Structure-Based Consensus Scoring Scheme for Selecting Class A

Article pubs.acs.org/jcim

Structure-Based Consensus Scoring Scheme for Selecting Class A Aminergic GPCR Fragments Á dám A. Kelemen,† Róbert Kiss,‡ György G. Ferenczy,† László Kovács,§ Beáta Flachner,∥,⊥ Zsolt Lőrincz,∥,⊥ and György M. Keserű*,† †

Medicinal Chemistry Research Group, ‡Research Group for Neurodegenerative Disease Drug Discovery, and ⊥Laboratory of Structural Biophysics, Research Centre for Natural Sciences, Hungarian Academy of Sciences, 2 Magyar tudósok krt., Budapest, H-1117, Hungary § Infarmatik, Inc., 113 Barksdale Professional Center, Newark, Delaware 19711, United States ∥ TargetEx, Kft.; 4/a. Kápolna köz, Dunakeszi, 2120, Hungary S Supporting Information *

ABSTRACT: Aminergic G-protein coupled receptors (GPRCs) represent well-known targets of central nervous-system related diseases. In this study a structure-based consensus virtual screening scheme was developed for designing targeted fragment libraries against class A aminergic GPCRs. Nine representative aminergic GPCR structures were selected by first clustering available X-ray structures and then choosing the one in each cluster that performs best in selfdocking calculations. A consensus scoring protocol was developed using known promiscuous aminergic ligands and decoys as a training set. The consensus score (FrACSfragment aminergic consensus score) calculated for the optimized protein ensemble showed improved enrichments in most cases as compared to stand-alone structures. Retrospective validation was carried out on public screening data for aminergic targets (5-HT1 serotonin receptor, TA1 trace-amine receptor) showing 8−17-fold enrichments using an ensemble of aminergic receptor structures. The performance of the structure based FrACS in combination with our ligand-based prefilter (FrAGS) was investigated both in a retrospective validation on the ChEMBL database and in a prospective validation on an in-house fragment library. In prospective validation virtual fragment hits were tested on 5-HT6 serotonin receptors not involved in the development of FrACS. Six out of the 36 experimentally tested fragments exhibited remarkable antagonist efficacies, and 4 showed IC50 values in the low micromolar or submicromolar range in a cell-based assay. Both retrospective and prospective validations revealed that the methodology is suitable for designing focused class A GPCR fragment libraries from large screening decks, commercial compound collections, or virtual databases.

■

INTRODUCTION

based method would provide predictions on binding modes that might support further fragment optimizations.4 Multitarget docking is a well-known approach that is capable of achieving higher enrichments of diverse actives compared to single-target docking.5−8 Considering the similar characteristics of aminergic orthosteric bindig sites we propose, that a representative collection of aminergic GPCR structures could provide hits that may also fit to aminergic receptors not included in the screening model. However, the use of multiple structural data results in increased computational time and large volumes of data for processing.9 Consequently the 19 available aminergic structures (Table 1) were clustered by the RMSD values of their binding site residues and in parallel were all subjected to self-docking in order to select an adequate number

1

In our recent study we developed a ligand-based desirability function to enrich libraries with compounds potentially active on aminergic G-protein coupled receptor (GPCR) targets. The fragment aminergic GPCR score (FrAGS) provides computationally inexpensive, physicochemical property-based filtering of extensive public or proprietary databases consisting of hundreds of thousands to millions of entries. Fragments satisfying a FrAGS score of at least 5 may serve as inputs for structurebased docking methods. Since the first GPCR structure was solved in 2000,2 an emerging number of class-A aminergic GPCR receptor X-ray structures have been published3 (Table 1). The large number of available structures raises the possibility for using their combinations in docking experiments to identify fragments for aminergic receptors. In addition to selecting actives in virtual screening campaigns such a structure© XXXX American Chemical Society

Received: September 29, 2015

A

DOI: 10.1021/acs.jcim.5b00598 J. Chem. Inf. Model. XXXX, XXX, XXX−XXX

Article

Journal of Chemical Information and Modeling Table 1. Human Aminergic X-ray Structures Available (as of December 31, 2014) PDB ID

receptor name

stoichiometry

ligand

funcionality

resolution (Å)

3UON 4MQS 4MQT 2RH1 3D4S 3NY8 3NY9 3NYA 3PDS 4GBR 4LDE 4LDL 4LDO 3P0G 2R4R 2R4S 3KJ6 4QKX 4IAQ 4IAR 4IB4 4NC3 3PBL 3RZE

M2 muscarinic acetylcholine receptor M2 muscarinic acetylcholine receptor M2 muscarinic acetylcholine receptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor β2-adrenoceptor chimeric 5-HT1B-BRIL serotonin receptor chimeric 5-HT1B-BRIL serotonin receptor chimeric 5-HT2B-BRIL serotonin receptor 5-HT2B serotonin receptor D3 dopamine receptor H1 histamine receptor

monomer hetero 2-mer−AB hetero 2-mer−AB monomer monomer monomer monomer monomer monomer hetero 2-mer−AB hetero 2-mer−AB hetero 2-mer−AB hetero 2-mer−AB hetero 2-mer−AB hetero 3-mer−ABC hetero 3-mer−ABC hetero 3-mer−ABC hetero 2-mer−AB homo 2-mer−A2 monomer monomer monomer monomer monomer

QNB IXO, LY2119620 IXO, LY2119620 CAU TIM JRZ JSZ JTZ ERC CAU 1WV 1WV 1WV P0G

antagonist agonist agonist, allosteric modulator agonist agonist inverse agonist inverse agonist neutral antagonist irreversible agonist agonist agonist agonist agonist agonist

35V 2GM ERM ERM ERM ETQ 5EH

covalent agonist agonist agonist agonist agonist antagonist antagonist

3.0 3.5 3.7 2.4 2.8 2.8 2.8 3.2 3.5 4.0 2.8 3.1 3.2 3.5 3.4 3.4 3.4 3.3 2.8 2.7 2.7 2.8 2.9 3.1

palmitic acid, tris buffer, phosphate ions, sodium ions, chloride ions, sulfate ions, glucose, maltose, cholesterol, and cocrystallized antibodies) and using Schrödinger’s Protein Preparation Wizard30 with default settings, including the assignment of bond orders, adding missing side chains, and missing hydrogens, creating disulfide bonds, removing crystalline waters, and applying protomer-optimization, and restrained minimization by OPLS_2005 force field. Selection of Receptor Structures for Consensus Docking. All 19 available aminergic ligand−receptor complexes were evaluated by examining the self-docking RMSD of their cocrystallized ligands, the number of noncovalent interactions recovered, and by conformational clustering of the receptor binding sites. From each resulting cluster the best performing structure was selected, thus picking an adequate number of structures for the optimization of the scoring method. The ligands were prepared for docking by generating possible predicted ionization states (at a pH range of 7.0 ± 2.0), tautomers, and stereoisomers using Schrödinger’s Ligand Preparation Wizard31 with default settings. Grid generation and self-dockings were performed with the Grid Generation application and Glide Docking SP (standard precision),32 respectively. RMSD values were calculated by superimposing the docked ligands using Schrödinger’s Atom Specification Language. Additionally the 2D interaction maps of the binding sites were downloaded from PDB,29 and the number of recovered secondary interactions was considered in the selection process. Conformational clustering of receptors with multiple experimental structures (muscarinic M2, adrenergic β2, 5-HT1B, and 5-HT2B) was performed by the average-type linkage method available in Schrö dinger’s Clustering of Conformers script33 using the RMSD-matrix calculated on atomic RMSD of the 5 Å proximity of the ligand. The resulting self-docking scores, ligand RMSD values, and the corresponding cluster trees are shown in Supporting

