Quaternary structure predictions and estimation of mutational effects on the free energy of dimeriza

JournalofStructuralBiologyjournalhomepage:www.elsevier.com/locate/yjsbiQuaternarystructurepredictionsandestimationofmutationaleffectsonthefreeenergyofdimerizationoftheOMPLAprotein

DanieleDell’Orco,DanieleCasciari,FrancescaFanelli*DepartmentofChemistry,DulbeccoTelethonInstitute(DTI),UniversityofModenaeReggioEmilia,ViaCampi183,41100Modena,Italyarticleinfoabstract

ThisstudyrepresentsanextensiontotheoutermembranephospholipaseAprotein(OMPLA)ofthedock-ing-basedprotocolspreviouslydevelopedforquaternarystructurepredictionsoftransmembraneoligo-mericproteinsandforestimatingmutationaleffectsonthethermodynamicsofprotein–proteinandprotein–DNAassociation.PredictionsofthelikelyarchitectureofOMPLAhomo-dimerswerecarriedouton31differentformsofthemonomer,30ofwhichwerevariantsoftheunboundstate.Inallthetestcasesbuttheonescharac-terizedbycombineddeletionsofthe98–110and145–153segments(L2andL3,respectively),native-likecomplexescouldbepredicted,independentoftheboundorunboundstateofthestructuralmodel,ofsidechainconformationandpresenceorabsenceofaminoaciddeletionsattheputativeinter-monomerinterface.Theprotocolforestimatingmutationaleffectsonthethermodynamicsofprotein–proteinassociationprovedeffectiveaswell.Infact,itwaspossibletoestimatecorrectlytheeffectsoffivemutantsonthefreeenergyofdimerizationofthesulfonylatedformofOMPLA.TheintegrityofL2andeitheroneoftheL1,L3andL4loopsturnedouttobemoreimportantthansulf-onylationfortheachievementofthenativedimericarchitecture.Ontheotherhand,sulfonylationseemstobeessentialforafavorabledimerizationenergetics.Ó2008ElsevierInc.Allrightsreserved.Articlehistory:Received5February2008Receivedinrevisedform1May2008Accepted2May2008Availableonline23May2008Keywords:MolecularrecognitionRigid-bodydockingZDOCKComputationalmodeling1.IntroductionIncreasingevidencehighlightsproteindimerizationasabiolog-icalmechanismemployedbybothsolubleandmembraneproteinstocarryoutmanyregulationtasks.Nevertheless,therearecur-rentlynotmanystructuraldataonthefactorsthatgoverndimer-izationinmembraneproteins.OneofthefewexceptionsisrepresentedbytheoutermembranephospholipaseA(OMPLA),abacterialintegralmembraneenzyme,whichhydrolyzesthefattyacidacyl-esterbondsofawidevarietyofphospholipids(ScandellaandKornberg,1971),andmorerecentlyhasbeenfoundtobeimplicatedinvirulence(Bukholmetal.,1997;Grantetal.,1997).TheactivityofOMPLAisregulatedbyamechanismofreversibledimerization(Dekkeretal.,1999,1997).Indeed,thecrystalstruc-tureofdimericOMPLArevealedthatthe12strandedb-barrelfoldedproteincanachievecompleteactivesitesonlyupondimer-ization(Fig.1,Snijderetal.,1999).Ithasbeenalsoobservedthatcalciumbindingatthedimerinterfaceisnecessaryforcatalysis(Snijderetal.,2001a)butnotforthestabilityofthedimer(Stanleyetal.,2006).Furthermore,dimerizationleadstotheformationofsubstrate-bindingcleftsattheouteredgeofthedimerinterface,*Correspondingauthor.Fax:+390593733.E-mailaddress:fanelli@unimo.it(F.Fanelli).1047-8477/$-seefrontmatterÓ2008ElsevierInc.Allrightsreserved.doi:10.1016/j.jsb.2008.05.006withthesubstrateacylchainmakingextensivecontactswithbothmonomers(Snijderetal.,1999).Indeed,ithasbeenfoundthathexadecylsulfonylfluoride(HSF),asubstrateanalogthatbindsspecificallyandcovalentlytoSer144leadingtothehexadecylsulf-onylated(HDS)formoftheprotein,actsasanactive-site-directedinhibitorforOMPLA(Horrevoetsetal.,1991)andgreatlycontrib-utestothefreeenergyofproteindimerization(Stanleyetal.,2006).Inthisrespect,sedimentationequilibriumanalyticalultra-centrifugationledtotheestimateofanapparentfreeenergyofdimerization(DG)forthesulfonylatedwild-type(WT)ofÀ7.25kcal/molevenintheabsenceofcalcium,whereasthecom-pletelyunmodifiedOMPLAwasfoundonlyinitsmonomericform,atleastwithintheexperimentallimitsoftheapproach(Stanleyetal.,2006).Thehigh-resolutioninformationarisingfromtheX-raycrystal-lographicstructuresofbothmonomericanddimericOMPLA(Snij-deretal.,2001a,1999)makestheproteinauniquetest-systemtoinvestigatetheforcesinvolvedinprotein–proteininteractionsinvolvingb-barrelfoldedtransmembraneproteins.Moreover,ther-modynamicstudiesandsite-directedmutagenesisexperimentstargetingkeyresiduesintheOMPLAdimerinterfacerevealedvalu-ableinformationonthedimerizationenergetics,whichcannowbeemployedasausefulcomplementtostructuraldata(StanleyandFleming,2007;Stanleyetal.,2006).156D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162Fig.1.CrystalstructureoftheOMPLAdimer(PDBcode1QD6,Snijderetal.,1999).Twoviewsofthedimerareshown,seeninadirectionparallel(top)andperpendicular(bottom)totheputativemembranesurface.ThesidechainsofY92,Q94andS96,whicharethetargetofinsilicomutagenesis,areshownasredspheres.ThetwoHDSchains,onefromeachmonomer,areshownasgrayspheres.Finally,calciumions,onefromeachmonomer,arerepresentedasbluespheres.