doi:10.1016/j.jmb.2010.02.007 J. Mol. Biol. (2010) 397, 1042–1054 Available online at www.sciencedirect.com Non-additivity of Functional Group Contributions in Protein–Ligand Binding: A Comprehensive Study by Crystallography and Isothermal Titration Calorimetry Bernhard Baum 1 †, Laveena Muley 2 ‡, Michael Smolinski 2 ‡, Andreas Heine 1 †, David Hangauer 2 ⁎‡ and Gerhard Klebe 1 ⁎† 1 Department of Pharmaceutical Chemistry, Philipps-University Marburg, Marbacher Weg 6, 35032 Marburg, Germany 2 Department of Chemistry, University at Buffalo, The State University of New York, Buffalo, NY 14260, USA Received 9 November 2009; received in revised form 21 January 2010; accepted 1 February 2010 Available online 12 February 2010 Additivity of functional group contributions to protein–ligand binding is a very popular concept in medicinal chemistry as the basis of rational design and optimized lead structures. Most of the currently applied scoring functions for docking build on such additivity models. Even though the limitation of this concept is well known, case studies examining in detail why additivity fails at the molecular level are still very scarce. The present study shows, by use of crystal structure analysis and isothermal titration calorimetry for a congeneric series of thrombin inhibitors, that extensive cooperative effects between hydrophobic contacts and hydrogen bond formation are intimately coupled via dynamic properties of the formed complexes. The formation of optimal lipophilic contacts with the surface of the thrombin S3 pocket and the full desolvation of this pocket can conflict with the formation of an optimal hydrogen bond between ligand and protein. The mutual contributions of the competing interactions depend on the size of the ligand hydrophobic substituent and influence the residual mobility of ligand portions at the binding site. Analysis of the individual crystal structures and factorizing the free energy into enthalpy and entropy demonstrates that binding affinity of the ligands results from a mixture of enthalpic contributions from hydrogen bonding and hydrophobic contacts, and entropic considerations involving an increasing loss of residual mobility of the bound ligands. This complex picture of mutually competing and partially compensating enthalpic and entropic effects determines the non-additivity of free energy contributions to ligand binding at the molecular level. © 2010 Elsevier Ltd. All rights reserved. Edited by J. E. Ladbury Keywords: non-additivity of functional group contributions; ligand–protein interactions; crystal structure analysis; isothermal titration calorimetry; thrombin Introduction *Corresponding authors. E-mail addresses: [email protected]; [email protected]. † Contributed to the kinetic and microcalorimetric characterization of the compounds, the crystal structure determination and analysis and the collaborative interpretation of the data. ‡ Contributed the design and synthesis of the compounds (correspondence regarding the design and synthesis should be addressed to D.H.) as well as collaborative interpretation of the data. Usually, the interactions experienced between a protein and a bound small molecule ligand are noncovalent in nature. Deeper insights, which allow better quantification of the energetics of the involved molecular recognition phenomena, would significantly enhance the ability to predict and rationally design small-molecule ligands that bind to specific sites on macromolecules, such as proteins.1 In recent years, substantial advances in the structural determination of protein–ligand complexes have been achieved by crystallography and NMR spectroscopy. Despite this impressive body of 0022-2836/$ - see front matter © 2010 Elsevier Ltd. All rights reserved. 1043 Non-additivity of Functional Group Contributions structural information, it is still a significant challenge to deduce binding affinity from structure. A plethora of computational methods have been developed to predict affinity; however, the performance and reliability of such scoring methods, e.g. as used in docking, is still in need of major improvement.2 What is the reason for this still unmet challenge? The reliable prediction of binding affinity has proven to be very difficult because the protein–ligand binding affinity is not simply a summation of collective weak non-covalent interactions such as multiple hydrogen bonds, burial of hydrophobic surface area, van der Waals interactions or the fixation of molecular degrees of freedom. The binding affinity also involves solvation/desolvation terms and various entropy contributions originating from the protein, ligand and solvent molecules.3,4 Experimental techniques such as isothermal titration calorimetry (ITC) provide the overall thermodynamic profile of protein–ligand interactions, but these data can be difficult to interpret, since they cover the binding event globally. Hence deconvolution of the binding affinity into the individual contributions arising from the interacting partners and the solvent is a very challenging task.5 A significant limitation of many current computational approaches is the assumption that binding affinity, as contributed by each individual interactions to the total binding affinity, is additive and does not depend on the surrounding interactions.6 This rather pragmatic oversimplification ignores the fact that the binding affinity is a Gibbs free energy and hence includes an important entropy term. Even though separation of energy terms into (pairwise) individual contributions is a popular approximation, e.g. usually done in force-fields, this is, in principle, not valid for free energy and thus also for entropy or enthalpy.7,8 Instead, we have to deal with non-additivity or cooperativity, as coined by Williams.9,10 Two ligand fragments mutually linked together into one entity can result in a ligand with a binding affinity greater than the sum of its parts. The latter situation has been defined as positively cooperative for protein–ligand interactions by Williams et al.11,12 and should be distinguished from the original definition given by Monod et al.,13 who used “cooperativity” to describe the cross-talk between multiple binding sites in protein complexes. Conversely, a binding event can be negatively cooperative when the affinity is decreased while two ligand parts experience simultaneous interactions with the protein.11,12 This fundamental phenomenon has thus far been pragmatically ignored in favor of the simpler additivity approach, well in line with frequently applied concepts in chemistry on the basis of simple and convenient additive models. Here, a series of thrombin