of representative structures suitable for docking based virtual screening. In the present study we have performed retrospective virtual fragment screening of a training set with actives derived from GPCR-SARfari.10 Docking scores of the fragments in different receptor structures were compared by their relative ranking scores and combined into a summed vote. The combination of protein structures was optimized at different ranking and voting cut-offs to improve the performance compared to standalone structures. The resulting ranking scheme was validated on screening data11 available for 5-HT1 and TA1 receptors. The consecutive application of the ligand(FrAGS) and structure-based (FrACS) methods was tested on an in-house fragment library consisting of 1183 compounds. Biological activity of 36 virtual screening hits was assessed on 5HT6 receptor in vitro. We identified six active fragments corresponding to a hit rate of 16.7%. The best hits were projected to IC50 measurement, resulting in four new fragment inhibitors of 5-HT6 receptors.

■

MATERIALS AND METHODS Preparation of Receptor Structures. Three structures of the human M2 muscarinic acetylcholine receptor (PDB IDs: 3UON, 4MQS, 4MQT),12,13 15 structures of the human β2 adrenoceptors (2RH1, 3D4S, 3NY8, 3NY9, 3NYA, 3P0G, 3PDS, 4GBR, 4LDE, 4LDL, 4LDO, 3P0G, 2R4R, 2R4S, 3KJ6, 4QKX),14−23 2−2 structures of human 5-HT1B (4IAQ, 4IAR),24 and 5-HT2B receptors (4IB4, 4NC3)25,26 and the Xray structures of human dopamine D3 receptor (3PBL)27 and human histamine H1 receptor (3RZE)28 were collected from the PDB database.29 Two adrenergic receptor apo structures (2R4R and 2R4S) and one structure (4QKX) containing a covalent agonist ligand were excluded from our analysis. The remaining structures were first prepared for docking by keeping the monomeric receptor, removing crystallization appendices (fatty acids, diethylene glycol, triethylene glycol, butanediol, B


Article

Journal of Chemical Information and Modeling Information Figures S1, S2, and S3. The first selection criterion was defined by best ligand RMSD value, with a cutoff 0.5 Å followed by the preferable Glide score as second criteria. Conformational clustering of the 12 β2-adrenergic receptor structures resulted in three clusters at 10.27 Å merge distance. One representative structure was selected from each cluster based on their self-docking Glide scores and ligand RMSD values. The β2 adrenergic structure (4LDE) with agonist ligand P0G was selected from the first cluster showing a remarkable self-docking result (RMSD = 0.2 Å, Glide score = −13.605; all eight original binding site interactions recovered). In the case of the second cluster, 2RH1 was selected, containing S-Carazolol agonist as ligand. Despite the moderate docking score, 2RH1 structure was favored over 3PDS due to its lower ligand RMSD value and better recovery of crucial interactions. The third selected structure was 3NY9 cocrystallized with an inverse agonist that showed the best RMSD value (0.5 Å) in selfdocking. The three available structures of the M2-muscarinic receptor (3UON, 4MQS, 4MQT) showed conformational difference at 18.2 Å merge distance, resulting in two main clusters, from among which 3UON and 4MQT structures were selected for docking studies. 4MQT was favored over 4MQS due to its docking RMSD (0.2 Å). Out of the two 5-HT1B and two 5-HT2B structures 1−1 representatives (4IAQ and 4IB4, respectively) were chosen based on their RMSD values. For the human D3 (3PBL) and H1 (3RZE) receptors the single available X-ray structures were selected. Altogether nine aminergic receptor structures were used in the optimization of the screening protocol. Compilation of the Training Set. The active ligand set was extracted from GPCR-SARfari (ChEMBL DB, version 3.00, 2012 June),10 containing 947 914 entries. Counter ions were removed and the highest available biological activity data was associated with each compound. The resulting 166 699 entries were filtered by ligand-efficiency metrics. Size-independent ligand-efficiency (SILE) was calculated and used as an activityfilter with SILE34 ≥ 1.951 resulting in 144 759 entries. Fragment-sized molecules with 8−22 non-hydrogen (heavy) atoms were kept, resulting in 10 477 fragments with listed binding activity data on several GPCRs. Here, 2183 compounds were considered to be promiscuous aminergic fragments, having SILE ≥ 1.95 on at least four different aminergic receptors. Afterward a diverse subset (692 fragments in total, clustered by Tanimoto distance of 0.3) was extracted to represent distinct molecular recognition types in the training set. The diversity selection was realized in Knime.com AG’s Konstanz Information Miner (Knime),35 using Tanimoto-based distance matrix calculator (based on ChemAxon’s Chemical Hashed Fingerprint; 36 binary type; length 1024 bit), Hierarchical Clustering (DistMatrix; linkage type complete linkage), and Hierarchical Cluster Assigner (distance threshold 0.3).35 The inactive training set was extracted from the opensource database ChEMBL (version 16),37 out of which the same 5000 fragments were filtered as in our previous study.1 The training set of 692 active and 5000 inactive fragments were prepared for docking with the same ligand preparation method used for the cocrystallized ligands resulting in 2381 active and 10 227 inactive generated structures (12 608 entries in total). The training sets were subjected to standard precision docking into the nine previously selected structures using Schrödinger’s Virtual Screening Workflow script.32 Consensus Ranking. The simultaneous docking of 12 608 entries into nine different structures resulted in nine Glide

Scores for each entry. In the case of multiple generated states for a single compound, the Glide Scores of the best states were kept. Docking score results for standalone receptor structures were sorted ascending by Glide score values. Rank numbers were assigned to each entry according to their ranking, resulting in nine different ordinal numbers for every single entry. The nine different ranks belonging to a single entry were translated into votes using a ranking cutoff. The Rank(x)% criterion was defined as a percent of upper ranking boundary. A compound receives a vote number of 1 if its docking score is in the top (x) % and 0 otherwise. The resulting votes were summed into a voting score (FrACS, fragment aminergic consensus score) with a range between 0 and 9. The developed scoring scheme involved: 1. Docking into nine structures, nine Glide Scores for each compound, sorting by Glide Scores. 2. Ranking docking scores per structure. 3. Voting rank numbers per each entry by the Rank(x)% criterion per structure. 4. Summing the votes (FrACS) per entry. 5. Sorting by FrACS, applying certain cut-offs (ranging between 0, 1, 2, ..., 9). Retrospective validations were evaluated by enrichment factors, calculated at each FrACS cutoff. Enrichments are subject to error. For this reason we have calculated 95% confidence intervals for every single enrichment factor value of the consensus rankings. For the evaluation of the 95% confidence intervals of the enrichment factors, we used an analytical error definition.38,39 An additional benefit of applying multiple receptor structures is to achieve higher enrichment and diversity of actives relative to single structure protocols. Enrichments achieved with standalone receptor structures were also calculated as reference (see Supporting Information Table S1). Calculation of the FrAGS Desirability Scores. In our previous study1 we demonstrated the efficiency of ligand-based FrAGS desirability scores for discriminating promiscuous aminergic GPCR fragments from decoys. FrAGS scores are based on six physicochemical properties (logD at pH = 7.4, TPSA, nitrogen and oxygen atom count, strongest basic pKa, and rotatable bond count) calculated using JChem for Excel functions (ChemAxon).36 These properties showed remarkable differences in case of promiscuous aminergic GPCR fragments with respect to randomly sampled fragments. The distribution functions of these six properties were translated into desirability functions. The six desirability functions define desired and undesired property ranges and provide a rather permissive boundary to avoid the exclusion of compounds with properties close to a cutoff limit.40 Desirability functions were defined using Knime Rule Engine,35 and the calculated desirability scores were summed up into the FrAGS score.

