ContentslistsavailableatScienceDirect
Journal
of
Chromatography
A
jou rn al h om ep a g e : w w w . e l s e v i e r . c o m / l o c a t e / c h r o m a
Application
of
microstructured
membranes
for
increasing
retention,
selectivity
and
resolution
in
asymmetrical
flow
field-flow
fractionation
Maria
Marioli
a,∗,
Ü.
Bade
Kavurt
b,
Dimitrios
Stamatialis
b,
Wim
Th.
Kok
aaAnalyticalChemistryGroup,van’tHoffInstituteforMolecularSciences,UniversityofAmsterdam,P.O.Box94157,1090GDAmsterdam,theNetherlands b(Bio)artificialOrgans,DepartmentofBiomaterialsScienceandTechnology,TechMedInstitute,UniversityofTwente,P.O.Box217,7500AEEnschede,the Netherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received10April2019
Receivedinrevisedform25June2019 Accepted2July2019
Availableonline3July2019 Keywords:
Field-flowfractionation Flowovergrooves AF4
Computationalfluiddynamics Microstructuredmembranes Proteinseparation
a
b
s
t
r
a
c
t
Inthepresentproof-of-conceptstudy,wedemonstratethatretentiontime,selectivityandresolution canbeincreasedinasymmetricalflowfield-flowfractionation(AF4)byintroducingmicrostructured ultrafiltrationmembranes.Evenlyspacedmicron-sizedgrooves,thatareplacedperpendiculartothe channelflowontheaccumulationwallofafield-flowfractionationsystem,causeadecreaseinthe zonevelocitywhichisstrongerforlargersolutes.Thishasbeendemonstratedinthermalfield-flow fractionation,andweprovethatthisisalsothecaseinAF4.Weexaminethehypothesistheoretically andexperimentally,bybothcomputationalandphysicalexperiments.Bymeansofmomentanalysis,we derivetheoreticallyasetofequationswhich,undercertainconditions,describethemasstransportand relateretentiontime,selectivityandplateheighttothedimensionsofthegrooves.Physicalexperiments arecarriedoutusingmicrostructuredpolyethersulfonemembranesfabricatedbyhotembossing,andthe experimentalresultsarecomparedwithcomputationalfluiddynamicsexperiments.
©2019TheAuthor(s).PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense
(http://creativecommons.org/licenses/by/4.0/).
1. Introduction
Asymmetrical flow field-flow fractionation (AF4), the most appliedsubtechniqueofthefield-flowfractionation(FFF)family, isanestablishedanalyticalmethodtoseparatemacromolecules andnanoparticlesaccordingtotheirhydrodynamicsizeundermild conditions[1–3].Thecouplingwithvariousphysicalandchemical detectorshascontributedsignificantlytoitspopularityasitcan providevaluableinformationsuchasmolecularweight distribu-tion,sizedistribution,conformationandchemicalcompositionin asinglerun[4].Considering therapidgrowthinbiotechnology, nanotechnologyandpolymerengineering,itisevidentthatAF4is goingtowitnessafurthergrowthinapplicationsinthecoming years.Inthisregard,itisworthwhiletoproposeandinvestigate possiblenewtechnicaldevelopmentsthat mayimprove perfor-mance.
Inthisstudyweinvestigatethepossibilityofincreasing reten-tion time, selectivity and resolution by using microstructured ultrafiltration(UF)membraneswithparallelgroovesontheir
sur-∗ Correspondingauthor.
E-mailaddress:M.Marioli@uva.nl(M.Marioli).
face (Fig. 1). However, considering that AF4 is a very flexible techniquewhere severalparameterscanbealtered tooptimize separation,firstajustificationshouldbegivenfortheusefulness ofsuchadevelopment.
AccordingtotherigorousFFFtheory,theretentiontimeof well-retained(withretentionratio<0.1)componentsinAF4isequal to[5], tR= w 2 6Dln
1+ V˙c ˙Vout B (1)where wisthechannelthickness, ˙Vcthecross-flowrate, ˙Voutthe
channeloutletflow rateandBthefractionoftheaccumulation areaafterthefocusingpoint.Therefore,theselectivityofapairof well-retainedsolutesequalstheratiooftheirdiffusioncoefficients,
˛=tR,2 tR,1 =
D1
D2
(2)
andconsequently,itcannotbealteredbychangingthe experimen-talparameters.Resolutioncanbeimprovedbyreducingtheplate https://doi.org/10.1016/j.chroma.2019.07.001
Fig.1.AF4withmicrostructuredmembranes.
heightwhich,basedonthenonequilibriumtheory(for<0.1),is equalto[6],
H= 24D2
v
0u3crw
(3) whereucr isthecross-flowvelocitythoughtthemembrane and
v
0 is thecross-sectional mean carrier velocity. Hence, a highcross-flowvelocitydecreasesplateheight.However,itmaylead toadsorptiononthemembrane and massoverloading for sen-sitivemacromolecules.Inaddition,highflowratesarehindered bythetransmembranepressurewhenultrafiltration(UF) mem-braneswithverylowmolecularweightcut-off(MWCO)areused toseparatesmallmacromolecules.
Thesolutescanberesolvedatlowercross-flowratesby increas-ingtheretentiontime,sinceaminimumtimeisrequiredtoachieve separation [7], which couldbe accomplished by increasingthe cross-flowtooutlet flowratioor thespacerthickness[8].Very highcross-flowtooutletflowratiosareimpractical,particularly forUFmembraneswithlowMWCO,andmaydistorttheparabolic flowprofile[9].Inaddition,theuseofathickerspacerresultsin higherrequiredfocusingtimesand moredilutionwitha subse-quentdecreaseinsensitivity[7].Moreover,alowaspectratiob/w (<30),wherebis thechannel breadth,mayaggravateedge and endeffectsincreasingplateheightandreducingrecovery[10,11]. Therefore,itcouldbebeneficialtoinvestigateamethodthatcould increase retention and resolution without altering the optimal cross-flow,spacerthicknessandcross-flowtooutletflowratio.
Theconceptofanaccumulationwallwithmicron-sizedgrooves inFFFhasbeenintroducedin1978byGiddingsetal.[12]asan attempttoincreaseretentionforsmallanalytesinthermal field-flowfractionation(ThFFF).Inaddition,groovedsurfaceshavebeen incorporatedinmicrofluidicchannelsforvariousotherapplications suchastoenablemixing[13]andtoseparatecellsand micropar-ticles[14].Navierhasdescribed thatmacroscopicallytherough surfaceisequivalenttoasmoothsurfacewithpartialslip[15–17]. Infact,forthisreason,asmallslipmightexistontheflatmembrane oftheAF4channel,asa resulttotheporosity,but itis negligi-bleforUFmembranes[18].NanostructuredUFmembraneshave beenfabricatedbynano-imprintinglithography[19–21], where membraneswerehot-embossed,andmicrostructuredpolymeric materialshave been developed withphase separation [22–24], whereapolymersolutioniscastoverapatternedmold.
