Protein-protein interactions and reduced excluded volume increase dynamic binding capacity of dual salt systems in hydrophobic interaction chromatography

Deploying two salts in hydrophobic interaction chromatography can significantly increase dynamic binding capacities. Nevertheless, the mechanistic understanding of this phenomenon is lacking. Here, we investigate whether surface tension or ionic strength govern dynamic binding capacities of the chromatographic resin Toyopearl Butyl-650 M in dual salt systems.

Journal of Chromatography A 1649 (2021) 462231

ContentslistsavailableatScienceDirect

journalhomepage:www.elsevier.com/locate/chroma

Leo A Jakoba, Beate Beyera,b, Catarina Janeiro Ferreirab, Nico Lingga,b, Alois Jungbauera,b,∗,

Rupert Tscheließniga

a Department of Biotechnology, University of Natural Resources and Life Sciences, Vienna, Muthgasse 18, A-1190, Austria

b Austrian Centre of Industrial Biotechnology, Muthgasse 18, Vienna A-1190, Austria

a r t i c l e i n f o

Article history:

Received 28 February 2021

Revised 26 April 2021

Accepted 28 April 2021

Available online 7 May 2021

Keywords:

HIC

Mixed electrolytes

Dynamic binding capacities

Breakthrough curves

Adsorption isotherms

Self-avoiding random walk

a b s t r a c t

Deployingtwosaltsinhydrophobicinteractionchromatographycansignificantlyincreasedynamic bind-ingcapacities.Nevertheless,themechanisticunderstandingofthisphenomenonislacking.Here,we in-vestigatewhethersurfacetensionorionicstrengthgoverndynamicbindingcapacitiesofthe chromato-graphicresinToyopearlButyl-650Mindualsaltsystems.Small-angleX-rayscatteringwasemployedto analyzethemodel proteinsandthe protein-resinadductintherespectivedualsaltsystems.The dual saltsystemsincorporatesodiumcitrateandasecondarysodiumsalt(acetate,sulfate,orphosphate).As modelproteins,weusedlysozyme,GFP,andamonoclonalantibody(adalimumab)

Moreover,fortheprotein-resinadduct,wedeterminedthemodelparametersofaself-avoidingrandom walkmodelfittedintothepairdensitydistributionfunctionoftheSAXSdata.Ionicstrengthismore pre-dictivefordynamicbindingcapacitiesinHICdualsaltsystemsthansurfacetension.However,dynamic binding capacitiesstill differby upto 30 %betweenthe investigated dualsaltsystems The proteins exhibitextensiveprotein-protein interactionsinthe studieddualsaltHICbuffers.We founda correla-tionofprotein-protein interactionswiththe well-knownHofmeister series.Forsystemswithelevated protein-proteininteractions,adsorptionisothermsdeviatefromLangmuirianbehavior.Thishighlightsthe importanceoflateral protein-protein interactionsinproteinadsorption, wheremonomolecular protein layersareusuallyassumed.SAXSanalysisoftheprotein-resinadductindicatesaninversecorrelationof thebindingcapacityandtheexcludedvolumeparameter.Thisisindicativeofthedepositionofproteins

inthecavitiesofthestationaryphase.Wehypothesizethatincreasingprotein-proteininteractionsallow theformationofattractiveclustersandmultilayersinthecavities,respectively

© 2021 The Author(s) Published by Elsevier B.V ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/)

1 Introduction

Senczuk et al.(2009) describedthe positiveeffect ofso-called

dual saltbuffer systems on dynamic binding capacities (DBC) in

hydrophobicinteractionchromatography(HIC).Thosedualsalt

sys-temsshowedincreaseddynamicbindingcapacitiescomparedtoa

singlesaltsystem[1]which hasbeenconfirmedby other groups

[2– ] Hackemann et al [5] has shown that dual salt systems

can either synergistically increase or decrease binding capacities

in adsorptionisotherms.Altogether, afundamentalunderstanding

∗ Corresponding author at: Department of Biotechnology, University of Natural

Resources and Life Sciences, Vienna, Muthgasse 18, A-1190, Austria

E-mail address: alois.jungbauer@boku.ac.at (A Jungbauer)

ofhowtwo differentbufferspromotebetterbindingthanasingle onehasnotyetbeenprovided.Commonly,akosmotropicbufferis addedtotheproteinsolutiontopromotebinding.Theadditionofa chaotropicsaltwouldbecounterintuitiveaccordingtothecurrent theoryexplainingtheadsorptionofproteinsinHIC[6].BothMüller

etal.[2]andBaumgartneretal.[3]postulatedthatmixinga kos-motropicsaltforpromotingbindingtothehydrophobicstationary phasesurfaceandchaotropicsalt,whichispossiblyincreasingthe proteinsolubility,shouldbethepreferredstrategywhensettingup mixedsaltbuffersystemsforchromatography.Thecurrent under-standingislackingafundamentalexplanationofthemechanism Thesurfacetensionincrementofthesaltinthebinding buffer andthesaltinginandoutpropertiesgoverntheadsorptionof pro-teins in HIC,as describedin the solvophobic theory [6] In

gen-https://doi.org/10.1016/j.chroma.2021.462231

0021-9673/© 2021 The Author(s) Published by Elsevier B.V This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )

eral,thistheory describestheinteraction behavior ofamore

po-larsolvent,in thiscasethe mobilephase andalesspolarsolute,

thesampleprotein,byconsideringthechangesinthesystem’sfree

energycausedby theindividualprocessesinvolved.Thestructural

forcesofwaterformedbyhydrogenbonding,inthiscontext,

rep-resent a low energy state In contrast, the water molecules near

the stationaryphase’s hydrophobic surfaceare inan energetically

"loaded" state.Theproteinbindingtothehydrophobicsurface

re-duces the surfacearea incontact withthe water molecules.The

energy released as a consequence of this can be describedas a

function of thechange in available free surfacearea A andthe

surfacetensionofthemobilephaseγ:

This means that the retention in both reversed-phase

chro-matography and HIC increases with the mobile phase’s surface

tension [6,7] Based on this concept, higher hydrophobic energy

andthusahighersurfacetensionofthemobilephaseshouldalso

translateintohigherproteinbindingcapacitiesofthecolumn

Another parameterthat could influenceretention andbinding

capacityinHICisionicstrength.Thisparameterdescribesthetotal

concentration ofionsinasolution.Thus,itcanbevastlydifferent

forsolutionscontainingidenticalmolarconcentrationsofdifferent

saltsdependingonthevalencesofthesaltsinquestion.Theionic

strength Iofa solutioncanbecalculatedbasedonthe Lewisand

Randallequation:

I=1

2

n



i

with n representingthe number of ions inthe solution, i

repre-sentingone specificion, ibeingthecorresponding concentration

ofioniinmol l−1 ,and idenotingthevalenceofioni

In orderto determine the ionicstrength, theconcentration of

the ions has to be determined using the Henderson-Hasselbalch

equation,definedas:

pH=p K a+log [A−]

where [HA] isthe molar concentration ofthe unassociated weak

acid and [A−]is themolar concentration oftheacid’s conjugate

base

Apart frominteractions between theprotein andthe HIC

sta-tionary phase [8,9],it is well knownthat ions modulate

protein-proteininteractions [10–14].Althoughspeculationsabout

protein-proteininteraction-basedmultilayerformation[15]andcluster

for-mation [16] can be found in literature, experimental evidence is

scarce for those phenomena in HIC.However, interactive protein

clusters have already been reported for other surfaces Langdon

et al [17] showed that attractive protein-protein interactions

re-sponsible for cluster formation of BSA on a hydrophilic surface

In the case of the presence of protein-protein interactions, the

Langmuir adsorptionisotherm model isno longervalid sincethe

non-interactivity of the adsorbate is a prerequisite for its

appli-cability[18].Meng etal.[19]haveshownthat theisotherm type

shifted betweenLangmuirand Freundlich type dependingonthe

saltconcentration.Moreover,theyhavehypothesizedthat

protein-protein interaction is responsible for Freundlich type isotherms

Besides Freundlich type isotherms, the Brunauer-Emmett-Teller

(BET) theorydescribesmultilayeradsorptionprotein

chromatogra-phy[20,21]

As an analyticaltool,small-angle x-rayscattering(SAXS) gives

a unique insightinto the native solution structure ofproteins It

allows the investigation ofthe intramolecularand intermolecular

structure ofproteins,suchasthemedium resolutionprotein

con-formation [22,23]andprotein-proteininteractions[12,14],

respec-tively More recently, SAXS has beenutilized for online

monitor-ingofthe proteinlayerthickness[24] andbinding conformations [25] in chromatographic systems In classical polymer chemistry, SAXS experiments allow the characterizationof polymers Fractal models can be used to describe linear and branched polymers,

characterizingthepolymer’sinter-monomerconformational distri-bution.Thisincludesseveralparameters,suchastheexcluded vol-umeand the path length in-between the monomers [26] In this work,wemodelthechromatographicresinasaself-avoiding ran-dom walk (SARW) with andwithout proteins bound The result-ing parameters are then interpreted to gain an understanding of thebindingtopology.Theseexperimentsareperformedwithresin slurriesusingapipettingrobot[27]

Asmodelproteinsforthisstudy,amonoclonalantibody (adal-imumab), lysozyme, and Green Fluorescent Protein (GFP) were used,since they havepreviously been describedindual salt sys-tems Senczuk et al postulated that their observations mightbe dueto specific interactions of theantibodies withthe stationary phase[1].LysozymewasfirstproposedbyMülleretal.asan addi-tionalmodelproteinforstudyingdualsaltbuffersystems.Ithasa basicpI(10.7[28]),similartomostmonoclonalantibodies[2]and adalimumab’s (7.9-9.1[29]) GFP was addedbecause of its acidic range(pI=5.8[30]).Thus,iftheclaimofincreasedbinding capac-itywithdualsaltsystemsalsoholds forGFP,thiswouldstrongly indicatethat thepIofthe sampleprotein doesnot influence sta-tionaryphasebindinginmixedsaltsystems.Furthermore,the cho-sen model proteins differ significantly in regards to their molar mass,havingmolarmassesof14.3kDa (lysozyme[28]), 26.9kDa (GFP[30])and148kDa(adalimumab[31])

Ultimately,thisstudyaims toidentifywhethersurfacetension

or ionic strength is the primary driving force fordynamic bind-ingcapacitiesinHIC.Forthatpurpose,wepreparedcitratebuffers containingasecondarysalt(acetate,phosphate,orsulfate)and var-ied theconcentrationsof thesesaltsto obtainbuffers with iden-tical surface tension Dynamic binding capacities of a Toyopearl Butyl-650M HIC column were determined for the systems with identicalsurfacetension.Similarly,wepreparedbufferswithmore

or less the same ionic strength by variation of the citrate con-centration.Forthosesystems,theequilibriumanddynamic bind-ingcapacities weredetermined.SAXS wasusedtoinvestigatethe impacton themodel proteinsolutionstructure (suchasthe pro-tein structure and protein-protein interaction) and the protein-resintopologywhenboundtothechromatographicresin.For mod-ellingtheprotein-chromatographicresinadduct,wehavederiveda SARWmodelthat wasthen fittedtothe pairdensitydistribution function(PDDF)oftheadduct

2 Theory

2.1 SARW model

We followthearguments ofHammouda[26],Zimm [32],and Beaucage[33].Weconsideralinearpolymerchainfirst;itconsists

ofn elements First, we define a segment of reference It can be any segment, i The probability of finding another segment, j of

thesamemoleculeis[26]:

π1

i j(r)=4πr2 (3/2 πr−2 )3 /2 exp

−3/2r2 r−2 

Then, we link the inter-segment distance, r, and the average inter-monomer distance,   We follow Hammouda and put it

2 =a2 |i − j|2 ν [26]. Herein resembles the inter-segment

dis-tance,andν givestheexcluded volumeparameter whilea isthe statisticalsegmentlength Ifweputtheexcluded volume param-eterto 1, we get the probability to find two pairs i, j of a non-self-avoiding randomchain.It is easy toshow that the Eq.(4)is normalizedtoone,∫∞

0 drπ1

i j(r)=1.Thelinearpolymerchainis fi-nite and consists ofN segments; still following the argument of

L.A Jakob, B Beyer, C Janeiro Ferreira et al Journal of Chromatography A 1649 (2021) 462231

Zimm,wegivethePDDFofthisparticularconstruct:

p(r)=∫N

0 dn(N − n) π1

The norm of it equals =∫∞

0 dr p(r)=N2 /2 It seems incor-rect asfromanyN segmentlong chain,random, orrandom

self-avoidingcanpairN(N-1)/2nonidenticalsegments.Thus,wecorrect

thenormandfindthePDDF:

p(r)= (N− 1)

The equationisstill inappropriateasinnonidenticalpairs,the

lowerboundaryoftheintegrationovernmustreadoneandnot0

ThentheappropriatePDDFreads:

p(r)= N N

− 1

N

∫

1 dn(N − n) π (n) π1

Pleasenoteoneimportantthing.Thesegmentsareequally

dis-tributed, π (n)=1 What if they are not? What if specific

seg-ment pairs are not to be taken into account? What if the

seg-ments are fractally distributed, and their probability is given by

π (n)=(nλ )c? We follow the arguments of Hammouda [26], we

introduce a fractal distribution of n Moreover, we compute the

norm:

∞

∫

0 dr p c λ( .|r)=λc



c+1− N

c+2− N c+1 

(c+1) (c+2) (8)

It is straightforward to show that in =0, the norm equals:

N(N-1)/2 We proceed and give pair density of a self-avoiding

random walk explicitly Therefore, we introduce a set of

ab-breviations: α=c − ν2 +1

ν , α=α+1 ν, β= 3 r2

2 b2, β=βN−ν, γ =

3

π3 b r32( c2 +3 c+2 )(1 −N ) N1− 32ν

ν( N + c+1 ) , and then find for the PDDF for an

ensemble ofself-avoiding random walks, withfractal distributed

pairs:

p c

λ(b, N,ν, c|r)=γN c+2 

E α

β

−E α

β

+N32ν(N E α( β )−E α( β ) )

