Word Search

A new crystal structure fragment-based pharmacophore method for G protein-coupled receptors

Description
A new crystal structure fragment-based pharmacophore method for G protein-coupled receptors
Categories
Published
of 9
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  13  A new crystal structure fragment-based pharmacophore method 4  for G protein-coupled receptors 567  Kimberley Fidom a,1 Q1  , Vignir Isberg a,1 , Alexander S. Hauser a,1 , Stefan Mordalski b,c , Thomas Lehto a , 8  Andrzej J. Bojarski b , David E. Gloriam a, ⇑ 9  a Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences, University of Copenhagen, Denmark 10  b Department of Medicinal Chemistry, Institute of Pharmacology, Polish Academy of Sciences, Poland 11  c Faculty of Biochemistry, Biophysics and Biotechnology, Jagiellonian University, Krakow, Poland 1213 1 5 a r t i c l e i n f o 16  Article history: 17 Received 30 June 201418 Received in revised form 9 September 201419 Accepted 26 September 201420 Available online xxxx21  Keywords: 22 Pharmacophore23 Virtual screening24 Fragment-based drug design25 G protein-coupled receptor26 Drug discovery27 2 8 a b s t r a c t 29 Wehavedeveloped anewmethodfor the buildingof pharmacophores for Gprotein-coupledreceptors, a 30 major drug target family. The method is a combination of the ligand- and target-based pharmacophore 31 methods and founded on the extraction of structural fragments, interacting ligand moiety and receptor 32 residue pairs, fromcrystal structure complexes. We describe the procedure tocollect a library withmore 33 than250fragments covering29residuepositions withinthegenerictransmembranebindingpocket.We 34 describe how the library fragments are recombined and inferred to build pharmacophores for new 35 targets. A validating retrospective virtual screening of histamine H 1  and H 3  receptor pharmacophores 36 yieldedarea-under-the-curves of0.88and0.82,respectively. Thefragment-basedmethodhastheunique 37 advantage that it can be applied to targets for which no (homologous) crystal structures or ligands are 38 known. 47% of the class A G protein-coupled receptors can be targeted with at least four-element phar- 39 macophores. Thefragment libraries canalsobe usedtogrowknownligands or for rotamer refinement of  40 homology models. Researchers can download the complete fragment library or a subset matching their 41 receptor of interest using our new tool in GPCRDB. 42   2014 Published by Elsevier Inc. 43 444546 47  1. Introduction 48  1.1. Traditional pharmacophore modeling methods and 49  chemogenomics 50  Pharmacophores have a widespread use in drug design for 51  ligand identification through virtual screening and subsequently 52  duringtheleadoptimizationphase[1]. Apharmacophorehasbeen 53  defined by IUPAC as ‘‘an ensemble of steric and electronic features 54  thatisnecessarytoensuretheoptimalsupramolecularinteractions 55 with a specific biological target and to trigger (or block) its biolog- 56 ical response’’ [2]. It represents the 3D map of shared elements 57 (charge, aromatic, hydrogen bonding, etc.) across different ligand 58 chemotypes that interact with complementary residues within 59 the biological target. Pharmacophores can be constructed based 60 on the superimposing of ligand 3D conformers. Identification of  61 the bioactive conformation can be challenging, especially for 62 ligandswithmanyrotatablebonds,butcanbeexploredbysynthe- 63 sis of conformationally restricted analogs or docking into a target 64 structure. Thus, this approach relies on the availability of known 65 ligands. Pharmacophores can also be built from the 3D structures 66 of ligand-target complexes, which allow for precise identification 67 and placement of one element for each ligand–protein interaction. 68 This type of model has the advantage that the binding mode of the 69 ligandisknownmakingit possibletocombinethepharmacophore 70 method with structure-based drug design and validation by target 71 proteinmutagenesis. Althoughhomologymodelsof apostructures 72 have been explored [3], the target-based approaches typically rely 73 ontheavailabilityofa3Dstructure,preferablyinthecorrectactiva- 74 tion state. http://dx.doi.org/10.1016/j.ymeth.2014.09.0091046-2023/   2014 Published by Elsevier Inc.  