approach exchange in cation chromatography, variants Part I: Saltgradient Method development the separation for of monoclonal antibodycharge Journal of Pharmaceutical and Biomedical Analysis

(1)

ContentslistsavailableatScienceDirect

Journal of Pharmaceutical and Biomedical Analysis

jou rn al h om e p a g 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

Method development for the separation of monoclonal antibody charge variants in cation exchange chromatography, Part I: Salt gradient approach

Szabolcs Fekete

^a,∗

, Alain Beck

^b

, Jen ˝o Fekete

^c

, Davy Guillarme

^a

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

bCenterofImmunologyPierreFabre,5AvenueNapoléonIII,BP60497,74160Saint-Julien-en-Genevois,France¹

cBudapestUniversityofTechnologyandEconomics,DepartmentofInorganicandAnalyticalChemistry,Szt.Gellérttér4.,1111Budapest,Hungary

a r t i c l e i n f o

Articlehistory:

Received2July2014

Receivedinrevisedform27August2014 Accepted27August2014

Availableonline8September2014

Keywords:

Ionexchange Saltgradient Monoclonalantibody Methoddevelopment Cetuximab

a b s t r a c t

Ionexchangechromatography(IEX)isahistoricaltechniquewidelyusedforthedetailedcharacterization oftherapeuticproteinsandcanbeconsideredasareferenceandpowerfultechniqueforthequalitative andquantitativeevaluationofchargevariants.WhenapplyingsaltgradientIEXapproachformono- clonalantibodies(mAbs)characterization,thisapproachisdescribedastime-consumingtodevelopand product-speciﬁc.Thegoalofthisstudywastotacklethesetwobottle-necks.

BymodelingtheretentionofseveralcommercialmAbsandtheirvariantsinIEX,weprovedthatthe mobilephasetemperaturewasnotrelevantfortuningselectivity,whileoptimalsaltgradientprogramcan beeasilyfoundbasedononlytwoinitialgradientsofdifferentslopes.Lastbutnotleast,thedependence ofretentionvs.pHbeingpolynomial,threeinitialrunsatdifferentpHwererequiredtooptimizemobile phasepH.Finally,only9hofinitialexperimentswerenecessarytosimultaneouslyoptimizesaltgradient proﬁleandpHinIEX.Thedatacanthenbetreatedwithcommercialmodelingsoftwaretoﬁndoutthe optimalconditionstobeused,andaccuracyofretentiontimespredictionwasexcellent(lessthan1%

variationbetweenpredictedandexperimentalvalues).

Second,wealsoprovedthatgenericIEXconditionscanbeappliedforthecharacterizationofmAbs possessingawiderangeofpI,from6.7to9.1.Forthispurpose,astrongcationexchangecolumnhasto beemployedatapHbelow6andusingaproportionofNaClupto0.2M.Undertheseconditions,allthe mAbswereproperlyelutedfromthecolumn.Therefore,saltgradientCEXcanbeconsideredasageneric multi-productapproach.

1. Introduction

Duetotheincreasingnumberofapprovedmonoclonalantibod- ies(mAb)inthepharmaceuticalareaandthenumberofbiosimilars potentiallyenteringthemarket,theneedforanalyticaltechniques adaptedfortheirdetailedcharacterizationhasincreased[1].The intrinsicmicro-heterogeneityisofmajorconcernwithmAbsand shouldbe criticallyevaluated becausedifferences in impurities and/ordegradationproductscouldleadtoserioushealthimplica- tions[2].

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

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

1 http://www.cipf.com.

In the production of mAbs, the final product often exhibits a number of variations from theexpected or desired structure [3]. These alterations may result from either known or novel types of posttranslational modifications or from spontaneous, non-enzymaticproteindegradationleadingtochargeand/orsize heterogeneity. Common modifications of theprimary sequence includeN-glycosylation[4],methionineoxidation[5],proteolytic fragmentation, and deamidation [6,7]. It has been shown that charge variants of therapeutic proteinscan have very different bioactivity[8].The complete characterizationof an intactmAb isdifficulttoachieve;therefore,variousenzymes,suchaspepsin andpapain,areoftenusedtoobtainmAbfragmentsandfacilitate theinvestigationofitsmicro-heterogeneity[9].Papainisprimar- ilyusedtocleavemAbsintothreefragmentsattheheavychain (HC)hingeregion,onecrystallizablefraction(Fc)andtwoidentical antigen-bindingfraction(Fab)fragmentsof∼50kDaeach,while http://dx.doi.org/10.1016/j.jpba.2014.08.035

(2)

pepsingeneratesF(ab)2fragments of∼100kDa.These typesof digestionareoftencalledlimitedproteolysis(LP).Thereduction ofdisulﬁdebondsisalsocommonlyusedtoproducetwoLCsand twoHCs.

Ingeneral,theidentity,heterogeneity,impuritycontent,and activityofeachnewbatchofmAbsshouldbethoroughlyinves- tigated before release. This examination is achieved using a widerangeofanalytical methods,includingionexchangechro- matography(IEX),reversed-phaseliquidchromatography(RPLC), hydrophobic interaction chromatography (HIC), size exclusion chromatography(SEC),sodiumdodecylsulfatepolyacrylamidegel electrophoresis(SDS-PAGE), capillary isoelectric focusing (cIEF), capillary zone electrophoresis (CZE), circular dichroism (CD), Fouriertransforminfraredspectroscopy(FT-IR),ﬂuorescencespec- trophotometry(FL),andmassspectrometry(MS).Thegoalofthis multi-methodstrategyistodemonstratethesimilaritybetween productionbatchesofmAbbypreciselycharacterizingtheprimary, secondary,andtertiarystructureofthemAbs[10,11].

