mechanism of potassium channel selectivity revealed by na and...

20
Mechanism of potassium channel selectivity revealed by Na + and Li + binding sites within the KcsA pore Ameer N. Thompson, Ilsoo Kim, Timothy D. Panosian, Tina M. Iverson, Toby W. Allen, Crina M. Nimigean Supplementary Information Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Upload: others

Post on 16-Mar-2021

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

MechanismofpotassiumchannelselectivityrevealedbyNa+andLi+bindingsiteswithintheKcsAporeAmeerN.Thompson,IlsooKim,TimothyD.Panosian,TinaM.Iverson,TobyW.Allen,CrinaM.Nimigean

SupplementaryInformation

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 2: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

SupplementaryFigure1|LowaffinityblockingsitesforLi+intheKcsAvestibule.a,AviewoftheselectivityfilterandvestibuleofNoK‐Li+showing|Fo|‐Fc|stylesimulatedannealingomitdensity(greenmesh)contouredto4σforthewatermolecules(redspheres)observedinthevestibule.Mapswerecalculatedbyomittingthewatermolecules,anda5Åradiusaroundthewatermoleculesfromthephasecalculation.StickmodelsoftwoopposingKcsAsubunitshighlighttwoputativeLi+blockingsitesintheNoK‐Li+structure:b,betweentheupperringofwatermoleculesandthesidechainhydroxylsofThr‐75,andc,withinthecenteroftheeightwatermoleculesinthevestibule.Li+ionsareshowningreen,watermoleculesofthefirsthydrationshell(notobservedinthedensity)areshowningrey,andwatermoleculesofthesecondhydrationshell(observedinthedensity)areshowninred.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 3: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

SupplementaryFigure2|ComparisonofNoK­Li+toLowK­Li+andotherpublishedstructuresofKcsA.TheNoK‐Li+structurewascomparedtoLowK‐Li+and8othermodelsofKcsAdepositedintheProteinDataBank(PDB):Low‐K+(PDBID=1k4d),High‐K+(PDBID=1k4c),NoK‐Na+(PDBID=2itc),High‐Tl+(PDBID=1r3k),Low‐Tl+(PDBID=1r3k),Rb+(PDBID=1r3i),Cs+(PDBID=1r3l)andBa+(PDBID=2itd).a.Rootmeansquared(RMS)deviationsofCα‐atomswerecalculatedusingLSQKAB5andplottedasafunctionofresiduenumber.Theblackbarrepresentstheregionoftheselectivityfilter.b.Atablesummarizingthemodelingofionsintheselectivityfilterofeachstructure.c.QuantificationofelectrondensitywithintheKcsAselectivityfilter.ElectrondensitypeakheightsofcompositeomitmapsweremeasuredusingtheprogramPEAKMAX5andwerenormalizedtothepeakcorrespondingtothebackbonecarbonylofCys88(onchainBoftheFABfragment).Theionlistedaboveeachgraphistheionmodeledatthesite.Forreferencetheoreticalratiosofpeakheights(proportionaltonumberofelectronsineachion)areplottedasdottedlines:K+:O=14:10=1.4,HOH:O=10:10=1.0,Na+:O=10:10=1.0.ThesequantificationssupporttheassignmentofwatermoleculesintheselectivityfilterofboththeNoK+‐Li+andLowK+‐Li+structures.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 4: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

SupplementaryFigure3|Selectivityfilterbackbonealignment.Stereoviewofcompositeomitelectrondensitycontouredto1.25sigma(bluemesh)foronemonomeroftheselectivityfiltermodeledasyellowsticksfromeithera,NoK‐Li+orb,LowK‐Li+.TheselectivityfilterLow‐K+wasalignedtoeitherstructureusingthealigncommandinPyMolandisshownasgreysticksforreference.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 5: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

SupplementaryFigure4|Convergenceanderrorplotsforfreeenergycalculations.Panelsa,canderevealtimeconvergenceofumbrellasamplingfor3samplefreeenergyprofilescorrespondingtothoseofpathsI,IIandII’fortheLi+ioninFig.6ofthemaintext.Panelsb,dandfshowthesamefreeenergyprofiles,butwheretimeissplitintotwosamplesof0.5ns,withthedifferencerepresentingameasureoferror.Therootmeansquare(RMS)differenceisreportedineachgraph.PathIIIisalsoshownindwithanarbitraryoffsetforclarity(matchingS1/S3minima).

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 6: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

Positions and Solvation Coordination number K+ Na+ Li+ Bulk Water

H2O Rmax

7.1±0.2 3.62 Å

5.9±0.5 3.22 Å

4.2±0.3 2.82 Å

H2O

5.4±1.2

4.8±1.2

3.8±0.6

C=O - - - Thr OH 1.6±0.9 1.1±0.3 0.4±0.6

Cavity (S1/S3/Cavity)

Total 7.0±1.4 5.9±1.2 4.2±0.8 Mean and range of z See Fig.4b.

H2O 1.6±0.4 2 2 C=O 4 4 3.9±0.1 Thr OH 2.8±0.7 0 0

S4 region (S0/S2/S4)

Total 8.4±0.9 6 5.9±0.1 Mean and range of z -5.5 [-6.6,-4.3] -4.5 [-5.3,-3.7] -4.2 [-4.7,-3.6]

H2O 1.9±0.2 1.2±0.4 2.5±1.0 C=O 4.0±0.1 3.5±0.5 1.8±0.3 Thr OH 2.7±0.4 1.9±0.2 0.7±0.7

S4 Cage (S0/S2/S4)

Total 8.5±0.4 6.6±0.6 5.0±1.2 Mean and range of z -5.5[-6.0,-4.8] -5.4[-6.0,-4.8] -5.5[-6.1,-4.7]

H2O 2.2±0.2 2 2 C=O 4 4 4 Thr OH 0.1±0.1 0 0

S4 Plane (S0/S2/S4)

