Article pubs.acs.org/JACS
How Nothing Boosts Affinity: Hydrophobic Ligand Binding to the Virtually Vacated S1′ Pocket of Thermolysin Stefan G. Krimmer,† Jonathan Cramer,† Johannes Schiebel, Andreas Heine, and Gerhard Klebe* Department of Pharmaceutical Chemistry, University of Marburg, Marbacher Weg 6, 35032 Marburg, Germany S Supporting Information *
ABSTRACT: We investigated the hydration state of the deep, well-accessible hydrophobic S1′ specificity pocket of the metalloprotease thermolysin with purposefully designed ligands using high-resolution crystallography and isothermal titration calorimetry. The S1′ pocket is known to recognize selectively a very stringent set of aliphatic side chains such as valine, leucine, and isoleucine of putative substrates. We engineered a weak-binding ligand covering the active site of the protease without addressing the S1′ pocket, thus transforming it into an enclosed cavity. Its sustained accessibility could be proved by accommodating noble gas atoms into the pocket in the crystalline state. The topology and electron content of the enclosed pocket with a volume of 141 Å3 were analyzed using an experimental MAD-phased electron density map that was calibrated to an absolute electron number scale, enabling access to the total electron content within the cavity. Our analysis indicates that the S1′ pocket is virtually vacated, thus free of any water molecules. The thermodynamic signature of the reduction of the void within the pocket by growing aliphatic P1′ substituents (H, Me, iPr, iBu) reveals a dramatic, enthalpy-dominated gain in free energy of binding resulting in a factor of 41 000 in Kd for the H-to-iBu transformation. Substituents placing polar decoy groups into the pocket to capture putatively present water molecules could not collect any evidence for a bound solvent molecule.
■
INTRODUCTION The displacement of water molecules from hydrophobic surfaces and pockets is considered to be one of the main driving forces behind protein folding, biochemical recognition, and the interaction of drug molecules with their target proteins.1,2 The energetically favorable association of nonpolar interaction partners from aqueous solution is generally referred to as the “hydrophobic effect”. Originally, this effect was attributed to the antipathy between hydrocarbons and water.3 The strong self-attraction of water was made responsible for the favorable free energy associated with the transfer of hydrocarbons from the water to the pure hydrocarbon phase. Moreover, unusually large heat capacity contributions were found for their separation from the water phase. These contributions turned out to be directly proportional to the boundary surface area and thus the number of solvated water molecules in the first solvation shell of the nonpolar solute.4 It is commonly believed that the water molecules solvating the hydrophobic surface of a solute have a lower entropy content compared to solvent molecules in the bulk phase, at least at temperatures close to ambient.5,6 In consequence, the displacement of these orientationally restrained water molecules into bulk by an apolar solute increases their degrees of freedom and gives rise to an entropically favorable contribution to the binding event.2,7,8 Over the past, numerous, sometimes © 2017 American Chemical Society
controversial models have been suggested to link the observed macroscopic thermodynamic signature with the molecular origin of the hydrophobic effect.8 However, more and more studies give merit to the hypothesis that the general assumption underlying the classical interpretation of the “hydrophobic effect” is not sufficient in all cases. Instead, the molecular origins of solvent effects on binding events seem to be more complicated and reliant on the specific topology of the apolar surface patches of both binding partners.9−16 In multiple instances it has been reported that the binding of an apolar ligand to a hydrophobic protein pocket can also result in an enthalpically favorable binding signature. A possible explanation for these observations requires the presence of disordered water molecules in the respective protein pockets, whose entropy content does not significantly differ from the bulk phase. The thermodynamics of ligand binding to such poorly hydrated pockets is consequently governed by the enthalpically favorable formation of new solvent−solvent interactions. Thus, in order to understand solvent effects in protein−ligand interactions, it is important to realize that these can occur with very different extent leading to deviating thermodynamic signatures depending on the given system. Received: May 16, 2017 Published: July 11, 2017 10419
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society
design, but also suggest that vacated protein pockets have evolved in nature to enhance the specificity of enzymes toward their natural substrates.
As an extension to this concept of nonclassical hydrophobic effects, it has been proposed that hydrophobic protein cavities may even remain completely unsolvated in the unbound state.17−22 In such cases, the energetic penalty for maintaining a vacated cavity within a folded protein is counterbalanced by an even more unfavorable energy of hydration. Consequently, a ligand that is able to address such a vacuous protein site must experience an enormous increase in potency. The sparse experimental evidence collected so far suggests that unsolvated protein pockets can indeed occur in biological systems. While the debate about the hydration state of a hydrophobic cavity in human interleukin-1β is not completely resolved,23−26 it has been shown by magnetic resonance that bovine β-lactoglobulin contains a vacated apolar pocket.21 The hydrophobic calyx of 315 Å3, which is known to bind large apolar molecules like palmitate, was found to be completely empty in magnetic relaxation dispersion experiments. Molecular dynamics simulations support this observation. For an engineered apolar cavity comprising a volume of 160 Å3 in a T4 lysozyme variant, it was found that water only enters the void at elevated pressures, which are associated with protein unfolding.22 Molecular dynamics simulations also suggest that protein cavities can undergo total or partial dewetting transitions in their active site.18−20 An intriguing case is highlighted by a simulation of cyclooxygenase 2.20 The narrow, apolar funnel, where the natural ligand arachidonic acid can bind, was found to remain unsolvated throughout the duration of the simulation. After artificial hydration of this area, bound water molecules evacuated the protein pocket within 100 ps of simulation time. Some of our previous experiments suggested that the deep, hydrophobic S1′ specificity pocket of the metalloprotease thermolysin (TLN) may only be poorly hydrated in the presence of a ligand.27 In the current contribution, we investigate the hydration state of this cavity in detail using an entirely experimental approach comprising ligand design, X-ray crystallography, and isothermal titration calorimetry (ITC). In particular, we engineered a weak-binding ligand that covers the active site of the protease without addressing the S1′ pocket, thus transforming it into an enclosed cavity characterized by a volume of 141 Å3. That the pocket is still accessible, even in the crystalline state, was proven by accommodating noble gas atoms in the pocket. The part of an experimentally phased electron density map on an absolute electron number scale covering the enclosed pocket was then analyzed and correlated to internal reference cavities within the enzyme. Furthermore, the effect of inserting and growing an aliphatic P1′ ligand portion into the pocket was investigated by crystallography and ITC. In addition, substituents with polar decoy groups to capture putatively present water molecules were oriented into the pocket; however, no solvent molecules could be traced even in the presence of such ligands. The reduction of the void within the pocket is accompanied by a dramatic, enthalpydominated gain in free energy of binding leading to an up to 41 000-fold improvement in Kd. Thus, we provide direct experimental evidence for the hypothesis that the S1′ pocket of TLN is vacated in the unbound state. Furthermore, we are able to directly investigate the thermodynamic effects that arise from the accommodation of a hydrophobic ligand moiety in a vacated protein pocket in a setting that is relevant for medicinal chemistry. The results of this study do not only have important implications for the targeted exploitation of similar phenomena in rational drug
■
RESULTS Isothermal Titration Calorimetry. ITC titrations of 2−4 (Figure 1) into TLN were performed in a direct manner
Figure 1. Congeneric series of phosphonamidate TLN ligands substituted with different P1′ groups. (A) Schematic binding mode of the parent ligand scaffold in complex with TLN. Protein residues and the zinc ion forming hydrogen bonds (orange dashed lines) with the parent scaffold are indicated (GOL = glycerol molecule from the cryo buffer bound to TLN). The S1 and S2′ pockets of TLN are wide open and well accessible to water molecules, whereas the S1′ pocket is deep and apolar. (B) P1′ substituents of 1−6 and their thermodynamic binding profiles as determined by global ITC analysis. The thermodynamic profiles, which are shown for 2, 3, and 4, were corrected for the heat of buffer ionization upon complex formation. A reliable determination of the buffer-corrected enthalpy/entropy partitioning of 1, 5, and 6 (displacement titrations) was impossible. Error bars represent the 95.4% confidence interval. Data values are listed in Table S1.