Thescopeofthisstudyistoconductaproof-ofconcept inves-tigationtoassesstheeffectofmicrostructuredmembranesonthe retentiontime,selectivityandresolutioninAF4.Ahotembossing methodwaschosenforthefabricationofthesemembranes.We sharefundamentaltheoryandexperimentalfindingsthat comple-mentandexpandthepreviousstudywithperpendiculargrooves inThFFF[12].
Fig.2. Left-handfigure:displayofthetheoreticalmodel.Right-handfigure:velocity profile(a)overaflatmembraneand(b)overagroovedmembranewherethevelocity zero-planeistakenontheedgeoftheridge(x=h).
2. Theory
2.1. Transportequationsandmomentanalysis
Here,wedescribeasimplifiedmodelthatenablesustoderivean analyticalsolutiontotheproblemofmassmigrationovergrooves inanAF4channel.Inthismodelthegroovesareformedby zero-widthridgeswithauniformheighthonthemembranesurface, perpendiculartotheflowdirection.Slipflowthroughthegrooves isneglected;thezero-velocityplanefortheaxialflow(
v
)istaken atthetopoftheridges(Fig.2).Thefollowingsimplificationshavebeenmade:
(1)Moleculardiffusionintheaxial(z-)directionisneglected. (2)Thedevelopmentoftheconcentrationprofileinthe
perpen-dicular(x-)directioniscompletebeforeelutionisstarted,bya precedingfocusingstepintheprocedure.
(3)Onlywell-retainedcompoundsareconsidered(withretention ratio <0.1). Such compounds are present predominantly closetotheaccumulationwall,wherethelinearpartoftheflow profileprevailsandthecross-flowvelocityucrmaybe
consid-eredasbeingequaltothefluidvelocitythroughthemembrane. For well-retainedcompounds,themathematicscanbe sim-plifiedsinceintegralsover theheightofthechannelcanbe takenfromx=0toinfinityinsteadoftotheupperwallposition (x=w),withgoodaccuracy.
(4)Therearenointeractionsbetweentheproteinandthe mem-brane.
(5)Flowconditionsarelaminar.Thisassumptionshouldholdtrue sincethepresenceofperpendiculargrooves,whicharesmall compared tothechannelthickness,reduceslocallytheflow velocity and decreasestheReynolds number [16]. Although eddiesmayexistinthecornersofthegrooves,theflowvelocity isverylowthereandthefluidisalmoststagnant.
Thetransportofacompoundi,withalocalconcentrationci=
ci(x,z,t),isgivenbythesimplifiedgeneraltransportequation
∂
ci∂
t =Di∂
2ci∂
x2 +ucr∂
ci∂
x −v
(x)∂
ci∂
z (4)whereDiisthediffusioncoefficientofthecompoundofinterest,
and
v
(x)thelocalaxialflowvelocity.Themoleculardiffusionterm alongthez-directionisneglectedintheRHSofEq.(4).Theplus signfor thesecondtermoftheRHSappearsbecausea positive valueistakenforucr,evenwhenthecrossflowisinthenegativex-direction.Theassumptionthattheanalytehasbeenintroduced inthechannelasafiniteplugleadstotheboundaryconditions
andtheassumptionthatthewallsofthechannelareimpermeable forthecompoundto
Di
∂
ci∂
x +ucrci=0 for x=0,w (4b) Twosetsofmomentsaredefined.Localmoments,thatdescribe themassdistributionofacompoundiinafluidlayeratacertain distancexfromthemembrane,aredefinedasmn,i(x,t)=
+∞ −∞znc
i(x,z,t) dz (5)
andoverallmoments,thatdescribethemassdistributioninthe axialdirectionintegratedovertheheightofthechannel,as
Mn,i(t) =
w 0 mn,i(x,t) dx≈ ∞ 0 mn,i(x,t) dx (6)MomentsexistwhentheintegralsconvergeinEq.(6),i.e.,whenit canbeassumedthattheconcentrationofacompoundiapproaches zerofastenoughwhenzgoestoplusorminusinfinity.Thiswillbe thecasewhenthecompoundwasintroducedinthechannelasa plugorpeakoffinitewidth.
WhenbothsidesofthegeneraltransportEq.(4)aremultiplied withznandintegratedoverzfromminustoplusinfinity,
expres-sionsareobtainedforthelocalmomentsofi
∂
mn,i∂
t =Di∂
2mn,i∂
x2 +ucr∂
mn,i∂
x +nv
(x)mn−1,i (7) ThethirdtermoftheRHSinthisequationisobtainedbypartial inte-grationwiththeassumptionthatznci(x,z,t)vanishesforz→±∞.
Whenalocalmoment(n−1)isknown,thisequationcanbeusedto evaluatethenextlocalmoment(n).IntegrationofEq.(7)overthe heightofthechannel,consideringboundarycondition(4b),gives anexpressionfortheoverallmomentMn,i
∂
Mn,i∂
t =n w0
v
(x)mn−1,idx (8)2.2. Thezerothmoment(massdistribution)
IntegratingbothsidesofEq.(4)overzfromminustoplusinfinity givesanexpressionforthelocalzerothmomentofi,
∂
m0,i∂
t =Di∂
2m0,i∂
x2 +ucr∂
m0,i∂
x (9)Undertheassumptionthatfocusingwascomplete,andasteady statewasreachedbeforetheexperimentwasstarted,bothsidesof Eq.(9)mustbezero,andthemassdistributionovertheheightof thechannelcanbefoundas
m0,i=m∗0,i·exp
−ucrDix
(10) wherem∗0,iisthezerothmomentonthemembranesurface(with x=0).Theexp.concentrationprofileextendsoutfromtheupper wall(x=w)here. Eq. (10)describesthewell-known exponen-tialconcentrationprofileontheaccumulationwallinFFF,witha characteristiclayerthickness
equaltoDi/ucr.Whenthecon-centrationoftheanalyteisscaledsoas m*
0,i=
ucr
Di
(11) theoverallzerothmomentM0,ibecomes1,andthehigheroverall
momentsareautomaticallynormalized.