(9)

Therein E n(z)=∫∞ 1 dt t −n exp(−zt) is the exponential integral

function

π (n)=(nλ )c accounts, within the integral for the average

numberofminimumpathswithapathlengthn[3]

2.2 Chromatographic stationary phase as a SARW

If we embed a random walk in a spherical volume, we

as-sumethat asphericalPDDFdistributestheminimumpaths’

aver-agenumberwitha pathlengthn.Think ofasphere thatisfilled

by randompoints,up toinfinitedensity.Then anyrandomly

cho-senpairwillhaveaminimumpaththatequalstheirEuclidnet

dis-tance.Thisistrueforahypotheticalresinabsentofanypore.The

introductionofporesandtheirdecorationbyproteinsisthen

mea-surable by the difference in their particular PDDF We introduce

thenormalizedprobabilitytoidentifyminimumpathsoflengthn,

r∝λn ,andR∝λN

π (n)=λ−1



3n5

16N6 −9n3

4N4 +3n2

N3



(10)

Finally,we obtain thePDDFfora hypotheticalresin.It

resem-blesaresinabsentofpores

p SARW(b, N,ν|r)∝1/16/ N6 p c

λ(b, N,ν,5|r)+3/4/ N4 p c

λ(b, N,ν,3|r)

+1/ N3 p c

WiththePDDF describing theSARWmodel(Eq.(11)),the ex-perimental PDDF p(r) can be fitted The fitting procedure min-imizes the difference between the experimental PDDF and the PDDFdescribingtheSARWbyadjusting

minargr p(r)−

ap SARW(b, N,νr)+c B r D

While parameters a, cB, and Dare due to the norm and the overallstochastic background,parametersb,Nandν characterize thesystem’smorphologyonasmallerscale

3 Material & methods

3.1 Buffer preparation

The saltsused forthe bufferstested inthe experiments were suppliedbyMerck (Germany)andwereall ofanalyticalgrade.All bufferswerepreparedfromstocksolutionsof1.5Mofsodium cit-ratemonobasic,1Mofsodiumphosphate,0.6Mofsodiumsulfate,

2Mofsodiumacetate,andthenadjustedtopH6withNaOH.The specific dual saltmixtures of 0.329 M of citrate + 0.5M of sul-fatewerepreparedfroma0.8Msodiumsulfatestocksolution.The bufferpreparationwasfollowedbyfiltrationusinga0.22μmfilter suppliedbyMerckMillipore(Ireland)

3.2 Model proteins

Lysozyme was obtained from Merck in the crystalline state GFP and the antibody were produced in-house and kept aslow ionicstrengthstocksolutionsat4°Cfortheexperiments’duration GFP was previously expressed in E coli and purified in a three-step chromatographic process In contrast, the monoclonal anti-body(mAb),an in-houseproducedadalimumab,wasexpressedin CHO and purified solely by protein A capture Forthe SAXS ex-perimentsanalyzing theproteininsolution,themonoclonal anti-body was purified using a HiLoad 26/600 Superdex 200 pg (Cy-tiva,Sweden).Themodelproteins havebeenanalyzed with high-performance size exclusion chromatography (HP-SEC) The corre-spondingchromatogramscanbefound intheSupplementary Ma-terial(Fig.S1)

3.3 Measurement of surface tension

The surface tension measurements were performed using the pendantdrop(PD)method,anopticalmethodfordeterminingthe surfacetensionofadropofliquidbyusingthedropprofile’s cur-vature The measurements of the differentsalt bufferswere per-formed using the Drop Shape Analyzer (Krüss, Germany) instru-ment The determinationof the surfacetensionusing the PD re-quiresthedroptobedistortedbygravity,whichisensuredby us-ing a tip large enough to support the neededdrop size (in this case, the needle had a diameter of 1.835 mm) Water was used

asareferenceatthebeginningofall setsofexperiments.Its sur-face tension is between 72 and 73 mNm−1 , depending on the surrounding temperatureand humidity conditions.The measure-ments were repeated at least three times each (each one is al-ready the average of one minute of measurements) The system wasalways flushedwiththeintended test buffer between differ-entbuffers’measurementsforfifteenminutestoensurethatthere were no traces of other buffers left in the tubes As determined

byapycnometer,boththebuffers’densityandthetemperatureof theroom weremeasured andtakenintoaccountby thesoftware KrüssAdvanced(Krüss,Germany)togetthemostaccurateresults possible

Forobtainingbufferswithcomparablesurfacetension,the sur-facetensionvaluemeasuredfor0.55Mcitratewasusedasa ref-erencepoint Theother buffers’ saltconcentrations,aspreviously

Table 1

Surface tension of the buffers used by Senczuk et al [1] (left-hand side), buffers with adjusted salt concentrations that resulted in similar surface tension values (right-hand side)

Starting Buffers as used by Senczuk et al [1] Surface Tension [mN ∗ m −1 ] Buffers with adjusted salt concentrations to achieve similar surface tension Surface Tension [mN ∗ m −1 ]

describedbySenczuk et al.,wereadjusted toachieveeithera

de-creaseoranincreaseinsurfacetension,whichwasthenconfirmed

by pendant drop measurements Based on these measurements,

thebufferslistedinTable1wereusedforchromatographic

exper-iments

3.4 Measurement of dynamic binding capacities

Dynamicbindingcapacitymeasurementsforproteinsamplesin

the different highsalt bufferswere performedusing a Toyopearl

Butyl-650 M(TosohBioscience,Germany)column.A4.8× 0.5cm

column with a column volume (CV) of 0.94 mland a 1.3× 1.0

cmcolumnwithaCV of1.02 mlwereused forthe breakthrough

(BT) experiments To test packing quality,1 % acetone (v/v) was

injected to evaluate the peak asymmetry.The asymmetry ranged

from1.2-1.6.Allchromatographicexperimentswerecarriedouton

anÄKTATM Pure25chromatographysystem(Cytiva,Sweden)

3.4.1 Column packing

A 10/20 tricorn columnhousing (Cytiva, Sweden) waspacked

withTOYOPEARLButyl-650M(Tosoh Cooperation,Japan)resin

us-ing50mMphosphatebufferwith1MofNaClaspackingbuffer.A

5 mlmin−1 flowratewaschosenforpackingbased onthe

man-ufacturer’s instruction manual Once the packing operation was

completed, the column wasequilibratedwith 5 – 10CVs of low

ionic strength buffer (50 mM ofphosphate buffer).While not in

use,bothcolumnswerestoredin20%(v/v)ethanolatroom

tem-perature

3.4.2 Breakthrough curves and calculation of DBC

All samples were transferred into the corresponding high salt

buffer beforethe experiment eitherby resolubilizing the

crystal-lized protein in the buffer (in the case of lysozyme) or diluting

the sampleproteinfroma stocksolution (forthe mAbandGFP)

