Identifying Interactions that Determine Fragment Binding at Protein

Subscriber access provided by UOW Library

Article

Identifying Interactions that Determine Fragment Binding at Protein Hotspots Chris John Radoux, Tjelvar S. G. Olsson, William Ross Pitt, Colin R. Groom, and Tom L. Blundell J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.5b01980 • Publication Date (Web): 04 Apr 2016 Downloaded from http://pubs.acs.org on April 8, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Medicinal Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Medicinal Chemistry

Identifying Interactions that Determine Fragment Binding at Protein Hotspots Chris J. Radoux*a,d, Tjelvar S. G. Olssona,b, Will R. Pittc, Colin R. Grooma, Tom L. Blundelld a) Cambridge Crystallographic Data Centre, 12 Union Road, Cambridge, CB2 1EZ, United Kingdom, Email: [email protected] b) John Innes Centre, Norwich Research Park, Colney Ln, Norwich, Norfolk NR4 7UH c) UCB, 208 Bath Rd, Slough, West Berkshire SL1 3WE d) Department of Biochemistry, University of Cambridge, Sanger Building, 80 Tennis Ct Rd, Cambridge CB2 1GA KEYWORDS Hotspots, Fragments, Structure-based drug design, Computer-aided drug design

Abstract

Locating a ligand-binding site is an important first step in structure-guided drug discovery, but current methods do little to suggest which interactions within a pocket are the most important for binding. Here we illustrate a method that samples atomic hotspots with simple molecular probes to produce fragment hotspot maps. These maps specifically highlight fragment-binding sites and their corresponding pharmacophores. For ligand-bound structures, they provide an intuitive visual guide within the binding site, directing medicinal chemists

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

where to grow the molecule and alerting them to sub-optimal interactions within the original hit. The fragment hotspot map calculation is validated using experimental binding positions of 21 fragments and subsequent lead molecules. The ligands are found in high scoring areas of the fragment hotspot maps, with fragment atoms having a median percentage rank of 97%. Protein kinase B and pantothenate synthetase are examined in detail. In each case the fragment hotspot maps are able to rationalise a Free-Wilson analysis of SAR data from a fragment-based drug design project.

Introduction Many proteins have pockets that have evolved to bind small molecules, and within these pockets are hotspots: areas that make a disproportionately large contribution to binding affinity1. Existing methods for hotspot prediction take a variety of approaches, resulting in different assumptions about their nature. These methods are grouped as follows: Atomic hotspot methods: GRID2 and SuperStar3 are two well established programs that are able to locate favourable positions for atomic probes on a protein surface. Atomic hotspots should not be confused with molecular hotspots, which are commonly referred to just as “hotspots”. Atomic hotspots can be thought of as a single interaction, which would not necessarily result in an environment suitable for ligand binding. GRID places an atomic probe at each point on a 3D grid placed over a protein, and calculates favourable positions for a given probe using force fields. SuperStar uses data from IsoStar4, a library of molecular contacts in the Cambridge Structural Database (CSD)5, to give a propensity for a given probe type at each grid point within the cavity. If an interaction between two groups at a certain distance and angle is favourable, it will occur more frequently in crystal structures and


Page 2 of 35

Page 3 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


therefore have a greater propensity in the SuperStar output. These methods are useful, but they find many regions of high interaction propensity throughout the protein surface and cannot be used directly for the detection of molecular hotspots. Consensus site methods: Sub-pockets that are found to bind a variety of chemically-diverse probe molecules can be referred to as consensus sites. The number of different probes that bind to the site is considered rather than the binding affinity of the probe. Consensus sites can be identified experimentally through Multiple Solvent Crystal Structures (MSCS)6–8 and fragment library screening9, or predicted computationally using the force field based method, FTMap10–13. FTMap uses 16 small molecule probes, which are either purely hydrophobic or contain one or two polar functional groups. FTMap ranks its hotspots by counting the number of probes that bind to a given cluster, resulting in a consensus site, reflecting results from experimental multiple solvent crystal structures. In one of their more recent papers14, Kozakov et al. suggest that hotspots are created by a mosaic of polar and apolar patches on the protein surface. The simplicity of the probes allows many of them to match the hotspot, making the single polar interaction (if required) and placing their carbon atoms in the hydrophobic region surrounding it. 'Unhappy' water site methods: The role of binding site waters is becoming increasingly prominent in computational methods applied to structure-guided drug discovery15–21. Molecular dynamics methods such as WaterMap19 calculate the thermodynamic properties of hydration sites, identifying 'unhappy water' sites. Mixed solvent molecular dynamics: MDMix22 uses three 20 ns molecular dynamics simulations, with one in the presence of 20% ethanol and another in 20% acetamide. These probes are highly miscible in water, removing the need for artificial potentials to prevent aggregation, and contain three common interaction types: hydrogen bond donor, hydrogen



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

bond acceptor and hydrophobic. They compare their results to GRID2. Without explicit solvation, they find that GRID locates too many polar interaction sites, which correspond to favourable water binding locations. In contrast, MDMix’s solvent probes are more selective in displacing water molecules that are displaced by ligands. However, none of the methods described give any indication of what the key interactions contributing to the hotspot are from a global search. Here we describe a new methodology for identifying hotspots that, starting from a global search of an apo protein, not only identifies fragment binding sites, but also the specific interactions that determine fragment binding. The method is validated using experimental binding positions of 21 fragments and subsequent lead molecules. Protein kinase B and pantothenate synthetase are examined in detail. Results and discussion For this work, hotspots were defined as “The minimum binding site that will bind a fragment, maintaining the fragment binding position once it has been elaborated.” To validate fragment-hotspot maps, the structures of 21 fragment-lead pairs collated by Ichihara21 were used as a dataset. For each fragment-lead pair, the fragment forms the core of the lead and its binding mode is retained. The fragment-lead pairs are bound to 15 different proteins, for which apo structures were also collated and protonated23. This was a further precaution over the original dataset to ensure there was no bias towards the known binding sites.