Abbreviations:  7TM, seventransmembrane domain; AFE, aromaticface-to-edge;AFF, aromatic face-to-face; AUC, area under the curve; ECL, extracellular loop;GPCR, G protein-coupled receptor; H 1 , histamine 1 receptor; H 3 , histamine 3receptor; HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; ROC plot,receiver operating characteristic plot; TM, transmembrane helix. ⇑ Corresponding author at: Jagtvej 162, 2100 Copenhagen, Denmark. E-mail addresses:  kimberleyfidom@gmail.com (K. Fidom), vis@sund.ku.dk(V. Isberg), alexshauser@gmail.com (A.S. Hauser), stefanm@if-pan.krakow.pl (S. Mordalski), thomas@leh2.dk (T. Lehto), bojarski@if-pan.krakow.pl (A.J. Bojarski), david.gloriam@sund.ku.dk (D.E. Gloriam). 1 These authors contributed equally. Methods xxx (2014) xxx–xxx Contents lists available at ScienceDirect Methods journal homepage: www.elsevier.com/locate/ymeth YMETH 3498 No. of Pages 9, Model 5G9 October 2014 Please cite this article in press as: K. Fidom et al., Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.09.009  75  Where no ligand or target structures are available, chemoge- 76  nomic techniques allow for the inference of chemical and biologi- 77  cal/pharmacological data from more characterized members of a 78  protein family based on the detection of local similarities within 79  theirbindingpockets[4–8].Chemogenomicshasbeenusedtocon- 80  struct pharmacophores by associating receptor sequence motifs 81  with ligand fragments based on mutagenesis-guided ligand dock- 82  ing in homology models [9]. Today, with a number of available 83  crystallizedligand-targetcomplexes, it isforthefirsttimepossible 84  to collect a high number of interacting ligand moiety – receptor 85  residue fragments and to develop a fragment-based pharmaco- 86  phore method applicable to the majority of G protein-coupled 87  receptors. Herein, we describe the generation of a comprehensive 88  fragment library, a protocol for building crystal structure 89  fragment-based pharmacophores and validation through virtual 90  screening for histamine H 1  and H 3  ligands. 91  1.2. G protein-coupled receptors Q2 92  G protein-coupled receptors (GPCRs) are membrane proteins 93  activatedbymanydiverseligandsincludinglight,ions,neurotrans- 94  mitters, lipids, carbohydrates, nucleotides, amino acids, peptides 95  and proteins [10]. GPCRs have been identified in eukaryotic cells 96  rangingfromyeast tohumans, whichhave  800GPCRgenes mak- 97  ing it the most abundant gene families in the human genome [11]. 98  Given their abundance and wide ligand recognition, it is not sur- 99  prising that GPCRs are involved in most physiological processes, 100  either alone or in concert with other signaling protein families. 101  Theirregulationofpathophysiologyindiversediseaseareas,acces- 102  sibilityat the cell surface and druggable binding sites, have earned 103  them a key role in medicine: more than 30% of the drug on the 104  market target GPCRs [12]. 105  1.2.1. Topology and activation mechanism 106  GPCRs share a bundle of seven transmembrane helices (7TM) 107  spanning the cell membrane and connected by three extra- (ECLs) 108  and intra-cellular loops. Ligands bind on the extracellular side: 109  largepeptidesmainlysuperficiallywithcontactstotheN-terminus 110  and loops and small molecules deeper within the 7TM [13]. The G 111  protein ( a –subunit) binds on the intracellular side to the lower 112  part of the 7TM bundle with additional contacts to ICL2 and prob- 113  ably the C-terminus [14–16]. 114  The GPCR structure dynamically fluctuates between different 115  conformational states stabilized by partial agonists, full agonists, 116  (neutral) antagonists or inverse agonists [17]. The fully activated 117  conformation requires binding of both a high-affinity agonist as 118  well as the G protein: both acting as allosteric modulators for the 119  other [14]. The structural differences between activated and inac- 120  tive GPCRs are large at the binding site of the G protein, TM5 and 121  6 move out by 10–14Å to open the cavity, but more limited in 122  the 7TM ligand binding site, involving a 2Å inward movement of  123  TM5 around 5.46  461 and a slight   1Å upward movement of  124  TM3 [14–16]. The rearrangements of helices have been proposed 125  to be associated with the transitions of a core of conserved hydro- 126  phobic residues stabilizing the inactive state towards the opening 127  of water channels seen in the active state [18]. Furthermore, an 128  ionic lock connecting the lower parts of TM3 and TM6 is broken 129  upon receptor activation [19]. 