• Aucun résultat trouvé

An evaluation of thermodynamic models for the prediction of drug and drug-like molecule solubility in organic solvents

N/A
N/A
Protected

Academic year: 2021

Partager "An evaluation of thermodynamic models for the prediction of drug and drug-like molecule solubility in organic solvents"

Copied!
18
0
0

Texte intégral

(1)

This is an author-deposited version published in:

http://oatao.univ-toulouse.fr/

Eprints ID: 6017

To link to this article

:

DOI:10.1016/J.FLUID.2011.06.032

URL: http://dx.doi.org/10.1016/J.FLUID.2011.06.032

To cite this version

:

Bouillot, Baptiste and Teychené, Sébastien and

Biscans, Béatrice (2011) An evaluation of thermodynamic models for the

prediction of drug and drug-like molecule solubility in organic solvents.

Fluid Phase Equilibria, vol. 309 (n°1). pp. 36-52. ISSN 0378-3812



O

pen

A

rchive

T

oulouse

A

rchive

O

uverte (

OATAO

)

OATAO is an open access repository that collects the work of Toulouse researchers and

makes it freely available over the web where possible.

Any correspondence concerning this service should be sent to the repository

administrator:

staff-oatao@listes.diff.inp-toulouse.fr



(2)

An

evaluation

of

thermodynamic

models

for

the

prediction

of

drug

and

drug-like

molecule

solubility

in

organic

solvents

Baptiste

Bouillot

,

Sébastien

Teychené,

Béatrice

Biscans

LaboratoiredeGénieChimique,UMRCNRS5503,BP84234CampusINP-ENSIACET,4alléeEmileMonso,31030ToulouseCedex4,France

a

b

s

t

r

a

c

t

Predictionofsolubilityofactivepharmaceuticalingredients(API)indifferentsolventsisoneofthemain issueforcrystallizationprocessdesign.Experimentaldeterminationisnotalwayspossiblebecauseof thesmallamountofproductavailableintheearlystagesofadrugdevelopment.Thus,oneinteresting perspectiveistheuseofthermodynamicmodels,whichareusuallyemployedforpredictingtheactivity coefficientsincaseofVapour–LiquidequilibriaorLiquid–Liquidequilibria(VLEorLLE).Thechoiceofthe bestthermodynamicmodelforSolid–Liquidequilibria(SLE)isnotaneasytaskasmostofthemarenot meantparticularlyforthis.Inthispaper,severalmodelsaretestedforthesolubilitypredictionoffive drugsordrug-likemolecules:Ibuprofen,Acetaminophen,Benzoicacid,Salicylicacidand4-aminobenzoic acid,andanothermolecule,anthracene,arathersimplemolecule.Theperformanceofpredictive (UNI-FAC,UNIFACmod.,COSMO-SAC)andsemi-predictive(NRTL-SAC)modelsarecomparedanddiscussed accordingtothefunctionalgroupsofthemoleculesandtheselectedsolvents.Moreover,themodelerrors causedbysolidstatepropertyuncertaintiesaretakenintoaccount.Theseerrorsareindeednot negli-giblewhenaccuratequantitativepredictionswanttobeperformed.ItwasfoundthatUNIFACmodels givethebestresultsandcouldbeanusefulmethodforrapidsolubilityestimationsofanAPIinvarious solvents.Thismodelachievestheorderofmagnitudeoftheexperimentalsolubilityandcanpredictin whichsolventsthedrugwillbeverysoluble,solubleornotsoluble.Inaddition,predictionsobtained withNRTL-SACmodelarealsoingoodagreementwiththeexperiments,butinthatcasetherelevance oftheresultsisstronglydependentonthemodelparametersregressedfromsolubilitydatainsingle andmixedsolvents.However,thisisaveryinterestingmodelforquickestimationslikeUNIFACmodels. Finally,COSMO-SACneedsmoredevelopmentstoincreaseitsaccuracyespeciallywhenhydrogen bond-ingisinvolved.Inthatcase,thepredictedsolubilityisalwaysoverestimatedfromtwotothreeorders ofmagnitude.Consideringtheuseofthemostaccurateequilibriumequationinvolvingthe1Cpterm, nobenefitswerefoundfordrugpredictionsasthemodelsarestilltooinaccurate.However,infunction ofthemoleculesandtheirsolidthermodynamicproperties,the1Cptermcanbeneglectedandwillnot haveagreatimpactontheresults.

1. Introduction

Solubilityofsolidcompoundsinsolventsisobviouslyoneofthe mostfundamentalphysicochemicalproperties,particularlyuseful ina widevarietyofphenomenarelevanttobiological, pharma-ceutical,environmentalandotherorganic,physicalandanalytical chemistry, and engineering sciences. From a simple thermody-namicpoint ofview,solubilityis thefugacityequalitybetween thesolidstateandtheliquidstateofthesolutemolecule.In prac-tice, it is the maximum amount of the crystallineform of the compoundthatcanbedissolvedinagivenvolumeofsolventat

∗ Correspondingauthor.

E-mailaddress:baptiste.bouillot@ensiacet.fr(B.Bouillot).

a given temperature and pressure. This stateis theexpression oftheGibbsfreeenergyofmixing,whichisthecombinationof twocontributions.Theenthalpiccontributionwhichreflectsthe solute–solute and solute–solventinteractions; and theentropic contributionwhichtakesintoaccountthemolecularsizeandthe potentialfor non-randomdistributionof moleculesin the mix-ture.Then,themolecularsizeandthepotentialfornon-random distributionarealsoimportant.Fromapracticalpointofview, sol-ubilityhasa considerableinfluenceonthechoiceofsolventfor performingpolymorphism,saltandco-crystalscreeninganditis thekeyparameterfortheprocessdesignofanewdrugby crys-tallization,asitwilldeterminetheamountofcoolingrequiredto yieldagivenamountofproductandwillinfactdetermineifcooling willprovideareasonableproductyieldandthusdefinethe crys-tallizationprocessitself(cooling,evaporation,orantisolvent).In

(3)

addition,attheearlystagesofanewdrugmoleculedevelopment, smallamountofproduct(typicallylessthan5mg)isavailable,so classicalsolubilitymethodscannotbeusedsincetheyareproduct consumingandareneitherrapidnorcosteffective.Toovercome theseissues,high-throughputsolubility-screeningprotocolshave beendevelopedbasedonsystemsusingeitherrobots[1]ordroplet microfluidic[2].Allthesemethodsrelyonkineticdetermination ofdissolutionandre-crystallizationsteps.Theprocessof dissolu-tionofacrystallinecompoundintoasolventgenerallyshowsan asymptoticprofile[3]:astheconcentrationofthecompoundin solutionincreases,theprocessbecomesprogressivelyslower,and saturationistheoreticallyneverreached.Inaddition,polymorphic phasetransitionmayoccurduringtheseexperiments,leadingto misinterpretationoftheresults.