Thestocksolutionconcentrationsweresetsothattheproteinwas

diluted atleast 1:5 inthe experimental buffer to achieve a final

load concentration ofapproximately5 g l−1 The precise

concen-tration ofthesamplesolutionwasthendetermined

spectrophoto-metricallybymeasuringtheabsorbanceat280nm

Forthechromatographicruns,thecolumnwasfirstequilibrated

inthecorrespondinghighsaltexperimentbuffer.Theflowratefor

the loading step was set to achieve a residencetime of 10 min

Sampleloadingwasfollowedbya5–10CVwashstepwiththe

ex-perimentbuffer.Forelution,alineargradientfrom0-100%Bwas

performedwithwaterasbufferBover10CV,followedby5CVat

100 % bufferB.ForcolumnCIP, 0.1M NaOHwasused.All

exper-iments were performedin atemperature-controlled room witha

temperaturerangingfrom21–25°C

For DBC calculations, the load’s absorbance value was

deter-minedinaby-passexperimentontheÄktasystem.Thisvaluewas

then treated asa 100 % breakthrough.The volume wasthen

de-termined,atwhich 10% oftheabsorbancevalue at100 %

break-through was reached (loaded volume 10%BT) Absorbance at 10 %

breakthrough was below 1 AU forall breakthrough experiments

From the volume at 10 % breakthrough, the void volume of the

columnandsystemweresubtracted.Theresultingvaluetimesthe concentrationoftheload( load)dividedbythevolumeofthe col-umnwastreatedastheDBCat10%breakthrough(DBC 10%):

DB C10% = (loadedvolum e10% BT −voidvolume)∗c load

3.5 Calculation of buffer ionic strength

The tested buffers’ ionic strength was calculated using Eqs (2) and (3) For preparing buffers with comparable ionic strengths,theionicstrengthvalueobtainedfor0.55Mcitratewas again used as a reference point The salt concentrations of the other bufferswereadjusted tomatchthat value.Since significant amountsofNaOHhadtobeusedtoadjusttheexperimentbuffers

toapHof6,thisalsohadtobeconsidered.Basedonthese calcu-lations,the bufferslistedinTable 1were usedforthe chromato-graphicexperimentsinvestigatingionicstrengthasapossible driv-ingforce

3.6 Adsorption isotherms

The procedure for the adsorption isotherms was based on a previous publication [25] Protein stock solutions were prepared

by mixing a concentrated protein stock (> 60 mgml−1 ), dH2 O, andsaltstocksolutionstoachievethedesiredbuffercomposition andaproteinconcentrationofapproximately7mgml−1 .The pro-tein stocksolution wasthen further diluted ina 96 UV Star Mi-croplate(GreinerBio-One,Austria)toachieveafinalconcentration range of0.5 mgml−1 – 5 mgml−1 with a total of tendifferent concentrations.Beforeaddingthechromatographicresin,theresin slurrywassettotheconcentrationof50%andwashedtwotimes withdH2 Oandsixtimeswiththe corresponding buffer 50μl of the 50 % slurry were added to the protein solutions to achieve

a totalvolume of250 μl anda slurryconcentration of10 % The chromatographicresinandthecorrespondingmodelproteinwere incubated for 24 h on a thermomixer (Thermo Fisher Scientific, Waltham, MA) at 950 rpmand 21.5°C The resultingsupernatant wasanalyzedspectrophotometricallyviaabsorbanceat280nmto determinetheproteinconcentration.When theplateauinthe ad-sorptionisothermwasnotreached,additionalmeasurementswere performedwitha3 4.5mgml−1 mobile phaseconcentrationat

a resin concentration of 5 % Adsorption isotherms incorporating suchdatapointsaremarkedinthecorrespondingfigure

TheLangmuir(Eq.(14))[18],BET(Eq.(15))[20]andFreundlich (16) [19] models were used to describe the adsorption isotherm data:

q= c q max ∗K a

1+q max ∗K a

(14)

q= (1− K L c q)mono(1− K K s c L c+K S c) (15)

whereqdescribesthebindingcapacityinmgproteinpermlresin,

c the mobile phase concentration in mgml−1 , qmax the maxi-mum bindingcapacityinmgproteinper mlresin,Ka the affinity

L.A Jakob, B Beyer, C Janeiro Ferreira et al Journal of Chromatography A 1649 (2021) 462231

constant oftheproteintowards thestationaryphaseinmlmg−1 ,

qmono thebindingcapacityofamonolayer,KS theaffinityconstant

towards thestationaryphase (equivalentto LangmuirKA ), KL the

affinity constant towards depositedlayers [20],KF the adsorption

constantinmlmg−1 andnF theadsorptionexponent[19]

Inthecaseofa distinctplateau,theLangmuirisotherm model

was used to fit the data Data with a second liftoff was fitted

with the BET adsorption isotherm model Data that showed

nei-ther asecond liftoff nor aplateauwasfitted withtheFreundlich

isotherm The fitted adsorption isotherm model was evaluated

basedontrendsinresiduals.Sinceprotein-proteininteractionmust

not be negligibleforthevalidity oftheLangmuirmodel[18]and

presentinthecaseoftheBETmodel[21],protein-protein

interac-tionswereevaluatedfromSAXSanalyticsofthemodelproteinsin

solution(Section2.7.1)

3.7 SAXS

All SAXS experiments were performed at the Elettra

syn-chrotron in Trieste,Italy.The scatteringvectorq (q = 4π sin(ϴ)

λ−1 , where ϴ is the scattering angle) ranged from 0.896–6.998

nm−1 atawavelengthofλ=0.154nm.Allproteinsolutionswere

preparedfromdH2 O,protein,andsaltstocksolutions.Therecently

described high throughput robot was used for all SAXS

experi-ments[27]

