Article pubs.acs.org/jmc

Rapid Identification of Ligand-Binding Sites by Using an AssignmentFree NMR Approach Yuya Kodama,†,‡,§ Koh Takeuchi,‡ Nobuhisa Shimba,§ Kohki Ishikawa,§ Ei-ichiro Suzuki,§ Ichio Shimada,*,‡,∥ and Hideo Takahashi*,‡,⊥ †

Japan Biological Informatics Consortium (JBIC), 2-3-26 Aomi, Koto-ku, Tokyo 135-0064, Japan Biomedicinal Information Research Center (BIRC), National Institute of Advanced Industrial Science and Technology (AIST), 2-3-26 Aomi, Koto-ku, Tokyo 135-0064, Japan § Institute for Innovation, Ajinomoto Co., Inc., 1-1 Suzuki-cho, Kawasaki-ku, Kawasaki-shi 210-8681, Japan ∥ Graduate School of Pharmaceutical Sciences, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan ⊥ Graduate School of Medical Life Science, Yokohama City University, 1-7-29 Suehiro-cho, Tsurumi-ku, Yokohama 230-0045, Japan ‡

S Supporting Information *

ABSTRACT: In this study, we developed an assignment-free approach for rapid identification of ligand-binding sites in target proteins by using NMR. With a sophisticated cell-free stable isotope-labeling procedure that introduces 15N- or 13Clabels to specific atoms of target proteins, this approach requires only a single series of ligand titrations with labeled targets. Using titration data, ligand-binding sites in the target protein can be identified without time-consuming assignment procedures. We demonstrated the feasibility of this approach by using structurally wellcharacterized interactions between mitogen-activated protein (MAP) kinase p38α and its inhibitor 2-amino-3-benzyloxypyridine. Furthermore, we confirmed the recently proposed fatty acid binding to p38α and confirmed the fatty acid-binding site in the MAP kinase insert region.



INTRODUCTION Recent developments in high-throughput screening (HTS) and in silico technologies allow rapid and efficient identification of active compounds from large-scale compound libraries.1,2 However, interactions between the compounds and target molecules identified require verification in independent experiments. NMR is a useful technique for this purpose as it allows a simple and direct investigation of protein−ligand interaction in aqueous solutions. Furthermore, NMR has the advantage of allowing identification of the compound-binding sites in the target proteins by using chemical shift perturbation experiments.3 The resulting information validates binding modes of active ligands and elucidates structural characteristics of these interactions. Nonetheless, NMR analyses are generally timeconsuming and challenging, mainly because of complicated assignment procedures for NMR spectra. In particular, the utility of NMR for large proteins is limited by lower sensitivity and greater complexity of resonances, which hamper complete assignment of the spectra. Here, we present an alternative assignment-free approach for rapid identification of compound-binding sites by using NMR. Although it shares basic concepts with our previous assignmentfree approach for protein−protein interactions,4 extensive modifications of the procedure accommodate smaller binding interfaces of low-molecular-weight compounds. The key feature of this approach is an elaborate labeling scheme that utilizes single-site 13C- or 15N-labeled amino acids in combination with newly developed computational protocols that are suitable for © 2013 American Chemical Society

small molecule interactions. In particular, a set of amino acids were carefully chosen and labeled for efficient detection of compound interactions. Ligand-binding sites were identified without prior assignment of resonances by using known 3D structures of target proteins and the number of shifted signals for each labeled amino acid in NMR titration experiments. In this study, we demonstrated the feasibility of this approach in analyses of the interaction between p38α MAP kinase and its inhibitor 2-amino-3-benzyloxypyridine.5,6 Furthermore, we confirmed a binding site for fatty acids, which are known to activate p38α, demonstrating rapid verification of a proposed biological interaction by using this assignment-free NMR approach.



RESULTS AND DISCUSSION Strategy for Assignment-Free Identification of Binding Sites. A schematic of the assignment-free site-identification strategy is shown in Figure 1. This strategy employs only a single NMR sample in which the target protein is labeled with single α-15N-labeled amino acid and several amino acids that are 13 C-labeled on specific atoms. In this study, we used α-15Nlabeled Phe, and [13Cε] Met, [13Cα/13Cβ] Ala, [13Cε] His, [13Cε] Tyr, and [13Cδ] Trp for 13C-labeled amino acids. The combination of α-15N- and 13C-labeled amino acids was designed to satisfy similar criteria to those developed in our Received: September 17, 2013 Published: October 30, 2013 9342