2.3. Thefirstmoment(meanretentiontime)
Themodelforthegroovedsurfaceusedhere(Fig.2),witha stagnantlayerof fluid determinedby thegrooveheight h,and approximatelylinearlyincreasingchannelflowratefromtheslip planeatthetopoftheridges,givesforthelocalaxialflowvelocity
v
(x)=0 for 0≤x≤h (12a)v
(x)=6(x−h)w
v
for x≥h (12b)WhenEqs.(10)–(12)aresubstitutedintoEq.(8),
∂
M1,i∂
t = w h 6(x−h) wv
ucr Di exp −ucr Dix dx (13)theaxialvelocity
v
iofthecompoundisobtained(withforsimplicity integrationtoinfinityinsteadoftox=w),v
i=∂
M1,i∂
t = 6Di ucrwv
exp −ucrh Di (14) Eq.(14)withh=0givesthewell-knownexpressionforthezone velocity over a flatmembrane. With a groovedmembrane, the velocitydecreasesexponentiallywiththeratiooftheridgeheight overthecharacteristiclayerthickness.Fortheretentiontimethe oppositecanbewrittentR,i=tR,iFLexp
+ucrh Di (15) wheretFLR,iistheretentiontimewithaflatmembrane,under
oth-erwise thesameconditions. Theretention time increasesmore stronglybythepresenceofthegroovesforcompoundswithasmall layerthickness,i.e.,formorestronglyretainedcompounds.
In the separation of two components, the selectivity ˛ is increasedwithincreasingridgeheightanditcanbewrittenas ˛=tR,2 tR,1 = D1 D2 exp
ucrh 1 D2 − 1 D1 =˛FLexp h 1 ˛FL−1 (16) where˛FListheselectivitywithaflatmembrane,and1the
char-acteristiclayerthicknessofthefirst,leastretainedcompound.In Fig.3a,thecalculatedeffectofthe(relative)heightofthegrooveson theretentiontimesandtheselectivityisshownfortwocompounds withdiffusioncoefficientsthatdifferbyafactorof√2.
2.4. Thesecondmoment(peakvariance)
Toevaluatetheinfluenceofthegroovedsurfaceonpeak broad-ening,first anexpressionfor thedevelopmentofthelocalfirst momentshastobederived.Inthis,wefollowtheapproachtaken byTaylorandArisintheirtreatmentofpeakbroadeningin cylin-dricalchannels,andinearlyworkofGiddingsondispersioninFFF [25].TheyfoundsolutionsforthegeneraltransportEq.(4)inthe formofasumoftransientfunctionsandastationaryfunction.The transientfunctionsdescribetheconcentrationchangesintimeand spacedirectlyafterthestartofthe’elution’andtheydependonthe initialconditions.Itwasshownthatthesetransientfunctionsdie outrapidly,andthatastationarysituationdevelopsinwhichthe localcentersofgravityatdifferentdistancesfromthewallare situ-atedinasteadyprofilearoundtheoverall(mean)centerofgravity ofthetransportedplugofthecompoundofinterest.Here,a solu-tionissoughtforEq.(4)describingonlythestationarysituation, i.e.,asolutionthatobliges
1 m0,i(x)
∂
m1,i(x)∂
t =∂
M1,i∂
t =v
i for all x (17)Fig.3.Theoreticalestimationofvariablesasafunctionoftherelativeridgeheight:a) increaseinselectivityandretentiontimefortwosoluteswithdiffusioncoefficients thatdifferbyafactorof√2(e.g.,monomeranddimer);retentiontimeshereare normalizedwiththeretentiontimeofthesmallersoluteforaflatmembranetFL
R,1b) increaseinplateheightc)increaseinresolution.
Asetofparticulatesolutionsform1,i(x)canbefoundthatsatisfyall
boundaryconditions. For0≤x<h, mA1,i(x,t) = 6
v
w t− Di u2 cr − x ucr exp −ucrh Di exp −ucrx Di (18a) andforx≥h, mB1,i(x,t) = 6v
w { (x−h)2 2Di + x−h u + t− Di u2 cr − x ucr exp −ucrh Di }exp−ucrx Di (18b)Theincreaseintimeoftheoverallsecondmomentcannowbe foundbysubstitutingEqs.(12b)and(18b)intoEq.(8)
∂
M2,i∂
t =2 w h 6v
0 w (x−h)m B 1,idx (19)Sincethefluidvelocityiszerofor0<x<h,Eq.(18a)doesnothave tobeincludedintheintegration.Centralizingoftheoverallsecond momentgivestheincreaseofthespatialvarianceintime
∂
2 z∂
t =∂
M2,i∂
t −2M1,i(t)∂
M1,i∂
t (20)andfinally,theplateheightHcanbeobtainedas H=
∂
z2/∂
t∂
M1,i/∂
t(21) ThefinalresultforHis
H=24D 2 i
v
0 u3 crw5 2exp +ucr Dih −32− ucr 2Dih exp −ucr Dih (22) Foraflatmembrane,withh=0,thesecondandthirdfactorsinthe RHSofEq.(22)areequalto1,andthewell-knownexpressionfor H(HFL)isobtained(Eq.(3)).
InFig.3btheincreaseoftheplateheightwiththerelativeridge heightisshown,andinFig.3ctheincreaseinresolutionoftwo soluteswithratioofdiffusioncoefficients√2isshown.Weobserve thatforgrooveheighth=1.51,thereisatwo-foldincreasein
res-olutionandafour-foldincreaseintheretentiontimeoftheless retainedcomponent.Forcomparison,thesameincreasein reso-lutioncouldbeachieved (withoutaltering thecross flow)by a two-foldincreaseofthespacerthicknessorapproximatelyten-fold increaseofthecross-flowtooutletflowratio.
3. Materialsandmethods 3.1. Samplesandcarriereluent
Bovineserumalbumin(BSA),␥-globulin,apoferritin, thyroglob-ulinandhemoglobinwerepurchasedbySigma–Aldrich(MO,USA). PBS0.15M(20mMduetosodiumphosphatesalts)withapHof 7.2wasusedasacarriereluentfortheAF4experimentsandas adiluentfortheproteins.Allproteinsampleswerepreparedata concentrationof1mg/mL.
3.2. Fabricationandcharacterizationofthemicrostructured(MS) membranes
Twosiliconmolddesignswithparallelgrooveswereusedfor preparationofthemicrostructuredmembranes.MoldI(LioniXBV, TheNetherlands)hadapatternedareaofdiameter15.1cmwith groovesof cavity widthc=50m,ridge width r=50m, ridge heighth=12mwhereasMoldII(MESA+cleanroom,Universityof Twente,TheNetherlands)hadapatternedareaofdiameter6.8cm withgroovesofc=30m,r=20mandh=25m. Polyethersul-fone(PES)membraneswith10kDaand30kDamolecularweight cut-off(MWCO)(Sartorius,Germany)wereusedforthemembrane patterningwithoutanypretreatment.