inhibitors are evaluated in depth in order to probe the non-additivity of noncovalent interactions. In order to unravel the individual interaction contributions, a series of closely related ligands was analyzed systematically so that their properties can be correlated with structural changes. Thrombin was selected as a model system because its binding site is well structured into three aligned subpockets and was used by us earlier for evaluating the binding properties of other systematically varied ligands.14,15 Thrombin is readily accessible in significant quantities, crystal structure analysis is well established, and ITC yields reliable measurements. Furthermore, thrombin has been a prominent drug target for many years and it is therefore well known to many medicinal chemists, modelers and structural biologists. It is suggested that the conclusions drawn from this protein are quite representative for protein– ligand binding in general. Target binding site and compound series As shown in Fig. 1, the designed series 1–4 of ligands firstly differ only in the presence or absence of a terminal amino group (X = H, NH2) competent to form a hydrogen bond with the carbonyl oxygen of Gly216. A second modification concerns the identity of the group that anchors the ligand in the S1 pocket. Either a meta-chlorobenzyl or a benzamidine moiety has been attached to the central L-prolyl portion, linked via an amide bond. The pivotal proline moiety fits well under Tyr60A and Trp60D of the 60-loop, similar to the natural substrate.16 Across the m-chlorobenzyl and the benzamidine series, the P2 portion was kept constant, whereas the third, and most interesting, variation is the increasing size of the hydrophobic side-chains R attached to occupy the virtually hydrophobic S3 pocket. The synthesis and the kinetic characterization of the different series in a photometric assay have been described.17 Overall, the binding data spread from two-digit micromolar to nanomolar binding. As expected, the P1-benzamidine analogs exhibit higher affinity compared to the m-chlorobenzyl derivatives. Furthermore, the attached terminal amino group increases binding affinity by a factor of 10–100, once similarly substituted derivatives are compared directly. In both P1 series (benzamidine and m-chlorobenzyl), increasing the size of the hydrophobic occupant for the S3 pocket from methyl to benzyl enhances the binding constant significantly. To understand the structure–activity relationship observed, we generated putative binding modes by molecular modeling17 and analyzed the hydrophobic surface shared between protein and ligands. The size of the hydrophobic contact surface was plotted against the free energy of binding (ΔGbind) as derived from the measured Ki values.17 Linear correlations were obtained for both the benzamidine and m-chlorobenzyl series with and without a terminal amino group (X = H, NH2), but with significantly different slopes (Fig. 1, data shown for the m-chlorobenzyl series with X = H, NH2). Well beyond experimental error, the series with the amino group exhibited a correlation with a slope almost twice as large as that of the series lacking the NH2 group.17 Undoubtedly, one would expect a significant contribution from the charge-assisted hydrogen bond formed by the most likely positively charged amino group in the 1044 Non-additivity of Functional Group Contributions Fig. 1. Compound series 1–4 studied as thrombin inhibitors. R1, substituent placed into the S1 pocket metachlorobenzyl or para-benzamidine; X = H or NH2; R2, substituent (a-l) placed into the S3 pocket; schematic binding mode and neighboring active site residues (S1, blue; S2, brown; S3/S4, red, non-specific peptide recognition, gray) are indicated. On the lower left the correlation of hydrophobic ligand surface area (in Å2) in contact with the protein as indicated by the model-built protein–ligand complexes is plotted against the Gibbs free energy of binding (in kJ/mol) as derived from the photometrically recorded binding constant Ki. A linear correlation is obtained for both inhibitor series 1 (X = H) and 2 (X = NH2) but the slopes of the lines are different. latter series to the carbonyl group of Gly 216. Assuming independence of the contributions added by the different interactions, one might have expected this to result in a parallel translated shift of the linear correlation towards better binding but not in a different slope. However, as shown in Fig. 1 the slope with X = NH2 is changed significantly, indicating cooperativity between the hydrogen bond and hydrophobic binding in the S3 subsite. This observation of non-additivity of functional group contributions becomes even more apparent when analyzing the binding data of the following four inhibitors from the m-chlorobenzyl series. Replacement of the terminal methyl group at the n-propyl chain (1c) by a phenyl ring (1l) results in an affinity increase of – 3.1 kJ/mol. Attachment of the amino group to the n-propyl derivative (2c) leads to an enhancement of – 9.6 kJ/mol. Surprisingly, the compound exhibiting both modifications (2l) features a binding enhancement of – 18.6 kJ/mol. This value is –5.9 kJ/mol larger than the ΔG value expected from a pure summation of the individual contributions described above. This remarkable contribution of non-additivity can be explained only by a pronounced influence of cooperative effects. These findings stimulated us to embark on a detailed crystal structure analysis and ITC of the structural and thermodynamic features that govern binding of these ligands to thrombin. Results and Discussion Here, we want to discuss the properties of the 12 compounds of the series (1c, 1e, 1l; 2c, 2e, 2l; 3c, 3e, 3l; 4c, 4e, 4l) shown in Fig. 2. All attempts to co-crystallize the ligands 1c, 1e and 1l from the m-chlorobenzyl series lacking the terminal amino group were unsuccessful. This observation is likely to be due to the limited solubility of these inhibitors along with their low binding affinity. Also, successful soaking into thrombin crystals would require rather high concentrations of solvated ligand in aqueous solution. Nevertheless, for nine of these inhibitors (series 2, 3, 4), crystal structure analysis and ITC data could be recorded, and for those of series 1 we will focus on the kinetically determined binding affinities. The data for 2c, 2l, 4c, and 4l were presented earlier.14 Kinetic data The free energy of binding ΔG0 as derived from the kinetic inhibition constants (Table 1) was also analyzed through a thermodynamic cycle (Figure 2). The thermodynamic cycle for the benzamidines (series 3, 4) also shows non-additivity, similar to the