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

on binding of ligands, were counted for each isotopically labeled amino acid type and were assumed to reflect the numbers of amino acid residues present near the ligand-binding site. Calculations of ΔωRMS and ωt values are presented in the Experimental Section. In addition to protein labeling, this strategy requires 3D structures of target proteins. These structures may include ligand-free forms (referred to as apo structure) or target proteins in complex with other ligands after removing the bound ligand (holo structure, Figure 1A). All hydrophobic surfaces of target proteins that could accommodate small hydrocarbons were computationally identified and recorded as grid points (Figure 1B). The procedure for generating candidate grid points is presented in the Experimental Section. For each grid point, the degree of consistency between (1) the number and type of shifted amino acid residues in the titration experiment and (2) the number and type of surfaceaccessible amino acids near the grid point were evaluated according to the procedure for obtaining penalty scores (Stot), as described in the Experimental Section (Figure 1C). Finally, clusters of the most consistent grid points that were large enough to accommodate the ligand of interest were identified as ligand-binding sites (Figure 1D). To confirm the generality of our labeling and computational algorithms, we conducted simulation studies prior to the experiment. In these simulations, the number of labeled amino acid residues within 5 Å of the ligand was counted as shifted residues. Using this data, we searched for ligand-binding sites of apo forms of the four target molecules according to the procedure described above. In all cases, we successfully identified ligand-binding sites (Supporting Information Figure S3), confirming that this approach provides sufficient information to unambiguously identify these sites. Specific Isotope Labeling of p38α. To minimize the use of isotope-labeled amino acids and metabolic scrambling, an in vitro expression system was used to produce p38α. Typically, isotope-labeled p38α was obtained from 2 mL of in vitro expression solution supplemented with 7.5 mg Met-13Cε, 4.5 mg Ala-13Cα/13Cβ, 7.8 mg His-13Cε, 9.1 mg Tyr-13Cε, 10.3 mg Trp-13Cδ, and 8.3 mg Phe-α-15N. The total cost of the labeled amino acids was about $157, which is comparable to 0.3 L of ul-2H/15N-labeled Escherichia coli culture media. Single in vitro expression reactions typically yielded 0.2−0.3 mM NMR samples. NMR spectra with correlated 1H−13C and 1H−15N, and sufficient S/N ratios for most resonances were obtained after 2−3 h in time-shared NMR experiments.4 To observe weaker signals with sufficient S/N ratios, the experiment was performed for 16 h, and 98% of expected resonances were identified.8 The number of observed overexpected resonances was 11/11, 25/25, 11/12, 15/15, 5/5, and 12/13 for Met-13Cε, Ala-13Cβ, His-13Cε, Tyr-13Cε, Trp-13Cδ, and Phe-α-15N, respectively. As discussed below, because of high molecular weight (41 k) and a region with significant line broadening, p38α is not an easy target protein. Accordingly, required experimental times and/or protein concentrations were significantly reduced with easier target proteins. Determination of the Binding Site for 2-Amino-3benzyloxypyridine Using the Assignment-Free Approach. The feasibility of our approach was tested on MAP kinase p38α and its inhibitor 2-amino-3-benzyloxypyridine. The structure of the ligand−protein complex was previously solved at a resolution of 2.2 Å (PDB ID: 1W7H)6 and was suitable for evaluation of the accuracy of the proposed method. Up to 1.6

Figure 1. Flowchart for the assignment-free ligand-binding site identification strategy. On the basis of the known protein structures (A), candidate grid points for ligand-binding sites are generated using EasyMIFs and SiteHound (B). Subsequently, consistency between experimental data (upper right panels) and the surrounding amino acid distribution is evaluated for each grid point. Grid point scores are represented as color gradations in the figure (cyan, highest; red, lowest (C)). In the final stage, the most consistent cluster of grid points is selected based on experimental consistency and the volume of the grid point cluster (D).

previous study.4 However, additional criteria for low-molecularweight ligands included (1) a high propensity of selected amino acids to exist at protein−ligand interfaces7 (see Supporting Information Figure S1), (2) abundance of at least one amino acid to ensure the complete coverage of protein surface by labeled amino acids, (3) minimal spectral overlap of 13C-labeled amino acids, and (4) commercial availability of labeled amino acids. Except for Ala, all commercially labeled amino acids met these criteria (1). However, these amino acids are usually not sufficient to cover entire protein surfaces. Thus, Ala, which meets the second criteria, was used to ensure sufficient NMR probe atoms on protein surfaces. To avoid signal overlap of amino acid types, the 13C-labeling site of each selected amino acid was carefully chosen by referring to the expected distributions of 13C/1H crosspeaks from the data of the Biological Magnetic Resonance Data Bank (BMRB, www.bmrb. wisc.edu; see Supporting Information Figure S2). To decipher possible overlaps between Ala-13Cβ and Met-13Cε, 13Cα/13Cβdoubly labeled Ala was used. In this way, resonances from Ala-13Cβ and Met-13Cε give different sign in constant-time HSQC spectra. We recommend the use of the same combination of labeled amino acids. However, other combinations may also be feasible. Ligands were then titrated with labeled proteins to obtain numbers and amino acid types of shifted resonances in 1H−15N and 1H−13C shift correlation spectra (Figure 1, upper right). The numbers of resonances, which showed the normalized chemical shift changes (ΔωRMS) over the threshold value (ωt) 9343

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

Figure 2. Assignment-free NMR titration experiments with p38α and 2-amino-3-benzyloxypyridine. (A) ΔωRMS of each resonance induced by the binding of 2-amino-3-benzyloxypyridine to specifically labeled p38α. The red line shows the threshold used for classification (ωt = 0.207). The definitions of ΔωRMS and ωt are described in the Experimental Section. (B) Overlay of selected regions in the HSQC spectra acquired for the specifically labeled p38α alone (black) and with 2-amino-3-benzyloxypyridine (red); signals with number signs in the Ala-13Cβ region indicate those originating from Met-13Cε. Asterisks in Phe-α-15N regions indicate the anti-TROSY components originating from His-13Cε, which happen to be in the same region in the time-shared experiment. Note the 1H/13C resonances were aliased in the 13C-indirect dimension in order to reduce 13C spectral width and to observe the 1H/15N resonances simultaneously in the same spectrum. Black arrows highlight significantly perturbed resonances, which are above the threshold shown in (A). In measurements of complexes, a 5.5-fold excess of 2-amino-3-benzyloxypyridine (1.6 mM) was added to p38α (0.29 mM).