Total 6.3±0.2 6 6 Mean and range of z -4.2[-4.9,-3.4] -4.2[-4.7,-3.6] -4.1[-4.5,-3.5]

Supplementary Table 1 | Ion solvation number and position analysis. Thenumber of oxygen atoms within distance Rmax of the ion are given for differentcoordinatingspecies.Bulk,CavityandS4regioncalculationsarebasedonunbiasedsimulationswheretheionwasfreetomove.TheS4cageandinplanecalculationswerebiasedwithplanarharmonicpotentials corresponding toRMS fluctuationof0.1Å.Uncertaintieslessthan0.01arenotshownaserrorbars.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 7: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

SupplementaryTable2|FreeEnergyPerturbationCalculations.Freeenergies(ΔG) for KNa and NaLi were calculated from dedicated FEP simulations andWHAManalysiswhileKLiwasderivedbyadditionofthesetwonumbers.ValuesofΔΔGforNaandLi,relativetoKandexpressedrelativetobulkareprovidedintheright hand columns. Cage and plane calculations were biased with harmonicpotentialsandthenunbiasedasdescribedinthesupplementarymethods.Valuesinbracketscome fromsimulationswith the1K4Cstructurewhereasallotherscomefromsimulationswith the1BL8 structure. Errorbars areone standarddeviationbased on 25‐50 blocks of 20ps. Values for free energy decomposition viathermodynamic integration(TI)dueto interactionswithwater,carbonylsandThrOH groups are also provided for each S4 calculation (contributions from otherinteractions, including with neighboring residues and other ions, were found tocontributelittletotheperturbationfreeenergiesandarenotreported).

FreeEnergies

ΔG(K+Na+) ΔG(Na+ Li+) ΔG(K+ Li+) ΔΔG(K+Na+) ΔΔG(K+ Li+)

Bulk

­17.7±0.1(­17.8±0.1)

­22.4±0.3(­22.6±0.3)

­40.1±0.3(­40.4±0.4)

­ ­

Cavity(S1/S3/Cavity)

­17.9±0.2(­18.2±0.2)

­21.7±0.6(­22.1±0.5)

­39.6±0.3(­40.3±0.6)

­0.2±0.2(­0.4±0.3)

0.5±0.6(0.1±0.7)

S4RegionUnbiased(S0/S2/S4)

­19.2±0.5(­19.0±0.4)

­25.6±0.2(­25.4±0.3)

­44.8±0.5(­44.5±0.5)

­1.5±0.5(­1.2±0.4)

­4.7±0.6(­4.1±0.5)

TIH2OTIC=O(53)TIThrOHTITotal

‐9.22±0.37‐10.93±0.44‐0.51±0.21­20.55±0.49

­14.10±0.20‐12.96±0.330.22±0.005­26.71±0.39

‐23.32±0.64‐23.89±0.55‐0.29±0.21­47.27±0.62

S4Cage(S0/S2/S4)

­13.7±0.3 ­21.9±0.5 ­35.6±0.6 4.1±0.4(3.2±0.6)

4.6±0.6(4.0±0.8)

TIH2OTIC=O(53)TIThrOHTITotal

‐5.32±0.64‐6.63±0.33‐2.85±0.24‐14.74±0.38

‐14.20±1.43‐6.68±0.89‐2.51±0.86­23.34±0.52

‐19.52±1.57‐13.31±1.57‐5.36±0.89­38.08±0.64

S4Plane(S0/S2/S4)

­23.9±0.2 ­26.5±0.2 ­50.4±0.3 ­6.1±0.3(­4.4±0.4)

­10.2±0.4(­7.9±0.6)

TIH2OTIC=O(53)TIThrOHTITotal

‐11.46±0.13‐14.14±0.250.23±0.01­25.19±0.28

‐14.50±0.24‐13.69±0.210.21±0.002­27.85±0.28

‐25.96±0.28‐27.83±0.330.44±0.01­53.04±0.40

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 8: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

1

SUPPLEMENTARYDISCUSSIONInteractionbetweenLi+andK+duringblockintheKcsAcavityLi+andNa+fastblock1aredependentonpermeantionconcentration.Theequilibriumblockingparameterswereextractedusingamodelthatassumestheblockerdissociatesfromthesitefasterthanthetimeresolutionavailable.Theresultisanapparentreductioninsinglechannelamplitudeproportionaltothefractionoftimetheblockeroccupiesitssite:

I(V ) =I0(V )

1+BKB

ap

, (Eq.S1)

whereBistheblockerconcentrationandKB

ap isitsapparentequilibriumdissociationconstant.KB

ap varieswiththemembranepotential,evidenceofvoltage‐dependentLi+block,expressedas:

KBap (V ) = KB

ap (0)exp −zFVRT

, (Eq.S2)

whereKBap(0)isthezero‐voltageblockingaffinityandzisthevoltage‐dependenceparameter,alsoknownastheeffectivechargemovementassociatedwiththeblock.ThedatainFig.2aarefittedusingeqs.S1andS2(redandbluelines).

TheaffinityforLi+decreaseswithanincreaseinK+concentrationasshowninFig.2b,consistentwithaprocesswhereK+competeswithLi+fortheblockingsite1.IfblockerandK+occupythecavityinamutuallyexclusivemanner,theapparentblockerdissociationconstant(KBap(0))willfollowacompetitionrelation:

KBap (0) = KB (0)+

KB (0)KK (0)

K , (Eq.S3)

whereK+isthepotassiumconcentration,andKB(0)andKK(0)aretheintrinsicaffinitiesat0mVforblockerandK+,respectively.Weusedeq.S3tofittheKBap(0)vsK+relationshipinFig.2b(redline).Theslopeofthelinearfit(theratioofKB(0)/KK(0))is2indicatingthattheblockingsiteprefersK+toLi+byafactoroftwo.AnidenticalanalysisdoneforNa+blockyieldedaratioof5,indicatingasitethatpreferredK+overNa+byafactorof5(dashedlineinFig.2b,from1).K+isstillpreferredoverbothLi+andNa+,butLi+appearstobepreferredtoNa+inthecavity.