■

RESULTS AND DISCUSSION Optimization of the Protein Ensemble for FrACS. Enrichments and (and corresponding statistical errors) calculated on Rank(x)% values of x = 1, 5, 10, 20, and 30% and summed vote cutoff values of 1−9 are shown in Table 2. The top 5−6-fold enrichments were found at Rank(5)% criterion and sum-vote cutoff 5−8. The performance of the consensus ranking protocol was assessed by comparing enrichments achieved with standalone structures and by using all nine structures together. The C


Article

10−3) 10−3) 10−3) 10−3) 10−3) 10−3) 10−4) 10−4) 10−4) × × × × × × × × × (6.4 (3.5 (2.6 (1.9 (1.4 (1.0 (6.7 (3.9 (1.5 4.8 4.6 3.8 3.2 2.6 2.2 1.9 1.6 1.3 81 200 288 362 426 492 572 634 679 10−2) 10−2) 10−3) 10−3) (na) (2.9 × (1.5 × (8.6 × (4.0 ×

FrACS cutoff. bNumber of selected compounds. cNumber of actives. d95% confidence interval. a

10 ) 10−2) 10−2) 10−3) 10−3) 10−3) 10−3) 10−3) (1.5 (2.0 (1.1 (7.7 (5.6 (4.2 (2.8 (1.4

× × × × × × × ×

−2

A S

138 361 618 944 1376 1875 2470 3245 4285 10−3) 10−3) 10−3) 10−3) 10−3) 10−3 10−3) 10−4) 10−4) × × × × × × × × × (9.5 (5.5 (3.9 (3.1 (2.3 (1.7 (1.2 (7.0 (3.2

EF (95% CI)

5.4 5.1 4.5 4.1 3.5 2.9 2.4 2.0 1.6 41 102 169 237 312 390 465 571 651

A S

62 164 308 481 731 1089 1616 2331 3415 10−2) 10−2) 10−3) 10−3) 10−3) 10−3) 10−3) 10−3) 10−4) × × × × × × × × × (2.6 (1.2 (7.9 (5.7 (4.2 (3.4 (2.5 (1.6 (7.3

EF (95% CI)

comparisons were made at the same top rank percentages of 1, 5, 10, 20, and 30%. Table S1 in the Supporting Information summarizes the enrichments of the nine, individual structures. The histamine H1 (3RZE) and the muscarinic M2 (4MQT) structures achieved the best enrichments for the training set highlighted in the Table S1 in the Supporting Information. While two structures (3PBL and 4LDE) showed only moderate enrichments, seven structures demonstrated 2 to 5-fold enrichment factors. Table 2 shows the corresponding enrichments by certain sum-vote cut-offs and Rank(x)% criteria for the consensus score. In most cases, we found that the consensus method yields better enrichments as compared to individual structures, in particular at FrACS ≥ 5 cutoff and Rank(5)% criterion. This result suggests that the combination of multiple receptor structures may provide promising ligands with diverse pharmacophore combinations. The highlighted rows in Table 2 show that within a 20−90 range of selected compounds at a Rank(5)% criterion, 10−60 actives are recovered yielding remarkable enrichment of 5−8. Validation of FrACS on Experimental Screening Data. Two screening data sets targeting aminergic receptors were downloaded from the PubChem public repository.9 Altogether 200 fragment-sized (number of heavy atoms between 8 and 22) confirmed agonists and antagonists with an activity threshold of 10 μM on 31 5-HT1A or 5-HT1E and 169 TA1 were collected. Experimentally inactive fragments (5-HT1 3100 and TA1 16 900) were randomly selected from the corresponding screening sets by keeping the proportion of actives at 1%. The 5-HT1 (3131 fragments) and TA1 (17 069 fragments) sets were prepared for docking with the ligand preparation method described above resulting in 37 167 generated structures. The screening data sets were subjected to docking into the nine previously selected structures using Schrödinger’s Virtual Screening Workflow script. FrACS consensus ranking scores were calculated for both validation sets (Table 3). The performance of the standalone structures for the serotonin-receptor targeted screening set is shown in Supporting Information Table S2. In this case the 5-HT2B structure (4IB4) highlighted in the table yielded the highest enrichment factors (7.7 to 16.3) at Rank(5)%. The FrACS score achieved similar enrichments (8.2 to 16.8) at Rank(1)% and Rank(5)%, as shown in Table 3. In case of the trace-amine receptor targeted HTS data set the best enrichments (2.5−3.5) were achieved with the 5-HT1B-receptor structure (4IAQ) highlighted in Supporting Information Table S2. Conversely the FrACS method, where an ensemble of structures was used, showed improved enrichments (Table 3) at the same RANK(x)% selections around Rank(5)%. Combining FrACS with the Ligand Based Prefilter FrAGSRetrospective and Prospective Validations. Proposing the consecutive application of ligand- and structure-based methods we screened fragment-sized molecules from the ChEMBL (version 16) database37 using the ligandbased desirability score FrAGS followed by the structure-based docking protocol FrACS. Salt-form duplicates were first removed from ChEMBL, and after fragment-size filtering 327 419 molecules remained containing the 2183 active training set molecules (intersection with GPCR SARfari10). Next we have calculated the FrAGS scores for this ChEMBL fragment set. The data set was sorted by descending desirability scores, and filtered by a FrAGS ≥ 5 cutoff. The resulting 28 816 fragments are predicted to be promiscuous aminergic compounds based on their physicochemical properties.

