dynamic undocking and the quasi-bound state as tools for...
TRANSCRIPT
![Page 1: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/1.jpg)
This is a repository copy of Dynamic undocking and the quasi-bound state as tools for drug discovery.
White Rose Research Online URL for this paper:http://eprints.whiterose.ac.uk/110037/
Version: Accepted Version
Article:
Sergio, Ruiz-Carmona, Schmidtke, Peter, Luque, F. Javier et al. (7 more authors) (2016) Dynamic undocking and the quasi-bound state as tools for drug discovery. Nature Chemistry. ISSN 1755-4349
https://doi.org/10.1038/nchem.2660
[email protected]://eprints.whiterose.ac.uk/
Reuse
Items deposited in White Rose Research Online are protected by copyright, with all rights reserved unless indicated otherwise. They may be downloaded and/or printed for private study, or other acts as permitted by national copyright laws. The publisher or other rights holders may allow further reproduction and re-use of the full text version. This is indicated by the licence information on the White Rose Research Online record for the item.
Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by emailing [email protected] including the URL of the record and the reason for the withdrawal request.
![Page 2: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/2.jpg)
1
DynamicUndockingandtheQuasi-BoundStateasToolsforDrugDesign1
2
SergioRuiz-Carmona,1PeterSchmidtke,2F.JavierLuque,1LisaBaker,3Natalia3
Matassova,3BenDavis,3StephenRoughley,3JamesMurray,3RodHubbard,3,4Xavier4
Barril1,5,*5
6
1 Institut de Biomedicina de la Universitat de Barcelona (IBUB) and Facultat de 7
Farmàcia, Universitat de Barcelona, Av. Joan XXIII s/n, 08028 Barcelona, Spain. 8
2 Discngine, 33 rue du Fauburg Saint-Antoine, 75011 Paris, France 9
3 Vernalis (R&D) Ltd, Granta Park, Cambridge, CB21 6GB, UK 10
4 YSBL, University of York, Heslington, York, YO10 5DD, UK 11
5 Catalan Institution for Research and Advanced Studies (ICREA), Passeig Lluís 12
Companys 23, 08010 Barcelona, Spain. 13
14*Sendcorrespondenceto:[email protected] 15
![Page 3: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/3.jpg)
2
There is a pressing need for new technologies that improve the efficacy and16
efficiency of drug discovery. Structure-based methods have contributed towards17
this goal but they focus on predicting the binding affinity of protein–ligand18
complexes,which is notoriously difficult. We adopt an alternative approach that19
evaluates structural, rather than thermodynamic, stability. Noting that bioactive20
moleculespresentastaticbindingmode,wedevisedDynamicUndocking(DUck),a21
fastcomputationalmethodtocalculatetheworknecessarytoreachaquasi-bound22
state,wheretheligandhasjustbrokenthemostimportantnativecontactwiththe23
receptor. This non-equilibrium property is surprisingly effective in virtual24
screening because true ligands form more resilient interactions than decoys.25
Notably,DUck isorthogonal todockingandother ‘thermodynamic’methods.We26
demonstrate the potential of the docking–undocking combination in a fragment27
screeningagainstthemolecularchaperoneandoncologytargetHsp90, forwhich28
weobtainnovelchemotypesandahitrateapproaching40%.29
30
31
32
33
34
35
36
37
38
39
40
![Page 4: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/4.jpg)
3
Structuralstabilityisafundamentalpropertyofprotein–ligandcomplexes.Though41
casesofdualbindingmodeshavebeenreported,1,2theyaregenerallynotdynamic,42
orinvolvepredominantlyhydrophobicinteractions,3whichlackdirectionalityand43
donotimposestrictgeometricconstraints.4Bycontrast,hydrogenbondsareideal44
to provide structural stability because they have sharp distance and angular45
dependencies.4Theircontributiontothefreeenergyofbinding(ΔGbind)isvariable46
butcanbesubstantial.5Importantly,theyoftenactasanchoringpointsinprotein–47
ligand complexes, providing the minimal binding unit through one or a few48
hydrogen bonds as demonstrated for fragment-sized ligands.6,7 We have49
previouslyshownthatcertainhydrogenbondspresentstrongoppositiontosmall50
structuraldistortionsandcanactaskinetic trapsbecause the local environment51
hinders the transition from a direct hydrogen bond to a water-bridged52
interaction.8Asanearlyunbindingevent,ruptureoftheso-calledwater-shielded53
hydrogen bonds can influence the whole dissociation process.8,9 Taken together,54
these observations suggest that hydrogen bonds are the main determinants of55
structural stability, and lead us to postulate that their resilience should provide56
informationabout thebindingpotentialofcandidate ligands.Thus,wesetout to57
investigatewhethertheworkrequiredtodisruptintermolecularhydrogenbonds58
canbeusedtopredictligandbinding.59
60
We will introduce DUck, a simplified computational procedure to calculate the61
work needed to break a key native contact, reaching a quasi-bound state (WQB).62
Then, we will show that active compounds are structurally stable and present63
higher WQB values than inactive ones. Finally, we demonstrate the use of this64
![Page 5: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/5.jpg)
4
property in virtual screening (VS) applications, showing thatDUck complements65
thethermodynamicperspectiveofferedbyexistingmethods.66
67
68
ResultsandDiscussion69
70
Simplifiedsimulationoftheearlydissociationstage71
To assess the hypothesis, we have devised Dynamic Undocking (DUck)72
simulations,where a key intermolecularhydrogenbond ispulled froman initial73
distanceof2.5Å(closecontact)to5.0Å(brokencontact).Inordertofocusonjust74
onespecifichydrogenbond,weusemodelreceptorscomprisingonlytheprotein75
residues that are within 6 Å of the given hydrogen bond (Figure 1A). The work76
necessarytocarryoutthesteeringprocessismonitored,andwedefinethequasi-77
bound (QB) state as the point along the simulation where the work profile78
presents the highest value. WQB is the work necessary to depart from the ideal79
hydrogenbondconfigurationandreachtheQBstate(Figure1B).Notably,thisisa80
non-equilibriumproperty,andthereisnoreasonwhyitshouldcorrelatewithany81
measurement of binding affinity. What is more, as the unbound state is not82
considered, WQB cannot inform about the binding free energy. Instead, this83
magnitude solely indicates if the interaction under investigation gives rise to a84
(local) minimum in the free energy landscape and estimates the depth of said85
minimum(SupplementaryFigure1).86
87
RelationshipbetweenWQBandbindingaffinity88
![Page 6: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/6.jpg)
5
Asan initialproofofconcept,weapplyDUcktoasetof41fragment-like ligands89
(<300Da)ofthecyclindependentkinase2(CDK2)withknownbindingmodeand90
halfmaximalinhibitoryconcentration(IC50)values.Thehingeregionofallkinases91
isahotspotforbinding,wheretheproteinbackboneoffersprivilegedhydrogen-92
bonding opportunities.10 For CDK2, the central hydrogen-bond donor (NH of93
Leu83)isthemostconservedinteractionsiteandwasusedtodefinethereaction94
coordinate. WQB presents only a weak correlation with binding affinity95
(Supplementary Figure 2), but the distribution of WQB values is clearly skewed96
(Figure2AandSupplementaryFigure3).Thus,65%ofweakbinders(IC50>1µM)97
presentWQBvaluesbelow6kcal/mol,whileallstrongbinders(IC50<1µM)pass98
this threshold. Ligand3FZ1,11 is the clear exception as it presents an almost flat99
dissociationprofile(WQB=0.12kcal/mol).Thisisexplainedbyanunsuitablylong100
(3.4Å) interactionwith thehinge region, involving amethoxy group,which is a101
poor hydrogen bond acceptor.4 Instead, this unusual ligand forms two charge-102
reinforcedhydrogenbondswithLys33andAsn132,fromwhichitdrawsstructural103
stability (Supplementary Figure 4). This shows that some ligands can use104
alternativeoradditional interactionpoints toattain structural stability, inwhich105
case,DUckcalculations(ascurrentlyimplemented)mayunderestimatethecostof106
breakingthenativecontacts.107
108
TofurtherexaminethesurprisingrelationshipbetweenbindingaffinityandWQB,109
weuse thebromodomainandextra-terminal (BET)BRD4-BD1asadditional test110
system.Theside-chainNofAsn140isawell-knownpharmacophoricpointofthis111
epigenetictarget,12anddefinesthekeyintermolecularhydrogenbond.Again,we112
observethesametrend, i.e.higherWQB formorepotent ligands,butwitha large113
![Page 7: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/7.jpg)
6
dispersion thatblurs correlation (SupplementaryFigures3 and5). Interestingly,114
the lowest WQB values (0, 1.1 and 1.7 kcal/mol) correspond to three kinase115
inhibitorswithoff-targetactivity for theBRD4-BD1.13Thus,achievingpotency in116
theabsenceofarobustanchoringinteractionispossible,butrare,whichsuggests117
thatitisanineffectivestrategy.118
119
DUckisveryeffectiveinvirtualscreening120
Wethenassesswhether theapproachcanbeused invirtual ligandscreeningby121
testingtheabilityofDUcktodistinguishtrueCDK2ligandsfromasetofcarefully122
selected decoys14 for which we had generated binding modes by docking. The123
distribution of WQB is strikingly different from the active set, with 61% of124
moleculespresentingvaluesbelow2kcal/moland49%below1kcal/mol(Figure125
2A).Thisindicatesthat,inspiteofformingthekeyhydrogenbond,thisinteraction126
is labile for most of the docking decoys, which would translate to an unstable127
bindingmode.We therefore propose that WQB can distinguish true ligands from128
inactivemolecules,asshowninthereceiveroperatingcharacteristics(ROC)curves129
(Figure2B).Todemonstratethewiderapplicabilityofthemethod,weconducted130
similar experiments with the adenosine A2A receptor (AA2R) and Trypsin, as131
representatives of G protein-coupled receptors (GPCR) and serine proteases,132
respectively (Figure 2B). Together with kinases (such as CDK2) these protein133
familiesincludealargepartofthecurrentandinvestigationaldrugtargets.15The134
key hydrogen bonds tracked by the DUck simulations involve the side-chain135
carbonyl of Asn253, in the case of AA2R, and the carboxylic acid of Asp189, for136
Trypsin.AsshowninFigure2B,theresultsforthesesystemsareevenbetterthan137
for CDK2, demonstrating that DUck is surprisingly effective in virtual screening.138
![Page 8: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/8.jpg)
7
Importantly, the performance improves consistently as sampling increases, but139
good enrichments can be obtained with as little as 2 DUck runs per ligand140
(SupplementaryFigure6).141
142
DUckisorthogonaltoexistingmethods143
TheseresultspositionDUckasanewmethodforvirtualscreening.But,asitaims144
topredictapropertythatisfundamentallydifferentfromthermodynamicstability,145