Microstructured (MS) membranes were prepared via hot embossing which was performed with an imprinter (Obducat, Sweden)inMESA+cleanroom(UniversityofTwente).The emboss-ingtemperature,pressureandtimewere120◦C,40barand180s, respectively and demoldingoccurredat 40◦C [20]. Surface and cross-section images of the microstructured membranes were takenby scanningelectron microscopy(SEM) equipment,XL30 ESEM-FEG(Philips,TheNetherlands)orJEOLJSM-6010LA(JEOL, Japan).MSmembraneI(Fig.4a)wasfabricatedbyhot-embossing a PES 10kDa membrane with the Mold I, and MS membrane II(Fig.4b)byhot embossinga PES30kDa membrane withthe MoldII.Membrane sampleswerewashed,dried,broken in liq-uidnitrogenforcrosssectionimagesandgold-sputteredforSEM imaging.
Fig.4.MicrostructuredmembranesandAF4channels:a)MSmembraneI(hotembossedwithMoldI)andChannelIb)MSmembraneII(hotembossedwithMoldII)and ChannelII.
Cleanwaterflux(Jw)valuesofthemembranesweremeasured
withdead-endAmiconStirredCell(Model8050,MerckMillipore, MA,USA)andultrapurewater(MilliQsystem,MerckMillipore). Measurementswereperformedatfourdifferenttransmembrane pressures(P)in therange of0.5–2bar,afterremoving ofthe membrane preservativesby immersingin water and after pre-compactionat2bar.Theweightofpermeatedwaterversustime wasmeasuredandthecleanwaterflux(JwinL/m2/h)was
calcu-latedforeachpressureconsideringtheeffectivemembranesurface area,whichwas13.4cm2 (Theareaisassumedasconstantafter
preparationofamicrostructuredsurface).Thecleanwater perme-ance(CWP,inL/m2/h/bar)ofthemembranewasdeterminedfrom
theslopeofJwversusPrelationship.
3.3. AF4experiments
TheAF4systemwasanEclipseDualTecsystem(Wyatt Technol-ogyEurope,Germany)connectedtoanAgilentHPLC1200system (AgilentTechnologies,Germany)thatconsistedofadegasser,an isocraticpump,aUVdetectorandanautosamplerequippedwith athermostat.Thetemperatureoftheautosamplerwassetat5◦C. TwoAF4trapezoidalchannelswereused,designatedasChannel IandII,oneforeachmembrane/moldsize(Fig.4).TheMS mem-braneswerecutwiththegroovesperpendicularandintheshape oftheporousfritwithsurgicalscissors.
ChannelIwasacommercialAF4channel(WyattTechnology Europe) which was used with the largerpatterned membrane (d=15.1cm).Ithadtip-to-tiplength13.3cmandaccumulationarea 15.6cm2(Fig.4a).Thenominalspacerthicknesswas250or350m.
Thefocus-flowwas1.5mL/minfor3minandthefocusingpointwas setat18%ofthechannellength.Theinjectedvolumewas10L (10ginjectedmass)andtheUVdetectionwasat280nm.
ChannelIIwasaminiaturizedchannelcreatedtotestthesmaller patternedmembrane(d=6.8cm).It hadtip-to-tiplength6.3cm andaccumulationarea7.24cm2 (Fig.4b).Itwascreatedusinga
commercialchannelmodifyingitsupperinlayandspacer.Inthe upperinlaytwointernalthreadsweremilledtoconnectthetubing fittingsfortheinletandoutlet.Thespacerwasfabricatedcutting
MylarA4sheetsofnominalthickness250and350m.The focus-flowwas0.8mL/minappliedfor3minandthefocusingpointwas setat18%.Theinjectedvolumewas5L(5ginjectedmass)and UVdetectionwasat220nm.
3.4. Computationalfluiddynamics(CFD)
Afiniteelementsolver,COMSOLMultiphysics5.2(COMSOLInc., MA,USA),wasusedtomodeltheAF4channelandsimulatethe proteinmigrationovertheflatandthepatternedmembrane.To reducethemodelintotwo dimensionsforlowercomputational cost,asymmetricalchannelwasmodelledinsteadofan asymmet-rical.Forthispurpose,asimplerectangulardomainwascreated, withaflatorgroovedbottomboundary.Ameshoffreetriangular elementswascreatedwithveryfineelements(<1m)inthe prox-imitytothebottomboundarytosimulateproteinmigrationwith highaccuracy.
Todescribetheflow,laminarflowofanincompressiblefluid wasusedand theboundaryconditions(inlets,outlets)wereset todefinechannel flowandcross-flowvelocities(itwasverified laterfromtheresultsthattheassumptionofthelaminarflowwas validbythecellReynoldsnumber).Thecross-flowvelocitywas dis-tributedhomogeneouslyalongthebottomboundary(membrane). Theoption“transportofdilutedspecies”(includingconvectionand diffusion)wasusedtosimulateproteinmonomeranddimer.The studyoftheflowprofilewassolvedasasteadystateproblemand theoutput(velocityfield)wasusedtosolvethetimedependent problemoftheproteinmigrationwithaBDF(Backwards Differen-tialFormula)solver.Therelativeandtheabsolutetoleranceswere setat10−4.Theinitialandthemaximumtimestepswereset0.001s and0.5s,respectively.
4. Resultsanddiscussion
4.1. Characterizationofthemicrostructuredmembranes
Themicrostructuredmembranes,designatedasMSmembrane Iand IIhadsimilarridgeheight,h∼12m,anddifferent
peri-Table1
ProteinrecoveryinAF4beforeandafterhotembossingoftheUFmembranes.AF4conditions: ˙Vc= ˙Vout=1mL/min. Recovery(%)±s.d.