130  1.2.2. A structurally conserved transmembrane binding cavity with 131  generic residue numbering  132  The structurally conserved 7TM cavity is very amenable to 133  structure-based drug design and so far 59 GPCRs have been 134  druggedwithsmallmolecules[20].TM3,5,6and7withoccasional 135  ligand interactions with TM2 and TM4 mainly defines the binding 136  site, whereas TM1 is (literally) an outsider. The binding site is 137 closed-off at the top by ECL2, which is anchored to TM3 (3.25) 138 via a conserved disulfide bridge. The conserved 7TM scaffold 139 allows for both structural and sequence alignment of receptors to 140 identify the corresponding residues located in the same position. 141 Of these, a well-defined subset of residue positions face inwards 142 the 7TM bundle and are accessible to ligands [8]. Given the con- 143 served structure, making up the placeholder, it is the variation in 144 receptor sequences that compose the unique binding site mosaic 145 determining ligand affinity and selectivity. 146 These residue positions can be compared across receptors by 147 using generic GPCR residue numbers [21–24], the most widely 148 known being the class A Ballesteros–Weinstein scheme [21] 149 recently adjusted for helical bulges and constrictions in the new 150 GPCRDB scheme [24]. Throughout this article, we combine these 151 two schemes as described in [24]. For example 5.42  43 first 152 denotes the helix, then the residue number within the sequence 153 (Ballesteros–Weinstein) and the latter the structural position cor- 154 rected for bulges and constrictions (GPCRDB). 155 2. Generation of a crystal structure fragment library  156  2.1. GPCR-ligand crystal structure complexes 157 There are six classes withinthe GPCR superfamily: Class A, B, C, 158 F (Frizzled) and Taste type 2 receptors. The first structures for the 159 classesB[25,26],C[27,28]andF[29]haveshownthatallsharethe 160 conserved fold with seven transmembrane helices (Section 1.2.1), 161 but they are positioned somewhat differently meaning that (phar- 162 macophore) elements within their respective binding cavities are 163 notreadilycomparable.Forthisreason,afragmentlibrarycanonly 164 be retrieved and used for the largest class, A, for which crystal 165 structures are available for many and diverse receptors. As of July 166 1st 2014, the Protein Data Bank (PDB) [30] contains 62 class A 167 GPCR structures with a (unique) ligand bound within the generic 168 transmembrane binding pocket. This encompasses 20 unique 169 receptors for acetylcholine [31,32], adenosine [33–37], adrenaline 170 [14,38–46], chemokines [47,48], dopamine [49], histamine [50], 171 neurotensin [51], opioids [52–55], proteinase [56], purines 172 [57,58], rhodopsin (retinal) [59–64], serotonin [65,66] and sphin- 173 gosine [67]. It ranges 47 unique ligands, whereas G proteins or 174 mimics are, so far, only available in complex with the  b 2  adrenore- 175 ceptor [14,68] and rhodopsin [15,16]. 176  2.2. Extraction of structural fragments: interacting ligand moiety – 177 receptor residue pairs 178 Crystalstructurecomplexesareidealforrationaldrugdesignas 179 they provide an atom-level view on ligand-target interactions, 180 including the bioactive conformation of the ligand as well as the 181 rotamers of binding residues. Interactions can be viewed directly 182 within the PDB entry (ligand chemical component section), as 2D 183 diagrams [69] or rotatable 3D structures [70]. However, to get 184 completeandcorrect informationtheproteinneeds tobe carefully 185 prepared in a molecular editing software, e.g. Schrödinger or MOE, 186 toaddhydrogenatoms,selectprotonationstates, orientsidechains 187 (e.g. flipping of Asn and Gln amides) and perform a restrained 188 energy minimization [71–73]. To allow cross-target comparisons, 189 there have to be consistent and well-defined criteria for each type 190 of ligand–protein interaction, including the directionality, i.e. 191 hydrogen bond donor vs. acceptor as well as aromatic edge or face 192 orientations. Table A.1 lists the hereinapplied maximumdistances 193 andbondanglesforeachoftheseinteractions.Aliphatichydropho- 194 bic interactions are not included, as they are not as spatially 195 defined and typically a whole cluster of residues is needed to sup- 196 port the ligand moiety. 