3.7.1 Proteins in solution

Theresultingproteinconcentrationwas5mgml−1 forthe

pro-teins’measurementsinsolution

20 μl oftheprotein solutionwaspipetted into themeasuring

cell,andatotalof12 imageswere measured.Foreachimage,the

exposuretimewas10 followedbya2 pausebetweenevery

im-age.Foreachsample,therespectivebufferwasmeasuredwithout

ananalyteforbackgroundsubtraction

3.7.2 Protein-chromatographic resin suspension

Forthesuspensionexperiments,theproteinconcentration was

5mgml−1 ,andthechromatographicresinslurrywaspreparedas

described in Section 2.6 The model proteins were GFP and the

monoclonalantibody.Theadsorptionexperimentswereperformed

at a proteinconcentration of 5mgml−1 anda slurry

concentra-tion of5% toachieve thechromatographicresin’s fullsaturation

Thereactionwasconductedin2mlEppendorfreactiontubes

(Ep-pendorf GmbH,Germany)ata totalvolumeof1ml Thereaction

was incubatedfor15 h on a thermomixer(Thermo Fisher

Scien-tific, Waltham, MA) at 900rpm androom temperature After

in-cubation, theresin slurrywasbriefly washed twotimeswiththe

respective buffer Forthemeasurements, the slurryconcentration

wassetto40%.Thesampleswerepreparedintriplicates

Forthe measurement,25μl of aslurrysuspension was

pipet-ted intothemeasuring cell.Toincrease thethroughputandkeep

thetime betweentheproteinincubationandtheactual

measure-ment to aminimum,20 imageswere recordedin atotal time of

20s Theexposuretime was950msforeach image,followedby

a 50ms pause betweenthe measurements.For eachsample, the

respectivebufferwasmeasuredwithoutananalyteforbackground

subtraction

3.7.3 Data treatment

Data evaluationwas performedusingthe program

Mathemat-ica 12.1 (Wolfram Research, Inc., USA) Intensities were averaged

over all 20 images for the sample and the background,

respec-tively.Afternormalizationat4.95-5.05nm−1 ,thebackgroundwas

subtracted from the scattering data, resulting in the background

corrected scatteringdata Qvalues ofdistinctive features and

re-gions ofthereciprocalspacewereconvertedtothereal-spacevia

Eq.17([34])

d=2 π

whered isthe real-space distance innm andq is thescattering vectorinnm−1

3.7.4 Plotting of the background-corrected scattering data

For the measurements of the protein in solution, the background-correctedscatteringdatawerenormalizedtoq=0.55

nm−1 andplottedtofacilitatethecomparisonofthelowandhigh

q-range For the measurements of the protein-chromatographic resin suspension, the background-corrected scattering data were normalizedto q = 0.09nm−1 Thecurves ofthe triplicates were stackedbymultiplyingtheintensityby1,101, and102 ,respectively,

tofacilitatethecomparisonbetweenthemeasurements

3.7.5 Pair density distribution function

ThePDDFp(r)ofscatteringdatawascalculatedvia aninverse Fouriertransform[35]:

I(q)=4πDmax ∫

I(q)isthescatteringintensityatthescatteringvectorq.Dmax is themaximumdimensionofcorrelated pairsand isthedistance betweenthecorrelatedpairs

The scattering data of the protein-chromatographic resin sus-pension was transformed to fit the SARW model The scattering data after background subtraction (Ie(q)) was fittedto the PDDF p(r)viaEq.(19):

whereI(q)iscalculatedaccordingtoEq.(18)tofindthePDDF de-scribingourdata(p(r)).Theminimumoftheargumentwas deter-mined byapplying theMathematicaFindArgMinfunction Only0

≤ p(r)were accepted inthe inverseFouriertransform Dmax was setto70andp(r)containedatotalof70datapoints(r=1,2,3… 70).Thisfittingprocedureresultedinexcellent fitsthroughoutall protein-chromatographicresinsuspensionexperiments,asseenin theoverlayoftheexperimentaldataandtheproducedfit (Supple-mentaryMaterial,Fig.S3,left-handside)

TheresultingPDDF(p(r))isthen furtherusedtofittheSARW modelderivedinSection3.Again,thedifferencebetweenp(r)(the experimental PDDF) and the PDDF of the SARW model is mini-mized (Eq (12)) Minimization is achievedby applyingthe Find-ArgMinfunction.Thisresultsinconsiderablygoodfitsfordistances

upto45nm(SupplementaryMaterial,Fig.S3,right-handside) Forcalculation ofthe theoreticalscatteringcurves, theatomic coordinates ofthe PDBsoflysozyme (1dpx), an IgG1monoclonal antibody(1hzh), andGFP (1gfl) were usedto calculate the theo-reticalPDDFbysummingupallpairdistancesofallatoms.The in-tensitieswerecalculatedforeveryscatteringanglebetween0.896 and3.000 nm−1 according to Eq.(18) The theoretical scattering curveswere usedasa benchmarkforattractive andrepulsive in-teractionsinthelowq-range

4 Results & discussion

4.1 Determination of buffer surface tension

ThebufferstestedinSenczuk et al. (2009)werereplicatedand their surface tension was measured (Table 1) Since the surface tensionvaluesvariedgreatlybetweenbuffers,theconcentrationof oneofthesaltsinthedual saltmixtures wasadjusted until sim-ilarsurfacetensionvalueswerereachedusingthesurfacetension measured for0.55 Mcitrate (73.5 mNm−1 ) as a referencepoint

Fig 1 Breakthrough curves for lysozyme (A, left) and mAb (B, right) at a sample concentration of 5 mg ∗ ml −1 using different buffer systems with com parable surface tension

as the mobile phase and a TOYOPEARL Butyl-650 M HIC column DBC was determined for a residence time of 10 min

andtargetvalue Basedonthesemeasurements, thebufferslisted

in Table1 (right-handside)were then chosenasthe appropriate

buffersforchromatographicexperiments forcomparingthe

bind-ing capacities ofa HICcolumn whendifferentdual saltmixtures

withsimilarsurfacetensionareusedasthemobilephases

At first glance, it might seemcounterintuitive that for two of

the dual salt buffer systems (citrate + sulfate and citrate +

ac-etate), the addition of0.3 M or0.5 M of the secondary salt

re-sulted in surface tension values that are almost identical to the

one obtained for 0.55 M citrate alone In this context, it has to

be statedthat the surfacetension of a mixed salt systemis not

thesumofthecontributionsoftheindividualsaltspresentinthe

mixture Instead ofbeing additive, the mixture’s surfacefree

en-ergy, which determines thesurfacetension,is reducedby an

ex-cessofthe componentwiththelower surfacefreeenergy,which

isenrichedinthesurfacelayer[36].Inadualsaltmixture,thesalt

withthelowersurfacetensionincrementdeterminesthemixture’s

surfacetension.Thisphenomenonwasalsoobservedby

Baumgart-neretal.Itledthemtostatethatintheirmixturesofkosmotropic

andchaotropic salt, "thesurfacetension seemsto be more

influ-encedbythechaotropicsalt"[3]

This behavior is also the reason why it was not possible to

achieveasurfacetensionvaluemoresimilartothereferencepoint

for the mixtureof citrate andphosphate, evenby further

reduc-ing the concentration ofphosphate presentinthe solutiondown

to 0.1 M It was, therefore, decided to keep the concentration

of phosphate at its original value of 0.5 M in order to have a

meaningful amountofsecondarysaltinthe solutionandinstead,

slightlydecreasetheamountofcitrateinthebuffer,whichresulted

in a surface tension value still within the acceptable rangeof ±

1mNm−1

4.2 Binding capacity in buffers with equal surface tension

Based on therelationship describedin Eq.(1), itcould be

ex-pectedthatdifferentbuffersatthesamepHandwithsimilar

sur-facetensionvalueswouldhavethesamehydrophobicenergyand,

hence,leadtothesamedynamicbindingcapacityoftheHICresin

This expectation was put to the test by measuring the dynamic

bindingcapacityofaToyopearlButyl650-Mcolumnforlysozyme

(Fig.1A)andthemAb(Fig.1B)inbreakthroughexperiments

us-ingthedualsaltbufferswithcomparablesurfacetension(Table1)

asmobilephases.Table2providesalistwiththeDBCvalues

cal-culatedat10%BTforalltheindividualcurves

Forallthedualsaltsystemsinvestigatedintheseexperiments, the measured binding capacity was noticeably higher than for citrate alone The resulting DBC values varied strongly between the differentbuffers(Fig.1 andTable 2) While thisconfirms, to somedegree,previousobservationsofdualsaltsystemsleadingto higherbindingcapacitiesinHIC,theresultsarestillslightly differ-enttowhatSenczuketal.reported.Ourstudyofthedualsalt sys-temwithphosphateasasecondarysaltdoesnotleadtothelargest increaseinbindingcapacity,aswaspreviouslyreported[1].Among the dual salt systems investigated, higher binding capacities did not correlatewiththe slightdifferences inbuffer surfacetension remainingafterconcentration adjustment.Therefore,itseems un-likelythat thesesmallvariationsinsurfacetensionare thecause fortheobservedphenomenon