(Forinterpretationofthereferencestocolorinthisfigurelegend,thereaderisreferredtothewebversionofthispaper.)Inthisstudy,wehaveextendedtotheOMPLAproteinthepro-tocolspreviouslydevelopedforquaternarystructurepredictionsofa-helicaltransmembraneoligomericproteins(Casciarietal.,2006)andforestimatingmutationaleffectsonthethermodynamicsofprotein–protein(Dell’Orcoetal.,2007a,b,2005)andprotein–DNAassociation(FanelliandFerrari,2006).DockingsimulationshavebeencarriedoutbothintheabsenceandinthepresenceofcalciumionsandsulfonylationtoinvestigatetheirstructuralandenergeticrolesinOMPLAdimerization.Onthesameline,deletionsofselectedaminoacidstretchesattheinter-monomerinterfacehavebeendonetoinferthekeyproteinpor-tionsthatdictatethenativedimericarchitecture.2.Methods2.1.ComputationalprotocolforquaternarystructurepredictionsRigid-bodydockingsimulationswerecarriedoutbymeansoftherigid-bodydockingalgorithmZDOCK2.1(ChenandWeng,2003).Thegrid-basedpairwisescoringfunctionimplementedinthisversionofthesoftwareaccountsforshapeandelectrostaticcomplementaritiesneglectingthedesolvationterm,whichissuit-ableforaqueousenvironments(ChenandWeng,2003).ThePair-wiseShapeComplementarity(PSC)scoringfunctionisnotexplicitlybaseduponproteinsurfacecurvatureorsurfacearea,butitrathercomputesthetotalnumberoftarget–ligandatompairswithinadistancecutoffminusaclashpenalty.Suchproce-dureenhancespairwisetarget–ligandcontactsand,byminimizingphysicaldiscontinuity,notablydecreasesthenumberoffalsepos-itivesinthedockingoutput.Theclashpenaltyinsteadaccountsforpotentialbadcontactsthatmayarisefromtherigid-bodyap-proach.Toaccountpartiallyfordifferentatomradii,thedistancecutoffisdefinedasaparameterDplusthereceptoratomradius.TheunfavorablecomponentofPSCislinearlyproportionaltothenumberofoverlappinggridpointsbetweenthereceptorandtheligand.ThedockingalgorithmperformsanefficientFastFourierTrans-formsamplingoftheentiresixdegreesoffreedominthetransla-tionalandrotationalspace,andfinallyprovidesascoreforeachdockingsolution,whichisrepresentativeofthecomplementarityofthecomplex.SimulationsaimedatpredictingthelikelyarchitectureintheOMPLAhomo-dimerfollowedacomputationalprotocolrecentlyproposed(Casciarietal.,2006).Indetail,simulationswerecarriedoutontheWTformextractedfromthe1QD6crystalstructureofthedimerat2.1Åresolution(i.e.abound–bounddockingcase),aswellason30differentformsoftheWTinitsunboundforms(i.e.unbound–unbounddockingcases).Theunboundformsofthemonomerincluded:(a)theoriginalstructuralmodelsoftheWTencodedas1QD5(Snijderetal.,1999)and1FW2(Snijderetal.,2001a),aswellasoftheN156Amutantencodedas1ILD,1IM0and1ILZ(Snijderetal.,2001b);allthesestructures,differentfromthedimeric1QD6,donotmissthe26–30segment;(b)atruncatedformoftheunboundmonomer1FW2(Trunc1,Table1),lackingthe26–30aminoacidstretch;and(c)threevariantsoftheoriginal1FW2andthreevariantsofTrunc1,inwhichanysidechainwassubjectedtorotationsaccordingtothelibrariesdevelopedbyDunbrackandKarplus(D&K)(DunbrackandKar-plus,1993),PonderandRichards(P&R)(PonderandRichards,1987)andSutcliffeandcoworkers(Sut)(Sutcliffeetal.,1987)(Table1).ThirteendifferentformsoftheTrunc1variantwerealsoprobed(Trunc2–14),whichwerecharacterizedbybackboneorsidechain-onlydeletionofthe–59(L1),98–110(L2),145–152(L3)and231–235(L4)aminoacidstretches,takeneithersingularlyorincombination(Table2).Deletionsoftheseloopregions,whichallparticipateintheinterface(Fig.2,Table2),wereinstrumentalininvestigatingtheirroleinOMPLAdimerization.Twoidenticalcopiesofthemonomerswereemployed,onewaskeptfixed(target),whereastheotherwasallowedtorotateandtranslatearoundthetarget(probe).Thebest4000solutionsfromeachrunwereretainedandrankedaccordingtotheZDOCKscore(ZD-score).Thesesolutionsweresubjectedtoafilter,the‘‘mem-branetopology”filter,whichdiscardsallthesolutionsthatviolatethemembranetopologyrequirements.Indetail,thefilterdiscardsallthesolutionscharacterizedbyadeviationanglefromtheorigi-nalz-axis(tiltangle)andadisplacementofthegeometricalcenteralongthez-axis(z-offset)abovedefinedthresholdvalues.Forthetiltangleandthez-offset,thresholdsof0.4radiansand6.0Åwere,respectively,employed.Thefilteredsolutionsfromeachrunweremergedwiththetargetprotein,leadingtoanequivalentnumberofdimersthatweresubjectedtoclusteranalysis.ClusteranalysiswasbasedontheQTclusteringalgorithm(Heyeretal.,1999)implementedintheFIPDsoftware(Casciarietal.,2006).Inthiscasestudy,thealgorithmfirstcalculatedthea-carbonatomRootMeanSquareDeviation(Ca-RMSD)foreachsuperimposedpairofdimers/oligomersandthenitcomputedthenumberofneighborsforeachdimer/oligomerbyusingathresholdCa-RMSD.Thedi-mer/oligomerwiththehighestnumberofneighborsisconsideredasthecenterofthefirstcluster.Alltheneighborsofthisconfigu-rationareremovedfromtheensembleofconfigurationstobecountedonlyonce.Thecenterofthesecondclusteristhendeter-D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162Table1