Evenwhenproductsareavailable,theexperimental determina-tionofsolubilitymustbepreparedandcompletedbypreliminary calculations.Indeed,in classicalsealedflaskmethods,saturated solutionsarepreparedwithanexcessofsolidatagiven temper-ature.Thesuspensionisthenfiltratedorcentrifugedtoensurea solutioncompletelyfreefromundissolvedsolute.First, bytheir nature,this traditional solubilitymethodis timeconsuming. In addition,thefiltrationorcentrifugationstepsareoftenperformed at a lower temperature than the saturationexperiments, what entailsashiftinsolubilityvalues.

Indeed,experimentalmethodsofsolubilitymeasurementsare timeconsuming,expensiveandoftenhavetechnicalproblems,in spiteoftheobtainedvaluesvaryinginthedegreeofaccuracyand precision.Toguidetheseexperimentsandtolimit misinterpreta-tion,thermodynamicmodellingofSolid–LiquidEquilibrium(SLE), usingaminimumofexperimentaldata,wouldbeofagreatvalue. BasedontheactualSLEthermodynamics,aquestionarises:canthe existingmodelsbeusedasguidefordesigningsolubility experi-ments?[4]showedthattheoriginalUNIFACmodelcouldpredict SLE.But,theresultswereaccurateonlyforrestrictedcompounds. [5]discussedthelimitsoftheUNIFACmodeland theequilibria equationsusedforSLE,butdidnotfocusonthetemperature depen-denceorcomparewithothermodels.[6]showedthatCOSMO-SAC isaninterestingmodeltocalculateapriorisolubilities,butitis notveryaccurateforSLE.Inaddition,thesepapersdidnot investi-gateneitherthecompleteequilibriumequationnortheinfluence oftemperatureonsolubilitypredictions.[7]usedNRTL-SACforSLE predictionandfounditpromising,butdidnotcomparedtheresults toamuchmoremodels.Morerecently,[8]comparedUNIFACmodel toCOSMO-RSmodelforsomeorganiccompounds.

The main objective of this paper is to test and analyze existingpredictive(UNIFAC,UNIFACmod.,COSMO-SAC)and semi-predictive(NRTL-SAC)thermodynamicmodelsforthepredictionof solubilityofsixcompounds(fivedrugsordrug-likemoleculesand onehydrocarbon)inseveralsolvents.Fourclassicdrugswere cho-senbecauseofthelargenumberofexperimentalsolubilitydataand ofthedifferentchemicalgroups(Ibuprofen,Acetaminophen, Ben-zoicacid,andSalicylicacid).Tocompletethisset,4-aminobenzoic acidwasalsochosengivenitsaminegroup,andanthracene,aquite simplemolecule(onlyACHgroups).

Inthis work,thethermodynamicmodelswerechosenbased ontheirvariousapproaches,andtheirabilitytobeusedwithor withoutexperimentaldata.UNIFACanditsmodificationsaregroup contributionmethods.Theyonlyrequirebinaryinteraction param-eters betweenthefunctional groups. Thesemodels relyonthe useofaspecificdatabaseofgroup–groupinteractionparameters. Whentheseparametersarenotavailable(formoleculescontaining cyclicSorNforinstance),thesemodelscannotbeused. COSMO-SAC,isbasedonquantummechanicsandrequiresprofilesofthe chargedensityaroundthestudiedmolecules(called-profile).The challengeisthedeterminationofthe“good”-profilewhichtakes intoaccounttherightconformationsofthemoleculeinthe

solu-tion.NRTL-SAC isbased onpolymerNRTL andtobeused,four parametersrepresentingthebehaviourofthemoleculeshavetobe calculatedfromexperimentaldataregression.Oncethese param-etershavebeencalculatedforonemolecule,theymaybeusedfor anyothersolubilitypredictionsinanysolvent.Thatiswhythere isnoneedtoregresseachtime,asinthecaseofmostthe semi-predictivemodelslikeUNIQUACorWilson.

Thesefourmodelswillbecomparedanddiscussedregarding solubilitypredictionatseveraltemperatures.Inaddition,the rela-tiveimportanceofthepuresolidstateproperties(i.e.themelting temperature,melting enthalpyandtheheat capacity difference betweenthesupercooledliquidandthesolid)onthe thermody-namicmodellingisdiscussed.

2. Theory

2.1. Equilibriumequation

Phaseequilibriaaredescribedbytheequalityofchemical poten-tials,,ineachphase.Thegoverningequationofsolubilityofthe compoundinagivensolventiswrittenasfollows:

S=Sat (1)

Thesuperscript‘S’denotesthesolidphase,andthesuperscript‘Sat’ thesaturatedsolution.Thechemicalpotentialofthesoluteinthe saturatedsolutionis:

Sat=0+RTln( SatxSat) (2)

where0 isthechemicalpotentialinthereferencestate,Tthe

temperature,and theactivitycoefficient.Theactivityofthesolute (a= x)inthesolventisobtainedbycombiningEqs.1and2: ln( SatxSat)= S−0 RT = gSg0 RT = 1gm RT (3)

where1gmisthechangeofpartialmolarGibbsenergyofthesolute

fromasolidstatetothereferencestateatconstanttemperatureand pressure.Thislastequationgivesarelationshipbetweensolubility andactivitycoefficients,whichcanfinallybewrittenas[9]: ln( SatxSat)= 1Hm(Tm) R



1 Tm −1 T



−1Cp(Tm) R

h

ln



Tm T



−Tm T +1

i

(4) where1Hmisthemeltingenthalpy,Tm,themeltingtemperature

and1Cpthedifferencebetweentheheatcapacityofthe

super-cooledmeltandtheheatcapacityofthesolid.Inpractice,thislast termcanbeverydifficulttomeasureespeciallywhensublimation, decomposition[10]orparallelreactionoccursduringmelting,and isoftenneglectedbecauseitisusuallythoughtthatitscontribution isnegligible.InthatcaseEq.(4)issimplifiedasfollows:

ln( SatxSat)= 1Hm(Tm) R



1 Tm −1 T



(5) 2.2. Groupcontributionmodels(classicandmodifiedUNIFAC)

UNIFACmethod,forUNIQUACfunctionalgroupactivity coef-ficient [11],is basedontheUNIQUAC model. Themain ideaof thisapproachisaseparationoftheactivitycoefficientofthepure compoundintotwoparts:

ln =ln C+ln R (6)

Cisthecombinatorialterm.Itrepresentstheentropic

contribu-tiontotheactivitycoefficientwhichtakesintoaccounttheshape andsizeofthemolecules.This C dependsonthemolefraction

(4)

Waalsarea(ri)andvolume(qi)(thesuperscript‘i’designatesthe compound): ln C i =ln 8i xi +5qiln i 8i +li−8i xi

X

j xj 5 rj−qj



− rj−1



(7) Ristheresidualpartwhichrepresentsthecontributionofinter

andintramolecularinteractions(enthalpiccontribution).Itisasum oftheactivitycoefficientsofthefunctionalgroupsweightedby theirnumberinsolution.

ln R i =

X

k

v

(i) k

h

lnŴk−Ŵ(i)k

i

(8) where

v

kand

v

i

karethenumberofgroupsofkindkinthemixture

andincomponenti.ŴkandŴikare,respectively,theresidual

activ-itycoefficientofgroupkinthemixtureandinasolutionofpure componenti.Theydependontheareaandsegmentfractionthe compoundsandonadjustablebinaryinteractionparametersamn

(theinteractionsbetweenthefunctionalgroups)thatareregressed usuallyfromVLEexperimentaldata:

lnŴk=Qk

"

1−ln

!

X

m 2m9mk

#

X

m 2m9km/

X

n 2n9nm

%

(9) with: 9mn=exp



−amn T



(10) InUNIFACmod.(Dortmund),by[12],amodifiedversionof UNI-FAC,issupposedtotakeabetteraccountofthetemperaturein theresidualactivitycontribution.Thenewcombinatorialtermand interactionparameterarewrittenas,respectively:

ln C i =1−ln



8 i xi



−5qi



1−8i i +ln



8i i



(11) and 9mn=exp



−amn+bmnT+cmnT 2 T



(12) UNIFACandmodifiedUNIFACarefrequentlyusedfortheprediction ofVapor-LiquidandLiquid-Liquidequilibriaforsimplemolecules. Thesemodelscanbeappliedtoallmoleculesprovidingthatthe binaryinteractionparametersexist,whichisnotalwaysthecase. Moreover,theconformationandisomerismofmoleculesarenot considered.

2.3. Segmentactivitycoefficientmodels 2.3.1. COSMO-SAC

COSMO-SACisbasedonCOSMO-RSmodel[13,14].First pub-lished by Lin and Sandler [15],this model is a combination of aquantumchemicaltreatmentofsolutesandsolventswith sta-tisticalthermodynamics procedurefor thedeterminationofthe molecularsurfaceinteractions.

TheCOSMO-SACmodelisbasedonthecalculationofthe sol-vationfree energy1G*sol whichdefines thefree energychange

comparingtoanidealsolutionatconstanttemperatureand pres-sure,whenthesolutemoleculesareintroducedintothesolvent.In COSMO-basedmodels,thistermisdividedintotwoparts.First,the solutechargesareturnedoff,anditsmoleculesareinsertedinto aperfect conductor(withaninfinitedielectricconstant



).This firststepgivesthecavitationformationfreeenergy,1G*cav,which

dependsonthemoleculespecificdimensionandrepresentsthe entropiccontributionoftheactivitycoefficient.Whenthecharges

areturnedon(thesolventbecomesarealsolvent),thefreeenergy ofrestoringthecharges,1G*res,correspondstotheenthalpic

con-tribution.Combining these two steps,theactivitycoefficient is expressedas:

ln i=ln C+

1G∗res

i/S −1G∗resi/i

RT (13)

Tocalculatetherestoringfreeenergy,a-profileofeachmolecule isneeded.Thisprofilecorrespondstotheprobabilityoffindinga surfaceofthemoleculewithachargedensity.Theyarecomputed fromquantumchemicalsimulations.Oncetheyhavebeen calcu-lated,themoleculesaredividedintosegments.Theseconceptual segmentsareportionsofthemoleculesurfacewiththesamecharge density.Theequationfortheactivitycoefficientisthenwritten: ln i=ni

X

m

pi(m)[lnŴS(s) −lnŴi(m)] + iC (14)

whereŴs(m)andŴi(m)aretheactivitycoefficientsofsegmentm

insolutionandinpureliquid,respectively.InCOSMO-SACmodel, theactivitycoefficientisobtainedbysummingthecontributions ofeachsegment.

The -profiles used in this work were taken from Mullins database [6,16]. Unlike other group-contribution methods, this modelisnotstrictlyapair-wiseadditiveone,sincethe-profiles takeintoaccountthegeometryofmoleculesandhydrogen bond-ing. As the -profiles take into account the geometry of the molecules,eachmoleculeanditsconformationhaveitsownprofile [16].

2.3.2. NRTL-SAC

NRTL-SAC model [17] is based on the original NRTL (Non-RandomTwo-Liquid)[18]andpolymerNRTL[19].TheNRTL-SAC modelcharacterizesthemoleculesintermsofpre-defined concep-tualsegments.Thesesegmentsaccountfortheinteractionsofeach moleculeinthesolution.[17]havedefinedfourtypesofconceptual segments:hydrophobicX,repulsiveY−andattractiveY+polar,and

hydrophilicZ.LikeinUNIFAC,theNRTL-SACmodelcomputesthe activitycoefficientforcomponentifromthecombinatorialterm C

andtheresidualterm R(Eq.(6)).Thecombinatorialtermisthen

calculatedfromtheFlory–Hugginsapproximationfortheentropy ofmixing: ln iC=ln8i xi +1−ri

X

j 8j rj (15) whereriand8iare,respectively,thetotalsegmentnumberand

thesegmentmolefractionofcomponenti.

Theresidualterm R iscalculatedfromthelocalcomposition

(lc)interactioncontribution[19]appliedtoeachsegment: ln iR=ln Ilc=

X

k rk,I

h

lnŴlck −lnŴlc,ik

i

(16)

Thesegment–segmentinteractionparametersandtheconceptual segmentvaluesofsolventsweredeterminedbytheregressionof experimentalvapor–liquidandliquid–liquidequilibriumdata[17]. Thesegmentvaluesforasoluteareobtainedbytheregressionof solubilitydatainatleastfoursolvents:onehydrophilic(i.e.water), onepolarattractor(i.e.alcohol),onepolardonor(i.e.ketone),and onehydrophobic(i.e.n-alkane).As suggestedby[17],it isalso possibletoestimatepuresolidpropertiesusingEq.(4)or(5). How-ever,inthisstudy,inordertocomparethemodelperformances withtheother,wechosetouseexperimentalsolidstateproperties, sincetheyareavailableintheliteratureandareeasilyaccessible experimentally.Oncethesegmentvaluesofthesoluteareobtained, theycanbeusedtopredictsolubilityinothersolventsorsolvent mixtures.

(5)

2.4. Howtheerrorsonthethermodynamicpropertiesimpactthe modelspredictions?

AsthesemodelsareusedwiththeSLEequation,thesolidstate propertiesandtheexperimentaluncertaintieshaveanimpacton theparameterdeterminationormodelprediction.Indeed,andas showninthenextsection,measurementsofthethermodynamic propertiesof thesolidsarescarcelyaccurateandtoinvestigate themodelperformances,theirexperimentaluncertaintieshaveto betakenintoaccount.Thus,thesolidstateproperties,1Hm,Tm

and1Cp,aredefinedbymeanvalues withstandard deviations.

Using Eq.(4), there arethree parameters(two forEq. (5)).The influenceoftheexperimentaluncertaintiesonsolubility predic-tionisevaluatedbyusingaclassicMonteCarlomethod.Solubility predictionsareperformedbychoosingrandomlyvectorsofmean valueandstandarddeviationcorrespondingtotheexperimental data,anddescribeditbyaGaussianlaw.Then,thepredicted sol-ubilitystandarddeviationiscalculated(aboutathousandrandom simulations).

3. Experimentalsolubilityandpurecompoundproperties 3.1. Referencedrugs

Inthis paper,MainlyformIofacetylp-aminophenol (parac-etamol),2-(p-isobutylphenyl)propionicacid(ibuprofen),benzoic acid, and2-hydroxybenzoic acid(salicylicacid) werechosenas modeldrugs.Thesemoleculeswerechosenmainlybecauseofa significantamountofsolubilityexperimentaldataavailableinthe literature[20–32]andtheycontainvarioususualfunctionalgroups (OH, COOH, NH, NH2).To complete this solubility database,

4-aminobenzoicacidwasconsideredbecauseofitsstructure(amine group)despitethereisfewdataavailable[31–34],andanthracene becauseofitssimple structure(ACHfunctionalgroups)andthe manysolubilitydataavailable[35–39]althoughitisnota drug. ThemoleculesstructuresareshowninFig.1.

3.2. Experimental

Inthisstudy,inspiteoftheexperimentalresultsbeingtaken from the literature, some data were missing. The solubility of ibuprofeninoctanolandheptane,paracetamolinheptaneat30◦C

weremeasured.Then,themeltingtemperatureandenthalpyof ibuprofen,paracetamol,salicylicacidandbenzoicacidwere mea-sured.Combiningwiththeliteraturedata,thegoalwastohave meanvaluesandstandarddeviationfortheseparameters.

3.2.1. Materials

Salicylic and benzoic acid were bought at Fisher Scientific (>99.5%),IbuprofenatBASF(Ibuprofen25>98.5%),and paraceta-molatSigma–Aldrich(>98%).Heptaneandoctanolwereboughtat AnalaRNORMAPUR(>99%).Theywereusedwithoutfurther purifi-cation.

Fig.1. (a)Ibuprofen,(b)paracetamol,(c)salicylicacid,(d)benzoicacid,(e) 4-aminobenzoicacidand(f)anthracene.

3.2.2. Determinationofthethermodynamicpropertiesofthe solids

Thethermodynamicpropertiesofthesolidsweredeterminedby DifferentialScanningCalorimetry(ThermalAnalysisQ2000DSC). Themeasuringmethodwasa1◦C perminuteramp from

ambi-enttemperatureto110◦C,220C,210Cand165Cforibuprofen,

paracetamol,salicylicacidandbenzoicacid,respectively.

3.2.3. Solubilitymeasurements

Solubilitymeasurementswerecarriedoutinwaterwith physi-calmixturesofenantiomersbytheanalyticalshake-flaskmethod, usingconstant-temperaturejacketedglasscells.Aconstant tem-peratureof30◦C(±0.1C)wasmaintainedwithacirculatingwater

bath.Thewayofpreparingthesaturated solutionswasslightly modifiedcomparedtotheclassicalmethod.Here,solubilitywas determinedbyaddingweighedquantitiesofpowdermixturetoa knownvolumeofdistilledwater,untilcrystalscannotbedissolved anymore.Smallamountsofsolid(about20mg)wereaddedeach timeandnoadditionwasperformedbeforethesolutionwasclear. Whentheequilibriumwasreached,solutionswerestirredfor24h. Theuncertaintyofsolubilitydetermination,1S,islowerthan 5%.

4. Results

4.1. Experimentalsolubilityandpurecompoundproperties

Concerning thethermodynamic properties ofthe solids,the experimentalresultsofthisworkandfoundintheliteratureare showninFig.2,andthemeanvaluesandstandarddeviationare presentedinTable1.

Table1

Meanvaluesandstandarddeviationofmeltingtemperatures,enthalpiesand1Cpofibuprofen,paracetamol,salicylicacid,benzoicacid,anthraceneand4-aminobenzoic

acidfromDSCexperiments(exceptthelasttwo)anddatatakenfromtheliterature[8,21,22,25,40–53].

Compound 1Hm(J/mol) Tm(K) 1Cp(J/(molK))

Ibuprofen[8,21,47,53] 25203.9(1577) 347.88(0.65) 48.8

Paracetamol[22,43,45–47,49,50] 27470.6(1720.3) 442.1(0.47) 87.4

Salicylicacid[25,47] 24626.12(2652.4) 431.37(0.12) None

Benzoicacid[40–42,44,48,51,52] 17350.34(774.7) 395.17(0.29) 58.43(0.81)

Anthracene[47,52] 28790(585.1) 489.18(0.89) 6.28

(6)

Fig.2.Meltingenthalpyof(a)ibuprofen,(b)paracetamol,(c)salicylicacidand(d)benzoicacidasafunctionofthemeltingtemperature.

Theexperimentalsolubilitydataobtainedinthisworkandfrom theliteratureare givenin Fig.3.The experiment errorsonthe solubilityarefrom1%to4%.

These solubility data will be used to test qualitatively and quantitativelythethermodynamicmodelsaccuracy.Thechosen criterionchosetoqualitativelyevaluatethemodelsisthe preserva-tionofthesolubilityrankingofanAPIinvarioussolvents(Table2). Thisscaleisdefined,inthiswork,asanorderedlistofsolvents fromthesolventinwhichtheAPIistheleastsolubletomost sol-uble.Quantitatively,themodelaccuracyisevaluatedbythemean squareerrorbetweenthemodelpredictionandtheexperimental molefractionsolubility.Thesensitivityofthemodelspredictionto thesolidstateproperties(1Hm,Tmand1Cp)havealsobeentaken

intoaccount.

4.2. OriginalUNIFACandmodifiedUNIFAC

Themeansquarerelativeerrorsobtainedforthesolubility pre-dictionusingUNIFACandUNIFACmodifiedforallthesolutesand solvents tested are reportedin Table 3. The experimentaldata usedarethesameasinFig.3.Foreachsolubilityprediction, stan-darddeviationwillbecomputedusingaMonteCarlomethodas explainedpreviously.Thesemeanrelativeerrorshavebeen calcu-latedusingthefollowingequation:

mse=1 n n

X

i=1



xi predicted−xi experimental xi experimental



2

Themeansquarerelativeerrors(mse)obtainedrangefrom0.07to 26.62forUNIFACandfrom0.137to322forUNIFACmod.Some pre-dictionsinparticularsolventsinducedhighmsevalues,sothatthe obtainednumbersdonottrulyrepresentthequalityofthe mod-els.For example,theupperpartoftherangesdecreasesto6.79 and18.04withoutchloroform.Despitetherelativegoodresults obtainedinthecaseofibuprofen,benzoicacidoranthracene,it canbestatedthatUNIFACseemsnotabletopredictquantitatively thesolubility.But,itisreallyinterestingtocomparethesolubility predictedordersofmagnitudetotheexperimental,andinthisway tocheckifthemethodsgivegoodapproximationsornot.

In addition, thepredictions obtainedat lower temperatures (20–25◦C)aremuchbetterthanthoseobtainedathigher

tempera-tures.TheoriginalUNIFACsolubilitypredictionsoftheconsidered compoundsasafunctionoftheexperimentalaregiveninFig.4.The behaviourobservedforUNIFACmod.isroughlythesame.Inthis paper,weusethetermUNIFACtodiscussaboutboththese mod-els.Inthisfigure,foronesoluteinonesolvent,itcanbeobserved thatthemorethetemperatureincreases,theless thepredicted solubilityisaccurate.Inaddition,thetemperaturedependenceof thebinaryparametersisintroducedinUNIFAC,byapolynomial ij=aijT2+bijT+cij.Evenifthisapproachgivesgoodresultsforsmall

moleculesandforVLE,itcannotrepresentthecomplexityofthe evolution ofinteractions andvolumetric properties (densityfor example)withtemperature.

UNIFACisapair-wiseadditivemodel,thusthetemperature evo-lutionofthespecificinteractions(likehydrogenbonds)cannotbe trulyrepresented.Moreover,evenifthebinaryinteraction

(7)

param-Table2

Experimentalsolubilitylogarithmrankingat30◦Candordersofmagnitude(insolubilitylogarithm)foribuprofen,paracetamol,salicylicacidandbenzoicacidinvarious

organicsolvents.

Ibuprofen Paracetamol

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−3;−2] Dichloromethane [−9;−8]

Cyclohexane [−2;−1.5] Ethylacetate [−6;−5]

Ethanol [−1.5;−1] Acetonitrile [−5;−4]

Toluene Dioxane

Ethylacetate Chloroform

Isopropanol Heptanol [−4;−3]

Acetone Methylethylketone

Octanol [−1;−0.5] Acetone Chloroform Butanol Propanol [−3;−2] Ethanol Methanol Dimethylsulfoxide [−1]

Benzoicacid Salicylicacid

Solvent Orderofmagnitude Solvent Orderofmagnitude

Hexane [−5;−4] Cyclohexane [−8;−7]

Cyclohexane Hexane

Heptane Chloroform [−7;−6]

Acetonitrile [−3;−2] Carbontetrachloride [−6;−5]

Carontetrachloride Xylene

Benzene Acetonitrile [−4;−3]

Heptanol [−2;−1.5] Aceticacid [−3;−2]

Acetone Methanol [−2;−1.5]

Octanol Ethylacetate

Isopropanol Ethanol

Butanol Octanol

N-Methylpyrrolidone [−1.5;−1] Acetone

Dioxane Ethylmethylketone

Table3

PredictionerrorsofUNIFAC,COSMO-SACandNRTL-SACforibuprofen,paracetamol,benzoicacidandsalicylicacidusingEqs.(4)and5,usingsolubilitydatainTable2..

Compound Model MSerrorEq.(5) Meanstandarddeviation MSerrorEq.(4)

Ibuprofen UNIFAC 0.07 11% 0.104

UNIFACmod. 0.137 15% 0.079

COSMO-SAC 1.150 8% 1.307

NRTL-SAClit. 0.201 12% 0.276

NRTL-SACthiswork1 0.202 10% –

NRTL-SACthiswork2 0.165 13% –

Paracetamol UNIFAC 2.306 31% 58.67

UNIFACmod. 1.165 30% 3.780(0.754b)

COSMO-SAC 25.187 19% 309(166b)

NRTL-SAClit. 0.367 26% 0.509

NRTL-SACthiswork1 76(0.9b)

NRTL-SACthiswork2 0.805(0.250b) 27%

-NRTL-SACthiswork3 9.034(2.750b)

Salicylicacid UNIFAC 26.62(0.195a) 29%

UNIFACmod. 322(3.01a) 36%

COSMO-SAC 15.03(5.01a) 66%

-NRTL-SAClit. 125(0.451a) 44%

NRTL-SACthiswork 361(0.436a) 49%

NRTL-SACthiswork 409(0.590a) 63%

Benzoicacid UNIFAC 0.180 9% 1.077

UNIFACmod. 0.439 8% 0.572

COSMO-SAC 1.136 33% 1.932

NRTL-SAClit. 0.107 9% –

NRTL-SACthiswork 0.114 9% –

4-Aminobenzoic UNIFAC 6.790 – –

acid UNIFACmod. 18.04 – –

COSMO-SAC 8.50 – –

NRTL-SACthiswork 0.672 – –

Anthracene UNIFAC 0.247 11% 0.506

UNIFACmod. 0.808 12% 1.388

COSMO-SAC 3.203 19% 3.80

NRTL-SACthiswork 3.631 19% –

amsewithoutchloroform. bmsewithoutdichloromethane.

(8)

Fig.3.Experimentalsolubilityof(a)ibuprofen[20,21],(b)paracetamol[22],(c)salicylicacid[23–27],(d)benzoicacid[28–30],(e)4-aminobenzoicacid[31–34]and(f) anthracene[35–39]invarioussolventsasafunctionofthereversetemperature(van’tHoffplot).

etersaretemperaturedependent(9mn),themoleculargeometry

andisomerismarenottakenintoaccount.Asmentionedinthe literature[5,8], thepredictedsolubilityof organicmolecules in nonpolarsolvents(heptane,cyclohexane,andtoluene)areingood agreementwiththeexperimentaldata.Inthecase ofpolar sol-vents,capableofforminghydrogenbonds,wefoundthesolubility to bealways underestimated for the chosen solvents. In

addi-tion,inthecaseofsolubilityinalcohols,theordersofmagnitude ofthepredictedsolubilitiesareveryclose.Thisprobablymeans thatinUNIFAC,thepair-wiseinteractioncoefficientsinvolvingthe OHgroupispredominantovertheinteractioncoefficientsofthe aliphaticchains(–CH2–CH3,-CH3groups).Thesolubilityordersof

magnitude,shown inTable4,seemstoconfirmthis statement. Fromaqualitativepointofview,theobtainedsolubilityranking

(9)

Fig.4.UNIFACpredictionsof(a)ibuprofen,(b)paracetamol,(c)salicylicacid,(d)benzoicacid,(e)4-aminobenzoicacidand(f)anthracenewithEq.(5)asafunctionofthe experimentalsolubility.

ofthestudiedmoleculesaregiveninTable4.Thistableshowsthat forlargesolubilityvalues(x>0.05),UNIFACgivessatisfactory qual-itativeresultsforpredictingsolubilityorderofmagnitudeandthe solubilityrankingisalmostpreserved.Forparacetamol,original UNIFAChasdifficultiestopredictsolubilitylowerthan0.05.The predictedordersofmagnitudearenotinagreementwith experi-mentaldata.

4.3. COSMO-SAC

Fromtheresults,presentedinFig.5andTable5,itcanbestated thatCOSMO-SACgivespoorresultsinthesolubilitypredictionof thedrugstested.Themeansquareerrorsrangefrom1.15to25, dependingonthemoleculeandonthepurecomponent proper-tieschosen.Notsurprisingly,therelativedeviationisparticularly

(10)

Table4

OriginalUNIFACsolubilityrankingatCor30◦Candordersofmagnitude(insolubilitylogarithm)foribuprofen,paracetamol,salicylicacidandbenzoicacidinvarioussolvents.

Ibuprofen Paracetamol

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−3;−2] Dichloromethane [−9;−8] Cyclohexane [−2;−1.5] Dioxane [−5] Ethanol Heptanol Isopropanol [−1.5;−1] Acetonitrile [−5;−4] Octanol DMSO Toluene Methanol

Ethylacetate Butanol

Acetone Ethylacetate

Chloroform Propanol

Methylethylketone Ethanol

Acetone [−4;−3]

Chloroform [−3;−2]

Benzoicacid Salicylicacid

Solvent Orderofmagnitude Solvent Orderofmagnitude

Hexane [−5;−4] Hexane [−9;−8]

Heptane Cyclohexane

Cyclohexane Carbontetrachloride

Carbontetrachloride [−4;−3] Xylene [−7;−6]

Benzene [−3;−2] Chloroform [−3]

Octanol Aceticacid

Isopropanol Acetonitrile

Acetonitrile Octanol [−3;−2]

Butanol Ethylacetate

Heptanol Ethylmethylketone

Dioxane [−2;−1] Acetone [−2;−1.5]

Acetone Ethanol

N-Methylpyrrolidone Methanol

Table5

COSMO-SACsolubilityrankingat30◦Candordersofmagnitude(insolubilitylogarithm)foribuprofen,paracetamol,salicylicacidandbenzoicacidinvarioussolvents.

Ibuprofen Paracetamol

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−3;−2.5] Dichloromethane [−6;−5]

Cyclohexane [−2] Chloroform [−5;−4]

Toluene [−1.5;−1] Acetonitrile [−3;−2.5]

Chloroform Ethylacetate

Ethylacetate [−1;−0.5] Heptanol

Octanol Butanol [−2.5;−2]

Ethanol Methanol

Isopropanol Propanol

Acetone Dioxane [−2;−1.5]

Ethanol

Methylethylmketone Acetone

Dimethylsulfoxide [>−1]

Benzoicacid Salicylicacid

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−5.5;−5] Hexane [−6;−5]

Cyclohexane Cyclohexane

Hexane Carbontetrachloride

Isopropanol [−5;−4] Xylene [−4]

Carbontetrachloride [−4;−3] Chloroform [−4;−3]

Benzene [−3;−2] Aceticacid [−3;−2]

Acetonitrile [−2;−1] Acetonitrile [−2;−1.5]

Octanol [−1;−0.5] Ethylacetate [−1.5;−1]

Butanol Methanol

Dioxane Octanol

Acetone Ethanol [>−1]

N-Methylpyrrolidone [>−0.5] Acetone

(11)

Fig.5.COSMO-SACpredictionof(a)ibuprofen,(b)paracetamol,(c)salicylicacid,(d)benzoicacid,(e)4-aminobenzoicacidand(f)anthracenewithEq.(5)asafunctionof experimentalsolubility.

highinsolventsinwhichsolubilityisverylow,e.g.paracetamolin ethylacetate.AsinthecaseofUNIFAC,themodelgives satisfac-toryresultswithaproticandapolarsolvents(hexane,heptane,and chloroform).Ontheopposite,inpolarsolventscapableofforming hydrogenbonds(alcohols,ketones,esters,...)withthesolute,the predictedsolubilityissystematicallyoverestimated.Inaddition,it canbenoticedthatthepredictedsolubilityvaluesofamolecule

inalcoholsarealwaysnearlythesame,whateverthesizeofthe alcoholmolecule(seeordersofmagnitudesinTable5).

AspreviouslyachievedfortheUNIFACmodel,thepredictions obtainedatthelowertemperatures (20–25◦C) aremuch better

thantheoneobtainedathighertemperatures(Fig.5).Inthismodel, theelectrostaticandhydrogenbondsparametersarenot temper-aturedependent.

(12)

Fromaqualitativepointofview(Table5andFig.5), COSMO-SACisquitereliableforthesolubilitypredictionofthemostsimple molecules,likealkanes.Thismodeloverestimatessolubilityin alco-holsandketonesand leadstoanonreliable solubilityscale. In addition,theordersofmagnitudearetwicetheexperimentalin mostcases.Consideringsolubilityranking,thismodelisnotasgood asUNIFAC.Toevaluatetheinfluenceofhydrogenbondonthe pre-dictingcapabilitiesofCOSMO-SAC,thehydrogenbondisexcluded fromthemodel,setting theconstantfor thehydrogenbonding interaction(chb)tozero.Thisconstantappearsinthecalculation ofthesegmentactivityŴ,in theexpressionof theelectrostatic interactions.

Theresults areshown in Fig. 6 and it canbe seenthat the predictedsolubilitiesare underestimatedin polarsolvents. The modelpredictionsarein goodagreementwiththesolubilityof polarmoleculesinpolarsolvents,whichishigherthanideal,mainly resultsfromspecificinteractionslikehydrogenbonds.Inaddition, theresultsobtainedwithandwithouthydrogenbondsalsosuggest thatinCOSMO-SACthehydrogenbondcontributionis overesti-mated.Thechbconstantisanadjustableparameterwhichhasbeen

optimizeduponmanyVLEdataofsmallorganicmolecules. How-ever,theorganicmoleculesinourstudyaremuchlarger,flexible, andcontainh-bondingsurfacesthatareburriedorhinderedsothat h-bondscannotform.Then,itislikelythatsuchsurfacesshallnotbe includedintheh-bondterminthemodel.Itwouldbevery interest-ingtore-optimizethistermforlargeorganicmoleculesusingSLE data(providingthatenoughexperimentaldataareavailable).Even ifEq.(4)isconsidered,thisdoesnotimprovetheresultsasthe1Cp

termincreasesthepredictedsolubility(seeSection4.5).Inaddition, evenifstereoisomerismcanbetakenintoaccountinCOSMO-SAC (eachisomerhasitsownsigmaprofile),molecularselfandcross associationsarenotconsidered.It iswellknownthatibuprofen formshydrogen-bondeddimerinsolution:twocomposedofthe sameenantiomersofR–RandS–SandtheracemateofR–S.The self-associationofdrugmoleculesmaydecreasethenumberoffree molecularsitesavailabletoformhydrogenbondswiththesolvent, whichmayexplaintheoverestimationofthecontributionof hydro-genbondstothecalculationoftheactivitycoefficient.Sucheffects arealsofoundforbenzoicacid(Self-associationandhydrationof

Fig.6.COSMO-SACpredictionofparacetamolinethanolandmethanolwith hydro-genbonds(straightlines),andnohydrogenbonds(dashedlines)asafunctionof thetemperatureusingEq.(5).

benzoicacidinbenzene[54]orinalessextentforcarboxylicacid likeaspirin).