resonances, which comprise 5% of all observed resonances. In summary, titration experiments yielded RAla = 3, RMet = 1, RTrp = 0, RTyr = 0, RHis = 0/1, and RPhe = 0/1 for further analysis. Two combinations of His and Phe reflected the number of unobserved resonances for those residues (see Experimental Section). Candidate grid points for 2-amino-3-benzyloxypyridine were generated as described in the Experimental Section. The spacing between grid points was 0.5 Å. Penalty scores for each grid point were then evaluated, and three clusters of grid points with Stot = 1, which corresponds to no inconsistencies between experimental data and simulations, were identified on the surface of p38α (Figure 3A). It should be noted that Stot cannot

mM 2-amino-3-benzyloxypyridine was titrated with 0.29 mM MAP kinase p38α to determine the number of residues of each amino acid type at the binding site. The ΔωRMS values of resonances originating from each residue are shown in Figure 2A, with the spectra corresponding to the region of each amino acid type in Figure 2B. In these experiments, 3 Ala and 1 Met resonances were significantly affected, although no remarkable change was observed for Tyr, His, Trp, or Phe resonances (Figure 2A). The distribution of ΔωRMS is shown in Supporting Information Figure S4. As shown in Figure 2 and Supporting Information Figure S4, resonances for which ΔωRMS values were above the threshold value (ωt) were limited and had considerable offsets from the distribution of unperturbed 9344

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

Figure 3. Identification of the 2-amino-3-benzyloxypyridine binding site in p38α. (A) The grid points with Stot equal to 1 are shown in red on the surface of the apo structure of p38α. (B) The complex structure of p38α and 2-amino-3-benzyloxypyridine (PDB ID: 1W7H) was superimposed on the apo structure. Green and gray ribbons represent the backbone trace for holo and apo forms of p38α, respectively. The surfaces of the most consistent clusters of the grid points are shown in red, whereas compounds identified in the complex structure are shown in blue. All structural figures were prepared using the Discovery Studio 3.5 (Accelrys, Inc.).

experiments were translated as RAla = 1, RMet = 1, RTrp = 1, RTyr = 0, RHis = 1/2, and RPhe = 0/1 for decanoic acid (Supporting Information Figure S5). The chemical shift perturbation induced by decanoic acid was smaller than that of the 2amino-3-benzyloxypyridine interaction. This probably reflects differences in the chemical structure of ligands containing an aromatic moiety that induces greater chemical shift changes compared to hydrocarbon chains with large anisotropy of aromatic groups in magnetic fields. In addition, we could not be able to reach 100% bound condition in case of decanoic acid because significant deterioration of NMR signals was observed at high fatty acid concentration. On the basis of these data, the binding site was searched using three different coordinates of p38α: the apo form (PDB ID: 1R39), the 2-amino-3-benzyloxypyridine binding form (PDB ID: 1W7H), and the β-OG binding form (PDB ID: 2NPQ).14 Among these p38α forms, the most consistent cluster with a volume greater than decanoic acid was located at the β-OG binding site of the corresponding structure (Figure 4). These data strongly suggest that decanoic acid binds the MAP kinase insert region in solution. Titration of β-OG also induced chemical shift changes in a manner similar to decanoic acid (Figure 4C,D). One tryptophan residue, Trp197, was located at the decanoic acid (or β-OG) binding site and was considered important for lipid binding.14,17 To confirm that the shifted resonance in the decanoic acid titration corresponded to Trp197 (Figure 4D), we substituted Phe for Trp197 (W197F mutant). Indeed, the shifted resonance was not present in the spectra of the W197F mutant (Figure 4E). These results suggest that the binding site for decanoic acid was reliably identified using our assignment-free approach. In addition, the successful determination of the decanoic acid binding site, even in the case for low ligand-bound population and less perturbing ligand chemical structures, would indicate the robustness of our approach. It should also be noted that most of the drug-like compounds have an aromatic moiety and thus induce larger chemical shift change compared to natural ligands. Therefore, our approach would be most suitable in medicinal chemistry approaches. Comparisons with Other Ligand Binding Analyses. Computational protocols that rely solely on in silico calculations

be zero as there are two missing resonances in the analysis. The volumes of these three clusters of grid points were 176, 75, and 26 Å3, respectively; only one cluster had a volume greater than that of 2-amino-3-benzyloxypyridine (153 Å3) and other clusters were obviously too small to accommodate the compound. This most consistent cluster comprised 170 grid points, which represented 1.0% of all grid points before selection and was proximal to the binding site for 2-amino-3benzyloxypyridine in the complex X-ray structure (Figure 3B). Distances between centroids of grid points and bound ligands (Dc) and minimum distances between centroids of grid points and any heavy atom of the bound ligand (Dmin) were 5.32 and 2.81 Å, respectively. While we used the apo structure of p38α (PDB ID: 1R39)9 in lieu of complex structure information, the precision of the method could also be validated using structural information of the p38α−ligand complex. When structures of holo p38α (PDB ID: 1W7H) were used for calculations, Dc and Dmin were reduced to 4.60 and 1.10 Å, respectively. Identification of Fatty Acid-Binding Sites in p38α. Recent investigations have demonstrated that some fatty acids and their derivatives such as arachidonic acid and decanoic acid activate the p38 cascade.10−13 A study by Diskin et al.,14 in which a n-octyl-β-D-glucopyranoside (β-OG) molecule cocrystallized with p38α, revealed a conformational change on binding of β-OG to the MAP kinase insert region. They also suggested that arachidonic acid binds to the same site as β-OG, and they discussed functional modulation of p38α by direct interaction with lipids. Taken together, biochemical data and crystal structures of p38α in complex with phosphatidylinositol ether lipid analogue (PIA) 23, PIA24, and perifosine, provide insights into the activation mode of p38α.15,16 These data indicate that the MAP kinase insert region might be a novel allosteric site that modulates p38α function and that fatty acids may also bind the MAP kinase insert region. To confirm this, we identified fatty acid-binding sites of p38α using our assignment-free approach. To detect binding of decanoic acid, which reportedly activates the p38 cascade,11 p38α was labeled in the same way as in the 2-amino-3-benzyloxypyridine experiment. The ΔωRMS values of resonances originating from each amino acid residue are presented with experimental data in Supporting Information Figure S5. Data from titration 9345