Page 4 of 35

Page 5 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 1. Apo CDK2 with atomic propensities (left), weighted atomic propensities (centre) and fragment hotspot map (right). The hydrophobic map is shown in yellow, hydrogen-bond donor in blue and hydrogen-bond acceptor in red. A magenta ligand is included for reference. Fragment hotspot maps are created by first calculating grid-based atomic propensities for donor, acceptor and hydrophobe probes, and then weighting these scores by the buriedness of the grid point. These weighted atomic propensities are then sampled by three simple molecular probes, the scores of which are mapped back onto three new grids corresponding to donor, acceptor and apolar maps. Figure 1 shows these three stages in the calculation, and demonstrates how each step removes false positives until those coinciding with the fragmentbinding site remain. For validation, fragment hotspot maps were created for the apo structures, to which the fragment and lead-bound structures were aligned. Scores were assigned to the fragment and lead atoms by assessing whether the atom was a hydrogen-bond donor, acceptor or hydrophobic, then reading the score from the corresponding map. In addition to being able to highlight the fragment binding site, the highest scoring interactions predicted from the apo structure are often those made by the fragment, with the subsequent lead molecule picking up moderate scoring interactions. An example can be seen



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

in Figure 2, which shows HSP90 with a lead molecule developed by fragment linking. The portion circled in blue coincides with the more potent of the two fragments, and can be seen to occupy the highest scoring region of the map. Only one of the two acceptors is predicted as the fragment binds to HSP90 via bridging water molecules, which were excluded for the purpose of the validation. If a water molecule is known or predicted to be important for binding it can be included in the calculation. For HSP90, inclusion of the relevant water molecules allows identification the remaining interactions of the fragment (Figure SI1). The second fragment is shown to have a much lower score, however the crystal structure with both fragments bound shows it to stack on top of the first fragment, likely contributing to its binding.

Figure 2. Left: HSP90 with a lead molecule (magenta) and apolar (yellow to grey volume), donor (blue surface) and acceptor (red surface) hotspot maps. Centre: A closer look at the ligand with the protein removed. Right: 2D schematic showing how the scores are distributed throughout the lead molecule, with the primary fragment circled in blue, and the second in orange. Scores are assigned based on their atom type (e.g. acceptor nitrogen read from the acceptor map). Bright yellow regions indicate scores > 17, purple indicates scores in the range 14-17 and scores lower than 14 are not highlighted. The choice of 17 and 14 are discussed below.


Page 6 of 35

Page 7 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Fragment PDB code: 2W1D

Fragment PDB code: 2OHM

Fragment PDB code: 2V00

Lead PDB code: 2W1G

Lead PDB code: 2OHU

Lead PDB code: 2VA7

Apo PDB code: 4J8N

Apo PDB code: 1W50

Apo PDB code: 1W50

Frag LE: 0.60

Frag LE: 0.33

Frag LE: 0.32

Lead LE: 0.43

Lead LE: 0.24

Lead LE: 0.36

Fragment PDB code: 2W70

Fragment PDB code: 1VYZ

Fragment PDB code: 2VTA

Lead PDB code: 2W71

Lead PDB code: 1VYW

Lead PDB code: 2VU3

Apo PDB code: 2J9G*

Apo PDB code: 1HCL

Apo PDB code: 1HCL

Frag LE: 0.53

Frag LE: 0.58

Frag LE: 0.54

Lead LE: 0.41

Lead LE: 0.41

Lead LE: 0.47



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Fragment PDB code: 3CCB

Fragment PDB code: 2QFO

Fragment PDB code: 3FT5

Lead PDB code: 3CCC

Lead PDB code: 2QG0

Lead PDB code: 3FT8

Apo PDB code: 1J2E

Apo PDB code: 1YES

Apo PDB code: 1YES

Frag LE: 0.45

Frag LE: 0.55

Frag LE: 0.56

Lead LE: 0.54

Lead LE: 0.30

Lead LE 0.39

Fragment PDB code: 2WI2

Fragment PDB code: 3E62

Fragment PDB code: 3FU0

Lead PDB code: 2WI7

Lead PDB code: 3E64

Lead PDB code: 3FH7

Apo PDB code: 1YES

Apo PDB code: 4ZIM*

Apo PDB code: 3B7S*

Frag LE: 0.48

Frag LE: 0.56

Frag LE: 0.21

Lead LE: 0.33

Lead LE: 0.41

Lead LE: 0.39


Page 8 of 35

Page 9 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Fragment PDB code: 3CIZ