m-chlorobenzyl series non-additivity described earlier, though somewhat less pronounced.17 1045 Non-additivity of Functional Group Contributions Fig. 2. Thermodynamic cycle for 12 compounds (1-4, b, d, l) as discussed in this study. All free energies of binding ΔG0 (italics in kJ/mol) to thrombin are derived from the kinetically determined Ki values. The ΔΔG differences between compounds, adjacent in the diagram, are indicated along the connecting arrows. The m-chlorobenzyl series is shown in the back (light blue), and the p-benzamidine series is shown in front (orange). Crystal structures, B-factor analysis An indication of the residual dynamic properties of the inhibitors, while bound to the active site, can be obtained by considering the temperature factors determined for the different thrombin-inhibitor complexes (Table 1). The crystallographically observed temperature or B-factors (closely related to the so-called atomic displacement parameter u; B = 8 π2 u2) reflect the fluctuation of the atoms about their average positions and provide at least qualitative information about the residual mobility of molecular portions in a crystal structure. Their magnitude is reflected in structure refinement. The more an atom, and thus its electron density, is scattered as a result of increased vibrations or scattering over multiple static arrangements, the more out-of-phase, and thus attenuated, are the X-rays from different parts of an atom diffracted. Therefore, at higher diffraction angles, the diffracted intensity falls off more than that calculated for a point atom of given diffraction power. This fall-off is approximated by an exponential function that includes the abovementioned B-factor. As they relate to the electron density in a particular region of a crystal, they will correlate significantly with the population parameters assigned to, for example, a hosted ligand. Furthermore, it is hardly possible to distinguish static from dynamic disorder. Usually, protein crystals are studied at 100 K where motions are reduced significantly. However, it is assumed that flash cooling applied to protein crystals still Table 1. Compilation of Ki values (photometric assay), thermodynamic data from microcalorimetry (ITC, ΔG, ΔH, –TΔS) in Tris buffer, PDB codes, mean B-factor as observed in the different crystal structures (for inhibitor and binding site residues) and the B-factor ratio of inhibitor versus binding site residues Inhibitor 2c 2e 2l 3c 3e 3l 4c 4e 4l Ki (μM) ΔG (kJ/mol) ΔH (kJ/mol) –TΔS (kJ/mol) PDB code Resolution (Å) B Inhibitor (Å2) B Binding site (Å2) B Inhibitor/ B Binding site 6.8 ± 1.6 0.54 ± 0.19 0.18 ± 0.15 5.70 ± 1.20 1.28 ± 0.27 0.75 ± 0.19 0.18 ± 0.03 0.02 ± 0.01 0.004 ± 0.001 - 31.3 ± 1.9 - 35.1 ± 0.3 - 35.4 ± 0.2 - 32.8 ± 0.3 - 37.4 ± 1.2 - 37.8 ± 0.8 - 40.1 ± 0.2 - 42.9 ± 0.2 - 46.1 ± 0.6 - 33.5 ± 0.3 - 42.7 ± 2.9 - 37.1 ± 1.1 - 27.0 ± 1.4 - 22.7 ± 1.7 - 28.6 ± 0.5 - 38.7 ± 0.7 - 34.5 ± 0.1 - 40.1 ± 2.9 2.2 ± 1.9 7.5 ± 3.0 1.7 ± 0.3 - 5.8 ± 1.7 - 14.7 ± 2.9 - 9.3 ± 0.8 - 1.4 ± 0.6 - 8.4 ± 0.3 - 6.1 ± 3.4 2ZFP 2ZGB 2ZC9 2ZI2 2ZIQ 2ZHQ 2ZGX 2ZNK 2ZDA 2.25 1.60 1.58 1.65 1.65 1.96 1.80 1.80 1.73 35.1 21.4 15.7 23.6 21.9 18.5 18.2 17.2 11.4 18.5 19.5 14.5 14.9 16.1 14.7 16.3 14.6 10.9 1.90 1.10 1.08 1.58 1.36 1.26 1.12 1.18 1.06 1046 provides a most likely frozen static picture of protein dynamics at ambient temperature. Nevertheless, some special care is needed in the interpretation of B-factors. We analyzed the Bfactors separately for each inhibitor relative to the neighboring amino acids exposed to the binding site. This calibration was done as an approximate normalization to correct for any superimposed quality difference in the diffraction properties of the various crystal structures being compared. Furthermore, if the ratio of B-factors of ligand atoms relative to those of the binding pocket residues is close to 1, equal thermal motion of ligand and the binding cavity is assumed and full occupancy of the inhibitors in the crystal is suggested. Deviations from this ratio indicate higher or lower mobility, often experienced by parts of a ligand, provided full occupancy is given for the considered ligand. As there might be some concerns about the inherent correlation of B-factors with partial/full occupancy, we studied this aspect in great detail, as described.15 With this analysis we can rule out concerns about partial population of the highly soluble and tightly binding benzamidines. Consequently, we are confident in our interpretation of a high ratio of the B-factors of ligand portions relative to the binding pocket residues as an indicator for enhanced residual mobility of the ligand or its fragments in the active site. Non-additivity of Functional Group Contributions In order to illustrate the magnitude of the thermal parameters determined, a color-coding with respect to the B-factors of active site residues and ligand atoms has been assigned and is shown in Figs. 3 to 6 (blue, low; green, medium; and red, high thermal motion). The protein residues are indicated by the colored solvent-accessible surface and the ligand by a color-coded stick representation. Fig. 3 shows clearly that the benzamidine moiety is firmly fixed in the S1 pocket (low temperature factors) and the proline ring bound below the 60's loop shows rather low thermal motion in the crystal. Much higher Bfactors are observed for the lipophilic side-chain binding in the S3 subpocket of the enzyme, though the difference electron density allows unambiguous location of all ligand atoms during refinement of the structure. The derived crystallographic and thermodynamic data enable us to compare individual inhibitors and to find possible explanations for the phenomenon of cooperativity. ITC data, protonation correction Before the comparison of thermodynamic data it has to be determined if any putatively overlaid protonation effects upon inhibitor binding are occurring. To detect potential changes in protonation states, titration experiments were performed Fig. 3. The crystallographically determined binding mode of 3e in complex with thrombin. Colors are assigned to all atoms according to their temperature factors, from blue (low B-factor) to green to yellow and to red (high B-factor). Protein residues are shown by solvent-accessible surface, ligand atoms in stick representation. The ligand is shown with the Fo – Fc difference electron density (green mesh) at a contour level of 1.5 σ. Non-additivity of Functional Group Contributions 1047 Fig. 4. The crystallographically determined binding mode of 3c and 4c in complex with thrombin. Colors are assigned to all atoms as described for Fig. 3. The ratio of