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

Figure 4. Identification of the decanoic acid-binding site in p38α. (A) Only grid points with Stot less than 3 are shown. Orange and yellow represent Stot = 2 and Stot = 3, respectively. There is no grid point for Stot = 1. (B) Enlarged view of the MAP kinase insert region in p38α, with the surface of the most consistent cluster of grid points shown in red. The positions of the β-OG molecule and Trp197 of p38α in the X-ray structure of the complex are represented by cyan and green stick models, respectively. (C,D) Overlay of the Trp region in the HSQC spectra acquired for free p38α (black) and in the presence of (C) β-OG or (D) decanoic acid. Bound spectra are shown in red in both panels. Complex spectra were recorded with a 3-fold excess of sodium decanoate (0.75 mM) or with a 0.67 mol equiv of β-OG (0.29 mM). (E) Trp197 resonance was assigned by comparing the spectra of the wild-type (black) with W197F (red).

to predict ligand-binding sites have been extensively developed.7,18−46 The success rates of these approaches are generally good for a deep, well-defined pocket. For instance, MolSite, which was recently reported by Fukunishi et al.,47 succeeded to predict ligand-binding sites with high accuracy (80−99%). However, the shallow interaction sites that do not have obvious deep pockets remain difficult to predict. In

contrast, the present approach is based mainly on experimental data and, therefore, identifies such ligand-binding sites more robustly. Another approach that uses both NMR data and computational analyses was reported by McCoy and Wyss,48 who discussed the potential use of their method for identification of assignment-free binding sites. However, their approach requires multiple samples with varied selective 15N9346

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

suboptimal exchange properties (see below), we obtained reasonable quality spectra for 41 kDa MAP kinase p38α within 16 h. Although amide resonances originating from the MAP kinase insert region are likely to be observed in p38α, it is difficult to assign such resonances using conventional 3D NMR techniques.56 This may be because of the chemical exchange phenomenon that relates to dynamic features such as conformational flexibility of the region and enhanced transverse relaxation during lengthy 3D NMR experiments. These complications are common in analyses of catalytic enzyme cores and proteins with functional conformational flexibility. Hence, our assignment-free approach may be beneficial for such analyses. Finally, our approach offers rapid validation of binding sites for candidate compounds of drug discovery. This method robustly generates ligand binding information for regions with significant flexibility, which hinders conventional NMR analyses with fast transverse relaxation.

labeling and multiple titration experiments reflecting the number of samples. In addition, their method may only be suitable for ligands with aromatic rings, as interpretation of experimental data relies on chemical shift changes induced by ring currents. In contrast, our approach requires only a single specifically labeled sample and a titration experiment and does not demand prerequisite ligand structures. Straightforward application of our approach might be difficult when target proteins exhibit significant conformational changes on ligand binding, especially when the ligand binding pocket is occluded in the apo structure and opens with ligand binding. In fact, the fatty acid binding site of p38α was successfully identified when we used the β-OG binding form of p38α as the template and the other p38α structures which do not have a fully formed fatty acid binding site were not suitable to identify the fatty acid binding site. A careful comparison between the structures with and without the fatty acid binding site revealed that the backbone conformation around the binding site was almost identical, whereas the several side chains exhibited significant conformation changes to form the binding site. From this notion, we would speculate that binding site can be identified by using our approach if we use the backbone structure without side chains as the template and regenerate side chain conformations afterward using computational methods. Neverthless, the demonstration of above ideas is rather out of scope in this research paper and this should be addressed as a separated paper in the future. In addition, we should point out that the previously reported computational methods for predicting open structures may also be appropriate to search hidden binding sites.49−53 Moreover, when structural data for target proteins is not available, homology models may compensate. Our strategy allows some uncertainty of modeled structures, as exemplified by similarities of results obtained from apo and holo 2-amino-3-benzyloxypyridine binding structures. Nevertheless, with predicted structure templates, data can be confirmed using methods such as site-specific mutagenesis. In addition, it should be noted that our proteinobserved strategy would be complementary to the ligandobserved, assignment-free strategies for ligand-binding-site identification such as SOS-NMR and INPHARMA.54,55