Consideringthelargesizeofantibodiesandtheminorstruc- turaldiversitybetweenthevariants,theexistenceofthesevariants imposesa great challengefortheirchromatographicseparation [3]. IEX chromatography is a non-denaturing technique widely used for the separation and isolation of protein charge variants for subsequent characterization. Among the different IEX modes,cation-exchangechromatography(CEX)isoneofthebest approachesformAbpuriﬁcationandcharacterization[12].Cation exchangechromatographyisconsideredasthegoldstandardfor chargesensitiveantibodyanalysis,howevermethodparameters, suchascolumntype,mobilephasepH,andsaltconcentrationgra- dient,oftenneedtobeoptimizedforeachindividualantibody[13].

Inthe90s,Moorhouseetal.wereamongthefirsttodemonstrate thepotentialofCEXformAbcharacterization[14].FabandFcfrag- mentswereseparatedwithsufficientresolutionandidentifiedwith MS.TheC-terminallysinevariabilityoftheFcandtheN-terminal glutamine-pyroglutamate variability of the Fab were observed.

AmorerecentstudyshowedthesuitabilityofCEXforstudying complexdegradationprocessesinvolvingvariousIgG1molecules [15].Thismethodwasparticularlyusefulforcharacterizingpro- tein variantsformedin thepresence of saltsunder accelerated storageconditions.Theimportanceofthisassaywasfurtherillus- tratedbycharacterizationoflight-induced degradationsofmAb formulations.AnotherstudypresentedthesigniﬁcanceofIEXin theanalysisofoxidizedmAbsamples[16].Inadditiontocation- exchange,anion-exchangechromatography(AEX)wasalsoapplied andfoundsuitablefortheseparationofthemorebasicoxidized variantsofintactmAbs[16].

Inthelate1970s,chromatofocusing(with internalpHgradi- ent)wasrecognizedasthechromatographicanalogytoIEF[17–19].

Chromatofocusinghasbeendemonstratedtobeusefulforseparat- ingproteinisoformsduetoitshighresolvingpowerandabilityto retaintheproteinnativestate[20,21].Therearehoweversomelim- itationstothistechniquesuchasthecostofpolyampholytebuffers, columnregenerationtimeandtheinﬂexibilityincontrollingthe slopeofpHgradient[20,22,23].Alternatively,pHgradientcanbe conductedexternallybypre-columnmixingoftwoelutingbuffers atdifferentpHvaluesconsistingofcommonbufferspecies[3].The externallyinducedpHgradienthasrecentlybeenappliedforsepa- rationofdeamidatedvariantsofamAb,resolvingC-terminallysine isoformsofamAbaftertreatingwithcarboxypeptidaseBandalso fortheanalysisofchargevariantsofintactmAbs[6,20,24].Accord- ingtotheliterature,ionicstrengthbasedIEXseparations(classical saltgradientmode)haveexcellentresolvingpowerandrobustness, butareproductspeciﬁcandtime-consumingtodevelop[24].On theotherhand,pHgradientbasedseparationusingaCEXcolumn isdescribedasamultiproductchargesensitiveseparationmethod formonoclonalantibodies[3,24,25].

ThehighcomplexityoftheanalyticalworksrelatedtomAbs requiresnew,moresimpleandgenericwaysofjudgingthequality andthevariabilityofthequalityofmAbproducts.Thisisoneof themajorrequirementsforHPLCmethodsaccordingtoQualityby Design(QbD)principles.Wecanexpectagreatstepforwardthe simpliﬁcationofHPLCmethodstoahigherdegreeofﬂexibilityin lifescience,especiallyinpharmaceuticaldevelopmentwork.For thispurpose,computerbasedchromatographicmodelscanexplain complexdetailsmuchfasterandhelpsinjudgingthequalityofmAb products.

In this contribution, the possibilities and limitations of the classicalsaltgradientapproachisstudiedanddiscussed.Theappli- cability of salt gradientsas a generic multi-product method is presentedfor 10therapeutic mAbs,approved bothbytheFood andDrugAdministration(FDA)andtheEuropeanMedicineAgency (EMA) and possessing a wide range of isoelectric points (pI) between6.7 and9.1.Uptonow,themethoddevelopmentpro- cedureinionexchangechromatographywasempiricalandtime consuming.Inourwork,theimpactofmobilephasepH,gradient steepnessandtemperatureonretention,peakcapacityandselec- tivitywasstudiedindetailsusingsixselectedmodelmAbsandtheir variants(i.e.trastuzumab,panitumumab,natalizumab,rituximab, adalimumabandcetuximab).Basedontheobservedeffectsonres- olution,asixrunsbasedinitialexperimentalsetupwassuggested forsuccessfulmethodoptimizationinthesaltgradientmode.Per- forminggradientrunswithtwogradienttimesandthreemobile phasepHona100×4.6mmcolumnallowedafastandreliable optimizationoftheseparation.Thisoptimizationwasperformed bycomputersimulation usingDryLabmodeling softwareand a custommademodel.

2. Experimental

2.1. Chemicalsandcolumns

WaterwasobtainedwithaMilli-QPuriﬁcation Systemfrom Millipore(Bedford,MA,USA).1M2-(N-morpholino)ethanesulfonic acid(MES)solution(BioReagent),1Msodiumhydroxide(NaOH) solutionandsodiumchloride(NaCl)(BioChemika)werepurchased fromSigma–Aldrich(Buchs,Switzerland).FDAandEMAapproved therapeuticIgG monoclonalantibodiesincludingpanitumumab, natalizumab, cetuximab, bevacizumab, trastuzumab, rituximab, palivizumab, adalimumab, denosumab and ofatumumab were kindlyprovidedbytheCenterofImmunologyPierreFabre(Saint- JulienenGenevois,France).Papain(fromCaricapapaya),usedfor fragmentationofmAbswasobtainedfromSigma–Aldrich(Buchs, Switzerland).

YMC BioPro SP-F 100×4.6mm, 5␮m non-porous strong cationexchangecolumnwaspurchasedfromStagroma(Reinach, Switzerland).