the mean B-factor of ligands with respect to the active-site residues is indicated and the difference in the thermodynamic data, as recorded by ITC, is given. The lengths of the hydrogen bonds between the ligands and Gly216 are given in Å. from different buffer solutions. A negligible heat of protonation has been observed experimentally for 4l.14 As discussed in detail elsewhere,14 this finding results from a rather complex protonation inventory that covers mutually compensating effects arising from deprotonation of His57 and protonation of the primary amino function of the ligand. Because a very similar pattern, and thus very similar pKa values are given with respect to titratable groups, we assume similar behavior, also resulting in buffer independence for the analogous inhibitors 4c and 4e. With some care, similar arguments can be used for 2c, 2e and 2l from the m-chlorobenzyl series because here the mutually compensating functionalities (His57 and amino group) are also present. However, a similar mutual protonation inventory is impossible for the inhibitors 3c, 3e and 3l because they do not feature the primary amino group. In an earlier study, we showed by a direct comparison of 2j and its N-acetylated analog that the release of 0.6 mol protons (most likely originating from His57) occurs upon ligand binding. Therefore, we assume similar heat effects to be superimposed to the heat of binding of 3c, 3e and 3l, as these ligands do not Fig. 5. The crystallographically determined binding mode of 3e and 4e in complex with thrombin. Colors are assigned to all atoms as described for Fig. 3. The ratio of the mean B-factor of ligands with respect to the active site residues is indicated and the difference in the thermodynamic data, as recorded by ITC, is given. The lengths of the hydrogen bonds between the ligands and Gly216 are indicated in Å. 1048 Non-additivity of Functional Group Contributions Fig. 6. The crystallographically determined binding mode of 3l and 4l in complex with thrombin. Colors are assigned to all atoms as described for Fig. 3. The ratio of the mean B-factor of ligands with respect to the active site residues is indicated and the difference in the thermodynamic data, as recorded by ITC, is given. The lengths of the hydrogen bonds between the ligands and Gly216 are indicated in Å. feature a titratable primary amino group that can pick up the protons released from His57. As this release will give rise to an additional heat signal, the experimentally observed heat of binding of 3c, 3e and 3l has to be corrected for the heat of ionization associated with the release of 0.6 mol of protons by His57 and the uptake of the same amount by the buffer (Table 2). The heat of ionization for a histidyl moiety accounts for 30 kJ/mol.18 If His57 releases 0.6 mol of protons upon complex formation, this release should be endothermic and involve about 18 kJ/mol. The experimentally measured heat signal is lowered by this value, and thus the authentic 0 is more enthalpic. In addition, the released ΔHbind protons are picked up by the buffer. Therefore, a correction for an uptake of 0.6 mol of protons with a heat of ionization for Tris buffer of 48 kJ/mol is necessary.19 Consequently, the exothermic proton uptake by the buffer contributes 28.8 kJ/mol to the observed enthalpy. Accordingly, for 3c, 3e and 3l 0 is in total the corrected enthalpy of binding ΔHbind balance 10.8 kJ/mol lower than the experimentally 0 for observed heat signal in Tris buffer. Thus, ΔHbind 3c appears to be – 16.2 kJ/mol, for 3e it accounts for 0 = – 17.8 kJ/ – 11.9 kJ/mol and 3l binds with ΔHbind mol. The entropic contribution to binding –TΔS0 can be calculated as the difference from the observed ΔG0 (Table 2). (indicated by an arrow), which is observed also in the complexes with 3c and 4c. Refinement results in a low temperature factor for this water molecule. With respect to the ligands, the introduction of the amino group (from 3c to 4c) leads to a significant decrease in the B-factor ratio from 1.58 to 1.12. Obviously, 4c loses some part of its mobility indicated by thermal motion compared to 3c. It can be assumed that an entropically beneficial binding event correlates with high residual mobility of the inhibitor in the binding site. This residual mobility has a significant influence on the entropy component reflected in the total free energy of binding. The observed fixation of 4c compared to 3c corresponds to an entropic disadvantage of –TΔΔS = 15.2 kJ/mol of the binding process. It is overcompensated by an enthalpic gain of ΔΔH = –22.5 kJ/mol occurring upon the formation of an additional, possibly charge-assisted hydrogen bond between the amino function of 4c and the carbonyl group of Gly216. The second hydrogen bond, already present in 3c which lacks the amino group, is formed between the ligand carbonyl group and the amide NH of Gly216. Its length is shortened only slightly by the additional charge- Correlation of structural and thermodynamic data of 3c/4c Table 2. Thermodynamic data as corrected for superimposed protonation steps for 3c, 3e and 3l The inhibitors 3c and 4c with a small hydrophobic alkyl R2 side-chain exhibit virtually the same binding mode (Figure 4). As detected in the highresolution crystal structure of uncomplexed thrombin,15 the S3 pocket hosts one water molecule Inhibitor 3c 3e 3l a ΔG0 (kJ / mol) ΔH0bind (kJ/mol) – TΔS0 (kJ/mol) -32.8 ± 0.3 a -37.4 ± 1.2 -37.8 ± 0.8 - 16.2 ± 1.4 - 11.9 ± 1.7 - 17.8 ± 0.5 -16.6 ± 2.9 -25.5 ± 2.9 -20.0 ± 0.8 Standard deviations as for the uncorrected values. Non-additivity of Functional Group Contributions assisted hydrogen bond. In summary, the data for 3c and 4c provide an example of a classical enthalpy/entropy compensation.20,21 Furthermore, differences in the solvation/desolvation properties of 3c and 4c will also have an impact on the observed thermodynamic profiles. Correlation of structural and thermodynamic data of 3e/4e A similar comparison performed for the ligand pair 3e and 4e with a branched medium-size alkyl R2-side-chain shows a slightly different picture (Fig. 5). In the case of 3e, the aliphatic side-chain seems to experience some kind of “conflict of interest”. In order to form an optimal lipophilic contact with the surface of the S3 pocket, and to fully displace the hosted water from the S3 pocket, which is observed in uncomplexed thrombin and the complexes with 3c and 4c, a rupture of an optimal hydrogen bond formed between the ligand carbonyl group and the amide nitrogen of Gly216 is required. This contact is expanded in the complex with 3e to a distance of 3.6 Å (Fig. 5, left). Upon introduction of the amino group in 4e, the complete dual ladder β-sheet-like hydrogen bond network between inhibitor and Gly216 is formed with two shorter distances (3.0 Å and 3.2 Å). However, they prevent the isopropyl side-chain from experiencing optimal lipophilic contacts with the hydrophobic binding pocket. Also in the present case of 3e/4e, the B-factor ratio decreases from 1.36 to 1.18, which most likely results from tighter fixation of 4e in the active site. This results in an entropic disadvantage for the binding of 4e. It is suggested that this effect is slightly more pronounced for the branched medium size than for the short chain alkyl R2 side-chain (–TΔΔS: 3e/4e = 17.1 versus 3c/4c = 15.2 kJ/mol). Possibly, the larger branched group has to abandon more degrees of freedom than the smaller alkyl substituent. Correlation of structural and thermodynamic data of 3l/4l A third ligand pair that was evaluated is 3l and 4l exhibiting a large aromatic R 2 side-chain. The strongest cooperative effects have been observed for this pair. In the present case, the introduction of the amino group results in hardly any differences of the crystallographically observed binding modes, except the additionally formed charge-assisted hydrogen bond (3.1 Å) and a slight reduction of the adjacent hydrogen bond (distance 3.3 Å versus 2.8 Å) between the ligand carbonyl and the nitrogen of Gly216. Again, we observe a significant decrease in the B-factor ratio from 1.26 to 1.06 (Figure 6), indicating that an enthalpy-/entropy compensation is in operation in the present complexes. Interestingly enough, the entropic penalty of – TΔΔS = 13.9 kJ/mol for the ligand pair 3l/4l is slightly lower than in the examples with the aliphatic side-chains (3c/4c and 3e/4e). Presumably, the 1049 large benzyl side-chain is already rather tightly fixed via hydrophobic contacts to the S3 pocket in 3l, even in the absence of the additional amino group of 4l. This view is supported by an almost identical binding mode exhibited by both ligands, which suggests that the introduced charge-assisted hydrogen bond has less of an influence on the differences in the residual mobility of the respective complexes. Thermodynamic cycle of the benzamidine series 3 and 4 Figure 7 shows the thermodynamic cycle for the benzamidine derivatives with and without a terminal amino group. In all examples the introduction of the NH2-group (3c/4c, 3e/4e, 3l/4l) enhances binding affinity by about – 7 kJ/mol. The introduced hydrogen bond reveals a strong enthalpic signal (– 22 kJ/mol) which is largely compensated by an entropic disadvantage of about – 15 kJ/mol. Interestingly, the enthalpic contribution remains virtually unchanged across the series, whereas the entropic contribution changes appear to be less uniform. As discussed, this observation is presumably explained by the deviating residual mobility and number of degrees of freedom experienced by the P3 substituents of different size in the ligands. In both series replacing short-chain by branched medium-size alkyl and large aromatic R2 side-chain reveals small increases in ΔG of varying amounts (– 4.6/– 2.8 and – 0.4/– 3.2 kJ/mol; Fig. 7). More interesting is the fact that these ΔΔG changes factorize into much larger absolute contributions of enthalpy and entropy across the series, which indicates pronounced mutual enthalpy/entropy compensation. Remarkably, this enthalpy/entropy compensation occurs with reversed sign going from short-chain to branched medium-size alkyl R2 sidechain and from branched medium-size alkyl to large aromatic R2 side-chain. This observation finds an explanation by considering the molecular detail of the binding mechanism. The short chain to branched medium-size alkyl replacement (3c→3e and 4c→4e) involves the release of a water molecule from the S3 pocket, an entropically favorable and enthalpically disfavorable process. 22 Subsequent substitution of the branched alkyl group by the larger aromatic moiety (3e→3l and 4e→4l) is enthalpically beneficial but entropically disfavored. The larger substituent experiences much better van der Waals contacts with the protein, however, at a cost of a significant loss of its residual mobility. From an entropic point of view the latter effect is disfavorable. The stepwise optimizing of interaction geometry by increasingly sacrificing ligand mobility is best seen in the case of 3e. The branched alkyl group experiences elevated mobility. Presumably, this ligand tries to find a compromise between satisfactory burial of its hydrophobic surface and formation of a reasonably short hydrogen bond. MD simulations performed on 1d, 2d, 1l and 2l, described elsewhere,17 are in full agreement with this hypothesis. Furthermore, in an earlier study we 1050 Non-additivity of Functional Group Contributions Fig. 7. The thermodynamic cycle for six p-benzamidine derivatives 3c, 3e, 3l and 4c, 4e, 4l. All binding data are taken from ITC experiments, ΔG0 (in italics), the relative differences (ΔΔ) between compounds, adjacent in the diagram, are indicated along the connecting arrows: ΔΔG0, black; ΔΔH0, blue;–TΔΔS0, red. All values are given in kJ/mol. experienced a similar behavior for a ligand with the same scaffold bearing a P3 cyclohexylamino group.23 The binding of this ligand is also governed by a dynamic balance between formation of a hydrogen bond to Gly216CO or, upon rupture of this bond, efficient occupation of the S3 pocket. The inventory between residual mobility, optimization of intermolecular contact geometry and partial desolvation is responsible for the observed cooperativity or non-additivity of different functional group contributions in the present compound series. To a substantial degree, the various contributions are leveled out by enthalpy/entropy compensation. Obviously, a ligand that is already firmly fixed to its receptor site has fewer degrees of freedom (and thus less residual entropy) to lose upon introduction of additional functional groups that will further amplify the attractive interactions to the host protein. In the present case, the mutual affinity-enhancing interactions are of different natures (hydrogen bonding or lipophilic contacts), but we can expect that they reveal an equivalent influence once they constrain the ligand degrees of freedom. This effect, in another context, has been described as the zippertype mechanism, and can be traced in detail in our example. It provides instructive insights into the fine details of molecular recognition. Summary and Conclusions The present systematic study of ligand binding to thrombin gives a structural and thermodynamic explanation for the phenomenon of cooperativity or non-additivity of functional group