Discovery and Evaluation of Anti-Fibrinolytic Plasmin Inhibitors

Jun 27, 2017 - The best overlay between the data set molecules and the reference (1) was determined using the Tcombo value as optimization criterion. ...
1 downloads 10 Views 3MB Size
Article pubs.acs.org/jcim

Discovery and Evaluation of Anti-Fibrinolytic Plasmin Inhibitors Derived from 5‑(4-Piperidyl)isoxazol-3-ol (4-PIOL) Thomas C. Schmidt,*,† Per-Olof Eriksson,‡ David Gustafsson,§ David Cosgrove,∥ Bente Frølund,# and Jonas Boström† †

Cardiovascular and Metabolic Diseases, Innovative Medicines and Early Development, AstraZeneca, Pepparedsleden 1, SE 43183 Mölndal, Sweden ‡ Structure and Biophysics, Discovery Science, Innovative Medicines and Early Development, AstraZeneca, Pepparedsleden 1, SE 43183 Mölndal, Sweden § Emeriti Pharma, AB, AZ Bioventure Hub, Pepparedsleden 1, SE 43183 Mölndal, Sweden ∥ Discovery Sciences, Chemistry Innovation Centre, Mereside 30S391, Alderley Park, Macclesfield SK10 4TG, United Kingdom # Department of Drug Design and Pharmacology, University of Copenhagen, DK 2100 Copenhagen, Denmark S Supporting Information *

ABSTRACT: Inhibition of plasmin has been found to effectively reduce fibrinolysis and to avoid hemorrhage. This can be achieved by addressing its kringle 1 domain with the known drug and lysine analogue tranexamic acid. Guided by shape similarities toward a previously discovered lead compound, 5-(4-piperidyl)isoxazol-3-ol, a set of 16 structurally similar compounds was assembled and investigated. Successfully, in vitro measurements revealed one compound, 5-(4-piperidyl)isothiazol-3-ol, superior in potency compared to the initial lead. Furthermore, a strikingly high correlation (R2 = 0.93) between anti-fibrinolytic activity and kringle 1 binding affinity provided strong support for the hypothesized inhibition mechanism, as well as revealing opportunities to fine-tune biological effects through minor structural modifications. Several different ligand-based (Freeform, shape, and electrostatic-based similarities) and structure-based methods (e.g., Posit, MM/GBSA, FEP+) were used to retrospectively predict the binding affinities. A combined method, molecular alignment using Posit and scoring with Tcombo, lead to the highest coefficient of determination (R2 = 0.6).



an N-terminal protein domain, five kringle domains with one lysine binding site each, and a serine protease domain (Figure 1).9,10 The serine protease domain constitutes the catalytically active domain, while the kringle domains are important for protein−protein interactions such as binding to fibrin.11 Plasminogen has been reported to adopt a predominantly compact conformation in which only one of the lysine binding sites, that of K1, is accessible. The lysine binding sites of K2− K5 are all buried between adjacent domains and thus inaccessible.12,14 In addition, the lysine binding site of K1 was also reported to have the highest affinity compared to K2−K5

INTRODUCTION The degradation of blood clots by the breakdown of fibrin fibers is a natural control mechanism within the hemostatic system.1 Excessive fibrinolytic activity however is related to disorders such as heavy menstrual bleeding2 or complications during dental surgery for hemophiliacs.3,4 Reducing fibrinolysis is also a means of avoiding excessive blood loss during general surgery5,6 or major trauma.7 The fibrin-degrading enzyme involved in fibrinolysis is plasmin. Its zymogen form, plasminogen, binds to, e.g., fibrin, and is there activated by plasminogen activators t-PA (tissue plasminogen activator) and u-PA (urokinase plasminogen activator).8 In this way, plasminogen is given a localized enzymatic activity. Plasmin is a serine protease and comprises © 2017 American Chemical Society

Received: May 8, 2017 Published: June 27, 2017 1703

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

accepting hydrogen bonds from the Arg35, Tyr64, and Arg71 residues (Figure 2). Consequently, addressing the binding site of K1 with TXA and EACA was found to inhibit fibrinolysis.13,18,19 In absence of the inhibitors, the K1 domain can bind to C-terminal lysine residues anchoring plasmin to its substrate, e.g., fibrin.11 The binding is followed by a large-scale conformational change that exposes the previously shielded binding sites of K2−K5. It also allows for the activation of the zymogen which requires cleavage of the Arg561/Val562 bond, resulting in the catalytically active two-chain form of the serine protease held together via a disulfide bond.10,12 TXA can been administered orally and is the active ingredient in drugs such as Amicar,20 Cyklokapron,3 or Lysteda.2 However, the high required dose of several grams per day (Amicar: 5 × 1 g per day;20 Cyklokapron: 25 mg/kg four times per day;3 Lysteda: 3 × 1.3 g per day2) is associated with a high frequency of side effects such as headaches, nasal symptoms, or back, abdominal, and muscle pain.2,3,20 Despite attempts to reduce these adverse effects, no other fibrinolysis inhibitors have yet reached the market. Searching for novel lead structures, Boströ m et al.21 discovered 5-(4-piperidyl) isoxazol-3-ol (1), and optimizations lead to its derivatives: 5-(3-benzyl-4-piperidyl)isoxazol-3-ol and 5-[(2S,4R)-2-isobutyl-4-piperidyl]isoxazol-3-ol (AZD6564).17 While structurally vastly different from TXA (Tables 1, 2, and 3), 1 was found to exhibit a very high similarity with respect to the three-dimensional shape (Tshape) and electrostatic potential (Telstat) of TXA (Table 3). As the discovery of 1 was a direct result of using the similarity property principle24 on a known molecule (TXA), we set out to further investigate this approach in the identification of other potential plasmin inhibitors. The basis for our investigations is a data set of 16 molecules which was assembled based on similarities with respect to 1 regarding the chemical structures and three-dimensional shape, thus expected to address the desired target. A Clot-Lysis assay was used to investigate if the compounds indeed possess the antifibrinolytic activity expected from the similarities with compound 1. An NMR binding assay was employed to determine affinity of the compounds to the lysine binding site of K1 and may thus provide support for the hypothesized mode of action. A diversity of theoretical methods, e.g., simple molecular descriptors, relative energies, docking scores, and threedimensional shape and electrostatic-based similarity models, were employed. Linear regression models were used to evaluate if any of the methods could be used as a predictive tool to estimate accurate binding affinities for this data set.

Figure 1. Compact structure of plasminogen (pdb code 4a5t12) comprising the N-terminal protein chain (NTP, gray), five kringle domains (K1 yellow, K2 cyan, K3 violet, K4 brown, K5 green), and serine protease domain (blue). The lysine binding site of K1 is highlighted in red, and the complexed ligand tranexamic acid (TXA, overlay with pdb code 1ceb13) in ice blue.

with respect to the binding of small lysine analogues such as εaminocaprotic acid (EACA) and tranexamic acid (TXA).15 A crystal structure of the K1 domain complexed with TXA (pdb code 1ceb)13 reveals that the inhibitor, residing in its zwitterionic tautomeric form, nicely complements the electrostatics of the binding site and establishes hydrogen bonds with the side chains of nearby residues. These findings are supported by theoretical investigations on the interaction between the molecule and its enzyme environment.16 The charged side chains of Asp55 and Asp57 exhibit a highly negative electrostatic potential, while those of Arg35 and Arg71 lead to a highly positive electrostatic potential (Figure 2, residue indices according to Cheng et al.17). Furthermore, the charged amine of the ligand is capable of acting as a 2-fold hydrogen bond donor toward the two Asp55 and Asp57 residues, while the carboxylate of TXA is nearly perfectly complexed by



METHODS Data Set. The 16 molecules used within this study (depicted in Table 4) were selected to investigate the structure activity relationship around the molecular framework of the lead compound 1. Resembling the scaffold of 1 as closely as possible and expected to act via the same hypothesized mode of action, all compounds were assumed to be able to address the lysine binding site of K1 and thus to exhibit anti-fibrinolytic activity. The compounds were found in the literature as well as from the AstraZeneca corporate collection. It should be noted that all compounds, as the lead (1), include a heterocyclic ring system that potentially could mimic the carboxylic acids functionality of TXA and a basic amine moiety. The molecules

Figure 2. Binding of TXA to the lysine binding site of the kringle 1 domain of human plasminogen. The coordinates were taken from the crystal structures pdb code 4cik17 and 1ceb.13 Indexing of the active site residues is according to Cheng et al.17 1704

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Table 1. Molecules Identified To Exhibit Anti-Fibrinolytic Activity by Binding to the Lysine Binding Site of the Kringle 1 Domain of Human Plasmin

Theoretical Methods. The theoretical methods employed within this work were aimed to investigate whether calculations can be used to predict the experimental binding affinities and in how far the similarity property principle could be confirmed for the data set molecules. To set a baseline, we compare with the use of simple molecular descriptors. Different methods for aligning and scoring the compounds were evaluated, including free energy estimations obtained from state-of-the-art MM/ GBSA and FEP calculations carried out within the Prime and FEP+ program package, respectively.31−34 QSAR Model Using Simple Molecular Descriptors. First, several simple descriptors based on the chemical structure of the compounds (such as molecular weight, number of hydrogen bond donor and acceptor sites, polar surface area, and number of rotatable bonds) were computed and correlated with the experimental pKd values (kringle 1, NMR). The molecular volume and polar surface area were computed using the OEChem Toolkit.35 While the polar surface area was computed using an atom-based additive scheme,36 molecular volumes were generated from three-dimensional structures. The Omega conformer generator37 was used to generate a conformational ensemble, out of which the lowest energy minimum was selected. Partition coefficients of the compounds in octanol/ water were also computed via the OEChem Toolkit according to the XlogP algorithm.38 To account for the combination of several descriptors, a principal component analysis was carried out, and the results were correlated with the experimental data. Three Dimensional Similarities. On the basis of the molecular alignments described below, 3D similarities with respect to 1 were computed and correlated with the experimental data. The measures used were the Tanimoto values for similarity in three-dimensional volume (Tshape),39,40 chemical features (Tcolor) represented by so-called color atoms according to the implicit Mills Dean force field,41 and a

Table 2. 2D Similarities of Compounds Shown and Measured by the Tanimoto Value between Their Fingerprints (166 bit MACCS)a TXA 4-PIOL EACA AZD6564

TXA

4-PIOL

EACA

AZD6564

1 0.33 0.67 0.34

0.32 1 0.29 0.86

0.82 0.33 1 0.33

0.33 0.88 0.36 1

a

Values above the diagonal are computed for the nonpolar species; those below the diagonal are computed for the zwitterionic tautomers.

Table 3. Three-Dimensional Similarities between Compounds and TXA as a Reference Derived via an Overlay of Omega Geometries to the X-ray Crystal Structure of TXA as in 1ceb23 molecule

Tshape

Tcolor

Tcombo

Telstat

TXA 4-PIOL EACA AZD6564

0.94 0.77 0.84 0.57

0.87 0.39 0.70 0.33

1.73 1.15 1.52 0.79

0.99 0.96 0.23 0.90

are modeled in their zwitterionic states throughout the study, unless otherwise stated (Table 4). Experimental Procedures. Experimental investigations of the compounds in Table 4 were carried out using two complementary approaches. The binding affinity of the molecules to the lysine binding site of K1 was measured in a competition assay monitored using NMR. The anti-fibrinolytic activity was assessed in a functional assay using human platelet poor blood plasma and measurements were performed using optical spectroscopy. A detailed explanation of the experimental procedures can be found in the Supporting Information. 1705

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Table 4. Chemical Structures of 1 and 15 Structurally Related Compounds Used within This Study Depicted in Their Putative Zwitterionic Ionization State upon Binding to the K1 Lysine Binding Site

combination of the two (Tcombo = Tshape + Tcolor), as well as the Tanimoto value for the overlap in the electrostatic potential (Telstat). The molecular electrostatic potentials were calculated and compared with the OEChem Toolkit35 using van der Waals radii according to Bondi,42 a dielectric constant of 1.0 and conformation-dependent AM1 charges including the bond charge correction.43,44 To account for solvation effects on the electrostatic properties, a second calculation was performed changing the dielectric constant to 80.0 and setting the salt concentration to its default value of 0.04. Ligand-Centric QSAR Model Using Freeform. Molecular alignments needed for the similarity comparisons were obtained in several ways. In a ligand-centric approach, the Freeform tool45 was used to determine the global free energy minimum conformation in solution for compound 1, which was then used as a reference geometry. The data set molecules (1− 16) were prepared by providing their zwitterionic tautomers as SMILES strings to Freeform, and corresponding ensembles of conformers were generated. The best overlay between the data set molecules and the reference (1) was determined using the Tcombo value as optimization criterion. Furthermore, Freeform provides relative energies, relative free energies, and potential strain energies for each conformer. These energies were also correlated with the experimental pKd values. Structure-Centric QSAR Model Using Crystallographic Ligand Geometry. Molecular alignments derived from a (protein) structure-centric approach were generated as

following. The ligand geometry (17), a substituted derivative of 1, within the crystal structure 4cik17 was used as reference. 1 was prepared by deleting the benzyl substituent of ligand 17, leaving the rest of the ligand unchanged. Hydrogen atoms were added using the VIDA molecular visualizer,46 including accounting for the zwitterionic states of the reference compound. Conformational ensembles of the data set molecules were obtained starting from SMILES strings using the OEChem toolkit35 and the Omega conformer generator.47 It should be noted that, in contrast to the previous approach via Freeform, minimizations of the Omega-derived conformations were not carried out as they are not expected to yield geometries closer to the bioactive conformations.48 The best overlay between the data set molecules and the reference (1) was determined using the Tcombo value as optimization criterion, identical to the ligand-centric protocol described above. Structure-Centric QSAR Model via Docking. Molecular alignments derived from another structure-centric approach were realized by pose prediction with the Posit tool49 using the X-ray structure with pdb code 4cik.17 Preparation of the ligand and protein were performed using the make_receptor tool.50 The side chain of Arg35 was modeled in the second of the two, equally probable, orientations provided in the pdb file, as this geometry is better suited for the formation of hydrogen bonds with the ligand. As described above, the 16 data set molecules were provided as SMILES strings (in their zwitterionic states), and three-dimensional geometries were calculated within Posit. 1706

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Table 5. Experimentally Determined Binding Affinities to the Lysine Binding Site of K1 (NMR pKd) and Anti-Fibrinolytic Activity in the Clot-Lysis Assay (Clot-Lysis pIC50)a molecule

1

2

3

4

5

6

7

8

NMR pKd Clot-Lysis pIC50 Clot-Lysis n molecule

7.0 ± 0.3 6.0 ± 0.1 9 9

5.0 ± 0.2 4.1 ± 0.1 2 10

6.7 ± 0.2 6.2 ± 0.1 3 11

6.3 ± 0.1 5.8 ± 0.1 3 12

4.0 ± 0.2 3.6 ± 0.1 2 13

5.0 ± 0.2 4.8 ± 0.1 2 14

7.3 ± 0.1 6.2 ± 0.1 3 15

6.0 ± 0.1 5.6 ± 0.1 3 16

NMR pKd Clot-Lysis pIC50 Clot-Lysis n

5.6 ± 0.2 4.9 ± 0.1 2

6.0 ± 0.1 5.3 ± 0.1 3

6.3 ± 0.2 5.3 ± 0.1 2

4.5 ± 0.1 4.2 ± 0.1 2

5.4 ± 0.2 5.0 ± 0.1 3

6.4 ± 0.1 5.7 ± 0.1 3

6.9 ± 0.1 6.1 ± 0.1 3

6.8 ± 0.1 5.9 ± 0.2 3

The errors given are for the NMR assay, the error of fit of the recovery curve (n = 1), and for the Clot-Lysis assay; the standard deviations of pIC50 values from multiple measurements (Clot-Lysis n). a

Figure 3. Correlation between the experimentally determined binding affinities to the lysine binding site of K1 (NMD pKd) and anti-fibrinolytic activities (Clot-Lysis pIC50) of the 16 compounds.

parameters have been obtained via the force field builder tools within Maestro.58 The actual free energy perturbation calculations were prepared using the FEP mapper tool.57 MM/GBSA. The Posit docking poses and the corresponding receptor were used as initial structures using the MM/GBSA approach. Using the MM-GBSA module together with Prime,31,32 binding free energies for all ligands were estimated applying the minimization or the hierarchical sampling method. To account for the flexibility of the receptor, a region of 5 Å around the ligand was defined as flexible.

While Posit provides a classification ranking of the docked poses (bad, moderate, good, and great) and a probability value, differentiated measures were obtained via the OEChem Toolkit35 using the Chemgauss3,51,52 Chemgauss4,53,54 Chemscore,55 PLP,56 and Shapegauss53 scores which were then correlated to the experimental pKd values. Finally, the obtained poses were investigated for their similarity in Tcombo with respect to the reference compound 1 using the OEChem Toolkit,35 and the determined values were correlated with the experimental binding affinities. All correlations between theoretical measures described above and the experimental pKd values were performed in form of a linear regression model. The quality of the regression model is assessed via the corresponding R2 value as well as in the mean and maximal unsigned error between the, retrospectively, predicted pKd values and the actually measured ones. Free Energy Perturbation Method. The receptor and the obtained docking poses from Posit were used as input to FEP+57 in order to calculate relative free energies of the compounds within the lysine binding site of K1. The receptor was prepared using the protein preparation wizard, and missing



RESULTS Experimental Affinity and Activity Data. The molecules depicted in Table 4 all closely resemble the chemical structure of 1 and were thus considered to be potential binders to the K1 lysine binding site. The binding affinities were determined in competition experiments against a reporter molecule, and measurements were done via NMR (see Methods for details). Their dissociation constants are given in Table 5 (NMR pKd). Because the binding to K1 is hypothesized to be a determining factor in the inhibition of fibrinolysis, all 16 molecules were also investigated for their anti-fibrinolytic activity in a Clot-Lysis 1707

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Figure 4. No correlation found between similarities and the experimental pKd values (NMR, kringle 1). The similarities were computed as Tcombo values for the aligned free energy minima using that of 1 as reference.

Table 6. Quality of Linear Regression Models Assessed by the Coefficient of Determination (R2) as Well as by the Mean and Maximum Error in Predicted pKd Values Method to compute molecular geometries and overlay with reference

Measured property

Linear regression (R2)

Mean unsigned error in predicted pKd

Maximal unsigned error in predicted pKd

none Freeform global free energy minima aligned via Tcombo

simple descriptors Erel dG Tcombo Telstat (ε = 1.0) Telstat (ε = 80.0)

0.65 0.75 0.77 0.78 0.67 0.74

>1.19 1.97 1.96 1.94 1.62 2.02

X-ray (1) + Omega conformers aligned via Tcombo

Tcombo Telstat (ε = 1.0) Telstat (ε = 80.0)

0.39 0.09 0.00

0.60 0.79 0.76

1.53 1.77 1.92

Posit

Chemgauss3 Chemgauss4 Chemscore PLP Shapegauss Tcombo Telstat (ε = 1.0) Telstat (ε = 80.0)

0.16 0.28 0.05 0.11 0.00 0.58 0.14 0.05

0.72 0.63 0.73 0.70 0.91 0.47 0.76 0.76

1.72 1.57 2.14 1.99 2.71 1.38 1.71 2.04

FEP+

dG

0.16

0.71

1.55

MM/GBSA

dG (minimization) dG (hierarchical sampling)

0.06 0.20

0.75 0.59

2.13 1.90

assay using human blood plasma (see Table 5, Clot-Lysis pIC50). The compounds in the data set were found to exhibit a range of anti-fibrinolytic activities and to bind to K1 to various degrees. Most compounds were found to have lower binding affinities than the lead compound 1. However, three compounds (3, 15, and 16) yielded comparable binding affinities, and one compound (7) was found to even surpass that of 1. In agreement with the hypothesis, compounds exhibiting a high affinity to K1 were also found to have a stronger anti-fibrinolytic activity. Compound 16 was deter-

mined to be similarly potent as 1, and the activities of compounds 3, 7, and 15 were found to exceed that of 1. A linear regression between the two measures is shown in Figure 3. Confirming our hypothesis, the anti-fibrinolytic activity and the capability to bind to the K1 binding site was found to correlate with each other. In fact, the correlation between the two experimentally measured values is astonishingly high. The coefficient of determination is R2 = 0.93 for a linear regression. Hence, the binding affinity of the compounds to K1 could not only be confirmed as an important factor for anti-fibrinolytic 1708

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Figure 5. Correlation between similarities and experimental pKd values. The similarities were computed as Tcombo values of the best alignment of Omega conformers to the geometry of 1 derived from the crystal structure 4cik.17

Figure 6. Overlay of the docking (Posit) poses for all 16 compounds with the K1 lysine binding site (pdb code 4cik). Left: Three dimensional shape of the binding pocket colored by its electrostatic potential. Right: Hydrogen bond network between the compounds and nearby protein residues.

comparison of their electrostatic potentials exhibits almost no diversity between the 16 compounds with Telstat values all exceeding 0.9. This is a consequence of the molecules being modeled as zwitterions with two opposing charges dominating the comparison of the electrostatic potentials. Another consequence arising from the two opposing charges is that the minimum geometries do not resemble the bioactive conformation of 1. Hence, a reasonable correlation between any of the Tanimoto values and the binding affinities could not be found (Figure 4 and Table 6). Structure-Centric QSAR Model Using Crystallographic Ligand Geometry. Using the conformational ensemble generated with Omega and comparing them to the geometries of 1 prepared from the crystal structure also led to high similarities in shape (Tshape > 0.8) and chemical resemblance (Tcolor > 0.6; Tcombo > 1.6). A modest correlation was found (R2 = 0.39, see Figure 5). The comparison of the electrostatic potentials of the 15 molecules with respect to that of 1 resulted in Telstat values close to 1.0 indicating almost identity between all compounds. This is a result of the molecules being modeled in their zwitterionic tautomers, a prerequisite for addressing the

activity but also was found to translate almost directly into antifibrinolytic activity. QSAR Model Using Simple Molecular Descriptors. Performing a linear regression between the simple molecular descriptors and the measured binding affinities resulted in almost no correlation for most descriptors. A modest correlation could only be found for the polar surface area of the molecules, where the R2 value was determined to be close to 0.40 (Table 3). A principal component analysis was carried out on the molecular descriptors. The resulting linear regression model from the PCA did however not show to be superior compared to the use of different descriptors individually. Ligand-Centric QSAR Model Using Freeform. The conformational ensembles obtained by using Freeform contained geometries that could well be superimposed with the calculated global free energy minimum in the aqueous solution of 1 as indicated by Tshape values of 0.7 and larger. While the molecular shape was found to be very similar among the 16 compounds, the structural diversity is reflected in the larger spread of Tcombo values ranging from 1.2 to 2.0. The 1709

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Figure 7. Linear regression between the ligand similarity measure (Tcombo) calculated for the poses obtained from Posit and the experimentally measured binding affinities (kringle 1, NMR pKd).

Figure 8. Linear regression between the computed binding free energies (FEP+) and the experimentally determined binding affinities (NMR pKd). R2 is low (0.16), indicating no correlation.

poses were rescored using the OEChem Toolkit35 and the scoring functions Chemgauss3,51,52 Chemgauss4,51,52 Chemscore,55 PLP,56 and Shapegauss.53 The scoring functions are developed to cover intermolecular interactions and provide estimates for the binding affinities. However, the correlations between the computed scores and the experimental pKd values are very low for any of the employed functions. The highest correlation was found for the Chemgauss3 and Chemgauss4 scoring functions (R2 ≈ 0.25), while using Chemscore, PLP, or Shapegauss resulted in R2 values below 0.10 (Table 6). This is remarkable as the correlation of experimental binding affinities with respect to the Tcombo values for the Posit poses lead to the highest coefficient of determination (R2 = 0.58; Figure 7) within this study. Therefore, in the present case, the ligand-based similarity measures clearly outperform the structure-based docking scoring functions. Estimates on Binding Affinities via Free Energy Perturbation Methods. The FEP calculations yielded relative

highly polar lysine binding site. The two opposing charges, however, clearly dominate the comparison of electrostatic potentials as their relative magnitude masks any other influences arising, e.g., from the exchange of heteroatoms between 1 and 5. Hence, no distinction between the compounds based on the Telstat values is possible, independent of the choice of ε, and no correlation with the experimental binding affinities could be found (R2 < 0.15). Structure-Centric QSAR Model via Docking. The molecular alignments obtained from the Posit docking poses for all 16 compounds (Figure 6) exhibit high similarities with respect to the reference molecule (Tshape > 0.9), especially with respect to the heterocycles. Thus, they also form a topologically identical network of hydrogen bonds with the binding site residues. In line with the almost perfect spatial overlap, the Tcombo (>1.6) and Telstat (>0.9) measures indicate extraordinarily high similarity between the molecules. All poses were ranked at least “good” within Posit and were assigned reasonable probabilities ranging from 0.42 to 0.72. All docking 1710

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

involving nitrogen or oxygen atoms.55,59 This can however play an important role such as in the case of compounds 1 and 6. Both molecules exhibit topologically identical hydrogen bond networks with the receptor; however, as 2 provides a nitrogen atom as the acceptor for the hydrogen bonds donated by Arg35 and Arg71, this position is replaced by an oxygen atom in 6. The resulting hydrogen bonds (N−H···O) are about 5−10 kcal/mol weaker than their analogues for compound 1 bridging between two nitrogen atoms (N−H···N).60,61 Because of two hydrogen bonds being affected, the interaction between 6 (pKd = 5.0) and K1 is much weaker than that of 1 (pKd = 7.0), which is also reflected in their binding affinities. The same effect can be observed for 5 (pKd = 4.0), which is found to have the lowest affinity within the series. None of the used scoring functions fully accounts for this observation. Encouragingly, the linear regression using the Tcombo measure for the molecular alignment obtained from the Posit docking poses resulted in a reasonably high R2 value of 0.58 and thus yielded a decently accurate prediction of pKd values (Table 6). Posit takes into account the steric demands of the binding pocket as well as flexibility in the geometry of the compounds.62 Thereby, the obtained docking poses appear to resemble the putative binding geometries of the compounds more closely, leading to a more reliable model compared to the ligand-centric one. The putative bioactive geometries of the data set compounds in complex with the lysine binding site of the K1 domain obtained from docking support the hypothesis of a common binding mode. It is characterized by the complementary electrostatic properties between the compounds and the binding site and a network of hydrogen bonds common between all data set molecules. Disturbances of the hydrogen bond network, e.g., by the exchange of the isoxazole nitrogen atom (compounds 5 and 6) or the electrostatic properties of the interface with the target (compounds 5 and 9), lead to significantly reduced affinities. Other modifications that do not interfere with these points are widely tolerated. In the future, we intend to investigate if improvements on the ligand-centric approaches can be achieved by extending the current rigid-body alignment of query compounds by a method allowing for a flexible alignment and refinements in the preparation of the reference molecule. Free Energy Perturbation Methods. Despite its status as state-of-the-art method for the estimation of binding free energies,34,63 the computed values exhibit hardly any correlation with the experimental measurements. This indicates the challenges of force field-based methods to describe heteroaromatic compounds or highly charged species. MM/GBSA. Just as the free energy perturbation approach, the MM/GBSA method, too, failed to reproduce the trend of binding affinities observed in the experiment, most probably due to the same challenges as described above.

binding free energies for the 16 compounds spanning a range of around 10 kcal/mol. A correlation of these energies with the experimental binding affinities only resulted in a R2 value of 0.16 (Figure 8). Estimates on Binding Affinities via MM/GBSA. The MM/GBSA method using the minimization approach resulted in binding free energies between −30 and −20 kcal/mol. A significant correlation with the experimental data could not be found. The coefficient of determination obtained from the hierarchical approach (R2 = 0.20) does imply some weak correlation; however, the predicted trend of binding energies is opposite to that from the experimental measurements.



DISCUSSION The observed correlation between the experimentally determined binding affinities and anti-fibrinolytic activities supports the hypothesis of a common mode of action between all 16 compounds by blocking the lysine binding site of the K1 domain. A common binding mode is also supported by the large degree of similarity between the molecules (Tshape values greater than 0.8; Tcombo larger than 1.6). The binding affinities as well as the anti-fibrinolytic activity were found to differ widely spanning a range of about 3 log units each, which indicates that small structural changes in the chemical structures can be used to fine-tune biological effects. QSAR Model Using Simple Molecular Descriptors. The comparison of simple molecular descriptors and the experimental binding affinities was performed to rule out the possibility of a straightforward prediction of pKd values. The almost complete lack of any correlation show that neither individual descriptors nor any combination thereof can be used as predictors for binding affinity or activity. Ligand-Centric QSAR Model Using Freeform. The ligand-centric approach using Freeform, which does not require the demanding determination of a crystal structure of the lead compound complexed with the target, was hampered by the fact that the global minimum of 1 does not correspond to the bioactive form observed in the crystal structure. While the putative bioactive conformation of 1 resembles an extended conformation with the isoxazole moiety in a favorable equatorial position with respect to the pyridine ring, Freeform overestimated the attraction between these charges leading to a compact conformer with the isoxazole in the axial position to be considered as global minimum. The same holds true for all compounds in the data set, which upon binding have to adopt other geometries than those of their (Freeform) calculated global free energy minima. Hence, no correlation with the experimental binding affinities could be found, neither for the obtained energies nor for similarities between the computed three-dimensional geometries (Table 6, Freeform). Structure-Centric QSAR Model Using Crystallographic Ligand Geometry. Accounting for the putative bioactive conformations by comparing the conformational ensembles generated using Omega and using the geometry of 1 derived from the crystal structure 4cik17 as reference, a higher R2 value (Table 6, X-ray (1) + Omega) with the measured pKd values could be obtained. Structure-Centric QSAR Model via Docking. All employed scoring functions fail in predicting the experimental binding affinities accurately. We think that this is due to intrinsic deficiencies of the used scoring functions. For example, while hydrogen bonding is included in the scoring functions, they do not properly discriminate between hydrogen bonds



CONCLUSION The investigated compounds were found to exhibit a range of anti-fibrinolytic activities. Three of the compounds (3, 15 and 16) show activities on par with the previously discovered lead molecule (1), while a fourth compound (7) was found to be even more potent, making it one of the most anti-fibrinolytically active small molecules known today. Providing solid evidence for the mechanistic hypothesis, the anti-fibrinolytic activity increased with an increase in the binding affinity to the K1 domain of plasmin. The almost 1711

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling

Clot-Lysis assay; B.F. synthesis; T.C.S. and J.B. theoretical investigations.

perfect correlation between the two measurements is striking and strongly supports that the binding affinity to K1 is a decisive factor to obtain anti-fibrinolytic activity. It is also in line with the hypothesis of a compact conformation of plasminogen,12 in which the other kringle domains (K2−K5) are buried within the protein and thus inaccessible for binding. The benefit of improved affinity to the K1 domain may, for example, show added value in the quest for novel fibrinolysis inhibitors with improved overall properties viz. permeable compounds without the side effects often associated with high therapeutic doses.17 The docking methods alone failed to generate predictive models. However, using the similarity principle to model the binding affinities of the compounds based on shape similarities with respect to an active reference compound was found to be a useful approach. The methods used to obtain the molecular alignments, as well as the choice of the reference geometry, were also found to play significant roles in the quality of the derived models. For example, while comparing the Freeform approach with that using the crystallographic geometry of the reference compound, it clearly shows that using a molecular geometry that resembles the bioactive conformation of the reference molecule as closely as possible is beneficial. While using the X-ray reference geometry resulted in a decent correlation with the experimental data, extending the ligandcentric approach with information on the protein environment using the Posit docking tool improved the predictive capability further. That is, the molecular alignment obtained from Posit, subsequently scored by Tcombo, resulted in the most predictive model (R2 = 0.58). This underlines the importance of using all available information when modeling experimental data. In summary, while the predictions of binding affinities using the methods in the current study were found to be reasonable, it should be noted that the compounds all include heterocyclic systems, which are notoriously challenging for such predictions.



Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS The authors thank Andrey Frolov for giving insightful comments and suggestions during preparation of the manuscript.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jcim.7b00255. SMILES strings of all molecules, synthesis protocol for compound 14, description of the experimental determination of kringle 1 NMR binding affinities, description of the experimental determination of Clot-Lysis activity, scatterplots illustrating the correlation between the theoretically computed properties with the kringle 1 NMR pKd values, and table of compound pairs used to create the FEP map. PDF) Jmol image. (MOL) Jmol image. (MOL) Jmol image. (MOL)



REFERENCES

(1) Gale, A. J. Current Understanding of Hemostasis. Toxicol. Pathol. 2011, 39, 273−280. (2) Ferring Pharmaceuticals Inc. Lysteda - Prescribing Information. December 7, 2016. http://www.ferringusa.com/wp-content/uploads/ 2016/07/LystedaPI_3.2016.pdf (accessed 2016−12−07). (3) Pfitzer Canada, Inc. Cyklokapron - Prescribing Information. December 7, 2016. https://www.pfizer.ca/sites/g/files/g10028126/f/ 201603/CYKLOKAPRON_PM_E_185812_22_Feb_2016.pdf (accessed 2016−12−07). (4) Forbes, C. D.; Barr, R. D.; Reid, G.; Thomson, C.; Prentice, C. R. M.; Nicol, G. P. M.; Douglas, A. S. Tranexamic Acid in Control of Haemorrhage after Dental Extraction in Haemophilia and Christmas Disease. Br Med. J. 1972, 2, 311−313. (5) Veien, M.; Sørensen, J. V.; Madsen, F.; Juelsgaard, P. Tranexamic Acid given Intraoperatively Reduces Blood Loss after Total Knee Replacement: A Randomized, Controlled Study. Acta Anaesthesiol. Scand. 2002, 46, 1206−1211. (6) Benoni, G.; Lethagen, S.; Fredin, H. The Effect of Tranexamic Acid on Local and Plasma Fibrionolysis during Total Knee Arthroplasty. Thromb. Res. 1997, 85, 195−206. (7) Binz, S.; McCollester, J.; Thomas, S.; Miller, J.; Pohlman, T.; Waxman, D.; Shariff, F.; Tracy, R.; Walsh, M. CRASH-2 Study of Tranexamic Acid to Treat Bleeding in Trauma Patients: A Controversy Fueled by Science and Social Media. J. Blood Transfus. 2015, 2015, 1− 12. (8) Alkjaersig, N.; Fletcher, A. P.; Sherry, S. The Mechanism of Clot Dissolution by Plasmin. J. Clin. Invest. 1959, 38, 1086−1095. (9) Novokhatny, V. V.; Kudinov, S. A.; Privalov, P. L. Domains in Human Plasminogen. J. Mol. Biol. 1984, 179, 215−232. (10) Wiman, B.; Wallén, P. Structural Relationship between “Glutamic Acid” and “Lysine” Forms of Human Plasminogen and Their Interaction with the NH2-Terminal Activation Peptide as Studied by Affinity Chromatography. Eur. J. Biochem. 1975, 50, 489− 494. (11) Wiman, B.; Collen, D. Molecular Mechanism of Physiological Fibrinolysis. Nature 1978, 272 (5653), 549−550. (12) Xue, Y.; Bodin, C.; Olsson, K. Crystal Structure of the Native Plasminogen Reveals an Activation-Resistant Compact Conformation. J. Thromb. Haemostasis 2012, 10, 1385−1396. (13) Mathews, I. I.; Vanderhoff-Hanaver, P.; Castellino, F. J.; Tulinsky, A. Crystal Structures of the Recombinant Kringle 1 Domain of Human Plasminogen in Complexes with the Ligands E-Aminocaproic Acid and Trans-4-(Aminomethyl)cyclohexane-1-Carboxylic Acid. Biochemistry 1996, 35, 2567−2576. (14) Lerch, P. G.; Rickli, E. E.; Lergier, W.; Gillessen, D. Localization of Individual Lysine-Binding Regions in Human Plasminogen and Investigations on Their Complex-Forming Properties. Eur. J. Biochem. 1980, 107, 7−13. (15) Markus, G.; Priore, R. L.; Wissler, F. C. The Binding of Tranexamic Acid to Native (Glu) and Modified (Lys) Human Plasminogen and Its Effect on Conformation. Thromb. Res. 1979, 254, 1211−1216. (16) Mladenovic, M.; Arnone, M.; Fink, R. F.; Engels, B. Environmental Effects on Charge Densities of Biologically Active Molecules: Do Molecule Crystal Environments Indeed Approximate Protein Surroundings? J. Phys. Chem. B 2009, 113, 5072−5082.

AUTHOR INFORMATION

Corresponding Author

*Phone: +46 31 7065251. E-mail: thomas.schmidt@ astrazeneca.com. ORCID

Thomas C. Schmidt: 0000-0001-5728-0160 Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. P.-O.E., NMR binding measurements; D.G., 1712

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling (17) Cheng, L.; Pettersen, D.; Ohlsson, B.; Schell, P.; Karle, M.; Evertsson, E.; Pahlén, S.; Jonforsen, M.; Plowright, A. T.; Boström, J.; Fex, T.; Thelin, A.; Hilgendorf, C.; Xue, Y.; Wahlund, G.; Lindberg, W.; Larsson, L.-O.; Gustafsson, D. Discovery of the Fibrinolysis Inhibitor AZD6564, Acting via Interference of a Protein−Protein Interaction. ACS Med. Chem. Lett. 2014, 5, 538−543. (18) Winn, E. S.; Hu, S.-P.; Hochschwender, S. M.; Laursen, R. A. Studies on the Lysine-Binding Sites of Human Plasminogen. Eur. J. Biochem. 1980, 104, 579−586. (19) Lucas, M. A.; Fretto, L. J.; McKee, P. A. The Binding of Human Plasminogen to Fibrin and Fibrinogen. J. Biol. Chem. 1983, 258, 4249− 4256. (20) Clover Pharmaceuticals Corp. Amicar - Prescribing Information. December 7, 2016. http://www.akorn.com/documents/catalog/ package_inserts/49411-052-08.pdf (accessed 2016−12−07). (21) Boström, J.; Grant, J. A.; Fjellström, O.; Thelin, A.; Gustafsson, D. Potent Fibrinolysis Inhibitor Discovered by Shape and Electrostatic Complementarity to the Drug Tranexamic Acid. J. Med. Chem. 2013, 56, 3273−3280. (22) Frolund, B.; Kristiansen, U.; Brehm, L.; Hansen, A. B.; Krogsgaard-Larsen, P.; Falch, E. Partial GABAA Receptor Agonists. Synthesis and in Vitro Pharmacology of a Series of Nonannulated Analogs of 4,5,6,7-Tetrahydroisoxazolo[4,5-C]pyridin-3-Ol. J. Med. Chem. 1995, 38, 3287−3296. (23) Mathews, I. I.; Vanderhoff-Hanaver, P.; Castellino, F. J.; Tulinsky, A. Crystal Structures of the Recombinant Kringle 1 Domain of Human Plasminogen in Complexes with the Ligands E-Aminocaproic Acid and Trans-4-(Aminomethyl)cyclohexane-1-Carboxylic Acid. Biochemistry 1996, 35, 2567−2576. (24) Ginn, C. M. R.; Turner, D. B.; Willett, P.; Ferguson, A. M.; Heritage, T. W. Similarity Searching in Files of Three-Dimensional Chemical Structures: Evaluation of the EVA Descriptor and Combination of Rankings Using Data Fusion. J. Chem. Inf. Comput. Sci. 1997, 37, 23−37. (25) Frølund, B.; Jørgensen, A. T.; Tagmose, L.; Stensbøl, T. B.; Vestergaard, H. T.; Engblom, C.; Kristiansen, U.; Sanchez, C.; Krogsgaard-Larsen, P.; Liljefors, T. Novel Class of Potent 4-Arylalkyl Substituted 3-Isoxazolol GABAA Antagonists: Synthesis, Pharmacology, and Molecular Modeling. J. Med. Chem. 2002, 45, 2454−2468. (26) De Amici, M.; Frølund, B.; Hjeds, H.; Krogsgaard-Larsen, P. Analogues of the Low-Efficacy Partial GABAA Agonist 4-PIOL. Syntheses and in Vitro Pharmacological Studies. Eur. J. Med. Chem. 1991, 26, 625−631. (27) Krehan, D.; í Storustovu, S.; Liljefors, T.; Ebert, B.; Nielsen, B.; Krogsgaard-Larsen, P.; Frølund, B. Potent 4-Arylalkyl-Substituted 3Isothiazolol GABAA Competitive/Noncompetitive Antagonists: Synthesis and Pharmacology. J. Med. Chem. 2006, 49, 1388−1396. (28) Jansen, M.; Rabe, H.; Strehle, A.; Dieler, S.; Debus, F.; Dannhardt, G.; Akabas, M. H.; Lüddens, H. Synthesis of GABAA Receptor Agonists and Evaluation of Their A-Subunit Selectivity and Orientation in the GABA Binding Site. J. Med. Chem. 2008, 51, 4430− 4448. (29) Krall, J.; Kongstad, K. T.; Nielsen, B.; Sørensen, T. E.; Balle, T.; Jensen, A. A.; Frølund, B. 5-(Piperidin-4-Yl)-3-Hydroxypyrazole: A Novel Scaffold for Probing the Orthosteric Γ-Aminobutyric Acid Type A Receptor Binding Site. ChemMedChem 2014, 9, 2475−2485. (30) Møller, H. A.; Sander, T.; Kristensen, J. L.; Nielsen, B.; Krall, J.; Bergmann, M. L.; Christiansen, B.; Balle, T.; Jensen, A. A.; Frølund, B. Novel 4-(Piperidin-4-Yl)-1-Hydroxypyrazoles as Γ-Aminobutyric AcidA Receptor Ligands: Synthesis, Pharmacology, and Structure− Activity Relationships. J. Med. Chem. 2010, 53, 3417−3421. (31) Jacobson, M. P.; Pincus, D. L.; Rapp, C. S.; Day, T. J. F.; Honig, B.; Shaw, D. E.; Friesner, R. A. A Hierarchical Approach to All-Atom Protein Loop Prediction. Proteins: Struct., Funct., Genet. 2004, 55, 351− 367. (32) Jacobson, M. P.; Friesner, R. A.; Xiang, Z.; Honig, B. On the Role of the Crystal Environment in Determining Protein Side-Chain Conformations. J. Mol. Biol. 2002, 320, 597−608.

(33) Harder, E.; Damm, W.; Maple, J.; Wu, C.; Reboul, M.; Xiang, J. Y.; Wang, L.; Lupyan, D.; Dahlgren, M. K.; Knight, J. L.; Kaus, J. W.; Cerutti, D. S.; Krilov, G.; Jorgensen, W. L.; Abel, R.; Friesner, R. A. OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J. Chem. Theory Comput. 2016, 12, 281−296. (34) Wang, L.; Wu, Y.; Deng, Y.; Kim, B.; Pierce, L.; Krilov, G.; Lupyan, D.; Robinson, S.; Dahlgren, M. K.; Greenwood, J.; Romero, D. L.; Masse, C.; Knight, J. L.; Steinbrecher, T.; Beuming, T.; Damm, W.; Harder, E.; Sherman, W.; Brewer, M.; Wester, R.; Murcko, M.; Frye, L.; Farid, R.; Lin, T.; Mobley, D. L.; Jorgensen, W. L.; Berne, B. J.; Friesner, R. A.; Abel, R. Accurate and Reliable Prediction of Relative Ligand Binding Potency in Prospective Drug Discovery by Way of a Modern Free-Energy Calculation Protocol and Force Field. J. Am. Chem. Soc. 2015, 137, 2695−2703. (35) OEChem Toolkit; Openeye Scientific Software, Inc.: Santa Fe, NM, 2017. (36) Ertl, P.; Rohde, B.; Selzer, P. Fast Calculation of Molecular Polar Surface Area as a Sum of Fragment-Based Contributions and Its Application to the Prediction of Drug Transport Properties. J. Med. Chem. 2000, 43, 3714−3717. (37) Hawkins, P. C. D.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T. Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database. J. Chem. Inf. Model. 2010, 50, 572−584. (38) Wang, R.; Fu, Y.; Lai, L. A New Atom-Additive Method for Calculating Partition Coefficients. J. Chem. Inf. Comput. Sci. 1997, 37, 615−621. (39) Grant, J. A.; Pickup, B. T. A Gaussian Description of Molecular Shape. J. Phys. Chem. 1995, 99, 3503−3510. (40) Grant, J. A.; Gallardo, M. A.; Pickup, B. T. A Fast Method of Molecular Shape Comparison: A Simple Application of a Gaussian Description of Molecular Shape. J. Comput. Chem. 1996, 17, 1653− 1666. (41) Mills, J. E. J.; Dean, P. M. Three-Dimensional Hydrogen-Bond Geometry and Probability Information from a Crystal Survey. J. Comput.-Aided Mol. Des. 1996, 10, 607−622. (42) Bondi, A. Van Der Waals Volumes and Radii. J. Phys. Chem. 1964, 68, 441−451. (43) Jakalian, A.; Bush, B. L.; Jack, D. B.; Bayly, C. I. Fast, Efficient Generation of High-Quality Atomic Charges. AM1-BCC Model: I. Method. J. Comput. Chem. 2000, 21, 132−146. (44) Jakalian, A.; Jack, D. B.; Bayly, C. I. Fast, Efficient Generation of High-Quality Atomic Charges. AM1-BCC Model: II. Parameterization and Validation. J. Comput. Chem. 2002, 23, 1623−1641. (45) Freeform; Openeye Scientific Software, Inc.: Santa Fe, NM, 2013. (46) VIDA; Openeye Scientific Software, Inc.: Santa Fe, NM. (47) Omega; Openeye Scientific Software, Inc.: Santa Fe, NM, 2004. (48) Boström, J.; Greenwood, J. R.; Gottfries, J. Assessing the Performance of OMEGA with Respect to Retrieving Bioactive Conformations. J. Mol. Graphics Modell. 2003, 21, 449−462. (49) Posit; Openeye Scientific Software, Inc.: Santa Fe, NM, 2011. (50) MakeReceptor; Openeye Scientific Software, Inc.: Santa Fe, NM, 2015. (51) McGann, M. FRED Pose Prediction and Virtual Screening Accuracy. J. Chem. Inf. Model. 2011, 51, 578−596. (52) McGann, M. FRED and HYBRID Docking Performance on Standardized Datasets. J. Comput.-Aided Mol. Des. 2012, 26, 897−906. (53) McGann, M. R.; Almond, H. R.; Nicholls, A.; Grant, J. A.; Brown, F. K. Gaussian Docking Functions. Biopolymers 2003, 68, 76− 90. (54) McGaughey, G. B.; Sheridan, R. P.; Bayly, C. I.; Culberson, J. C.; Kreatsoulas, C.; Lindsley, S.; Maiorov, V.; Truchon, J.-F.; Cornell, W. D. Comparison of Topological, Shape, and Docking Methods in Virtual Screening. J. Chem. Inf. Model. 2007, 47, 1504−1519. (55) Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee, R. P. Empirical Scoring Functions: I. The Development of a Fast Empirical Scoring Function to Estimate the Binding Affinity of Ligands 1713

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714

Article

Journal of Chemical Information and Modeling in Receptor Complexes. J. Comput.-Aided Mol. Des. 1997, 11, 425− 445. (56) Verkhivker, G. M.; Bouzida, D.; Gehlhaar, D. K.; Rejto, P. A.; Arthurs, S.; Colson, A. B.; Freer, S. T.; Larson, V.; Luty, B. A.; Marrone, T.; Rose, P. W. Deciphering Common Failures in Molecular Docking of Ligand-Protein Complexes. J. Comput.-Aided Mol. Des. 2000, 14, 731−751. (57) FEP+; Schrodinger, LLC: New York, 2017. (58) Maestro; Schrodinger, LLC: New York, 2017. (59) Scoring - Toolkits -- Python. https://docs.eyesopen.com/ toolkits/python/dockingtk/scoring.html (accessed 2017−06−22). (60) Steiner, T. The Hydrogen Bond in the Solid State. Angew. Chem., Int. Ed. 2002, 41, 48−76. (61) Grabowski, S. J. What Is the Covalency of Hydrogen Bonding? Chem. Rev. 2011, 111, 2597−2625. (62) Kelley, B. P.; Brown, S. P.; Warren, G. L.; Muchmore, S. W. POSIT: Flexible Shape-Guided Docking For Pose Prediction. J. Chem. Inf. Model. 2015, 55, 1771−1780. (63) Sherborne, B.; Shanmugasundaram, V.; Cheng, A. C.; Christ, C. D.; DesJarlais, R. L.; Duca, J. S.; Lewis, R. A.; Loughney, D. A.; Manas, E. S.; McGaughey, G. B.; Peishoff, C. E.; van Vlijmen, H. Collaborating to Improve the Use of Free-Energy and Other Quantitative Methods in Drug Discovery. J. Comput.-Aided Mol. Des. 2016, 30, 1139−1141.

1714

DOI: 10.1021/acs.jcim.7b00255 J. Chem. Inf. Model. 2017, 57, 1703−1714