Fragment PDB code: 1W84

Fragment PDB code: 1W7H

Lead PDB code: 3CJ5

Lead PDB code: 1WBT

Lead PDB code: 1W83

Apo PDB code: 3MWV

Apo PDB code: 1WFC

Apo PDB code: 1WFC

Frag LE: 0.25

Frag LE: 0.37

Frag LE: 0.28

Lead LE: 0.31

Lead LE: 0.32

Lead LE: 0.27

Fragment PDB code: 3IMG

Fragment PDB code: 1Y2B

Fragment PDB code: 3ET0

Lead PDB code: 3IUE

Lead PDB code: 1Y2K

Lead PDB code: 3ET3

Apo PDB code: 3COV

Apo PDB code: 3SL3

Apo PDB code: 1PRG

Frag LE: 0.38

Frag LE: 0.48

Frag LE: 0.33

Lead LE: 0.29

Lead LE: 0.51

Lead LE: 0.31



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Fragment PDB code: 2UW3

Fragment PDB code: 2UVX

Fragment PDB code: 2VIN

Lead PDB code: 2UW7

Lead PDB code: 2VO6

Lead PDB code: 2VIW

Apo PDB code: 4C33

Apo PDB code: 4C33

Apo PDB code: 3OY5

Frag LE: 0.48

Frag LE: 0.59

Frag LE: 0.32

Lead LE: 0.45

Lead LE: 0.48

Lead LE: 0.32

Table 1. A summary of the 21 fragment-lead pairs collated by Ichihara21. The 2D structures of the leads have been mapped with the atomic scores calculated from fragment hotspot maps. Bright yellow regions indicate scores > 17, purple indicates scores in the range 14-17 and scores lower than 14 are not highlighted. The part of the molecule corresponding to the fragment is outlined in blue, and secondary fragments or small molecules present in the fragment crystal structure that were later incorporated into the lead are outlined in orange. Each image is titled by the protein name and accompanied by related PDB codes and the ligand efficiencies (LE) of the fragment (Frag LE) and the lead (Lead LE). PDB codes labelled with * are structures with the natural substrate (unrelated small molecule inhibitor in the case of JAK2), as an apo structure was not available.

∆


Page 10 of 35

Page 11 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Equation 1: Heavy atom count is the number of non-hydrogen atoms in the molecule. The units for ligand efficiency are assumed to be kcal mol-1 atom-1 throughout this paper, and ligand efficiency is displayed as a positive value.

Max. Score

uPA: 2vin

Difference in Score

58.6

99.7

P38a: 1w7h

99.9

HSP90: 3ft5

96.6

Aurora-Kinase: 2w1d

74.0

BACE1: 2ohm

90.6

99.6

97.5

99.8

BACE1: 2v00

-2.854

98.4

98.4

Protein Kinase B: 2uw3

Biotin Carboxylase: 2w70

-3.403 -3.143

24.5

99.9

Protein Kinase B: 2uvx

-4.108 -3.583

99.7

PDE4: 1y2b

Pantothenate Synthetase: 3img

-4.136

99.8

99.0

PPAR: 3et0 DPP-IV: 3ccb

99.8

99.8

-7.368 -5.528

99.9

HSP90: 2wi2

NS5-RNA 0.0 Polymerase: 3ciz

99.9

85.7

-7.751 -7.631

99.9

99.9 88.8

JAK2: 3e62

-7.852 99.9

99.9

HSP90: 2qfo

-9.139

99.8

99.9

CDK2: 2vta

CDK2: 1vyz

99.9

99.9

P38a: 1w84

LTA4H: 3fu0

-9.822

96.8 99.9 99.8

-2.669

97.5

-2.111 99.9 99.9 99.7 99.2

100.0

97.5

-2.100 -1.679 -0.767 -0.731 -0.188 -0.029

Figure 3: Left: The highest scoring fragment atom (blue) and lead atom (orange) for each fragment-lead pair, labelled by their percentage ranking compared to all grid points with a score greater than 0. At least part of the fragment is expected to interact with a hotspot, therefore the highest scoring atom is used to determine whether the ligand is interacting with



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

a hotspot. Most fragments had their highest scoring atom in the top 1% of scoring grid points. Right: Bar graph showing the (average lead atom score) – (average fragment atom score) for each fragment-lead pair. In all cases the fragment scores more highly than the lead. The NS5 RNA polymerase fragment binding site was the lowest scoring of the dataset. The highest scoring atom of the fragment only had a score of 8.6, as the fragment bound to a moderately scoring region known as the thumb site, away from the large highly scoring catalytic region described as the palm24. The fragment had the lowest experimentally determined affinity out of the dataset, in the mM range, and inspection of the electron density showed that the fragment was poorly resolved. The temperature factors (B-factors) of the fragment atoms ranged from 32 to 42 compared to the surrounding residue atoms, which ranged from 12 to 24. This suggests that pockets should not be thought of as “hot or not”, but rather a continuum, where moderate scoring regions are able to bind fragments, albeit very weakly. Our method aims to locate the interactions that drive fragment binding starting from a global search of the protein. It is therefore important that the method avoids returning false positive binding sites. Here it was assumed that the experimental binding sites from the datasets are the only true positive binding sites, and any others sites detected were treated as false positives. During the calculation of the atomic propensities, the cavity detection process eliminates much of the protein surface. After sampling the atomic propensities with the molecular probes and generating maps based on the highest scoring poses of each probe, the majority of probes were placed within a small subset of cavities, eliminating even more of the protein surface. In order to check whether the fragments were found in the highest scoring regions, the atomic scores were compared only to grid points that had at least one probe atom placed