4.3 The ionic strength of the buffers

Theresultsdescribedintheprevioussectionindicatedthatthe surfacetensionofthemobilephasesolutionmightnotbethe de-cisive influencing factor when it comes to the dynamic binding capacities of a HIC column Thus the influence of ionic strength

onproteinbindingwasinvestigated.Thesaltconcentrationinthe buffersystemswasadjustedtoionicstrengthvaluescomparableto thereferencebuffer(0.55McitratepH6.0)

Eqs.(2)and(3)wereused tocalculatetheionicstrength The citrate concentration in the buffers was then adjusted to get a valuethatcloselymatchedthereference(ionic strengthof3.1M) Forthe buffer containingthesecondary saltsulfate,we have de-cidedtoadjustthesecondarysaltconcentrationto0.5Mtomatch thesecondarysaltconcentrationofalldualsaltsystems.SincepH adjustmenttopH6.0requiredtheadditionofsignificantamounts

ofNaOH, which,whentakeninto account,led tothe newcitrate concentrationsandionicstrengthvalueslistedinTable3

4.4 Binding capacity in buffers with equal ionic strength

TheDBC wasstudiedwithlysozyme,GFP,andmAbatsample concentrationsofapprox.5mgml−1 (Fig.2).Dynamicbinding ca-pacitiesdiffersubstantially betweenthemono-anddualsalt sys-tems(Table4) ForlysozymeandGFP,thebreakthroughcurvesof dualsaltsystemsgroupclosertogether.FormAb,dynamicbinding capacitiesdiffervastlydependingonthesecondarysalt.Altogether, differencesare lesspronounced compared tothe buffersofequal surfacetension,especiallyinthecaseoflysozyme.Allproteins ex-hibitthe lowestbinding capacityin themono saltbuffer 0.55M

L.A Jakob, B Beyer, C Janeiro Ferreira et al Journal of Chromatography A 1649 (2021) 462231

Table 2

Comparing capacities at 10 % BT for lysozyme, mAb, and GFP when solubilized in buffers sharing comparable surface tension The DBC was determined for a residence time of 10 min Differences between the lowest and highest binding capacities are shown, where either all buffers or only dual salt buffers are compared to each other

Buffer Buffer Surface tension [mN ∗ m −1 ] DBC 10% for lysozyme [mg ∗ml −1 ] DBC 10% for mAb [mg ∗ml −1 ]

Table 3

New citrate concentrations calculated to achieve dual salt systems sharing the same ionic strength con- sidering the citrate buffer as a reference (3.1 M)

Buffer Citrate concentration [M] Ionic strength after pH adjustment [M]

Citrate + 0.50 M Acetate 0.463 2.9 Citrate + 0.50 M Phosphate 0.441 2.8 Citrate + 0.50 M Sulfate 0.329 2.8

Fig 2 Breakthrough curves for lysozyme (A, top left), mAb (B, top right) and GFP (C, bottom left) at a sample concentration of approx 5 mg ∗ ml −1 using different buffer systems with matching ionic strength as the mobile phase and a TOYOPEARL Butyl-650 M HIC column DBC was determined for a residence time of 10 min

sodium citrate The breakthroughcurves with the secondary salt

sulfateinducethehighestdynamicbindingcapacitiesforlysozyme

andGFP,whereasitranksclosesecond formAb.Besides,itis

dif-ficult to deduce trends forthe investigated systems, andfurther

analyticsareneededtogainbetterunderstandingofdrivingforces

governingbindingtothestationaryphase

4.5 Adsorption behavior, internal structure, protein-protein interactions, and binding topology in buffers with equal ionic strength

The breakthrough experiments showed that ionic strength seems to be the more decisive factor for the DBC Nevertheless, ionicstrengthaloneisnotsufficientlydescribingthephenomenon Therefore,we haveconductedSAXS and adsorption isotherm

ex-Table 4

Comparing capacities at 10 % BT for lysozyme, mAb, and GFP when solubilized in buffers sharing comparable ionic strength The DBC was deter-

mined for a residence time of 10 min Differences between the lowest and highest binding capacities are shown, where either all buffers or only

dual salt buffers are compared to each other

Buffers DBC 10 % for lysozyme [mg ∗ml −1 ] DBC 10% for mAb [mg ∗ml −1 ] DBC 10% for GFP [mg ∗ml −1 ]

periments to investigate possibleexplanations forthe differences

indynamicbindingcapacities.Firstly,wehypothesizethatthe

pro-tein structure could be alteredin the respective buffer, resulting

ineitheran expandedorcollapsedconformation.Thiswouldthen

result in modulation of the protein’s footprint on the

chromato-graphicresinandthereforecausedifferencesinthedynamic

bind-ing capacities Alternatively, protein-proteininteractions could be

responsible for modulating the surface coverage, allowing closer

packingwhenprotein-proteininteractionsareattractiveandlooser

packing when protein-protein interactions are repulsive,

respec-tively.Moreover,attractiveprotein-proteininteractioncouldtrigger

multilayerformation.Inordertoinvestigatetheinternalstructure

andintermolecularinteractions,themodelproteinswereanalyzed

via SAXS Furthermore, adsorption isotherms were performed to

evaluate the impact of protein-proteininteraction on protein

ad-sorption.Lastly,theprotein-resinadductwasanalyzedusingSAXS

Theself-avoidingrandomwalkmodelwasfittedintothepair

den-sitydistributionfunction.Theresultingmodelparameterswere

an-alyzedtoinvestigatetheproteintopologyonthechromatographic

resin

4.5.1 SAXS: proteins in buffers of equal ionic strength

InFig.3,SAXStracesofthemodelproteinsintheinvestigated

mono and dual saltbuffers are shown Moreover, the theoretical

scatteringprofileofPDBcrystalstructures1dpx,1hzhand1gfl are

depicted Notably,the intermediary andhighq-range ofall SAXS

curves(~ 0.4 nm−1 < q)are comparabletothecrystalstructure’s

theoreticalscatteringcurve.However, noiseincreasessubstantially

atq=1.5nm−1 ,resultinginmoresignificantdeviationsfromthe

theoreticalscatteringcurve.Thisisbelievedtobeduetothehigh

electronic contrast.SinceSAXStracesarecomparablebetween0.4

and1.5nm−1 ,real-spacedistances of4.1-15.7nmare accordingly

(astheirreciprocalrelationisgivenbyEq.(17),whichincludesthe

intramoleculardistancesofmAbandGFP(Dmax mAbandGFP:16.4

nm [37] and 7 nm [38]) but exceeds that of lysozyme (Dmax of

lysozyme: 4.0nm [39]) Thisindicates comparableintramolecular

structuresofmAbandGFP>4.1nminallinvestigatedbuffer

sys-tems

In the low q-range (q > 0.2 nm−1 ), the scattering intensities

differsubstantially formAbindifferentHICbuffers(Fig.3B) For

lysozymeandGFP(Fig.3A&B),differencesinthelowq-rangeare

observablebutlesspronounced.Generally,thelowq-rangeis

dom-inatedbylong-rangecorrelations,indicatingtherespectivebuffer’s

modulationofprotein-proteininteractions.Toclassifywhetherthe

interactions are attractive or repulsive, the theoretical scattering

profiles ofthe crystalstructures ofthe corresponding model

pro-teins were calculated andcompared to the experimental data in

the low q-range Lysozyme shows attractive interactions (Fig 3

A), whereasmAbshowsbothattractive,neutralandrepulsive

be-havior, respectively (Fig 3 B) ForGFP, no orminor repulsive

in-teractions can be observed in the respective mono or dual salt

buffers.Trendstowardsattractionandrepulsioncorrelatewiththe

pI of the model protein: the acidic GFP (pI = 5.8 [30]) exhibits

no or weak repulsive interactions, mAb (pI = 7.9-9.1 [29]) both

Fig 3 SAXS profiles of lysozyme (A), the mAb (B), and GFP (C) in solution (5

mg ∗ ml −1 ) Attractive and repulsive categorizations are referred to as the theoreti- cal scattering profile of the corresponding PDB Respective PDBs are visualized in the top right corner for each protein

L.A Jakob, B Beyer, C Janeiro Ferreira et al Journal of Chromatography A 1649 (2021) 462231

pronounced attractiveandrepulsiveinteractions,respectively, and

lysozyme(pI=10.7[28])aredominatedbyattractiveinteractions

inthedualsaltbuffers

The attractivity(andvice versarepulsion)induced by the

sec-ondarysaltfollowsatrend:thepresenceofdivalentanions(SO4 2 −

andHPO4 2 −) induce thehighestattractive/lowestrepulsive forces

followed by the monovalent acetate anion This trend is in line

withtheHofmeisterseries[13].Themono-anddualsaltsystem’s

comparison revealsinconsistencieswiththeHofmeister series: at

pH 6, citrate2 − and citrate3 − are the predominant anion species

inaqueoussolution[40]andratherkosmotropicanions.(citrate3 −

> SO4 2 − > HPO4 2- >citrate2 − >CH3 COO− >citrate− [13,41,42])

However, thesingle saltsodium citrate buffer induces higher

re-pulsive/lower attractive interactions than the citrate and acetate

system

Ultimately, the SAXS analysisofthe proteinsin therespective

buffer indicates that the internal structure of mAb and GFP >

4.1 nm is comparable Moreover, protein-protein interactions

de-pend on the kosmotropicnature ofthe secondary anion andthe

pIofthe protein.mAb systemsgenerallyspan thebroadestrange

of protein-protein interactions, ranging fromthe repulsive to the

attractive regime Lysozyme systems are strictly in the attractive

regime, whereas GFP showsno to slightly repulsive interactions

Attractive interactions correlate withdynamic binding capacities,

as highly attractive systems (such as the systems with the

sec-ondarysaltsulfate)coincidewithhigherdynamicbinding

capaci-ties.Morerepulsivesystems(especiallycitratealone)coincidewith

low dynamic binding capacities For mAb,both the variations in

dynamicbindingcapacity(30% formAb’sdualsaltsystems

com-pared to11–14 % forGFP andlysozyme, asseen inTable 4) and

protein-protein interactions are high (Fig 3), whereas they are

smaller forthe other two proteins The singlesalt system 0.550

Mcitrateshowsaninterestingbehavior.Judgingfromthe

protein-proteininteractiondataalone,wewouldpostulategenerallylower

bindingcapacities thanthedualsaltsystem,asthecitratesystem

isratherrepulsive(Fig.3).However,thedifferenceforcitratealone

tothesystemwiththehighestbindingcapacityis57–61%,butthe

differencebetweenthelowestandhighestbindingcapacityranges

from11–30%forthedualsaltsystems(Table4).Althoughweonly

haveaqualitativemeasureforprotein-proteininteractionsathand,

thisvastdifferencecannot beexplainedintheprotein-protein

in-teractionanalysis(Fig.3).Thisunderlinestheneedfora

quantita-tive comparisonofprotein-proteininteractionsanddynamic

bind-ingcapacities

Altogether, we hypothesize that protein-protein interactions

could explain high dynamicbinding capacitiesand play acrucial

role inprotein adsorption.Inthe followingsection,we willfocus

on the implications of protein-proteininteractions inprotein

ad-sorption in generalandinvestigatewhetherthe binding mode of

theproteinisinfluenced

4.5.2 Isotherms in buffers with equal ionic strength

Equivalent to the breakthrough curves (Fig 2), adsorption

isotherms weredetermined forthemodel proteinsin mono-and

dual salt buffers of equal ionic strength (Fig 4) Generally, the

rankingofthebindingcapacitiesintheadsorptionisotherm

exper-imentsiscomparabletothebreakthroughcurvesforGFPandmAb

For lysozyme,however, thisis not the case exceptfor the mono

saltbuffer.The0.55Mcitratebufferinducesthelowestbindingin

theadsorptionisothermsandbreakthroughexperiments

As discussed above, most model proteins exhibit

protein-proteininteractionsintheinvestigatedsystems,whereGFPshows

the weakestprotein-proteininteractions.Factoringinthe

protein-protein interactions fromour SAXS analysis, Langmuiradsorption

isotherm behavior isnotexpectedforsystemsexhibiting

protein-Fig 4 Adsorption isotherms for lysozyme (A, top), mAb (B, middle), and GFP (C,

bottom) A total volume of 250 μl was incubated for 24 h in 96 well plates at a slurry conc of 10 % and 5 %, respectively Data points where a resin concentration of

5 % where used are denoted with a star 95 % confidence intervals are displayed in the corresponding color Time effects were tested by reducing the incubation time

to 3 h for the mAb in 0.441 M citrate & 0.5 M phosphate As seen in Fig S2, Sup- plementary Material, the difference between 3 and 24 h is small

proteininteractions, whichis trueforthemajority ofthe experi-ments(Fig.4)