contributions. Four series of closely related ligands (1-4, Fig. 1) have been investigated. These series exhibit at P1 a m-chlorobenzyl or benzamidine moiety. Next to the P3 position, they either present an amino group capable of forming a charge-assisted hydrogen bond to Gly216CO or this group is lacking. For the S3 pocket, a broad range of hydrophobic substituents of significant size difference has been tested. A convincing linear relationship is obtained by correlation of hydrophobic surface areas buried in the model-built complexes upon binding with free energy changes. This finding supports the idea that most of the currently used scoring functions evaluate such a burial term by awarding a constant amount of binding affinity per Å2 of hydrophobic contact area. A derived adjustable parameter is utilized to correlate hydrophobic contact surface area to ΔG changes either to transfer scoring functions among different protein targets or similarly in setting up a QSAR analysis. However, most striking in the present case is the fact that hydrophobic contact surface contributions differ depending upon the presence or the absence of the amino group. Considered in detail, this observation contradicts simple additivity of functional group contributions and supports pronounced cooperative effects. Analysis of the individual crystal structures and factorizing the free energy into enthalpy and entropy demonstrates that the binding affinity of the ligands results from a mixture of enthalpic contributions from hydrogen bonding and hydrophobic contacts, and entropic considerations involving an increasing loss of residual mobility of the bound ligands. These conclusions match with a recent comprehensive 1051 Non-additivity of Functional Group Contributions database survey.24 However, even within a series of ligands that agrees well with the above-mentioned linear burial versus free energy correlation, the derived additivity can be artificial. This fact finds an explanation in differences in the residual mobility of bound ligand portions. Therefore, a property such as the degree of surface burial calculated for the static picture of a protein–ligand complex must be misleading and does not consider dynamic properties. This fact becomes obvious regarding the reversal enthalpy/entropy compensation and the pronounced cooperative effects within the series. As a matter of fact, many molecular phenomena determining ligand binding are governed by pronounced enthalpy/entropy compensation. Thus, they do not become transparent in ΔG and will not be reflected in a free energy correlation. Therefore, even incorrectly modeled binding modes can lead to correct estimations of binding affinity from structure. Supposedly, this is the explanation of why empirical scoring functions still work reasonably well. However, if the properties of one member of the series deviate significantly from pronounced enthalpy/entropy compensation, this complex will fall out of the correlation based on a too simply modeled (static) geometry. In conclusion, if we want to have scoring functions of greater reliability we have to incorporate descriptors that model residual mobility and cooperative effects (or appropriate entropic terms) correctly, apart from correct solvation/desolvation phenomena. Materials and Methods Kinetic inhibition The kinetic inhibition of human α-thrombin (isolated from Beriplast®, CSL Behring, Marburg, Germany) was determined by monitoring absorbance at 405 nm using the chromogenic substrate Pefachrom tPa (LoxoGmbH, Dossenheim, Germany) as described,25 and applying the following conditions: 50 mM Tris–HCl, pH 7.4, 154 mM NaCl, 5% (v/v) DMSO, 0.1% (w/v) PEG 8000 at 25 °C using different concentrations of substrate (182, 91, and 45 μM) and inhibitor (36.4, 27.3, 18.2, and 9.1 μM for the weakest inhibitor and 3.6, 2.7, 1.8, and 0.9 nM for the tightest binder). The activity of thrombin was adjusted by diluting (∼ 1:300) 50 μg/ml thrombin solution with 154 mM NaCl until linear conversion of the substrate could be detected over 5 min in an appropriate absorption window (0.2∼0.8). The assay was stopped after 3 min by the addition of concentrated acetic acid and absorption in each well was corrected for the blank value. The Ki values (n ≥ 3) were determined as described.26 was freshly prepared for each experiment by dialysis of a thrombin sample in the buffer used for titration experiments and adjustment of DMSO final concentration to 2.5%. ITC measurements were routinely done at 25 °C in 50 mM Tris–HCl, pH 7.8, 100 mM NaCl, 2.5% DMSO, 0.1% PEG 8000. Inhibitor solutions (0.03–0.25 mM, depending on the individual ligand) were degassed for ∼ 10 min immediately before use and titrated into the stirred cell (1.3513 ml) containing a thrombin solution (0.006– 0.02 mM) once the baseline was stable. The injection sequence consisted of an initial injection of 1.5 μl of ligand solution to prevent artifacts arising from the filling of the syringe (not used in data fitting), followed by injection of 7–12 μL each at 300 s intervals until complete saturation of the enzyme binding sites was achieved. Raw data were collected and the area under each peak was integrated, followed by correction for heats of dilution and mixing by subtracting the final baseline consisting of small peaks of the same size to zero. The data were analyzed with ORIGIN Software (Microcal Inc.) by fitting a single-site binding isotherm,27 which yields ΔH0bind (enthalpy of binding) and KD (dissociation constant). Measurements were done in at least triplicate; KD was reproducible to within 10% and ΔH0bind was reproducible to within 5%. The buffer dependence of ΔH0bind was tested earlier for the present compound series.14 Crystallisation and soaking Human α-thrombin was dissolved in crystallisation buffer (100 mM sodium phosphate, 350 mM NaCl, 10 mM benzamidine, pH 7.5) and dialyzed against the same buffer overnight. The sample was concentrated to 5 mg/mL and 200 μL of this solution were mixed with 20 μL of an aqueous solution (500 mg/mL) of Hirugen (Bachem, Bubendorf, Switzerland). After incubation for 10 h at 4 °C, crystallisation was carried out at 4 °C by the vapour-diffusion, hanging-drop method using 28% PEG 8000 and 100 mM sodium phosphate (pH 7.5). Microseeding was done after equilibration for 16 h. For soaking, a 1:3 (v/v) mixture of 10–40 mM inhibitors in DMSO and crystallisation buffer was prepared, in which medium-sized crystals without visible imperfections were soaked for 6–24 h. Data collection and processing Crystals were prepared for data collection at 110 K using 20% (w/v) glycerol in crystallisation buffer as a cryoprotectant solution. The