixing between the different crystal structures to create new conformations of the binding site. It is possible that backbone and loop conformations could be treated in a similar fashion in regions of low structural stability. This is an excellent treatment for regions where the flexibility is too large for simple averaging.

examined. The first was energy-weighted averaging of the grids; this method smoothed the large repulsive terms by using only the attractive components of the potential in problematic regions like the black-hashed regions in Fig (3). The second method was geometry-weighted averaging. In this treatment, the average positions of the atoms of the receptor were used to construct a grid, as in Combine Conformations in Fig. (2). For regions were the variance was large the same regions where FlexE uses rotameric libraries the contributions were calculated for each position, and the concept of fractional occupancy was used for the averaging onto the grid. Both methods performed equally well for the chosen test systems and were an incredible improvement over using a single conformation of the receptor. A recent study employing Autodock carefully examined four choices for combining grids [47]. A grid of mean values was shown to be too restrictive because of the dominance of the repulsive terms (as discussed above). A minimum energy grid was generated by choosing the lowest and most favorable values at each point on the grid; this is similar to a grid described by the maximum binding site in Fig. (3). The minimum grid overestimated the favorable regions, and one would expect that the size of the binding site would be overestimated in some cases. Lastly, two methods for creating a weighted average of the grids were examined. Again, both weight-averaged grids performed well. Combining probes and template points can range from retaining only the regions of consensus to using all possible combinations of the points. Carlson, McCammon, and coworkers [29,48] presented a method using probe molecules from MPS to determine pharmacophore models for database searching. The structures were overlaid with respect to two essential residues (catalytic residues are usually, but not always, located in regions with high structural stability [12]). Regions of consensus were identified where probes were conserved over many of the conformations of the receptor. This focus helps to identify inhibitors that must complement the most consistent (rigid and less forgiving) features of the binding site. The positions of the probes in the conserved sites were averaged, and no limitations were placed on the flexible regions. Such models have the potential to identify inhibitors that bind to many conformations of the protein, even conformations that require significant changes in the flexible regions. Using consensus regions also reduced the pharmacophore model to a reasonable number of sites (5 or 6 for the test system). Kuhn and coworkers [35,38] have developed SLIDE based on template points. The algorithm is used to sort through the many combinations of template points possible for the flexible receptor. A multilevel hash-table scheme is used to save time and make the search of so many combinations possible. Searches with SLIDE are capable of docking a ligand within seconds. The issue of bridging water molecules within the receptor has not been addressed. These positions can be calculated through geometric rules [49,50] or taken from crystal structures or MD simulations. It may be appropriate to include some water molecules as part of the receptor itself when docking small molecule probes, numerating template

Fig. (4). A schematic representation of the technique employed in FlexE [45]. Regions of the protein that are in good agreement are averaged, and the orientations of the flexible side chains (A and B) are retained as a set, much like a rotameric library. The method allows for mixing between the MPS to create new conformations of the receptor (e.g., the white orientation for A and a gray or black orientation for B).

In fact, simple averaging can be misleading. Examine the grids from the second column of Fig. (2). If these were averaged, the large repulsive energy associated with steric clashes with the binding site (all gray regions in Combine Grids) would heavily bias the mean. The resulting available space would be restricted to the region in white, greatly reducing the size of the binding site and the conformational space accessible to the ligand. The first example of using MPS employed the program DOCK and was based on grids [46]. The authors recognized the need for careful consideration in combining the grids. Two methods were

1576

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

Heather A. Carlson

points, or generating grids. Usually, there are many possible positions for water molecules, and their addition greatly increases the number of possible conformations of the receptor. At this time, SLIDE is capable of incorporating flexible, explicit water molecules [49,51], but no systematic study has been done regarding water molecules and multiple protein structures.