5.5 6.2 5.4 5.0 4.6 4.0 3.5 2.8 2.1 10 28 56 97 152 207 291 412 572

A S

15 37 86 161 273 421 693 1198 2215 na 7.1 6.0 5.8 5.4 5.1 4.4 3.6 2.6

EF (95% CI) A

0 6 14 29 57 98 155 255 435 0 7 19 41 86 158 291 591 1354

S EF (95% CI)

na na na na 4.1 5.7 5.5 4.8 3.6 0 0 0 0 1 9 20 51 159

A S

F

0 0 0 0 1 13 30 88 368

Rank(20)% Rank(10)% Rank(5)%

d c

Rank(1)%

b a

Table 2. Enrichments Achieved by FrACS on Training Set

9 8 7 6 5 4 3 2 1

Rank(30)%

EF (95% CI)

Journal of Chemical Information and Modeling

D


a

E

F

0 2 5 12 33 86 201 510 1859

3 12 30 57 111 198 358 660 1280

S

Sb

33.7 (1.78) 8.4 (0.3) 16.8 (0.2) 10.6 (0.1) 8.2 (7.5 × 10−2) 6.1 (4.6 × 10−2) 4.0 (2.7 × 10−2) 2.3 (1.5 × 10−2) 1.6 (7.3 × 10−2)

0 0 0 0 0 4 10 16 36

A

na 0.0 0.0 0.0 0.0 4.7 5.0 3.2 2.0

(na) (na) (na) (na) (3.4 × (2.2 × (1.1 × (4.3 ×

10−2) 10−2) 10−2) 10−3)

EF (95% CI)

Rank(1)%

1 1 5 6 9 12 14 15 20

EF (95% CI)d

Rank(1)%

Ac

7 32 110 246 475 886 1723 3327 6765

S

137 425 754 1120 1508 1886 2221 2601 2920

S 6.6 2.9 2.1 1.5 1.5 1.3 1.2 1.1 1.0

0 2 6 9 16 25 43 64 98

A

× × × × × × × × ×

10−2) 10−2) 10−2) 10−3) 10−3) 10−3) 10−3) 10−3) 10−3)

0.0 6.3 5.5 3.7 3.4 2.9 2.5 1.9 1.5

(na) (6.6 × (3.2 × (1.8 × (1.2 × (7.7 × (4.9 × (2.9 × (1.4 × 10 ) 10−2) 10−2) 10−2) 10−3) 10−3) 10−3) 10−3)

−2

EF (95% CI)

(6.2 (2.2 (1.3 (8.7 (5.9 (4.3 (3.2 (2.2 (1.2

EF (95% CI)

Rank(5)%

9 12 16 17 22 24 26 28 30

A

Rank(5)%

42 175 435 877 1544 2571 4183 6666 10649

S

1644 2900 2999 3028 3052 3065 3077 3097 3122

S 0.98 (5.9 × 10−3) 1.01 (1.7 × 10−3) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0)

EF (95% CI)

3 7 20 34 47 66 80 95 135

A 7.2 (6.0 × 10−2) 4.0 (2.2 × 10−2) 4.6 (1.4 × 10−2) 3.9 (8.7 × 10−3) 3.1 (5.6 × 10−3) 2.6 (3.7 × 10−3) 1.9 (2.3 × 10−3) 1.4 (1.5 × 10−3) 13 (7.3 × 10−4)

EF (95% CI)

Rank(10)%

TA1

A

Rank(10)%

5-HT1

16 29 31 31 31 31 31 31 31

FrACS cutoff. bNumber of selected compounds. cNumber of actives. d95% confidence interval.

9 8 7 6 5 4 3 2 1

9 8 7 6 5 4 3 2 1

Fa

Table 3. FrACS Enrichments Achieved on PubChem Validation Sets

289 1019 2012 3338 4950 6973 9234 11881 14593

1855 3114 3126 3127 3127 3127 3127 3127 3127

S

S 0.98 (5.2 × 10−3) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0)

EF (95% CI)

14 41 63 80 93 103 126 142 161

A 4.9 4.0 3.2 2.4 1.9 1.5 1.4 1.2 1.1

(1.8 (8.0 (4.6 (2.9 (1.9 (1.4 (9.2 (6.0 (2.8

× × × × × × × × ×

10−2) 10−3) 10−3) 10−3) 10−3) 10−3) 10−4) 10−4) 10−4)

EF (95% CI)

Rank(20)%

18 31 31 31 31 31 31 31 31

A

Rank(20)%

1054 3006 5116 7269 9342 11384 13170 14803 16281

1855 3114 3126 3127 3127 3127 3127 3127 3127

S

S

A

EF (95% CI) 0.98 (5.2 × 10−3) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0) 1.0 (0.0)

37 66 89 103 120 139 149 159 166

A

3.6 2.2 1.8 1.4 1.3 1.2 1.1 1.1 1.0

(7.5 (3.1 (1.9 (1.3 (9.4 (6.5 (4.8 (3.1 (1.6

× × × × × × × × ×

10−3) 10−3) 10−3) 10−3) 10−4) 10−4) 10−4) 10−4) 10−4)

EF (95% CI)

Rank(30)%

18 31 31 31 31 31 31 31 31

Rank(30)%

Journal of Chemical Information and Modeling Article


0

0

0

3

8

36

135

451

2071

8

7

6

5

4

3

2

1

S

9

c

26

4

0

0

0

0

0

0

140

A

d

f

g

39

3

0

680 1481 3195 7792

4.4 [0.6 (3.6 × 10−3)]

8.6 [1.2 (2.7 × 10−3)]

10.1 [1.4 (1.3 × 10−3)]

267

106

S

0.0 (na)

0.0 (na)

0.0 (na)

na

na

na

EF [EF (95% CI) ]

e

Rank(1)%

9

3

1

0

469

234

107

51

17

A

1105 2198

9.5 [1.3 (3.7 × 10−3)] 11.2 [1.5 (2.4 × 10−3)]

9.0 [1.2 (5.2 × 10−4)]

11.0 [1.5 (1.0 × 10−3)] 12504

6684

3843

541

12.7 [1.7 (6.7 × 10−3)]

10.8 [1.5 (1.6 × 10−3)]

207

7

11.5 [1.6 (1.1 × 10−2)]

S

45

na 50.0 [6.8 (13.8)]

EF [EF (95% CI)]

Rank(5)%

2

1

728

454

291

173

87

40

13

A

(3.3

(6.0

(9.1

(1.3

(1.9

(2.7

(4.1

(7.7

(3.9

21.4 [2.9 × 10−2)] 6.7 [0.9 × 10−3)] 9.4 [1.3 × 10−3)] 11.1 [1.5 × 10−3)] 11.8 [1.6 × 10−3)] 11.8 [1.6 × 10−3)] 11.4 [1.5 × 10−4)] 10.2 [1.4 × 10−4)] 8.7 [1.2 × 10−4)]

EF [EF (95% CI)]

Rank(10)%

18498

12703

8909

6134

3936

2298

1219

484

103

S

45

10

1022

733

575

438

317

187

101

A

(1.9

(3.3

(4.7

(6.5

(9.1

(1.3

(1.9

(3.2

(7.1

14.6 [1.9 × 10−3)] 13.9 [1.9 × 10−3)] 12.4 [1.7 × 10−3)] 12.2 [1.7 × 10−3)] 12.1 [1.6 × 10−4)] 10.7 [1.4 × 10−4)] 9.7 [1.3 × 10−4)] 8.7 [1.2 × 10−4)] 8.3 [1.1 × 10−4)]

EF [EF (95% CI)]

Rank(20)%

21899

17360

13673

10453

7657

5251

3137

1538

456

S 56

1165

909

774

653

524

382

269

145

A

18.4 [2.5 (3.6 × 10−3)] 14.1 [1.9 (1.7 × 10−3)] 12.9 [1.7 (1.1 × 10−3)] 10.9 [1.5 (7.3 × 10−4)] 10.3 [1.4 (5.4 × 10−4)] 9.4 [1.3 (4.1 × 10−4)] 8.5 [1.2 (3.1 × 10−4)] 7.9 [1.1 (2.2 × 10−4)] 8.0 [1.1 (1.1 × 10−4)]

EF [EF (95% CI)]

Rank(30)%

a

FrAGS prefiltering step EF = 6.7 (4.1 × 10−4) (95% confidence interval). bFrACS cutoff. cNumber of selected compounds. dNumber of actives. eEnrichment factor (Sequential). fEnrichment factor (FrACS). g95% confidence interval.