EXPERIMENTAL SECTION

Materials. All isotope-labeled chemicals were purchased from Cambridge Isotope Laboratories Inc. (MA, USA). All other chemicals were purchased from Wako Pure Chemical Industries, Ltd. (Osaka, Japan), NACALAI TESQUE, Inc. (Kyoto, Japan), or Sigma-Aldrich Co. (MO, USA), unless otherwise indicated. Protein Synthesis, Purification, and Single-Point Mutation. For production of specifically labeled p38α, the human MAPK14 gene was inserted into the pIVEX2.4d plasmid (Roche Diagnostics GmbH; Mannheim, Germany) with an N-terminus 6× His-tag and a digestion site for tobacco etch virus (TEV) protease between the His-tag and the MAPK14 gene. Mixtures of specific 13C-labeled amino acids ([13Cε] Met, [13Cα/13Cβ] Ala, [13Cε] His, [13Cε] Tyr, and [13Cδ] Trp) and [α-15N] Phe were prepared according to the protocol for the RTS amino acid sampler (Roche Diagnostics GmbH). In vitro synthesis of the labeled p38α was carried out using an Escherichia coli-based cellfree system (RTS 500 ProteoMaster E. coli HY kit; Roche Diagnostics GmbH) at 303 K, according to the manufacturer’s protocol. Synthesized p38α was purified by affinity chromatography with a Ni-NTA Sepharose 6 Fast Flow column (GE Healthcare UK Ltd.; Buckinghamshire, England). After cleavage of the fusion tag using TurboTEV Protease (Accelagen Inc.; CA, USA), the sample was applied to a HisTrap HP column (GE Healthcare) to remove digested His-tag and protease. Flow-through fractions were applied to a Superdex 200 10/300 GL (GE Healthcare), with a running buffer containing 50 mM sodium phosphate (pH 6.8), 150 mM sodium chloride, and 5 mM dithiothreitol. Site-specific mutations were introduced using a QuikChangeII sitedirected mutagenesis kit (Agilent Technologies, Inc.; CA, USA). NMR Spectroscopy and Titration Experiments. All NMR spectra were recorded at 303 K on a Bruker AVANCE-600 instrument equipped with a cryo-cooled probe (Bruker BioSpin GmbH; Karlsruhe, Germany). 1H−15N and 1H−13C shift correlations were simultaneously obtained in time-shared NMR experiments.4 The experiment was implemented using TROSY type coherence transfer in indirect dimension for backbone 1H−15N and aromatic 1H−13C shift correlations. Protein samples were concentrated to 0.25−0.43 mM in NMR buffer containing 5 mM deuterated 4-(2-hydroxyethyl)-1piperazineethanesulfonic acid (HEPES) (pH 6.8), 150 mM sodium chloride, and 5 mM dithiothreitol in 1H2O/2H2O (95:5). 2-Amino-3benzyloxypyridine (Thermo Fisher Scientific Inc.; MA, USA; > 97% purity determined by using HPLC) was dissolved in deuterated DMSO at 40 or 160 mM. Sodium decanoate (Sigma-Aldrich Co.; ≥ 98% purity determined by using GC) and β-OG (Tokyo Chemical Industry Co., Ltd.; Tokyo, Japan; > 96% purity determined by using GC) were dissolved in NMR buffer at 50 and 20 mM, respectively. Maximum concentrations of 2-amino-3-benzyloxypyridine, sodium



CONCLUSION In this study, we developed an assignment-free NMR approach for rapid identification of ligand-binding sites and demonstrated the feasibility of our approach by determining the 2-amino-3benzyloxypyridine binding site of p38α MAP kinase. Furthermore, we structurally confirmed fatty acid-binding sites of p38α. The present strategy utilizes sophisticated and selective stable isotope-labeling and 3D structures of target proteins. Cell-free expression systems provide methods of choice for amino acidselective labeling of target proteins. Advantages of expressing labeled proteins using cell-free systems include (1) reduced cost of expressing labeled proteins using the minimal quantities of expensive stable isotopes and (2) evasion of metabolic amino acid scrambling, which might complicate NMR analyses. As for NMR experiments, this strategy requires only a few time-shared HSQC measurements with and without ligands. The additional transverse-relaxation period introduced by timeshared element is not detrimental for sensitive detection of protein resonances. In most cases, only a few hours are required to obtain reasonably good spectra using a state-of-the-art NMR spectrometer. Indeed, despite its high molecular weight and 9347

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

decanoate, and β-OG in the NMR samples were 1.6, 0.75, and 0.29 mM, respectively. Evaluation of Chemical Shift Change. The normalized chemical shift change of each resonance upon ligand binding was calculated using the following equation, with chemical shift distributions of the correlated nuclei (e.g., 1H/15N or 13C) as described previously.57 ΔωRMS =

1 N

so that the number of unobserved resonances would not bias the penalty score. Grid points with smaller Stot had higher chances of being the ligandbinding site. A simple Python script used for calculation of Stot for each grid point is shown in the Supporting Information (section SI 1). Clusters of grid points with minimal Stot and cluster volumes greater than or equal to those of ligands were defined as ligand-binding sites.



⎛ Δωi ⎞2 ∑⎜ ⎟ di ⎠ i ⎝

S Supporting Information *

In this equation, N represents the number of nuclei correlated in each resonance (2 for 2D and 3 for 3D experiments); di denotes the nucleus-specific value, which corresponds to the standard deviation of the chemical shift distribution of the nucleus i (1H, 13C, or 15N) in each resonance, and Δωi represents the chemical shift change in nucleus i. The di values used in this study are described in the Supporting Information (Table S1). The threshold value (ωt) for the number of significantly shifted resonances is defined as

Script used in analysis, chemical shift distributions of each nucleus of certain amino acid residues, propensity of each amino acid-type residue to exist in a protein−ligand interface, selection of 13C-labeling sites that avoid signal overlaps in HSQC spectra, distribution of ΔωRMS in the 2-amino-3benzyloxypyridine titration experiment, results of the simulation study, and titration experiment data of p38α with decanoic acid. This material is available free of charge via the Internet at http://pubs.acs.org.



ωt = m + s where m is the average of normalized chemical shift changes (ΔωRMS) and s is the standard deviation of ΔωRMS. The numbers of resonances that showed the normalized chemical shift changes (ΔωRMS) over the threshold (ωt) on binding of the ligand were counted for each isotopically labeled amino acid type; this reflected the numbers of residues of each amino acid type present near the ligand-binding site. Generation of Candidate Grid Points. Three-dimensional structures of p38α were prepared from crystal structures (PDB ID: 1R39 for free form, 1W7H for 2-amino-3-benzyloxypyridine binding form, and 2NPQ for β-OG binding form) by reconstructing the missing coordinates using the Swiss-PdbViewer58 4.0.1 function Scan Loop Database. The coordinate with the lowest force field energy after energy minimization of reconstructed and peripheral residues was used. Candidate sites for ligand binding in p38α were computationally searched using an energy-based approach with EasyMIFS and SiteHound,59−61 which propose all hydrophobic surfaces suitable for binding of small molecule inhibitors. In the candidate grid search using EasyMIFS, spacing between grids was set to 0.5 Å. The methyl probe, which is the default probe in EasyMIFS, was used to search for inhibitor-binding sites, whereas the CH2 probe was used to identify fatty acid-binding sites. Grid points generated by EasyMIFS were further processed using SiteHound to remove all grid points that had unfavorable interaction energies. The unfavorable interaction energy threshold was set to −8.0. Other parameters were set at values recommended in the User’s Guide. Determination of Binding Site. To score each grid point, a virtual sphere (VS) with radius rv was calculated:

AUTHOR INFORMATION

Corresponding Authors

*For I. S.: phone/fax, +81-3-3815-6540; E-mail, [email protected]. *For H.T.: phone/fax, +81-45-508-7213; E-mail, hid@tsurumi. yokohama-cu.ac.jp. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Dr. H. Hanzawa for the p38α gene and Drs. Y. Mizukoshi and Y. Tokunaga for technical advice. This work was supported by a grant from the Japan New Energy and Industrial Technology Development Organization (NEDO) and the Ministry of Economy, Trade, and Industry (METI; to I.S.). This work was partly supported by Grants-in-Aid for Scientific Research (KAKENHI) grant nos. 24370048 (to H.T. and K.T.), 24102524 (to H.T.), and 25121743 (to K.T.) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) and Japan Society for the Promotion of Science (JSPS).



ABBREVIATIONS USED β-OG, n-octyl-β-D-glucopyranoside; HEPES, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid; MAP, mitogen-activated protein



rv = rseb + rcsp Here, rseb is the radius of the smallest enclosing ball of the ligand, as calculated using ligand coordinates from PubChem3D (pubchem.ncbi. nlm.nih.gov), and rcsp is the expected distance of chemical shift perturbations (CSP). In this study, rcsp was set at 4 Å. The number and types of surface-accessible amino acid residues (solvent accessibility >2%) within VS were counted for each grid point, and penalty scores (Stot) reflecting inconsistency of amino acid compositions with those indicated by CSP in experimental data were calculated as follows: Stot =

ASSOCIATED CONTENT

REFERENCES

(1) Mayr, L.; Bojanic, D. Novel trends in high-throughput screening. Curr. Opin. Pharmacol. 2009, 9, 580−588. (2) Villoutreix, B.; Eudes, R.; Miteva, M. Structure-based virtual ligand screening: recent success stories. Comb. Chem. High Throughput Screening 2009, 12, 1000−1016. (3) Cavanagh, J.; Fairbrother, W. J.; Palmer, A. G., III; Skelton, N. J.; Rance, M. Protein NMR Spectroscopy, 2nd ed.; Academic Press: London, 2006. (4) Kodama, Y.; Reese, M. L.; Shimba, N.; Ono, K.; Kanamori, E.; Dötsch, V.; Noguchi, S.; Fukunishi, Y.; Suzuki, E. I.; Shimada, I.; Takahashi, H. Rapid identification of protein−protein interfaces for the construction of a complex model based on multiple unassigned signals by using time-sharing NMR measurements. J. Struct. Biol. 2011, 174, 434−442. (5) Gill, A.; Frederickson, M.; Cleasby, A.; Woodhead, S.; Carr, M.; Woodhead, A.; Walker, M.; Congreve, M.; Devine, L.; Tisi, D.;

∑ |Va − Ra| a

Here, Va is the number of isotope-labeled amino acid type a within VS of the grid point and Ra is the number of significantly perturbed resonances of amino acid type a. In the case that not all expected resonances were observed for some reasons (broadening and/or overlapping), all possible penalty scores were calculated and averaged 9348