Page 12 of 35

Page 13 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


there. For each atom in the dataset, its score was ranked against all qualifying grid points of the protein it was calculated from, and was represented as a percentage rank. Fragments were found to rank more highly than the lead atoms, with a median rank of 97% compared to 72% for the lead atoms (those outside the fragment core). This demonstrates that the fragment-hotspot maps are not simply locating ligand-binding sites, but are picking out the hotspot within those sites. To aid with visualising the output and assessing whether a hotspot is present, a cut-off for a predicted hotspot was calculated. As it is possible that the fragment is larger than the hotspot it binds to, the upper quartile of atomic scores for each fragment was used. The median of these values across the dataset was 17, and the lowest was 14. Contouring at these two levels allows visualisation of not only where on the protein ligands are likely to bind, but also where within that pocket the fragment will bind and which interactions will drive binding. This can be seen in Figure 4, where areas of acceptor and apolar propensity > 17 suggest the interactions leading to fragment binding, with areas > 14 matching lead atoms and remaining fragment atoms. Only the binding site of PDE4 contains maps scoring above these contour levels, therefore this information does not come at a cost of being unable to identify the binding site from a global search.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 4. Left PDE4 with a fragment in magenta and a lead molecule in cyan. The maps have been contoured once at 14 (semi-transparent) and again at 17 (almost solid). Many of the fragment atoms coincide with high scoring apolar (yellow) and acceptor (red) regions, suggesting these interactions drive fragment binding. The cyan lead atoms and some of the fragment atoms extend into regions of the pocket that only make the lower contouring level. The fragment’s NH is not predicted by the maps as it is facing the solvent. Right At this contouring level, only the binding site contains any surfaces, despite starting from a global search. There are some limitations to using a single apo structure to calculate the maps. One example is p38α, where the morpholines of the leads occupy the DFG-out pocket, which is not present in the apo structure. This is not a solely a problem for static methods, as the limited sampling and atomic restraints of the previously described molecular dynamics methods would fail to predict this conformational change starting from an apo structure. Although unable to predict pockets opening a priori, the speed of this method makes it convenient to recalculate the fragment hotspot maps from the new ligand-bound structure. We are currently looking at calculating fragment hotspot maps for ensembles to assess the improvement over a single static structure.


Page 14 of 35

Page 15 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Kozakov and colleagues25 have shown previously that hotspots are less sensitive to conformational change. Looking specifically at protein-protein interaction hotspots, they found that even if substantial conformational change was required for ligand binding, the hotspots were still detectable from the apo structure. This has also been the case for this method in terms of binding site location; however the locations of individual interactions can shift upon ligand binding due to induced fit effects. If the high scoring interactions are used to suggest residues or interaction types to be used in a structure-based pharmacophore screening program, the radius of the pharmacophore sphere is generally large enough to take these induced-fit effects into account. Protein Kinase B Verdonk et al26 used a fragment growing approach to design inhibitors of protein kinase B (PKB), and performed a Free-Wilson analysis to provide group contributions to binding. They used group efficiency (GE), defined in equation 2, to evaluate whether a group increased potency enough to justify the number of heavy atoms it contained.

∆∆ ∆

Equation 2. ∆∆ is the change in the free energy of binding between two compounds,

∆ is the change in the number of heavy atoms.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

A

B 0.28 1.6 0.54

0.42 0.32 1.5 C

D

Figure 5. A: Inhibitor of protein kinase B labelled with the group efficiency of each highlighted group26. B: Fragment hit. C: Intermediate structure during fragment growth. D: Full compound. Hydrophobic map is shown in dark grey to yellow to show moderate to high scoring regions. Donor hotspots are shown as a blue surface and acceptor hotspots are shown as a red surface. The atoms seem displaced from the maps as the global alignment of the proteins did not manage to align the binding site well. Their results are summarised Figure 5. The pyrazole is estimated to have a group efficiency of 1.5, second only to the single-atom chloro group. Both the acceptor and the donor interactions are predicted in the fragment hotspot map, and the carbons are located within the


Page 16 of 35

Page 17 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


main hydrophobic hotspot. Comprising of only 5 atoms, this small fragment binds efficiently, as expected. The first phenyl group does not make any specific interactions, but is located mostly within the main hydrophobic hotspot, reflected in a GE of 0.42. The methyl group has one of the lowest group efficiencies of the groups, and was ultimately removed from the molecule. From the maps it can be seen to extend slightly outside of the main hotspot. Addition of EtNH2 to the phenyl ring yields one of the more group-efficient additions to the molecule. This is easily rationalised from the map in Figure 6C, as the primary amine occupies a region of high scoring donor propensity. The second phenyl group is given a group efficiency of 0.28, which is the lowest of all groups. This could be an underestimate, as addition of the phenyl group prevents the primary amine from making the interaction it found previously as can be seen in Figure 6D. This is one example out of many where addition of a group does not make a simple additive increase in potency27. If the fragment binding site is known, it no longer makes sense to do a global search of the protein unless additional binding sites are of interest. Instead, the calculation can be run with the fragment included and the binding site defined. This increases the speed of the calculation and limits the information to the cavities of interest to the medicinal chemistry programme. The run time for this calculation is only a few minutes.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 6. Hydrophobic fragment hotspot map calculated with the fragment left in the binding site, which was defined prior to the calculation. This allows for finer sampling of the pocket in much less time as a global search of the protein is no longer needed. The result of this calculation is shown in Figure 6, where it is much more obvious which direction the fragment should be grown in order to make the most efficient addition to the molecule, with part of the phenyl group and the chloro group placed in highly scoring areas. As the chloro group is a single atom placed in a highly scoring region, it is understandable why it is the most group efficient addition. The chloro atom also affects the electronics of the phenyl ring, and could again be an example where addition of a group does not lead to simple additive increase in potency27. Pantothenate synthetase The lead molecule in this dataset for pantothenate synthetase (PDB code: 3IUE) has been recently revisited28 and the group efficiency (GE) of each group determined to highlight which part of the molecule should be developed further. Figure 8 shows how the GE is distributed throughout the molecule, and how each intermediate structure compares to the fragment hotspot map.