Furthermore,Fig.2cshowsthatthevoltagedependenceofLi+blockremainsunchangedovera10‐foldchangeinK+concentration,incontrasttothesignificantincreaseinthevoltagedependenceofNa+blockoverthesameK+range(dashedline,from1).OuroriginalmodeltoexplaintheincreaseforNa+invokedinteractionbetweenNa+andthepermeantionsintheporethatappearstobelessstrongforLi+.Interestingly,thiscouldresultfromamoreshallowbindingofLi+inthecavity,asseenintheMDsimulations(Fig.4).Thesedatapointtoapermeationandblock

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 9: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

2

modelwhereLi+andNa+bindatsomewhatdifferentlocationswithintheporeandinteractdifferentlywiththepermeantions.KineticmodelsfortheeffectofLi+andNa+ondecreasingburstdurationsKcsAburstdurationsaredramaticallyshortenedinaconcentrationandvoltage‐dependentmannerbyapplicationofintracellularNa+andLi+.Aswestateinthemaintext,anobviousmechanismforthisisthatbindingofNa+/Li+inthechannelpore,atthesitethatleadstotheobservedfastblock,inducesinactivation.ThisisshowninSchemeI.SchemeIIpresentsanalternativemechanism,moreconsistentwithourmoleculardynamicsandX‐raycrystallographydata,thatpositsthattheburstdurationsareshortenedbecauseNa+orLi+bindtoanadditionalbindingsiteintheporewithhigheraffinity,blockingtheflowofK+ionsonalongtimescale,similartotheinactivationtimescale.Inbothschemes,theclosed(C)toopen(O)transitiondescribesthegatingwithintheburst,whereIistheinactivatedstateresponsibleforthelongclosedinactivatedintervals,Bfisthefastblockedstatethatcontributestothedecreaseincurrentamplitudeduetothefaston‐andoff‐rates(kBf,kuf)oftheblocker,andBistheslow‐blockedstate,responsibleforthedecreaseinburstdurations.

SchemeI

SchemeII

ThemodelinSchemeIpredictsthattheobservedeffectofdecreasingtheburstdurationswilloccuroverasimilarrangeofconcentrationsasthefastblockeffect.Thehalf‐maximalconcentrationthatinducesthedecreaseinburstdurationsis~1mMat100mV(forbothNa+andLi+)whileaminimumof10mMLi+/Na+isrequiredtoseeanyeffectat100mVondecreasingthesingle‐channelcurrentamplitude.ThissuggeststhatSchemeImaynotbethecorrectmodeltoexplainthedata.

InSchemeII,themeanburstdurationisinverselyproportionaltotheon‐rateoftheblockerontheslow‐blocksite(kB)(Eq.1),askBfistoofasttoinduce

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 10: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

3

interruptionsinthecurrentonthegatingtimescale.AlthoughinthisschemekBisnotdirectlydependentontheconcentrationoftheblockerbutontheoccupancyofstateBf,wewillapproximatekBwithasecondorderrateconstant(becausetheconcentration‐dependentprocessisveryfastanddoesnotaffectslowprocesses),bothconcentrationandvoltage‐dependent(Eq.1),astheblockingsiteisinthetransmembraneelectricfield.ByfittingtheburstdurationdatainFig.3a,bwithEq.1,weobtainedanintrinsicblockeron‐rateof0.11M‐1sec‐1forNa+and0.07M‐1sec‐1forLi+,andavoltagedependenceofthisrateof0.5forNa+and0.5forLi+(constrained,seeFig.3legend).ThesenumbersillustratethatalthoughtheNa+/Li+induceddecreaseinburstdurationsatthesingle‐channellevelappearsquiterobustatlargevoltages,theintrinsicrateofthismodeledslow‐blockprocess(kB)isstillveryslowcomparedtotherateofK+permeation.Unfortunately,wecannotquantitativelyanalyzetheoff‐rateconstantsanddetermineaKdasitisimpossibletodistinguishinactivationeventsfromslowblockedeventsundertheseconditionsinthischannel.Theeffectivegatingchargeofthisslow‐blockprocess(z=0.5)issimilartothatofthefastblock(z=0.4),suggestingthattheyoccuratlocationsclosetoeachotherinthepore.Crystallographically­modeledLi+­bindingsitesintheKcsAcavityWhiletheassignmentofwatermoleculesat2.8Åresolutioncanbechallenging,electrondensityforeightputativeorderedwatermoleculesisobservedinthevestibuleoftheNoK‐Li+structure(Fig.5a,SupplementaryFig.1).ThesewatermoleculesareinpositionssimilartothoseofthehighresolutionKcsAstructures2,3wheretheycoordinateK+orNa+(Fig.5b).ThenearesthydrogenbonddonororacceptortothewatermoleculesintheupperringoftheshellisthesidechainhydroxylofThr75,whichis4.5Åaway.Thisistoofartoformahydrogenbondinginteraction.DuetotheresolutionofourNoK‐Li+structure,thelackofhydrogenbondingbetweenthesewatersandtheprotein,andtheproximityofthewaterstothefourfoldcrystallographicaxis,whereartifactscanbemagnified,extracarewastakentoensurethatthesedensitiestrulyrepresentwatermoleculesandnotnoise.Weusedbothcompositeomitmaps(Fig.5a,b)andsimulatedannealingomitmaps(SupplementaryFig.1a)toprovidetwocomplementarymethodsofreducingmodelbiasatthislocation.Inbothcases,theelectrondensitiesappearwell‐defined.Inaddition,thetemperaturefactorsofthesewatermolecules(52Å2forthewatermoleculesnearertheselectivityfilterand51Å2forthelowerringofwatermolecules)matchcloselythoseoftheneighboringaminoacids(57Å2),consistentwiththedensitiesbeingwatermolecules.Finally,watermoleculeshavebeenpreviouslymodeledatthesepositionsinhigherresolutionstructuresofKcsA,andhavebeenproposedtobefunctionallyrelevant.Thus,thesespecificlocationshavebeenpreviouslyshowntobeimportantwaterfocusingsitesdespitethelackofhydrogenbondstotheprotein.Alltheabovelinesofevidence,togetherwiththeelectrophysiologicaldata(Fig.1‐3)andmoleculardynamicscalculations(Fig.4and6)thataccompanythisstructure,allowustoassignthesedensitiesaswatermoleculesratherthannoiseresultingfromthecrystallographic4‐foldandwehavemodeledtheminthismanner.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 11: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

4

Sincenoneofthe8modeledorderedwatermoleculesinthecavityoftheNoK‐Li+structureishydrogenbondedtotheprotein,wesuggestthattheirpositionsarestabilizedbythepresenceofanionthatisnotobservableinanx‐rayexperiment,i.e.Li+.ThesewatermoleculescanformanidealizedsecondaryhydrationshelltoatetrahedrallycoordinatedLi+.Theprimaryhydrationshelllacksfourfoldsymmetrybutispositionedalongthefour‐foldcrystallographicaxis.Inaddition,theorientationofthistetrahedronmaybeconformationallydegeneratesuchthatitisnotobservedintheelectrondensity.FortheLowK‐Li+structure,theelectrondensityofthewatermoleculeswithinthecavityisnotstatisticallysignificant.Crystallographically­modeledwatersintheselectivityfilteroftheNoK­Li+structureThreespheresofelectrondensityareobservedwithintheselectivityfilteroftheNoK‐Li+structureonthecrystallographicfourfoldaxis(Fig.5a).ThesedensitiesarelocatednearS1,S3,andS4,theS‐sitesinK+‐containingKcsAstructures2,3wheredensitieswereobservedinother,higher‐resolutionKcsAstructures(Fig.5).Wehypothesizethatthesedensitiesresultfromwatermoleculesfortworeasons.First,thecrystallizationwasperformedinLi+astheonlymonovalentcationsandLi+isinvisibleatthisresolution.Second,wequantifiedthesedensitiesinordertocomparethemwithsimilarlylocateddensitiesinotherKcsAstructures,bynormalizingthemtotheoxygenatomofCysB88(SupplementaryFig.2).CysB88providesanidealreferencemeasurementbecauseitisnotinvolvedinbinding,iswellordered,andisconsistentbetweenallfourstructures.Whiletherearecaveatstointerpretationsresultingfromthisanalysis(theatomisfarfromthefilter,itiscovalentlybound,thereisvariabilityintheresolutionofthedatasets,andtheionsorwatermoleculesintheselectivityfiltermayhavepartialoccupancy),bynormalizingthedensitypeakstoareferencepointwithsimilarproperties,weovercomethechallengeofcomparingstructuresofdifferentresolutionswhileaccountingforthevariationinsignaltonoise.

QuantificationofthedensityattheS3andS4sitesintheNoK‐Li+andLowK‐Li+structuresshowsthatthesearenearlyequalinmagnitudeandofsimilarintensitytoacarbonyloxygen(SupplementaryFig.2).Incontrast,intheS4siteoftheLow‐K+andatallS‐sitesintheHigh‐K+structureKcsAstructures2,3(Fig.5a,b,c,SupplementaryFig.2),thedensitiesarestrongeraswouldbeanticipatedforK+ions.OurcrystallographicdataarethusbestexplainedbytheassignmentofwatermoleculesintheselectivityfilteroftheNoK‐Li+structure.WehypothesizethatthewatermoleculesatS3andS4arerequiredtocompletetheshellofahexa‐coordinatedLi+boundattheB‐site,withtheotherfourligandscontributedbyfourThr75carbonyloxygenatoms(Fig.5).

WhiletheselectivityfilteroftheLowK‐Li+structureappearstoadopttheconductiveconformation,otherfeaturesoftheelectrondensitysuggestthatamixedstateispresentinthiscrystal,withsomeKcsAmoleculeshavingLi+andothershavingK+inthepore.Reflectingthislikelyaveragingovertwostates,theelectrondensitywithintheselectivityfilter(Fig.5b)islessclearlyresolvedthanitisfortheNoK‐Li+structure(Fig.5).Thepatternoftheintensitiesofelectrondensitywithin

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 12: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

5

theporeofLowK‐Li+issimilartoNoK‐Li+anddistinctlydifferentfromLow‐K+(Fig.5b,SupplementaryFig.2).Whenallofthesedensitiesaremodeledaswatermolecules,thetemperaturefactorsrefinetoapproximatelythatofthesurroundingpeptidechain.Bycomparison,modelinginthemannerofLow‐K+,withaK+atS1andS4andawatermoleculeatS3,resultsinsignificantlyelevatedtemperaturefactorsascomparedthesurroundingatoms.Takentogether,thissuggeststhatevenin3mMK+,KcsApreferentiallybindsLi+ionsattheBsiteratherthanadoptingacollapsedporeconformation.Paradoxically,thedecreaseinelectrondensityqualityaccompaniesaslightimprovementinresolutionfortheLowK‐Li+structure(seeTable1).FreeEnergyPerturbationCalculationsResultsforaverageandrangeofpositionsfordifferentionsineachlocationoftheporeareincludedinSupplementaryTable1.PositionsarenotlistedinthecavitybecauseofthevastrangeoflocationsthatarebetterdescribedbythedistributionsinFig.4bofthemaintextobtainedusingumbrellasampling.Errorbarsarestandarddeviationsbasedon25‐50blocksof20ps,andmayoverestimatetheuncertaintycomparedtoastandarderrorofmeans.SupplementaryTable1alsoincludesaveragecoordinationanalysisfromthesesimulationsinbulk,cavityandS4,includingbiasedS4simulationswheretheionwasheldeitherintheplaneorinthecageofS4(seesupplementarymethods).Interestingly,thesmallerLi+iondoesnotsiton‐axiswhenheldintheenergeticallyunfavorablecagesite,butinsteadmoveslaterallyforoptimumcoordinationbyfewerproteinligandsandwater.

SupplementaryTable2reportsallfreeenergycalculationsfortheperturbationsfromK+toNa+,Na+toLi+andthecumulativeperturbationfromK+toLi+.Thesenumbersarealsogivenrelativetobulkwaterasameasureofselectivityofeachsite.TheS4regionoverallisslightlyselectiveforNa+overK+andstronglyselectiveforLi+ions.Separatesimulationswheretheionisconstrainedtobeincageorinplane(B‐site)revealastarkdifferenceintheselectivityofthesetwodifferentcoordinationgeometries.AnalysisofthesesimulationswasalsousedtogeneratethefreeenergyprofilesofFig.4cinthemaintext.DecompositionsintocontributionsformparticularinteractionsviathermodynamicintegrationarealsoprovidedinSupplementaryTable2. Thereareseveraldeterminantsofthein‐planeB‐sitelocationforNa+andLi+intheS4regionofthefilter.Simplistically,anionsmallerthanK+residinginS4musteithermoveupwardtofillfreevolume(duetoitssignificantlysmallerionicradius)orelsesufferthecostsofadownwardmovementofboundwaterandionsinS0‐S3abovetheion.Secondly,thein‐plane6‐ligandsiteprovidesatleastthebulkcoordinationneedsofthesesmallerions(basedonbulkwaterandignoringdifferencesinligandchemistry),withnoobviousobstructionsforclosecoordination,yetdoesnotmatchtherawcoordinationneedsofthelargerK+ion.Decompositionofthecalculatedfreeenergyvaluesshowsthation‐waterandion‐carbonylinteractionscontributemoretothestabilizationofNa+andLi+whenintheplanecomparedtothecagesite,i.e.thein‐planesitecanmoreeffectivelystabilizeasmallerionduetostrongerinteractionswiththosecoordinatinggroups.Weproposethat,inadditiontotheseenergeticcontributions,thestronglycoordinating

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 13: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

6

watermoleculestothesmallionsactasobstaclestotheinwardmotionofligandstotheionwheninthecage4.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 14: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

7

SUPPLEMENTARYMETHODS

DetailedMolecularDynamicsmethodsThesimulationsystem,showninFig.4a,consistsoftheKcsAproteinimmersedinadipalmitoylphosphatidylcholine(DPPC)membranewithexplicitwatermoleculesandwasconstructedusingextensionstothemembranebuilderprotocoldescribedelsewhere6.TheKcsAproteinstructurewastakenfromPDBfiles1bl8or1k4c(fortwoindependentsetsofcalculations),threeboundK+ionsandrequiredwatermoleculeswereplacedintheselectivityfilterandenoughwatermoleculestofilltheentirechannellumenwereaddedbyoverlayingpre‐equilibratedwaterandkeepingonlythoseinsidethepore.115hydratedDPPClipidswererandomlychosenfromapre‐equilibratedmembraneandoptimizedaroundtheproteinwithanasymmetrytoaccountforthenon‐cylindricalshapeoftheprotein(65lipidsinthebottomleafletand50inthetopleaflet).Themembrane‐proteinsystemwasthenplacedbetweenbulkelectrolytesolutionsof∼150mlKCl,correspondingtoanadditional13K+and23Cl‐ions(tobalancenetcharge)andatotalof7630watermoleculesand∼43,770atomsoverall.TheprogramCHARMM7version32b2,withthePARAM27forcefield8,9(DPPClipids10andTIP3Pwater11)wasusedforallMDsimulations.ElectrostaticswerecomputedusingtheParticleMeshEwaldalgorithm12withoutcutoff,a12ÅLennard‐Jonestruncation,andbondstoHatomsweremaintainedwiththeSHAKEalgorithm13.Simulationswereperformedunderconstantnormalpressure(1atm)withanextendedLagrangianalgorithm1415andtemperature(330K,abovethegelphasetransitiontemperatureforDPPC)wascontrolledbyaNose‐Hooverthermostat16,17.Hexagonalperiodicboundaryconditionswithxy‐translationlengthof77.0Åandaverageheightof~90Åwaspreservedwithpistonmass(750.0amu)andcollisionfrequency5ps‐1,normaltothemembrane(z‐direction).Topreventtheproteinfromdriftinginthexyplane,acylindricalconstraintof5kcal/mol/Å2wasappliedtothecenterofmass(COM)ofKcsAandaplanarconstraintinz‐directionwasalsoappliedtoCOMoflipidmembranewithforceconstant5kcal/mol/Å2(neitheroftheseconstraintshavinganyeffectonresults). Withregardstothechoiceofionparametersandtheiraccuracy,wehavepaidcarefulattentiontotheagreementwithrelevantexperimentalbenchmarks,whereavailable.ItisknownthatCHARMMdescribeshydrationfreeenergieswellwithoutanyforcefieldmodification18,19.ModificationshavebeenmadeintheLennard‐Jonesparameterfortheioninteractionwithfiltercarbonylstoensurethatnotonlyisthecorrecthydrationfreeenergyofeachionachieved,butthatsolvationinN‐methylacetamide(NMA)isclosetoexperimentalbenchmarkstoensurecorrectenergeticsforsolvationbyproteinbackbone19,20.ThechosenparametersarethoseusedpreviouslyforK+andNa+thatapproximatelymatchwaterandNMAsolvationfreeenergies,withasmallpreferencefortheioninNMA21.TheabsolutefreeenergiesofK+solvationinwaterandNMAare‐79.6kcal/mol18and‐83.2kcal/mol,respectively2223:valueswhichprovidereasonableK+ioniccurrentcharacteristicsforthechannel24.BasedonFEPcalculations(K+Na+andNa+Li+)usingboxesof

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 15: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

8

400wateror200NMAwith22windowsof4ns(SergeiNoskov,personalcommunication;allvalueswithstandarddeviationsof∼0.3kcal/mol)weestimateabsolutesolvationfreeenergiesof‐98.1and‐99.9kcal/molforNa+inwaterandNMA,respectively.Despitenothavingthisexperimentalbenchmark,weusethesameapproachforLi+withparametersthatgivehydrationandNMAsolvationfreeenergiesof‐121.9and‐123.3kcal/mol,respectively(previouslyparameterizedbyS.Noskov21).Whilesomeerrormayexistinthesefreeenergies,weremarkthatthelargebindingandrelativefreeenergies(K+‐Li+)intheB‐site,of5‐10kcal/mol,suggestrobustresults.

Priortoanyfreeenergycalculationsthemembrane‐proteinsystemwassimulatedfor30nswithoutanyconstraints.AS1/S3/cavityconfigurationofK+ionsinaconductingstate(withallVal76carbonylspointingintothepore)waschosenforfurtherstudies.PreviouslyithasbeenreportedthatthecarbonylgroupsatVal76flipawayfromtheselectivityfilterduringMDsimulations19,25‐27,possiblyassociatedwithC‐typeinactivation27‐29.InordertosampleappropriateconductingconformationsofKcsAduringallofthefollowingsimulations,aharmonicdihedralconstraintwithforceconstantof0.0030kcal/mol/deg2,centeredon‐90°,wasappliedtoψdihedralanglesofeachVAL76ofthefoursubunitsofKcsA.Usingthisconstraint,afurther700psofdynamicswasusedtoequilibratethesystembeforefreeenergycalculations.TogenerateastartingS0/S2/S4configurationweexchangedionsandneighboringwatermolecules,carriedoutconstrainedminimizationandthenafurther700psofequilibrationwasperformedbeforeanyfreeenergysimulation.

Freeenergyperturbation(FEP)andpotentialofmeanforce(PMF)­umbrellasamplingcalculationsWehavecarriedouttwotypesoffreeenergysimulationinthisstudy:FEP30,31,wherewealchemicallymutateionstofindtherelativefreeenergiesofK+,Na+andLi+,asdonepreviouslybyseveralgroups19,21,22,25,32‐36,aswellasPMF‐umbrellasampling37calculationswhereweextractfreeenergyprofilesalongdifferentpositionalcoordinateswithbiasedsampling.Allcalculationswerecarriedoutwithamultipleionconfigurationinthechannel(3K+ions,or2K+ionswith1Na+orLi+ion),asexplainedinthetext,withexcessKClsolutioninthebaths.FEP:WefirstcarriedoutunconstrainedFEPsimulationsoneitherthecavityioninaS1/S3/cavityconfiguration,orontheS4ioninanS0/S2/S4configurationtodeterminetherelativefreeenergies,ΔG,ofK+,Na+andLi+ineachofthosesitesandcomparedtoasimilarcalculationinbulkwater(neartheedgeofthesimulationbox).Wefoundoptimumconvergenceiftheperturbationswerecarriedoutintwosteps,fromK+toNa+andthenfromNa+toLi+ineachsite.Alinearcouplingschemewasusedwith11simulationwindowsfromλ=0to1.Toavoidhysteresis,perturbationswerenotcarriedoutsequentially.Instead,all11windowsweresimulatedconcurrentlyondifferentcomputersuntilfreeenergieswerewellconverged,excludinginitialdataof20‐40psforequilibrationwitheachperturbedHamiltonian.Simulationsforeachwindowwerecarriedoutfor500ps‐1nseach,representing5‐10nsofsimulationforeachΔGcalculation.Therelativefree

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 16: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

9

energieswerecalculatedusingtheweightedhistogramanalysismethod(WHAM38,39).

Inadditiontounbiasedsimulationswithineachsite,wecarriedoutconstrainedFEPsimulationsof500ps/windowusingtheS0/S2/S4configurationwiththeS4ionheldineitherthecageof8carbonylsorintheplaneofthefourThr75carbonylsoftheS4site.Planarharmonicpotentialswiththreedifferentforceconstantsof60kcal/mol/Å2,10kcal/mol/Å2,and2.5kcal/mol/Å2,allowingRMSDfluctuationsof∼0.1,0.25,and0.5Å,respectively,wereappliedtoensuretheionstayedclosetotheCOMofthecageformedby8carbonyloxygenofTHR75(Cage)orneartheplanedefinedbytheCOMofthe4carbonyloxygenatomsofTHR75.Theconstraintwasthenrigorouslyunbiasedbyaddingthefreeenergy,−ΔGcons ,forturningofftheconstraint,−ΔUcons ,inpost‐trajectoryanalysis:

consln exp( / )kT U kTΔ ,wherethebracketssignifyensembleaveragecalculationwithaHamiltonianthatincludestheconstraint,kisBoltzmann’sconstantandTisthetemperature.Thiswascomputedforλ=0and1endpointsonlyandwaswellconverged.

ThemultiplebiasedFEPtrajectoriesaboveforthein‐planeandin‐cageconfigurationswithinS4constituteaseriesofbiasedsimulationsforK+,Na+andLi+,aswellasallhybridionsbetween.Byusingendpoints,λ=0and1,fromtheK+Na+andNa+Li+perturbationswehaveenoughdatatoproduceaPMFacrosstheS4site,fromcagetoplane(6umbrellasamplingwindows,withdifferentconstraints,periontype).EmployingWHAManalysisonthebiasedtrajectories,wehavecomputedthefreeenergyprofileacrossS4foreachiontype.Becausethecoordinatesoftheconstraintswererelativetothecenterofmassofeither4or8ligandcarbonyloxygenatoms,profilesweretransformedintoacommoncoordinateframeforthisstudy,whichisthezcomponentofthepositionoftheionrelativetotheCOMoftheselectivityfilterwithresultsshowninFig.4bofthemaintext.UmbrellaSampling:Wehavecarriedoutseveralumbrella‐samplingcalculationstoexplorekeypathsinthemultipleionpermeationmechanisminthepresenceofNa+andLi+ions(showninFig.6).InthepresenceofK+ionsinS0andS2,orinS1andS3,wehavemovedanion(K+,Na+orLi+),atequilibrium,fromdeepinthecavityatz=‐15Å(definedrelativetotheCOMoftheselectivityfilter)upabovetheS4sitewithz=‐3Åinaseriesof25umbrellasamplingwindows,spaced0.5Åapart,withforceconstants10kcal/mol/Å2.Eachwindowwassimulatedfor1ns,yielding10‐30ns/calculation.Because,forsomeextremepositionsofthislowerion,theuppertwoions(ions1and2),maybeunstable,thosetwoionswerekeptintheirplacewithasteepflat‐bottomharmonicpotentialthatisnotfeltunlesstheionsattempttoleavetheirsites.Theallowedrangeofthecenterofmassofthosetwoions,z12,was0‐2ÅforS0/S2(arangethateasilyencompassestheS1/S3freeenergyminimumseenpreviously19and3‐6ÅforS0/S2,alsoeasilyencompassingthatstate’sminimuminthefreeenergysurface.UmbrellasamplingwasalsousedtoobtainfreeenergyprofilesformovingapairofK+ionsinS1/S3toS0/S2whilethethirdion(K+,Na+orLi+)wasresidingeitherinthecavityortheS4site.Ifthelowerionwaskeptinthecavity,thiswasachievedwithasteepflat‐bottompotentialactingatz=‐7Åsuchthattheioncouldobtainanequilibriumdistributionof

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 17: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

10

positionsthroughoutthecavity,butnotentertheS4site.AsimilarconstraintwasusedifwewishedtokeeptheionintheS4siteandnotletitfalloutintothecavity.ThesecalculationswererepeatedwherethelowerioncouldventureanywhereinthecavityorS4site(unconstrained)astheuppertwoionsweremovedatequilibrium(toexploreaconcertedmovementoflowerandupperions).Thechosenreactioncoordinateforthese2‐ionPMFcalculationswastheCOMoftheupperionpair(twoKionsresidingatS1/S3orS0/S2orinbetween)withcoordinatez12relativetotheCOMoftheselectivityfilter.Atotalof15windowsimulationswereneededtospan‐1to6Åin0.5Åintervalswiththesameforceconstantasgivenabove. ConvergenceandErrorEstimatesforFreeEnergyProfilesWehavecalculatedthetimeconvergenceforthefreeenergyprofilespresentedinFig.6ofthemaintext,from0‐1ns,todemonstratethatallprofileshaveconvergedtowithinasmallfractionofakcal/mol(seeleftpanelsofSupplementaryFig.4).Wehavealsodoneablockanalysisbydividingthe1nsinto2partsandusingthedifferenceinthetwoprofilesasameasureoferror.InSupplementaryFig.4rightpanels,weshowtheLi+iononly.Wechosetoshowtheseasanexamplebecausetheyrepresenttheworstconvergenceofallcalculations.Inthiscase,pathsI,II,II’andIIIrevealedRMSDerrorsof0.7,1.4,1.6and1.0kcal/mol,respectively.FortheK+ion(notshown)theseerrorswere0.6,0.5,0.6and0.6kcal/mol,respectively,andfortheNa+ion(alsonotshown)theywere0.4,1.0,0.7and0.6kcal/mol,respectively.RMSDerrorsforFig.4cofthemaintextwerecalculatedinasimilarway.

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 18: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

11

REFERENCES1. Nimigean,C.M.&Miller,C.Na+blockandpermeationinaK+channelof

knownstructure.JGenPhysiol120,323‐35(2002).2. Morais‐Cabral,J.H.,Zhou,Y.&MacKinnon,R.Energeticoptimizationofion

conductionratebytheK+selectivityfilter.Nature414,37‐42(2001).3. Zhou,Y.F.,Morais‐Cabral,J.H.,Kaufman,A.&MacKinnon,R.Chemistryofion

coordinationandhydrationrevealedbyaK+channel‐Fabcomplexat2.0angstromresolution.Nature414,43‐48(2001).

4. Allen,T.W.,Blizniuk,A.,Rendell,A.,Kuyucak,S.&Chung,S.H.ThePotassiumChannel:Structure,SelectivityandDiffusion.J.Chem.Phys.112(2000).

5. CollaborativeComputationalProject,N.TheCCP4Suite:Programsforproteincrystallography.ActaCrystallographicaSectionDBiologicalCrystallography50,760‐763(1994).

6. Woolf,T.B.&Roux,B.Structure,energetics,anddynamicsoflipid‐proteininteractions:AmoleculardynamicsstudyofthegramicidinAchannelinaDMPCbilayer.Proteins24,92‐114(1996).

7. Brooks,B.R.etal.CHARMM:Aprogramformacromolecularenergyminimizationanddynamicscalculations.JournalofComputationalChemistry4,187‐217(1983).

8. MacKerellJr.,A.D.etal.All‐atomempiricalpotentialformolecularmodelinganddynamicsstudiesofproteins.JournalofPhysicalChemistryB3586–3616(1998).

9. Feller,S.E.&MacKerellJr.,A.D.Animprovedempiricalpotentialenergyfunctionformolecularsimulationsofphospholipids.JPhysChemB104,7510‐7515(2000).

10. Biltonen,R.L.&Lichtenberg,D.Theuseofdifferentialscanningcalorimetryasatooltocharacterizeliposomepreparations.ChemPhysLipids(1993).

11. Jorgensen,W.L.,Chandrasekhar,J.,Madura,J.D.,Impey,R.W.&Klein,M.L.Comparisonofsimplepotentialfunctionsforsimulatingliquidwater.JChemPhys79,926‐935(1983).

12. Darden,T.A.&Pedersen,L.G.Molecularmodeling:anexperimentaltool.EnvironHealthPerspect101,410‐2(1993).

13. Ryckaert,J.P.,Ciccotti,G.&Berendsen,H.J.C.NumericalintegrationoftheCartesianequationsofmotionofasystemwithconstraints:Moleculardynamicsofn‐Alkanes.JCompPhys,327‐341(1977).

14. Anderson,H.C.J.Moleculardynamicssimulationsatconstantpressureand/ortemperature.ChemPhys(1980).

15. Feller,S.E.,Zhang,Y.H.,Pastor,R.W.&Brooks,B.R.Constantpressuremoleculardynamicssimulation‐theLangevinpistonmethod.JChemPhys,4613‐4621(1995).

16. Hoover,W.G.Canonicaldynamics:Equilibriumphase‐spacedistribution.PhysRevA,1695‐1697(1985).

17. Nose,S.Aunifiedformulationoftheconstanttemperaturemoleculardynamicsmethods.JChemPhys,511‐519(1984).

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 19: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

12

18. Beglov,D.&B.,R.Finiterepresentationofaninfinitebulksystem:Solventboundarypotentialforcomputersimulations.JChemPhys,9050‐9063(1994).

19. Berneche,S.&Roux,B.EnergeticsofionconductionthroughtheK+channel.Nature414,73‐7(2001).

20. Allen,T.W.,Andersen,O.S.&Roux,B.IonPermeationthroughaNarrowChannel:Usinggramicidintoascertainall‐atommoleculardynamicspotentialofmeanforcemethodologyandbiomolecularforcefields.BiophysJ,3447‐3468(2006).

21. Noskov,S.Y.&Roux,B.ControlofionselectivityinLeuT:twoNa+bindingsiteswithtwodifferentmechanisms.JMolBiol377,804‐18(2008).

22. Noskov,S.Y.,Berneche,S.&Roux,B.Controlofionselectivityinpotassiumchannelsbyelectrostaticanddynamicpropertiesofcarbonylligands.Nature431,830‐4(2004).

23. Lamoureux,G.&Roux,B.Absolutehydrationfreeenergyscaleforalkaliandhalideionsestablishedfromsimulationswithapolarizableforcefield.JPhysChemB110,3308‐22(2006).

24. Berneche,S.&Roux,B.AmicroscopicviewofionconductionthroughtheK+channel.ProcNatlAcadSciUSA100,8644‐8(2003).

25. Allen,T.W.,Kuyucak,S.&Chung,S.H.MoleculardynamicsestimatesofiondiffusioninmodelhydrophobicandKcsApotassiumchannels.BiophysChem86,1‐14(2000).

26. Shrivastava,I.H.,Tieleman,D.P.,Biggin,P.C.&Sansom,M.S.K(+)versusNa(+)ionsinaKchannelselectivityfilter:asimulationstudy.BiophysJ83,633‐45(2002).

27. Berneche,S.&Roux,B.Agateintheselectivityfilterofpotassiumchannels.Structure,591‐600(2005).

28. Yellen,G.Themovingpartsofvoltage‐gatedionchannels.QRevBiophys31,239‐95(1998).

29. Cuello,L.G.,Jogini,V.,Cortes,D.M.&E.,P.ConferencePoster.BiophysJ,1788(2008).

30. Zwanzig,R.W.Hightemperatureequationofstatebyaperturbationmethod:I.Nonpolargases.ChemPhys22,1420‐1426(1954).

31. Kollman,P.FreeEnergyCalculations:ApplicationstoChemicalandBiochemicalPhenomena.Chem.Rev.,2395‐2417(1993).

32. Allen,T.W.,Kuyucak,S.&Chung,S.H.MoleculardynamicsstudyoftheKcsApotassiumchannel.BiophysJ77,2502‐16(1999).

33. Aqvist,J.&Luzhkov,V.Ionpermeationmechanismofthepotassiumchannel.Nature404,881‐4(2000).

34. Burykin,A.,Kato,M.&Warshel,A.ExploringtheoriginoftheionselectivityoftheKcsApotassiumchannel.Proteins52,412‐26(2003).

35. Bostick,D.L.&Brooks,C.L.,3rd.SelectivityinK+channelsisduetotopologicalcontrolofthepermeantion'scoordinatedstate.ProcNatlAcadSciUSA104,9260‐5(2007).

36. Fowler,P.W.,Tai,K.&Sansom,M.S.TheselectivityofK+ionchannels:testingthehypotheses.BiophysJ95,5062‐72(2008).

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703

Page 20: Mechanism of potassium channel selectivity revealed by Na and …pbsb.med.cornell.edu/pdfs/Nimigean_1_Supp_Article_2.pdf · 2010. 10. 5. · each ion) are plotted as dotted lines:

13

37. Torrie,G.M.&Valleau,J.P.NonphysicalsamplingdistributionsinMonteCarlofree‐energyestimation:Umbrellasampling.J.Comput.Phys.,187–199(1977).

38. Kumar,S.,Bouzida,D.,Swendsen,R.H.,Kollman,P.A.&Rosenberg,J.M.Theweightedhistogramanalysismethodforfree‐energycalculationsonbiomolecules.I.Themethod..J.Comput.Chem.Rev.13,1011–1021(1992).

39. Roux,B.Thecalculationofthepotentialofmeanforceusingcomputersimulations.CompPhysComm(1995).

Nature Structural & Molecular Biology: doi:10.1038/nsmb.1703