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

O’Reilly, M.; Seavers, L.; Davis, D.; Curry, J.; Anthony, R.; Padova, A.; Murray, C.; Carr, R.; Jhoti, H. Identification of novel p38alpha MAP kinase inhibitors using fragment-based lead generation. J. Med. Chem. 2005, 48, 414−426. (6) Hartshorn, M.; Murray, C.; Cleasby, A.; Frederickson, M.; Tickle, I.; Jhoti, H. Fragment-based lead discovery using X-ray crystallography. J. Med. Chem. 2005, 48, 403−413. (7) Soga, S.; Shirai, H.; Kobori, M.; Hirayama, N. Use of amino acid composition to predict ligand-binding sites. J. Chem. Inf. Model. 2007, 47, 400−406. (8) Vogtherr, M.; Saxena, K.; Hoelder, S.; Grimme, S.; Betz, M.; Schieborr, U.; Pescatore, B.; Robin, M.; Delarbre, L.; Langer, T.; Wendt, K.; Schwalbe, H. NMR characterization of kinase p38 dynamics in free and ligand-bound forms. Angew. Chem., Int. Ed. Engl. 2006, 45, 993−997. (9) Patel, S.; Cameron, P.; Frantz-Wattley, B.; O’Neill, E.; Becker, J.; Scapin, G. Lattice stabilization and enhanced diffraction in human p38 alpha crystals by protein engineering. Biochim. Biophys. Acta 2004, 1696, 67−73. (10) Reddy, M.; Thimmalapura, P.; Lanting, L.; Nadler, J.; Fatima, S.; Natarajan, R. The oxidized lipid and lipoxygenase product 12(S)hydroxyeicosatetraenoic acid induces hypertrophy and fibronectin transcription in vascular smooth muscle cells via p38 MAPK and cAMP response element-binding protein activation. Mediation of angiotensin II effects. J. Biol. Chem. 2002, 277, 9920−9928. (11) Collins, Q.; Xiong, Y.; Lupo, E. J.; Liu, H.; Cao, W. p38 Mitogen-activated protein kinase mediates free fatty acid-induced gluconeogenesis in hepatocytes. J. Biol. Chem. 2006, 281, 24336− 24344. (12) Nony, P.; Kennett, S.; Glasgow, W.; Olden, K.; Roberts, J. 15SLipoxygenase-2 mediates arachidonic acid-stimulated adhesion of human breast carcinoma cells through the activation of TAK1, MKK6, and p38 MAPK. J. Biol. Chem. 2005, 280, 31413−31419. (13) Kong, J.; Klassen, S.; Rabkin, S. Ceramide activates a mitochondrial p38 mitogen-activated protein kinase: a potential mechanism for loss of mitochondrial transmembrane potential and apoptosis. Mol. Cell. Biochem. 2005, 278, 39−51. (14) Diskin, R.; Engelberg, D.; Livnah, O. A novel lipid binding site formed by the MAP kinase insert in p38 alpha. J. Mol. Biol. 2008, 375, 70−79. (15) Gills, J. J.; Castillo, S. S.; Zhang, C.; Petukhov, P. A.; Memmott, R. M.; Hollingshead, M.; Warfel, N.; Han, J.; Kozikowski, A. P.; Dennis, P. A. Phosphatidylinositol ether lipid analogues that inhibit AKT also independently activate the stress kinase, p38alpha, through MKK3/6-independent and -dependent mechanisms. J. Biol. Chem. 2007, 282, 27020−27029. (16) Tzarum, N.; Eisenberg-Domovich, Y.; Gills, J. J.; Dennis, P. A.; Livnah, O. Lipid molecules induce p38alpha activation via a novel molecular switch. J. Mol. Biol. 2012, 424, 339−353. (17) Perry, J.; Harris, R.; Moiani, D.; Olson, A.; Tainer, J. p38alpha MAP kinase C-terminal domain binding pocket characterized by crystallographic and computational analyses. J. Mol. Biol. 2009, 391, 1−11. (18) Laskowski, R. A. SURFNET: a program for visualizing molecular surfaces, cavities, and intermolecular interactions. J. Mol. Graphics 1995, 13 (323−330), 307−308. (19) Laskowski, R. A.; Luscombe, N. M.; Swindells, M. B.; Thornton, J. M. Protein clefts in molecular recognition and function. Protein Sci. 1996, 5, 2438−2452. (20) Peters, K. P.; Fauck, J.; Frommel, C. The automatic search for ligand binding sites in proteins of known three-dimensional structure using only geometric criteria. J. Mol. Biol. 1996, 256, 201−213. (21) Hendlich, M.; Rippmann, F.; Barnickel, G. LIGSITE: automatic and efficient detection of potential small molecule-binding sites in proteins. J. Mol. Graphics Modell. 1997, 15 (359−363), 389. (22) Liang, J.; Edelsbrunner, H.; Woodward, C. Anatomy of protein pockets and cavities: measurement of binding site geometry and implications for ligand design. Protein Sci. 1998, 7, 1884−1897.

(23) Brady, G. P., Jr.; Stouten, P. F. Fast prediction and visualization of protein binding pockets with PASS. J. Comput.-Aided Mol. Des. 2000, 14, 383−401. (24) Lichtarge, O.; Sowa, M. E. Evolutionary predictions of binding surfaces and interactions. Curr. Opin. Struct. Biol. 2002, 12, 21−27. (25) Joughin, B. A.; Tidor, B.; Yaffe, M. B. A computational method for the analysis and prediction of protein:phosphopeptide-binding sites. Protein Sci. 2005, 14, 131−139. (26) Laurie, A. T.; Jackson, R. M. Q-SiteFinder: an energy-based method for the prediction of protein−ligand binding sites. Bioinformatics 2005, 21, 1908−1916. (27) Coleman, R. G.; Salzberg, A. C.; Cheng, A. C. Structure-based identification of small molecule binding sites using a free energy model. J. Chem. Inf. Model. 2006, 46, 2631−2637. (28) Glaser, F.; Morris, R. J.; Najmanovich, R. J.; Laskowski, R. A.; Thornton, J. M. A method for localizing ligand binding pockets in protein structures. Proteins 2006, 62, 479−488. (29) Huang, B.; Schroeder, M. LIGSITEcsc: predicting ligand binding sites using the Connolly surface and degree of conservation. BMC Struct. Biol. 2006, 6, 19. (30) Chen, B. Y.; Bryant, D. H.; Fofanov, V. Y.; Kristensen, D. M.; Cruess, A. E.; Kimmel, M.; Lichtarge, O.; Kavraki, L. E. Cavity scaling: automated refinement of cavity-aware motifs in protein function prediction. J. Bioinform. Comput. Biol. 2007, 5, 353−382. (31) Cheng, A. C.; Coleman, R. G.; Smyth, K. T.; Cao, Q.; Soulard, P.; Caffrey, D. R.; Salzberg, A. C.; Huang, E. S. Structure-based maximal affinity model predicts small-molecule druggability. Nature Biotechnol. 2007, 25, 71−75. (32) Kahraman, A.; Morris, R. J.; Laskowski, R. A.; Thornton, J. M. Shape variation in protein binding pockets and their ligands. J. Mol. Biol. 2007, 368, 283−301. (33) Kawabata, T.; Go, N. Detection of pockets on protein surfaces using small and large probe spheres to find putative ligand binding sites. Proteins 2007, 68, 516−529. (34) Pettit, F. K.; Bare, E.; Tsai, A.; Bowie, J. U. HotPatch: a statistical approach to finding biologically relevant features on protein surfaces. J. Mol. Biol. 2007, 369, 863−879. (35) Weisel, M.; Proschak, E.; Schneider, G. PocketPicker: analysis of ligand binding-sites with shape descriptors. Chem. Cent. J. 2007, 1, 7. (36) Xie, L.; Bourne, P. E. A robust and efficient algorithm for the shape description of protein structures and its application in predicting ligand binding sites. BMC Bioinformatics 2007, 8 (Suppl 4), S9. (37) Zhong, S.; MacKerell, A. D., Jr. Binding response: a descriptor for selecting ligand binding site on protein surfaces. J. Chem. Inf. Model. 2007, 47, 2303−2315. (38) Brylinski, M.; Skolnick, J. A threading-based method (FINDSITE) for ligand-binding site prediction and functional annotation. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 129−134. (39) Harris, R.; Olson, A. J.; Goodsell, D. S. Automated prediction of ligand-binding sites in proteins. Proteins 2008, 70, 1506−1517. (40) Kalidas, Y.; Chandra, N. PocketDepth: a new depth based algorithm for identification of ligand binding sites in proteins. J. Struct. Biol. 2008, 161, 31−42. (41) Kim, D.; Cho, C. H.; Cho, Y.; Ryu, J.; Bhak, J.; Kim, D. S. Pocket extraction on proteins via the Voronoi diagram of spheres. J. Mol. Graphics Modell. 2008, 26, 1104−1112. (42) Ming, D.; Cohn, J. D.; Wall, M. E. Fast dynamics perturbation analysis for prediction of protein functional sites. BMC Struct. Biol. 2008, 8, 5. (43) Capra, J. A.; Laskowski, R. A.; Thornton, J. M.; Singh, M.; Funkhouser, T. A. Predicting protein ligand binding sites by combining evolutionary sequence conservation and 3D structure. PLoS Comput. Biol. 2009, 5, e1000585. (44) Huang, B. MetaPocket: a meta approach to improve protein ligand binding site prediction. OMICS: J. Integr. Biol. 2009, 13, 325− 330. (45) Coleman, R. G.; Sharp, K. A. Protein pockets: inventory, shape, and comparison. J. Chem. Inf. Model. 2010, 50, 589−603. 9349