2  K. Fidom et al./Methods xxx (2014) xxx–xxx YMETH 3498 No. of Pages 9, Model 5G9 October 2014 Please cite this article in press as: K. Fidom et al., Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.09.009  197  We have systematically annotated all class A GPCR structures 198  that have a ligand bound within the generic transmembrane bind- 199  ing pocket (Section 2.1) and extracted structural fragments – pairs 200  of one ligand moiety interacting with one receptor residue. Frag- 201  mentswerenotcollectedfromnon-classAGPCRcrystalstructures, 202  as the backbone structural differences would give too large offset 203  in the placement of pharmacophore elements. Fig. 1 illustrates 204  theextractionoffragmentsfromtheagonistBI-167107– b 2 -adren- 205  ergic receptor complex (PDB: 3SN6). Here, hydrogen bond donor 206  (HBD) fragments are extracted for the interactions between the 207  amine/D3.32, amide NH/S5.43  44, phenol hydroxyl/S5.43  44 208 and amine/N7.39. Hydrogen bond acceptor (HBA) fragments are 209 derived for phenol hydroxyl/S5.46  461 and  b -hydroxyl/N7.39 210 residues. One aromatic face-to-edge (AFE) fragment is extracted 211 for the left benzene/6.52 interaction, whereas the right benzene 212 ring is excluded because its centroid is 4.8Å away from the edge 213 of W3.28, thus exceeding the 4Å limit (Table 1). Finally, a cationic 214 (ICT) fragment is extracted for the amine/D3.32 interaction. In 215 total, seven fragments couldbe derivedfromthis crystal structure. 216 A few special cases need to be considered in the fragment 217 extraction process. First, the same receptor residue can interact 218 with more than one ligand moiety, as seen in Fig. 1 for D3.32 219 and N7.39 (amine and  b -hydroxyl) and S5.43  44 (amide NH 220 and phenol hydroxyl). Similarly, an aromatic ligand moiety often 221 interacts with several aromatic receptor residues and, to facilitate 222 the matching to new target sequences and structures; individual 223 fragments should be extracted for each interaction pair. Hydrogen 224 bond directionalities, of e.g. hydroxyls, can be ambiguous. In some 225 cases a moiety can be recognized as either acceptor or donor after 226 analysis of the intra-receptor hydrogen bonding patterns. How- 227 ever, where the analysis of the binding cavity suggests that a 228 newligand may have one or the other, it is advantageous to incor- 229 porate both (pharmacophore) elements. Finally, where receptor- 230 interacting water molecules can be replaced by a ligand moiety, 231 they too should constitute hydrogen bond acceptor and/or donor 232 fragments. 233  2.3. Composition of the fragment library 234 The collected fragment library comprises in total over 250 frag- 235 ments,eachwithaligandmoietythatrepresentsapharmacophore 236 feature interacting with a specific amino acid and residue position 237 intheconservedtransmembranehelices.Fig.2showsthefragment 238 library, which covers 29 structural positions, each with one to five Fig. 1.  Extraction of fragments (small boxes) from the BI-167107 (green) –  b 2 adrenergic receptor (blue) complex crystal structure (PDB: 3SN6). The annotationcriteria for ligand-receptor interactions are listed in Table A.1. The figure was produced in PyMOL  [94]. The generic residue numbers in Figs. 1–3 and 5 are denoted using the Ballesteros–Weinstein [21] scheme complemented by theGPCRDB scheme [24], which corrects for bulges and constrictions. Fig. 2.  