integrase from the ACD. To date, this is the only use of MPS methods that has been verified by experimental testing. In 1999, Stultz and Karplus investigated the use of fully flexible proteins with MCSS methods for probe molecules [54]. They note that complementary maps of the binding site of HIVp changed when flexibility was introduced. Some of the most interesting advancements to incorporating protein flexibility have been made by Kuhn and coworkers. In 2000, they introduced SLIDE, an improved database-screening tool [51]. SLIDE uses template points in docking. The template points are generated with geometric rules for hydrophobic regions, hydrogen-bond donors, and hydrogen-bond acceptors. Their most recent improvement to SLIDE has been the addition of backbone flexibility [35]. Flexible regions of the backbone are identified with a graph theory technique (the FIRST algorithm), and sets of alternate positions for the backbone are generated through random sampling. The backbone motion is limited to a reasonable range by the local van der Waals interactions and the hydrogen-bonding network. SLIDE has been used to search the Cambridge Structural Database (CSD) and the National Cancer Institute (NCI) database to identify potential new inhibitors of the human progesterone and estrogen receptors, glutathione transferase, HIVp, DHFR, and human uracil-DNA glycosylase [51]. Also in 2000, Broughton introduced a technique using MD simulations to generate many conformations of proteininhibitor complexes [55], as opposed to the use of an unbound protein as in studies by Carlson et al. [29]. A composite grid was created by overlaying the structures with respect to the bound inhibitors. A weighted-average of the individual grids from each of the protein conformers provided the composite grid used in docking studies. This method showed significant improvement over using a single crystal structure to evaluate known ligands of DHFR. The GRID/CPCA method was also introduced in 2000 [56]. Consensus principle component analysis (CPCA) is used to analyze the grids of multiple protein structures. An interesting twist to this study is the application to several crystal structures for each member of a family of serine proteases (thrombin, trypsin, and factor Xa). The consensus regions were used to analyze for similarities, but more importantly, the differences were related to the structural aspects that define the specificity for the systems. Lamb et al. have examined a similar family of serine proteases (trypsin, chymotrypsin, and elastase), specifically addressing the need to create combinatorial libraries specific to protein families [4]. Their techniques for library design should complement the CPCA work. Additional studies by Chen and Zhi promote the use of ligand-protein inverse docking where a single ligand is docked against many proteins could also work well with the design of family-specific libraries [57]. In 2001, Clauen et al. introduced FlexE [45], a new version of their docking software FlexX. FlexX is capable of handling ligand flexibility in docking [58], and FlexE introduces protein flexibility by employing MPS from

METHODS THAT INCLUDE MPS IN SBDD These studies have only appeared in the past few years, and they are presented below based on the year that they appeared in the literature. Given that the use of rotameric libraries for side chains was first attempted in 1994 [37], it is impressive that the first treatment of protein flexibility with ensemble-based methods appeared only three years later. The first MPS method was presented in 1997 [46] to be used with the original ligand-macromolecule docking program, DOCK [26,52]. They used composite grids to incorporate multiple crystal or NMR solution structures of protein-ligand complexes. That study focused on four protein systems: HIVp (five crystal structures of inhibitor complexes), ras p21 (five bound crystal structures), uteroglobin complexed with a polychlorobiphenyl metabolite (25 NMR structures), and bovine retinol binding protein (five co-crystals). Protein conformations were superimposed with respect to residues near the binding site. Composite grids improved the accuracy of identifying known inhibitors and, surprisingly, increased the speed of the docking simulations used in database searching. In 1998, Sudbeck et al. proposed a simple method to qualitatively incorporate multiple crystal structures [53]. They overlaid the structures of nine inhibitor complexes of HIV reverse transcriptase (using the C of the most stable region, residues 97-213). This mapped the area encompassing all of the bound inhibitors and described most of the available space in the receptor. The composite picture of the receptor was not used for docking small molecules, but it provided a reference point for qualitatively evaluating the predicted conformations of small molecules docked to a single, rigid structure of the reverse transcriptase. In 1999, Carlson et al. presented a dynamic pharmacophore model of HIV integrase based on MPS taken from an MD simulation [29,48]. Small molecule probes were docked to each protein structure. The maps were overlaid with respect to the essential residues to identify regions where probes were conserved over many of the conformations of the receptor. The MPS model was based on those conserved interactions (focusing on the conserved regions achieves a similar goal as weight-averaged grids). A database search using that model correctly identified many of the most active, known inhibitors of the integrase. A static pharmacophore model, developed in the standard fashion using a single, rigid crystal structure, was unable to identify any of the known inhibitors. The dynamic pharmacophore model based on MPS was used to screen the ACD for potential inhibitors, and subsequent experimental testing verified the prediction of several new inhibitors of HIV