weinvestigateitscomplementaritywithmoleculardocking,amethodwithalong146
and successful history of application in virtual screening.16,17 Using the rDock147
software,18 we find that docking scores have no correlation with WQB, and good148
dockingscorersarenearlyas likelytopresenta lowresistancetodissociationas149
the rest of the decoys (Figures 2C, 2D and Supplementary Figure 7). As such,150
molecular docking and dynamic undocking can be considered orthogonal (i.e.151
perfectlycomplementary)andtheintersectionbetweenbothtechniquesdefinesa152
region highly enriched in true ligands. We have also performed extensive153
calculations with other virtual screening tools (Glide docking, MMPBSA and154
MMGBSAre-scoring).Theresults,summarisedinSupplementaryFigures8and9,155
confirmthatDUckiscomplementarytoallofthem.Infact,asweobtainlowWQB156
valuesformanydecoyswithgoodscoresbyallothermethods,DUckpost-filtering157
delivers several fold improvementevenwhenapplied toa consensus listby two158
independent ‘thermodynamic’ approaches (Figures 2E, 2F and Supplementary159
Figure10).Theseresults support the idea that structural stabilityof thebinding160
mode,justlikegoodchemicalcomplementarity,isanecessary–butnotsufficient–161
condition for binding. By imposing both conditions simultaneously, we can162
multiply theeffectivenessof structure-basedvirtual screening.At thesame time,163
![Page 9: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/9.jpg)
8
usingWQBasapost-docking filtermeans thatonly thebest-scoringsubsetof the164
virtual chemical collection needs to be reassessed by DUck simulations, thus165
improvingcomputationalefficiency.166
167
Fragmentdiscoverywithintandemdocking-undockingcalculations.168
To demonstrate the power of the docking-undocking combination, we have169
applied the method prospectively for the identification of small molecules that170
bindthemolecularchaperone,HeatShockProtein90KDa(Hsp90).Thisoncology171
target has been a test-bed and paradigm in fragment and structure-based drug172
design.19WithhundredsofHsp90-ligandcomplexesdepositedintheProteinData173
Bank (PDB), discovery of novel chemotypes is very challenging. We focused on174
fragment-like molecules, as this may be the most efficient way to discover new175
leadsandtogeneratescaffold-hopingideas.20,21Acollectionof280000fragment-176
sizedmoleculeswasdockedtotheATPbindingsiteofHsp90.Adiversesetof139177
molecules from the best 450 (top 0.16%) was then selected and each one was178
subjectedto100DUckrunstoobtainfullyconvergedWQBvalues(notethatfewer179
DUck runs would have given similar results (Supplementary Figure 11)). The180
distributionofWQBvalues(Figure3A,SupplementaryFigure12)showsthateven181
attheupperlimitofthedockingscoredistributionalargeproportionofputative182
ligandspresentlowresistancetodissociation,with32%,50%and80%presenting183
WQBbelow3,4and6kcal/mol,respectively.Wepurchasedallthemoleculesfrom184
the high stability set (WQB > 6 kcal/mol) that were available (n=21). They were185
testedusing threedifferent ligand-observedNuclearMagneticResonance (NMR)186
experiments, in the absence or presence of a known competitor to confirm that187
fragmenthitsbindat the targetsite.19Eightoutof the21molecules(38%)were188
![Page 10: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/10.jpg)
9
confirmed as true hits (Table 1). Crucially, for the same system and screening189
method,thehitrateobtainedwithageneralfragmentscreeninglibraryis4.4%.22190
Therefore, theDUck-basedvirtualscreening increasestheefficiencybynearlyan191
orderofmagnitude.This is similar to optimal virtual fragment screening results192
reportedforothersystems.23.InordertobetterassessthecontributionofDUckto193
thesuccessrate,wealsopurchasedandcollecteddatafor15moleculesfromthe194
mediumstabilityset(WQBbetween3and6kcal/mol)and11fromthelowstability195
set (WQB < 3 kcal/mol). Only one molecule from these sets was a hit and,196
importantly,itsWQBvalueisveryclosetotheupperthreshold(5.6kcal/mol).This197
confirmsthatDUckfalsenegatives(i.e.activemoleculeswithlowWQB)arerare,an198
ideal property for a screening method. Hit rates for the three categories are199
summarizedinFigure3B.200
201
Toassessthevalueofthehitsasstartingpoints,wehavecomparedtheirchemical202
structures toexistingHsp90 ligands, finding lowsimilarity inall cases (Table1).203
Bindingmodedeterminationandanalysisofthemaininteractionsthatdefinethe204
chemical scaffold offers a more precise assessment of their novelty. Crystal205
structures for 3 of the fragment hits were determined by X-ray crystallography206
(Figure 4 and Supplementary Figure 13). This confirmed that the docking pose207
used as starting position for the DUck experiments was correct, particularly208
regarding the key interaction that was being monitored (side-chain of Asp93).209
Compound1isthemostpotentfragmenthit(dissociationconstantKD=77µM)and210
has a ligand efficiency (LE) of 0.33 kcal/mol per non-hydrogen atom, similar to211
other Hsp90 fragment hits that have been evolved into very efficient lead212
compounds.24Many2-aminopyrimidineshavebeendescribedasHsp90ligands,19213
![Page 11: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/11.jpg)
10
confirmingthepotentialofthefragmenthit,buttherelativelackofnoveltywould214
adviseagainstusing this fragmentas startingpointat this stage.Compound2 is215
lesspotent(KD=320µM)butequallyefficient(LE=0.32)byvirtueofhavingfewer216
atoms.Inthiscase,thekeyinteractionwithAsp93ismediatedbyanaminothiazole217
moiety, which is unprecedented and would constitute a good starting point to218
developnewchemicalentities.Compound3(KD=700µM;LE=0.25)belongstothe219
well-known family of resorcinol inhibitors, which includes the clinical candidate220
NVP-AUY922,19butprovidesaninterestingexampleofscaffoldhopping,wherethe221
oximeactsabioisostericreplacementofthefive-memberedringsincludedascore222
scaffold in the patents. Compounds 4, 5 and 6 also represent completely novel223
startingpoints,astheirscaffoldisuniqueamongstHsp90inhibitors.Thebinding224
mode could not be confirmed experimentally, but is likely correct because two225
independent methods deemed the molecules active based on the predicted226
geometry(Theirpredictedbindingmodescanbe found inSupplementaryFigure227
14).228
229
Conclusions230
231
Insummary,wehavedemonstratedthattheconceptofstructuralstabilitycanbe232
usedveryeffectivelyinstructure-baseddrugdesign,complementingthestandard233
focus on binding free energy. Hydrogen-bonding groups in the active site are234
privileged structures to fix the ligand in place, particularly when they act as235
bindinghotspotsandcanformwater-shieldedhydrogenbonds.8Theworkneeded236
tobreaksuchinteractions(WQB)isveryusefultodetecttrueligandseventhoughit237
is a non-equilibrium property that is not expected to correlate withΔGbind. This238
![Page 12: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/12.jpg)
11
intriguing fact may reflect the nature of proteins, which have been designed to239
bind their natural ligands not only with high affinity and selectivity, but also240
forming structurally stable complexes. Thus, it will be important to test the241
approach on other types of supramolecular assemblies. Dynamic Undocking242
(DUck), aparticular implementationof steeredmoleculardynamics, allowsus to243
calculateWQB in a very efficientmanner.DUck canbe used in combinationwith244
existing‘thermodynamic’approachestomultiplytheireffectiveness.Thedocking-245
undocking combination has proven particularly useful for virtual fragment246
screening,deliveringnovel,diverseandsuitablestartingpointswithahitrateof247
38%.Atpresent,wefocusonasinglekeyhydrogenbondtoestimateWQB,which248
requires previous knowledge and has a critical impact on the outcome. Future249
investigations should address the extension of the method to multiple sites and250
other interaction types to improve performance and avoid reliance on extrinsic251
decisions.DUck inherits the intrinsic limitationsofstructure-basedmethods(e.g.252
protein flexibility, quality of the force-field) and may have some of its own (e.g.253
longrangeeffects,steeringconditions).Furthertestswillrevealitstruepotential,254
butconsideringthatitisorthogonaltoexistingmethodsandcomputationallyvery255
efficient, we expect that it will be rapidly adopted by the structure-based drug256
design community and adapted to other biotechnological applications involving257
non-covalentcomplexes.258
259
260
![Page 13: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/13.jpg)
12
METHODS261
262
DynamicUndocking263
Dynamic Undocking (DUck) is a particular type of Steered Molecular Dynamics264
(SMD),25whereweforcetheruptureofanintermolecularhydrogenbondformed265
between a pre-defined interaction point in the receptor and a complementary266
atom in the ligand.Additionally,weuse amodel receptor that includes only the267
minimalsubsetoftheproteinnecessarytopreservethelocalenvironmentaround268
the hydrogen bond that is being monitored. This transformation minimizes the269
influenceofperipheralinteractions,thussimplifyingthedissociationpathwayand270
facilitatingconvergence(SupplementaryFigure15).Asanaddedbonus,itspeeds271
upthecalculationsbyafactorof5(SupplementaryTable2).Thefirstandessential272
stepistoidentifyanatomofreferenceintheprotein,whichmustformahydrogen273
bondwithall(ormost)knownligands.Forwell-knownsystems,liketheonesused274
here, it can be identified from a structural superimposition of all the available275
protein-ligand complexes. On novel binding sites, it may be identified with a276
quantitative hot spot identification method.26. Then, the model receptor is277
generated from a representative 3D structure of the protein by selecting all278
residueswithatleastoneatomwithin6Åoftheatomofreference.Theselectionis279
visuallyinspectedand,ifneeded,additionalresiduesthataredeemednecessaryto280
preservethelocalenvironmentareincludedintheselection.Unselectedresidues281
are eliminated and truncated side chains are acetylated or N-methylated, as282
needed. Interstitial water molecules, if present, are preserved. The PDB codes,283
reference interactionpointsand the listofproteinresiduesandwatermolecules284
for each system are listed in Supplementary Table 3. Given the model receptor285
![Page 14: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/14.jpg)
13
(protein chunk) and a set of ligands properly oriented (docking poses or286
superimposed X-ray geometries), a MOE27 SVL script developed in house287
automaticallyperformsthefollowingsteps:1)CalculatesAM1-BCCchargesforthe288