Page 18 of 35

Page 19 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


A

042

B

0.75

0.17 0.17

C

D

E

F

P2

P1

Figure 7. A Group efficiency for the groups that make up the lead molecule for pantothenate synthetase (PDB code 3IUE)28. B-E Crystal structures of each iteration of the hit-to-lead development used to calculate the GE (PDB codes 3IMG, 3ISJ, 3IUB, 3IUE) overlaid with fragment hotspot maps. F Recent inhibitor of pantothenate synthetase (PDB code: 4MUK), which fills the more highly scoring P1 pocket.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

The fragment hit shown in Figure 7B makes two specific interactions, the methoxy oxygen matches an area of high acceptor propensity and the NH forms a hydrogen bond with a sulfate ion. If the sulfate ion is included in the fragment hotspot map calculation, the NH is located in a region of high donor propensity. However, although present in the crystallisation conditions, the sulfate is not present in the isothermal titration calorimetry (ITC) measurement and therefore does not help rationalise the high group efficiency of the fragment. We assume there is another ion in the ITC experiment that is able to bridge the interaction in the same manner as the sulfate, discussed in the supporting information. Addition of the sulphonamide moiety shown in Figure 7C has a GE of just 0.17. This is perhaps surprising as one oxygen of the sulfonamide is placed in a very high scoring region, and the NH forms a hydrogen bond with a bridging water molecule interacting with Gly-158. This water molecule is displaced upon binding of the transition state analogue shown in Figure 9, so it is possible that this water is not particularly tightly bound, resulting in the lower GE. The remaining sulfoxide and carbonyl oxygen atoms are left interacting with the solvent, therefore again not contributing to the binding energy and reducing the GE. The methyl pyridine moiety added in Figure 7D is also not very group efficient. The ring occupies a moderately high scoring hydrophobic pocket, but fails to satisfy any of the polar interactions of the protein or the acceptor nitrogen of the pyridine. The final addition of the ethanoic acid group is highly group efficient. The oxygen atoms are placed in an area of very high acceptor propensity, resulting in a strong increase in binding energy for a small increase in atom count. From the GE analysis, the methyl pyridine group was highlighted as the best place for optimisation. Toluene was found to be the most group efficient change, with a GE of 0.35 compared to 0.17. However the crystal structures of the new more potent compounds showed


Page 20 of 35

Page 21 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


that they no longer bound to the P2 pocket, but instead bound to the P1 pocket, as labelled in Figure 7E. The molecule was developed further by adding a trifluoromethyl group and an extra carbon between the ring and the sulphonamide linker to completely fill the P1 pocket. The molecule has a reported IC50 of 250 nM. As is clear in the figure, the P1 pocket is a hydrophobic hotpot and is also capable of binding fragments. The P1 pocket does have polar interactions available, but the lead molecule does not interact with any of these. Looking again at the percentage ranking of atoms across the whole dataset, splitting the atoms into their corresponding interaction types shows the apolar atoms to rank more highly than donor and acceptor atoms for both fragments and leads (Figure 8). In most cases a few very highly scoring apolar regions correlated strongly with fragment binding locations. Polar interactions were more likely to be left unsatisfied by the fragments, but this does not mean they were unimportant for binding. The fragments in the dataset were typically flat and unable to match all the interactions highlighted by the fragment hotspot maps, exemplified in Figure 9.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 8. Box and violin plots showing the percentage rank for fragment and lead atoms split by interaction type. For both fragment and lead atoms, the apolar atoms are the most highly ranking. Very few polar lead atoms were found to reside in highly scoring areas. In contrast, when the maps were compared to a transition state analogue almost all of the predicted high scoring interactions were satisfied. This is in line with the findings by Higueruelo et al.29, who used scissor plots30 to describe the interactions of both synthetic and natural ligands in proteins. Natural ligands were found to maintain a greater ratio of polar to apolar contacts, with the number of polar atoms in the ligand correlating with the number of polar contacts. In contrast, synthetic ligands tended to find a few polar interactions with the remaining heteroatoms unmatched by the protein. For the lead atoms in our dataset, relatively few polar atoms scored above zero. This could be a physical illustration of Hann’s31 complexity constraint or the result of the tendency in medicinal chemistry programs to add hydrophobicity to increase potency.


Page 22 of 35

Page 23 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


The two fragments bound to pantothenate synthetase each have two polar groups, both of which are satisfied. When using X-ray crystallography for a primary fragment screen, having a smaller library of simpler fragments like these is feasible as the elaboration of the fragments can be visualised in three-dimensions; the structural data allows complexity to be built into the molecule to gain more affinity and selectivity by rational design during lead optimisation. However, some opt for larger collections of more complex fragments, for which X-ray crystallography is too low throughput. Pharmacophores derived from the fragment hotspot maps could make the use of X-ray crystallography with these libraries feasible. Rather than relying on the promiscuity of simple fragments, the larger library could be filtered to give a subset that matches the hotspot interactions of the target. This may yield higher quality and more polar hits that would be more forgiving if hydrophobicity was then added to increase potency.

Figure 9: Pantothenate synthetase showing Left: two bound fragments in cyan (PDB code: 3img) and Right: A transition state analogue in magenta (PDB code: 1n2h). Hydrophobic map is shown in dark grey to yellow to show moderate to high scoring regions. Donor hotspots are shown as a blue surface and acceptor hotspots are shown as a red surface. Although all the interactions of the fragments are satisfied, they leave many of the protein



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

interactions unsatisfied. The more 3D and flexible transition state analogue satisfies the majority of interactions predicted by the fragment hotspot maps. One interaction not satisfied in either case is found to bind an ethanol molecule (in magenta sticks); suggesting this is still a genuine hotspot interaction. Conclusion Identification of hotspots and their specific interactions can be used to evaluate the ligandability of a pocket and suggest which interactions fragments and larger ligands will need to make to bind there. Current methods use hotspots to assess either ligandability of sub pockets from a global search of the protein13, or provide interaction information for a predefined binding site22. The relationships between hotspots, fragments and ligandability, and how they all relate to fragment hotspots maps are illustrated in Figure 10.


Page 24 of 35

Page 25 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 10: Relationship between hotspots, fragments, ligandability and hotspot maps. The red boxes show established ideas from the literature, green show those explored in this paper, grey will be tested in future work. We have presented a new method capable of providing interaction information from a global search of an apo protein crystal structure. As our method does not rely on molecular dynamics, results can be calculated within minutes rather than hours on an ordinary laptop. Ligand atoms were consistently found in the highest scoring grid points; fragment atoms had a median percentage rank of 97% and lead atoms 72%. This method will be useful to both medicinal chemists and molecular modellers. For the medicinal chemist, the maps will be a visual guide to the most important interactions within the pocket. Once a structure containing a hit is available, it will be easy to determine whether any of the existing groups are suboptimal. The maps will suggest the direction in which the hit should be grown, and which types of interaction are required. For the molecular modeller, the maps will complement existing virtual screening methods. With the most important interactions highlighted, existing pharmacophore methods can be used to screen for molecules capable of making these essential interactions. Equally, the maps can be used to generate constraints for docking, steering the docking towards occupying the hotspot and ensuring the right interactions are made. Currently, this requires the maps to be visually inspected and then the docking constraints or structure-based pharmacophores to be generated manually, however automatic workflows are being developed. Fragment hotspot maps could also be used as a direct score of ligandability. As fragment hit rate has been shown to correlate with ligandability1, a computational method that can predict fragment binding may also be able to predict ligandability. This has been demonstrated recently13 by using hotspots predicted by FTMap, combined with the volume of the pocket



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

and the density of hotspots. As the scores from our method are comparable across targets, due to their probabilistic origin of sampling SuperStar propensities, future work will determine whether the scores can provide the relative ligandability of a wide range of targets. Method The full method for the generation of fragment hotspot maps has been summarised in Figure 11.

Figure 11. Flow chart describing the main steps taken to generate the fragment hotspot maps and validate them with the experimental structures. Dataset It is assumed that at least part of a fragment must interact with a hotspot in order to bind efficiently enough to be detectable by X-ray crystallography. Fragment-binding positions were therefore used as a standard for hotspot prediction, with the understanding that it is possible that only part of the fragment will be located within the hotspot. In order to


Page 26 of 35

Page 27 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


discriminate between hotspots and the rest of the ligand binding site, atoms from lead-like molecules were also examined and compared to the fragments. The dataset collated by Ichihara et al21 contains crystal structures from fragment-based drug design projects, where lead molecules developed from the fragment hit retains the fragment binding position. Affinity data were available for each of the fragments and leads. Here, this dataset of fragment-lead pairs was extended further to include apo structures, on which all calculations were performed, removing bias towards the binding site. Protonated structures were retrieved from the Protoss server23, which also protonates the ligand using the context of the binding site, and had all waters and small molecules removed. For each protein, the protonation was checked manually, and then the fragment and lead bound structures were globally aligned with the apo structure. Only the fragment binding monomer was used. Atomic Hotspot Prediction To generate fragment hotspots, atomic hotspots must first be calculated. In this study we used SuperStar to calculate the atomic hotspots, although, in principle, any grid-based approach could potentially be used. Three maps were generated:

•

Hydrophobic – Aromatic CH probe

•

Donor – Uncharged NH probe

•

Acceptor – Carbonyl O probe

Normally the binding site would need to be defined prior to the SuperStar calculation; however no information about the binding site was used in this case. SuperStar uses the LIGSITE32 algorithm to detect cavities, and in the absence of a starting coordinate or residue



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

from which to grow the cavity, LIGSITE is run on the whole protein. It gives each grid point a buriedness score between zero (completely solvent exposed) and seven (completely buried). SuperStar then provided atomic propensities for cavities that contained grid points with a LIGSITE score of five or above. In order to find areas where high interaction propensity coincides with buried pockets, the SuperStar maps are weighted by the LIGSITE score for each grid point. The weighted SuperStar maps show propensity throughout the protein, but in order to find fragment hotspots, the atomic propensities were sampled with molecular probes. Fragment Hotspot Prediction As only hydrophobic, donor and acceptor maps are calculated, probes containing either all carbons or carbons with a single donor or acceptor heteroatom are able to sample the maps fully. The probes, shown in Figure 12, were chosen to reflect hotspot environments. All three probes have the same shape: the polar atoms represent a functional group attached to the ring and toluene is used for the apolar probe. The large but flat rings are selected to find tight hydrophobic environments, with polar interactions at the deepest part for the polar probes. The probes may be too large to sample very small pockets accessible to alkyl chains, but were chosen as they resulted in fewer false positives when performing a global search. Smaller probes could be used to give a deeper exploration of pockets highlighted by the default probes. The bond orders of the probes are ignored, and it is just the atom types that are used to assign a score from one of the three weighted SuperStar maps.