DockingresultsforOMPLAdimerizationPDBa1QD61QD51ILD1IM01ILZ1FW2ModelbOriginalOriginalOriginalOriginalOriginalOriginalOrig-P&ROrig-D&KOrig-SutTrunc1Trunc1-P&RTrunc1-D&KTrunc1-SutRes.c2.102.172.802.982.502.60Fsold968472887290757210210413467135NCe1012121197111191314914Cpopf37302527253928363427242833MTbestg0.1350.4080.4760.7660.4760.4140.4220.2790.4600.4140.6290.2790.498Bestsolh38122251371132834ZDbesti24.5823.5822.2825.7021.1822.7423.8825.29.8822.4821.8222.5224.74157RMSDj0.711.782.221.982.191.321.311.072.031.621.681.212.04ProteinDataBankcodeofthestructuralmodelemployedasamonomer.Molecularformofthemonomer.Theword‘Original’anditsabbreviation,‘Orig’,referstothecrystalstructurenotsubjectedtoanymodification.Theword‘‘Trunc1”meansthatthe26–30aminostretchismissing.cResolution(Å)ofthestructuralmodel.dNumberoffilteredsolutions.eNumberofclusters.fNumberofsolutionsinthemostpopulatedclusterthatcoincideswithcluster1.gMemTopindexofthebestscoredsolutionfromthemostpopulatedcluster.hRankorder,accordingtotheZD-score,ofthebestscoredsolutionfromthemostpopulatedcluster.iZD-scoreofthebestscoredsolutionfromthemostpopulatedcluster.jCa-RMSD(Å)betweenexperimental(1QD6)andpredicteddimer.baTable2

Dockingresultsforthetruncatedformsofthe1FW2structureModelaTrunc2Trunc3Trunc4Trunc5Trunc6Trunc7Trunc8Trunc9Trunc10Trunc11Trunc12Trunc13Trunc14Trunc9hdsTrunc10hdsTrunc11hdsTrunc13hdsTrunc14hdsaDeletionbL1L2L3L4L2-scL3-scL2+L3-scL2+L3L1+L2L1+L3L1+L4L2+L4L3+L4L2+L3L1+L2L1+L3L2+L4L3+L4Fsolc114661169096787393681061069177616963NCd1312913812129147111191098108Cpope341516281633201316192714231916192120MTbestf0.1350.70.6070.4601.066*0.4140.456———0.456————0.408——Bestsolg425623521182*12221———57————19——ZDbesth21.5017.9418.1020.7018.16*22.0218.04———19.30————18.94——RMSDi1.502.702.502.313.00*1.662.28———2.03————1.00——Molecularvariantsofthe1FW2unboundcrystalstructure.Indetail:(a)Trunc2ischaracterizedbythedeletionofL1(–59segment);(b)Trunc3ischaracterizedbydeletionofL2(98–110segment);(c)Trunc4ischaracterizedbydeletionofL3(145–152segment);(d)Trunc5ischaracterizedbydeletionofL4(231–235segment);(e)Trunc6ischaracterizedbysidechain-onlydeletioninL2;(f)Trunc7ischaracterizedbysidechain-onlydeletioninL3;(g)Trunc8ischaracterizedbysidechain-onlydeletioninbothL2andL3;(h)Trunc9–14arecharacterizedbyallthepossiblepairwisetruncationsofL1,L2,L3andL4;(i)sulfonylatedformsofTrunc9–11andTrunc13,14.bLoopdeletioncharacterizingthe1FW2variants;‘‘sc”standsfor‘‘sidechain-only”deletions.cNumberoffilteredsolutions.dNumberofclusters.eNumberofsolutionsinthemostpopulatedcluster,thatcoincideswithcluster1inallthecasestudies,exceptfortheTrunc9case,inwhichthefirstandsecondclustersareequipopulated.fMemTopindexofthebestscoredsolutionfromthemostpopulatedcluster(Cluster1).Onlydataconcerningnative-likesolutions(RMSD62.5Å)ornear-native-likesolutions((RMSD63.0Å)areshown.Theasterisksmeanthatthebestsolutionhasbeenextractedfromthesecondclusterthathasapopulationsimilartothatoffirstone(15solutions).Thesymbol‘‘—”meansthatnosolutionwithaRMSD67Åcouldbefoundinanyofthefirsttwomostpopulatedclusters.gRankorder,accordingtotheZD-score,ofthebestscoredsolutionfromthemostpopulatedcluster.Forthedefinitionoftheasterisks,seeaboveatpoint(f).hZD-scoreofthebestscoredsolutionfromthemostpopulatedcluster.Forthedefinitionoftheasterisks,seeaboveatpoint(f).iCa-RMSD(Å)betweenexperimental(1QD6)andpredicteddimer.Forthedefinitionoftheasterisks,seeaboveatpoint(f).minedinthesamewayasforthefirstcluster,andthisprocedureisrepeateduntileachstructureisassignedtoacluster.TheCa-RMSDthresholdforeachpairofsuperimposeddimerswassetequalto3.0Å.AlltheaminoacidresiduesinthedimerwereincludedinCa-RMSDcalculations.Toidentifytheclusterofsolutionswiththebestmembranetopology,namelythosecharacterizedbythelowestvaluesofbothtiltandz-offset,wedefinedtheMemTopindex,qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiMemTop¼htiltnori2þhZoffnori2,wherehtiltnoriandhZoffnoriare,respectively,thenormalizedtiltangleandthez-offsetaveragedoverallthemembersofagivencluster.Normalizationofeachtiltangleandz-offsetvaluewascarriedoutbydividingeachvaluefortherespectivecutoffvalue,namely0.4radians,forthetiltangle,angle158D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–1622.2.ComputationalprotocolforassociationfreeenergyestimationsDockingsimulations(bymeansoftheZDOCKprogram)aimedatestimatingmutationaleffectsontheOMPLAdimerizationfreeenergyfollowedanapproachextensivelyprobedonprotein–pro-tein(Dell’Orcoetal.,2007a,b,2005)andprotein–DNAassociation(FanelliandFerrari,2006).Inthiscasestudy,simulationswerecar-riedoutontheWTandfivemutatedformsofOMPLAextractedfromthecrystalstructureofthe1QD6dimer.Mutationsinvolvethreeresiduesatthedimerinterface,namelyY92mutatedinF,S96mutatedinA,andQ94mutatedinA,NandE(Table3andFig.1).AminoacidreplacementsweredoneonboththeCandDchainsinthecrystallographicdimer,withnofurtherenergymini-mization.TheHDSchain,covalentlylinkedtoS144inboththeCandDmonomers,wasincludedinthedockingsimulationsoftheWTandthefivemutants.ThesamesetofsixsulfonylatedOMPLAvariantsweresimulatedinthepresenceofboundcalciumions,oneforeachmonomer.Furthermore,onlyfortheWT,dockingsimula-tionswereadditionallycarriedoutonthenon-sulfonylatedformbothinthepresenceandintheabsenceofcalciumions.Theinclu-sionoftheHDSchainandofthecalciumionsrequiredaparame-trization,consistingintheadditionoftherelativeatomicradiiandCHARMM22atomiccharges,whichweremissingintheZDOCKparameterset.Threeindependentsetsofdockingrunswereper-formedforeachcomplex,onestartingfromtheexperimentalcoor-dinatesandtheothertworandomizingtheinitialpositionsoftheprobe,inafashionpreviouslydescribed(Dell’Orcoetal.,2007a,b).Ineachdockingrun,theWTormutatedOMPLAmonomercorre-spondingtotheCchainfromthedimericcomplexwasemployedasatarget,whereastheonecorrespondingtotheDchainservedasaprobe.Fromeachdockingrun,thebest4000solutions,rankedaccordingtotheZD-score,wereretained.Asfordockinganalysis,accordingtopreviousworks(Dell’Orcoetal.,2007a,b,2005),weselectedasnative-likestructuresallthedockedcomplexescharacterizedbyaCa-RMSDlowerthan1.0Åfromthenativedimer.ItisworthnotingthatsuchaCa-RMSDva-lueislowerthanthe2.5ÅCa-RMSDcutoffemployedfordefiningthenative-likesolutionsarisingfromquaternarystructurepredic-tions(Casciarietal.,2006).TheemploymentofastrictCa-RMSDcutoffintheprotocolforfreeenergyestimationswasdictatedbytheneedtoachieveahighlyhomogeneousclusterofnative-likesolutionsforthecomputationsoftheaverageZD-score(ZDavg).In-deed,foreachmolecularsystem,weconsideredtheZD-scoreaver-agedoverthescoresofallthenative-likecomplexesresultingfromFig.2.The1FW2crystalstructureisshown,coloredingray.Orange,red,green,purpleandbluecolorsindicate,respectively,the26–30,–59,98–110,145–152and231–235aminoacidsegments,whichallparticipateintheinter-monomerinterfaceandhavebeendeletedeithersingularlyorincombinationintheTrunc1–14forms.(Forinterpretationofthereferencestocolorinthisfigurelegend,thereaderisreferredtothewebversionofthispaper.)and6.0Å,forthez-offset.Theoptimalvalueforsuchindexiszero,whereasthemaximumvalueistherootsquareof2.Similartoourpreviousstudy(Casciarietal.,2006),native-likesolutionswereconsideredthosecharacterizedbyCa-RMSDsfromthenativecomplex62.5Åafteroptimalsuperimposition.Thiscut-offisthesameastheoneemployedtoidentifyZDOCKhits(ChenandWeng,2003;Chenetal.,2003)withthedifferencethat,inourapproach,optimalsuperimpositionandRMSDcalculationsin-volvealltheCa-atomsinthecomplexinsteadoftheinterfaceCa-atomsonly.Table3

InvitroandinsilicodataconcerningOMPLAdimerizationVariantaWTY92FS96AQ94AQ94NQ94EUnmod-WTaDGexp1bDGpredcÀ7.32À7.08À7.23À6.À6.72À5.12>0ZDavg1d26.8326.6926.8726.4026.4825.8820.99Bsol1e1111111DGexp2fÀ8.31±0.25À8.34±0.21À8.±0.23À7.83±0.05À8.02±0.09À8.14±0.40À5.68±0.14ZDavg2g27.4526.9126.9326.3526.6527.4722.34Bsol2h1111111À7.25±0.23À7.00±0.07À7.63±0.21À6.30±0.23À6.67±0.07À5.86±0.22(Monomer)WTandmutatedformsofOMPLA.Invitroandinsilicodetermineddatainthefirstsixrowsrefertothesulfonylatedforms,whereasthelastrowreferstotheunmodifiedWT(StanleyandFleming,2007).bInvitrodeterminedfreeenergiesofassociation(kcal/mol)intheabsenceofCaCl2(StanleyandFleming,2007).cEstimatedfreeenergiesofassociation(kcal/mol)baseduponthecross-validationtestandconcerningthesulfonylatedformswithoutcalcium;for‘‘Unmod-WT”DGpredictionhasbeenbaseduponthelinearcorrelationequationinFig.4A.dAverageZD-scorecomputedoverthemembersofthenative-likeclusterforthesulfonylatedformswithoutcalcium.eRankorderofthebestscoredsolutioninthenative-likeclusterintheabsenceofcalcium.Forthesulfonylatedforms,thetophitisfoundasthebestscoredonefromanyofthethreeparalleldockingruns,whereasfortheUnmod-WTthetophitisfoundfromonlyonerun.fInvitrodeterminedfreeenergiesofassociation(kcal/mol)inthepresenceofcalcium