ligand.282)Assignsparm@Frosst29atomtypesandnon-bondedparameterstothe289
ligand. 3) Identifies the ligand atom that is hydrogen-bonded to the protein’s290
referenceatom(basedondistanceandtype).4)Writesinputandexecutionfilesto291
carryout theMDsimulationswithAMBER30.5)CallsAMBER’s tLeaptogenerate292
valid topology and coordinate files for each individual receptor-ligand complex.293
For the protein, theAMBER force field 99SB is used. Each system is placed in a294
cuboidboxspanningat least12Åmorethanthefurthestatomineachdirection.295
The box is then filled with TIP3 water molecules to create periodic boundary296
conditions.Whenneeded,Na+orCl- ionsareaddedtoforcetheneutralityofthe297
wholesystem.MDsimulationconditions(wherenon-default)areasfollows:1)At298
allstages,harmonicrestraintswithaforceconstantof1kcal/mol·Å2areplacedon299
all non-hydrogen atoms of the receptor to prevent structural changes. 2)300
Spontaneousruptureofthekeyhydrogenbondduringnon-steeredsimulationsis301
preventedwithagradual restraint fordistancesbeyond3Å (parabolicwithk=1302
kcal/mol·Å2between3Åand4Åandlinearwithk=10kcal/mol·Åbeyond4Å).3)303
AllequilibrationandsimulationstepswererunusingaLangevinthermostatwitha304
collisionfrequencyof4ps-1andthecutofffornon-bondedinteractionswassetto305
9Å. 4) Bonds involving hydrogen are constrained using SHAKE.31 In order to306
equilibrate the system the following steps are executed: 1)Energyminimization307
for1000cycles.2)Assignmentofrandomvelocitiesat100Kandgradualwarming308
to300Kfor400psintheNVTensemble.3)Equilibrationofthesystemfor1nsin309
theNPTensemble (1atm,300K).At this stage, the first SMDsimulations canbe310
![Page 15: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/15.jpg)
14
executed. We run two SMDs from the same restart file, but at different311
temperatures(300Kand325K)toensurethatthetrajectoriesproceeddifferently.312
TheSMDlasts500ps,duringwhichtimethedistancebetweenthekeyhydrogen313
bonds is steered from2.5Å to5.0Å (constant velocity of 5Å/ns)with a spring314
constantof50kcal/mol·Å2.Wehave tested slowervelocities and the results are315
essentially unchanged (SupplementaryFigure16). The spring constant had little316
influenceandonalimitedtestsetweobtainedessentiallyidenticalresultsinthe317
range k=10 kcal/mol·Å2 to k=1000 kcal/mol·Å2. We have also investigated the318
importance of the specific reaction coordinate by using the closest contact319
between CDK2 Leu83:O and the ligand (instead of Leu83:N). The WQB values320
obtainedwiththesedifferentatomsofreference(locatedonly3Åapart)presenta321
highcorrelation(r2=0.75;SupplementaryFigure17).Bycontrast,whentheatoms322
of reference involve completely different part of the ligand, the results are323
uncorrelated(SupplementaryFigure18). Togeneratediversestartingpoints for324
SMDtrajectories,weperform1nsunbiasedMDsimulationandrepeattheprocess325
as many times as desired (e.g. 50ns unbiased MD simulations are needed to326
execute 100 SMD trajectories). All simulations were performed with Amber 12327
adapted for running inGPUsandexecutedeither in-housewithNVIDIAGeForce328
TITAN X GPUs or at the Barcelona Supercomputing Center using NVIDIA Tesla329
M2090 GPUs. The simulations took 24 minutes (unbiased MD) or 30 minutes330
(SMD)ofwallclocktimepernanosecond(averagevaluesforthesystemstestedon331
theTITANGPUs).WorkprofilesoutputtedbytheSMDsimulationsareprocessed332
asexplainedinthemaintexttoobtainWQBvalues.Variousmethodscouldbeused333
to obtain free energies from the SMD work, but they have strict convergence334
requirements,arecomputationallymuchmoreexpensiveandtheresultsareonly335
![Page 16: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/16.jpg)
15
valid if the reaction coordinate is mechanistically correct.25 Instead, we simply336
assume that WQB is an upper limit to the equivalent magnitude in free energy337
(ΔGQB).InordertogetascloseaspossibletoΔGQB,werunmultipleSMDreplicas338
andtaketheoveralllowerWQBastherepresentativevalue.Notethatwehaveused339
very conservative settings, favouring sampling over computational efficiency.340
Basedonconvergenceanalysis(SupplementaryFigures6and11)andothertests,341
weproposetheprotocolshowninSupplementaryFigure19forvirtualscreening.342
LessthanoneGPUhourper ligandwouldbenecessarytodiscardapproximately343
80% of candidate ligands and produce a reasonable estimate of WQB for the344
remainingones.Bycomparison,ahigh-throughputimplementationofMM-PBSA(1345
ns of sampling) would require at least 3 GPU hours plus 20 CPU minutes per346
ligand.347
348
Hsp90virtualscreening349
A collection of 280000 purchasable fragment-sized molecules (<250 Da), were350
dockedtotheATPbindingsiteofHsp90withanoptimizedprotocol,wherethekey351
hydrogenbondwithAsp93isenforced.18Next,wegroupedthe1000topscoring352
moleculesinto400clustersbasedonchemicalsimilarityandvisuallyinspectedthe353
top-scoring molecule within each cluster to select 139 molecules that were354
subjected to DUck simulations. Docking score was the main selection criterion,355
with90moleculesoriginatingfromthetop200andallofthemwithinthetop450.356
Additional criteria included high predicted aqueous solubility and chemical357
diversity. The selected molecules were subjected to 100 DUck calculations. We358
divided the molecules in three categories according to their resistance to359
dissociation:weak(WQB<3;N=44;32%),medium(3<WQB<6;N=67;48%)and360
![Page 17: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/17.jpg)
16
strong(WQB>6;N=28;20%).Wetestedallthemoleculesthatwecouldbuyfrom361
the strong set. For comparison, we also purchased and tested 15 molecules of362
medium and 11 from the low stability sets. The chemical structures of the 47363
compoundsareshowninSupplementaryFigure12.364
365
ScreeningbyNMR366
IdentificationofcompoundswhichbindtotheATPsiteofHsp90αwasperformed367
as described previously.32,33 Briefly, a number of 1D 1H NMR experiments (STD,368
water-LOGSY, relaxation filtered) were used to identify interactions between369
compoundsandtheprotein;apotentcompetitor(PU3)wasthenaddedinorderto370
block the ATP binding site. Compounds which bound and were then displaced371
wereidentifiedasinteractingspecificallywiththeprotein.34Moleculesactiveinall372
experimentswereconsideredbonafidehits,whilethosegivingapositiveresponse373
inoneortwoexperimentswereconsideredunconfirmedhitsbecausechangesin374
NMR signal are not necessarily related to binding. All NMR experiments were375
performed on a BrukerAvIII HD 600 MHz NMR spectrometer at 298K; pulse376
sequences included an excitation sculpting module in order to suppress bulk377
water.Samplescontained500µMligandand10µMHsp90αin20mMtrispH7.5,378
50mMNaCl1mMfreshlypreparedDTTandcontained10%D2O.379
380
X-Raycrystallographicstudies381
Protein was produced and crystallized as previously described.35 For the382
successfulcrystals,datawerecollectedat100Konanin-houseBrukerD8Venture383
TXS Generator with a Bruker Photo 100 detector and were subsequently384
processedusingSAINT&SADABS.Thecrystalsbelong to thespacegroups I222.385
![Page 18: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/18.jpg)
17
The structures were solved by molecular replacement using a previously solved386
Hsp90αproteinmodel(PDBcode:1UY6;PU3ligandandsolventremoved)andthe387
programAMoRe.36 Twenty cycles of rigid-body then restrained refinementwere388
carriedoutusingtherefinementprogramREFMAC537followedbymodelbuilding389
andsolventadditionusingthemoleculargraphicsprogramCOOT.38Theprogress390
of the refinement was assessed using Rfree and the conventional R factor. Once391
refinementwascompletedthestructureswerevalidatedusingvariousprograms392
from the CCP4i package.39 Full data collection and refinement statistics are393
presentedinSupplementaryTable4.394
395
Methodologicaldetailsconcerningthecreationofthedatasets,moleculardocking,396
MMPBSAandMMGBSAcalculations,andsurfaceplasmonresonanceexperiments397
areprovidedasSupplementaryInformation.398
399
400
401
402
![Page 19: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/19.jpg)
18
ACKNOWLEDGEMENTS403
We thank Carles Galdeano for helpful discussions and manuscript revision. We404
thank the Barcelona Supercomputing Center for access to computational405
resources. This work was financed by the Spanish Ministerio de Economia406
(SAF2012-33481, SAF2015-68749-R), the Catalan government (2014 SGR 1189)407
andtheICREAAcademiaprogram(FJL).408
CorrespondenceandrequestsformaterialsshouldbeaddressedtoX.B.409
410Authorcontributions411412F.J.L,R.H.andX.B.conceivedtheoverallstrategyofthestudy.P.S.andX.B.413conceivedandimplementedtheDUckapproach.S.R.C.andX.B.carriedoutand414analysedallcomputationalwork.B.D.performedNMRexperiments.L.B.415performedX-raycrystallographyexperiments.N.M.performedSPRexperiments.416Allauthorsdiscussedtheresults,designedexperimentsandwrotethepaper.417418419CodeAvailabilitystatement420421Astep-by-stepguideonhowtoprepare,executeandanalyseDUcksimulations,422alongwithscriptsusedtoautomatetheprocesscanbefoundonthefollowing423website:http://www.ub.edu/bl/undocking424425426427REFERENCES4284291. Kuhnert,M.etal.TracingBindingModesinHit-to-LeadOptimization:430
Chameleon-LikePosesofAsparticProteaseInhibitors.Angew.Chem.Int.Ed.431Engl.54,2849–2853(2015).432
2. Krohn,A.,Redshaw,S.,Ritchie,J.C.,Graves,B.J.&Hatada,M.H.Novel433bindingmodeofhighlypotentHIV-proteinaseinhibitorsincorporatingthe434(R)-hydroxyethylamineisostere.J.Med.Chem.34,3340–3342(1991).435
3. Smith,L.J.,Gunsteren,W.F.Van&Allison,J.R.Multiplebindingmodesfor436palmitatetobarleylipidtransferproteinfacilitatedbythepresenceof437proline12.ProteinSci.22,56–64(2013).438
4. Bissantz,C.,Kuhn,B.&Stahl,M.AMedicinalChemist’sGuidetoMolecular439Interactions.J.Med.Chem.5061–5084(2010).doi:10.1021/jm100112j440
5. Klebe,G.Applyingthermodynamicprofilinginleadfindingandoptimization.441Nat.Rev.DrugDiscov.14,95–110(2015).442
![Page 20: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/20.jpg)
19
6. Ferenczy,G.G.&Keserű,G.M.Thermodynamicsoffragmentbinding.J.443Chem.Inf.Model.52,1039–45(2012).444
7. Kozakov,D.etal.Liganddeconstruction:Whysomefragmentbinding445positionsareconservedandothersarenot.Proc.Natl.Acad.Sci.U.S.A.112,446E2585-2594(2015).447
8. Schmidtke,P.,Luque,F.J.,Murray,J.B.&Barril,X.Shieldedhydrogenbonds448asstructuraldeterminantsofbindingkinetics:applicationindrugdesign.J.449Am.Chem.Soc.133,18903–10(2011).450
9. Colizzi,F.,Perozzo,R.,Scapozza,L.,Recanatini,M.&Cavalli,A.Single-451moleculepullingsimulationscandiscernactivefrominactiveenzyme452inhibitors.J.Am.Chem.Soc.132,7361–71(2010).453
10. Noble,M.E.M.,Endicott,J.a&Johnson,L.N.Proteinkinaseinhibitors:454insightsintodrugdesignfromstructure.Science303,1800–5(2004).455
11. Anderson,D.R.etal.BenzothiopheneinhibitorsofMK2.Part1:structure-456activityrelationships,assessmentsofselectivityandcellularpotency.Bioorg.457Med.Chem.Lett.19,4878–81(2009).458
12. Filippakopoulos,P.&Knapp,S.Targetingbromodomains:epigeneticreaders459oflysineacetylation.Nat.Rev.DrugDiscov.13,337–356(2014).460
13. Ember,S.W.J.etal.Acetyl-lysinebindingsiteofbromodomain-containing461protein4(BRD4)interactswithdiversekinaseinhibitors.ACSChem.Biol.9,4621160–1171(2014).463
14. Mysinger,M.M.,Carchia,M.,Irwin,J.J.&Shoichet,B.K.Directoryofuseful464decoys,enhanced(DUD-E):Betterligandsanddecoysforbetter465benchmarking.J.Med.Chem.55,6582–6594(2012).466
15. Rask-Andersen,M.,Almén,M.S.&Schiöth,H.B.Trendsintheexploitationof467noveldrugtargets.Nat.Rev.DrugDiscov.10,579–90(2011).468
16. Shoichet,B.K.Virtualscreeningofchemicallibraries.Nature432,862–5469(2004).470
17. Brooijmans,N.&Kuntz,I.D.Molecularrecognitionanddockingalgorithms.471Annu.Rev.Biophys.Biomol.Struct.32,335–73(2003).472
18. Ruiz-Carmona,S.etal.rDock:AFast,VersatileandOpenSourceProgramfor473DockingLigandstoProteinsandNucleicAcids.PLoSComput.Biol.10,474e1003571(2014).475
19. Roughley,S.,Wright,L.,Brough,P.,Massey,A.&Hubbard,R.E.Hsp90476inhibitorsanddrugsfromfragmentandvirtualscreening.Top.Curr.Chem.477317,61–82(2012).478
20. Hajduk,P.J.&Greer,J.Adecadeoffragment-baseddrugdesign:strategic479advancesandlessonslearned.Nat.Rev.DrugDiscov.6,211–9(2007).480
21. Joseph-McCarthy,D.,Campbell,A.J.,Kern,G.&Moustakas,D.Fragment-481basedleaddiscoveryanddesign.J.Chem.Inf.Model.54,693–704(2014).482
22. Chen,I.-J.&Hubbard,R.E.Lessonsforfragmentlibrarydesign:analysisof483
![Page 21: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/21.jpg)
20
outputfrommultiplescreeningcampaigns.J.Comput.Aided.Mol.Des.603–484620(2009).doi:10.1007/s10822-009-9280-5485
23. Teotico,D.G.etal.DockingforfragmentinhibitorsofAmpCbeta-lactamase.486Proc.Natl.Acad.Sci.U.S.A.106,7455–7460(2009).487
24. Murray,C.W.etal.Fragment-baseddrugdiscoveryappliedtoHsp90.488Discoveryoftwoleadserieswithhighligandefficiency.J.Med.Chem.53,4895942–55(2010).490
25. Chipot,C.&Pohorille,A.FreeEnergyCalculationsTheoryandApplicationsin491ChemistryandBiology.FreeEnergyCalculationsTheoryandApplicationsin492ChemistryandBiology(SpringerBerlinHeidelberg,2007).493
26. Alvarez-Garcia,D.&Barril,X.Molecularsimulationswithsolvent494competitionquantifywaterdisplaceabilityandprovideaccurateinteraction495mapsofproteinbindingsites.J.Med.Chem.57,8530–9(2014).496
27. ChemicalComputingGroupInc.MolecularOperatingEnvironment(MOE),4972014.09.(2015).498
28. Jakalian,A.,Jack,D.B.&Bayly,C.I.Fast,efficientgenerationofhigh-quality499atomiccharges.AM1-BCCmodel:II.Parameterizationandvalidation.J.500Comput.Chem.23,1623–41(2002).501
29. Bayly,C.I.,McKay,D.&Truchon,J.F.AnInformalAMBERSmallMolecule502ForceField:parm@Frosst.(2011).503
30. Case,D.A.etal.AMBER12.(2012).504
31. Ryckaert,J.-P.,Ciccotti,G.&Berendsen,H.J.C.Numericalintegrationofthe505cartesianequationsofmotionofasystemwithconstraints:molecular506dynamicsofn-alkanes.J.Comput.Phys.23,327–341(1977).507
32. Brough,P.aetal.Combininghitidentificationstrategies:fragment-based508andinsilicoapproachestoorallyactive2-aminothieno[2,3-d]pyrimidine509inhibitorsoftheHsp90molecularchaperone.J.Med.Chem.52,4794–809510(2009).511
33. Baurin,N.etal.Designandcharacterizationoflibrariesofmolecular512fragmentsforuseinNMRscreeningagainstproteintargets.J.Chem.Inf.513Comput.Sci.44,2157–2166(2004).514
34. Davis,B.inProtein-LigandInteractions:MethodsandApplications(eds.515Williams,M.A.&Daviter,T.)1008,389–413(SpringerScience+Business516Media,2013).517
35. Wright,L.etal.Structure-activityrelationshipsinpurine-basedinhibitor518bindingtoHSP90isoforms.Chem.Biol.11,775–85(2004).519
36. Navaza,J.AMoRe:anautomatedpackageformolecularreplacement.Acta520Crystallogr.Sect.AFound.Crystallogr.50,157–163(1994).521
37. Murshudov,G.N.,Vagin,A.A.&Dodson,E.J.Refinementofmacromolecular522structuresbythemaximum-likelihoodmethod.ActaCrystallogr.Sect.DBiol.523Crystallogr.53,240–255(1997).524
![Page 22: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/22.jpg)
21
38. Emsley,P.&Cowtan,K.Coot:model-buildingtoolsformoleculargraphics.525ActaCrystallogr.Sect.DBiol.Crystallogr.60,2126–2132(2004).526
39. Collaborative,C.P.&others.TheCCP4suite:programsforprotein527crystallography.ActaCrystallogr.D.Biol.Crystallogr.50,760(1994).528
529
![Page 23: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/23.jpg)
22
TABLES530531Table1.Summaryofresultsforthe9Hsp90NMRClass1hits.Chemicalstructures532ofallcompoundsareshowninSupplementaryTable1.533534
ID MW
Docking DUck SPR
Kd
(mM)
PDB
Simb
ChEMBL
Sim.bScore(Ranka)
Score(Ranka)
1* 248.7-25.0(79)
9.1(10)
772XDX(0.37)
CHEMBL1340447
(0.44)
2* 221.3-25.0(73)
8.2(11)
3202WI6(0.29)
CHEMBL1536318
(0.54)
3* 230.2-26.7(19)
11.3(1)
7004EFU(0.32)
CHEMBL1458840
(0.51)
4 240.3-26.4(22)
7.4(16)
7303WHA(0.29)
CHEMBL1542436
(0.37)
5 165.2-23.8(128)
8.1(12)
-4EFT(0.27)
CHEMBL1313412
(0.28)
6 206.3-23.3(138)
9.5(5)
-3HHU(0.42)
CHEMBL2103879
(0.42)
7 236.3-25.4(51)
7.8(15)
-3B24(0.31)
CHEMBL1375884
(0.36)
8 224.7-25.3(58)
7.0(22)
-2XDX(0.35)
CHEMBL1383799
(0.37)
22 237.3-28.3(2)
5.6(33)
-3O0I(0.27)
CHEMBL1834092
(0.33)
535*XraystructuresolvedaPositionwithinthelist536of149moleculesthatwereevaluatedwith537DUck.bHsp90structureinthePDBor538compoundwithHsp90activityinChEMBL(as539of23/03/2016)withtheclosestsimilarityto540thefragmenthit.Similarity(valuesin541parentheses)wascalculatedwithOpenBabel542usingtheFP2fingerprint.543544 545
![Page 24: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/24.jpg)
23
FIGURECAPTIONS546547
548
Figure 1. Calculation of WQB. a. The receptor is idealized as a model system549
containing only the local environment around a key intermolecular hydrogen550
bond. b. Representative work profiles obtained from dynamic undocking551
simulationsforastrong(black)andaweak(grey)ligand.Thequasi-boundstateis552
definedas thepointwith thehighestenergyrelative to the idealhydrogenbond553
geometry.554
555556
Figure 2. Application of the quasi-bound approximation to ligand ranking. a.557
DistributionofWQBvaluesofpotentCDK2 ligands(IC50<1µM;darkgrey),weak558
CDK2 ligands (IC50 > 1µM; light grey) and non-binding decoys (black). Points559
indicatepopulationvalues,fromwhichthesmoothlinesareextrapolated.b.ROC560
curves for the CDK2 (black), A2AR (red) and Trypsin (green) DUD sets. Plotted561
results correspond to 2 DUck runs per ligand. AUC values are shown in562
Supplementary Figure 6. c. Docking score vs. WQB values for active (red) and563
inactive (black or gray) compounds in the CDK2 retrospective virtual screening564
dataset. The quadrant in orange highlights the area corresponding to top 25%565
dockingscoreandtop25%WQBvalues,whereoptimalenrichmentfactors(EF)are566
achieved.d.Forthesameset,distributionofWQBvaluesfortheactivecompounds567
(red), all decoys (black) and decoys in the top 25% docking score (gray). e.568
DistributionofWQBvaluesofCDK2actives(red)anddecoys(gray)rankedinthe569
top25%bytwoindependentdockingprograms(rDockandGlide).f.Distribution570
ofWQBvaluesofCDK2actives(red)anddecoys(gray)rankedinthetop25%both571
byMMPBSAandtherDockdockingprogram.572
573
574Figure3.AdditionalanalysesoftheprospectiveapplicationofDUckinHsp90.a.575
DistributionofWQBvaluesfor139topdockingscorers(palegray),47compounds576
withinthissetthatwerepurchased(darkgray),andthe9compoundsdetectedas577
active.b.PiechartsshowingthehitratesforthesetofcompoundswithhighWQB578
(top),mediumWQB(middle)andlowWQB(bottom).Theareainblackcorresponds579
tobonafidehits,darkgrayrepresentscompoundsthatgiveapositivesignalin1580
or2NMRexperiments,palegraycorrespondstoinactivecompounds.Labels581
indicatethenumberofcompoundsofeachclass.Chemicalstructuresareshownin582
SupplementaryFigure12.583
584
![Page 25: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/25.jpg)
24
Figure4.Experimental(grey)andpredicted(orange)bindingmodesofthe585
fragmenthits.a.Compound1,theRMSDofthewholemoleculeis2.58Åduetoa586
conformationalchangeoftheproteinnexttothep-toluenering.Thepyridinering587
andbondedatoms,wherethekeyinteractionoccur,haveaRMSDof0.54Åb.588
Compound2hasaRMSDof0.54Åc.Compound3hasaRMSDof1.55Å,all589
hydrogenbondinteractionsarepreserved.590
591
![Page 26: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/26.jpg)
![Page 27: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/27.jpg)
![Page 28: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/28.jpg)
![Page 29: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/29.jpg)
![Page 30: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/30.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-1-
SupplementaryInformation
DynamicUndockingandtheQuasi-BoundStateasToolsforDrugDesign
SergioRuiz-Carmona,PeterSchmidtke,F.JavierLuque,LisaBaker,NataliaMatassova,Ben
Davis,StephenRoughley,JamesMurray,RodHubbard,XavierBarril
![Page 31: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/31.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-2-
INDEX
SUPPLEMENTARYMETHODS.................................................................................................3
DATASETS....................................................................................................................................3
MOLECULARDOCKINGWITHRDOCK.................................................................................................4
MOLECULARDOCKINGWITHGLIDE..................................................................................................4
MMGBSAANDMMPBSA............................................................................................................4
SURFACEPLASMONRESONANCE.......................................................................................................5
SUPPLEMENTARYREFERENCES..............................................................................................6
SUPPLEMENTARYFIGURES....................................................................................................7
SUPPLEMENTARYFIGURE1.............................................................................................................7
SUPPLEMENTARYFIGURE2.............................................................................................................7
SUPPLEMENTARYFIGURE3.............................................................................................................8
SUPPLEMENTARYFIGURE4.............................................................................................................9
SUPPLEMENTARYFIGURE5.............................................................................................................9
SUPPLEMENTARYFIGURE6...........................................................................................................10
SUPPLEMENTARYFIGURE7...........................................................................................................11
SUPPLEMENTARYFIGURE8...........................................................................................................12
SUPPLEMENTARYFIGURE9...........................................................................................................13
SUPPLEMENTARYFIGURE10.........................................................................................................14
SUPPLEMENTARYFIGURE11.........................................................................................................15
SUPPLEMENTARYFIGURE12.........................................................................................................16
SUPPLEMENTARYFIGURE13.........................................................................................................19
SUPPLEMENTARYFIGURE14.........................................................................................................20
SUPPLEMENTARYFIGURE15.........................................................................................................21
SUPPLEMENTARYFIGURE16.........................................................................................................22
SUPPLEMENTARYFIGURE17.........................................................................................................23
SUPPLEMENTARYFIGURE18.........................................................................................................24
SUPPLEMENTARYFIGURE19.........................................................................................................24
SUPPLEMENTARYFIGURE20.........................................................................................................25
SUPPLEMENTARYFIGURE21.........................................................................................................26
SUPPLEMENTARYFIGURE22.........................................................................................................27
SUPPLEMENTARYFIGURE23.........................................................................................................28
SUPPLEMENTARYFIGURE24.........................................................................................................29
SUPPLEMENTARYFIGURE25.........................................................................................................29
SUPPLEMENTARYFIGURE26.........................................................................................................30
SUPPLEMENTARYTABLES....................................................................................................31
SUPPLEMENTARYTABLE1.............................................................................................................31
SUPPLEMENTARYTABLE2.............................................................................................................32
SUPPLEMENTARYTABLE3.............................................................................................................32
SUPPLEMENTARYTABLE4.............................................................................................................33
SUPPLEMENTARYTABLE5.............................................................................................................34
SUPPLEMENTARYTABLE6.............................................................................................................35
![Page 32: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/32.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-3-
SUPPLEMENTARYMETHODS
Datasets
Whenpossible,datasetsweregearedtowardsfragment-sizedligandsbecausethey
present more scaffold diversity, make fewer peripheral interactions that could
maskthemaininteractionsandbecauseFragment-BasedDrugDiscovery(FBDD)
approaches are increasingly important ashit identification strategy.1,2 ForCDK2,
allligandswithmolecularweightbelow300Daandknownbindingaffinity(IC50)
wereextractedfromthePDB.3Toincreasethediversityofthedataset,allligands
wereclusteredat75%similarityusingtheMACCSfingerprintsasimplementedin
MOE(ChemicalComputingGroupInc.,2015)andonlythecentroidswereusedto
definetheactiveset.ThecompositionofthedatasetisdescribedinSupplementary
Table 5. It shouldbenoted that this is a noisy dataset becausedata sources are
veryheterogeneousandIC50valueshaveanindirectrelationshipwithdissociation
constants.4Assuch,itshouldonlybeusedtodetecttrends.Inordertoassessthe
significanceof thecorrelation,wehavealso investigatedthecorrelationbetween
IC50 and molecular weight (Supplementary Figure 20). For retrospective VS
experiments,apoolof30decoysperactivefragmentwasobtainedwiththeDUD-E
decoygenerator,5whichputstogetherasetofputatively inactivemoleculeswith
physicochemical properties very similar to active ones. For BRD4, as it was
designedtostudythecorrelationbetweenexperimentalbindingaffinityandWQB,
only the ligands with known binding mode and measured IC50 or KD were
considered(relationshipwithmolecularweightreportedinSupplementaryFigure
21).Thecrystalstructureofeachligand-proteincomplexwasobtainedfromPDB
andusedas input for subsequent calculations.The compositionof thedataset is
described in Supplementary Table 6. In the case of AA2AR, as there are few
structuresinthePDB,theactivefragmentsweretakenfromtheDUD-Ebenchmark
set.5TherestoftheprocedureisthesameasdescribedforCDK2.ForTrypsin,we
found that few ligandshave a lowmolecularweight sowedidnot filterby size.
Instead,arandomsubsetof2000activesanddecoyswasselectedfromDUD-E.In
the case of Hsp90, all candidate molecules originate from a unified collection
generatedinhousefromthecommerciallibrariesoffivepreferredvendors(Specs,
Enamine,LifeChemicals,PrincetonBiomolecularsandAsinex).Inthiscaseweset
anupperlimitof250Da,obtaining280000candidatefragments.Allligandswere
![Page 33: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/33.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-4-
prepared for docking using Schrödinger’s Ligprep6with the following options
differentthandefault:neutralizeandionizeatpH7withathresholdof+-1witha
maximumof6tautomersand8stereoisomersgenerated.
MolecularDockingwithrDock
ForCDK2,AA2ARandTrypsin, the3Dstructureused todefine thereceptorwas
obtained fromtheDUD-Ebenchmarkset.5MOE7wasused togeneratemol2 files
thatcanbereadbyrDock,ourdockingengine.8ForHsp90,weusethesamecavity
definition and docking protocol described previously.8 In all systems,
pharmacophoricrestraintswereusedtoensurethatthekeyinteractionpointwas
matchedby everymolecule in thedataset, asdefined in SupplementaryTable 3.
rDock was run with the default parameters for standard docking. 50 individual
dockingprocesseswereexecutedperligand,thusensuringthatthelowest-energy
bindingmode is identified. The best-scoring solution is accepted as the putative
bindingmode.Ligandsthatdonotfulfillthepharmacophoreareidentifiedbythe
restraintpenaltyandeliminated fromthedataset (i.e.notconsidered in theROC
curvesoranyotheranalysis).
MolecularDockingwithGlide
Inordertodemonstratethatourmethodologyprovidesanadvantageregardlessof
the docking program used, we also run CDK2, AA2R and trypsin systems with
Glide.9 The generationof the cavitywithGlidewasperformedusing coordinates
defined as in rDock docking and default parameters. Pharmacophoric restraints
were defined to force all ligands to make a hydrogen bond as defined in
SupplementaryTable3.Glidedockingwasrunwithdefaultparametersandwith
pharmacophoric restraints (Supplementary Figures 22, 23 and 24). The best
dockingposeforeachligandwasselectedandusedasinputforDUck.
MMGBSAandMMPBSA
MMGBSA and MMPBSA calculations using AMBER12 software were also
performedandcomparedagainsttherestofmethods.Eachligandwassimulated
for 5 ns with the full size receptor of CDK2 using the same MD configuration
defined in the section above (Supplementary Figures 24 and 25). For each
![Page 34: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/34.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-5-
simulation, a total of 25 snapshots separated by 200 ps were used and the free
energieswereaveragedover theensembleof conformations.All the calculations
wereperformedwithdefaultparameterswiththeexceptionofthefollowing:the
GBmodelusedisoneofthedevelopedbyOnufrievetal.10(igb=2)andtheatomic
radiiaresetupaccordingtothetopology(radiopt=0).
Surfaceplasmonresonance
Surface plasmon resonance (SPR) experiments have been done mainly as
described before.11,12 All measurements were performed on a Biacore T200
instrument(BiacoreGEHealthcare)at20°ConSeriesSNTAchips.25mMHEPES
pH7.4, 175mMNaCl, 0.01%P-20, 0.025mMEDTAand1%DMSOwasusedas a
runningbuffer.HSP90proteinwasproducedasdescribedpreviously.Chipsurface
wasgeneratedwithmulti-His-taggedHsp90proteinas inreference.11Thesensor
surfacewasregeneratedby0.35MEDTAand45%DMSOwithadditional60sec
injections of 0.1mg/mL trypsin and0.5M imidazole. In some experiments, the
protein was further stabilized on NTA surface by covalent amine coupling as
advisedbymanufacturer.Screeningoffragmentswasconductedindoseresponse
titrationsofninetwo-folddilutedexperimentalpointswiththetopconcentration
of500µM.Eachfragmenthasbeentestedatleastthreetimes.Dataprocessingwas
performed using BIAevaluation 2.1 (Biacore GE Healthcare Bio-SciencesCorp) or
Scrubber2 (BioLogic) software. Sensorgrams were double referenced prior to
global fitting of the concentration series to a Steady State Affinity model.
RepresentativesensorgramsareshowninSupplementaryFigure26.
![Page 35: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/35.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-6-
SUPPLEMENTARYREFERENCES
1. Joseph-McCarthy,D.,Campbell,A.J.,Kern,G.&Moustakas,D.Fragment-basedleaddiscoveryanddesign.J.Chem.Inf.Model.54,693–704(2014).
2. Hajduk,P.J.&Greer,J.Adecadeoffragment-baseddrugdesign:strategicadvancesandlessonslearned.Nat.Rev.DrugDiscov.6,211–9(2007).
3. Berman,H.M.etal.TheProteinDataBank.NucleicAcidsRes.28,235–242(2000).
4. Klebe,G.Applyingthermodynamicprofilinginleadfindingandoptimization.Nat.Rev.DrugDiscov.14,95–110(2015).
5. Mysinger,M.M.,Carchia,M.,Irwin,J.J.&Shoichet,B.K.Directoryofusefuldecoys,enhanced(DUD-E):Betterligandsanddecoysforbetterbenchmarking.J.Med.Chem.55,6582–6594(2012).
6. Schrödinger.LigPrep,version2.9.Schrödinger,LLC(2014).
7. ChemicalComputingGroupInc.MolecularOperatingEnvironment(MOE),2014.09.(2015).
8. Ruiz-Carmona,S.etal.rDock:AFast,VersatileandOpenSourceProgramforDockingLigandstoProteinsandNucleicAcids.PLoSComput.Biol.10,e1003571(2014).
9. Friesner,R.aetal.Glide:anewapproachforrapid,accuratedockingandscoring.1.Methodandassessmentofdockingaccuracy.J.Med.Chem.47,1739–49(2004).
10. Onufriev,A.,Bashford,D.&Case,D.A.ExploringProteinNativeStatesandLarge-ScaleConformationalChangeswithaModifiedGeneralizedBornModel.ProteinsStruct.Funct.Genet.55,383–394(2004).
11. Meiby,E.etal.Fragmentscreeningbyweakaffinitychromatography:comparisonwithestablishedtechniquesforscreeningagainstHSP90.Anal.
Chem.85,6756–66(2013).
12. Murray,J.B.,Roughley,S.D.,Matassova,N.&Brough,P.a.Off-ratescreening(ORS)bysurfaceplasmonresonance.Anefficientmethodtokineticallysamplehittoleadchemicalspacefromunpurifiedreactionproducts.J.Med.
Chem.57,2845–2850(2014).
13. Anderson,D.R.etal.BenzothiopheneinhibitorsofMK2.Part1:structure-activityrelationships,assessmentsofselectivityandcellularpotency.Bioorg.
Med.Chem.Lett.19,4878–81(2009).
14. O’Boyle,N.M.etal.OpenBabel:AnOpenchemicaltoolbox.J.Cheminform.3,(2011).
![Page 36: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/36.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-7-
SUPPLEMENTARYFIGURES
SupplementaryFigure1
Graphicalrepresentationof thequasi-boundstate inrelationtothedissociationprocess.The macroscopic constants describing the behavior of a non-covalent complex aredetermined by the relative free energies of three states (bound, transition state andunbound). States in-between are theoretically irrelevant, somolecules1, 2 and3wouldhave the same kinetic and thermodynamic constants. The Quasi-bound state is merelydesigned to probe the slope around the bound state, obtaining an approximation to thestructuralstabilityofthebindingmode.Wefindthattrueligandsaremorelikelytohaveaprofilelike1,whereasmanydecoyshaveprofilessimilarto2or3.
SupplementaryFigure2
WQBvaluesvs.experimentallydeterminedactivities(expressedasLog(IC50)), forasetof41Fragment-likeCDK2ligandstakenfromthePDB.Ligand3FZ1isshowninredandnotincluded in the correlation.As shownbelow, this liganddoesnot fulfill the condition ofusingthehingeregionasattachmentpoint.
![Page 37: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/37.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-8-
SupplementaryFigure3
DistributionofWQBvaluesasa functionofbindingaffinity(IC50), fortheCDK2(top)andBRD4set(bottom).CompoundswiththesamebindingaffinitypresentawidedistributionofWQBvalues,butthereisatendencytowardshighervaluesformorepotentcompounds.Mostnotably,verylowWQBvaluesarerareforpotentligands.
![Page 38: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/38.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-9-
SupplementaryFigure4
Binding mode of ligand in PDB structure 3FZ1. This ligand is unusual because itsinteraction with the hinge region is labile. Structural and SAR data confirms that thisinteractionisnotimportantforpotency.13Instead,thisligandformstwocharge-reinforcedhydrogen bonds with Nζ of Lys33 and Oδ1 of Asn132, from which it draws structuralstability(WQB=10.50kcal/mol;SupplementaryFigure18).NotethattheIC50reportedinthePDBforthiscompoundiswrong.Thecorrectvalueis146nM.13
SupplementaryFigure5
WQBvaluesvs.experimentallydeterminedactivities(expressedasLog(IC50)), forasetof30 BRD4 ligands taken from the PDB. The points in red have not been included in thecorrelation. They correspond to three kinase inhibitors that bind to BRD4 as anunintended secondary target and present extremely low resistance to breaking theinteractionwithNδ2ofAsn120(PDBcodes4O74,4O77&4O7E).
![Page 39: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/39.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-10-
SupplementaryFigure6
ROC curves (left) and semilog-ROC curves (right) of the retrospective virtual screeningexperimentsonCDK2(top),AA2R(middle)andTrypsin(bottom).Thegreylineindicatesthebaseline(randomselection).ForCDK2,theresultscorrespondingto2,8and22DUckruns are reported. For AA2R, the results corresponding to 2, and 8 DUck runs arereported.ForTrypsin,only2DUckrunswereexecuted.AUCvaluesareinsetintheplots.
![Page 40: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/40.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-11-
SupplementaryFigure7
Docking(rDock)scorevs.WQBvaluesforactive(red)andinactivecompounds(blackorgray)intheretrospectivevirtualscreeningdatasetsforCDK2(top),AA2AR(middle)andTrypsin(bottom).Thesidepanelsshowthedistributionofactive(red)andinactive(black)compoundsforeachindividualmethod(dockingtotheleft,DUckatthebottom).Graypoints(centralpanel)andgrayline(inferiorpanel)representthedecoyswithadockingscorewithinthetop25%.Theorangesquarehighlightstheareacorrespondingtotop25%dockingscoreandtop25%WQBvalues,whereoptimalenrichmentfactors(EF)areachieved.
![Page 41: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/41.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-12-
SupplementaryFigure8
Docking score vs. WQB obtained for two different programs on the CDK2 test set. EachmoleculewasdockedwithrDock(top)orGlide(bottom)andthebindingmodegeneratedby each program was used as starting geometry for DUck simulations. In both cases,dockingscoresareorthogonaltoWQBandahighproportionofgoodscorershaveverylowWQBvalues.Theintersectionbetweenmethodsdefinesasubsethighlyenrichedinactivemolecules.Twointersectinglevelsarepresentedperprogram:top25%docking+top25%DUck(left);andtop25%docking+top12%DUck(right).
![Page 42: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/42.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-13-
SupplementaryFigure9
MMPBSA and MMGBSA-calculated ΔGbind vs. WQB on the CDK2 test set. The rDock-generatedbindingmodewasusedas startingpoint formoleculardynamics simulations,which where then processed to obtain MMPBSA and MMGBSA binding free energies. Inboth cases, the calculatedΔGbind values are orthogonal to WQB and a high proportion ofgoodMM(PB/GB)SAscorershaveverylowWQBvalues.Theintersectionbetweenmethodsdefinesasubsethighlyenrichedinactivemolecules.Twointersectinglevelsarepresentedpermethod:top25%MM(PB/GB)SA+top25%DUck(left);andtop25%MM(PB/GB)SA+top12%DUck(right).
![Page 43: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/43.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-14-
SupplementaryFigure10
FilteringbyWQBincreasesperformanceevenafterconsensusscoring(CDK2testset).Theleft panels show a scatter plot of rDock score vs. Glide score (top) and rDock score vs.MMPBSA-calculatedΔGbind (bottom).Molecules ranked in the top25%bybothmethods(highlightedarea)arethenbinnedaccordingtotheirWQB(rightpanels,alsoshowninthemaintext).FilteringbyWQBwould increasetheenrichmentfactor inacut-offdependentmanner.
![Page 44: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/44.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-15-
SupplementaryFigure11
Percentageofactivemoleculesinthetop21(outof47compoundstested)asafunctionofthenumberofDUckruns.Atthescreeningstagewecarriedout100DUcksimulationsperligand,obtainingahitrateof38%.Retrospectively,wetook50randomcombinationsofN={2,4,6,8,10,20,50} DUck runs and calculated the hit rates that would have beenobtained.Averagesare representedas filled circles and labeledwith their actual values.Thebarsspanfromthemaximumtotheminimumvalues.
Number of DUck Runs
Hit R
ate
●
0.31
●
0.318
●
0.329
●
0.338
●
0.356
●
0.376
●
0.381
●
0.381
0.0
0.1
0.2
0.3
0.4
0.5
2 4 6 8 10 20 50 100
![Page 45: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/45.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-16-
SupplementaryFigure12
Chemical structure of the tested compounds. Duck Class refers to strong, medium andweak binders (1, 2 & 3, respectively). NMR Class 1 are true binders. The rest areconsideredinactive.TherealnumberscorrespondtorDockscore(left)andWQB(right).
Cl
N
N
N
N
DUck Class: 1
NMR Class: 1
-24.9700 9.1000
ID: 1
S
N
O
N
N
DUck Class: 1
NMR Class: 1
-25.0300 8.2000
ID: 2
O
O
N
O
N
DUck Class: 1
NMR Class: 1
-26.6200 11.3000
ID: 3
S O
N
N
NN
DUck Class: 1
NMR Class: 1
-26.4500 7.4000
ID: 4
O
N
N
N
DUck Class: 1
NMR Class: 1
-23.7700 8.2000
ID: 5
S
NN
N
N
DUck Class: 1
NMR Class: 1
-23.2600 9.5000
ID: 6
S
NO
N
NN
DUck Class: 1
NMR Class: 1
-25.4500 7.8000
ID: 7
Cl
N N
N
N
N
N
DUck Class: 1
NMR Class: 1
-25.3500 7.0000
ID: 8
S
N
ON
NN
N
N
DUck Class: 1
NMR Class: 2
-28.0400 7.3000
ID: 9
O
N
O
N
O
DUck Class: 1
NMR Class: 2
-27.1200 6.4000
ID: 10
S NO
N
N
N
N
DUck Class: 1
NMR Class: 2
-26.1400 9.8000
ID: 11
S
N
N
N N
N N+
DUck Class: 1
NMR Class: 3
-25.4000 7.0000
ID: 12
S
NO
N
O
NN
DUck Class: 1
NMR Class: 3
-25.0100 6.4000
ID: 13
N
N
N
N
N
N N
NN
DUck Class: 1
NMR Class: nb
-26.0000 6.5000
ID: 14
O
O
N
N
+
DUck Class: 1
NMR Class: nb
-25.9100 6.7000
ID: 15
S
O
NN
NN
N
DUck Class: 1
NMR Class: nb
-24.7300 6.5000
ID: 16
O
N
NN
DUck Class: 1
NMR Class: nb
-24.4600 10.3000
ID: 17
N
NN
NN
N
+
DUck Class: 1
NMR Class: nb
-28.4400 7.2000
ID: 18
N
N N
N
DUck Class: 1
NMR Class: nb
-25.9200 9.2000
ID: 19
S ON
N
N
N
N
DUck Class: 1
NMR Class: nc
-26.3400 7.3000
ID: 20
![Page 46: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/46.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-17-
SupplementaryFigure12(cont)
S
N
N
N
DUck Class: 1
NMR Class: r
-24.0100 8.0000
ID: 21
S
NO
N
N
NN
DUck Class: 2
NMR Class: 1
-28.2700 5.6000
ID: 22
S N
N
NN
S
DUck Class: 2
NMR Class: 2
-26.8600 4.2000
ID: 23
N
N N
N
DUck Class: 2
NMR Class: 2
-25.3300 3.6000
ID: 24
S O
N
N
N
N
N
DUck Class: 2
NMR Class: 2
-25.3200 3.9000
ID: 25
O
N
N
N
N
N
DUck Class: 2
NMR Class: 3
-25.0900 5.5000
ID: 26
N
N
N
N
N
+
+
DUck Class: 2
NMR Class: 3
-27.3400 3.5000
ID: 27
O
N
N
N
N
DUck Class: 2
NMR Class: 3
-27.5500 3.2000
ID: 28
O
N
N
O
N
N
DUck Class: 2
NMR Class: nb
-25.4600 4.7000
ID: 29
O
O
N
N
O
DUck Class: 2
NMR Class: nb
-25.0900 5.5000
ID: 30
O
N
NN
NN
N
N
DUck Class: 2
NMR Class: nb
-26.7600 4.4000
ID: 31
O
N
N
N
+
DUck Class: 2
NMR Class: nb
-26.4700 3.1000
ID: 32
S
N
NN
N
N
DUck Class: 2
NMR Class: nb
-25.1000 4.9000
ID: 33
S
O
N
N
N
N
DUck Class: 2
NMR Class: nb
-24.8300 4.8000
ID: 34
O
O
O
NN
N
N N
N
DUck Class: 2
NMR Class: nb
-24.3200 4.4000
ID: 35
O
N
N
N
N
NN
DUck Class: 2
NMR Class: nb
-26.8200 3.0000
ID: 36
S
N
NN
N
N
DUck Class: 3
NMR Class: 2
-27.9100 0.8000
ID: 37
O
O
N
N
O
N
DUck Class: 3
NMR Class: 2
-26.2500 1.4000
ID: 38
S
N
N
NN
DUck Class: 3
NMR Class: 2
-25.4700 0.5000
ID: 39
S
NNN
O
O
O
DUck Class: 3
NMR Class: 3
-25.3700 1.0000
ID: 40
![Page 47: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/47.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-18-
SupplementaryFigure12(cont)
O
N
O
NN
DUck Class: 3
NMR Class: nb
-26.4800 2.3000
ID: 41
O
N
N N+ +
DUck Class: 3
NMR Class: nb
-25.7700 2.5000
ID: 42
S
O
O
N
N+
DUck Class: 3
NMR Class: nb
-24.9900 0.0000
ID: 43
SO O
N
N
N
O
+
DUck Class: 3
NMR Class: nb
-24.9700 0.0000
ID: 44
ONN
NN
N
N
DUck Class: 3
NMR Class: nb
-24.9100 0.0000
ID: 45
O
N
N
N
N
O
DUck Class: 3
NMR Class: nb
-23.2300 2.0000
ID: 46
O
N
N
N
N
NN
DUck Class: 3
NMR Class: nb
-26.1100 1.7000
ID: 47
![Page 48: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/48.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-19-
SupplementaryFigure13
CrystalstructureofHsp90incomplexwithcompounds1(top),2(middle)and3(bottom).The2fofcelectrondensitymapsaredisplayedatthe1.0Sigmalevel(Carve=1.7).
![Page 49: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/49.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-20-
SupplementaryFigure14
Predictedbindingmodesforcompounds4,5and6(fromlefttoright).
![Page 50: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/50.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-21-
SupplementaryFigure15
Dependenceof the resultson the sizeof the receptor.WQB valuesofCDK2 ligandswerecalculated using the whole protein as receptor and plotted against the results obtainedwithatruncatedsystem(top).WQBvaluesobtainedwiththetruncatedsystemrepresentalowerbound to thoseobtainedwith the full system.This indicates thatwhen thewholesystem is included, WQB may not reflect the contribution of the interaction underinvestigation. Potentially, this may give rise to false positives. Noteworthy, the virtualscreening results are comparable to those obtainedwith the truncated system (bottom;comparewithSupplementaryFigure8).
![Page 51: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/51.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-22-
SupplementaryFigure16
Dependenceoftheresultsonthesteeringvelocity.Twodifferentvelocitiesarecompared:5Å·ns-1(usedthroughthiswork)and1.25Å·ns-1.Slowervelocitiesmeanmoresamplingand, potentially, lower WQB values. The high correlation (r2=0.94) indicates that thestandardconditions(v=5Å·ns-1)produceconvergedresults.
![Page 52: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/52.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-23-
SupplementaryFigure17
Dependenceoftheresultsonthechoiceofreactioncoordinate.WQBvaluesobtainedusingtwodifferentatomsof reference in thehingeregionofCDK2arehighlycorrelated(top)andaffordsimilarenrichmentfactorsinretrospectivevirtualscreening(bottom;comparewith Supplementary Figure 8). The atoms used as reference (Leu83:N in the x-axis andLeu83:O in the y-axis) are part of the hinge and located in close proximity (3Å). Mostligandsformahydrogenbondwithbothatomsatthesametime.PointsinredrepresentligandsthatonlyformahydrogenbondwithLeu83:N.
![Page 53: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/53.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-24-
SupplementaryFigure18
WQBofCDK2ligandspullingtheamineofLys33andcomparisonwithWQBvaluesobtainedforthehingeregion(inkcal/mol).OnlythoseligandscapableofformingahydrogenbondwithNZofLys33havebeenconsidered.Itshouldbenotedthatthispartoftheactivesitepresents large conformational diversity between structures. In consequence, the DUckresultsmaybelessreliablethanforthehingeregion.
PDB Code
WQB (O Leu83)
WQB
(Nζ Lys33)
1OIQ 4.48 5.18
3BHT 6.59 9.65
3BHV 6.94 0.00
3EJ1 5.77 2.06
3FZ1 0.12 10.50
3QTQ 6.66 5.56
3QTW 9.76 5.91
3TIY 3.47 0.00
SupplementaryFigure19
Proposedprotocol forDUck-basedvirtual screeningandcomparisonwithMMPBSA.Thesmallersizeofthesystemspeedsupcalculationsbyafactoror5(SupplementaryTable2),also permitting shorter equilibration times. Each ligand undergoes equilibration and atleast twoSMDs(45GPUminutes).MoleculeswithWQB aboveagiven threshold (e.g. t=6kcal/mol) would then proceed to N cycles of unbiased MD + SMD simulations (42 GPUminutespercycle).AsimilarprotocolforMMPBSAwouldrequireatleast2GPUhoursofequilibration followed by N cycles of 1ns MD simulation and MMPBSA calculation ofrepresentativesnapshots(1GPUhour+20CPUminutespercycle).
![Page 54: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/54.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-25-
SupplementaryFigure20
CDK2testset:correlationbetweenLogIC50andWQB isnotcausedbyMolecularWeight.ThecorrelationbetweenWQBandMW(r2=0.07)islowerthanthecorrelationbetweenWQBandLogIC50 (r2=0.23) or betweenLogIC50 andMW (r2=0.36).Redpoints (discussed inSupplementaryFigure2)areexcludedfromallcorrelations.
![Page 55: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/55.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-26-
SupplementaryFigure21
BRD4testset:correlationbetweenLogIC50andWQB isnotcausedbyMolecularWeight.ThecorrelationbetweenWQBandMW(r2=0.17)islowerthanthecorrelationbetweenWQBand logIC50 (r2=0.26) or between LogIC50 and MW (r2=0.43). Red points (discussed inSupplementaryFigure5)areexcludedfromallcorrelations.
![Page 56: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/56.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-27-
SupplementaryFigure22
ROCcurvescomparisonofDUck(instandalonemode)withunbiaseddockingwithGlideandrDockforthethreetestsystems:CDK2,AA2ARandTrypsin.
AA2AR CDK2
Trypsin
![Page 57: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/57.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-28-
SupplementaryFigure23
ROC curves comparison of DUck (in standalone mode) with pharmacophore-guideddockingwithGlideandrDockforthethreetestsystems:CDK2,AA2ARandTrypsin.
AA2AR CDK2
Trypsin
![Page 58: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/58.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-29-
SupplementaryFigure24
DUck postfiltering improves early enrichment. Semilogarithmic ROC curves for theretrospective virtual screeningof CDK2, obtainedwith thebest-performingprogram forthis test set (Glide),alone or in combination with three different postfiltering methods:DUck(left);MMPBSA(middle)andrDock(right).LigandswereinitiallyrankedaccordingtoGlide’sscore.Then,movedtothebackofthelistiftheywerenotinthetop12%oftherescoringmethod.ThisshowsthattheGlide-DUckcombinationissuperiortoGlidealone.For this test set the effect ismost prominent in the top 1% to 5%of the library. Glide-MMPBSA combination is provided for comparison and affords very similar results. TheGlide-rDockcombinationdoesnotimproveearlyenrichment.
SupplementaryFigure25
ROCcurvescomparisonofDUck(instandalonemode)withMMPBSAandMMGBSAforCDK2
CDK2
![Page 59: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/59.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-30-
SupplementaryFigure26
Examplesof typical sensorgrams (left column)and steady stateplots (right column) forthe binding of the fragment hits to Hsp90. Fragments were tested in a 2-fold dilutionseries starting at 500uM or 250uM concentrations. Steady state values were calculated4secondsbeforetheinjectionstoppedandplottedagainsttheconcentration.TheKDvaluewascalculatedbyfittingthedatatoasteadystateaffinitymodel(BiacoreT200evaluationsoftwareGEHealthcare)
![Page 60: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/60.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-31-
SUPPLEMENTARYTABLES
SupplementaryTable1
Chemicalstructuresandsummaryofresultsforthe9Hsp90NMRClass1hits.
ID Structure MW
Docking DUck
Xray
SPR
Kd
(mM)
PDB
Simb
ChEMBL
SimbScore RankaWQBRanka
1
248,72 -24,97 79 9,1 10 Yes 772XDX(0.37)
CHEMBL1340447
(0.44)
2
221,29 -25,03 73 8,2 11 Yes 3202WI6(0.29)
CHEMBL1536318
(0.54)
3
230,22 -26,62 19 11,3 1 Yes 7004EFU(0.32)
CHEMBL1458840
(0.51)
4
240,33 -26,45 22 7,4 16 - 7303WHA(0.29)
CHEMBL1542436
(0.37)
5
165,20 -23,77 128 8,1 12 - -4EFT(0.27)
CHEMBL1313412
(0.28)
6
206,27 -23,26 138 9,5 5 - -3HHU(0.42)
CHEMBL2103879
(0.42)
7
236,30 -25,45 51 7,8 15 - -3B24(0.31)
CHEMBL1375884
(0.36)
8
224,66 -25,35 58 7,0 22 - -2XDX(0.35)
CHEMBL1383799
(0.37)
22
237,29 -28,27 2 5,6 33 - -3O0I(0.27)
CHEMBL1834092
(0.33)
aPositionwithinthelistof149moleculesthatwereevaluatedwithDUck.bHsp90structureinthePDBorcompoundwithHsp90activityinChEMBL(asof23/03/2016)withtheclosestsimilaritytothefragmenthit.Similarity(valuesinparentheses)wascalculatedwithOpenBabelusingtheFP2fingerprint.14
![Page 61: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/61.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-32-
SupplementaryTable2
Number of atoms of the investigated systems. On average, using a protein chunk withexplicit solvation produces a system 20% in size relative to the whole protein. Ascomputational times scale linearlywith thenumberofparticles, this representsa5-foldgaininefficiency.
System
NumberofAtoms
FullSystem ProteinChunkforDUckProtein PeriodicBoxa Proteinb PeriodicBoxa,b
Hsp90 3291 30387 527 (16,0%) 9415 (31,0%)Cdk2 4578 46803 345 (7,5%) 9110 (19,5%)AA2AR 4603 73039 525 (11,4%) 8815 (12,1%)Trypsin 3231 26721 335 (10,4%) 9696 (36,3%)Average 3926 44238 433 (11,0%) 9259 (20,9%)
aProteinsolvatedwithTIP3watermoleculesusingAmber’stleapprogram.Inallcases,theperiodicsystemisatruncatedoctahedralbox,thedistanceparameteris12.0andtheclosenessparameteris0.65.bValuesinparenthesesarepercentageofatomsrelativetothefullsystem.
SupplementaryTable3
DetailofthereceptordefinitionusedinDUcksimulations.WaterandresiduenumbersweretakenfromthecorrespondingPDBfile.
SystemReference
Atom
PDBCode
(Chain)Proteinresiduesincludedasreceptor
Water
Molecules
CDK2 LEU 83 NH 1CKP (A)
ILE10 VAL18 LYS20 ALA21 VAL29 VAL30
ALA31 LEU32 VAL64 PHE80 GLU81 PHE82
LEU83 HIS84 GLN85 ASP86 LEU133 LEU134
ILE135 ASN136 ALA144
-
AA2AR ASN 253 ND2 3EML (A)
LEU167 PHE168 GLU169 VAL172 PRO173
MET174 MET177 VAL178 ASN181 PHE182
TRP246 LEU247 PRO248 LEU249 HIS250
ILE251 ILE252 ASN253 CYS254 PHE255
THR256 PHE257 HIS264 ALA265 PRO266
LEU267 MET270 TYR271 LEU272 ALA273
ILE274
-
Trypsin ASP189 OD1 2AYW (A)
HIS57 LEU99 ASP102 ASP189 SER190 CYS191
GLN192 GLY193 ASP194 SER195 VAL213
SER214 TRP215 GLY216 SER217 GLY219
CYS220 ALA221A GLN221 LYS224 PRO225
GLY226 VAL227 TYR228 THR229
1017 1096
1098 1101
Hsp90 ASP93 OD2 2YED (A)
GLU47 LEU48 ILE49 SER50 ASN51 SER52
SER53 ASP54 ALA55 LEU56 ASP57 LYS58
ILE78 ILE91 VAL92 ASP93 THR94 GLY95 ILE96
GLY97 MET98 GLY137 PHE138 VAL150 ILE151
THR152 GLY183 THR184 LYS185 VAL186
2043 2045
2049 2105
2107
BRD4 ASN140 ND2 3U5L (A)
TRP81 PRO82 PHE83 GLN84 GLN85 PRO86
VAL87 ASP88 ALA89 LYS91 LEU92 ASN93
LEU94 TYR97 ILE101 PRO104 MET105
THR131 ASN135 CYS136 TYR137 TYR139
ASN140 ASP144 ASP145 ILE146 MET149
-
![Page 62: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/62.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-33-
SupplementaryTable4
DatacollectionandrefinementstatisticsforHsp90incomplexwithCompounds1,2and3.RfreeistheRfactorcalculatedusing5%ofthereflectiondatachosenrandomlyandomittedfromtherefinementprocess,whereasRcrystiscalculatedwiththeremainingdatausedintherefinement.Rmsbondlengthsandanglesarethedeviationsfromidealvalues;thermsdeviationinBfactorsiscalculatedbetweencovalentlybondedatoms.
Compound 1 2 3
Datacollectionstatistics
Resolution(Ǻ) 2.20 2.00 2.10Spacegroup I222 I222 I222Celldimensions(Ǻ)a=b=c=
66.8790.2998.33
64.9688.4199.06
68.9888.1896.90
No.molecules/asymmetricunit 1 1 1Solventcontent(%) 57.25 54.73 57.41Measuredreflections 66152 66886 62479Uniquereflections 15401 19011 17526Completeness:Overall/inhrba(%) 99.5/98.5 96.7/90.9b 99.4/99.9
MeanI/σI:Overall/inhrb 11.2/2.8 11.1/1.3 8.33/0.95
Rmerge:Overall/inhrb(%) 0.083/0.315 0.048/0.412 0.074/0.555
Refinementstatistics
Rfree(%) 24.0 30.8b 27.6Rcryst(%) 19.1 22.1 22.4RmsDeviations:Bonds(Ǻ)Angles(o)BFactor(Ǻ2)
0.0181.9204.679
0.0191.9585.536
0.0192.0466.415
PDBCode 5FNC 5FND 5FNFahrb:highestresolutionbin.bDiffractiondataforthisstructurewascollectedfromacrystalthatdidnotcryo-freezecorrectlythereforethedatainsomeoftheresolutionbinswasofalesserqualitythantheequivalentdatacollectedfromtheothertwocrystals.Thisislikelytohaveimpactedtherefinementstatisticsforthisstructure.
![Page 63: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/63.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-34-
SupplementaryTable5
List of ligands in the CDK2 test set. Ligands highlighted in red are not included in thecorrelationplottedinSupplementaryFigure2andSupplementaryFigure25.
PDB No.Atoms MW IC50 (uM) Log IC50 WQB (kcal/mol)
1E1V 21 247.303 12.00 1.08 5.53
1E1X 22 251.292 1.30 0.11 3.19
1JSV 23 265.293 2.00 0.30 6.93
1JVP 19 233.274 1.60 0.20 5.59
1OIQ 23 271.325 2.90 0.46 4.98
1PF8 21 242.26 0.03 -1.51 6.88
1PXJ 16 206.267 13.00 1.11 1.76
1PXK 19 249.293 2.20 0.34 4.84
1PXM 23 298.365 0.06 -1.22 7.32
1VYW 24 291.355 0.04 -1.43 12.13
1VYZ 19 227.268 0.29 -0.54 12.19
1W0X 25 298.351 5.00 0.70 3.97
1WCC 10 129.55 350.00 2.54 3.33
2BTR 19 261.344 0.10 -1.02 6.50
2BTS 24 300.417 0.02 -1.70 8.94
2C4G 22 270.294 1.15 0.06 13.18
2C5O 17 207.275 6.50 0.81 3.12
2CLX 21 218.221 3.50 0.54 6.94
2EXM 17 203.249 78.00 1.89 7.84
2R3H 19 239.282 20.00 1.30 4.18
2VTA 10 118.139 185.00 2.27 6.05
2VTH 18 223.249 120.00 2.08 1.97
2VTJ 22 286.739 1.90 0.28 1.68
2VTL 16 187.203 97.00 1.99 7.89
2VTM 11 144.137 1000.00 3.00 1.68
2VTN 22 262.246 0.85 -0.07 9.11
2VTR 16 234.67 1.50 0.18 5.34
3BHT 20 241.255 0.01 -1.96 6.28
3BHV 26 293.291 0.08 -1.10 7.30
3EJ1 20 252.281 0.12 -0.92 6.41
3FZ1 24 278.352 0.15 -0.84 0.12
3PXY 22 233.233 5.90 0.77 3.66
3QQK 21 259.328 15.00 1.18 7.87
3QTQ 21 262.332 3.10 0.49 6.67
3QTR 24 295.361 0.93 -0.03 10.90
3QTW 24 296.349 0.65 -0.19 11.51
3R8Z 21 262.332 49.00 1.69 6.57
3RZB 20 236.294 100.00 2.00 9.31
3TIY 20 220.185 17.00 1.23 3.38
3TIZ 23 265.314 150.00 2.18 2.23
4EZ3 25 296.305 45.00 1.65 2.33
![Page 64: Dynamic undocking and the quasi-bound state as tools for ...eprints.whiterose.ac.uk/110037/1/NChem_2016_prepint.pdf · Dynamic undocking and the quasi-bound state as tools for drug](https://reader031.vdocuments.us/reader031/viewer/2022022521/5b2411a37f8b9aff278b488c/html5/thumbnails/64.jpg)
SupplementaryInformation–DynamicUndockingandtheQuasi-BoundState
-35-
SupplementaryTable6
List of ligands in the BRD4 test set. Ligands highlighted in red are not included in thecorrelationplottedinSupplementaryFigure5andSupplementaryFigure26.
PDB No.Atoms MW IC50 or Kd (nM) Log IC50 WQB (kcal/mol)
3MXF 31 458.00 49.00 -1.31 6.63
3U5J 22 308.77 2460.00 0.39 7.00
3U5L 23 323.78 640.00 -0.19 9.12
4A9L 22 325.38 30000.00 1.48 4.78
4C66 23 343.85 79400.00 1.90 3.92
4CFK 23 307.35 1830.00 0.26 3.13
4CFL 23 306.36 1330.00 0.12 3.63
4E96 24 347.39 136.00 -0.87 4.51
4HBV 13 241.09 23000.00 1.36 2.51
4HBW 18 269.32 4800.00 0.68 5.98
4HBX 20 295.36 1900.00 0.28 5.53
4HBY 22 317.36 4400.00 0.64 5.42
4HXR 21 338.41 4100.00 0.61 6.54
4HXS 23 346.42 4100.00 0.61 4.90
4J0R 22 295.34 386.00 -0.41 7.70
4J0S 22 295.34 382.00 -0.42 7.10
4LR6 13 174.20 33000.00 1.52 3.85
4LZS 15 208.26 16000.00 1.20 2.84
4MEN 20 267.33 125000.00 2.10 3.84
4MEO 22 292.34 250000.00 2.40 5.73
4MEQ 17 225.25 250000.00 2.40 4.62
4O72 30 413.49 1000.00 0.00 9.00
4O74 38 521.66 25.00 -1.60 1.67
4O77 25 331.35 2500.00 0.40 1.10
4O78 30 406.44 4600.00 0.66 4.07
4O7A 23 349.17 19000.00 1.28 3.72
4O7E 24 313.36 5700.00 0.76 0.00
4PCE 19 253.34 7000.00 0.85 4.30
4PCI 19 252.31 7500.00 0.88 4.84
4UYD 15 205.22 79400.00 1.90 4.03