Thefragmentlibrary’scompositionshowing,foreachresidueposition,theincludedresidues(top),associatedpharmacophoreelementsandtheirnumberoffragments(middle)andamatchingtothesequenceof allclassAreceptors(bottom). InthematchingtoclassAsequences, similarresiduesareonlythosethatcanorienttheinteractingatoms in the same place; e.g. Ser and Thr for hydrogen bonds; His, Phe, Trp and Tyr for an aromatic edge/face-to-face interactions; and Phe for Tyr face-to-edge interactions. K. Fidom et al./Methods xxx (2014) xxx–xxx  3 YMETH 3498 No. of Pages 9, Model 5G9 October 2014 Please cite this article in press as: K. Fidom et al., Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.09.009  239  (for 6.58) residue types (amino acids). Each residue is associated 240  with one or more pharmacophore elements. D3.32 has the largest 241  number of fragments (60) and these are distributed over the com- 242  bined HBD-ICT (36), HBD (20) and ICT (4) element types. Fig. 3 243  shows the location of the residue positions within the binding site 244  andtheircoveragebythefragmentlibrary(color-coding).Thefrag- 245  ment library covers most residue positions in the binding cavity 246  and all helices except TM1 and TM4, which rarely participate in 247  ligand binding (Section 1.2.1). Fig. 4 plots the distribution of the 248  fragments over the element types, with the most frequent being 249  HBD (96), HBA (61), aromatic (44) and combined HBD-ICT (37). 250  In assessingthe coverageof the libraryit is interestingto see to 251  which extent it matches the potential targets – all class A GPCRs. 252  The bottom part of  Fig. 2 shows the class A consensus sequence 253  and the number and percentages of identical and similar residues, 254  respectively. ‘‘Similar’’residuesweredefinedasthosethatmediate 255  the same ligand–protein interaction. This is Ser and Thr for hydro- 256  gen bonds; His, Phe, Trp and Tyr for an aromatic edge/face-to-face 257  interactions;andPheforTyrface-to-edgeinteractions. Onaverage, 258  a residue position contains amino acid(s) that match similar resi- 259  duesin24%of theclassAGPCRs. If requiringatleastfourelements 260  when building a pharmacophore, 137 (47%) of the class A GPCRs 261  can be targeted. This number will increase in the future, as addi- 262  tional receptor structures are determined. 263  Our fragmentlibrarycan becomparedto that of Klabundeet al. 264  [9]. Our study includes a significantly larger number of receptor 265  structures, 62 compared to 13, and fragments: 250 instead of 43. 266  The number of pharmacophore elements cannot be compared to 267  Klabunde et al. due to differences in the hydrophobic interactions, 268  which included non-aromatic contacts and listed multi-residue 269  motif rather than individual amino acids. As a further note, 10 of  270  thereceptorstructuresinthestudybyKlabundeetal.werehomol- 271  ogy models, whereas this analysis includes only crystal structures 272  and should therefore have a higher precision in the placement of  273  the pharmacophore features. 274  3. Pharmacophore model building  275  The construction of pharmacophores from the fragment library 276  consists of five steps: selection of a crystal structure template, 277  matchingandsuperimposingoffragments,selectionofrepresenta- 278  tivefragments,placementofpharmacophoreelements,refinement 279  and validation. 280  3.1. Selection of a crystal structure template for the 7TM binding cavity 281 backbone 282 Thecrystalstructuretemplatewillmainlyservetosuperimpose 283 the fragments to the 7TM backbone in building of the pharmaco- 284 phore,butcanoptionallyalsobeusedtogenerateatargetreceptor 285 structure model, and the considerations are the same for both 286 approaches. In selecting the template, the most important factor 287 is the activationstate of the receptor, as it affects the relativeposi- 288 tionsofthetransmembranehelices[18].Thus,modelsintendedfor 289 identification of agonists and antagonists should apply active and 290 inactive templates, respectively. Next, the degree of receptor 291 homology, i.e. sequence similarity, should be assessed although 292 this is not as crucial as in homology modeling, as only the back- 293 bone, not sidechains, need to be conserved. When several compa- 294 rable templates are available, ligand structural similarity and 295 structure resolution should be weighted. The overall template 296 selection process has been facilitated by the implementation of a 297 crystal structure browser in GPCRDB [74] that allows for swift fil- 298 teringof allstructures, whichareorderedaftersequencesimilarity 299 to the target of interest. 