datasets for 2e, 3c, 3e, 4e and 4l were collected with synchrotron radiation at BESSY (Berlin, Germany) beamlines 14.1 on the Marmosaic 225 mm CCD and 14.2 on the Mar CCD 165 mm detector. Diffraction data for 2l were collected at DESY (Hamburg, Germany) beamline X13 on the Mar CCD 165 mm detector. Diffraction data for 2c, 3l and 4c were collected on RIGAKU copper rotating anode (RU300) at 50 kV, 90 mA using an R-AXIS IV++ image plate system. Data processing and scaling were performed using the HKL2000 package.28 ITC experiments Structure determination and refinement The ITC experiments were done with an MCS titration calorimeter (Microcal, Inc., Northampton, MA).27 Concentrations of inhibitor stock solutions in DMSO were determined by the weight of the corresponding hydrochlorides. The final concentration was achieved by diluting 1:40 in the experimental buffer. Protein solution The coordinates of human thrombin (PDB code 1H8D)29 were used for initial rigid body refinement of the protein molecules followed by repeated cycles of conjugate gradient energy minimization, simulated annealing and B-factor refinement using the CNS program package.30 1052 Table 3. Data collection and refinement statistics for the nine complex structures determined in this study Complex PDB entry A. Data collection and processing No. crystals used Wavelength (Å) Space group Unit cell parameters a, b, c (Å) β (°) Matthews coefficient (Å3/Da) Solvent content (%) B. Diffraction dataa Resolution range (Å) a THR-2e 2ZGB THR-2l 2ZC9 THR-3c 2ZI2 THR-3e 2ZIQ THR-3l 2ZHQ THR-4c 2ZGX THR-4e 2ZNK THR-4l 2ZDA 1 1.54178 C2 1 0.91841 C2 1 0.97803 C2 1 0.91841 C2 1 0.91841 C2 1 1.54178 C2 1 1.54178 C2 1 0.91841 C2 1 0.91841 C2 70.5, 71.3, 72.9 100.6 2.3 48 69.8, 71.4, 72.6 100.3 2.3 46 70.4, 71.3, 72.9 100.7 2.3 47 70.5, 71.6, 72.6 100.6 2.3 47 70.2, 71.6, 72.3 100.3 2.3 46 70.1, 71.4, 72.9 100.8 2.3 47 70.4, 71.5, 72.5 100.7 2.3 47 70.2, 71.6, 72.5 100.5 2.3 46 70.3, 71.5, 72.4 100.5 2.3 46 20 – 2.25 (2.29 – 2.25) 15,884 (786) 8.2 (31.7) 93.8 (90.3) 2.2 (2.0) 12.1 (2.8) 20 – 1.60 (1.63 – 1.60) 45,709 (2273) 3.5 (31.9) 99.5 (100) 3.2 (3.1) 29.5 (3.9) 30 – 1.58 (1.61 – 1.58) 46,641 (1725) 3.4 (23.5) 95.9 (71.1) 2.2 (1.7) 24.9 (3.4) 20 – 1.65 (1.68 – 1.65) 41,212 (1681) 3.2 (23.8) 96.5 (77.8) 2.0 (1.7) 20.8 (2.5) 20 – 1.65 (1.68 – 1.65) 41,871 (1963) 4.2 (23.3) 98.6 (91.0) 2.8 (2.3) 24.2 (3.5) 20 – 1.96 (2.01 – 1.96) 24,126 (1668) 7.2 (25.7) 93.7 (96.7) 2.8 (2.7) 12.6 (2.8) 20 – 1.80 (1.83 – 1.80) 32,453 (1558) 4.8 (32.6) 99.1 (96.2) 2.5 (2.4) 20.7 (2.5) 20 – 1.80 (1.83 – 1.80) 32,354 (1535) 2.8 (9.1) 98.8 (95.1) 2.7 (2.4) 29.4 (6.0) 20 – 1.73 (1.76 – 1.73) 35,607 (1692) 9.0 (25.2) 97.6 (93.1) 2.9 (2.2) 14.8 (3.3) 10–2.25 14 832/709 10–1.60 43 780/2187 10–1.58 42 070/2197 10–1.65 39 496/1 973 10–1.65 40 527/2 022 10–1.96 22,286/1070 10–1.80 31,065/1536 10–1.80 31,481/1536 10–1.73 32,402/1691 17.6/29.8 19.3/23.1 18.0/23.3 18.2/22.5 18.1/22.5 19.3/27.2 17.8/22.9 17.9/22.9 17.2/22.8 15.4/26.3 18.6/22.2 17.1/22.0 17.1/21.2 17.3/21.3 17.7/25.4 16.6/21.5 17.4/21.9 16.9/21.4 251 2 22 142 251 2 24 185 251 2 27 271 252 2 23 192 252 2 25 179 251 2 28 154 250 2 24 194 251 2 26 220 252 2 29 268 0.005 2.0 0.009 2.5 0.010 2.6 0.009 2.6 0.009 2.5 0.006 2.1 0.008 2.4 0.008 2.5 0.008 2.5 86.0 84.6 86.1 85.5 84.0 85.2 86.5 85.2 86.6 13.0 14.9 13.4 14.0 15.6 14.8 13.0 14.3 13.4 1.0 0.5 0.5 0.5 0.4 - 0.5 0.4 - 23.8 23.0 28.1 23.2 21.4 29.9 20.7 15.7 31.2 19.5 23.6 27.7 21.7 21.9 28.6 20.1 18.5 25.7 21.6 18.2 27.9 19.6 17.2 27.4 17.1 11.4 26.7 Numbers in parentheses characterize the highest resolution shell. Non-additivity of Functional Group Contributions Unique reflections R(I)sym (%) Completeness (%) Redundancy I/σ(I) C. Refinement Resolution range (Å) Reflections used in refinement (work/free) Final R value for all reflections (work/free) [%] Final R value for reflections with F N 4 σ F (work/free) (%) Protein residues Sodium ions Inhibitor atoms Water molecules RMSD from ideal Bond lengths (Å) Bond angles (°) Ramachandran plot Residues in most-favoured regions (%) Residues in additionally allowed regions (%) Residues in generously allowed regions (%) Mean B-factor Protein (Å2) Inhibitor (Å2) Water molecules (Å2) THR-2c 2ZFP Non-additivity of Functional Group Contributions Refinement at later stages was done with the program SHELXL.31 Here, at least 20 cycles of conjugate gradient minimization were done with default restraints on bonding geometry and B-values; 5% of all data were used for the Rfree calculation. Amino acid side-chains were fitted into σ-weighted 2Fo – Fc and Fo – Fc difference electron density maps using Coot.32 After the first refinement cycle, water molecules and subsequently ions and ligand were located in the electron density and added to the model. Restraints were applied to bond lengths and angles, chiral volume, planarity of aromatic rings and van der Waals contacts. Multiple side-chain conformations were built in case an appropriate electron density was observed and maintained during the refinement, and if the minor populated side-chain showed at least 10% occupancy. During the last refinement cycles, riding hydrogen atoms were introduced without additional parameters. The final models were validated using PROCHECK.33 Data collection, unit cell parameters and refinement statistics are given in Table 3. Analysis of temperature factors was done with Moleman, 34 distances were measured using SYBYL 8.0 (Tripos Inc., St. Louis, MO). Figures were prepared using Isis Draw (MDL, San Leandro, CA) and Pymol 0.99 (DeLano Scientific, Palo Alto, CA). Coordinates and structure factors of all X-ray structures have been deposited in the Protein Data Bank and the accession codes are given in Table 3. Protein Data Bank accession numbers Coordinates and structure factors have been deposited in the Protein Data Bank with the following accession numbers: THR-2d complex 2ZGB; THR-3c complex 2ZI2; THR-3l complex 2ZHQ; THR-4e complex 2ZNK. Acknowledgements We kindly acknowledge CSL Behring, Marburg, for supplying us with generous amounts of human thrombin from the production of Beriplast®. We thank the beamline support staff at DESY (Hamburg, Germany) and BESSY (Berlin, Germany) for their advice during data collection and the BMBF (support code 05ES3XBA/5) for generously supporting travel to BESSY/Berlin. Author Contributions. L.M., M.S. and D.H. contributed the design and synthesis of the compounds (correspondence regarding the design and synthesis should be addressed to D.H.) as well as collaborative interpretation of the data. B.B., A.H. and G.K. contributed to the kinetic and microcalorimetric characterization of the compounds, the crystal structure determination and analysis and the collaborative interpretation of the data. Correspondence regarding these aspects should be addressed to G.K. References 1. Whitesides, G. M. & Krishnamurthy, V. M. (2005). Designing ligands to bind proteins. Q. Rev. Biophys. 