(Figure S1A−C), while ligands 1, 5, and 6 (Figure 1) were analyzed applying a displacement protocol (Figure S1D−F). The thermodynamic profiles of 2−4 were corrected for protontransfer reactions upon complex formation via measurement in several buffers displaying different ionization enthalpies (Figure S2).28 Such corrections were not performed for displacement titrations of 1, 5, and 6, because this would have resulted in 10420
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society
S1′ pocket. The total electron content integrated over the entire volume of the S1′ cavity is 6.6 e−. Reference Cavity 1. Reference cavity 1, shown in Figure 2C, is a highly polar cavity comprising a volume of 59 Å3. It contains three water molecules that establish multiple hydrogen bonds (2.8−3.2 Å) to adjacent polar functional groups of TLN residues. The observed firm density peaks comprise spherical shapes and have been refined to fully occupied water molecules with low B factors of 11−12 Å2. The integrated total electron content within this reference cavity is 19.4 e−. Reference Cavity 2. Reference cavity 2, shown in Figure 2D, surrounds a volume of 93 Å3. In the lower part of the pocket, a strong electron density peak originating from a water molecule (refined B factor of 10 Å2) is detected. This water molecule establishes three hydrogen bonds with Tyr81 (backbone O, 2.9 Å), Arg90 (backbone O, 2.7 Å), and Ser92 (backbone N, 2.8 Å). In contrast, the upper part of the cavity has a highly apolar character. Apart from the described peak assigned to the hydrogen-bonded water molecule, a second, far less intense electron density peak is observed in the center of the large apolar cavity (maximum concentration of 0.62 e−/Å3, Figure 2D). The distance of this peak to that of the already assigned water molecule is 3.2 Å. The integrated total electron content within reference cavity 2 is 13.4 e−. Reference Cavity 3. Reference cavity 3, shown in Figure 2E, is, apart from the S1′ cavity, the largest cavity which remains nonsolvated in the refined structure of TLN-1. Its volume of 16 Å3 is significantly smaller than that of the S1′ cavity. Its character is primarily apolar, and it is flanked by polar atoms of Asn112 (backbone O), Val121 (backbone O), and Gly123 (backbone N). The integrated total electron content within reference cavity 3 is 0.9 e−. Xenon and Krypton Binding to the S1′ Cavity of TLN1. Pressure derivatization of TLN-1 with xenon and krypton was performed at 9 and 5 bar, respectively. The positions of bound xenon and krypton in the crystal structure can be unambiguously identified in the electron density due to their anomalous scattering properties.33 To optimize the anomalous signal of xenon, a wavelength of 1.4 Å ( f″ = 6.3 e−) had been chosen for data collection, resulting in a strong anomalous signal without significantly losing scattering power due to the long wavelength and air absorption of the X-ray beam (Table 3). In the case of krypton, two data sets of the same crystal were collected directly above and below the K absorption edge of krypton to unambiguously identify bound atoms of the noble gas (Table 3). Even though the difference between the wavelengths is only 0.0121 Å, the change of the anomalous signal of krypton is drastic, whereas its mFo−DFc electron density virtually remains unaffected (Figure S8). All data sets of the TLN-1 noble gas derivatives were collected with high redundancy to maximize the accuracy of the anomalous signal (Table 1). For both noble gases, binding was observed in the S1′ cavity (refined occupancy of xenon: 17%, krypton: 8%) as well as in reference cavity 3 (refined occupancy xenon: 68%, krypton: 20%). The binding of xenon within the upper, highly apolar part of the deeply buried reference pocket 3 has been previously observed by us (unpublished results, PDB code 3LS7). Both types of noble gases populate the same position in the S1′ cavity as well as in the upper part of reference cavity 3. In the S1′ cavity, van der Waals interactions are established to the side chains of Val139, His142, Glu143, Ile188, Leu202, Arg203, and the portion of 1 covering the S1′ pocket (Figure 3). No significant adaptations of cavity-lining residues of the
large experimental uncertainties, rendering the thermodynamic parameters unreliable even for mutual comparison. Thus, in these cases, only the Gibbs free energy is discussed. The thermodynamic profiles of 1−6 are displayed in Figure 1B. The affinity strongly increases over 4.5 orders of magnitude (expressed in terms of the binding constants as listed in Table S1) with the growing size of the hydrophobic P1′ group (1 → 2 → 3 → 4). In particular, the addition of a single methyl group to 1 yielding 2 (ΔΔG°1→2= −12.3 kJ mol−1) and the addition of two methyl groups to 2 yielding 3 (ΔΔG°2→3 = −9.2 kJ mol−1) result in substantial affinity boosts. With respect to the partitioning in ΔH° and −TΔS°, the affinity increase from 2 → 3 → 4 is the result of a sharp increase in ΔH°, which is only partially compensated by a decrease in −TΔS°. The affinities of 5 and 6 fall between those of 1 and 2. The amino-substituted 6 clearly shows an improved affinity relative to the hydroxylsubstituted 5. Crystal Structure Analysis. In addition to the already published crystal structures of the TLN-2 (3FVP), TLN-3 (3FLF), and TLN-4 (4H57) complexes,27 we succeeded in crystallizing TLN-1, TLN-5, and TLN-6 (Table 1). Furthermore, we also obtained structures of TLN-1 in complex with xenon (TLN-1-Xe) and krypton (TLN-1-Kr). Except for the native structure of TLN-1, which was experimentally phased, all other crystal structures were phased by the standard molecular replacement technique. Shape, Polarity, and Solvent Content Analysis within the S1′ Cavity and within Three Internal Reference Cavities of TLN-1. The structure of TLN-1 was experimentally phased using a zinc multi-wavelength anomalous dispersion (MAD) data set (Table 2 and Table S7). The experimental phasing of TLN-1 without any further leastsquares refinement steps resulted in a very clear, high-quality electron density (Figure S3). We decided to apply density modification techniques (solvent flattening and histogram matching) on the experimental phases to further improve their accuracy. Since the analyzed cavities are completely buried within the core of TLN and exhibit a narrow shape, the electron density within the cavities is not affected by the applied phase improvement techniques. The experimentally phased density map was put on an absolute electron number scale (cf. Experimental Section), and the total electron content within the S1′ cavity and three internal reference cavities of different solvation state was calculated. The protein model coordinates displayed in Figure 2A−E are taken from the experimentally phased, fully refined, and deposited model of the native TLN-1 (PDB code 5M9W), whereas the superimposed electron density map is the experimentally phased electron density without any further model-based refinement. In the following, the cavities and electron density maps are described relative to the view of Figure 2B−E. S1′ Cavity. The S1′ cavity of TLN-1 (Figure 2B) comprises a volume of 141 Å3. The top of the cavity is exclusively formed by apolar amino acid side chains of protein residues in addition to the ligand atoms of the P1′ group and the leucine P2′ portion of 1. The mid-to-lower left part of the cavity is mainly apolar, except for Asp138, which is inaccessible for hydrogen bonding due to its buried geometry. The polar side chains of Glu143, Asp170, and Arg203 describe the right-hand surface portion of the cavity. However, apart from the side chain of Glu143, they are all involved in saturating polar contacts and thus not available for hydrogen bonding with putative occupants of the 10421
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
10422
89.6 9.3 0.7 0.4 13.8 20.5 30.3
88.9 10.0 0.7 0.4 11.7 14.2 28.1
11.3 13.3 26.2
87.8 11.1 0.7 0.4
11.1 12.6 27.2
87.8 11.1 0.7 0.4
0.009 1.1
38.24−1.25 87232/4591 10.5/12.5 0.95 9.0 316 4/1 0/0 30 386
50.00−1.25 (1.32−1.25) 91830 (14553) 6.6 (49.5) 9.4 99.9 (99.3) 24.9 (24.7) 35.2 (7.0)
P6122 92.8, 92.8, 130.4 0.91841 2.4 48
TLN-5 (5LVD)
12.3 19.3 28.7
88.1 11.1 0.4 0.4
0.010 1.2
43.67−1.30 76939/4050 11.2/13.9 0.94 10.9 316 4/1 0/0 30 380
50.00−1.30 (1.38−1.30) 80990 (12879) 5.3 (49.0) 10.5 99.8 (99.6) 9.8 (8.7) 26.1 (4.4)
P6122 92.8, 92.8, 129.8 0.91841 2.3 47
TLN-6 (5MA7)
Numbers in brackets report values for the highest resolution shells. bCalculated using the program Matthews_coef from the CCP4 suite.29 cFigure of Merit. dCalculated using PROCHECK.30 The Ramachandran outlier Thr26 of TLN is described in the literature.31 eCalculated using MOLEMAN.32
a
0.013 1.2
0.012 1.2
0.011 1.2
43.55−1.33 71796/3779 11.1/13.9 0.94 10.6 316 4/1 0/2 28 384
(C) Refinement 46.24−1.44 56290/2963 10.8/14.1 0.94 11.0 316 4/1 2/0 28 332
43.60−1.21 95148/5007 10.7/12.7 0.95 9.2 316 4/1 0/0 28 344
resolution range (Å) no. reflections used in refinement (work/free) final R values for all reflections (%) (work/free) FOMc phase error (deg) protein residues no. calcium/zinc ions no. xenon/krypton atoms no. inhibitor atoms no. water molecules RMSD from ideality: bond lengths (Å) bond angles (deg) Ramachandran plot:d residues in most favored regions (%) residues in additionally allowed regions (%) residues in generously allowed regions (%) residues in disallowed regions (%) mean B factors (Å2):e protein non-hydrogen atoms inhibitor water molecules
50.00−1.33 (1.41−1.33) 75575 (11969) 8.1 (49.9) 10.1 100.0 (99.7) 20.7 (19.5) 26.0 (6.3)
50.00−1.21 (1.28−1.21) 100158 (15893) 6.2 (48.8) 9.9 99.9 (99.5) 12.9 (12.5) 24.7 (4.9)
resolution range (Å) no. unique reflections Rsym (%) Wilson B factor (Å2) completeness (%) redundancy ⟨I/σ(I)⟩
TLN-1-Kr (5M5F)
complex (PDB code) (A) Data Collection and Processing P6122 P6122 92.5 92.5, 129.9 92.4, 92.4, 130.2 1.40000 0.85940 2.3 2.3 47 47
TLN-1-Xe (5M69)
(B) Diffraction Data 50.00−1.44 (1.53−1.44) 59253 (9235) 10.2 (48.6) 12.3 99.0 (97.2) 37.2 (37.1) 28.0 (8.1)
P6122 92.5, 92.5, 130.5 0.91841 2.3 47
space group unit cell parameters: a, b, c (Å) wavelength (Å) Matthews coefficient (Å3/Da)b solvent content (%)b
TLN-1 (5M9W)
Table 1. X-ray Data Collection and Refinement Statisticsa
Journal of the American Chemical Society Article
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society Table 2. Zinc MAD Data Set of TLN-1 Used for Experimental Phasinga space group unit cell parameters: a, b, c (Å) wavelength (Å) resolution range (Å) no. unique reflections Rsym (%) completeness (%) redundancy ⟨I/σ(I)⟩ anomalous correlation (%) SigAno a
peak
inflection
remote
P6122 92.5, 92.5, 131.0 1.28196 50.00−1.50 (1.59−1.50) 91983 (11525) 4.3 (17.2) 91.4 (71.0) 7.8 (7.5) 31.0 (9.8) 67 (38) 1.905 (1.085)
P6122 92.6, 92.6, 131.0 1.28306 50.00−1.50 (1.59−1.50) 91926 (11423) 3.6 (14.5) 91.3 (70.3) 7.8 (7.5) 37.0 (12.1) 50 (23) 1.495 (0.957)
P6122 92.6, 92.6, 131.0 0.91841 50.00−1.32 (1.40−1.32) 137883 (17564) 5.4 (46.0) 93.2 (73.4) 7.9 (7.5) 24.2 (4.2) 16 (0) 1.020 (0.777)
Numbers in parentheses are values for the highest resolution shells. Additional quality statistics are given in Table S7.
Figure 2. Analysis of the experimentally phased electron density map of TLN-1 within selected cavities. (A) Ribbon model of the refined TLN-1 indicating the solvent-excluded surface of the S1′ cavity and of the internal reference cavities 1−3 as dark gray mesh. Water molecules within these cavities are shown as magenta spheres. The zinc ion is shown as a dark blue sphere, and the four calcium ions of TLN are shown as gray spheres. The TLN-bound ligand 1 is indicated as a light orange stick model. (B−E) Depiction of the experimentally phased absolute-scale electron density map detected within the (B) S1′ cavity (volume, 141 Å3; total electron content, 6.6 e−), (C) reference cavity 1 (volume, 59 Å3; total electron content, 19.4 e−), (D) reference cavity 2 (volume, 93 Å3; total electron content, 13.4 e−), and (E) reference cavity 3 (volume, 16 Å3; total electron content, 0.9 e−). The electron density map is displayed as blue, cyan, or red mesh, indicating three different contour levels in e−/Å3. Selected density peaks are labeled with their concentration maximum. Cavity-lining TLN residues are shown as thin gray stick models. The solvent excluded surfaces of the cavities are indicated in semi-transparent gray. In panel B, the TLN-bound 1 is shown as a light orange stick model, and the zinc ion is shown as a dark blue sphere. In panels C and D, water molecules are shown as magenta spheres and hydrogen bond interaction distances as orange dashed lines (labeled in Å). The determined volume and total electron content of the cavities are summarized in Table S8. The electron density maps of all four cavities are shown at six different contour levels from −0.2 to 0.3 e−/Å3 in Figures S4−S7.
Comparison between the S1′ Cavities of TLN-1 to TLN-6. In none of the refined, σ-scaled mFo − DFc electron densities of the six crystal structures TLN-1 to TLN-6 could
noble gas derivatized TLN-1 are observed compared to the native structure of TLN-1. 10423
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society
Table 3. Data Sets of Noble-Gas-Derivatized TLN-1 Used for the Determination of the Anomalous Signal (Friedel Pairs Separated)a TLN-1-Kr space group unit cell parameters: a, b, c (Å) wavelength (Å) resolution range (Å) no. unique reflections Rsym (%) completeness (%) redundancy ⟨I/σ(I)⟩ a
above K edge
below K edge
TLN-1-Xe
P6122 92.4, 92.4, 130.2 0.85940 50.00−1.33 (1.41−1.33) 142833 (23002) 7.8 (48.7) 99.9 (99.7) 11.0 (10.2) 19.2 (4.6)
P6122 92.5, 92.5, 130.2 0.87150 50.00−1.35 (1.43−1.35) 136781 (22091) 8.2 (46.6) 99.9 (99.6) 8.6 (8.0) 15.6 (4.0)
P6122 92.5, 92.5, 130.0 1.40000 50.00−1.44 (1.53−1.44) 111687 (17774) 9.9 (48.0) 99.2 (97.6) 19.8 (19.3) 20.9 (5.9)
Numbers in parentheses stand for the highest resolution shells.
acid side chains of TLN. Whereas the P1′ OH group of the serine derivative is found in only one populated conformation forming a hydrogen bond to the side chain of Glu143 (2.6 Å, Figure 4E), three conformations (occupancy a, 42%; b, 27%; and c, 31%) can be assigned to the P1′ amino function of 6 in the refined model (Figure 4F). Conformations b and c place the amino group within hydrogen-bonding distance to the side chain of Asn112 (3.0 Å) and Glu143 (2.5 Å).
■
DISCUSSION Analysis of the Experimentally Phased Electron Density within the S1′ Cavity of TLN-1. Even with highresolution data as in the current case, the analysis of the solvation pattern within a protein cavity by X-ray crystallography is by no means straightforward, since highly mobile (“disordered”) or partially occupied water molecules are difficult to detect. A highly mobile water molecule may be characterized by the lack of a well-defined, sufficiently deep energy minimum on the free energy landscape, resulting in a widely distributed, rather blurred electron density lacking a clearly defined center.34 Model bias introduced into the electron density via the use of phases transferred from the refined structure of a related complex can furthermore obscure the detection of such weak density signals. In particular, the use of molecular replacement as phasing technique can result in a significant impact of model bias.35−37 Experimentally phased electron densities have the advantage that no model bias is arbitrarily introduced by the application of predefined model phases.25,38,39 Here, it is important to distinguish between “specific” model bias, e.g., bias toward the structure of the search fragment used for molecular replacement, and “general” model bias, e.g., the assumption of a flat solvent region that all density modification programs make. Specific model bias is much more dangerous because it can lead to false conclusions about the biochemistry involved, so it should be avoided where possible. General model bias only affects the overall quality of the model, and in the case of density modification usually does more good than harm, though it is critically dependent on the resolution and solvent content and on whether non-crystallographic symmetry (NCS) can be exploited. If the anomalous and dispersive signals are weak, the combination of MAD and density modification can still be a good way of obtaining a highquality structure that is not biased by preconceived ideas and existing structures (G. M. Sheldrick, personal communication). Therefore, to reliably detect traces of electron density originating from highly mobile or partly occupied water molecules, we applied an elaborate experimental phasing
Figure 3. Xenon binding site in the S1′ cavity of TLN-1. The xenon derivatized crystal structure TLN-1-Xe (dark gray) is superimposed on the native crystal structure of TLN-1 (light gray). The center of the bound xenon atom is shown as teal sphere, its van der Waals radius is indicated as teal mesh, and van der Waals interactions (≤4.6 Å) to cavity lining atoms are indicated as dark gray dotted lines. Cavitysurface lining TLN residues are shown as thin sticks, the bound 1 is shown as thick stick model. The solvent-excluded surface of the S1′ cavity is indicated in semi-transparent white. The anomalous map is displayed in gold at a contour level of 5σ, and the detected peaks are labeled with their intensity maximum and their related atom. The crystal structure of TLN-1-Kr is shown in Figure S8.
any clearly defined electron density features attributable to a bound water molecule be detected within the S1′ cavity. Even in the complexes of TLN with the ligands exhibiting the polar P1′ groups (5 and 6), the unoccupied part of the cavity remains seemingly empty, even though, as indicated in Figure 4, sufficient space is given to accommodate a water molecule. Figure 4 displays the crystal structure models of TLN-1 to TLN-6 and the residual S1′ cavities. With growing P1′ portion (1 → 4), the top part of the cavity is gradually occupied, until only the bottom part of the cavity remains unoccupied (Figure 4A−D). Accordingly, the volume of the S1′ cavity is gradually decreasing from 141 Å3 to 48 Å3. The hydrophobic P1′ groups of TLN-2, TLN-3 and TLN-4 form multiple hydrophobic van der Waals interactions to the pocket-shape determining amino 10424
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society
Figure 4. S1′ cavity in crystal structures: (A) TLN-1, (B) TLN-2, (C) TLN-3, (D) TLN-4, (E) TLN-5, and (F) TLN-6. The solvent-excluded surface of each S1′ cavity is displayed as dark gray mesh labeled with its volume in Å3. Cavity-lining amino acids are displayed as gray stick models. Ligands 1−6 are shown as thick stick models in light orange. The zinc ion is displayed as a dark blue sphere. Hydrogen bonds formed between TLN and the P1′ group of 5 and 6 are indicated as orange dashed lines (labeled in Å). The three conformations of the P1′ group of 6 are labeled with a−c. All three conformations of the latter P1′ substituent were considered for the calculation of the residual cavity volume.