300  3.2. Matching and superimposing the fragments to the target of  301 interest  302 Herein, we have made available a fully automated procedure in 303 GPCRDBtosuperimposefragmentsontoareceptorstructure(user- 304 provided pdb file). This selects fragments by matching library res- 305 idue positions and amino acids, listed in Fig. 2, with a sequence 306 alignment to the target. As fragments are re-combined across the 307 entire class A (>292 receptors in human), this puts stringent 308 demands on the sequence alignments, which have been gapped 309 to compensate for helix bulges and constrictions that otherwise 310 offset the generic residue positions and numbering [24]. As dis- 311 cussedinSection2.3,fragmentscanalsobeinferredbetween‘‘sim- 312 ilar’’ residues, i.e. those that can mediate the same ligand 313 interaction. For each matched position, the fragment residues are 314 superimposed onto the receptor backbone (C-alpha, C and N 315 atoms) and this directs the residue side-chains and the interacting 316 ligand moieties into the binding site. The results can be down- 317 loaded as a single pdb file containing the receptor structure and Fig. 3.  The helix box diagram depicts the seven transmembrane helices from thetop. The number of fragments for each position is illustrated with color-coding:yellow: 1–4, light green: 5–9 and dark green: 10 or more. The GPCRDB residuenumbering is used for residue numbering [24]. Positions 7.31 and 7.32 are locatedvery high are not included in this figure. The helix box diagram was produced inGPCRDB [74] and residue numbers added in InkScape. Fig. 4.  Pharmacophore library element types with their total numbers and relativepercentages shown in the inner and outer rings, respectively.4  K. Fidom et al./Methods xxx (2014) xxx–xxx YMETH 3498 No. of Pages 9, Model 5G9 October 2014 Please cite this article in press as: K. Fidom et al., Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.09.009  318  all superimposed matching fragments. To ensure compatibility 319  with all molecular software, the titles of fragments contain the 320  annotated data; residue position, amino acid, receptor gene name, 321  PDBid,pharmacophoreelementtype,waterinteraction(optional); 322  separated by an underscore. It is also possible to download the 323  complete library for use in locally installed molecular software. 324  3.3. Selection of representative fragments and placement of  325  pharmacophore elements 326  Most interactions are represented by multiple fragments 327  (Fig. 2). Fig. 5A shows the 36 fragments for the double HBD–ICT 328  interactions with D3.32. In the vast majority of fragments, the 329  Asp rotamers are very similar and the two corresponding ICT and 330  HBD moieties, respectively, make up rather well defined clusters. 331  Thus, the pharmacophore feature can be placed in the middle 332  and its size (matching tolerance) adjusted to cover most of the 333  density or all fragments. Fig. 5B shows the fragments for F/Y3.33 334  and F6.52 that have close or overlapping aromatic moieties. This 335  applies to several positions, often as a result of the same ligand 336  moiety being associated with several conserved receptor residues. 337  Tobe representative, it can benecessarytocopy or moveonefrag- 338  ment to the center. Instead of trying to cover all fragments, an 339  alternative strategy is to select a subset of fragments that were 340  derived from the receptors with the highest homology/sequence 341  similarity. Furthermore, Fig. 5C shows that S5.46  461 contains 342  a clear separation of agonist- and antagonist bound rotamers. This 343  is very interesting as it allows for the construction of different 344  pharmacophores with selectivity towards the desired activity. 345  The use of vectors for hydrogen bonding interactions is optional, 346  but recommended as they typically improve the specificity in the 347  virtual screening. Finally, to avoid ligand-receptor clashes, exclu- 348  sion volumes should be added on proximal (e.g. 6Å) backbone 349  atoms. 