Protein Flexibility is an Important Component

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

1577

crystallography, see Fig. (4). In combining the MPS, the parts of the structure that are in good agreement are averaged, and the many orientations in the flexible regions are retained as a set, much like a rotamer library. The method was tested with 10 sets of proteins: aldose reductase, alphamomorcharin, carboanhydrase II, carboxypeptidase, DHFR, isocitrate dehydrogenase, mandelate racemase, ricin, seryl tRNA synthase, and trypsin. Each set contained seven to 16 crystal structures, totaling 105 structures in all. Using FlexE, 83% of the ligands were placed within 2.0 RMSD of their crystallographic position. FlexE was faster than a series of individual FlexX calculations with each protein structure of the set, but the average time to dock a ligand with FlexE was still 5.5 minutes. In 2001, Anderson et al. combined two structures of thymidylate synthase to identify regions of flexibility, called soft spots [59]. An algorithm was presented for identifying these regions from a single protein structure, and a procedure called PLASTIC was used to provide a collection of possible conformations of the protein target, based on rotameric libraries. The study shows good improvement over the use of a single, rigid protein conformation. Lastly, the most recent report of a study using MPS was made by sterberg et al. in 2002 [47]. The study was based on 21 crystal structures of HIVp complexes. The ligands were all peptidomimetic inhibitors. Autodock was used in the study, and the authors systematically compared four choices for combining grids. To set a base line for comparison to the performance of using a single protein structure, all 21 inhibitors were docked to all 21 (rigid) protein structures. It was found that all but one of the inhibitors were capable of docking in good agreement with their crystallographic coordinates when using the protein derived from that inhibitors crystal structure. For the cases where inhibitors were bound to protein conformations from crystal structures determined with other bound ligands, 13 of the 21 inhibitors were relatively insensitive to which protein conformation was chosen; all protein conformations yielded good docked structures. However, eight of the 21 inhibitors were problematic for most of the protein conformations. The four methods for combining grids were discussed at length above. When docking the 21 inhibitors to the combined grids, a grid of mean values was shown to be too restrictive because of the dominance of the repulsive terms, and the minimum grid overestimated the favorable regions. The two methods for producing weight-averaged grids performed well for every one of the 21 inhibitors.

ABBREVIATIONS ACD CMC CPCA CSD DHFR HIVp MCSS MD MPS MUSIC NCI NMR RMSD SAR SBDD = Available Chemical Directory = Comprehensive Medicinal Chemistry = Consensus principle component analysis = Cambridge Structural Database = Dihydrofolate reductase = HIV-1 protease = Multiple copy simultaneous search = Molecular dynamics = Multiple protein structures = Multi-unit search for interacting conformers = National Cancer Institute = Nuclear magnetic resonance = Root mean squared deviation = Structure-activity relationship = Structure-based drug discovery

REFERENCES
[1] [2] [3] [4] Carlson, H.A.; McCammon, J.A. Mol. Pharmacol. 2000, 57 , 213. Bhm, H.-J.; Stahl, M. Curr. Opin. Chem. Biol. 2000, 4, 283. Dean, P.M.; Zanders, E.D.; Bailey, D.S. TRENDS in Biotech. 2001, 19 , 288. Lamb, M.L.; Burdick, K.W.; Toba, S.; Young, M.M.; Skillman, A.G.; Zou, X.; Arnold, J.R.; Kuntz, I.D. Proteins 2001, 42 , 296. Greenidge, P.A.; Carlsson, B.; Bladh, L.-G.; Gillner, M. J. Med. Chem. 1998, 41 , 2503. Bhm, H.J.; Klebe, G. Angew. Chem. Int. Ed. Engl. 1996, 35 , 2588. Zheng, Q.; Kyle, D.J. Drug Discovery Today 1997, 2, 229. Walters, W.P.; Stahl, M.T.; Murcko, M.A. Drug Discovery Today 1998, 3, 160. Zheng, Q.; Kyle, D.J. Proteins 1994, 19 , 324. Sandak, B.; Wolfson, H.J.; Nussinov, R. Proteins 1998, 32 , 159. Freire, E. Proc. Natl. Acad. Sci. USA 1999, 96 , 10118. Luque, I; Freire, E. Proteins 2000, S4 , 63. Mosberg, H.I. Biopolymers 1999, 51 , 426. Klebe, G. J. Mol. Med. 2000, 78 , 269.