(StanleyandFleming,2007).gAverageZD-scorecomputedoverthemembersofthenative-likeclusterforthesulfonylatedformsinthepresenceofcalcium.hRankorderofthebestscoredsolutioninthenative-likeclusterinthepresenceofcalcium.Forthesulfonylatedforms,thetophitisfoundasthebestscoredonefromanyofthethreeparalleldockingruns,whereasfortheUnmod-WTthetophitisfoundfromonlyoneofthethreeruns.D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162159thethreeindependentdockingrunsandconstitutingthenative-likeensemble.Theseaveragescoreswerethenemployedinthecorrelationanalysiswiththeinvitrodetermineddimerizationfreeenergies(Table3).Inallthedockingsimulationscarriedoutinthisstudy,a128Â128Â128pointgridwitha1.2Åspacingwasusedfordig-italizingtheinteractingmolecules.Moreover,arotationalsam-plingintervalof6°wasemployed.3.Results3.1.PredictionsoftheOMPLAdimericarchitectureInthisstudy,wehaveextendedtoab-strandedtransmembraneproteintheprotocolconceivedforquaternarystructurespredic-tionsofa-helicaltransmembraneproteins(Casciarietal.,2006).Suchacomputationalprotocolconsistsofanumberofdensedock-ingsamplings,startingfromthedockingoftwoidenticalcopiesofagivenmonomer(Casciarietal.,2006).PredictionsofthelikelyarchitectureofOMPLAhomo-dimershavebeencarriedoutbyprobing31differentformsofthemonomer,30ofwhicharevari-antsoftheunboundstate,i.e.unbound–unbounddockingcases(seeSection2fordetails).Eachstructuralmodelofthemonomerhasbeensubjectedtothepredictionprotocolconsistinginarunofdockingsimulationsfollowedbymembranetopology-basedfil-teringandclusteranalysis.TheresultsofdockinganalysisaresummarizedinTables1and2.ByconsideringalltogethertheresultsofdockingrunsreportedinTable1,filteringsavedonaverage2.3%ofthebest4000solu-tionsprovidedbyeachrun.Onaverage,33%ofthefilteredsolu-tionsgroupinonecluster,whichissignificantlymorepopulatedthantheothers,theaveragenumberofclustersperrunbeing10.92(Table1).ThebestscoredsolutionfromthisclusterhasagoodMemTopindex(almostalwayslowerthan0.5)andfallswith-inthetop30solutionsoutof4000(Table1).Inmostcasesitfallswithinthetop10solutions.Thus,foreachdockingrun,apredicteddimerhasbeenachievedbyemployingthebestscoredsolutionfromthemostpopulatedcluster(Table1).Allthepredicteddimersholdnative-likeintermolecularcontacts,theCa-RMSDbetweennative1QD6structure(Snijderetal.,1999)andpredictedcomplexbeingsignificantlylowerthan2.5Å.Asexpected,thebound–boundcaseresultsinthelowestCa-RMSD.ThegoodnessofthefitbetweennativeandpredictedstructuresforthreecasestudiescanbevisualizedinFig.3,whichshowsthatthedivergencesinRMSDessentiallylayattheinterfaceinvolvingL1,L2andL3.Itisworthnotingthatthenative-likesolutionswereachievedintheabsenceoftheHDSchain,indicatingthatthestructuraldeterminantsfortheachievementofthenativedimerareheldbytheinteractingmonomers.Thus,toinvestigatewhichproteinregionisneededforZDOCKtopredicttheproperdimericarchitec-ture,13differentvariantsofthetruncatedformof1FW2(Trunc1)wereprobed,characterizedbywholeorsidechain-onlydeletionsoftheL1,L2,L3andL4segments(seeTable2fordetails),whichcontributetothedimerinterfaceinthe1QD6crystalstructure(Fig.2).TheresultsofdockingsimulationssuggestthatdeletionsofL1,L3orL4donotpreventZDOCKfrompredictinganative-likearchitecture.Infact,foralltheTrunc2,Trunc4andTrunc5formstheCa-RMSDbetweennativeandpredicteddimeris62.5Å(Table2).Amongthesethreecases,theworstpredictionintermsofZD-scoreconcernsdeletionofL3(Trunc4,Table2).DeletionofL2,leadingtotheTrunc3form(Table2),resultsinapredictioncharac-terizedbyaCa-RMSDslightlyhigherthanthecutoffof2.5Å(Table2).Alsothesidechain-onlytruncationofL2,intheTrunc6form,failsinpredictinganative-likecomplex,thebestsolutionfromthemostpopulatedclusterholdingaCa-RMSDof3.00Å.Collec-Fig.3.Superimpositionbetweennative(green)andpredicted(blue)dimers.Thenativestructureisthe1QD6crystalstructure(Snijderetal.,1999).Thetop,middleandbottompanelsrefer,respectively,tothemonomerextractedfromthe1QD6structure(Ca-RMSD=0.71Å),thetruncatedformof1FW2(Ca-RMSD=1.62Å)andtheoriginalformof1FW2(Ca-RMSD=1.32Å).(Forinterpretationofthereferencestocolorinthisfigurelegend,thereaderisreferredtothewebversionofthispaper.)tively,singleloopdeletionsemphasizetheimportantroleofL2and,byalesserextent,ofL3indictatingthedimerarchitecture.Wethensearchedfortheminimalpairwiseloopdeletionthatcouldimpedenative-likepredictions.AllthepossiblepairwisedeletionsofthefourloopsexceptfortheL1+L4one(Trunc12,Ta-ble2)failedinfindinganative-likeinthefirstmostpopulatedcluster(sign‘‘—”inTable2).Forthesetruncatedforms,nosolutionwithaCa-RMSDlowerthan7.0Åcouldbefoundinanysolutioncluster.ToinvestigatewhetherthepresenceofsulfonylationcouldmaketheTrunc9–11andTrunc13–14formsabletoproducena-160D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162tive-likedimers,dockingsimulationswerealsocarriedoutinthepresenceoftheHDSmolecule.Inallthecasesexceptforsulfony-latedTrunc11,thecomputationalapproachwasunabletopredictnative-likesolutions(sign‘‘—”inTable2).Collectively,thesere-sultssuggestthattheintegrityofalmostallthepossiblepairwisecombinationsofL2,L3andL4andnotsulfonylationareessentialforZDOCKtopredictthenativedimericarchitecture.Thebackboneconformationoftheseloopsdoesnotchangeongoingfromtheun-boundtotheboundstates(seethecrystalstructures).3.2.PredictionsofmutationaleffectsonthefreeenergyofOMPLAdimerizationInthisstudy,wehavealsoemployedthecomputationalpro-tocolrecentlydevelopedforestimatingmutationaleffectsonthethermodynamicsandkineticsofassociationofwater-solubleproteins(Dell’Orcoetal.,2007b,2005),transmembranea-helicalproteins(Dell’Orcoetal.,2007a),andprotein–DNAsystems(FanelliandFerrari,2006).Theonlyrequirementoftheapproachistheavailabilityofthecrystalstructureoftheprotein–proteinorprotein–DNAcomplex.Briefly,theapproachconsistsinper-formingthreeparalleldockingrunsforeachWTandmutatedform,bycollectingatotalnumberof12,000solutions(4000solutionsfromeachrun).Amongsttheseoutputsolutions,theclusterofnative-likesolutionsisthenindividuated,whichcol-lectsallthesolutionsshowingaCa-RMSDfromtheexperimen-talcomplexlowerthan1Å.Thenative-likeclusterisinstrumentalincomputingtheZDavg(seeMethods).Asdiscussedelsewhere(Dell’Orcoetal.,2007a,b,2005;FanelliandFerrari,2006),theemploymentofaverages,insteadofsinglevaluesovercomes,atleastinpart,thelow-resolutionofthecomplexesinvolvingthemodeledmutants,bytakingintoaccountalsopos-sibleslightchangesinthebindingmodesofthemutantstruc-turescomparedtotheWT.TheZDavgisthenemployedincorrelativeanalyseswithinvitrodeterminedthermodynamicorkineticconstants.ThreeparallelrunsofdockingsimulationshavebeencarriedoutontheWTandfivemutatedformsofOMPLAextractedfromthe1QD6crystalstructureofthedimer(Snijderetal.,1999).Mutationsinvolvethreeaminoacidresiduesatthedimerinterface,namelyY92mutatedinF,Q94mutatedinA,NandEandS96mutatedinA(Table3).IntheWTstructure:(a)Y92fromonemonomerisinvolvedinvanderWaalsinterac-tionswithF68fromtheothermonomer;(b)Q94fromonemonomerisinvolvedintwoH-bondinginteractionswithQ94fromtheothermonomerand(c)S96isnotinvolvedinrelevantinter-monomerinteractions(Fig.1).TheHDSinhibitoriscova-lentlyboundtoS144fromeachmonomer.Its16-carbonatomchainisthebiggestcontributortotheinter-monomerinterface,beinginvolvedinvanderWaalsinteractionswithL40,F69,L71,W98,F109,V263andL265.Thechoiceofthesemutantsisduetotheveryrecentavailabilityofinvitroestimatedfreeenergiesofdimerization(DGexp)bothintheabsenceandpres-enceofsulfonylationatS144andofcalciumions(StanleyandFleming,2007).Asexpectedbasedupontheimportantinvolve-mentofHDSintheinterface,theabsenceofsulfonylationcou-pledwiththeabsenceofboundcalciumionsimpairsdimerizationforalltheconsideredforms(StanleyandFleming,2007).Inthesulfonylatedprotein,thethreedifferentreplace-mentsofQ94,whichareexpectedtobreakthenativeinter-monomerH-bonds,arecharacterizedbyareductionofthedimerizationfreeenergy.TheQ94Emutationisthemostdetri-mentalinthisrespect,presumablyduetotherepulsiveelectro-staticeffectofthetwoapproachingnegativecharges.Incontrast,theconservativemutationofY92tophenylalanine,ortheala-ninesubstitutionforS96,whichisnotstronglyinvolvedintheinterface,donotaffectthedimerizationfreeenergy(Table3,StanleyandFleming,2007).ThepresenceofboundcalciuminthesulfonylatedformsoftheWTandmutantssignificantlyreducesthevariationsinDGexp,thusmaskingtheeffectofmutationscomparedtothecalcium-freeforms(Table3).Infact,fortheWTandthefivemutantsconsideredinthiswork,theDGexpvariationinthepresenceandintheabsenceofcalciumis0.81kcal/moland1.77kcal/mol,respectively(Table3,StanleyandFleming,2007).TheresultsofdockingsimulationsconcerningthesulfonylatedformsoftheWTandthefiveOMPLAmutantsintheabsenceandpresenceofboundcalciumionsarereportedinTable3.Inallthetestedcases,theclusterofnative-likesolutionscontainsthesolu-tionN1,namelythetophitoutofthe4000solutionsprovidedbyeachrun(Table3).Forthedatasetofsulfonylatedformsintheabsenceofcalcium,agoodcorrelationequationbetweenDGexp1andZDavg1wasob-tained(R=0.96,N=6;Fig.4A).Moreover,inspiteofthefewdatapoints,thecross-validationtest(Table3andFig.4B)showsagoodpredictivepower(R=0.88).Incontrast,aworsecorrelationwasachievedinthepresenceofcalcium(R=0.65,N=5).Thisisdue,atleastinpart,tothesignificantlylowervariationofDGexp2com-paredtoDGexp1(Table3).Collectively,thecalciumeffectofimprovingthedimerizationfreeenergyofalltheOMPLAvariantscomparedtothecalcium-freeforms,isaccountedforbythedock-ingscore,thoughinanunderestimatedmanner(Table3).Underes-timationofcalciumeffectsmaybedue,atleastinpart,tothenecessaryneglectofdesolvationintheemployedversionofthedockingalgorithm.Desolvationis,indeed,inconsistentwiththeassociationprocessesthatoccurinmembranes,butisprobablynecessarytoproperlyaccountfortheeffectsofinterfaceionatomsonthethermodynamicsofassociation.Tochallengetheabilityofthepredictiveprotocoltoestimatethedetrimentaleffectondimerizationexertedbytheabsenceofsulfonylation,onlyfortheWT,additionaldockingsimulationswerecarriedoutonthenon-sulfonylatedform,bothintheabsenceandinthepresenceofcalcium(Unmod-WT,Table3).Alsointhesecases,anative-likeclustercouldbeobtainedholdingthetophitoutof4000.However,differentlyfromthesulfonylatedforms,thetoprankedsolutioncouldbefoundonlyinoneofthethreesetsof4000solutions.Intriguingly,byemployingtheZDavg1concerningtheunmodifiedWTformintheabsenceofcalcium(ZDavg1=20.99,Table3)andthecorrelationequationreportedinFig.4A,apositivefreeenergyispredicted,consistentwiththeimpairmentindimer-izationinferredfromtheinvitromeasurementsontheunmodifiedcalcium-freeforms(StanleyandFleming,2007).Takentogether,dockingsimulationsandcorrelativeanalysissuggestthatsulfony-lationisnotessentialforZDOCKtoreconstitutethenativedimerbutisessentialfortheachievementofadegreeofcomplementaritypermissivetodimerization.4.DiscussionTheresultsofthiscasestudyrepresentasuccessfulextensiontotheOMPLAproteinofthepredictionprotocoldevelopedandvali-datedona-helicalmembraneproteins.Infact,inallthetestcases,native-likecomplexescouldbepredicted,independentoftheboundorunboundstateofthestructuralmodel,ofsidechaincon-formationandpresenceorabsenceofaminoaciddeletionsattheputativeinter-monomerinterface(Table1).Thus,theresultsachievedfortheOMPLAproteinstrengthentheeffectivenessofthepredictionprotocol,consistingofdensedockingsamplings,startingfromtwoidenticalcopiesofagivenmonomer,followedbymembranetopologyfilteringandclusteranalysis(Casciarietal.,2006).Filteringandclusteranalysisarefundamentalinthisrespect.Ononehand,filteringisessentialindiscardingfalseD.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162161A-5.5B-4.8-5.2-6-5.6 (kcal/mol)(kcal/mol)exp1 -6.5-6-6.4-6.8-7.2exp1ΔG-7.5ΔG-7-7.6-825.82626.2ZD-826.4avg126.626.827-8-7.6-7.2-6.8ΔG-6.4-6-5.6-5.2-4.8pred (kcal/mol)Fig.4.PredictionofmutationaleffectsonOMPLAdimerization.(A)Thecorrelationbetweenfreeenergyofdimerization(DGexp1),withexperimentalerrorbars(StanleyandFleming,2007),andaveragedockingscores(ZDavg1)isshown.Thelinearcorrelationequationis:DGexp1=38.1À1.7ZDavg1,N=6,R=0.96.(B)ExperimentalversuspredictedDG’sforthesixOMPLAvariants.Predictionsrefertoaleave-one-outcross-validationtest.Thelinearcorrelationequationis:DGexp1=À2.1+0.7DGpred,N=6,R=0.88.positivesthatrepresentmorethan97%ofthebestscoredsolutions,thusimprovingthepredictivepoweroftheprotocol.Ontheotherone,clusteranalysis,performedonthefilteredsolutions,isinstru-mentalinindividuatingthenative-likesolutions.Thelattergener-allyfallwithinthemostpopulatedclustercharacterizedbyagoodmembranetopology,namelybyaMemTopindexlowerthan0.5.Similartotheresultsachievedfora-helicaltransmembranepro-teins(Casciarietal.,2006)thebestscoredsolutionfromthemostpopulatedclusterfallswithinthetop30solutionsoutofthe4000collected.Inmostcases,theyfallwithinthetop10(Table1).Takentogether,clusterpopulationanddockingscoreconfirmtobethemostimportantcriteriafordetectingnative-likesolutionsbothfora-helical(Casciarietal.,2006)andb-strandedtransmembraneproteins.Effectivepredictionsofnative-likearchitectureswereindepen-dentofthepresenceofsulfonylationorcalciumions,butturnedouttostronglydependontheintegrityoftheL1,L2,L3andL4loops.Indeed,intheabsenceofsulfonylchain,anypossiblepair-wisedeletionsoftheseloopsexceptfortheL1+L4onepreventedZDOCKfromreconstitutingthenative-likedimer.ThepresenceofsulfonylationwasabletorestoretheZDOCKabilitytopredictana-tive-likedimeronlyfortheTrunc11form.TheseresultssuggestthatanypairwisecombinationofL2witheitheroneoftheremain-ingthreeloopsisessentialforZDOCKtopredictthenativearchi-tectureoftheOMPLAdimer,independentofthepresenceorabsenceofsulfonylation.Sulfonylation,however,turnedouttobeessentialfortheachievementofadegreeofcomplementaritypermissivetodimerization.ThiswasinferredbytheenergeticsanalysisofOMPLAassociation.Inthisrespect,followinganap-proachpreviouslydevelopedforestimatingmutationaleffectsonthefreeenergyofprotein–proteinandprotein–DNAassociation(Dell’Orcoetal.,2007a,b,2005;FanelliandFerrari,2006),wecouldbuildacorrelativemodelbetweenaveragedockingscoresandinvitrodeterminedfreeenergyofassociationfortheWTandfiveOMPLAmutantsintheirsulfonylatedforms.Thismodelpredictedimpairmentindimerizationforthenon-sulfonylatedformoftheWT,consistentwiththeresultsofinvitroexperiments.Inconclusion,theresultsofthisstudyextendandcorroborateourapproachestoquaternarystructurepredictionsandestimationofmutationaleffectsonthefreeenergyofprotein–proteindimer-izationinmembrane.AnimportantinferencefromthisstudyisthattheintegrityofL2andeitheroneoftheL1,L3andL4loopsandnotsulfonylationisrequiredfortheachievementofthenativedimericarchitectureofOMPLA.Ontheotherhand,sulfonylationisessentialforafavorabledimerizationenergetics.AcknowledgmentThisstudywassupportedbyaTelethon—ItalyGrantn.S00068TELU.ReferencesBukholm,G.,Tannaes,T.,Nedenskov,P.,Esbensen,Y.,Grav,H.J.,Hovig,T.,Ariansen,S.,Guldvog,I.,1997.ColonyvariationofHelicobacterpylori:pathogenicpotentialiscorrelatedtocellwalllipidcomposition.Scand.J.Gastroenterol.32,445–4.Casciari,D.,Seeber,M.,Fanelli,F.,2006.Quaternarystructurepredictionsoftransmembraneproteinsstartingfromthemonomer:adocking-basedapproach.BMCBioinform.7,340.Chen,R.,Weng,Z.,2003.Anovelshapecomplementarityscoringfunctionforprotein–proteindocking.Proteins51,397–408.Chen,R.,Tong,W.,Mintseris,J.,Li,L.,Weng,Z.,2003.ZDOCKpredictionsfortheCAPRIchallenge.Proteins52,68–73.Dekker,N.,Tommassen,J.,Verheij,H.M.,1999.BacteriocinreleaseproteintriggersdimerizationofoutermembranephospholipaseAinvivo.J.Bacteriol.181,3281–3283.Dekker,N.,Tommassen,J.,Lustig,A.,Rosenbusch,J.P.,Verheij,H.M.,1997.DimerizationregulatestheenzymaticactivityofEscherichiacolioutermembranephospholipaseA.J.Biol.Chem.272,3179–3184.Dell’Orco,D.,DeBenedetti,P.G.,Fanelli,F.,2007a.Insilicoscreeningofmutationaleffectsontransmembranehelixdimerization:insightsfromrigid-bodydockingandmoleculardynamicssimulations.J.Phys.Chem.B111,9114–9124.Dell’Orco,D.,DeBenedetti,P.G.,Fanelli,F.,2007b.Insilicoscreeningofmutationaleffectsonenzyme-proteicinhibitoraffinity:adocking-basedapproach.BMCStruct.Biol.7,37.Dell’Orco,D.,Seeber,M.,DeBenedetti,P.,Fanelli,F.,2005.Probingfragmentcomplementationbyrigid-bodydocking:insilicoreconstitutionofcalbindinD9k.J.Chem.Inf.Mod.45,1429–1438.DunbrackJr.,R.L.,Karplus,M.,1993.Backbone-dependentrotamerlibraryforproteins.Applicationtoside-chainprediction.J.Mol.Biol.230,3–574.Fanelli,F.,Ferrari,S.,2006.PredictionofMEF2A–DNAinterfacebyrigidbodydocking:atoolforfastestimationofproteinmutationaleffectsonDNAbinding.J.Struct.Biol.153,278–283.Grant,K.A.,Belandia,I.U.,Dekker,N.,Richardson,P.T.,Park,S.F.,1997.MolecularcharacterizationofpldA,thestructuralgeneforaphospholipaseAfromCampylobactercoli,anditscontributiontocell-associatedhemolysis.Infect.Immun.65,1172–1180.Heyer,L.J.,Kruglyak,S.,Yooseph,S.,1999.Exploringexpressiondata:identificationandanalysisofcoexpressedgenes.GenomeRes.9,1106–1115.162D.Dell’Orcoetal./JournalofStructuralBiology163(2008)155–162Asn156AlaofoutermembranephospholipaseA:functionoftheAsn–Hisinteractioninthecatalytictriad.ProteinSci.10,1962–1969.Snijder,H.J.,Ubarretxena-Belandia,I.,Blaauw,M.,Kalk,K.H.,Verheij,H.M.,Egmond,M.R.,Dekker,N.,Dijkstra,B.W.,1999.Structuralevidencefordimerization-regulatedactivationofanintegralmembranephospholipase.Nature401,717–721.Stanley,A.M.,Fleming,K.G.,2007.Theroleofahydrogenbondingnetworkinthetransmembranebeta-barrelOMPLA.J.Mol.Biol.370,912–924.Stanley,A.M.,Chuawong,P.,Hendrickson,T.L.,Fleming,K.G.,2006.EnergeticsofoutermembranephospholipaseA(OMPLA)dimerization.J.Mol.Biol.358,120–131.Sutcliffe,M.J.,Hayes,F.R.,Blundell,T.L.,1987.Knowledgebasedmodellingofhomologousproteins,partii:rulesfortheconformationsofsubstitutedsidechains.ProteinEng.1,385–392.Horrevoets,A.J.,Verheij,H.M.,deHaas,G.H.,1991.InactivationofEscherichiacoliouter-membranephospholipaseAbytheaffinitylabelhexadecanesulfonylfluoride.Evidenceforanactive-siteserine.Eur.J.Biochem.198,247–253.Ponder,J.W.,Richards,F.M.,1987.Tertiarytemplatesforproteins.Useofpackingcriteriaintheenumerationofallowedsequencesfordifferentstructuralclasses.J.Mol.Biol.193,775–791.Scandella,C.J.,Kornberg,A.,1971.Amembrane-boundphospholipaseA1purifiedfromEscherichiacoli.Biochemistry10,4447–4456.Snijder,H.J.,Kingma,R.L.,Kalk,K.H.,Dekker,N.,Egmond,M.R.,Dijkstra,B.W.,2001a.StructuralinvestigationsofcalciumbindinganditsroleinactivityandactivationofoutermembranephospholipaseAfromEscherichiacoli.J.Mol.Biol.309,477–4.Snijder,H.J.,VanEerde,J.H.,Kingma,R.L.,Kalk,K.H.,Dekker,N.,Egmond,M.R.,Dijkstra,B.W.,2001b.Structuralinvestigationsoftheactive-sitemutant