F

b

Table 4. Sequential Screening Enrichments Achieved by the Consecutive Application of FrAGSa and FrACS on the ChEMBL Fragment Set

Journal of Chemical Information and Modeling Article

F


Article

Journal of Chemical Information and Modeling FrAGS filtering enriched the actives by 6.7-fold by retrieving 1285 out of 2183 actives. The resulting 28 816 fragments were then subjected to the default Ligand Preparation method resulting in 87 176 molecules that were docked into the nine aminergic structures. In the case of multiple generated states for a single compound, the Glide Scores of the best states were translated into rankings, and votes were assigned according to the Rank(x)% criteria. The resulting votes were summed up, and enrichmentsshown in Table 4were examined at every FrACS cutoff (1−9). In a realistic scenario, sequential filtering may start with similarly sized data sets consisting of about 105 fragments as was the case for our ChEMBL fragment training set (327 419 entries). The FrAGS score may serve as a prefiltering step with low computational cost to select compounds with favorable physicochemical properties for class A aminergic GPCR affinity. The structure-based FrACS method can be applied as a second filtering step to further enrich promising actives for experimental library screenings. Our studies on the ChEMBL data set showed 8−18-fold total enrichment in the case of selecting 4−500 fragments. Enrichments increased with increasing Rank(x)% criterion, showing their dependence on the data set size. Furthermore, the preselected fragments already satisfy main features for aminergic promiscuity, thus more permissive Rank(x)% criterion should be applied. Our results showed the best enrichments of the known actives at Rank(30)%. The prospective validation was carried out on our in-house fragment library (number of heavy atoms between 8 and 22, clogP ≤ 3, H-bond donors ≤ 3, H-bond acceptors ≤ 3, number of rotatable bonds ≤ 3, TPSA ≤ 60 Å) containing 1183 fragments. First, the FrAGS desirability scores were calculated, and fragments satisfying FrAGS ≥ 5 criterion were kept. The resulting 211 prefiltered fragments are expected to be enriched in aminergic promiscuous fragments according to their physicochemical features. These desirability-scored hits were subjected to Ligand Preparation with default settings, and the resulting 2126 isomers were docked into the 9 aminergic receptors. The docked fragments were ranked by Glide Scores, and nine votes were assigned to each fragment at Rank(30)% value. As described above, the sum-vote cutoff may be defined according to the experimental screening capacity. We selected fragments at sum-vote cutoff ≥ 6. The 36 virtual screening hits were subjected to antagonist activity measurements in a cellbased assay against 5-HT6 receptor.41,42 We have identified eight in vitro hits with remarkable inhibitory levels (biological activities are summarized in Table 5), two of them appeared to be PAINS frequent hitters43 (phenolic Mannich-bases44). The remaining six fragment hits, corresponding to a hit rate of 16.7% possess pharmacophore features of typical aminergic compounds. We selected compounds 4, 5, 7, and 8 for IC50 measurements. The dose−response curves are shown in Figure 1. We have found that our virtual screening fragment hits are new 5-HT6 receptor ligands with low micromolar IC50 values. These fragment hits represent novel structural classes compared to the ligands in the X-ray structures (collected in the Supporting Information Table S3). Both the “spiro” compound 4 (2′-(3-fluorophenyl)spiro[indoline-3,3′-pyrrolidin]-2-one) and the N-heterocycle-containing tryptaminederivatives are structurally diverse from the known crystallized ligands. We have accomplished a counter-screen for compounds 5, 7, and 8 against a peptidergic target Cannabioid Receptor

Table 5. Antagonist Efficacy of FrACS Hits against 5-HT6 Receptora

a

5-HT6 serotonin receptor inhibition assay,42 antagonist efficacy at 50 μM final concentration 100%; same inhibition with the control antagonist, 0%; no inhibition (agonist only); negative control 1% DMSO; positive control 10 μM SB271046.

Subtype-1 (CB1) in order to check specificity of compounds toward aminergic subfamily. The compounds were found to be virtually inactive against CB1 (Table S4). Binding Mode Analysis of Active Fragments in 5-HT6 Homology Models. In a recent study45 Vass and colleagues have analyzed a 5-HT6 homology model (built using h5-HT2B in complex with ergotamine as template (PDB ID: 4IB4))46 by membrane-embedded molecular dynamics simulations. A chemometric analysis of enrichments from retrospective virtual screening studies resulted in nine MD-frames containing SB742457 selective 5-HT6 antagonist47 as ligand. We projected our primary hits (compounds 4, 5, 7, and 8) to docking studies using the nine 5-HT6 model structures. The experiments were performed with prepared ligands using Glide Single Precision docking. Poses corresponding to the best Glide scores were compared to the orientation and interactions of SB-742457 as reference. Most important interaction of the reference ligand is the H-bond formed between the protonated basic amine moiety in the piperazine ring and the highly conserved Asp1063.32. Further two H-bonds are formed between the sulfonyl-oxygens of the ligand and Asn2886.55 and Ser1935.43. The quinoline ring occupies a hydrophobic cavity (noted as first hydrophobic pocket), surrounded by three aromatic residues (Trp2816.48, Phe2846.51, Phe2856.52). The second hydrophobic cavity is formed by Val1073.33, Ala1574.56, Leu1604.59, Pro1614.60, Leu1644.63, etc., around the phenylmoiety of SB-742457. Four representative docking poses of our primary hits are shown in Figure 2. Twenty-eight out of the thirty-six docking poses were found in an orientation that is closely analogous to the ones of SB742457 in all the nine frames. Characteristic interactions are G


Article

Journal of Chemical Information and Modeling

Figure 1. Dose−response curves of compounds 4, 5, 7, and 8 obtained on 5-HT6 receptors.

Figure 2. Representative docking poses of compounds 4, 5, 7, and 8.