350  3.4. Refinement and extension of the pharmacophore 351  When high affinity/potency ligands are already known for the 352  target, they can be used for example to adjust the positions of  353  pharmacophore elements to better separate between agonist and 354  antagonist ligands. Furthermore, tested inactive analogs with 355  unique substituents can be used to place additional exclusion vol- 356  umes. Optionally, also new elements can be added to the pharma- 357  cophore if ligand SAR and/or mutagenesis have identified moieties 358  that are crucial to achieve high potency or selectivity. However, as 359  these moieties are not included in the fragment library, great care 360  needs tobetakenintheir placement in3D. Thebioactive(binding) 361  conformation may be identified from conformationally restricted 362 ligandsand/ordockingintoreceptorstructuremodels,whichcould 363 be retrieved from GPCRDB [74] or easily built online with the 364 GPCRM [75], GoMoDo [76] or GPCR-SSFE [77] web servers. 365 Whereas the ligand- and target structure-based techniques offer 366 to bring in complementary useful information, extensive editing 367 ofthefragment-basedpharmacophoresmayriskoverfittingreduc- 368 ing the ability to identify new chemotypes. 369 Mutagenesis data can be found in GPCRDB, whereas reference 370 GPCR ligands can be retrieved from several public databases. The 371 IUPHAR guide to pharmacology database [78] offers expert anno- 372 tated GPCR ligands with potency values determined in at least 373 two independent laboratories. The NIMH Psychoactive Drug 374 Screening Program  K  i  database contains a collection of   K  i  values 375 for psychoactive targets including many GPCRs. Much of this data 376 was generated in-house in a standardized fashion, whereas other 377 data is derived from the literature and needs to be validated by 378 reading the source publication. Finally, ChEMBL contains more 379 than145,000GPCRligandsfrompublications,patentsandacquired 380 databases [79]. However, this data varies significantly in quality 381 and it is therefore always necessary to filter this data, e.g. for high 382 assay confidence (score 8 or 9), dose–response data ( K  i/d , XC50, 383 pXC50) and sub-micromolar ligands. The demands on size and 384 quality are tightly linked to the use of the reference ligands. In 385 the pharmacophore refinement phase, it can be more effective to 386 use a focused set of structurally diverse ligands instead of a large 387 structurallyredundantcollection.However,inpharmacophoreval- 388 idation by retrospective virtual screening (Section 3.5) and the 389 plotting of Receiver Operating Characteristic (ROC) plots (Fig. 7), 390 typically a larger reference database which can make the analysis 391 less sensitive to (limited) errors. 392  3.5. Validation of the pharmacophore model (retrospective virtual 393 screening) 394 Before application, pharmacophores should be evaluated 395 through a retrospective virtual screening that assesses the ability 396 to selectively retrieve known ligands from a database that also 397 contains a large number of decoys – presumed inactive com- 398 pounds. To get a fair comparison relevant for the coming prospec- 399 tive screening, the decoys should be property-matched based on 400 reference ligand and/or pharmacophore properties (e.g. number 401 of aromatic rings and charges). As always, all compounds need to 402 be prepared to desalt, generate charge states, add hydrogen atoms 403 and generate tautomers, stereoisomers and 3D conformations. The 404 assessment should focus on the top percentages to simulate a pro- 405 spective screening, for which only a limited number of top-scoring 406 compounds would normally be purchased (after applying addi- 407 tional filters and clustering). Fig. 5.  (A)TheD3.32carboxylicacidcansimultaneouslyinteractwithtwofragmentclusters(grayandgreen)andhasdoublespatiallydistinctICTandHBDelements.(B)TheF/Y3.33(gray)andF/H6.52(green)fragmentsinteractwithcloseoroverlappingaromaticmoieties;oftenidenticalderivedthesamecrystalstructure;bestrepresentedbyonecommon (manually placed) central moiety. (C) Two different S5.46  461 rotamers, left and right, exist in active (gray) and inactive (green) state structures, respectively.These correspondto thetwomost commonrotamers inthe standardMaestro rotamer libraryandhave a frequencyof 42%and26%, respectively. Thefigure was producedinMaestro [95]. K. Fidom et al./Methods xxx (2014) xxx–xxx  5 YMETH 3498 No. of Pages 9, Model 5G9 October 2014 Please cite this article in press as: K. Fidom et al., Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.09.009
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks