development approach Analysis of recombinant monoclonal antibodies by RPLC: Toward a genericmethod Journal of Pharmaceutical and Biomedical Analysis

ContentslistsavailableatSciVerseScienceDirect

Journal of Pharmaceutical and Biomedical Analysis

jo u r n al h om epag e :w w w . e l s e v i e r . c o m / l o c a t e / j p b a

Analysis of recombinant monoclonal antibodies by RPLC: Toward a generic method development approach

Szabolcs Fekete

, Serge Rudaz

, Jen ˝o Fekete

, Davy Guillarme

aSchoolofPharmaceuticalSciences,UniversityofGeneva,UniversityofLausanne,Boulevardd’Yvoy20,1211Geneva4,Switzerland

bBudapestUniversityofTechnologyandEconomics,DepartmentofInorganicandAnalyticalChemistry,Szt.Gellérttér4,Budapest1111,Hungary

a r t i c l e i n f o

Articlehistory:

Received24April2012

Receivedinrevisedform29May2012 Accepted15June2012

Available online 23 June 2012

Keywords:

Monoclonalantibodies Core–shell

UHPLC DryLab

Methoddevelopment Optimizationsoftware

a b s t r a c t

Monoclonalantibodies(mAbs)areanemergingclassoftherapeuticagentsthathaverecentlygained importance.ToattainacceptablekineticperformancewithmAbsinreversedphaseliquidchromatogra- phy,thereisaneedtoworkwiththelastgenerationofwide-poresub-2␮mfullyporousorcore–shell particlesstationaryphases.Inaddition,temperatureintherange60–90^◦Cwasfoundtobemandatoryto limitadsorptionphenomenonofmAbsandtheirfragments.Agenericmethoddevelopmentstrategywas proposedtoaccountfortheselectivity,efficiency,recovery,andthepossiblethermaldegradation.This studyalsodemonstratedthatthegradientsteepnessandtemperaturecannotbeoptimizedusingvan’t Hofftypelinearmodels.Similarly,thecommonlinearsolventstrengthmodelalsogeneratedsomeerror inpredictingtheretentiontimes.Incontrast,whenquadraticmodelswereemployed,theprediction accuracyofretentiontimeswasfoundtobeexcellent(relativeerrorbetween0.5and1%)usingarea- sonablenumberofexperiments(9or6experimentsforoptimizationofgradienttimeandtemperature, whichrequiresbetween6and8h).TwoseparationsofmAbsfragmentswereperformedtodemonstrate thereliabilityofthequadraticapproach.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Ithasbeendemonstratedthatreverse-phaseliquidchromatog- raphy(RPLC)isoneofthemostpromisinganalyticaltechniques inthefield of peptideand proteinanalysis[1,2].Theefficiency obtainedwithRPLCisgenerallysuperiortothatofion-exchange (IEX)orsize-exclusion(SEC)chromatography[1].Furthermore,the separationtime canbeconsiderablyshortenedwithRPLC com- pared toIEX or SEC. Anotheradvantage is the straightforward couplingofthistechniquetovariousionizationmethodsformass spectrometry(MS)detection.

Inlaboratoryapplications(e.g.,biopharmaceuticalqualitycon- trol,industrial issues),separationsofproteinswithverysimilar molecularweightsandnearlyidenticalstructures(conformations) areoftenperformed.Veryoften,thedifferenceinphysicochemical propertiesisrelativelysmallandthereforesimilarretentionbehav- iorsareexpectedforthedifferentforms.Inmanycases,selectivity cannotbeimproved,soefficiencymustbeincreasedtoenhance theresolution.Therefore,thestationaryphase andtemperature becomethetwomostrelevantparametersinmethoddevelopment.

Byusingthemostadvancedstationaryphases,suchascore–shell

∗Correspondingauthor.Tel.:+41223796334;fax:+41223796808.

E-mailaddresses:szfekete@mail.bme.hu,szabolcs.fekete@unige.ch(S.Fekete).

typematerials,sub-2␮mporousparticlesorwide-poremonolithic columns,theseparationpowercanbeincreasedconsiderably[3–7].

Arecentsystematicstudydemonstratedtheeffectofcolumnlength onpeakcapacityofintactproteinseparations[8].

The number ofapproved monoclonalantibodies (mAbs)has beengrowingcontinuouslyinthepharmaceuticalfield.Antibod- iesarelargetetramericglycoproteinsofapproximately150kDa, composedoffourpolypeptidechains:twoidenticalheavychains (≈50kDa)andtwoidenticallightchains(≈25kDa)thatarecon- nected through several inter- and intra-chain disulfide bonds at the hinge region. The resulting tetramer has two similar halvesthat form a Y-likeshape [9]. Functionally,mAbs consist of tworegions:thecrystallizable fraction(Fc) and theantigen- bindingfraction(Fab)[10].Becausethisstructureismadeoffour polypeptidechains,monoclonalantibodiescandisplayconsider- ablemicro-heterogeneity.Thereareseveralcommonmodifications that produce charge variants (or isoforms) (e.g., deamidation, C-terminallysinetruncation,N-terminalpyroglutamation,methi- onineoxidation, andglycosylationvariants)andsizevariantsof thepeptidechains(e.g.,aggregationorincompleteformationof disulfidebridges).Duetotheincreasingimportanceofthisclass oftherapeuticcompounds,thedevelopmentofanalyticalmeth- odsfortheirdetailedcharacterizationisanactiveareaofstudy.

Complete proteolyticdigestionof mAbs (peptide mapping) fol- lowedbygradientRPLC/MSanalysisisthemethodofchoicefor 0731-7085/$–seefrontmatter© 2012 Elsevier B.V. All rights reserved.

http://dx.doi.org/10.1016/j.jpba.2012.06.021

Fig.1. Schematicviewofthelimitedproteolyticdigestionandreductionofmonoclonalantibodies.

theidentificationandquantificationofchemicalmodificationsof mAbs[11,12].However,peptidemappingistimeconsumingand caninduceputativemodifications duringthelengthyandcom- plexsamplepreparation[12].Alternatively,theanalysisoflarge mAbfragments,suchasFab,Fc,F(ab)₂,HCandLC,requiresvery littlesamplepreparationandcanprovideahigh-throughputalter- nativetopeptidemapping(thesamplepreparationprocedureof mAbfragmentsissummarized inFig.1).For thesereasonsand duetoadvancesinRPLCcolumnsandinstrumentation,thesec- ondapproachiscurrentlypreferredoverthetraditionalpeptide mapping.

InthedevelopmentofRPLCmethodsforsmallmoleculecharac- terization,computermodelingprogramsareemployedtoimprove theanalyticalthroughputandobtaindetailedinformationabout methodspecificity.Themostsuccessfulandwidespreadmodel- ingprogram(DryLab,Molnar-Institute,Berlin,Germany)optimizes theDesignSpacebymeasuringandvisualizingtheeffectofdiffer- entparametersincludingmobilephasecomposition,gradienttime andshape,pH,ionicstrength,ternaryeluent,additiveconcentra- tionsortemperature[13].Forthispurpose,theprogramsuggests arelativelywell-definednumberofexperimentsforaparticular stationaryphase.Furthermore,itcanpredicttheseparationinside theDesignSpacebasedonchangesinthemobilephasecomposi- tion,modeofelution(eitherisocraticorgradient),temperature, and pH,as wellas column parameters such ascolumn length, internaldiameter,particlesizeandflowrate[14].Inthecaseof largebiomolecules,suchasmAbs,conformationalchangesinpro- teinsthatoccurinresponsetovaryingchromatographicconditions couldbeverycomplex,leadingtounpredictablebehaviorbasedon thechemical-chromatographicrelationship(e.g.,LSStheory).Thus, thereisaneedtoevaluatethevalidityofthemodelsemployedby thesoftware.

Thisstudytookadvantageoftherecentlycommercializedand veryefficientcolumnspackedwitheither wide-poresub-2␮m, fully porousstationary phase or the last generation core–shell particles.TheretentionpropertiesoflargemAbfragmentswere investigated. Selectivity, efficiency and possible thermal degra- dationswereallconsideredanda genericmethoddevelopment approachwasproposed.Thecommonlyusedlinearretentionmod- els(linearsolventstrengthorvan’tHoffmodels)weremodified intoquadraticmodels,resultinginveryaccuratepredictionofthe retentiontimes.Moreover,thegoalofthisstudywastheevaluation ofmethodoptimizationsoftwaresuchasDryLabforthesepara- tionofantibodyfragmentsbyusingcustomizedmodelsduringthe optimizationprocess.Thisnewapproachallowsa veryfastand

accuratesystematicmethoddevelopmentfortheefficientsepara- tionofmAbsfragments.

2. Experimental

2.1. Chemicalsandcolumns

Acetonitrile (gradient grade) was purchased from Sigma–Aldrich(Buchs,Switzerland).Waterwasobtainedwitha Milli-QPurificationSystemfromMillipore(Bedford,MA,USA).

IgG monoclonal antibodies, including rituximab (MabThera) andbevacizumab (Avastin),werepurchasedfromRoche(Roche Pharma,Switzerland),andpanitumumab(Vectibix)waspurchased fromAmgen(Switzerland).Forthefragmentationofmonoclonal antibodies,dithiothreitol(DTT)andpapain(fromCaricapapaya) wereobtainedfromSigma–Aldrich (Buchs,Switzerland).Triflu- oroaceticacid(TFA)waspurchased fromSigma–Aldrich(Buchs, Switzerland).

AcquityBEH300 C18and C4columnswitha particlesizeof 1.7␮m(150mm×2.1mm,300 ˚A) were purchased fromWaters (Milford,MA,USA).AnAerisWidepore(WP)C18columnpacked with3.6␮mcore–shellparticles(150mm×2.1mm)wasagener- ousgiftfromPhenomenexInc.(Torrance,CA,USA).

2.2. Equipmentandsoftware

All measurements were performed using a Waters Acquity UPLC^TM systemequippedwitha binarysolventdeliverypump, anautosamplerand aUVand/orfluorescence(FL)detector.The WatersAcquitysystemincludesa5␮lsampleloopanda0.5␮lUV flow-cellanda2␮lFLflow-cell.Theloopisdirectlyconnectedto theinjectionswitchingvalve(noneedleseatcapillary).Thecon- nectiontubebetweentheinjectorandcolumninletwas0.13mm I.D.and250mmlong(passivepreheatingincluded),andthecap- illarylocatedbetweenthecolumnanddetectorwas0.10mmI.D.

and150mmlong.Theoverallextra-columnvolumes(Vext)were approximately13␮land 15␮l,as measuredfromthe injection seatoftheauto-samplertotheUVandFLdetectorcells,respec- tively. The measured dwell volume wasapproximately 100␮l.

Dataacquisitionandinstrumentcontrolwasperformedusingthe EmpowerPro2Software(Waters).Calculationanddatatransfer wereachievedusingExceltemplates.

Method optimization was performed using DryLab® 2010 Pluschromatographicmodelingsoftware(Molnar-Institute,Berlin, Germany).

Fig.2.Theeffectoftemperature(andstationaryphase)ontheapparentgradientretentionfactorofrituximabheavychain(A)andlightchain(B)fragments.Columns:Aeris WPC18,AcquityBEH300C18andC4(150mm×2.1mm),injectedvolume:0.5␮l,detection:fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%

TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:from30%to37%BontheAerisWPC18andBEH300C4columns,andfrom32%to39%ontheBEH300C18 columnin6min,flowrate:0.3ml/min.Temperaturewasvariedfrom40to80^◦ContheBEH300columnsandfrom40to90^◦ContheAerisWPC18column.

2.3. Apparatusandmethodology

2.3.1. Mobilephasecompositionandsamplepreparation

For the gradient separation of mAbs and its fragments, the mobile phase “A” consistedof 0.1%TFA in water, whereas the mobilephase“B”was0.1%TFAinacetonitrile.

TheintramoleculardisulfidebondsofIgGmonoclonalantibod- ieswerereducedusingDTT.Forthispurpose,0.05mgofDTTwas addedto100␮lofaconcentratedmAbsolutionandincubatedat 30^◦Cfor60min.Proteinswerecompletelyconvertedintothelight andheavychaincomponents.

Thedigestionofrituximabandbevacizumabwasinitiatedbythe additionofpapain(dilutedto100␮g/mlinwater)toreachafinal protein:enzymeratioof100:1(m/m%).Thedigestionwascarried outat37^◦Cfor3h.Thefinaldigestionvolumeof200␮lwasdirectly injectedusinglowvolumeinsertvials.

2.3.2. Investigationofretentionpropertiesofantibodyfragments Heavychain,light chain,FabandFc fragmentsof rituximab, bevacizumabandpanitumumabwereelutedinthegradientmode.

Toinvestigatetheeffectoftemperature(Fig.2),shortgradient runs(6min)wereperformedonthethreecolumnsatdifferenttem- peraturesfrom40^◦Ctotheuppertemperaturelimitofthecolumn (in10^◦Csteps).TheBEH300C4andC18materialsarestableup to80^◦C,whereastheAerisWPC18isstableupto90^◦C(accord- ingtothevendors).AlthoughtherecoveryofmAbfragmentsisnot acceptableat40and50^◦C,ourpurposewastoillustratetheeffectof temperatureontheretentiontimesusingarelativelybroadrange oftemperatures.Settingthemobilephasetemperaturetolessthan 60^◦Cisnoteffectivefortheseparationoftheselargefragments.

Fortherituximabsamples,alineargradientfrom30to37%B wasemployedwiththeAerisWPC18andBEH300C4columns, whereaswiththeBEH300 C18column,thegradient wasvaried from32to39%Btomaintainasimilarapparentgradientretention factor.Becausepanitumumab isslightlymorehydrophobicthan rituximab,thegradientprogramswereadjustedtoachievecompa- rableapparentretentionproperties.Forthepanitumumabsamples, alineargradientfrom31to39%BwasemployedwiththeAerisWP C18andBEH300C4columns,whereasagradientof33to41%B wasusedwiththeBEH300C18column.Forthebevacizumabsam- ples,agenericgradientof31to40%Bprovidedsuitableretention timesforallthreecolumns.Inthisstudy,shortgradientsepara- tions(6minon15-cmlongcolumns)wereutilizedtoavoidany possiblethermaldegradationofthesamplesatelevatedtempera- tures.Theflowratewassetto0.3ml/min.Thechromatogramswere

recordedinbothUV(215nm)andfluorescencemode(excitationat 280nm,emissionat360nm).Theretentionpropertiesonthedif- ferentstationaryphaseswereevaluatedbyplottingthelogarithm oftheapparentgradientretentionfactors(logkapp)asafunctionof 1/T(reciprocatetemperature).

Toevaluatetheeffectofgradientsteepnessontheretention properties(Fig.3),differentgradienttimesweretestedatagiven temperature.Agenericlineargradient,startingfrom30%to40%

B,wasemployedataflowrateof0.35ml/minforallsamples.The gradienttimewasvariedfrom4,to8,12and16min(at70and 90^◦C).BothUVandfluorescencedetectionmodeswereused.The observedlog(kapp)valueswereplottedasafunctionoftheloga- rithmofgradienttime.

2.3.3. Systematicmethodoptimization

Snyderandco-workersdemonstratedtheutilityofinitialbasic runsformultifactorialexperimentaldesignsin1990s[15].Thegen- eralapproachistosimultaneouslymodeltheeffectoftemperature andgradientsteepnessontheselectivityofapreviouslyselected RPcolumn[16,17].Forconventionalstandardborecolumns(I.D.of 4.6mm)withalengthof15and25cm,at1–2ml/minflowratewith a5–100%ACN-watergradient,twogradientsoftg₁=20–30minand tg2=60–90minshouldbeemployedtoobtainaccuratepredictions fromthelinearsolventstrength(LSS)model.Themodelingsoft- wareisthenusedtogenerateresolutionmapsthatshowthecritical resolutionoftheseparatedpeaks[18].Inthismanner,thegradi- entprogramandcolumntemperaturecanberapidlyandefficiently optimized.

Recent studieshave showedthat mAb fragments (IgG1 and IgG2)generallyeluteusinga30–40%ACN(containing0.1%TFA)gra- dientatelevatedtemperatures[4,19].Inultrahigh-pressureliquid chromatography(UHPLC),narrowborecolumns(2.1mmI.D.)are generallyusedtoincreasethesensitivity,reducefrictionalheating effectsanddecreasethesolventandsampleconsumption.Bytak- ingintoaccountthefactthat(i)onlya10%changeinBproducesan adequategradientforelutingallthedifferentmAbfragmentvari- antsand(ii)that2.1-mmcolumnsareused;thenapplyingtherules ofgeometricalmethodtransfer[13,20],thefollowingconclusions canbedrawn.For150mm×2.1mmcolumns,gradienttimesinthe rangeoftg₁=4mintotg₂=12min(ataflowrateof0.35ml/min, startingfrom30to40%B)shouldprovideappropriateinitialdata forconstructingresolutionmapsandpredictingretentiontimes.

It was recently demonstrated [19] that the useof elevated temperatures(upto80–90^◦C)isnecessaryfortheRPLCseparation of mAb fragments due to the adsorption phenomena on both

Fig.3.Theeffectofgradienttime(atagiventemperature)ontheapparentgradientretentionfactorofrituximabheavychainandlightchainfragments(A),andofbevacizumab FcandFabfragments(B)andthedeviationbetweenlinearandquadraticfittingincaseoftherituximablightchain(C).Column:AerisWPC18(150mm×2.1mm),injected volume:0.5␮l,detection:fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:

from30%to40%B,flowrate:0.35ml/min.Thegradienttime(steepness)wasvariedas4,8,12and16min(at70and90^◦C).

silica-basedand hybridstationaryphases.Atelevatedtempera- tures,thermaldegradationispossibleandbecomes relevantfor gradienttimeslongerthan20min[19].Acompromise mustbe foundbetweentheresidencetime andseparation temperature.

Therefore,theuseoftg₁=4minandtg₂=12mingradientswere employed to avoid issues with stability. Finally, the effect of temperatureonselectivityandresolutionshouldbeinvestigated onlyinalimitedtemperaturerange(e.g.,T=20^◦C).Themobile phasetemperatureshouldthusbesettoT1=70^◦CandT2=90^◦C (orT₁=60^◦CandT₂=80^◦Cdependingonthethermalstabilityof thestationaryphase).

Asdiscussed in Section 3, becauselinear models cannot be usedtodescribethebehavioroflargebiomolecules(25–50kDa), quadraticmodelswereappliedfortheoptimization.A3²factorial designwaschosentosimultaneouslyoptimizethegradientsteep- nessandmobilephasetemperature.Theeffectofthetwofactors wasinvestigatedat3levels.Gradienttimewasequidistantlyset totg₁=4min,tg₂=8minandtg₃=12minwhiletemperaturewas settoT1=70^◦C,T2=80^◦C andT3=90^◦C. Twoexamplesdemon- stratethereliabilityofthesecond-degreepolynomialmodel.The firstexampleshowstheoptimizationprocessfortheRPLCsepara- tionofbevacizumabFcandFabfragments(Section3.3.1),whereas thesecondexampleconfirmstheaccuracyofthisapproachwhen appliedtotheseparationofthelightchainandheavychainvariants ofrituximab(Section3.3.2).

3. Resultsanddiscussion

Whendealingwithlowmolecularweightanalytes,themost commonstrategyinmethoddevelopmentconsistsofselectinga suitablestationaryphasechemistry,organicmodifierandmobile phasepH.Thegradientprogramandmobilephasetemperature aresubsequentlytunedascomplementaryparametersforasecond leveloptimization.

InthecaseoflargebiomoleculessuchasmAbfragments,the methoddevelopmentrulesaredifferentandsomeadditionalcon- straintsshouldbeconsidered.First,ahighlyefficientwide-pore stationaryphasemustbeused,butthechemistryofsuchcolumns isverylimitedandonlyasmallnumberofC4,C8orC18materials arecurrentlyavailable.Toattainsuitablepeakshapeswithmaxi- mumefficiency,themobilephasecontaining0.1%TFAistypically recommended.Finally,acetonitrileshouldbeselectedasthesol- ventbecauseitprovidesalowerbackpressurewhenusingcolumns packedwithsub-2␮mparticles,incomparisonwithmethanolor isopropanol,andisparticularlyfavorableintermsofkineticper- formance.

Based on these considerations, this study was conducted usingrecentlydevelopedstationaryphases,namely,theAcquity BEH300 and Aeris Widepore C4 or C18, and a mobile phase consisting of water with 0.1% TFA and acetonitrile with 0.1%

TFA. The mobile phase temperature and gradient steepness

wereoptimizedtoachievetheoptimum separationofthemAb fragments.

3.1. Retentionbehaviorofantibodyfragments 3.1.1. Effectoftemperatureontheretentionproperties

InRPLC,theeffectoftemperatureontheretentionfactor(k)can generallybeexpressedbythevan’tHoffequation:

logk=−H RT +S

R +logˇ (1)

where H is the enthalpychangeassociated with thetransfer ofthesolute betweenphases, S isthecorrespondingentropy change,Ris themolargasconstant,T istheabsolutetempera- tureandˇisthephaseratioofthecolumn.Whenlog(k)isplotted against1/T,theenthalpyiscalculatedfromtheslopeofthecurve.

Withneutralcompounds,thevan’tHoffplotsexhibitalinearrela- tionship.However,a quadratic dependence oflog(k)versus1/T wasobservedoverawiderangeoftemperaturesinanotherstudy [21],whichemployedsilicaandnon-silicabasedstationaryphases (thisphenomenonhasnotbeenfullyexplainedyet).Theeffectof temperatureontheretentionofpartiallyionizedcompoundsthat existintwoformscanalsobedescribedbyEq.(1).However,both enthalpyandentropyareexpectedtobedifferentforthetwoforms (i.e.,molecularandionizedforms);asaresult,bothHandScan varywithtemperaturewhenbothformsarepresentinsignificant quantities[21].Melanderetal.[22]andCastellsetal.[23]have developedcomplexrelationshipstodescribetheretentionbehav- iorofapartiallyionizedsolutewithauniqueacid–baseequilibrium, basedontheassumptionthat theretentionfactorisconsidered astheweightedaverageoftheretentionfactorsoftheindividual forms.

Theeffectoftemperatureontheretentionbehaviorbecomes morecomplex forlargebiomolecules.Depending onthestabil- ityofthesecondarystructure,themoleculesunfoldtoadifferent extentandinteractwiththestationaryphasewithvariedstrength [24].Duetothedifferentconformation-dependentresponseofpro- teinsatelevatedtemperatures,thechangeintheretentiontimes canbevery different[25,26].Therefore, temperatureoffersthe abilitytoadjust theselectivityoftheseparation. Ifa proteinis completelydenaturedintoarandomcoilconformation,itwillbe elutedasasinglesharppeak.However,undercertainconditions, thenativeconformationand/orotherintermediateconformations mayalsobepresentduringtheanalysis.Eachofthesewillinter- act differently with the stationary phase, resulting in varying retentiontimesandmultiplepeaksinthechromatogram[27–33].

Insomecases,irreversibleconformationalchangescanoccurby changingthetemperature.Irreversibletemperature-inducedcon- formationaltransitionsmayhavebeenresponsiblefortheobserved peaksplittingofproteinsunderRPLCconditions[34].

Herein,shortgradientrunswereperformedonthreedifferent columntypesatdifferenttemperatures.Thechangeintheappar- entgradientretentionfactorsofantibodyfragmentsusingdifferent stationaryphases(C18 and C4,core–shelltype silicabased and fullyporoushybrid)wasinvestigated.Thelogarithmofapparent gradientretentionfactorslog(kapp)wasplottedasa functionof 1/T.AllfragmentsofthethreedifferentmAbsexhibitedsignificant deviationfromtheconventionallinearleast-squaredmodel.More surprisingly,therelationshipbetweenretentionandtemperature dependedonthestationaryphaseemployed.Fig2providesarep- resentativeexampleofthechangeinretentionbehaviorofheavy andlightchainofrituximabfragments.Experimentaldatashows anobviousconcavecurvaturewhenlog(kapp)isplottedasafunc- tionof1/T,andthedatacanbereadilyfitwithpolynomialfunctions (e.g.,quadraticmodel).Thechangeintheretentionbehaviorisalso verysimilarfortheheavychainfragmentonallstationaryphases,

whereas thelight chainshows differentbehavior dependingon thenatureof thecolumn.WhenusingtheBEH300 C4material, thechangeintheretentionasfunctionof1/Tappearstobemore linearthanwhenusingtheBEH300C18andAerisC18columns.

Thus,theintrinsicpropertiesofthestationaryphase(e.g.,surface physicochemicalproperties)canplayanimportantroleatelevated temperatures.Anon-linearrelationshiponsilica-basedcolumnsis oftenexplainedbytheso-called“phasetransition”phenomenon [21],which isdue toa conformationalchangein thestationary phase fromasolid-like (lowtemperature)toa liquid-like(high temperature)state.Inmanycases,thistransitioncanbediffuse andoccuroveralargetemperaturerange.

Avarietyofphenomenamayberesponsibleforthenon-linear relationshipbetweenlog(k)and1/Tofpoly-chargedanalytes(mAb fragments),including:(i)possibleconformationalchangesinthe protein,(ii)changesinthedissociationrateofthefunctionalgroups withtemperatureand(iii)thephasetransitionofthestationary phase.Withinthetemperaturerangeinvestigated(40–90^◦C),the changeinapparentretentionfactor(log(k_app))ofmAbfragments asafunctionof1/Tcanbedescribedbythequadraticmodels.Chro- matographicoptimizationsoftwarecommonlyuseslinearmodels forthelog(k)−1/Tdependencebasedontwoinitialruns.Asdis- cussedearlier,thisapproachcannotbeappliedfortheantibody fragments.

3.1.2. Theeffectofgradienttime(gradientsteepness)onthe retention

InRPLC,theinteractionbetweenthestationaryphaseandthe analyte is mediated predominantly through hydrophobic inter- actionsbetweenthenonpolaraminoacidresiduesofmAbsand theimmobilizedn-alkylligands.DuringRPLCanalysis,thegradi- entelutionmodeispreferred,whereinthesolutesareelutedin orderofincreasingmolecularhydrophobicity.Becausethereten- tionofmAbsisstronglydependentonsmallchangesinthesolvent strength,verysmallchanges(<1%)intheorganicmodifiercontent couldleadtoasignificantshiftintheretention.Thus,isocraticcon- ditionsareimpractical,andgradientelutionisarequirementfor theseparationofrealmAbsamples.

InRPLC,theLSSmodelisthewidelyacceptedfordescribingthe retentionofanalytesasafunctionofthevolumefraction(˚)of theBsolvent.Forthegradientelutionmode,thefollowinggeneral equationcanbewritten:

logk∗=logkw−S∗ (2)

wherek*isthemedianvalueofkduringgradientelutionwhen thebandhasreachedthecolumnmid-point,k_wisthevalueofk inpurewater,Sisaconstantforagivencompound(slopeofthe curve)and˚*isthecorrespondingvalueof˚.Thedependenceof k*onthegradienttime(t_g)isalsotypicallypresented,whichcan bedescribedusingthefollowingequation[35,36]:

k∗= tg

1.15t0S (3)

wheret0isthecolumndeadtime.Forpracticalreasons,modeling softwaresuchasDryLabtransformsthevariablesofkork*into log(k)orlog(k*)whenbuildingamathematicalmodel.According toEqs.(2)and(3),log(k*)shouldexhibitalineardependencewhen plottedagainstthelogarithmofgradienttime(whichisrelatedto thegradientsteepness)inthecaseof“regular”samples.

TheLSSmodel generallyprovidesagood descriptionfor the retentionbehaviorofnumeroustypesofanalytes.Insomecases, deviationfromalinearmodelcanbeobserved.Becausetheconfor- mationofproteinscanvaryduringtheelution,linearrelationships cannotbeused.Theeffectofgradientsteepness(gradienttime)on theretentionofheavyandlightchainfragments,aswellastheFab andFcfragments,ofthethreemAbswasinvestigated.Thegradient

whencreatingalinearmodel.Forthequadraticfit,alltheexperi- mentalpointsweretakenintoaccount.Then,retentiontimeswere modeledonthebasisoftwomodelsfordifferentgradienttimes (i.e.,3,10and20min).Fig.3Cdemonstratesthata1.8%difference betweenthepredictedandobservedretentiontimeswascalcu- latedforthe3mingradient,whereasthedifferencebetweenthe twomodelswas2.3and2.1%forthe10-and20-mingradients, respectively.

Inconclusion,wecanconfirmthattheLSSmodelcandescribe the retention behavior of antibody fragments in the RPgradi- entelutionmodewithinapproximately1.0–2.5%error,whichis reasonable.However,whenmoreprecisepredictionisrequired, quadraticmodelshouldbeconsidered.

3.2. CreatingatwodimensionalquadraticDryLabmodel

Theoptimizationsoftwarepackagesgenerallyemployalinear modelforthesimultaneousoptimizationoft_gandT.Thepolyno- mialrelationshipoftwovariablescanbewrittenas:

y=b0+b1x1+b2x2 (4) whereyistheresponse(retentiontimeoritstransformation),x₁ andx₂ arethemodelvariables,e.g.,tgandT,whereasb₀,b₁,b₂ arethemodelcoefficients.Asobservedpreviouslywithantibody fragments,itispreferredtousethequadratic modeltoachieve maximumaccuracyinthepredictionofretentiontimes.Ageneral quadraticmodelfortwovariablescanbewrittenas:

y=b0+b1x1+b2x2+b11x²₁+b22x²₂+b12x1x2 (5) TheDryLabsoftwarewasusedforfurthermethodoptimization bycreatinganewtwodimensionalmode.Retentiontimeswere transformedintoretentionfactors,andthequadraticmodelwas chosenforbothvariables (temperatureandgradienttime).This modewasmodeled ona rectangularregionin the tg−T plane determinedby3temperatureand3 gradienttimes(steepness).

Inthiscase, themodelrequires themeasurement oftheeffects ofthevariablesat3levels.Hence,thisapproachislikea3²fac- torialdesign,which needs9initialexperimentalrunstocreate amodel.Thegradienttimewassettotg₁=4min,tg₂=8minand tg3=12min,whereasthetemperaturewasvariedfromT1=70^◦C, toT₂=80^◦CandT₃=90^◦C.Fig.4showstheschematicrepresen- tationoftheexperimentaldesignutilizedinthetwo-dimensional quadraticmodel.Afterperformingtheinputexperimentalruns,the data(retentiontimes,peakwidthsandpeaktailingvalues)were importedintoDryLabandpeaktrackingwasperformed.Subse- quently,anoptimizationwascarriedoutontheresolutionmaps, whereinthesmallestresolutionvalue(Rs)ofanytwocriticalpeaks inthechromatogramwasplottedasafunctionoftwosimultane- ouslyvariedexperimentalparameters.

Fig.4.Experimentaldesignofthetwo-dimensionalquadraticmodelfortheopti- mizationofmAb(IgG)fragmentRPLCseparation.

Theaccuracyofthis2dimensionalquadraticmodelwasver- ified by comparing the predicted and experimentally derived chromatograms(retention timesand resolution) under optimal conditions.

3.3. Accuracyofretentiontimeandresolutionpredictions 3.3.1. OptimizationoftheseparationofFabandFcfragments

PapainisathiolproteasethatcleavesIgGantibodiesattheheavy chainhingeregionintothreefragments:oneFcandtwoidenti- calFabfragments (Fig.1).Theseparation ofthesedomains has facilitatedtheinvestigationofthemicro-heterogeneityofhuman monoclonalantibodies,includingtheconfirmationofchemicaland post-translationalmodifications,suchastheN-terminalcycliza- tion, oxidation, deamidation, and C-terminal processed lysine residues[37,38].Thisstudydescribesafastandefficientmethodfor thedeterminationofvariantsanddegradationproductsofarecom- binantmAb(bevacizumab)fromacommercialsolution,usingthe separationpowerofnewwide-porecore–shelltypecolumns(Aeris Wideporeof 150mm long).ThenativemAb wasdigestedwith papainandtheaimofthemethoddevelopmentwastoseparate asmanyvariantsoftheFabandFcfragmentsaspossible,within theshortestachievableanalysistime.Threeinitialgradientswith differentslopeswerecarriedout atthreecolumn temperatures (asdescribedinSection2.3.3).Fig.5providesthechromatograms obtainedduringthe9initialruns.Notethatrelativelylargedevia- tionsinthepeakareas(andsumofpeakareas)areexpectedwhen trackingthepeaksbecauseofrecoveryissueswithlargeantibody fragmentsatlowtemperatures. Moreover,therecoveryofthese fragmentsdependsontheirmolecularweight(size).Incontrast,the reproducibilityofretentiontimes,derivedfromconsecutiveruns ataconstanttemperature,wasexcellent.Theresultispresented inFig.6asaresolutionmap.Asshown,the11-mingradientwas foundtoprovidethehighestresolutionwhenthecolumntemper- aturewaskeptat90^◦C.RPLCanalysiswasthenperformedusing theoptimumpredictedconditions,andtheresultingexperimen- talchromatogramsareprovidedinFig.7,alongwiththepredicted data.

The accuracy of the quadratic approach (using 4-, 8- and 12-minbasicgradientrunswithB=10%)wasevaluatedusing the150mm×2.1mmcolumn.Thepredictedandexperimentally derivedchromatograms(retentiontimesandresolution)arecom- pared in Table 1, which reveals good agreement between the

Fig.5.Experimentalchromatogramsofthe9initialruns(BevacizumabFcandFabfragments).Column:AerisWPC18(150mm×2.1mm),injectedvolume:0.5␮l,detection:

fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:from30%to40%B,flowrate:

0.35ml/min.Gradienttimeandtemperatureweresetas4min,70^◦C(A),8min,70^◦C(B),12min,70^◦C(C),4min,80^◦C(D),8min,80^◦C(E),12min,80^◦C(F),4min,90^◦C(G), 8min,90^◦C(H)and12min,90^◦C(I).Peaks:1–3:pre-Fcpeaks,4:Fc,5,6:post-Fcpeaks,7–9:pre-Fabpeaks,10:Fab,11–13:post-Fabpeaks.

Table1

Experimentalretentiontimesandresolutionsvs.predictedfromthetwo-dimensionalgradienttime–temperaturequadraticmodelofbevacizumabfragments(FcandFab).

Peaks Retentiontime Resolution

Experimental Predicted Difference^a Abserror%^b Experimental Predicted Difference^a Abserror%^b

Pre-Fc1 2.99 2.97 0.02 0.67 5.34 4.93 0.41 7.68

Pre-Fc2 3.45 3.41 0.04 1.16 1.11 1.27 −0.16 14.41

Pre-Fc3 3.60 3.54 0.06 1.61 0.84 0.96 −0.12 14.29

Fc 3.68 3.64 0.04 1.11 2.02 2.46 −0.44 21.78

Post-Fc1 3.89 3.85 0.04 0.95 0.58 0.65 −0.07 12.07

Post-Fc2 3.96 3.92 0.04 1.09 7.22 6.84 0.38 5.26

Pre-Fab1 4.78 4.75 0.03 0.54 2.89 2.42 0.47 16.26

Pre-Fab2 5.15 5.08 0.07 1.30 0.5 0.6 −0.10 20.00

Pre-Fab3 5.24 5.17 0.07 1.39 0.67 0.82 −0.15 22.39

Fab 5.35 5.3 0.05 0.92 1.36 1.25 0.11 8.09

Post-Fab1 5.49 5.45 0.04 0.75 1.01 0.57 0.44 43.56

Post-Fab2 5.60 5.54 0.06 1.00 1.02 0.95 0.07 6.86

Post-Fab3 5.70 5.69 0.01 0.18

Average 0.97 Average 16.05

aDifference=experimental−predicted.

b%error=[(experimental−predicted)/predicted]×100.

theretentiontimeswas∼1.0%,whichisconsideredanexcellent predictionusingsuchrapidgradientprofiles.Themeanerrorinthe predictedresolution(Rs)was16.1%.Theerrorintheresolutionval- uescontainstheretentiontimeerroraswellastheuncertaintyof peakwidthandpeaksymmetryprediction.Thus,thispredictionis consideredreliableandthesuggestedfastgradientruns(4,8and 12minfora150-mmlongnarrowborecolumn)canbeappliedin routinework,resultinginsignificanttimesavings.Inthiscase,the timespentformethoddevelopmentwasapproximately8h(3gra- dienttimes×3temperatures×3samples).Thepredictedmethod wasthenexperimentallyverifiedandthefinalseparationrequired onlyan11-minlineargradient,whereas aseparation ofsimilar qualityusingconventionalcolumnswouldrequireatleast60min.

Byusingthemostadvanced,highlyefficient150-mmlongnarrow

differentcolumntemperatures(asdescribedinSection2.3.3).

Fig.8providestheexperimentalchromatogramsofthe9basic inputruns.Theresultsarepresentedintheformofaresolution mapinFig.9,whichrevealsthattheseparationperformedat81^◦C usinga14-mingradientprovidesthehighestresolution.Thepre- dictedoptimumconditionswerethenevaluatedexperimentally.

Theresultsdemonstratethatthepredictedretentiontimeswere practicallyidenticalwiththoseobtainedexperimentally,withan averageerrorintheretentiontimesof0.5%(seeTable2).Thereso- lutionwasalsopredictedwithsufficientaccuracy(averageerror of ∼20%).In this manner,computer-assisted simulation canbe appliedwithhighprecisionfor theseparation ofantibodyfrag- mentsbyusingtwodimensionalquadraticmodels.Fig.10provides acomparisonbetweenthepredictedandexperimentallyobtained

Fig.7.PredictedandexperimentalchromatogramsofBevacizumabFcandFabfragmentsoptimizedbyquadraticmodel.Column:AerisWPC18(150mm×2.1mm),injected volume:0.5␮l,detection:fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:

from30%to40%B,flowrate:0.35ml/min.Gradienttime:11min,T=90^◦C.Peaks:1–3:pre-Fcpeaks,4:Fc,5,6:post-Fcpeaks,7–9:pre-Fabpeaks,10:Fab,11–13:post-Fab peaks.

Fig.8. Experimentalchromatogramsofthe9initialruns(RituximabLCandHCfragments).Column:AerisWPC18(150mm×2.1mm),injectedvolume:0.5␮l,detection:

fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:from30%to40%B,flowrate:

0.35ml/min.Gradienttimeandtemperatureweresetas4min,70^◦C(A),8min,70^◦C(B),12min,70^◦C(C),4min,80^◦C(D),8min,80^◦C(E),12min,80^◦C(F),4min,90^◦C(G), 8min,90^◦C(H)and12min,90^◦C(I).Peaks:1:pre-LCpeak,2:LC,3–5:post-LCpeaks,6.7:pre-HCpeaks,8:HC,9–10:post-HCpeaks.

Fig.9. Two-dimensionalresolutionmapofthecolumntemperature(^◦C)against gradienttime(tg,min)fortheseparationofRituximabLCandHCfragments.

chromatograms.Inthisexample,themethoddevelopmentprocess tookwaslessthan6h.

3.3.3. Comparisonofquadraticandlinearmodels

Asshown in Section 3.1,the changein the retention factor (log(k)) oftheantibody fragmentscaused by temperature(1/T)

cannot be described accurately using linear functions,whereas the log(k_app) vs. gradient time log(t_g) exhibits a nearly linear correlation. Accordingly, the accuracy of the two dimensional linear and the combined linear-quadratic model were com- pared to that of the quadratic model. In the linear model, 4 initial runs were considered, namely: tg₁=4min, tg₂=12min and at T₁=70^◦C, T₂=90^◦C. In the combined model, 6 initial runs were considered, and the effect of gradient time was investigatedat2levels(tg₁=4min,tg₂=12min),whereasthetem- perature wasevaluated at3 levels(T₁=70^◦C, T₂=80^◦C and T₃

=90^◦C).

ByusingthelinearmodelfortheexampledescribedinSection 3.3.1,theaverageerrorintheretentiontimepredictionwas4.1%, whereasthecombinedmodelproducedanaverageerrorof1.5%

(∼1%errorwasobservedwiththequadraticmodel).Forthesec- ondexample(seeinSection3.3.2),thelinearmodelproducedan averageerrorof3.8%,whereasthecombinedmodelproduceda1.7%

errorfortheretentiontimeprediction(∼0.5%errorwasobserved withthequadraticmodel).Inconclusion,thecombinedmodelcan alsoprovideacceptableaccuracy(∼1.5–2%error)andrequiresonly

Fig.10.ExperimentalandpredictedchromatogramsofRituximabLCandHCfragmentsoptimizedbyquadraticmodel.Column:AerisWPC18(150mm×2.1mm),injected volume:0.5␮l,detection:fluorescence(excitationat280nm,emissionat360nm).MobilephaseA:0.1%TFAinwater,mobilephaseB:0.1%TFAinacetonitrile.Gradient:

from30%to40%B,flowrate:0.35ml/min.Gradienttime:14min,T=81^◦C.Peaks:1:pre-LCpeak,2:LC,3–5:post-LCpeaks,6.7:pre-HCpeaks,8:HC,9–10:post-HCpeaks.

Table2

Experimentalretentiontimesandresolutionsvs.predictedfromthetwo-dimensionalgradienttime–temperaturequadraticmodelofrituximabfragments(light-andheavy chain).

Peaks Retentiontime Resolution

Experimental Predicted Difference^a Abserror%^b Experimental Predicted Difference^a Abserror%^b

Pre-LC 2.38 2.41 −0.03 1.18 6.37 8.73 −2.36 37.05

LC 2.96 2.98 −0.02 0.57 4.95 5.35 −0.40 8.08

Post-LC1 3.49 3.5 −0.01 0.29 2.59 3.01 −0.42 16.22

Post-LC2 3.82 3.82 0.00 0.13 1.63 2.69 −1.06 65.03

Post-LC3 4.03 4.05 −0.02 0.45 15.21 17.04 −1.83 12.03

Pre-HC1 5.92 5.93 −0.01 0.12 0.89 0.87 0.02 2.25

Pre-HC2 6.06 6.04 0.01 0.25 0.75 0.87 −0.12 16.00

HC 6.15 6.14 0.01 0.18 1.79 1.41 0.38 21.23

Post-HC1 6.41 6.43 −0.02 0.27 7.3 7.19 0.11 1.51

Post-HC2 7.40 7.53 −0.14 1.83

Average 0.52 Average 19.93

aDifference=experimental−predicted.

b %error=[(experimental−predicted)/predicted]×100.

6initialruns.Thus,thetimerequiredfortheinitialrunscanbe shortenedbyafactorof1/3(6runsvs.9runs).

4. Conclusion

ThecurrenttrendsinRPLCanalysisofmAbsinvolvetheuseof limitedfragmentation(proteolysisandreduction),whichisbased ontheseparationoftheheavyandlightchainvariantsaswellasthe Fab–Fcfragments.Therecentadvancesinwide-porecolumns,such asthehighlyefficient,fullyporoussub-2␮mBEH300orcore–shell typeAerisWideporecolumns,offerhighresolutionpowerforthe separationoftheselargefragments.

Fortunately, these mAb (IgG) fragments elute in a relative narrowgradientrange(e.g.,30–40%acetonitrileatelevatedtem- perature),whichmeansthatthemethodoptimizationprocedure doesnot require a wide range of scouting gradients(e.g., 0to 100%B). In general,it wasfoundthatinitialgradientsof 30to 40%Bwereappropriate formodeling theretentionbehavior of thefragments.Inaddition,itwasshownpreviouslythattheuse of elevated temperatures (e.g., 60–90^◦C) is necessary because of recovery issues due to column adsorption of the IgG frag- ments.Accordingly,theeffectoftemperatureonselectivitywas investigated usinga limited temperature range (e.g.,70–90^◦C), eliminating the need to perform initial gradient runs at low temperatures.