H


Article

Journal of Chemical Information and Modeling discussed here using the example frame annotated as “1001” in ref 44. The “spiro” compound 4 (2′-(3-fluorophenyl)spiro[indoline-3,3′-pyrrolidin]-2-one) interacts through the protonated secondary amine nitrogen of the pyrrolidine-ring in all frames. The benzene ring of the indoline scaffold occupies the same hydrophobic pocket, like the benzene-sulfonyl part of the reference ligand. The fluorophenyl part of compound 4 is oriented toward the same hydrophobic pocket (Phe2846.51, Phe2856.52, and Trp2816.48) occupied by the quinoline-ring of SBP-742457. Interestingly, an H-bond was formed between the backbone carbonyl oxygen of Asn2886.55 and the amidenitrogen of the indoline-2-one scaffold in case of 6/9 poses. Compound 5 (2-(2-(pyridin-2-yl)-1H-indol-3-yl)ethanamine) forms an H-bond with the conserved aspartate (Asp1063.32) in all poses, and π−π stacking interactions are formed between the pyridine-ring and Phe2856.52, Trp2816.48 hydrophobic residues; furthermore, the benzene part of the indole-ring occupies the same hydrophobic cavity as the benzene-sulfonyl part of SB742457. The pose of compound 7 (2-(2-(quinolin-6-yl)-1Hindol-3-yl)ethanamine) is in good agreement with the reference ligand, by forming the desired H-bond with Asp1063.32, and both of the hydrophobic (first and second) subpockets are occupied by its quinoline and indole rings, respectively. In the case of compound 8 (2-(7-chloro-2-(pyridin-4-yl)-1H-indol-3yl)ethanamine) an interesting halogen-bond is formed between the chlorine and the amino moiety of Asn2886.55 in six out of nine docking poses. The aromatic-stacking interactions and the formation of the H-bond with the aspartate are both wellrepresented.

set is recommended. In the case of smaller (103) libraries, more permissive sum-vote cutoffs between 5 and 7 are proposed. On the other hand, the selection of a larger output library can be advantageously performed by the combination of a more permissive Rank(x)% and a stricter sum-vote cutoff as it was demonstrated by the screening of the ChEMBL fragment database using Rank(30)% and a 6−9 sum-vote cutoff (Table 4). Our prospective validation on the consecutive application of FrAGS and FrACS protocols started from an in-house library of 1183 fragments and resulted in four 5-HT6 receptor antagonists showing good in vitro inhibitory activity with IC50 values in the micromolar to the submicromolar range. The identified hits have not been published yet as 5-HT6 inhibitors. Their structures contain characteristic aminergic/serotonergic pharmacophore features: basic nitrogen, carbocyclic, and Nheterocyclic scaffolds. The prospective validations of our ligand and structure based methods showed that the similar characteristics of the aminergic binding pockets allows the identification of frits active on an aminergic receptor being not included in our model. This finding underpins that the FrACS based virtual screening either itself or in combination with the FrAGS is an efficient tool for the design of focused libraries for class A aminergic GPCRs and ultimately to support the identification of fragment ligands for these receptors.

■

EXPERIMENTAL PROCEDURES Cell-Based Assays. CHO-K1 cells expressing stably the mitochondrially targeted aequorin (luminescent indicator) and Gα16 were used for the assay.42 The principle of the assay is that aequorin (derived from Aequorea victoria) complex emits blue light while binding Ca2+ ions. The aequorin and Gα16 expressing CHO-K1 cells were transiently transfected with plasmid harboring the gene expressing human 5-HT6 or CB1 respectively using Roche X-treme GENE HP DNA Transfection Reagent. Cells were grown for 48 h after transient transfection. One day before the assay the culture medium was changed to an antibiotic free medium and the cells were grown for an additional 6 h, then the cells were detached by gentle flushing with PBS/0.5 mM EDTA, recovered by centrifugation, and resuspended at 1 × 106 cells/mL density in assay medium (DMEM/HAM’s F12 with HEPES, without phenol red (Sigma, D6434), + 0.1% BSA + 2 mM glutamine) in a Falcon tube. Coelenterazine h (LifeTechnologies C6780) was added at a final concentration of 5 μM and the cells were incubated overnight at room temperature using constant shaking protected from light. Before the measurement the cells were diluted 2 fold in assay medium and incubated for 60 min. The measurement was executed in Optiplate TM-96 plates (PerkinElmer 6005290) and the luminescent emission was detected by an AppliscanTM (Thermo) plate-reader. Cell suspension (45 μL)/well + 5 μL antagonist (diluted in assay medium) was preincubated in the plate, the reaction was initiated with the addition of agonist EMD 386088 (50 μL) at a concentration of 50 nM. In case of CB1 measurement agonist CP55940 (50 μL) was added at a concentration of 25 nM. The time lapse curves of receptor activation signals were recorded well by well. After detection of the baseline (8 s) the agonist was injected in one well and the change of intracellular Ca2+ level released due to receptor activation was monitored for 45 s. Peak luminescence signal was used for evaluation of the measurements. For assay development and validation, SB271046 47 was used as antagonist control (Supporting Information, Figures S4 and S5) for 5-HT6 assay, and

■

CONCLUSIONS The emerging number of available aminergic GPCR X-ray structures raised the opportunity to apply them collectively for the design of focused aminergic fragment libraries. In this study a consensus score based virtual screening protocol was developed to identify orthosteric fragment sized class A aminergic GPCR ligands. Representative aminergic receptor structures were selected by clustering available X-ray structures and by doing self-docking experiments to choose the best performing structure from each cluster. The nine selected structures were subjected to parallel docking workflows, with an input training set of fragments compiled for our preceding ligand-based study.1 The docking scores for each individual receptor were translated to ranking lists, and the ranks were used to calculate votes at a ranking cutoff for each fragment entry. A consensus ranking vote score, called fragment aminergic consensus score (FrACS), was generated by the summation of the nine votes. The performance of the FrACS method was tested on the ChEMBL training set and on independent open-source validation sets extracted from the PubChem screening database. Our consensus ranking method achieved improved enrichments compared to docking into standalone receptor structures. The screening efficiency of FrACS has been demonstrated by analyzing experimental screening results for the 5-HT1 and TA1 receptors. Moreover, the sequential application of the FrAGS and FrACS protocols was also investigated. Our results showed that the application of FrACS either alone or following a FrAGS prescreening provides the most promising aminergic fragments when applied at a rather strict Rank(5)% criterion (see Tables 2 and 3 for the standalone and Table 4 for the sequential applications). On the other hand, in a sequential application the fine-tuning of the sum-vote cutoff according to the desired size of the filtered data I


Article

Journal of Chemical Information and Modeling AM25148 was used as antagonist control (Supporting Information Figures S6 and S7) respectively. All compounds were screened at a fixed concentration of 50 μM. The hit compounds with better than 100% inhibitory levels were selected for IC50 measurements.