When only the adsorption isotherm data is considered, the Langmuir model describesthe GFP adsorption isotherms reason-ablywell (Fig 4A).Consideringalsothe SAXSdata; GFPin solu-tion showed thelowest protein-proteininteraction of all investi-gated model proteins Only GFP in citrate and citrate plus phos-phateshowsweak repulsiveprotein-proteininteraction (Fig.3C) Since the protein-protein interaction analysis here is only quali-tative, it is challenging to state whether the measured

protein-protein interactions are highenough todiminish themodel’s

va-lidityortheycanbeneglectedtoallowforagoodfit

AdsorptionisothermsofthemAbonlyfollowLangmuir

behav-iorwhenacetateisemployedasasecondarysalt(Fig.4B), which

isinlinewiththeprotein-proteininteractiondatafromtheSAXS

analytics(Fig.3B) Whenphosphateandsulfateareemployed as

secondary salts, a non-Langmuirian ascent can be observed that

can befittedwell withtheFreundlich isotherm.When phosphate

isemployedasasecondarysalt,anon-Freundlichplateauis

even-tually reached, making both models unsuitable for the

descrip-tion of the isotherm For the secondary salt sulfate, however, a

plateau could not be reached Here, we could not collect data

athighermobilephaseconcentrationsduemethodological

limita-tions.Lastly,the0.55Mcitrate bufferinducestheFreundlichtype

binding for mAb This non-Langmuirian behavior is also in line

withour protein-proteininteractiondata sincethe mAbisinthe

repulsiveregimewhen0.55Mcitrateisusedasabuffer

The adsorption isotherm experiments with lysozyme reveal

Freundlich and BET behavior, respectively (Fig 4 A) For the

lysozyme experiments, non-Langmuirian behavior is also in line

with the SAXS data since a strictly attractive regime isobserved

for lysozyme in all investigated systems (Fig 3 A) Adsorption

isothermsthatfollowtheBETmodelindicatemultilayerformation,

but it isunclear whether themultilayer forming interactions are

reversibleorirreversible

Conclusively,wehypothesizethateitherthesurfacecoverageis

increasedor multilayerformationdoesoccur insystemsthat

fol-low the Freundlich and BET isotherm model, respectively, being

consistent with our protein-protein interaction data However, it

cannotbestatedwhetherreversibleself-associationorirreversible

aggregation occurs Furthermore, GFP in citrate only and citrate

plus phosphate could show pseudo-Langmuirian behavior or too

littlerepulsiveinteractiontoimpacttheproteinadsorption

4.5.3 SAXS: protein-resin adduct fitted via SARW model

For the analysis of the protein-resin adduct, the

chromato-graphicresinwasincubatedfor15hwitheithermAb,GFPoronly

buffer,respectively.TheresinsuspensionsweremeasuredviaSAXS

andaself-avoidingrandomwalkmodelwasfittedintothe

result-ing pair densitydistribution function after inverse Fourier

trans-formofthescatteringdata(Fig.S3,Supplementary Material).The

resulting model parameters are presented in Fig.5 A, as well as

Fig.S4(SupplementaryMaterial)

Fig.5 Ashowsthat theexcluded volumedecreases when

pro-tein(GFPandmAb)isloadedontotheresin.Whencomparingthe

bound modelprotein’simpact,the resultingexcluded volume

pa-rameter islower forresin incubatedwithmAbcompared toGFP

Besidestheimpactoftheloadedprotein,theexcludedvolume

pa-rameterdependsonthebufferingsystem.Foreithermodelprotein,

theexcludedvolumeparameterissignificantlyhigherinthemono

saltsystem(0.55sodiumcitrate)thanall other dualsaltsystems

Furthermore,theexcludedvolumeparameterislowestforsystems

incubatedwiththedualsaltbuffercitrateplussulfate.Thisbuffer

results in a significantly lower excluded volume parameter

com-pared to all others in mAb systems Moreover, it induces a

sig-nificantlylowerexcludedvolumeparameterforGFPsystems

com-paredtocitratealoneandcitrateplusacetate

Altogether,theexcludedvolume parametercorrelates inversely

withtheequilibriumbindingcapacitydetermined viathe

adsorp-tion isotherms This of course raises the question how protein

adsorption could impact the excluded volume parameter of the

adductasawhole.Generally,theexcludedvolumeparametercan

be correlated with the accessible surface area, as the accessible

surfaceareaencompassestheexcludedvolume[43].Therefore,we

believe that thereduction oftheexcluded volume parametercan

be best understood with the reduction of the accessible surface

Fig 5 A: Self-avoiding random walk (SARW) excluded volume parameter ( ν) de- duced from SAXS measurements of resin slurry (5 %) incubated with protein at 5

mg ∗ ml −1 for 15 h The average of three independent experiments is shown, includ- ing standard deviation B: Conceptual visualization of the impact of protein bind- ing on a SARW polymer As proteins deposit in the cavities of the chromatographic resin, the excluded volume parameter ( ν) of the protein-resin adduct decreases

area.Whenafractalobjectisconsidered,thisismostlikelycaused

by thedeposition of the proteinin thecavities of the chromato-graphic resin Deposition of proteins in the cavities of the chro-matographic resin woulddecrease overall accessible surfacearea (Fig.5B)

On the other hand, preferential binding of the protein to flat

orconvexregionsofthechromatographicresinwouldincreasethe accessiblesurfaceareaand,therefore,theexcludedvolume param-eterofthewholeobject,whichcouldnotbeobserved.This curva-turedependencywaspreviously highlightedinatheoretical work [44].There, concave hemicylindricalcarbon nanotubeswere sim-ulatedinwater,andthey weremorehydrophobicthan their con-vex counterpart When we now alsoconsider the SAXS analytics

of the proteins in solution, buffer-dependent protein-protein in-teractions could play a role in the topology of the protein-resin adduct.Protein-proteininteractions could leadto increased depo-sitionontoalreadyoccupiedcavities anddecreasedsurface cover-ageduetorepulsion,respectively Altogether,we believethat the excludedvolumeparameterdecreasesduetothedepositionofthe proteininthecavitiesofthechromatographicresin.Nevertheless, thishypothesisisonlybasedontheoreticalconsiderationsand de-mandsfurthervalidation

Similarly, the path length of the resulting self-avoiding ran-domwalkincreaseswhenmAbandGFPareloadedontotheresin, whereas the increase is more pronounced for mAb than GFP In contrast to the excluded volume parameters, only two buffering systemsshowsignificantlydifferentpathlengths,namelymAb in-cubatedwithcitratealoneexhibitedshorterpathlengthsthan cit-rateplussulfate(Fig.S4,SupplementaryMaterial)

5 Conclusion

TheionicstrengthofdualsaltHICbuffersisamoredecisive pa-rameterfordynamicbindingcapacitiesthantheir surfacetension However,dynamicbindingcapacitiesstilldifferupto30% depend-ingon thesecondarysaltemployed, andthemodelproteinused, evenwithcomparableionic strengthofthe bufferingsystems.To gainbettermechanisticinsightintodualsaltsystemsinHIC,SAXS