2.2. Equipmentandsoftware

Alltheexperimentswereperformedusinga WatersAcquity UPLC^TMsystemequippedwithabinarysolventdeliverypump,an autosamplerandﬂuorescencedetector(FL).TheWatersAcquity systemincludeda5␮lsampleloopanda2␮lFLﬂow-cell.Theloop isdirectlyconnectedtotheinjectionswitchingvalve(noneedle seatcapillary).Theconnectiontubebetweentheinjectorandcol- umninletwas0.13mmI.D.and250mmlong(passivepreheating included),andthecapillarylocatedbetweenthecolumnanddetec- torwas0.10mmI.D.and150mmlong.Theoverallextra-column volume(V_ext)isabout14␮lasmeasuredfromtheinjectionseat oftheauto-samplertothedetectorcell.Themeasureddwellvol- umeisaround100␮l.Dataacquisitionandinstrumentcontrolwas

(3)

performedbyEmpowerPro2Software(Waters).Calculationand datatransferringwasachievedbyusingExceltemplates.

ThemobilephasepHwasadjustedusingaSevenMultiS40pH meter(MettlerToledo,Greifensee,Switzerland).

Method optimization was performed using DryLab^® 2000 Pluschromatographicmodelingsoftware(Molnar-Institute,Berlin, Germany).

2.3. Apparatusandmethodology

2.3.1. Mobilephasecompositionandsamplepreparation

ForthegradientseparationofmAbsandtheirfragments,the mobilephase “A”consisted of 10mM MES in water, whilethe mobilephase“B”was10mMMESinwatercontaining1MNaCl.

ThepHofbothmobilephaseswasadjustedbyadding0.1MNaOH solutiontoreachtherequiredpH(pH=5.6,5.7,6.0,6.3,6.4and6.6).

Thedigestionofcetuximabwasinitiatedbyadditionofpapain (dilutedto100␮g/mlwithwater)toreachaﬁnalprotein:enzyme ratioof100:1(m/m%).Thedigestionwascarriedoutat37^◦Cfor3h.

Theﬁnaldigestionvolumewas200␮landdirectlyinjectedusing lowvolumeinsertvials.

2.3.2. Investigationofretentionpropertiesofantibodies

Intactantibodieswereelutedinsaltgradientmode.Forstudying theretentionpropertiesofintactmAbs,6ofthe10availableanti- bodieswereselectedbasedontheirtype(IgGclassandisotype) andcalculatedpI,namelypanitumumab(huIgG2,pI=6.7),natalizumab(hzIgG4,pI=8.6),cetuximab(chIgG1,pI=8.7),adalimumab (huIgG1, pI=8.8), trastuzumab (hzIgG1, pI=8.8) and rituximab (chIgG1,pI=9.1).OurpurposewastocoverthewholepIrangeand toincludechimeric(ch),humanized(hz)andhuman(hu)reference IgG1,IgG2andIgG4isotypesaswellinordertodrawoveralland reliableconclusions.

First,theeffectofsaltgradientsteepnessontheretentionwas evaluated.Differentgradienttimesweretestedatagivenmobile phasetemperatureandpH.Agenericlineargradient,startingfrom 0%to20%B(equivalentwith0–0.2MNaClgradient)wasapplied ataﬂowrateof0.6ml/minforallsamples.Thegradienttime(tg) wasvariedas10,15,20,30and40min(atT=30^◦CandatpH6.3).

Theobservedapparentretentionfactors(kapp)andpeakcapacity (Pc)valueswereplottedagainstthegradienttime(steepness).

Fortheinvestigationofmobilephasetemperature,15mingra- dientrunswerecarriedoutatpH6.3usingvarioustemperatures between30^◦C and the upper temperaturelimit of the column (60^◦C).TheretentionpropertiesofintactmAbsandtheircharge variantswereevaluatedbyplottingthelogarithmkappagainst1/T (Van’tHofftyperepresentation).Peakcapacityandresolutionwere alsostudiedasafunctionofmobilephasetemperature.

TheimpactofmobilephasepHinsaltgradientmodewasﬁnally evaluated byperforming 15min long gradientsat T=30^◦C and varyingthemobilephasepHbetween5.7and6.6.Thecommonly appliedmobilephasepHinsaltgradientCEXseparationofmAbs isbetween6and6.5,thereforeourselectedrangerepresentswell theconditionsofreal-lifeseparations[14,26,27].Again,thekapp,Pc

andresolutionwerestudiedasafunctionofmobilephasepH.

2.3.3. Systematicmethodoptimization

Initialbasicrunsformultifactorialexperimentaldesignswere alreadysuggestedinthe90sforreversedphaseliquidchromato- graphicmethodoptimization[28].Ageneralapproachconsistsin modelingsimultaneouslytheeffectofthemostimportantfactors e.g.gradientsteepnessandtemperatureonselectivitywithapre- viouslyselectedcolumn[29,30].Then,withthehelpofresolution mapsgeneratedbymodelingsoftware–whichshowthecritical resolutionofthepeakstobeseparated[31]–theselectedvari- ablescanberapidlyandefﬁcientlyoptimized.Thisapproachwas

currentlyappliedforthereversed phase separationof antibody variants[32].Inthisstudy,thisprocedurewasimplementedfor IEXsaltgradientbasedseparations.

Basedontheobservedeffectsofthefactorsonretentionandres- olutionofmAbspeaks,asixrunsbasedinitialexperimentalsetup wassuggestedformethodoptimizationinthesaltgradientmode.It wasindeedfoundthattheimpactoftemperatureonselectivityand resolutionwasnotsigniﬁcant.Performinggradientrunswithtwo gradienttimes(ast_g1=10min,t_g2=30min)andthreepH(aspH₁ 5.6,pH₂6.0,pH₃6.4)ona100×4.6mmcolumnallowedareliable optimizationoftheseparation.Theoptimizationwasperformed bycomputersimulationusingDryLabandacustommademodel.

Cetuximabpapaindigestedsampleswereinjectedtobuildupthe DryLabmodelandstudythepredictionaccuracyerror.Cetuximab isa heterogeneousmAbpossessingtwoN-glycosylationsitesin theheavychainandseveralchargevariantsincludingC-terminal lysinesand sialicacids[33].It isthereforeaparticularcomplex exampleformethoddevelopment.

2.3.4. Genericsaltgradientformultiproductanalysis

Afterstudyingtheretentionbehaviorofantibodies,ageneric saltgradientwasproposedthatallowedtheelutionandseparation ofallthetenmAbswithinreasonableanalysistime(20min).Alin- earNaClgradient,startingfrom0%to20%Bwasappliedataﬂow rateof0.6ml/minforallsamples.Themobilephasetemperature wassetatT=30^◦CandpHwasequalto5.7).Fluorescencedetection wascarriedoutatex=280andem=360nm.

3. Resultsanddiscussion

ToachieveanoptimalseparationofmAbs,theinfluenceofvar- ious parameterson separation, suchas pH,gradient steepness, temperatureorflowratehastobetakenintoaccount.Theclas- sicalIEXmodeemploysalinearsaltgradient.Inpreviousstudies, isocratic experiments wereused todeterminethe salt concen- tration dependenceof theproteinretention. Then,thegradient retentiontimeswerecalculatedfromtheestablishedsaltdepend- encemodelusingmathematicalfunctionsdescribingthegradient profile[34–36].

TheworkofSnyderandco-workersshowedthatIEXsystems follownon-linearsolventstrength(LSS)typemechanism[37,38].

Consequently,solute-speciﬁccorrectionfactorsarerequiredtouse LSSmodelforretentionpredictions,therebylimitingtheapplica- bilityoftheLSSmodel.Thenon-linearityofLSSmodelwasassessed bycomparingtheelutiondatatothestoichiometricdisplacement model(SDM)commonlyusedinIEX[39].Theretentionfactor(k) canbewritteninthefollowingwayaccordingtotheSDMmodel:

logk=logK−zlogC (1)

whereKisthedistributionconstant,zisassociatedwiththeprotein netchargeornumberofbindingsites(effectivecharge)andCisthe saltconcentration(thatdeterminestheionicstrength).Thenon- linearityofEq.(1)ismostpronouncedforsmallvaluesofz[38].

Ifz>6,anLSStypemodelmayprovidereliabledataforretention factor(retentiontime).

BesidetheLSSandSDM,severalothermodelsweredeveloped forIEXsuchastheslab[40–42],mechanistic[39]orstericmass actionmodels(SMA)[43].Recently,Schmidtetal.extendedthe SDMtodescribetheretentionbehaviorofamAbinlinearsaltand pHgradientelutioninCEX[44].ProteinretentioninIEXwasalso predictedbystructurebasedmodels[45,46].

Inadditiontosaltgradientprofile(steepness),themobilephase pHandflowratewerealsofoundtobecriticalparametersforthe IEXseparation(purification)ofmAbs.Toobtainsufficientbinding oftheproteinstothecationexchangeresin,thepHshouldbeset

(4)

atleast1pHunitbelowthepIvalue[26].Usually,lowerpHpro- videsmorepositivechargetotheproteinandincreasesretention.

Fromakineticpointofview,lowerflowrateswerefoundtooffer higherefficiencyformonoclonalantibodiesinCEX[47].Therefore, reductionofflowratecouldbeasolutiontoimproveresolutionof chargevariants.

3.1. Theeffectofsaltgradienttime(gradientsteepness)onmAbs retentioninCEX

ForCEXanalysisof mAbs,thegradient elutionmodeis pre- ferredin practice.Thesolutesare elutedin order ofincreasing bindingcharge(correlatesmoreorlesswiththepI)andequilibrium constant.TheretentionofmAbsinsaltgradientmodeisstrongly dependentonthesaltconcentration(gradientsteepness)–dueto therelativelyhighzvalue–andasmallchangecouldleadtosig- niﬁcantshiftintheretention.Forthisreason,isocraticconditions areimpractical,andgradientelutionismandatoryinreal-lifemAb separations.

ForlinearsaltgradientinIEX,thesaltconcentrationvarieswith timeduringthegradient,thereforeEq.(1)canbewrittenas:

logk=logK−zlog

C0+C tg

(2) whereC0isthesaltconcentrationatthebeginningofthegradient andCisthechangeinCduringthegradient.SimilarlytoRPLC, theLSSmodeldescribestheanalyteretentionasafunctionofthe volumefraction(˚)oftheBsolvent.Forgradientelutionmode,the followinggeneralequationcanbewritten:

logk^∗=logkw−S^∗ (3)

wherek*isthemedianvalueofkduringgradientelutionwhen thebandhasreachedthecolumnmid-point,kwisthevalueofkin purewater,Sisaconstantforagivencompound(slopeofthecurve describedinEq.(3))and˚*isthecorrespondingvalueof˚.Itis practicaltoshowthedependenceofk*onthegradienttime(tg).

Forthispurpose,thefollowingequationcanbederived[43,44,48]:

k^∗= tg

1.15t₀S (4)

wheret₀isthecolumndeadtime.Forpracticalreasons,modeling softwaresuchasDryLabgenerallydealwithtransformedvariables ofkork*tolog(k)orlog(k*)tobuildmathematicalmodels.Onthe basisofEqs.(3)and(4),log(k*)shouldfollowalinearmodelwhen plottedagainstthelogarithmofgradienttime(whichisrelatedto thegradientsteepness)incaseof“regular”samples.Forlinearsalt gradientinIEX,asimilarequationfork*canbederived[30]:

k^∗= tg

1.15[t0|z|log(Cf/C0)] (5)

where C_f is theconcentration of thecounter-ionat the end of thegradientprogram.PleasenotethatbothRPLCandIEXsepa- rationsvarywithgradientconditionsinasimilarway.However, becauseofthedifferencesinthedependenceofkonthemobile phase composition C in IEX (log–log relationship)versus RPLC (log–linearrelationship),theLSSmodelistheoreticallynotappli- cableforIEX.Nevertheless,asshowninEqs.(2)and(5),thehigher thez,thelowerthedeviationfromnon-linearityis.Therefore,the LSSapproachmaybeapplicableforlargeproteins(mAbs)possess- inganimportantnumberofcharges.Consequently,thelinearsalt gradientIEXseparationofmoleculeswithz≥3canbedescribed semi-quantitativelybytheLSSmodel[30,37,38].

Theeffectofgradientsteepness(gradienttime)ontheretention ofintactmAbsandtheirvariantswaspracticallyinvestigated.The gradienttime(steepness)wasvariedas10,15,20,30and40min(at T=30^◦CandatpH6.3).TheretentionofthesixselectedmAbsand

thevariantsoftrastuzumab,adalimumabandcetuximabshowed thesamebehavior.Fig.1illustratestheeffectofgradienttimeon theapparentretention(kapp)ofintactmAbsandchargevariants, asrepresentativeexamples.Therelationshipbetweenkappandtg

canbeperfectlydescribedbyﬁttingalinearfunction(R²>0.999 forallsolutes).Thislineartypebehaviorcanbeexpectedforlarge proteinspossessingseveralcharges,butsurprisinglyitseemstobe evenmorelinearinIEXthaninRPLC.Asshownin[32],theretention behaviorofmAbfragmentsshowedaslightdeviationfromlinear relationshipinRPLC.Theexperimentalpointsinlog(kapp)vs.log(tg) representationfollowedaslightlyconcavecurvature,whileitisnot thecaseinIEX.InIEXsaltgradientmode,itcanbeconcludedthata LSStypemodelperfectlydescribestheretentionbehaviorofmAbs.

Thereisnoneedforlogarithmicorpolynomialﬁtting(asitisoften thecaseinRPLCmode).

Theresultssuggest thatonlytwogradientrunsarerequired (e.g.witht_g1=10minandt_g2=30min)fortheoptimizationofa saltgradient.Then,theretentiontimescanbepredictedforany gradientprogram,duetothelinearbehavior.

3.2. TheeffectofmobilephasetemperatureonmAbsretentionin CEX

Theeffectoftemperatureonretentionfactor(k)cangenerallybe expressedinliquidchromatographywiththevan’tHoffequation:

logk=−H RT +S

R +logˇ (6)

where H representsthe enthalpychangeassociated withthe transfer of the solute between phases, S the corresponding entropychange,Rthemolargasconstant,Ttheabsolutetempera- tureinKelvinandˇthephaseratioofthecolumn.Whenlog(k)is plottedagainst1/T,theenthalpyisgivenbytheslopeofthecurve.

Withregularcompounds,thesevan’tHoffplotsfollowalinear relationship.However,aquadraticdependenceoflog(k)versus1/T overawiderangeoftemperaturewasnoticedbydifferentauthors usingsilica-basedaswellasnonsilica-basedstationaryphases[49].

Theeffectoftemperatureontheretentionofpartiallyionizedcom- poundswhichexistintwoforms(i.e.molecularandionizedforms) canalsobewelldescribedwithEq.(6).However,bothenthalpy andentropyareexpectedtobedifferentforthetwoformsandas aresult,bothHandScanvarywithtemperaturewhenbothforms arepresenttoasigniﬁcantextent[49].Withlargebiomolecules, theeffectoftemperatureonretentionsometimesbecomesmore complex.Dependingonthestabilityofthesecondarystructure,the moleculesunfoldtovariousextentsandhenceinteractwiththe stationaryphasewithvariousstrengths[50].Duetothedifferent conformation-dependentresponsesofproteinsatelevatedtem- peratures,thechangeinretentioncanbedifﬁculttoassess[51,52].

InRPLCseparationofproteins,temperatureisausefulparameter foradjustingselectivity.However,forIEXseparationsofproteins, theimpactoftemperaturewasnotreportedintheliterature.Data onthedistributioncoefﬁcientsasafunctionofmobilephasetem- peraturewereonlyreportedforaminoacidsinIEXmode[53].

Fig.2illustratestheobtainedvan’tHofftypeplots.Threeconclu- sionscanbedrawn.First,thelog(k)vs.1/Tshowslinearbehaviorin theinvestigatedtemperaturerange.Second,theslopeofthecurves ismuchlowerthaninRPLCmode.Indeed,theslopesinIEXsaltgra- dientmodewerecomprisedbetween−0.13and0.14K×10³,while forintactmAbsinRPLCmode,theslopesofvan’tHoffcurvesare typicallyaround0.5–1.0K×10³[32].Thissuggeststhattheeffect oftemperatureontheretentionpropertiesofmAbsislessimpor- tantinIEXmodethaninRPLCmode.Finally,theslopeoflog(k) vs. 1/T curves for related mAbs (e.g.charge variantsof a given mAb)ispracticallythesame.Thismeansthatselectivitycannotbe tunedbythetemperature,asillustratedinFig.2Bwiththeplotsof

(5)

Fig.1.Effectofgradienttime(steepness)ontheapparentretentionfactorofnativeantibodies(A)andantibodyvariants(B).Column:YMCBioProSP-F(100×4.6mm).Mobile phase:“A”,10mMMESpH6.3;“B”,10mMMESpH6.3+1MNaCl.Flowrate:0.6ml/min;gradient:0–20%Bin10,15,20,30and40min;temperature:30^◦C,detection:FL (280–360nm);injectedvolume:2␮l.

adalimumab,cetuximabandtrastuzumab.However,temperature hasanimpactonthepeakwidth(peakcapacity)–seeinSection 3.4–andthereforecouldmodifythequalityoftheseparation.

3.3. TheeffectofmobilephasepHonmAbsretentioninCEX Oneoftheprimaryvariables forvaryingIEXretentionisthe mobilephasepH[54].Indeed,retentionisstronglyaffectedbythe ionizationstateofacompound.InCEXmode,lowerpHincreases thenumberofpositivechargeonthemAbandincreasesthereten- tion.

Fig.3shows thedependencyofapparentretentionfactor of nativemAbsonmobilephasepH.MAbspossessingpIofatleast 8.6showlinearrelationshipsinthestudiedpHrange(5.7–6.6)and theslopesoftheapparentkvs.pHfunctionswereverysimilar.

However,panitumumabwithpI=6.7behavesinadifferentway.

Atmobilephase pHbeyond∼6.2, a cleardeviationfromlinear relationshipwasobserved.ClosetothepI,theretentiondecreases drastically,butasecondorderpolynomialfunctioncanappropri- atelyﬁttheobserveddata.Thecurveintersectsthex-axisataround pH∼6.65,thisvalueisveryclosetothecalculatedpIofpanitu- mumab.AtpHequaltopI,noretentionisexpected.Pleasenote,

thatpanitumumabisaspecialcaseasitpossessesanextremely lowpI,comparedtoothermAbs(pIbetween8and9).Therefore, thecommonlyusedpHinsaltgradientmodeisbetween6and6.8 andensuresappropriateretentioninCEXmode.

From theretention modeling of view,all the mAbs have to beconsidered.Chromatographicoptimizationsoftwarecommonly usesecondorderpolynomialmodelstodescribekvs.pHdepend- enceinRPLCmode,basedonthreeinitialruns.Onthebasisofthe resultsobtainedwithintactmAbs,thisapproachisprobablyalso appropriateforthemodelingofmAbsretentioninsaltgradientCEX mode.

3.4. Peakcapacity

Peakcapacityasafunctionofgradienttimeandtemperature wasalsoestimated.Peakcapacityisameasureoftheseparation powerthat includestheentire chromatographicspace together withthevariabilityofthepeakwidthoverthechromatogram.The generalexpressionforpeakcapacity(Pc)inliquidchromatogra- phy,assumesunitresolutionbetweenthesuccessivelyelutedpeaks [55].Inthisstudy,peakcapacitieswereexperimentallydetermined fromthegradienttime(tg)andtheaveragemeasuredpeakwidthat 50%height(w50%).Thefollowingequationwasusedtoestimatethe

(6)

Fig.2. Effectoftemperatureontheapparentretentionfactorofnativeantibodies(A)andantibodyvariants(B)(van’tHofftyperepresentation).Column:YMCBioProSP-F (100×4.6mm).Mobilephase:“A”,10mMMESpH6.3;“B”,10mMMESpH6.3+1MNaCl.Flowrate:0.6ml/min;gradient:0–20%Bin15min;temperature:30,40,50and 60^◦C;detection:FL(280–360nm);injectedvolume:2␮l.

peakcapacitybasedonpeakwidthathalfheight,correspondingto aresolutionofRs=1betweenconsecutivepeaks:

Pc=1+ tg

1.7·w50%

(7)

Toavoidtheimprecisionassociatedwiththemeasurementof peakwidthsaroundbaselineformAbswhichoftencontainclosely relatedvariants(i.e.heterogeneoussample),thepeakwidthwas measuredathalfheightinthisstudy.

Fig.3.EffectofmobilephasepHontheapparentretentionfactorofnativeantibodies.Column:YMCBioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMES;“B”,10mM MES+1MNaCl(pHwassetat5.7,6.0,6.3and6.6).Flowrate:0.6ml/min;gradient:0–20%Bin15min;temperature:30^◦C;detection:FL(280–360nm);injectedvolume:

2␮l.

(7)

Fig.4.Peakcapacityasafunctionofgradienttime(steepness)(A)andtemperature(B).Column:YMCBioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMESpH6.3;

“B”,10mMMESpH6.3+1MNaCl.Flowrate:0.6ml/min;detection:FL(280–360nm);injectedvolume:2␮l.

AsshowninFig.4A,peakcapacity(P_c)of∼50–60wasobserved with10minlonggradientwhilethelongest40mingradientpro- videdPc∼110–130.

Fig.4Bshows the change in peak capacity as a function of mobilephasetemperature.Aslightdecreaseinpeakcapacitywas observedformostmAbs(exceptnatalizumab).Thepeakcapacity foradalimumabchangedfrom75downto60whenincreasingthe temperaturefrom30to60^◦C.Thisobservationsuggeststhatlower temperatureismore favorablefortheIEXseparations ofmAbs.

Becauseselectivitydoesnotchangesigniﬁcantlywithtemperature inIEXandsincepeakcapacityissomewhathigheratlowertem- perature,betterresolutionisexpectedundersuchconditions.This isoppositetowhatiscommonlyobservedinRPLC.Indeed,when analyzingmAbsunderRPLCconditions,atemperatureofatleast 70–80^◦Cisrequiredtohavesharpsymmetricalpeaksandappro- priaterecovery[56].TheimprovementofpeakshapeinRPLCis relatedtotheimprovementofthemasstransferprocessthrough theporousparticles.InthecaseofIEXseparation,anon-porous hydrophilic polymer bead is used, therefore there is no trans- particlemasstransferprocess.Onlyeddydispersion,longitudinal diffusionandexternalﬁlmmasstransfercontributetobandbroad- ening.Thetwolattermaybeaffectedbymobilephasetemperature;

andincreaseintheﬁlmmasstransferandlongitudinaldiffusionis expected.Inourstudy,theimpactoflongitudinaldiffusionisproba- blynotnegligible,sincewedonothavetrans-particlemasstransfer

resistance.Maybe,thisenhancedlongitudinaldiffusioncanexplain theslightdecreaseinpeakcapacitywhenincreasingmobilephase temperature.OtherpossibleexplanationscouldbethechangeofpI andmobilephasepHwithtemperatureandon-columnaggregation atelevatedtemperature.

3.5. CreatingatwodimensionalDryLabmodelforCEX

Optimizationsoftwarepackagesgenerallyemploylinearmod- elsforthesimultaneousoptimizationoftwoorthreevariables.The polynomialequationfortwovariablescanbewrittenas:

y=b₀+b₁x₁+b₂x₂ (8) whereyistheresponse(retentiontimeoritstransformation),x₁ andx₂arethemodelvariablese.g.tgandTwhileb₀,b₁,b₂arethe modelcoefﬁcients.Asalreadydiscusseditisbettertousequadratic modelswithmAbs,tohaveadequatepredictionaccuracyofreten- tiontimesasafunctionofmobilephasepH.Thegeneralquadratic modelfortwovariablescanbewrittenas:

y=b0+b1x1+b2x2+b11x²₁+b22x₂²+b12x1x2 (9) Inthis study,DryLabwasusedfor furthermethodoptimiza- tionanddeterminationoftheunknowncoefﬁcientsofthemodel.

The software implements an interpretive approach, where the retention behavior is modeled using experimental information

(8)

Fig.5.Cetuximabpapaindigestedsample.Column:YMCBioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMES;“B”,10mMMES+1MNaCl.Flowrate:0.6ml/min;

gradient:0–20%B;temperature:30^◦C;detection:FL(280–360nm);injectedvolume:2␮l.Gradienttimes:tg1=10min,tg2=30min,pH15.6,pH26.0,pH36.4.

frominitialruns,andtheretentiontimesatotherconditionsare predictedinaselectedexperimentaldomain.Thisallowscalculat- ingthecriticalresolution,andaccordingly,theoptimalseparation canbefound[57].

AsshowninSections3.2and3.4,temperatureisnotanimpor- tantvariable.Itdoesnotinfluencesignificantlytheretentionand selectivity.Onthecontrary,thesaltgradientsteepnessandmobile phasepHappearasthemostimportantfactorstoadjustselectiv- ityandresolution.Regardingtemperature,itisbeneficialtowork atlowtemperaturetoguaranteethehighestpeakcapacity.Inour optimization,gradienttimeandpHwereselectedasmodelvari- ables,whiletemperaturewaskeptconstantat30^◦C.

A new two dimensional mode was created in DryLab software.Retentiontimesweretransformedtoretentionfactors,and quadraticandlinearmodelswerechosenforpHandgradienttime (steepness),respectively.Themodelingtakesplaceinarectangular regioninthetg–pHplanedeterminedbythreepHandtwogradient times(steepness)values.Then,themodelrequiresmeasuringthe effectsofthevariablesatthreedifferentpHlevelsandtwogradient

times.Hence,thisapproachnecessitatessixinitialexperimental runsforcreatingthemodel.Gradienttimesweresettotg1=10min andt_g2=30minwhilepHwasvariedaspH₁5.6,pH₂6.0andpH₃6.4.

Followingtheexecutionoftheinputexperimentalruns,theﬁgures ofmerit(i.e.retentiontimes,peakwidthsandpeaktailingvalues) wereimportedintoDryLabandpeaktrackingwasperformed.Peak trackingwasperformedonthebasisofpeak areas.Ithastobe mentionedthatinsomecases,slighttendencieswereobservedin peakareasasthedigestionprocesscannotbestoppedcompletely.

ThereforeinthecaseofdigestedmAbsamplesthepeaktracking processmaybemorecomplexthanforcommonsmallmolecules.

Next,theoptimizationwascarriedoutonthebasisofthecreated resolutionmap.Intheresolutionmap,thesmallestresolution(Rs) valueofanytwocriticalpeaksinthechromatogramwasplottedas afunctionoftwosimultaneouslyvariedexperimentalparameters.

Toestablish theaccuracy of this two dimensional quadratic model,thepredictedandexperimentallyderivedchromatograms (retentiontimesandresolution)undertheoptimalconditionswere compared.

(9)

3.6. OptimizationoftheseparationofFabandFcfragmentsof cetuximab

SeparationoftheFcandFabdomainshasfacilitatedinvestiga- tionofthemicro-heterogeneityofhumanmAbs(confirmationof chemicalandpost-translationalmodificationssuchasN-terminal cyclization, oxidation, deamidation, and C-terminal processed lysineresidues[58,59]).Thepresentexampledescribesafastand efficientmethoddevelopment appliedfor thedetermination of chargevariantsofarecombinantmAb(cetuximab),usingsaltgra- dientapproachinCEXmode.ThenativemAbwasdigestedwith papainandtheaimofthemethoddevelopmentwastoseparateas manyvariantsoftheFabandFcfragmentsaspossible,withinthe shortestachievableanalysistime.Thetwoinitialgradientswith differentslopeswerecarriedoutatthreepHvalues.Fig.5shows thechromatogramsofthesixinitialruns.

Thecorresponding resolution map is presentedin Fig. 6.As shown,a17mingradientwasfoundtoprovidethehighestresolu- tionwhenthemobilephasepHis∼5.6.Thepredictedoptimum condition was set and experimental chromatograms recorded.

Fig.7showsthepredictedandexperimentalchromatograms.

Toevaluatetheaccuracyofthisapproach(with10and30min initialgradient runs)appliedfor100×4.6mmcolumn,thepre- dictedand experimentalchromatograms(retentiontimes)were compared(Table1).Thepredictedretentiontimeswereingood agreementwiththeexperimentalones;theaverageretentiontime relativeerrorswas∼1.0%(seeTable1),whichcanbeconsideredas excellent.

Toconclude,thismethodoptimizationapproachcanbecon- sidered as reliable and the suggested initial experiments (10 and 30min gradient on a 100mm long standard bore column at pH 5.6, 6.0 and 6.4) can be applied in daily routine work, resulting in time saving. The time spent for method development in this example was approximately 9h (2 gradient

Fig.6. Cetuximabpapaindigestionresolutionmap(tg–pHmodel).Column:YMC BioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMES;“B”,10mMMES+1M NaCl.Flowrate:0.6ml/min;gradient:0–20%B;temperature:30^◦C;detection:FL (280–360nm);injectedvolume:2␮l.Gradienttimes:tg1=10min,tg2=30min,pH1

5.6,pH26.0,pH36.4.

times×3pH×3samples+equilibration), and then the predicted methodwasexperimentallyveriﬁed.

Howeverpleasenotethatformorecomplexsamples,theopti- mumconditionsforhighresolutionseparationscanbeshiftedto thelowerpHandlongergradienttimeranges.Thereforeforhigh resolutionseparationsanextendedmodelmightbeuseful.

3.7. GenericsaltgradientCEXmethodforvariousmAbs

ThemaincriticismofsaltgradientIEXseparationsisthatitis productspeciﬁcandtime-consumingtodevelop[24].Onthecon- trary,pHgradientbasedseparationusingaCEXcolumnisdescribed asa multi-productchargesensitiveseparationmethodformAb samples[3,24,25].Inthisstudy,wewantedtoprovethatsaltgradi- entseparationisalsosuitableformultiproductseparations(mAbs possessingvariouspI),providedthatusingappropriateconditions.

SimilarlytoSection3.6,thegradientsteepnessandmobilephase pHweresystematicallyvariedtoobtainasuitableseparationof 10mAbs and their variants. In addition, the impactof the salt

Fig.7.Comparisonofpredictedandexperimentalchromatograms.Column:YMCBioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMES;“B”,10mMMES+1MNaCl.

Flowrate:0.6ml/min;gradient:0–10%B;temperature:30^◦C;detection:FL(280–360nm);injectedvolume:2␮l.Gradienttimes:tg=17min,pH5.62.

(10)

Fig.8.Genericsaltgradient.Column:YMCBioProSP-F(100×4.6mm).Mobilephase:“A”,10mMMESpH5.7;“B”,10mMMESpH5.7+1MNaCl.Flowrate:0.6ml/min;

gradient:0–20%Bin20min;temperature:30^◦C;detection:FL(280–360nm);injectedvolume:2␮l.

fractionwasalsostudiedanditwasfoundthat0.2MNaClwassuf- ﬁcienttoelutemAbspossessingthehighestpI(∼9).Ontheother hand,toensureasufﬁcientretentionofmAbswithlowpI(∼6.5), themobilephasepHwaskeptunder6(e.g.pH5.7).

Fig.8showstheobtainedchromatogramsof10intactmAbs, andsuggeststhatsaltgradientCEXseparationmayalsobeade- quateformulti-productmAbseparations.Theoptimalconditions onastrongcationexchangerresinwerefoundas20minlonggra- dient(0–0.2MNaCl)atpH5.7andatamobilephasetemperature ofT=30^◦C.IfmAbswithlowerorhigherpIarealsoincludedinthe productionline,theconditionscanbeadjustedaccordingly.This exampleclearlyshowsthationicstrengthbasedCEXseparationof mAbsisnotproductspeciﬁc,andgenericconditionscaneasilybe foundforseveraldifferenttypesofmAbs.

Table1

Predictionaccuracy.ConditionsarethesameasspeciﬁedinFig.7.

Peak Retentiontime(min)

Experimental Predicted Difference Error(%)

B 2.68 2.67 0.01 0.26

1 5.33 5.26 0.07 1.41

2 5.68 5.60 0.08 1.34

3 6.10 6.03 0.07 1.16

4 6.43 6.38 0.05 0.77

5 6.86 6.85 0.01 0.09

6 7.29 7.27 0.02 0.23

7 7.88 7.90 −0.02 −0.24

8 8.24 8.19 0.04 0.55

9 8.66 8.65 0.01 0.07

10 9.02 9.00 0.02 0.18

11 9.55 9.55 0.00 −0.04

12 11.21 11.22 −0.01 −0.11

13 12.49 12.49 0.00 −0.01

14 14.19 14.16 0.03 0.24

Average 0.02 0.39

4. Conclusion

TherearetwomaincriticismswhenapplyingsaltgradientIEX forthecharacterizationofmAbs.Firstofall,itisgenerallylongand tedioustodevelopaCEXmethodandsecondly,theconditionsare notenoughgeneric(productspeciﬁcconditions).Thegoalofthis studywastotacklethesetwolimitations.

Inaﬁrstinstance,theretentionmodelswhenvaryingmobile phasetemperature,pHandgradientsteepnesswereassessedunder CEXconditions,thankstotherapeuticreferencemAbsand their chargevariants.It appearsthattemperaturewasnota relevant parameterfortuningselectivityandshouldbekeptat30^◦C,to achievehighresolvingpower.Becausetherelationshipbetween apparentretentionfactorsandgradienttime–inIEXmode–can bedescribedwithalinearfunction,onlytwoinitialgradientruns ofdifferentslopesarerequiredforoptimizingthesaltgradientpro- gram.Finally,secondorderpolynomialmodels(i.e.basedonthree initialruns)arerequiredtodescribekvs.pHdependence,forthe modelingofmAbsretentioninsaltgradientCEXmode.Basedon theseobservations,wedemonstratedthatthedevelopmentofa CEXmethodcanbeperformedrapidly,inanautomatedwaythanks toaHPLCmodelingsoftware,usingtwogradienttimesandthree mobilephasepH(e.g.10and30mingradientona100mmlong standardborecolumnatpH5.6,6.0and6.4).Suchaprocedurecan beappliedroutinelyandthetimespentformethoddevelopment wouldbeonly9h.Attheend,thedifferencesbetweenexperimen- talandpredictedretentiontimeswerelowerthan1%,makingthis approachhighlyaccurate.

Secondly,wehavealsodemonstratedthatgenericsaltgradient CEXconditionscanbeappliedforthecharacterizationof10mAbs possessingpIbetween6.7and9.1.Aproportionof0.2MNaClwas sufﬁcient toelute themostbasic mAbs fromthe strongcation exchangecolumn,whileapHlowerthan6wasappropriatetosuf- ﬁcientlyretainthemAbswithlowestpI.Thisexampleshowsthat saltgradientCEXcanbeconsideredasamulti-productapproach