Page 28 of 35

Page 29 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 12. The three molecular probes used to sample the weighted atomic propensity maps. The bonds of the probes are ignored, all carbons sample the aromatic CH propensity, the oxygen samples the carbonyl oxygen propensity, and the NH2 samples the uncharged NH2 propensity. The probes undergo 200 rotations, which are uniformly distributed on the surface of a sphere, then translated such that they centred on the heteroatom for the polar probes, or the methyl group of the toluene probe. As the atomic propensities are calculated prior to sampling, it is not necessary to sample the entire protein surface. All 200 rotations of a probe are placed with their central atom on the top 400 scoring grid points. For each pose, the atomic propensities are assigned to each atom from their corresponding map. Any atom that clashes with the protein has a score of zero and the pose is skipped, all remaining poses are assigned a score calculated by taking the geometric mean of their atomic scores. Three 0.5 Å grids, one for apolar, donor or acceptor atoms, are placed over the protein. Each grid point that contains a probe atom is set to the score of the probe, not including the carbon atoms for the polar probes. If multiple probes place atoms in the same grid point, the highest score is used. Validation Maps were created for all apo structures in the dataset, and scores assigned to both the fragment and lead atoms. Each atom in the ligand was categorised as either hydrophobic, donor or acceptor, and its score was read from the corresponding map. Sometimes the ligand atoms were slightly displaced from a hotspot, due to the alignment of the proteins and errors in the crystal structure. To accommodate for this, the highest scoring grid point within two grid points was assigned to the atom. Atoms that were in the maximum common substructure



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

match between the fragment and the lead molecules were assigned as “fragment atoms”, with the remaining atoms assigned as “lead atoms”. Supporting Information Availability Results for HSP90 with waters included. Discussion on possible interference from compounds in ITC conditions for Pantothenate Synthetase Corresponding Author Information *(C.R) Phone +44223 762 915; Email: [email protected] Acknowledgements Thanks to Dr John Liebeschuetz who was instrumental in the initiation of this project. CJR was funded by the British Biological Research Council (BB/L502686/1) and UCB. Abbreviations Used CSD, Cambridge Structural Database; DPP-IV, Dipeptidyl peptidase 4; GE, Group efficiency; HAC, Heavy atom count; ITC, Isothermal titration calorimetry; JAK2, Janus kinase 2; LTA4H, Leukotriene-A4 hydrolase; MSCS, Multiple solvent crystal structure; PDE4, Phosphodiesterase 4; uPA, Urinary-type Plasminogen Activator References (1)

Hajduk, P. J.; Huth, J. R.; Fesik, S. W. Druggability Indices for Protein Targets Derived from NMR-Based Screening Data. J. Med. Chem. 2005, 48 (7), 2518–2525.

(2)

Goodford, P. J. A Computational Procedure for Determining Energetically Favorable Binding Sites on Biologically Important Macromolecules. J. Med. Chem. 1985, 28 (7), 849–857.


Page 30 of 35

Page 31 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(3)

Verdonk, M. L.; Cole, J. C.; Taylor, R. SuperStar: A Knowledge-Based Approach for Identifying Interaction Sites in Proteins. J. Mol. Biol. 1999, 289 (4), 1093–1108.

(4)

Bruno, I. J.; Cole, J. C.; Lommerse, J. P. M.; Rowland, R. S.; Taylor, R.; Verdonk, M. L. IsoStar: A Library of Information about Nonbonded Interactions. J. Comput. Aided. Mol. Des. 11 (6), 525–537.

(5)

Allen, F. H. The Cambridge Structural Database: A Quarter of a Million Crystal Structures and Rising. Acta Crystallogr. Sect. B Struct. Sci. 2002, 58 (3), 380–388.

(6)

Allen, K. N.; Bellamacina, C. R.; Ding, X.; Jeffery, C. J.; Mattos, C.; Petsko, G. A.; Ringe, D. An Experimental Approach to Mapping the Binding Surfaces of Crystalline Proteins †. J. Phys. Chem. 1996, 100 (7), 2605–2611.

(7)

Mattos, C.; Bellamacina, C. R.; Peisach, E.; Pereira, A.; Vitkup, D.; Petsko, G. A.; Ringe, D. Multiple Solvent Crystal Structures: Probing Binding Sites, Plasticity and Hydration. J. Mol. Biol. 2006, 357 (5), 1471–1482.

(8)

English, A. C.; Groom, C. R.; Hubbard, R. E. Experimental and Computational Mapping of the Binding Surface of a Crystalline Protein. Protein Eng. Des. Sel. 2001, 14 (1), 47–59.

(9)

Hajduk, P. J.; Huth, J. R.; Tse, C. Predicting Protein Druggability. Drug Discov. Today 2005, 10 (23-24), 1675–1682.

(10)

Brenke, R.; Kozakov, D.; Chuang, G.-Y.; Beglov, D.; Hall, D.; Landon, M. R.; Mattos, C.; Vajda, S. Fragment-Based Identification of Druggable “Hot Spots” of Proteins Using Fourier Domain Correlation Techniques. Bioinformatics 2009, 25 (5), 621–627.

(11)

Hall, D. H.; Grove, L. E.; Yueh, C.; Ngan, C. H.; Kozakov, D.; Vajda, S. Robust Identification of Binding Hot Spots Using Continuum Electrostatics: Application to



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Hen Egg-White Lysozyme. J. Am. Chem. Soc. 2011, 133 (51), 20668–20671. (12)

Hall, D. R.; Ngan, C. H.; Zerbe, B. S.; Kozakov, D.; Vajda, S. Hot Spot Analysis for Driving the Development of Hits into Leads in Fragment-Based Drug Discovery. J. Chem. Inf. Model. 2012, 52 (1), 199–209.

(13)

Kozakov, D.; Hall, D. R.; Napoleon, R. L.; Yueh, C.; Whitty, A.; Vajda, S. New Frontiers in Druggability. J. Med. Chem. 2015, 58 (23), 9063–9088.