of 40 Å3 observed in interleukin-1β provided insight for the debate about the presence of a disordered water molecule in this volume.23,25,41,42 From the data analysis, Quillin et al. concluded that the cavity must indeed be empty.25 In the case of TLN, reference cavity 3 can be considered as an internal control for an entirely empty cavity (Figure 2E). This small, apolar cavity of 16 Å3 is unsuited to host a water molecule due to its size and polarity.43−45 With respect to the calculated volume of 16 Å3 for reference pocket 3, it must be considered that the calculation of the volume that is really available to host a water molecule is nontrivial and is actually further reduced adjacent to apolar residues.46,47 In other words, the solvent-probe radius adjacent to apolar residues is in fact larger than the commonly applied 1.40 Å. Consequently, the presence of a water molecule in reference cavity 3 can be certainly excluded. The electron concentration peak found within this cavity comprises a concentration maximum of 0.41 e−/Å3, and the total electron content is 0.9 e−. We consider this electron content the result of “spillover” of electrons from cavity-lining amino acids into the, to some degree fuzzy, volume of the empty cavity,25,38 e.g., due to a slight movement of the protein residues. Reference cavities 1 and 2 (Figure 2C,D) serve as an internal control for clearly solvated cavities. The calculated total electron content within reference cavity 1 (59 Å3) is 19.4 e−, a number significantly lower than expected for three fully occupied water molecules (30.0 e−). Similar as
protocol for TLN-1. Even though unbiased phases can be obtained by this procedure, another obstacle arises from the fact that the total electron number within the crystal unit cell F000 is impossible to determine experimentally in the case of proteins. In small-molecule crystallography, the content of the diffracting unit cell is usually easy to define, and thus the number of contributing electrons is clear. In the case of proteins showing a large solvent content, particularly in the channels passing through the crystal, such an assignment is impossible and thus can only be estimated. In consequence, electron densities in protein crystallography are typically σscaled,40 where zero σ corresponds to the average, numerically unknown electron concentration of the unit cell. Even though this step appears very reasonable and pragmatic in standard refinements of protein structures, it will make the assignment of an absolute number of electrons to a particular integrated volume virtually impossible. Therefore, we attempted to transform the experimentally phased electron density to an absolute electron number scale, where the zero value corresponds to vacuum. Following a similar approach, Liu et al. analyzed a hydrophobic cavity of 134 Å3 in L99A/M102L T4 lysozyme.38 In this case, analysis of the experimentally phased electron density on absolute scale revealed a water cluster of three water molecules with an occupancy of approximately 50% within the pocket. In another example, the analysis of an experimentally phased electron density of a central apolar cavity 10425
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society observed by Liu et al.,38 this is possibly the result of the electron spillover from within the cavity to the outside due to overlapping van der Waals radii of polar cavity-lining atoms and the tightly hydrogen-bonded water molecules. Reference cavity 2 (93 Å3) contains a total electron content of 13.4 e−. Apart from the unambiguous water density in the lower part of the cavity, a significant amount of blurred electron density with a peak maximum of 0.62 e−/Å3 is detected in the upper, hydrophobic part of the cavity (Figure 2D)which is large enough to host a phenol molecule in addition to the water molecule in the lower part of the cavity.48 The abovementioned secondary electron concentration maximum is within hydrogen-bonding distance (3.2 Å) to the modeled water molecule in the lower part of the cavity and could potentially originate from a second, very weakly populated water site. Overall, this results in a higher electron content than expected for the single fully occupied water molecule, overcompensating the expected spillover of electrons of the tightly bound water molecule to the outside of the cavity. Remarkably, the residual electron distribution within the S1′ cavity shows an electron concentration of maximally 0.46 e−/ Å3, a value close to the one found for the empty reference cavity 3 (0.41 e−/Å3). The total electron content within the S1′ cavity was determined to be 6.6 e−. If the electron content within the empty reference cavity 3 (0.9 e−) is extrapolated to the larger volume of the S1′ cavity (141 Å3/16 Å3 = 8.8 times larger), this corresponds to 7.9 e− (0.9 e− × 8.8), a number higher than the one for the experimentally determined electron content within the S1′ cavity (6.6 e−). This can be explained by the fact that relative to their volumes reference cavity 3 has a significantly larger surface than the S1′ cavity, resulting in a proportionally larger electron “spillover” (vide supra). Taking these considerations into account, we therefore propose that the TLN-1 S1′ cavity is indeed a completely unsolvated cavity. We also believe that this is not simply a consequence of the binding of 1 which seals the pocket from the top. Our complex structures of TLN1 with the noble gases (Figure 3) clearly demonstrate that the pocket is still well accessible in the crystal by particles as large as a xenon atom, likely due to the residual mobility of the complexes. Thus, smaller water molecules could easily access the S1′ pocket if their penetration would be favorable. Analyses of TLN-1 in Complex with Xenon and Krypton and of Crystal Structures TLN-1 to TLN-6. The noble gases xenon and krypton are known to preferentially bind to desolvated, hydrophobic protein cavities through weak van der Waals interactions.49,50 These atoms can, therefore, be used as experimental probes to detect such cavities. The S1′ cavity (Figure 3) and reference cavity 3 were revealed as noble gas binding sites in TLN-1. The fact that both noble gases were found binding to the S1′ cavity even though the derivatization pressure was kept low supports the hypothesis that both cavities do not contain significantly populated solvent molecules, which need to be displaced upon ligand accommodation. The increased occupancy of the gaseous probes in reference cavity 3 compared to the S1′ cavity can be attributed to the deeper burial of this cavity within the apolar interior of TLN. Thus, during the experimental depressurization phase that needs to be accomplished to transfer the crystal specimen from the pressuring cell to the liquid nitrogen, diffusion of the noble gases is slower. The reduced occupancy of krypton compared to xenon is attributable to the reduced polarizability of the former,51,52 the lower applied derivatization pressure, and the faster diffusion kinetics of the smaller krypton
atoms. Noble gas binding provides a crude estimate about the least available space within a given cavity, and provides information about the least detectable electron concentration in a certain volume.17,51 For instance, in TLN-1-Kr (Figure S8), the krypton atom (36 e−, van der Waals radius 2.0 Å, volume 33.5 Å3) bound to the S1′ cavity was refined to 8% occupancy. This corresponds to 2.9 electrons (36 e− × 0.08). Considering the refined temperature factor of 12.5 Å2 of this krypton atom, krypton will occupy a volume of about 58 Å3.17,34 Hence, if we transfer this detection limit to our cavity analysis, a putative water molecule with an occupancy as low as 29% (corresponding to 2.9 e−) and a distribution over a volume of 58 Å3 should be detectable by conventional refinement. If we apply this estimation rule to the S1′ cavities of complexes TLN-1 to TLN6, it is permissible to conclude that the residual volumes found in TLN-3 to TLN-6 (48−76 Å3, Figure 4) must be virtually empty. This is further supported by the observation that the S1′ cavities of TLN-3, TLN-4, and TLN-6 are divided by the P1′ groups of the bound ligands into two spatially separated cavities, reducing the mobility of a putatively present water molecule even further. Consequently, a potentially present but nonetheless undetectablewater molecule must be very weakly occupied to elude crystallographic detection completely. We believe that this minor displacement effect would be insignificant for the binding event and would hardly influence the overall thermodynamic profile. Thus, for the discussion of the thermodynamic signature, we refer to the empty state of the cavity. The polar groups of 5 and 6 were introduced into the S1′ pocket with the aim to provide a local anchor for hydrogen bonding for a potentially present and disordered water molecule. If such a water molecule would hypothetically be present in the complexes with the aliphatic ligands, this hydrogen-bonding decoy should result in the stabilization of a disordered water moleculeor possibly in the recruitment of a new water molecule from the bulk water phaseand make it crystallographically detectable.53,54 However, no clear electron density peak is observed in the crystal structures of TLN-5 or TLN-6. One reason why the polar groups might not be available to establish hydrogen bonds to a putative water molecule arises from the fact that they form hydrogen bonds to protein residues. In TLN-6, however, conformation a, which refines to the highest occupancy, orients toward the void of the cavity where it could experience hydrogen bonding (Figure 4F). In addition, even though the polarity of the S1′ cavities of TLN-5 and TLN-6 increases, the volume of the cavity is reduced, which decreases the probability to find a water molecule within a cavity.44 Thermodynamic Binding Profiles of 1−6 As Determined by ITC. The affinities of the investigated ligand series fall into the range between milli- to nanomolar binding (Table S1). Because of this broad range, it was necessary to apply different measurement protocols to obtain reliable calorimetric data. Ligands 2, 3, and 4 were measured by direct titration. Due to their low affinities, 1, 5, and 6 had to be characterized by displacement titrations. With this technique, the measurement accuracy of the binding constants is strongly dependent on how accurately the thermodynamic data of the applied reference ligand has been recorded. Any error in the thermodynamic profile of the reference ligand will propagate to the thermodynamic profile of the analyte. Consequently, such measurements usually result in large experimental errors, in particular considering the partitioning of enthalpy and 10426
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society entropy.55 The affinity of 1 is too low to accurately determine the enthalpy/entropy signature via displacement titration. Furthermore, 5 and 6, comprising polar P1′ groups, show very likely different changes in their protonation state compared to the ligands with the aliphatic P1′ groups. A superposition of a proton transfer reaction can alter the observed enthalpy of binding and therefore prevent a direct comparability of measured enthalpy values.56 Unfortunately, an appropriate enthalpy correction is not reliably feasible in a displacement titration scenario due to the resulting highly potentiating errors of the buffer-corrected thermodynamic profiles. Consequently, we decided to refrain from analyzing the enthalpy/entropy profiles of 1, 5, and 6, and instead solely report their affinities. This value is independent of buffer effects and allows accurate comparison with the more potent ligands. The addition of a single methyl group to the glycine derivative 1 (yielding 2) results in a 140-fold affinity increase (Figure 1B and Table S1). Adding two methyl groups to 2 (yielding 3) results in a further 40-fold affinity increase. Finally, the addition of a methyl group from 3 to 4 corresponds to a 7fold affinity increase. Based on a statistical evaluation, a 100fold affinity increase solely achieved by the addition of one single methyl group has been reported to be a very rare event with a probability of only 0.4%.57 In the reported cases, the favorable affinity increase resulted from the placement of a methyl group into hydrophobic pockets that the authors assumed to be entirely empty, in combination with an energetically favorable pre-organization of the ligand conformation in solution. This assumption is supported by MD simulations suggesting that the placement of hydrophobic groups into supposedly empty, hydrophobic pockets produces an extraordinarily favorable change of the Gibbs free energy of binding due to the absence of a cavity desolvation step.18,20 In the same manner, the extraordinary affinity increase from 1 to 2 (140-fold) can be attributed to the insertion of a methyl group into a hydrophobic cavity, that isas experiment confirms virtually empty. This effect is reduced, but nevertheless still pronounced when further methyl groups are grown into this volume (40-fold affinity increase between 2 and 3). From 3 to 4, the comparably low 7-fold affinity increase is the result of a highly favorable enthalpy term, that, however, is partly compensated by unfavorable entropy (Figure 1B). Usually, dispersive (van der Waals) interactions are less significant, because such interactions established between protein and ligand are largely canceled out by the required unfavorable disruption of dispersive interactions between protein and solvent.58 However, in the case of a vacated pocket that makes a desolvation step obsolete, dispersive interactions formed between protein and ligand upon complexation become determinant in terms of affinity.58 This is the case in the current studythere is no cost to desolvate the S1′ binding pocket. As a result, the contribution of the established dispersive interactions between protein and ligand to the enthalpy of binding increases with increasing P1′ chain length, and overall strongly affects binding affinity. However, with increasing P1′ chain length, the enthalpic signal is increasingly compensated by a decrease in entropy (Figure 1B). This is potentially the result of a loss of conformational degrees of freedom upon complex formation. The low affinities of 5 and 6 can be attributed to the large energetic penalty for the desolvation of their polar functional groups which is not overcompensated by the dispersive interactions with the cavity-lining residues. The hydrogen-bond interactions formed with Glu143 (and Asn112
in the case of 6) do not suffice to compensate for this loss. Thus, the free energy gain resulting from the establishment of dispersive interactions is largely compensated by the high cost to desolvate 5 and 6, overall lowering their affinity. Protein cavities can fulfill essential biological functions, e.g., conformational flexibility, and thus can represent more than “packing defects”.50,59,60 In the case of TLN (and many other metalloproteinases), the S1′ pocket is the most important pocket to discriminate substrates from non-substrates, and thus defines the specificity profile of the protease. A substrate that exhibits shape complementary and thus fills the hydrophobic S1′ pocket efficiently without requiring a large price for desolvation (e.g., a P1′ leucine, isoleucine, or valine side chain) will experience strong dispersive interactions with the protease. It thus gets bound and enzymatically processed. In contrast, if the pre-organized void in the protease is insufficiently filled by an either small or hydrophilic P1′ substrate portion, a pronounced affinity-reducing enthalpic penalty results. Thus, the unsolvated state of the TLN S1′ pocket is key for the substrate specificity of this protease. An entirely independent evidence for the existence of unsolvated protein binding pockets is provided by a current neutron diffraction study.61 Neutron scattering can reliably differentiate between hydrogen and deuterium. We recently determined the joint X-ray/neutron structures of two trypsin complexes. Crystal growth was performed with the protein in its hydrogen form, and crystals were subsequently exposed to fully deuterated buffer over up to 532 days before data collection. Under such conditions polar hydrogen atoms should exchange with deuterium atoms considering the large excess of deuterons compared to protons. However, as a precondition the polar groups must be accessible to D2O molecules. In a folded protein, polar hydrogens usually remain only at sites where they are involved in strong hydrogen bonds. Nevertheless, in a sterically accessible pocket of our trypsin complexes we found NH groups that are not involved in hydrogen bonds but hardly exchange to ND over 1.5 years of incubation in D2O. Only 7% deuterium could be found at this site. We therefore hypothesize that they hardly experienced any contact with D2O molecules in the folded protein indicating that these cavities are extremely rarely accessed by water molecules. We believe this is another independent indication that empty and hardly solvated pockets exist in folded proteins.
■
CONCLUSION The collected experimental data strongly suggest that the S1′ cavity of TLN in complex with ligands 1−6 is not solvated and virtually evacuated prior to ligand binding. We found no experimental evidence for the occupation with highly mobile water molecules, and the properties of the cavities (shape, volume, polarity) suggest that the cavity is indeed empty. To gain the required experimental evidence we engineered a weak, only 5 mM binding ligand that geometrically covers the active site of the protease without addressing the S1′ pocket, thus formally transforming it into a sealed cavity. This calls for the question whether this enclosure transfers the pocket only then into the vacated state. The following facts argue against this conclusion. We could prove the still sustained accessibility of the pocket by accommodating noble gas atoms in the pocket even in the crystalline state. The thermodynamic data were recorded in solution under equilibrium conditions where our 5 mM inhibitor will frequently dissociate and associate with the protein, giving unrestrained access to the intermediately fully 10427
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society
detector at 100 K.66,67 A three-wavelength MAD data set of TLN-1 (Table 2), which was applied for the determination of the experimental phases, was collected from another crystal at BESSY II on beamline 14.2 equipped with a Rayonix MX-225 pixel detector at 100 K. Data sets for crystal structures TLN-1-Xe and TLN-1-Kr (Table 3) were collected at Elettra (Triest, Italy) on beamline XRD1 at 100 K using a Dectris Pilatus 2M pixel detector. The data set for TLN-1-Xe was collected at λ = 1.4000 Å. Two data sets were collected for TLN-1-Kr from the same crystal (Table 3), the first at 0.8594 Å above the krypton K absorption edge (that is at 0.8655 Å), against which the structure was refined, and a second data set at 0.8715 Å below the krypton K absorption edge. The second data set was solely used for analysis of the change of the anomalous signal of krypton to unambiguously identify the bound atom. The anomalous maps for TLN-1-Xe (Figure 3) and TLN-1-Kr (Figure S8) were created with ANODE.68 Data Set Processing and Structure Determination. All data sets were indexed, integrated and scaled with XDS.69 The phases of TLN-1-Xe, TLN-1-Kr, TLN-5, and TLN-6 were determined by molecular replacement applying Phaser70 from the CCP4 suite (version 6.3.0).71 The crystal structure of the PDB entry 8TLN was used as molecular replacement search model.72 The phases for TLN-1 were determined experimentally by a MAD data set (Table 2) using the intrinsically bound zinc ion of TLN as anomalous scatterer. Experimental phasing was performed applying the programs SHELXC (data preparation), SHELXD (heavy atom substructure determination),73 and SHELXE (experimental phasing and density modification)74 as implemented in HKL2MAP.75 The anomalous signal d′′/σ of the peak data set was significant (1.42) to 1.62 Å. We estimated that the zinc ion contributed approximately 1.5% toward the anomalous signal, the calcium ions 0.7%, and the contribution from sulfur atoms is about 0.4%.76 Nevertheless, structure determination using the zinc as anomalous scatterer only was straightforward and we did not encounter any problems during this step as confirmed in the following statistics. Between all data sets the anomalous correlation coefficient was above 30% up to 1.57 Å. The best solution of the SHELXD substructure search was CCAll = 71.3 and CCWeak = 58.9. The substructure phases calculated by SHELXE from the original hand gave a final contrast of 0.56 and a connectivity of 0.81 (inverted hand: contrast 0.35, connectivity 0.66). The experimentally determined phases were combined with the amplitudes of a 1.21 Å native data set of TLN-1 (Table 1) followed by density modification with the program DM77 from the CCP4 suite applying solvent flattening and histogram matching. A starting model for conventional refinement was created with ARP/wARP,78 where 314 amino acids in a single polypeptide chains (99% sequence coverage) were successfully placed into the electron density map with a resolution of 1.21 Å. Model Building and Refinement. Crystal structure model refinement was performed with phenix.ref ine version 1.10.1-2155.79 Simulated annealing with default settings was performed as first refinement step. Subsequently, all crystal structure models were refined with riding hydrogen atoms added to protein residues, applying xyz refinement, individual anisotropic B factors for all atoms except for hydrogens, and occupancy refinement. Refinement cycles were alternated with model building steps in Coot.80 Ligand building was performed with MOE,81 ligand restraints were created with eLBOW.82 Graphical representations of the crystal structure coordinates and electron density maps (automatic ccp4 map normalization turned off) were created with PyMOL.83 Cavity computation and volume calculation. The cavities as displayed in Figure 4 were computed with POVME.84 Dummy atoms (DAs) with a radius of 1.40 Å were placed into the cavities (grid spacing 0.2 Å). The radius of 1.40 Å was chosen as it is frequently applied as the van der Waals radius of a water molecule in the literature.17,21,25,51,85 The solvent-excluded surfaces of the DA objects representing the cavities were then displayed with PyMOL. Since POVME can only calculate the solvent-accessible volume, the solventexcluded volumes of the DA objects describing the S1′ cavities of TLN-1 to TLN-6 (Figure 4) were calculated with the program 3V (radius of the DAs set to 1.40 Å).86 The volumes of the S1′ cavity and
accessible pocket. Two very weak low-micromolar ligands comprising a P1′ substituent with a H-bonding decoy group to capture putatively penetrating water molecules in the pocket did not show any evidence for hosted water molecules. Thus, it seems energetically more favorable to maintain a vacuum than to host one or several water molecules within the S1′ pocket. It has been discussed that the generation of vacated pockets in proteins is energetically very costly.62,63 Nevertheless, it has to be considered that the costs for producing such a void have to be afforded during protein folding and not during ligand or substrate binding. In the current case, the fact that the TLN S1′ cavity is empty has a major impact on the energetically highly favored accommodation of aliphatic P1′ side chains of either substrate or inhibitor molecules. The observed enthalpy-driven affinity enhancement with increasing size of the P1′ substituent (1 → 4) is mainly a result of the binding of aliphatic groups into a void, where no price for pocket desolvation has to be afforded. The decreasing affinity contribution of a growing side chain can be attributed to the augmenting desolvation penalty of the larger P1′ substituents and to a reduction in conformational flexibility. Addition of a polar group entirely destroys binding affinity due to an uncompensated desolvation penalty. Remarkably, the derivative with a P1′ benzyl side chain (PDB code 3FV4) that fills the S1′ pocket more efficiently than the P1′ leucine side chain of 4 is less potent, likely due to the higher desolvation costs for the aromatic side chain.27 In the case of TLN, the hydration state of the S1′ cavity seems to strongly contribute to the substrate specificity of the protease. The remarkable, about 41 000-fold increase in affinity upon the introduction of a hydrophobic isobutyl group into this pocket has major implications for medicinal chemistry. The identification of poorly hydrated cavities can represent a valuable strategy to gain overwhelmingly in binding affinity of a prospective drug molecule.
■
EXPERIMENTAL SECTION
Ligand Synthesis. The synthesis of 1−6 has been reported previously.64 Crystal Preparation and Soaking. Crystals were prepared with lyophilized TLN powder commercially obtained from Calbiochem (EMD Biosciences) according to the procedure described earlier.12 For crystal soaking of the low-affinity compounds 1, 5, and 6, TLN crystals were transferred into a soaking solution composed of 100 mM Tris-HCl, pH 7.5, 2 mM CaCl2, and 20% DMSO saturated with the respective ligand (ligand precipitate visible), followed by incubation for 24 h. Afterward, crystals were flash-frozen in liquid nitrogen after a brief immersion in a cryoprotectant solution saturated with the respective ligand, composed of 10 mM Tris-HCl, pH 7.5, 10 mM Ca(CH3COO)2, 20% DMSO, and 20% glycerol. Derivatization of TLN with Xenon and Krypton. For the noble gas derivatization of the TLN crystals (TLN-1-Xe and TLN-1-Kr), a pressure cell from Oxford Cryosystems (Long Hanborough, UK) was used.65 Before derivatization, TLN crystals were soaked with 1 in the above-mentioned soaking buffer for 24 h. To protect the crystals from drying out during the pressurization phase, the filter paper of the pressurization cell was drenched with soaking buffer. Subsequently, xenon derivatization was performed at 9 bar for 5 min, whereas derivatization with krypton was carried out at 5 bar for 5 min. Derivatization was conducted at relatively low pressure because higher pressure resulted in a strong increase in crystal mosaicity. After the incubation phase, pressure was quickly released, and TLN crystals were immediately flash-frozen in liquid nitrogen. Data Collection. Data sets for crystal structures TLN-1, TLN-5, and TLN-6 (Table 1) were collected at BESSY II (Berlin-Adlershof, Germany) on beamline 14.1 operating with a Dectris Pilatus 6M pixel 10428
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society of reference cavities 1−3, as displayed in Figure 2, used for the analysis of the electron density map, were created with DAs comprising a radius of 1.30 Å. The reduction of the DA radius compared to the DA radius applied for computing the cavities in Figure 4 (1.40 Å) was necessary, because otherwise the narrow parts of reference cavities 1 and 2, where hydrogen-bonded water molecules are closely bound to TLN residues, would not have been detected. Placement of the Experimentally Phased Electron Density Map of TLN-1 on an Absolute Electron Number Scale and Determination of the Total Electron Content within a Cavity. Two mathematical operations are required to transform the experimentally phased, σ-scaled map of TLN-1 (ccp4 format) characterized by arbitrarily small map voxel values to an absolute electron number density map where zero corresponds to vacuum thus allowing quantification of the total electron content within a given map volume. First, the values of the map voxels have to be set to the correct scale by applying a scaling coefficient. Subsequently, the still σscaled mapwhere zero corresponds to the average electron concentration of the unit cellhas to be shifted in a way that zero corresponds to vacuum. For that, the average electron concentration (e−/Å3) of the unit cell must be added to every map voxel. To have access to an absolute electron number density map as a reference, the σ-scaled density map of the fully refined model of TLN-1 was transformed to an absolute electron number density map applying the END map script developed by Lang et al.87 running with phenix.ref ine version 1.8.4−1492. This program computes absolute electron number density (END) maps from conventionally refined σ-scaled maps. However, it is dependent on a refinement program and relies on refined model phases for the calculation of the electron density map, thus it is not possible to directly use it to transform the experimentally phased σ-scaled density map to an absolute electron number scale without the introduction of model phases. The END map script also provided the structure factor F000 of TLN-1 (256 668 e−), which corresponds to the total electron content within the crystal unit cell when considering the structural model as well as (disordered) bulk solvent. Thereby, the average electron concentration within the unit cell was accessible by calculating the ratio F000/unit cell volume = 256 668 e−/967 353 Å3 = 0.26533 e−/Å3. We decided to use the Zn2+ ion (28 e−) that is intrinsically bound to TLN to derive the scale coefficient. To be independent of the theoretical electron number of the zinc ion and of the necessity of an integration mask covering the entire zinc ion without any electron spillover from inside to the outside of the mask (e.g., due to a slightly disordered zinc) or electron spillover from outside to the inside of the integration mask (e.g., from zinc-complexing residues), we determined the total electron content of a spherical map fragment describing the zinc