■

(7) Spyrakis, F.; Cavasotto, C. N. Open challenges in structure-based virtual screening: Receptor modeling, target flexibility consideration and active site water molecules description. Arch. Biochem. Biophys. 2015, 583, 105−19. (8) Sinko, W.; Lindert, S.; McCammon, J. A. Accounting for receptor flexibility and enhanced sampling methods in computer-aided drug design. Chem. Biol. Drug Des. 2013, 81, 41−49. (9) Andrews, S. P.; Brown, G. A.; Christopher, J. A. Structure-Based and Fragment-Based GPCR Drug Discovery. ChemMedChem 2014, 9, 256−275. (10) ChEMBL GPCR SARfari Home Page. https://www.ebi.ac.uk/ chembl/sarfari/gpcrsarfari (accessed February 1, 2014). (11) PubChem home page. https://pubchem.ncbi.nlm.nih.gov/ (accessed March 14, 2014). (12) Haga, K.; Kruse, A. C.; Asada, H.; Yurugi-Kobayashi, T.; Shiroishi, M.; Zhang, C.; Weis, W. I.; Okada, T.; Kobilka, B. K.; Haga, T.; Kobayashi, T. Structure of the human M2 muscarinic acetylcholine receptor bound to an antagonist. Nature 2012, 482, 547−551. (13) Kruse, A.C.; Ring, A. M.; Manglik, A.; Hu, J.; Hu, K.; Eitel, K.; Hubner, H.; Pardon, E.; Valant, C.; Sexton; Christopoulos, A.; Felder, C. C.; Gmeiner, P.; Steyaert, J.; Weis, W. I.; Garcia, K. C.; Wess, J.; Kobilka, B. K. Activation and allosteric modulation of a muscarinic acetylcholine receptor. Nature 2013, 504, 101−106. (14) Cherezov, V.; Rosenbaum, D. M.; Hanson, M. A.; Rasmussen, S. G. F.; Thian, F. S.; Kobilka, T. S.; Choi, H. J.; Kuhn, P.; Weis, W. I.; Kobilka, B.K.; Stevens, R. C. High-resolution crystal structure of an engineered human b2-adrenergic G protein-coupled receptor. Science 2007, 318, 1258−1265. (15) Hanson, M. A.; Cherezov, V.; Griffith, M. T.; Roth, C. B.; Jaakola, V. P.; Chien, E. Y.; Velasquez, J.; Kuhn, P.; Stevens, R.C. A specific cholesterol binding site is established by the 2.8 A° structure of the human b2-adrenergic receptor. Structure 2008, 16, 897−905. (16) Wacker, D.; Fenalti, G.; Brown, M. A.; Katritch, V.; Abagyan, R.; Cherezov, V.; Stevens, R. C. Conserved binding mode of human beta2 adrenergic receptor inverse agonists and antagonist revealed by X-ray crystallography. J. Am. Chem. Soc. 2010, 132, 11443−11445. (17) Rosenbaum, D. M.; Zhang, C.; Lyons, J. A.; Holl, R.; Aragao, D.; Arlow, D. H.; Rasmussen, S. G. F.; Choi, H. J.; Devree, B. T.; Sunahara, R. K.; Chae, P. S.; Gellman, S. H.; Dror, R. O.; Shaw, D. E.; Weis, W. I.; Caffrey, M.; Gmeiner, P.; Kobilka, B. K. Structure and function of an irreversible agonist-β(2) adrenoceptor complex. Nature 2011, 469, 236−240. (18) Zou, Y.; Weis, W. I.; Kobilka, B. K. N-terminal T4 lysozyme fusion facilitates crystallization of a G protein coupled receptor. PLoS One 2012, 7, e46039−e46039. (19) Ring, A. M.; Manglik, A.; Kruse, A. C.; Enos, M. D.; Weis, W. I.; Garcia, K. C.; Kobilka, B. K. Adrenaline-activated structure of β2adrenoceptor stabilized by an engineered nanobody. Nature 2013, 502, 575−579. (20) Rasmussen, S. G. F.; Choi, H. J.; Fung, J. J.; Pardon, E.; Casarosa, P.; Chae, P. S.; Devree, B. T.; Rosenbaum, D. M.; Thian, F. S.; Kobilka, T. S.; Schnapp, A.; Konetzki, I.; Sunahara, R. K.; Gellman, S. H.; Pautsch, A.; Steyaert, J.; Weis, W. I.; Kobilka, B. K. Structure of a nanobody-stabilized active state of the β(2) adrenoceptor. Nature 2011, 469, 175−180. (21) Rasmussen, S. G.; Choi, H. J.; Rosenbaum, D. M.; Kobilka, T. S.; Thian, F. S.; Edwards, P. C.; Burghammer, M.; Ratnala, V. R.; Sanishvili, R.; Fischetti, R. F.; Schertler, G. F.; Weis, W. I.; Kobilka, B. K. Crystal structure of the human b2 adrenergic G-protein-coupled receptor. Nature 2007, 450, 383−388. (22) Bokoch, M. P.; Zou, Y.; Rasmussen, S. G.; Liu, C. W.; Nygaard, R.; Rosenbaum, D. M.; Fung, J. J.; Choi, H. J.; Thian, F. S.; Kobilka, T. S.; Puglisi, J. D.; Weis, W. I.; Pardo, L.; Prosser, R. S.; Mueller, L.; Kobilka, B. K. Ligand-specific regulation of the extracellular surface of a G-protein-coupled receptor. Nature 2010, 463, 108−112. (23) Weichert, D.; Kruse, A. C.; Manglik, A.; Hiller, C.; Zhang, C.; Hubner, H.; Kobilka, B. K.; Gmeiner, P. Covalent agonists for studying G protein-coupled receptor activation. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 10744−10748.

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jcim.5b00598. Conformational clustering, self-docking scores and RMSD values; enrichments achieved by the standalone structures, structures of the compared screening hits, experimental procedures of the in vitro assay, 1H NMR spectra of the in vitro hits, PDB ligands in comparison with prospective validation hit compounds, and CB1 counter-assay results (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*Phone: +36-1-382-6821. E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors are thankful to Dávid Bajusz and all members of the Medicinal Chemistry Research Group (Research Centre for Natural Sciences) for valuable ideas and helpful discussions. This work was supported by National Brain Research Program KTIA-NAP-13-1-2013-0001.

■

ABBREVIATIONS FrAGS, fragment aminergic GPCR score; FrACS, fragment aminergic consensus score; GPCR, G-protein coupled receptor; ASL, atom specification language; RMSD, root mean square deviation; ChEMBL DB, database of the European Bioinformatics Institute (as part of European Molecular Biology Laboratory, EMBL); TPSA, topological polar surface area; HTS, high-throughput screening; SILE, size-independent ligand efficiency; EF, enrichment factor; 5-HT, 5-hydroxy-triptamine receptor; TA1, trace-amine receptor subtype-1; 95% CI, 95% confidence interval; MD, molecular dynamics; DMSO, dimethyl sulfoxide

■

REFERENCES

(1) Kelemen, Á .A.; Ferenczy, G. G.; Keserű , G. M. A desirability function-based scoring scheme for selecting fragment-like class A aminergic GPCR ligands. J. Comput.-Aided Mol. Des. 2015, 29, 59−66. (2) Palczewski, K.; Kumasaka, T.; Hori, T.; Behnke, C. A.; Motoshima, H.; Fox, B. A.; Le Trong, I.; Teller, D. C.; Okada, T.; Stenkamp, R. E.; Yamamoto, M.; Miyano, M. Crystal structure of rhodopsin: a G protein-coupled receptor. Science 2000, 289, 739−45. (3) Topiol, S.; Sabio, M. X-ray structure breakthroughs in the GPCR transmembrane region. Biochem. Pharmacol. 2009, 78, 11−20. (4) Bohacek, R. S.; McMartin, C.; Guida, W. C. The art and practice of structure-based drug design: a molecular modeling perspective. Med. Res. Rev. 1996, 16, 3−50. (5) Totrov, M.; Abagyan, R. Flexible ligand docking to multiple receptor conformations: a practical alternative. Curr. Opin. Struct. Biol. 2008, 18, 178−184. (6) Huang, S. Y.; Zou, X. Ensemble docking of multiple protein structures: considering protein structural variations in molecular docking. Proteins: Struct., Funct., Genet. 2007, 66, 399−421. J