(14)

Kozakov, D.; Grove, L. E.; Hall, D. R.; Bohnuud, T.; Mottarella, S. E.; Luo, L.; Xia, B.; Beglov, D.; Vajda, S. The FTMap Family of Web Servers for Determining and Characterizing Ligand-Binding Hot Spots of Proteins. Nat. Protoc. 2015, 10 (5), 733– 755.

(15)

Snyder, P. W.; Lockett, M. R.; Moustakas, D. T.; Whitesides, G. M. Is It the Shape of the Cavity, or the Shape of the Water in the Cavity? Eur. Phys. J. Spec. Top. 2014, 223 (5), 853–891.

(16)

Barillari, C.; Taylor, J.; Viner, R.; Essex, J. W. Classification of Water Molecules in Protein Binding Sites. J. Am. Chem. Soc. 2007, 129 (9), 2577–2587.

(17)

Kulp, J. L.; Pompliano, D. L.; Guarnieri, F. Diverse Fragment Clustering and Water Exclusion Identify Protein Hot Spots. J. Am. Chem. Soc. 2011, 133 (28), 10740– 10743.

(18)

Andrews, S. P.; Mason, J. S.; Hurrell, E.; Congreve, M. Structure-Based Drug Design of Chromone Antagonists of the Adenosine A2A Receptor. Medchemcomm 2014, 5 (5), 571.

(19)

Young, T.; Abel, R.; Kim, B.; Berne, B. J.; Friesner, R. A. Motifs for Molecular Recognition Exploiting Hydrophobic Enclosure in Protein-Ligand Binding. Proc. Natl.


Page 32 of 35

Page 33 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Acad. Sci. U. S. A. 2007, 104 (3), 808–813. (20)

Cheng, Y. K.; Rossky, P. J. Surface Topography Dependence of Biomolecular Hydrophobic Hydration. Nature 1998, 392 (6677), 696–699.

(21)

Ichihara, O.; Shimada, Y.; Yoshidome, D. The Importance of Hydration Thermodynamics in Fragment-to-Lead Optimization. ChemMedChem 2014, 9 (12), 2708–2717.

(22)

Alvarez-Garcia, D.; Barril, X. Molecular Simulations with Solvent Competition Quantify Water Displaceability and Provide Accurate Interaction Maps of Protein Binding Sites. J. Med. Chem. 2014, 57 (20), 8530–8539.

(23)

Bietz, S.; Urbaczek, S.; Schulz, B.; Rarey, M. Protoss: A Holistic Approach to Predict Tautomers and Protonation States in Protein-Ligand Complexes. J. Cheminform. 2014, 6 (1), 12.

(24)

Bressanelli, S.; Tomei, L.; Roussel, A.; Incitti, I.; Vitale, R. L.; Mathieu, M.; De Francesco, R.; Rey, F. A. Crystal Structure of the RNA-Dependent RNA Polymerase of Hepatitis C Virus. Proc. Natl. Acad. Sci. 1999, 96 (23), 13034–13039.

(25)

Kozakov, D.; Hall, D. R.; Chuang, G.-Y.; Cencic, R.; Brenke, R.; Grove, L. E.; Beglov, D.; Pelletier, J.; Whitty, A.; Vajda, S. Structural Conservation of Druggable Hot Spots in Protein-Protein Interfaces. Proc. Natl. Acad. Sci. 2011, 108 (33), 13528– 13533.

(26)

Verdonk, M. L.; Rees, D. C. Group Efficiency: A Guideline for Hits-to-Leads Chemistry. ChemMedChem 2008, 3 (8), 1179–1180.

(27)

Baum, B.; Muley, L.; Smolinski, M.; Heine, A.; Hangauer, D.; Klebe, G. NonAdditivity of Functional Group Contributions in Protein–Ligand Binding: A



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Comprehensive Study by Crystallography and Isothermal Titration Calorimetry. J. Mol. Biol. 2010, 397 (4), 1042–1054. (28)

Hung, A. W.; Silvestre, H. L.; Wen, S.; George, G. P. C.; Boland, J.; Blundell, T. L.; Ciulli, A.; Abell, C. Optimization of Inhibitors of Mycobacterium Tuberculosis Pantothenate Synthetase Based on Group Efficiency Analysis. ChemMedChem 2016, 11 (1), 38–42.

(29)

Higueruelo, A. P.; Schreyer, A.; Bickerton, G. R. J.; Blundell, T. L.; Pitt, W. R. What Can We Learn from the Evolution of Protein-Ligand Interactions to Aid the Design of New Therapeutics? PLoS One 2012, 7 (12), e51742.

(30)

Olsson, T. S. G.; Williams, M. A.; Pitt, W. R.; Ladbury, J. E. The Thermodynamics of Protein-Ligand Interaction and Solvation: Insights for Ligand Design. J. Mol. Biol. 2008, 384 (4), 1002–1017.

(31)

Hann, M. M.; Leach, A. R.; Harper, G. Molecular Complexity and Its Impact on the Probability of Finding Leads for Drug Discovery. J. Chem. Inf. Model. 2001, 41 (3), 856–864.

(32)

Hendlich, M.; Rippmann, F.; Barnickel, G. LIGSITE: Automatic and Efficient Detection of Potential Small Molecule-Binding Sites in Proteins. J. Mol. Graph. Model. 1997, 15 (6), 359–363, 389.


Page 34 of 35

Page 35 of 35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Fragment core

Table of contents graphic


Identifying Interactions that Determine Fragment Binding at Protein

Recommend Documents