38, 385–395. 1053 2. Warren, G. L., Andrews, C. W., Capelli, A. M., Clarke, B., LaLonde, J., Lambert, M. H. et al. (2006). A critical assessment of docking programs and scoring functions. J. Med. Chem. 49, 5912–5931. 3. Davies, T. G., Tame, J. R. & Hubbard, R. E. (2000). Generating consistent sets of thermodynamic and structural data for analysis of protein-ligand interactions. Persp. Drug Discov. Des. 20, 29–42. 4. Ladbury, J. E., Klebe, G. & Freire, E. (2010). Adding calorimetric data to decision making in lead discovery: a hot tip. Nat. Rev. Drug Discov. 9, 23–27. 5. Barratt, E., Bronowska, A., Vondrasek, J., Cerny, J., Bingham, R., Phillips, S. & Homans, S. W. (2006). Thermodynamic penalty arising from burial of a ligand polar group within a hydrophobic pocket of a protein receptor. J. Mol. Biol. 362, 994–1003. 6. Dill, K. A., Truskett, T. M., Vlachy, V. & Hribar-Lee, B. (2005). Modeling water, the hydrophobic effect, and ion solvation. Annu. Rev. Biophys. Biomol. Struct. 34, 173–199. 7. Brady, G. P. & Sharp, K. A. (1997). Entropy in protein folding and in protein-protein interactions. Curr. Opin. Struct. Biol. 7, 215–221. 8. Mark, A. E. & van Gunsteren, W. F. (1994). Decomposition of the free energy of a system in terms of specific interactions. Implications for theoretical and experimental studies. J. Mol. Biol. 240, 167–176. 9. Williams, D. H., Searle, M. S., Mackay, J. P., Gerhard, U. & Maplestone, R. A. (1993). Toward an estimation of binding constants in aqueous solution: studies of associations of vancomycin group antibiotics. Proc. Natl Acad. Sci. USA, 90, 1172–1178. 10. Williams, D. H., Stephens, E., O'Brien, D. P. & Zhou, M. (2004). Understanding noncovalent interactions: ligand binding energy and catalytic efficiency from ligandinduced reductions in motion within receptors and enzymes. Angew. Chem. Int. Ed. Engl. 43, 6596–6616. 11. Williams, D. H., Stephens, E. & Zhou, M. (2003). Ligand binding energy and catalytic efficiency from improved packing within receptors and enzymes. J. Mol. Biol. 329, 389–399. 12. Williams, D. H., Stephens, E. & Zhou, M. (2006). Ligand binding energy and enzyme efficiency from reductions in protein dynamics. J. Mol. Biol. 355, 760–767. 13. Monod, J., Wyman, J. & Changeux, J.-P. (1965). On the nature of allosteric transitions: a plausible model. J. Mol. Biol. 12, 88–118. 14. Baum, B., Muley, L., Heine, A., Smolinski, M., Hangauer, D. & Klebe, G. (2009). Think twice: understanding the high potency of bis(phenyl)methane inhibitors of thrombin. J. Mol. Biol. 391, 552–564. 15. Baum, B., Mohamed, M., Zayed, M., Gerlach, C., Heine, A., Hangauer, D. & Klebe, G. (2009). More than a simple lipophilic contact: a detailed thermodynamic analysis of nonbasic residues in the S1 pocket of thrombin. J. Mol. Biol. 390, 56–69. 16. Stubbs, M. T., Oschkinat, H., Mayr, I., Huber, R., Angliker, H., Stone, S. R. & Bode, W. (1992). The interaction of thrombin with fibrinogen. A structural basis for its specificity. Eur. J. Biochem. 206, 187–195. 17. Muley, L., Baum, B., Smolinski, M., Freindorf, M., Heine, A., Klebe, G. & Hangauer, D. (2010). Enhancement of hydrophobic interactions and hydrogen bond strength by cooperativity: Synthesis, modelling and molecular dynamics simulations of a congeneric series of thrombin inhibitors. J. Med. Chem. In the press. doi: 10.1021/jm9016416. 1054 18. LSBU. http://www.lsbu.ac.uk/biology/enztech/ph. html. 19. Steuber, H., Czodrowski, P., Sotriffer, C. A. & Klebe, G. (2007). Tracing changes in protonation: a prerequisite to factorize thermodynamic data of inhibitor binding to aldose reductase. J. Mol. Biol. 373, 1305–1320. 20. Dunitz, J. D. (1995). Win some, lose some: Enthalpyentropy compensation in weak intermolecular interactions. Chem. Biol. 2, 709–712. 21. Williams, D. H., Zhou, D. & Stephens, E. (2006). Ligand binding energy and enzyme efficiency from reductions in protein dynamics. J. Mol. Biol. 355, 760–767. 22. Dunitz, J. D. (1994). The entropic cost of bound water in crystals and biomolecules. Science, 264, 670. 23. Gerlach, C., Smolinski, M., Steuber, H., Sotriffer, C. A., Heine, A., Hangauer, D. G. & Klebe, G. (2007). Thermodynamic inhibition profile of a cyclopentyl and a cyclohexyl derivative towards thrombin: the same but for different reasons. Angew. Chem. Int. Ed. Engl. 46, 8511–8514. 24. Olsson, T. S. G., Williams, M. A., Pitt, W. R. & Ladbury, J. E. (2008). The thermodynamics of protein– ligand interaction and solvation: Insights for ligand design. J. Mol. Biol. 384, 1002–1017. 25. Sturzebecher, J., Sturzebecher, U., Vieweg, H., Wagner, G., Hauptmann, J. & Markwardt, F. (1989). Synthetic inhibitors of bovine factor Xa and thrombin comparison of their anticoagulant efficiency. Thromb. Res. 54, 245–252. 26. Dixon, M. (1972). The graphical determination of Km and Ki. Biochem. J. 129, 197–202. Non-additivity of Functional Group Contributions 27. Wiseman, T., Williston, S., Brandts, J. F. & Lin, L. N. (1989). Rapid measurement of binding constants and heats of binding using a new titration calorimeter. Anal. Biochem. 179, 131–137. 28. Otwinowski, Z. & Minor, W. (1997). Processing of Xray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307–326. 29. Skordalakes, E., Dodson, G. G., Green, D. S., Goodwin, C. A., Scully, M. F., Hudson, H. R. et al. (2001). Inhibition of human alpha-thrombin by a phosphonate tripeptide proceeds via a metastable pentacoordinated phosphorus intermediate. J. Mol. Biol. 311, 549–555. 30. Brunger, A. T., Adams, P. D., Clore, G. M., DeLano, W. L., Gros, P., Grosse-Kunstleve, R. W. et al. (1998). Crystallography & NMR system: a new software suite for macromolecular structure determination. Acta Crystallogr. D, 54, 905–921. 31. Sheldrick, G. M. & Schneider, T. R. (1997). SHELXL: high-resolution refinement. Methods Enzymol. 277B, 319–343. 32. Emsley, P. & Cowtan, K. (2004). Coot: model-building tools for molecular graphics. Acta Crystallogr. D, 60, 2126–2132. 33. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & Thornton, J. M. (1993). PROCHECK: a program to check the stereochemical quality of protein structures. J. Appl. Crystallogr. 26, 283–291. 34. Kleywegt, G. J., Zou, J. Y., Kjeldgaard, M. & Jones, T. A. (2001). Around O. In International Tables for Crystallography (Rossmann, M. G. & Arnold, E., eds), vol. F, pp. 353–356, 366–367. Kluwer Academic Publishers, Dordrecht.
© Copyright 2026 Paperzz