ion of the refined END map. The zinc coordinates of the refined crystal structure TLN-1 were used for the center of the sphere, and 1.39 Å was used as sphere radius. The total electron content within this map fragment gave 27.3 e− as calculated with an in-house script based on the program mapman.88 Subsequently, an identical integration mask was applied to the σ-scaled experimentally phased electron density for scaling of the latter data set by applying the program mapmask from the CCP4 suite.71 The scale coefficient (14.1) was derived in a way that after applying the scale coefficient followed by addition of the average electron concentration (0.26533 e−/Å3), integration of the map fragment covering the zinc of the experimentally phased map resulted in exactly 27.3 e−, just as observed in the map fragment of the refined END map. Visually, the experimentally phased electron density map was correctly shifted in a way that zero corresponds to vacuum (Figures S4−S7). If an electron concentration of ≥0.0 e−/Å3 is displayed (panels C of Figures S4−S7), the entire cavities are covered with electron density. In contrast, displaying higher (positive) or lower (negative) electron density concentrations for instance within the empty S1′ pocket (Figure S4) immediately reduces the visible residual electron density. For the calculation of the total electron content within a given volume, e.g., the volume of the S1′ cavity and reference cavities 1−3, a part of the map was cut out from the entire electron density map applying the program Coot via the command “mask map by atom
selection”, followed by integration over all voxels (including negative values) applying the aforementioned in-house script. For the S1′ cavity and reference cavities 1−3, the DA objects also used for displaying the contour regions of the cavities as shown in Figure 2 were used applying a radius of 1.3 Å. Measurement of the Thermodynamic Binding Profiles. For ITC measurements, a Microcal ITC200 device from GE Healthcare (Piscataway, NJ) was used. Measurements were performed with freeze-dried TLN powder from Calbiochem (EMD Biosciences), which was dissolved in buffer directly before measurement without further treatment. Measurement buffers (pH 7.5) were composed of 20 mM N-(2-acetamido)-2-aminoethanesulfonic acid (ACES), 3morpholino-2-hydroxypropanesulfonic acid (MOPSO), 2-(Nmorpholino)ethanesulfonic acid (MES), piperazine-N,N-bis(2-ethanesulfonic acid) (PIPES), or cacodylate buffer substance, 500 mM NaSCN, and 2 mM CaCl2. The salt NaSCN was selected because it strongly increases the solubility of TLN.89 The high concentration of 500 mM was necessary to provide solubility of TLN up to 250 μM that was necessary to enable a direct measurement of the low-affinity ligand 2. Since different concentrations of NaSCN are known to influence the measured thermodynamic binding parameters,55 this salt concentration was applied for all ligands to guarantee relative comparability of the thermodynamic binding parameters. The thermodynamic profiles of 3 (Table S4) and 4 (Table S5) were determined by direct titration at a TLN concentration of 50 μM, resulting in well analyzable isotherms with clear sigmoidal curvatures (Figure S1B,C). For the low-affinity compound 2 (Table S3), the concentration of TLN was increased to 250 μM, resulting in isotherms described by a c-value of 6 (Figure S1A). This allowed the experimental determination of the inflection point and the determination of reliable thermodynamic parameters. To determine the heat of ionization associated with the complex formation between TLN and 2, 3 or 4 (Figure S2), measurements were performed in ACES, MOPSO, HEPES, MES, PIPES, and cacodylate buffers. The affinities of the weak-binding compounds 1, 5, and 6 were determined by displacement titrations (Figure S2D−F and Table S6).90 50 μM TLN in cacodylate buffer solution was pre-incubated with different concentrations of weak ligand (1: 5, 8, 20 mM; 5: 5, 10, 15 mM; 6: 1.5, 2, 2 mM), followed by titration with the reference ligand 4. Different concentrations of the weak ligand (resulting in titration curves exhibiting different c-values) were applied to improve the accuracy of the binding parameter determination by global analysis of the ITC isotherms.91 Measurement peaks of the raw thermograms were extracted and integrated with NITPIC version 1.1.2.92,93 Global analysis of the ITC isotherms was performed with SEDPHAT version 12.1b.94,95 The thermodynamic profiles of the ligands measured in individual buffers were determined by global analysis applying the model “A+B ↔ AB Hetero-Association”. Correction for the heat of ionization of these thermodynamic profiles was performed using the model “A+B ↔ AB Hetero-Association Global Buffer Ionization Enthalpy Analysis”. For the displacement titrations, the model “A+B +C ↔ AB+C ↔ AC+B; competing B and C for A” was applied. ITC isotherm graphs were prepared with GUSSI.96
■
ASSOCIATED CONTENT
S Supporting Information *
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/jacs.7b05028. Isothermal titration calorimetry results, analysis of the experimentally phased TLN-1, and crystal structure of TLN-1-Kr, including Figures S1−S8 and Tables S1−S8 (PDF)
■
AUTHOR INFORMATION
Corresponding Author
*
[email protected] 10429
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society ORCID
(25) Quillin, M. L.; Wingfield, P. T.; Matthews, B. W. Proc. Natl. Acad. Sci. U. S. A. 2006, 103, 19749. (26) Vaitheeswaran, S.; Yin, H.; Rasaiah, J. C.; Hummer, G. Proc. Natl. Acad. Sci. U. S. A. 2004, 101, 17002. (27) Englert, L.; Biela, A.; Zayed, M.; Heine, A.; Hangauer, D.; Klebe, G. Biochim. Biophys. Acta, Gen. Subj. 2010, 1800, 1192. (28) Jelesarov, I.; Bosshard, H. R. J. Mol. Recognit. 1999, 12, 3. (29) Collaborative Computational Project Number 4. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1994, 50, 760. (30) Laskowski, R. A.; MacArthur, M. W.; Moss, D. S.; Thornton, J. M. J. Appl. Crystallogr. 1993, 26, 283. (31) Matthews, B. W. Macromolecules 1972, 5, 818. (32) Kleywegt, G. J.; Zou, J.-Y. In International Tables for Crystallography, Vol. F: Crystallography of biological macromolecules; Rossmann, M. G., Arnold, E., Eds.; Kluwer Academic Publishers: Dordrecht, The Netherlands, 2001; p 353. (33) Schiltz, M.; Fourme, R.; Prangé, T. Methods Enzymol. 2003, 374, 83. (34) Levitt, M.; Park, B. H. Structure 1993, 1, 223. (35) Read, R. J. In International Tables for Crystallography, Vol. F: Crystallography of biological macromolecules; Rossmann, M. G., Arnold, E., Eds.; Kluwer Academic Publishers, Dordrecht, The Netherlands, 2001; p 325. (36) Brünger, A. T.; Rice, L. M. Methods Enzymol. 1997, 277, 243. (37) Hodel, A.; Kim, S.-H. -H; Brünger, A. T. Acta Crystallogr., Sect. A: Found. Crystallogr. 1992, 48, 851. (38) Liu, L.; Quillin, M. L.; Matthews, B. W. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 14406. (39) Burling, F. T.; Weis, W. I.; Flaherty, K. M.; Brünger, A. T. Science 1996, 271, 72. (40) Urzhumtsev, A.; Afonine, P. V.; Lunin, V. Y.; Terwilliger, T. C.; Adams, P. D. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2014, 70, 2593. (41) Yu, B.; Blaber, M.; Gronenborn, A. M.; Clore, G. M.; Caspar, D. L. D. Proc. Natl. Acad. Sci. U. S. A. 1999, 96, 103. (42) Matthews, B. W.; Morton, A. G.; Dahlquist, F. W. Science 1995, 270, 1847. (43) Sonavane, S.; Chakrabarti, P. PLoS Comput. Biol. 2008, 4, e1000188. (44) Hubbard, S. J.; Gross, K. H.; Argos, P. Protein Eng., Des. Sel. 1994, 7, 613. (45) Wolfenden, R.; Radzicka, A. Science 1994, 265, 936. (46) Bhat, S.; Purisima, E. O. Proteins: Struct., Funct., Genet. 2006, 62, 244. (47) Li, A. J.-J.; Nussinov, R. Proteins: Struct., Funct., Genet. 1998, 32, 111. (48) English, A. C.; Groom, C. R.; Hubbard, R. E. Protein Eng., Des. Sel. 2001, 14, 47. (49) Prangé, T.; Schiltz, M.; Pernot, L.; Colloc’h, N.; Longhi, S.; Bourguet, W.; Fourme, R. Proteins: Struct., Funct., Genet. 1998, 30, 61. (50) Tanwar, A. S.; Goyal, V. D.; Choudhary, D.; Panjikar, S.; Anand, R. PLoS One 2013, 8, e77781. (51) Quillin, M. L.; Breyer, W. A.; Griswold, I. J.; Matthews, B. W. J. Mol. Biol. 2000, 302, 955. (52) Schiltz, M.; Shepard, W.; Fourme, R.; Prange, T.; de la Fourtelle, E.; Bricogne, G. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1997, 53, 78. (53) Adamek, D. H.; Guerrero, L.; Blaber, M.; Caspar, D. L. D. J. Mol. Biol. 2005, 346, 307. (54) Xu, J.; Baase, W. A.; Quillin, M. L.; Baldwin, E. P.; Matthews, B. W. Protein Sci. 2001, 10, 1067. (55) Krimmer, S. G.; Klebe, G. J. Comput.-Aided Mol. Des. 2015, 29, 867. (56) Baker, B. M.; Murphy, K. P. Biophys. J. 1996, 71, 2049. (57) Leung, C. S.; Leung, S. S. F.; Tirado-Rives, J.; Jorgensen, W. L. J. Med. Chem. 2012, 55, 4489. (58) Barratt, E.; Bingham, R. J.; Warner, D. J.; Laughton, C. A.; Phillips, S. E. V; Homans, S. W. J. Am. Chem. Soc. 2005, 127, 11827. (59) Lee, C.; Maeng, J. S.; Kocher, J. P.; Lee, B.; Yu, M. H. Protein Sci. 2001, 10, 1446. (60) Vallone, B.; Brunori, M. Ital. J. Biochem. 2004, 53, 46.
Gerhard Klebe: 0000-0002-4913-390X Author Contributions †
S.G.K. and J.C. contributed equally.
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS The authors would like to thank George M. Sheldrick, University of Göttingen, and Kevin Cowtan, University of York, for helpful discussion and advice during the revision of the manuscript. The authors thank James Holton (UCSF) for helpful discussions with the application of the END map script. Furthermore, the authors want to thank the MX beamline teams at BESSY II (Berlin-Adlershof, Germany) and Elettra (Triest, Italy) for advice during data collection, in particular Maurizio Polentarutti from Elettra Sincrotrone for help with the noble gas derivatization experiments. The receipt of a travel grant from the Helmholtz-Zentrum Berlin is gratefully acknowledged. This work was supported by grant 268145DrugProfilBind awarded to G.K. by the European Research Council (ERC) of the European Union.