Article

Journal of Chemical Information and Modeling (24) Wang, C.; Jiang, Y.; Ma, J.; Wu, H.; Wacker, D.; Katritch, V.; Han, G. W.; Liu, W.; Huang, X. P.; Vardy, E.; McCorvy, J. D.; Gao, X.; Zhou, X. E.; Melcher, K.; Zhang, C.; Bai, F.; Yang, H.; Yang, L.; Jiang, H.; Roth, B. L.; Cherezov, V.; Stevens, R. C.; Xu, H. E. Structural Basis for Molecular Recognition at Serotonin Receptors. Science 2013, 340, 610−614. (25) Wacker, D.; Wang, C.; Katritch, V.; Han, G. W.; Huang, X. P.; Vardy, E.; McCorvy, J. D.; Jiang, Y.; Chu, M.; Siu, F. Y.; Liu, W.; Xu, H. E.; Cherezov, V.; Roth, B. L.; Stevens, R. C. Structural features for functional selectivity at serotonin receptors. Science 2013, 340, 615− 619. (26) Liu, W.; Wacker, D.; Gati, C.; Han, G. W.; James, D.; Wang, D.; Nelson, G.; Weierstall, U.; Katritch, V.; Barty, A.; Zatsepin, N. A.; Li, D.; Messerschmidt, M.; Boutet, S.; Williams, G. J.; Koglin, J. E.; Seibert, M. M.; Wang, C.; Shah, S. T.; Basu, S.; Fromme, R.; Kupitz, C.; Rendek, K. N.; Grotjohann, I.; Fromme, P.; Kirian, R. A.; Beyerlein, K. R.; White, T. A.; Chapman, H. N.; Caffrey, M.; Spence, J. C.; Stevens, R. C.; Cherezov, V. Serial femtosecond crystallography of G protein-coupled receptors. Science 2013, 342, 1521−1524. (27) Chien, E. Y.; Liu, W.; Zhao, Q.; Katritch, V.; Han, G. W.; Hanson, M. A.; Shi, L.; Newman, A. H.; Javitch, J. A.; Cherezov, V.; Stevens, R. C. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. Science 2010, 330, 1091− 1095. (28) Shimamura, T.; Shiroishi, M.; Weyand, S.; Tsujimoto, H.; Winter, G.; Katritch, V.; Abagyan, R.; Cherezov, V.; Liu, W.; Han, G. W.; Kobayashi, T.; Stevens, R. C.; Iwata, S. Structure of the human histamine H1 receptor complex with doxepin. Nature 2011, 475, 65− 70. (29) RCSB PDB. www.rcsb.org (accessed March 14, 2014). Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T. N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. Protein Data Bank. Nucleic Acids Research 2000, 28, 235−242. (30) Protein Preparation, Version X; Schrödinger, LLC: New York, NY, 2011. (31) LigPrep, Version 2.5; Schrödinger, LLC: New York, NY, 2011. (32) Glide, Version 5.7; Schrödinger, LLC: New York, NY, 2011. (33) Clustering of Conformers; Schrödinger, LLC: New York, NY. http://www.schrodinger.com/scriptcenter/ (accessed June 19, 2014). (34) Nissink, J. W. M. Simple Size-Independent Measure of Ligand Efficiency. J. Chem. Inf. Model. 2009, 49 (6), 1617−22. (35) Knime Desktop (Konstanz Information Miner), version 2.9.1; 2014. (36) JChem for Excel, ChemAxon for Knime, ChemAxon fingerprint, version 5.2; Chemaxon: Budapest, Hungary, 2014. (37) ChEMBL home page. ftp://ftp.ebi.ac.uk/pub/databases/ chembl/ChEMBLdb/ (accessed 2014). (38) Nicholls, A. Confidence limits, error bars and method comparison in molecular modelling. Part 1: The calculation of confidence intervals. J. Comput.-Aided Mol. Des. 2014, 28, 887−918. (39) McGann, M.; Nicholls, A.; Enyedy, I. The statistics of virtual screening and lead optimization. J. Comput.-Aided Mol. Des. 2015, 29, 923−936. (40) Segall, M. D. Multi-Parameter Optimization: Identifying high quality compounds with a balance of properties. Curr. Drug Metab. 2012, 18, 1292−1310. (41) Marazziti, D.; Baroni, S.; Borsini, F.; Picchetti, M.; Vatteroni, E.; Falaschi, V.; Catena-Dell’Osso, M. Serotonin Receptors of Type 6 (5HT6): From Neuroscience to Clinical Pharmacology. Curr. Med. Chem. 2013, 20, 371−377. (42) Dobi, K.; Flachner, B.; Pukáncsik, M.; Máthé, E.; Bognár, M.; Szaszkó, M.; Magyar, C.; Hajdú, I.; Lő rincz, Z.; Simon, I.; Fülöp, F.; Cseh, S.; Dormán, G. Combination of Pharmacophore Matching, 2D Similarity Search, and In Vitro Biological Assays in the Selection of Potential 5-HT6 Antagonists from Large Commercial Repositories. Chem. Biol. Drug Des. 2015, 86, 864−880. (43) Whitty, A. Growing Pains in Academic Drug Discovery. Future Med. Chem. 2011, 3, 797−801.

(44) Baell, J. B.; Holloway, G. A. New Subsctructure Filters for Removal of Pan Assay Interference Compounds (PAINS) from Screening Libraries and for Their Exclusion in Bioassays. J. Med. Chem. 2010, 53, 2719−2740. (45) Vass, M.; Jójárt, B.; Bogár, F.; Paragi, G.; Keserű , G. M.; Tarcsay, Á . Dynamics and structural determinants of ligand recognition of the 5-HT6 receptor. J. Comput.-Aided Mol. Des. 2015, 29, 1137−1149. (46) Tarcsay, A.; Paragi, G.; Vass, M.; Jójárt, B.; Bogár, F.; Keserű , G. M. The impact of molecular dynamics sampling on the performance of virtual screening against GPCRs. J. Chem. Inf. Model. 2013, 53, 2990− 2999. (47) Upton, N.; Chuang, T. T.; Hunter, A. J.; Virley, D. J. 5-HT6 Receptor Antagonists as Novel Cognitive Enhancing Agents for Alzheimer’s Disease. Neurotherapeutics 2008, 5, 458−469. (48) Lan, R.; Liu, Q.; Fan, P.; Lin, S.; Fernando, S. R.; McCallion, D.; Pertwee, R.; Makriyannis, A. Structure−Activity Relationships of Pyrazole Derivatives as Cannabinoid Receptor Antagonists. J. Med. Chem. 1999, 42, 769−776.

K


Structure-Based Consensus Scoring Scheme for Selecting Class A

Recommend Documents