[5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

ACKNOWLEDGMENTS Many of the research groups cited in this minireview have numerous publications that involve computational modeling of molecular recognition processes; those more recent and relevant to protein flexibility and drug discovery through ligand-docking and pharmacophore modeling have been cited. Any oversight of relevant papers the literature is unintentional. This work is supported in part by the National Institutes of Health (GM 65378) and the Backman Young Investigator Program.

1578

Current Pharmaceutical Design, 2002, Vol. 8, No. 17

Heather A. Carlson

[15] [16] [17] [18] [19] [20] [21] [22]

Schweitzer, B.I.; Dicker, A.P.; Bertino, J.R. FASEB J. 1990, 4, 2441. Feeney, J. Angew. Chem. Int. Ed. Engl. 2000, 39 , 290. Cody, V.; Galitsky, N.; Rak, D.; Luft, J.R.; Pangborn, W.; Queener, S.F. Biochemistry 1999, 38 , 4303. Radkiewicz, J.L.; Brooks, C.L. J. Am. Chem. Soc. 2000, 122, 225. Pan, H.; Lee, J.C.; Hilser, V.J. Proc. Natl. Acad. Sci. USA 2000, 97 , 12020. Babine, R.E.; Bender, S.L. Chem. Rev. 1997, 97 , 1359. Sinha, N.; Nussinov, R. Proc. Natl. Acad. Sci. USA 2001, 98 , 3139. Bouzida, D.; Rejto, P.A.; Arthurs, S.; Colson, A.B.; Freer, S.T.; Gehlhaar, D.K.; Larson, V.; Luty, B.A.; Rose, P.W.; Verkhivker, G.M. Int. J. Quantum Chem. 1999, 72 , 73. Murray, C.W.; Baxter, C.A.; Frenkel, A.D. J. Comp.-Aided Mol. Design 1999, 13 , 547. Gane, P.J.; Dean, P.M. Curr. Opin. Struct. Biol. 2000, 10 , 401. Goodford, P.J. J. Med. Chem. 1985, 28 , 849. Ewing, T.J.A.; Kuntz, I.D. J. Comput. Chem. 1997, 18 , 1175. Morris, G.M.; Goodsell, D.S.; Halliday, R.S.; Huey, R.; Hart, W.E.; Belew, R.K.; Olson, A.J. J. Comput. Chem. 1998, 19 , 1639. Miranker, A.; Karplus, M. Proteins 1995, 23 , 472-. Carlson, H.A.; Masukawa, K.M.; Jorgensen, W.L.; Lins, R.D.; Briggs, J.M.; McCammon, J.A. J. Med. Chem. 2000, 43 , 2100. Kuntz, I.D.; Meng, C.E.; Shoichet, B.K. Acc. Chem. Res. 1994, 27 , 117. Joseph-McCarthy, D.; Fedorov, A.A.; Almo, S.C. Protein Eng. 1996 9, 773. Leclerc, F.; Karplus, M. Theor. Chem. Acc. 1999, 101, 131. Bemis, G.W.; Murcko, M.A. J. Med. Chem. 1999, 42 , 5095. Verdonk, M.L.; Cole, J.C.; Taylor, R. J. Mol. Biol. 1999, 289, 1093. Jacobs, D.J. Rader, A.J.; Kuhn, L.A.; Thorpe, M.F. Proteins 2001, 44 , 150. Thomas, B.E.; Joseph-McCarthy, D.; Alvarez, J.C. In Pharmacophore Perception, Development, and Use In Drug Design. Gner, O.F., Ed.; International University Line: La Jolla, CA, pp.352-367.