■
REFERENCES
(1) Chandler, D. Nature 2005, 437, 640. (2) Bissantz, C.; Kuhn, B.; Stahl, M. J. Med. Chem. 2010, 53, 5061. (3) Tanford, C. Proc. Natl. Acad. Sci. U. S. A. 1979, 76, 4175. (4) Murphy, K. P.; Privalov, P. L.; Gill, S. J. Science 1990, 247, 559. (5) Frank, H. S.; Evans, M. W. J. Chem. Phys. 1945, 13, 507. (6) Kauzmann, W. Adv. Protein Chem. 1959, 14, 1. (7) Olsson, T. S. G.; Williams, M. A.; Pitt, W. R.; Ladbury, J. E. J. Mol. Biol. 2008, 384, 1002. (8) Southall, N. T.; Dill, K. A.; Haymet, A. D. J. J. Phys. Chem. B 2002, 106, 521. (9) Homans, S. W. Drug Discovery Today 2007, 12, 534. (10) Breiten, B.; Lockett, M. R.; Sherman, W.; Fujita, S.; Lange, H.; Bowers, C. M.; Heroux, A.; Whitesides, G. M.; Al-Sayah, M.; Krilov, G. J. Am. Chem. Soc. 2013, 135, 15579. (11) Krimmer, S. G.; Betz, M.; Heine, A.; Klebe, G. ChemMedChem 2014, 9, 833. (12) Krimmer, S. G.; Cramer, J.; Betz, M.; Fridh, V.; Karlsson, R.; Heine, A.; Klebe, G. J. Med. Chem. 2016, 59, 10530. (13) Biela, A.; Nasief, N. N.; Betz, M.; Heine, A.; Hangauer, D.; Klebe, G. Angew. Chem., Int. Ed. 2013, 52, 1822. (14) Setny, P.; Baron, R.; McCammon, J. A. J. Chem. Theory Comput. 2010, 6, 2866. (15) Snyder, P. W.; Lockett, M. R.; Moustakas, D. T.; Whitesides, G. M. Eur. Phys. J.: Spec. Top. 2014, 223, 853. (16) Snyder, P. W.; Mecinovic, J.; Moustakas, D. T.; Thomas, S. W.; Harder, M.; Mack, E. T.; Lockett, M. R.; Héroux, A.; Sherman, W.; Whitesides, G. M. Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 17889. (17) Matthews, B. W.; Liu, L. Protein Sci. 2009, 18, 494. (18) Wang, L.; Berne, B. J.; Friesner, R. A. Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 1326. (19) Berne, B. J.; Weeks, J. D.; Zhou, R. Annu. Rev. Phys. Chem. 2009, 60, 85. (20) Young, T.; Abel, R.; Kim, B.; Berne, B. J.; Friesner, R. A. Proc. Natl. Acad. Sci. U. S. A. 2007, 104, 808. (21) Qvist, J.; Davidovic, M.; Hamelberg, D.; Halle, B. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 6296. (22) Collins, M. D.; Hummer, G.; Quillin, M. L.; Matthews, B. W.; Gruner, S. M. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 16668. (23) Ernst, J. A.; Clubb, R. T.; Zhou, H. X.; Gronenborn, A. M.; Clore, G. M. Science 1995, 267, 1813. (24) Yu, B.; Blaber, M.; Gronenborn, A. M.; Clore, G. M.; Caspar, D. L. Proc. Natl. Acad. Sci. U. S. A. 1999, 96, 103. 10430
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431
Article
Journal of the American Chemical Society (61) Schiebel, J.; Gaspari, R.; Sandner, A.; Ngo, K.; Gerber, H.-D.; Cavalli, A.; Ostermann, A.; Heine, A.; Klebe, G. Angew. Chem., Int. Ed. 2017, 56, 4887. (62) Eriksson, A. E.; Baase, W. A.; Zhang, X.-J.; Heinz, D. W.; Blaber, M.; Baldwin, E. P.; Matthews, B. W. Science 1992, 255, 178. (63) Kellis, J. T.; Nyberg, K.; Fersht, A. R. Biochemistry 1989, 28, 4914. (64) Cramer, J.; Klebe, G. Synthesis 2017, 49, 1857. (65) Oxford Cryosystems. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1999, 55, 724. (66) Mueller, U.; Darowski, N.; Fuchs, M. R.; Förster, R.; Hellmig, M.; Paithankar, K. S.; Pühringer, S.; Steffien, M.; Zocher, G.; Weiss, M. S. J. Synchrotron Radiat. 2012, 19, 442. (67) Mueller, U.; Förster, R.; Hellmig, M.; Huschmann, F. U.; Kastner, A.; Malecki, P.; Pühringer, S.; Röwer, M.; Sparta, K.; Steffien, M.; Ü hlein, M.; Wilk, P.; Weiss, M. S. Eur. Phys. J. Plus 2015, 130, 141. (68) Thorn, A.; Sheldrick, G. M. J. Appl. Crystallogr. 2011, 44, 1285. (69) Kabsch, W. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 125. (70) McCoy, A. J.; Grosse-Kunstleve, R. W.; Adams, P. D.; Winn, M. D.; Storoni, L. C.; Read, R. J. J. Appl. Crystallogr. 2007, 40, 658. (71) Winn, M. D.; Ballard, C. C.; Cowtan, K. D.; Dodson, E. J.; Emsley, P.; Evans, P. R.; Keegan, R. M.; Krissinel, E. B.; Leslie, A. G. W.; McCoy, A.; McNicholas, S. J.; Murshudov, G. N.; Pannu, N. S.; Potterton, E. A.; Powell, H. R.; Read, R. J.; Vagin, A.; Wilson, K. S. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2011, 67, 235. (72) Holland, D. R.; Tronrud, D. E.; Pley, H. W.; Flaherty, K. M.; Stark, W.; Jansonius, J. N.; McKay, D. B.; Matthews, B. W. Biochemistry 1992, 31, 11310. (73) Schneider, T. R.; Sheldrick, G. M. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2002, 58, 1772. (74) Sheldrick, G. M. Acta Crystallogr., Sect. A: Found. Crystallogr. 2008, 64, 112. (75) Pape, T.; Schneider, T. R. J. Appl. Crystallogr. 2004, 37, 843. (76) Hendrickson, W. A.; Ogata, C. M. Methods Enzymol. 1997, 276, 494. (77) Cowtan, K. Jt. CCP4 ESF-EACBM Newsl. Protein Crystallogr. 1994, 31, 34. (78) Lamzin, V. S.; Wilson, K. S. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1993, 49, 129. (79) Adams, P. D.; Afonine, P. V.; Bunkoczi, G.; Chen, V. B.; Davis, I. W.; Echols, N.; Headd, J. J.; Hung, L.-W.; Kapral, G. J.; GrosseKunstleve, R. W.; McCoy, A. J.; Moriarty, N. W.; Oeffner, R.; Read, R. J.; Richardson, D. C.; Richardson, J. S.; Terwilliger, T. C.; Zwart, P. H. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 213. (80) Emsley, P.; Lohkamp, B.; Scott, W. G.; Cowtan, K. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 486. (81) Molecular Operating Environment (MOE); Chemical Computing Group Inc.: Montreal, Canada. (82) Moriarty, N. W.; Grosse-Kunstleve, R. W.; Adams, P. D. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2009, 65, 1074. (83) The PyMOL Molecular Graphics System; Schrödinger LLC: New York. (84) Durrant, J. D.; Votapka, L.; Sørensen, J.; Amaro, R. E. J. Chem. Theory Comput. 2014, 10, 5047. (85) Gerstein, M.; Richards, F. M.; Chapman, M. S.; Connolly, M. L. In International Tables for Crystallography, Vol. F: Crystallography of biological macromolecules; Rossmann, M. G., Arnold, E., Eds.; Kluwer Academic Publishers: Dordrecht, The Netherlands, 2001; p 531. (86) Voss, N. R.; Gerstein, M. Nucleic Acids Res. 2010, 38, W555. (87) Lang, P. T.; Holton, J. M.; Fraser, J. S.; Alber, T. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 237. (88) Kleywegt, G. J.; Jones, T. A. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1996, 52, 826. (89) Inouye, K.; Kuzuya, K.; Tonomura, B. J. Biochem. 1998, 123, 847. (90) Zhang, Y.-L.; Zhang, Z.-Y. Anal. Biochem. 1998, 261, 139. (91) Zhao, H.; Piszczek, G.; Schuck, P. Methods 2015, 76, 137.
(92) Keller, S.; Vargas, C.; Zhao, H.; Piszczek, G.; Brautigam, C. A.; Schuck, P. Anal. Chem. 2012, 84, 5066. (93) Scheuermann, T. H.; Brautigam, C. A. Methods 2015, 76, 87. (94) Brautigam, C. A.; Zhao, H.; Vargas, C.; Keller, S.; Schuck, P. Nat. Protoc. 2016, 11, 882. (95) Houtman, J. C. D.; Brown, P. H.; Bowden, B.; Yamaguchi, H.; Appella, E.; Samelson, L. E.; Schuck, P. Protein Sci. 2007, 16, 30. (96) Brautigam, C. A. Methods Enzymol. 2015, 562, 109.
10431
DOI: 10.1021/jacs.7b05028 J. Am. Chem. Soc. 2017, 139, 10419−10431