BSA(66.5kDa) ␥-Globulin(150kDa) Apoferritin(443kDa) Thyroglobulin(669kDa)
Flatmembrane(10kDa) 89±2 86±2 87±3 78±4
MSmembraneI 22±4 35±5 86±1 76±3
Flatmembrane(30kDa) 20±3 60±5 82±4 71±2
MSmembraneII 9±1 11±3 84±2 72±3
odicity(i.e.,thesumofcavityandridgewidth),100and 50m respectively (Fig. 4). The shape of the patterns was rectangu-lar with round corners (MS membrane I) or ellipsoidal (MS membrane II)as therectangularcavities of themoldwere not completely filled during embossing. Although the polymer is heatedaboveits glasstransitiontemperaturein hot embossing processes,embossing wasperformed belowtheglasstransition temperature, since collapse of the pores and loss of perme-ance are reported in the literature for a PES membrane [20]. TheCWPs ofthe non-patterned membraneswereestimated as 150±20L/m2/h/bar (for membranes with 10kDa MWCO) and
271±113L/m2/h/bar (formembraneswith30kDaMWCO). The
CWPsofbothmembranesdecreasedafterhotembossing;MS mem-braneIhadCWPof74±1L/m2/h/barandMSmembraneIIhadCWP
of130±18L/m2/h/bar.
ProteinrejectionoftheMSmembraneswasevaluatedwiththe AF4system;therecoveryoffourproteinsofdifferentmolecular weight(66.5–669kDa)wasestimatedfromtheratioofthepeak areaof thefractionatedsampletothepeak areaofthe unfrac-tionatedsample.Thepeakareaoftheunfractionatedsamplewas estimatedfromthefractogramobtainedbyinjectingandeluting thesameamountoftheproteinwiththesamechanneloutletflow rate,withouttheapplicationof focusflow orcross-flowexcept forapoferritin.Thesolutionofapoferritincontainedlow molec-ularweightcomponentswhichwereUV-activeatthedetection wavelength,andthereforefocusflowwasappliedfortheirremoval. Forthisreason,therecoveryvaluesofapoferritinmaybeslightly overestimatedforallmeasurements(bothwithflatandwithMS membranes).TheexperimentalresultsaregiveninTable1;the recoveryofthesmallerproteins(BSAand␥-globulin)was signifi-cantlylowerfortheMSmembranes.
Thedecreaseinrecoveryafterhotembossingshouldindicate anincreaseintheactualMWCOratherthanproteinadsorption sincethePESmembranesusedinthisstudyarehydrophilicwith lowfoulingpropertiesforproteinsolutions.Thiswasconfirmedby injectingandfocusingforseveralminutesahighvolume(100L) ofaconcentratedsolution(30mg/mL)ofhemoglobin(∼65kDa) whichhasaredcolor.Itwasobservedthatthesamplewasfocused asanarrowbandwithaflatmembranewhileitwaspassingthrough thecross-flowwithanMSmembrane.Whenthemembranewas removedandvisuallyinspected,itwasnotstainedwhichwould indicateadsorption.
Theaforementionedresults(increaseinMWCOanddecrease inCWP)seemcontradictingsincelowerCWPisoftencorrelated withadecreaseinthesizeornumberoftheporesoftheselective (patterned)side.ApossibleexplanationisthattheCWPdecreases becauseof the membrane compaction (particularly in thearea ofthegrooves’valleyswhichexperiencethehigheststress dur-inghotembossing).Inaddition,theincreaseintheactualMWCO mightberelatedtoanincreaseoftheporesizeof thegrooves’ ridgesbecause of themembrane deformation or toother local defectsthatoccurduringimprinting/demoldingwhichare, how-ever, small enough to affect only the recovery of the smaller proteins.
Incontrastwithourobservations,Marufetal.[20]showedthat hotembossingcouldleadtosimilarCWPand lowerMWCOfor
anotherPESmembraneandamoldpatternedwithsmallergrooves (inthesub-micronrange).Perhapstheporedeformationtherewas minimalbecauseofthesmallersizeofthegrooves.However,the effectofthemembranecompactionontheCWPandthedifference inthestressdistributiononthevalleysandontheridgesduring hotembossinghavebeendiscussedinthesestudies[21].Overall ourresultsindicatethathot-embossingneedstobeoptimizedto avoidchangesoftheMWCOsincetheconceptwouldbe benefi-cialparticularlyforlowmolecularweightanalytes,andingeneral UF membraneswithhighsolventpermeability are preferredin AF4.
Using BSA as thecalibrant withknown diffusioncoefficient (6.21·10−11m2/s[26]),theactualchannelthicknessforthe
Chan-nel I and the Channel II witha flat membrane was estimated 305±6mand294±8mrespectively,andthediffusion coef-ficientofapoferritinwasestimated3.38·10−11m2/sfromEq.(1).
Thesevalueswereusedinthesimulations.However,MS mem-branesarealreadycompressedduetohotembossing,andhence anyadditionalcompressioncausedbythespacerisexpectedto besmall.Thiswouldresultinlargeractualchannelthicknessand consequentlyin longerretention times. Thedifference in com-pressionbetweentheflatandtheMSmembraneswasevidentby visualinspectionwhenthemembraneswereremovedfromthe channelandinspected.Unfortunately,themethodwithaprotein ofknowndiffusivitycannotbeappliedfortheMSmembranesas theretentiontime increasesbythepresenceofthegroovesfor well-retainedcompounds.However,inorder toassesscorrectly theeffectofgrooves,theactualchannelthicknessoftheMS mem-branes needs tobe measuredand we attempted this by other means.
First,themembranecompressibilitywasestimatedfromthe dif-ferenceinthethicknessofthecompressedandnon-compressed partofthemembranes,measuredbySEMandamicrometerscrew gaugewhentheywereremovedfromthechannel.The compres-sionthatoccurredwithaflatandaMSmembranewas∼50m and∼20mrespectively.Thiscorrespondsto11%largerchannel thicknesswiththeMSmembranes.However,thesemethodshave lowprecisionsincetheymeasureonlyaverysmallpartofthetotal membraneareawhenthemembranesaredry.Second,weapplied therapidbreakthroughmethod[27]inthefractogramsobtained fortherecoveryexperimentswithinjectionand elutionof thy-roglobulin(withouttheapplicationoffocusorcross-flow).Thevoid volumewasmeasured13%largerwiththeMSmembranewhich correspondstoa13%thickerchannel.Thisresultisinclose agree-mentwiththefirstmethod.However,bothmethodsdonotuse cross-flowwhichmightslightlyaffectthemembranecompression and/orswelling.
4.2. AF4experiments
Apoferritinandthyroglobulinwerechosenasthemodel pro-teinstoassesstheeffectofthegroovesonretentiontime,selectivity andplateheight,sincetheyexhibitedhighrecoverieswiththe pat-ternedmembranes(Table1).InFig.5thefractogramsofapoferritin, obtainedusingflatandMSmembranes,areoverlaidafter subtrac-tionofthetimethatwasrequiredforthefocusingstep.Forboth
Fig.5. ComparisonofflatandMSmembranesanalyzedwithflowrates ˙Vc= ˙Vout=1.0mL/minfora)ChannelIandMSmembraneIandb)ChannelIIandMSmembraneII.
Fig.6.CFDmodelfortheChannelII/MSmembraneIIsystem:a)Meshofthemodelinthebeginningofthechannel,b)velocityprofileoverthegrooves,c)concentration profileofapoferritinoverthegroovesandd)derivedconcentrationattheoutlet(rightboundary)foreverytimepointforthemonomeranddimer.
Table2
ComparisonofflatandMSmembranesforbothchannelswithrespecttotheplateheightofthemonomerH,theretentiontimeofthemonomertR,1andtheselectivitya betweenthemonomerandthedimer.Theerrorbarsaregivenat1levelandreflectthemembrane-to-membranereproducibility.
ChannelI
Apoferritin Thyroglobulin
˙Vc/ ˙Vout(mL/min) H(mm) tR,1(min) a H(mm) tR,1(min) a
Flatmembrane, w=350m 0.8/0.8 0.88±0.02 4.66±0.17 1.35±0.02 0.69±0.01 6.32±0.30 1.35±0.01 1.0/1.0 0.66±0.02 4.62±0.25 1.36±0.01 0.54±0.02 6.26±0.33 1.36±0.00 1.5/1.5 0.43±0.01 4.51±0.16 1.37±0.02 0.42±0.01 6.33±0.31 1.36±0.02 MSmembraneI, w=350m 0.8/0.8 1.03±0.01 7.78±0.04 1.42±0.01 0.87±0.02 11.15±0.24 1.46±0.01 1.0/1.0 0.78±0.03 8.09±0.08 1.45±0.00 0.69±0.01 11.45±0.18 1.51±0.01 1.5/1.5 0.65±0.04 8.25±0.23 1.49±0.01 – – – MSmembraneI, w=250m 0.8/0.8 1.27±0.02 4.59±0.04 1.44±0.01 0.91±0.04 6.57±0.15 1.45±0.02 1.0/1.0 0.93±0.05 4.92±0.37 1.46±0.02 0.79±0.02 6.63±0.10 1.48±0.01 1.5/1.5 0.64±0.00 5.00±0.04 1.51±0.01 0.54±0.01 7.39±0.04 1.53±0.02 ChannelII Apoferritin Thyroglobulin
˙Vc/ ˙Vout(mL/min) H(mm) tR,1(min) a H(mm) tR,1(min) a
Flatmembrane, w=350m 0.5/0.5 0.27±0.01 4.25±0.23 1.34±0.01 0.21±0.00 5.90±0.48 1.33±0.00 0.8/0.8 0.16±0.00 4.26±0.32 1.35±0.02 0.16±0.02 5.92±0.46 1.33±0.01 1.0/1.0 0.13±0.00 4.33±0.17 1.35±0.01 0.15±0.00 5.89±0.47 1.34±0.01 MSmembraneII, w=350m 0.5/0.5 0.43±0.01 8.88±0.14 1.49±0.01 0.39±0.02 13.16±0.13 1.50±0.01 0.8/0.8 0.33±0.01 9.90±0.08 1.55±0.01 0.36±0.01 14.56±0.09 1.57±0.02 1.0/1.0 0.29±0.00 12.22±0.05 1.59±0.01 – – – MSmembraneII, w=250m 0.5/0.5 0.53±0.01 4.61±0.03 1.48±0.00 0.47±0.00 6.80±0.14 1.53±0.02 0.8/0.8 0.34±0.02 5.47±0.08 1.56±0.02 0.32±0.01 8.28±0.18 1.63±0.00 1.0/1.0 0.29±0.01 5.74±0.02 1.62±0.01 0.30±0.00 8.88±0.20 1.68±0.01
channels/MSmembranesystemsandthesamespacerthicknessof 350m(Fig.5left-handfigures),thereisaconsiderableincrease inretentiontime,selectivityandresolutionbetweenmonomerand dimer.Althoughthereismorepeakbroadeningwiththepresence ofthegrooves,resolutionishigherbecauseofthehigherselectivity (asexpectedbythetheory,Fig.3).Consequently,thesame resolu-tioncouldbeachievedwiththeMSmembranesand applyinga lowercross-flowrate,oralternativelyusingathinnerspacer(Fig.5 right-handfigures).
Therefore,theMSmembranescouldbebeneficialasthesame retentionandresolutioncouldbeachievedwithlowercross-flow rates without the need to increase spacer thickness or to use impracticalcross-flowtooutletflowratios.Inpracticethatwould beparticularlyusefulforrelativelysmallsolutes(suchasBSAor evensmaller)sincelargersolutescanbeanalyzedwithoptimal spacerthicknessandflowrates,andthereforethereisaneedto fabricateMSmembraneswithlowerMWCO.Thechallengesand theprocedurestooptimizeAF4methodsforsmallsoluteswitha 300DaMWCOmembranehavebeenreported[28].Smaller macro-moleculesaretypicallyanalyzedbysizeexclusionchromatography (SEC)wheretheyexhibitverygoodresolution,butinsomecases, SECisnotsuitable,forinstance,whenthereisstrongnon-specific adsorptioninthechromatographicsupportandwhenlarge macro-moleculesco-existinthesamplethatneedtobeanalyzed.Inthe lastcase,across-flowprogramwithexponentialdecayshouldbe usedasthegrooveswouldcauseverystrongretentionforthelarge components.
Aseriesofexperimentswerecarriedoutusingdifferent cross-flowrateswhileretainingtheratio( ˙Vc/ ˙Vout=1);theresultsare
displayedinTable2.Theplateheightofthemonomerwas esti-matedfromthewidthathalfpeakheight.Anumberofconclusions maybedrawnfromtheseexperimentalresults.Althoughthereis alargedepartureoftheengineeredgroovesfromthetheoretical model(i.e.,noslip,infinitesimalridgeandrectangularshape),the underlyingconclusionswerefoundsimilar.
First,fromTable2,itcanbeseenthatforhigher ˙Vcandsame
˙Vc/ ˙Vout the retention time of the monomer and theselectivity
betweenmonomeranddimerweresimilarfortheflatmembranes asexpectedbythetheory(Eqs.(1)and(2))butincreasedfortheMS membranes.Thisisinlinewiththetheoreticalequationsderived bymomentanalysis(Eqs.(15)and(16)).Secondly,forthesame experimentalconditions,theincreaseinselectivitywashigherfor thyroglobulin(lower)anditwasindependentofthespacer thick-ness,aspredictedbythetheory.Lastly,theincreaseinretention timeandselectivityforthesamecross-flowvelocitywashigherfor theMSmembraneII,probablyduetothesmallerslipbecauseof thesmallerperiodicityofthegrooves.
It is howeverimportant tonotethat partof theincrease in retentiontime isa resultofthelargeractualchannelthickness withtheMSmembranes.Asitwasmentionedabove,thechannel thicknesswasestimated∼12%largerwithMSmembranes,which correspondsto∼25%longerretentiontimescausedbytheeffectof themembranecompression,astheretentiontimeisproportional tow2(Eq.(1)).Evenso,intheexperimentalresults(Table2)we
observeamuchhigherincreaseintheretentiontimes,namelyfrom 67%(forChannelIandacrossflowrateof0.8mL/min)to180%(for ChannelIIandacross-flowrateof1.0mL/min)forthemonomerof apoferritin,whichindicatesthattheeffectofthegrooveshasthe largestcontributiontotheincreaseintheretentiontime.Moreover, overloadingwasinvestigatedbyinjectingdifferentsamplemass, namely2g,10g,and20g,intheChannelI/MSmembraneI system;nooverloadingeffectwasobservedasretentiontime,plate heightandselectivitywerepracticallythesameforeveryexamined injectedmass.
4.3. Computationalfluiddynamics
FortheCFDexperiments,theminiaturized channel(Channel IIin Fig. 4)wasmodelled and themigration ofthe apoferritin monomeranddimerwassimulated.Thediffusioncoefficientsfor themonomerandfor thedimer ofapoferritinweretakenfrom theAF4experiments(wheretheratioofthediffusioncoefficients, andtherefore theselectivitywas1.34).Themodelwasverified byreducingsignificantlythesizeofthemeshelements,thetime
Table3
CFDexperimentsandcomparisonwiththeexperimentalresultsforapoferritin, ChannelIIandcross-flowrate0.5mL/min.
tR,1(s) tR,2(s) a
Flatmembrane-Exper. 255 342 1.34
Flatmembrane-CFD 261(SE=2%) 350(SE=2%) 1.34(SE<1%)
MSIImembrane-Exper. 533 794 1.49
MSmembraneII-CFD 720 1310 1.82
stepandthetolerances;allthesechangesdidnotalterthe reten-tiontimes(within0.2%).TheCFDmodelwasvalidatedcomparing theresults(retentiontimeandselectivity)withtheexperimental resultsobtainedforthenonpatternedmembrane.Goodagreement wasfound,thestandarderror(SE)was2%fortheretentiontime ofapoferritin (monomeror dimer)and<1%for theirselectivity (Table3).Theassumptionofthelaminarflowconditions,which wasusedforthemodel,wasjustifiedbytheresults;theReynolds numberwaslessthan0.05acrossthewholechannel(maximum inthemiddleofthechannelthickness)andlessthan2·10−4inside
thegrooves.
For the patterned membrane (MS membrane II) and ˙Vc=0.5mL/min, the flow and concentration profiles, and the
derived concentration at the outlet for each time point are depictedinFig.6.Itwasrevealedthattheexperimentalretention time and selectivity are much lower compared to the values predicted by the simulation (Table 3). This may be due to a non-uniform cross-flow velocity as a result of differences in themembranecompactionand/orintheporesizebetweenthe ridgesandthecavitiesofthemembrane’sselectivelayerasitwas discussedabove.
5. Conclusions
Todateonlyflat(non-patterned)UFmembraneshavebeenused inAF4asmicron-sizedfeaturesareconsideredharmfulforthe sep-aration.Wehavedemonstratedthatmicron-sizedgroovescouldin factimproveperformanceinAF4.Thiswasshownbyseveralmeans includingmomentanalysis,physicalexperiments,andCFD simu-lations.Ourresultsshowthatperpendiculargroovescanincrease retention,selectivityandresolution.Thissystemcouldbeusefulas macromoleculesandnanoparticlescouldbeanalyzedwithlower cross-flowrateswithouttheneedtousehigherspacerthickness orhigher cross-flowtooutletflow ratio.Thisconcept couldbe appliedonanyFFFsystemasithasbeenoriginallydemonstrated byGiddingsetal.forThFFF[12].
Thephysical experimentswerecarried outwith microstruc-turedUFmembranesfabricatedbyhot-embossing.Thisfabrication processcausedanincreaseintheactualMWCOofthemembrane (asindicatedbytheAF4experiments)buttheeffectofthegrooves couldbeshownwiththelargerproteinstandardsusedinthisstudy (apoferritin 443kDa and thyroglobulin 669kDa). However, this conceptcouldbeparticularlyusefulforsmallermacromolecules, andthereforefutureworkshouldbefocusedonthefabricationof microstructuredmembraneswithlowerMWCOandhighwater permeability. This couldbe achieved,for instance, by methods otherthanhotembossingsuchasphaseseparationoradditive tech-nologies(e.g.,3Dprintingofanon-porousorporousmaterialover anUFmembrane).AdditionalresearchwithCFDexperimentsof differentgrooveshapesanddimensionsisunderwaytoinvestigate theoptimalgroovestructure.
Acknowledgements
Thisworkwas partof theresearchprogram SmartSep with projectnumber 11400 which wasfinanced by theNetherlands
OrganizationforScientificResearch(NWO).Authorsalso acknowl-edgeLydiaBolhuis-Versteeg(UniversityofTwente)forherhelpon SEMimagingandWyattTechnologyEuropeforprovidingtechnical assistance.
References
[1]C.Contado,Fieldflowfractionationtechniquestoexplorethe“nano-world”, Anal.Bioanal.Chem.409(2017)2501–2518,http://dx.doi.org/10.1007/ s00216-017-0180-6.
[2]M.I.Malik,H.Pasch,Field-flowfractionation:newandexcitingperspectives inpolymeranalysis,Prog.Polym.Sci.63(2016)42–85,http://dx.doi.org/10. 1016/j.progpolymsci.2016.03.004.
[3]F.vonderKammer,S.Legros,E.H.Larsen,K.Loeschner,T.Hofmann, Separationandcharacterizationofnanoparticlesincomplexfoodand environmentalsamplesbyfield-flowfractionation,TrAC–TrendsAnal.Chem. 30(2011)425–436,http://dx.doi.org/10.1016/j.trac.2010.11.012.
[4]S.Podzimek,LightScattering,SizeExclusionChromatographyand AsymmetricFlowFieldFlowFractionation,JohnWiley&Sons,Inc.,Hoboken, NewJersey,2011.
[5]K.G.Wahlund,J.C.Giddings,Propertiesofanasymmetricalflowfield-flow fractionationchannelhavingonepermeablewall,Anal.Chem.59(1987) 1332–1339,http://dx.doi.org/10.1021/ac00136a016.
[6]A.Litzén,K.G.Wahlund,Zonebroadeninganddilutioninrectangularand trapezoidalasymmetricalflowfield-flowfractionationchannels,Anal.Chem. 63(1991)1001–1007,http://dx.doi.org/10.1021/ac00010a013.
[7]R.N.Qureshi,W.T.Kok,Optimizationofasymmetricalflowfield-flow fractionation(AF4),LCGCEur.23(2010)18–25.
[8]J.S.Yang,M.H.Moon,Flowoptimisationswithincreasedchannelthicknessin asymmetricalflowfield-flowfractionation,J.Chromatogr.A1581–1582 (2018)100–104,http://dx.doi.org/10.1016/j.chroma.2018.10.053. [9]J.M.Davis,Influenceofcrossflowhydrodynamicsonretentionratioinflow
field-flowfractionation,Anal.Chim.Acta246(1991)161–169.
[10]J.Janˇca,Field-flowfractionationinmicrofluidicchannelsoflowaspectratio, Int.J.Polym.Anal.Charact.22(2017)62–71,http://dx.doi.org/10.1080/ 1023666X.2016.1232558.
[11]P.S.Williams,S.B.Giddings,J.C.Giddings,Calculationofflowpropertiesand endeffectsinfield-flowfractionationchannelsbyaconformalmapping procedure,Anal.Chem.58(1986)2403–2408,http://dx.doi.org/10.1021/ ac00125a010.
[12]J.C.Giddings,L.K.Smith,M.N.Myers,Surfacebarriersforretention enhancementinfield-flowfractionation,Sep.Sci.Technol.13(1978) 367–385,http://dx.doi.org/10.1080/01496397808057108.
[13]A.D.Stroock,S.K.W.Dertinger,A.Ajdari,I.Mezic,H.aStone,G.M.Whitesides, Chaoticmixerformicrochannels,Science295(2002)647–651,http://dx.doi. org/10.1126/science.1066238.
[14]J.A.Bernate,C.Liu,L.Lagae,K.Konstantopoulos,G.Drazer,Vectorseparation ofparticlesandcellsusinganarrayofslantedopencavities,LabChip13 (2013)1086–1092,http://dx.doi.org/10.1039/c2lc40927e.
[15]C.L.M.H.Navier,Mémoiresurlesloisdumouvementdesfluides,Mémoiresde l’AcadémieR,DesSci.l’InstitutFr.6(1823)386–436.
[16]C.Y.Wang,Flowoverasurfacewithparallelgrooves,Phys.Fluids15(2003) 1114–1121,http://dx.doi.org/10.1063/1.1560925.
[17]T.Lee,E.Charrault,C.Neto,Interfacialsliponrough,patternedandsoft surfaces:areviewofexperimentsandsimulations,Adv.ColloidInterfaceSci. 210(2014)21–38,http://dx.doi.org/10.1016/j.cis.2014.02.015.
[18]P.S.Williams,Retentionratioandnonequilibriumbandspreadingin asymmetricalflowfield-flowfractionation,Anal.Bioanal.Chem.407(2015) 4327–4338,http://dx.doi.org/10.1007/s00216-015-8734-y.
[19]S.H.Maruf,M.Rickman,L.Wang,J.MerschIV,A.R.Greenberg,J.Pellegrino,Y. Ding,Influenceofsub-micronsurfacepatternsonthedepositionofmodel proteinsduringactivefiltration,J.Memb.Sci.444(2013)420–428,http://dx. doi.org/10.1016/j.memsci.2013.05.060.
[20]S.H.Maruf,L.Wang,A.R.Greenberg,J.Pellegrino,Y.Ding,Useof nanoimprintedsurfacepatternstomitigatecolloidaldepositionon ultrafiltrationmembranes,J.Memb.Sci.428(2013)598–607,http://dx.doi. org/10.1016/j.memsci.2012.10.059.
[21]S.H.Maruf,Z.Li,J.A.Yoshimura,J.Xiao,A.R.Greenberg,Y.Ding,Influenceof nanoimprintlithographyonmembranestructureandperformance,Polymer 69(2015)129–137,http://dx.doi.org/10.1016/j.polymer.2015.05.049. [22]M.Bikel,I.G.M.Pünt,R.G.H.Lammertink,M.Wessling,Shrinkageeffects
duringpolymerphaseseparationonmicrofabricatedmolds,J.Memb.Sci.347 (2010)141–149,http://dx.doi.org/10.1016/j.memsci.2009.10.015.
[23]L.Vogelaar,R.G.H.Lammertink,J.N.Barsema,W.Nijdam,L.A.M. Bolhuis-Versteeg,C.J.M.VanRijn,M.Wessling,Phaseseparation micromolding:anewgenericapproachformicrostructuringvarious materials,Small1(2005)645–655,http://dx.doi.org/10.1002/smll. 200400128.
[24]Y.-J.Won,J.Lee,D.C.Choi,H.R.Chae,I.Kim,C.H.Lee,I.C.Kim,Preparationand applicationofpatternedmembranesforwastewatertreatment,Environ.Sci. Technol.46(2012)11021–11027,http://dx.doi.org/10.1021/es3020309. [25]R.Aris,G.I.Taylor,Onthedispersionofasoluteinafluidflowingthrougha
tube,Proc.R.Soc.London.Ser.A:Math.Phys.Sci.235(1956)67–77,http://dx. doi.org/10.1098/rspa.1956.0065.
[26]N.Meechai,A.M.Jamieson,J.Blackwell,Translationaldiffusioncoefficientsof bovineserumalbumininaqueoussolutionathighionicstrength,J.Colloid InterfaceSci.218(1999)167–175,http://dx.doi.org/10.1006/jcis.1999.6401. [27]J.C.Giddings,P.S.Williams,M.A.Benincasa,Rapidbreakthrough
measurementofvoidvolumeforfield-flowfractionationchannels,J. Chromatogr.A627(1992)23–35.
[28]C.W.Cuss,I.Grant-weaver,W.Shotyk,AF4-ICPMSwiththe300Damembrane toresolvemetal-bearing“Colloids”<1kDa:optimization,fractogram deconvolution,andadvancedqualitycontrol,Anal.Chem.89(2017) 8027–8035,http://dx.doi.org/10.1021/acs.analchem.7b01427.