[37] [38] [39] [40] [41] [42]

Leach, A.R. J. Mol. Biol. 1994, 235, 345. Schnecke, V.; Swanson, C.A.; Getzoff, E.D.; Tainer, J.A.; Kuhn, L.A. Proteins 1998, 33 , 74. Schaffer, L.; Verkhivker, G.M. Proteins 1998, 33 , 295. Trosset, J.Y.; Scheraga, H.A. J. Comput. Chem. 1999, 20 , 244. Pak, Y.; Wang, S. J. Phys. Chem. B 2000, 104, 354. Verkhivker, G.M.; Rejto, P.A.; Bouzida, D.; Arthurs, S.; Colson, A.B.; Freer, S.T.; Gehlhaar, D.K.; Larson, V.; Luty, B.A.; Marrone, T.; Rose, P.W. Chem. Phys. Lett. 2001, 337, 181. Philippopoulos, M.; Lim, C. Proteins 1999, 36 , 87. Clarage, J.B.; Romo, T.; Andrews, B.K.; Pettitt, B.M.; Phillips, G.N. Proc. Natl. Acad. Sci. USA 1995, 92 , 3288. Clauen, H.; Buning, C.; Rarey, M.; Lengauer, T. J. Mol. Biol. 2001, 308, 377. Knegtel, R.M.A.; Kuntz, I.D.; Oshiro, C.M. J. Mol. Biol. 1997, 266, 424. sterberg, F.; Morris, G.M.; Sanner, M.F.; Olson, A.J.; Goodsell, D.S. Proteins 2002, 46 , 34. Carlson, H.A.; Masukawa, K.M.; McCammon, J.A. J. Phys. Chem. A 1999, 103, 10213. Raymer, M.L.; Sanschagrin, P.C.; Punch, W.F.; Venkataraman, S.; Goodman, E.D.; Kuhn, L.A. J. Mol. Biol. 1997, 265, 445. Makarov, V.A.; Andrews, B.K.; Pettitt, B.M. Biopolymers 1998, 45 , 469. Schnecke, V.; Kuhn, L.A. Persp. Drug Disc. Design 2000, 20 , 171. Kuntz, I.D.; Blaney, J.M.; Oatley, S.J.; Langridge, R.; Ferrin, T.E. J. Mol. Biol. 1982, 161, 269. Sudbeck, E.A.; Mao, C.; Vig, R.; Venkatachalam, T.K.; Tuel-Ahlgren, L.; Uckun, F.M. Antimicrob. Agents Chemother. 1998, 42 , 3225. Stultz, C.M.; Karplus, M. Proteins 1999, 37 , 512. Broughton, H.B. J. Mol. Graphics Model. 2000, 18 , 247 Kastenholz, M. A.; Pastor, M.; Cruciani, G.; Haaksma, E.E.J., Fox, T. J. Med. Chem. 2000, 43 , 3033. Chen, Y.Z.; Zhi, D.G. Proteins 2001, 43 , 217. Kramer, B.; Rarey, M.; Lengauer, T. Proteins 1999, 37 , 228. Anderson, A.C.; ONeil, R.H.; Surti, T.S.; Stroud, R.M. Chem. & Biol. 2001, 8, 445.

[43] [44] [45] [46] [47] [48] [49]

[23] [24] [25] [26] [27]

[28] [29]

[50] [51] [52] [53]

[30] [31] [32] [33] [34] [35] [36]

[54] [55] [56] [57] [58] [59]

RELATED ARTICLES RECENTLY PUBLISHED IN CURRENT PHARMCEUTICAL DESIGN Hoare, S.R.J. and Usdin, T.B. Molecular Mechanisms of Ligand Recognition by Parathyroid Hormone 1 (PTH1) and PTH2 Receptors. Curr. Pharm. Des., 2001, 7(8), 689-713. Wouters, J. and Ooms, F. Small Molecule Crystallography in Drug Design. Curr. Pharm. Des., 2001, 7(7), 52945. Yamada, S.; Yamamoto, K. and Masuno, H. StructureFunction Analysis of Vitamin D and VDR Model. Curr. Pharm. Des., 2000, 6(7), 733-48.

Das könnte Ihnen auch gefallen