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Journal of Medicinal Chemistry

Article

(46) Kawabata, T. Detection of multiscale pockets on protein surfaces using mathematical morphology. Proteins 2010, 78, 1195− 1211. (47) Fukunishi, Y.; Nakamura, H. Prediction of ligand-binding sites of proteins by molecular docking calculation for a random ligand library. Protein Sci. 2011, 20, 95−106. (48) McCoy, M. A.; Wyss, D. F. Spatial localization of ligand binding sites from electron current density surfaces calculated from NMR chemical shift perturbations. J. Am. Chem. Soc. 2002, 124, 11758− 11763. (49) de Groot, B. L.; van Aalten, D. M.; Scheek, R. M.; Amadei, A.; Vriend, G.; Berendsen, H. J. Prediction of protein conformational freedom from distance constraints. Proteins 1997, 29, 240−251. (50) Seeliger, D.; Haas, J.; de Groot, B. L. Geometry-based sampling of conformational transitions in proteins. Structure 2007, 15, 1482− 1492. (51) Eyrisch, S.; Helms, V. Transient pockets on protein surfaces involved in protein−protein interaction. J. Med. Chem. 2007, 50, 3457−3464. (52) Eyrisch, S.; Helms, V. What induces pocket openings on protein surface patches involved in protein−protein interactions? J. Comput.Aided Mol. Des. 2009, 23, 73−86. (53) Seeliger, D.; de Groot, B. L. Conformational transitions upon ligand binding: holo-structure prediction from apo conformations. PLoS Comput. Biol. 2010, 6, e1000634. (54) Hajduk, P. J.; Mack, J. C.; Olejniczak, E. T.; Park, C.; Dandliker, P. J.; Beutel, B. A. SOS-NMR: a saturation transfer NMR-based method for determining the structures of protein−ligand complexes. J. Am. Chem. Soc. 2004, 126, 2390−2398. (55) Orts, J.; Tuma, J.; Reese, M.; Grimm, S. K.; Monecke, P.; Bartoschek, S.; Schiffer, A.; Wendt, K. U.; Griesinger, C.; Carlomagno, T. Crystallography-independent determination of ligand binding modes. Angew. Chem., Int. Ed. Engl. 2008, 47, 7736−7740. (56) Vogtherr, M.; Saxena, K.; Grimme, S.; Betz, M.; Schieborr, U.; Pescatore, B.; Langer, T.; Schwalbe, H. NMR backbone assignment of the mitogen-activated protein (MAP) kinase p38. J. Biomol. NMR 2005, 32, 175. (57) Vallurupalli, P.; Hansen, D.; Kay, L. Structures of invisible, excited protein states by relaxation dispersion NMR spectroscopy. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 11766−11771. (58) Guex, N.; Peitsch, M. SWISS-MODEL and the SwissPdbViewer: an environment for comparative protein modeling. Electrophoresis 1997, 18, 2714−2723. (59) Ghersi, D.; Sanchez, R. EasyMIFS and SiteHound: a toolkit for the identification of ligand-binding sites in protein structures. Bioinformatics 2009, 25, 3185−3186. (60) Hernandez, M.; Ghersi, D.; Sanchez, R. SITEHOUND-web: a server for ligand binding site identification in protein structures. Nucleic Acids Res. 2009, 37, W413−416. (61) Ghersi, D.; Sanchez, R. Improving accuracy and efficiency of blind protein−ligand docking by focusing on predicted binding sites. Proteins 2009, 74, 417−424.

9350

dx.doi.org/10.1021/jm4014357 | J. Med. Chem. 2013, 56, 9342−9350

Rapid identification of ligand-binding sites by using an assignment-free NMR approach.

In this study, we developed an assignment-free approach for rapid identification of ligand-binding sites in target proteins by using NMR. With a sophi...
2MB Sizes 0 Downloads 0 Views