At last, even if the predicted solubilityis quantitatively far fromtheexperimentaldata,Fig.6showsthat themodel nicely predictsthesolubilitytemperaturedependence.Fordesigning a crystallizationprocess,theoneofthemostimportantparameter isthesupersaturation:thedrivingforceoftheprocess,definedas theratiobetweentheinitialandfinalconcentration(S=C/C*).For instance,asshowninFig.6,theexperimentalsupersaturationfor paracetamolinmethanolfrom30◦Cto0CisS=1.54andthe

pre-dictedisS=1.5(similarresultsareobtainedforallthemolecules testedinthepaperinsmallalcoholmolecules).

Table6

NRTL-SACsegmentswhen1Cpisneglected(Eq.(5)).

Compounds X Y− Y+ Z mse Numberofsolvents

Ibuprofen[55] 1.038 0.051 0.028 0.318 1.055 19

Ibuprofen(thiswork1)a 0.507 0.297 0.350 0 0.005 6

Ibuprofen(thiswork2)b 0.484 0 0.267 0.210 0.0188 9

Paracetamol[56] 0.498 0.487 0.162 1.270 – 8

Paracetamol[7] 0.416 0.016 0.168 1.86 – 5

Paracetamol(thiswork1)c 0.265 0.576 0.668 0.717 0.029 6

Paracetamol(thiswork2)d 0.369 0.618 0.521 1.095 0.859 14

Paracetamol(thiswork3)e 0.275 1.036 1.159 0.777 0.033 6

Salicylicacid[55] 0.726 0.176 0 0.749 0.774 18

Salicylicacid(thiswork1)f 0.911 1.762 0.129 0.329 0.182 6

Salicylicacid(thiswork2)g 1.006 2.29 0 0.451 0.0617 6

Benzoicacid[17] 0.524 0.089 0.45 0.405 0.160 7

Benzoicacid[55] 0.494 0 0.336 0.468 0.292 26

Benzoicacid(thiswork)h 0.576 0 0.121 0.543 0.144 6

4-aminobenzoicacidi 0 0 4.229 2.321 0.222 5

Anthracenej 1.486 0 0 0.564 0.180 5

aInacetone,cyclohexane,chloroform,ethanol,octanolandethylacetateat30

C.

bInallthesolventconsideredinthisstudy(seeFig.3)at30C.

c Inacetone,ethylacetate,chloroform,toluene,ethanolandwaterat30C. d Inallthesolventsofthiswork(seeFig.3)at30C.

eInethanol,propanol,acetone,methylethylketone,waterandtolueneat30C. f Inethanol,ethylacetate,cyclohexane,acetone,acetonitrileandxyleneat30C. gInethanol,ethylacetate,acetone,xylene,acetonitrileandwaterat30C.

h Incyclohexane,acetonitrile,1-butanol,benzene,n-methylpyrrolidoneandwater,at30

C.

i Inethylacetate,ethanol,dimethoxyethane,THFanddioxane,at30C.

(13)

4.4. NRTL-SAC

TotestthepredictionabilitiesofNRTL-SAC,thesegments val-ues(X,Y−,Y+ andZ)weretakenfromtheliterature[7,55,56].It

canbenoticedthatsegmentvaluesfoundintheliteraturecan sig-nificantlydifferforthesamecompound.Forinstance,asshownin Table6,ChenandCrafts[56]andMotaetal.[7]havepublished differentsegmentvaluesforparacetamol.Thesedifferencescanbe notonlyduetothevaluesusedforthepurecomponent parame-ters(1Hm,Tm,and1Cp)determinedexperimentallyorregressed

simultaneouslywiththesegments,butalsoduetothenatureand thenumberofsolventsusedintheregression.

Inordertotestthemodelsensitivitytothequadruplet[XY−

Y+Z],calculationswereperformedwithdifferentsolventsforall

thesolutesinordertoobtainnewvalues.Allthesenewvaluesand thecorrespondingmeansquareerrorsaregivenin Table6.The sensitivityofthemodeltotheexperimentalsolidstateproperties isperformedusingtheMonteCarlomethod.

Fromaquantitativepointofview,themodelgivesresultswith a mserangingfrom0.107to409or3.63 when neither chloro-formnordichloromethaneareconsidered(seeTable3).NRTL-SAC predictionsusing[55]segmentvaluesversusexperimental solu-bilityisrepresentedinFig.7,andusingsegmentfromthisworkin Fig.8(quadruplet1foribuprofen,quadruplet2forparacetamoland

quadruplet2forsalicylicacid).Duetotheempiricalnatureofthe model,NRTL-SACseemstobelessaffectedbyerrorsonthe ther-modynamicpropertiesthantheothermodels.Theresultsobtained areingoodagreementwiththeexperimentaldata,andmostofthe resultsarebeneathamseof4.However,themserrorsseemquite hightodoquantitativepredictions.

The model accuracy of the resultsstrongly depends onthe quadruplets[XY−Y+Z]usedasparameters.Infact,thechoiceof

theseparametersisacriticalsteptousethismodel.

Thesegmentsvaluescanbeunderstoodasaweightgiventoeach moleculebehaviour:hydrophilic,hydrophobic,polarattractiveand repulsive.Theyrepresenta“meanbehaviour”ofthemoleculein solutioncomputedusingexperimentaldata.Thesevaluesdepend onthequalityandquantityofthesolubilitydatausedtoregress thefourparameters.Theywillnotrepresenttherealbehaviourof theAPIinallthesolventsiftheexperimentaldataarenottaken judiciously.However,thechoiceoftheexperimentaldataisnot obvious.NRTL-SAC using segmentscalculated in somealcohols willnotnecessarilygivegoodpredictionsinotheralcoholsandit givesevenworseresultsinothersolventlikeketonesoralkanes. Sometestsperformedonparacetamolshowedthatthepredictions obtainedusingtheparametersregressedusingfoursolubilitydata inpropanolandbutanol(twotemperatures)gavethesameresults asthoseobtainedwithquadruplet3ofthiswork.Infact,allalcohols

Fig.7.NRTL-SACpredictionof(a)ibuprofen,(b)paracetamol,(c)salicylicacid,(d)benzoicacid,(e)4-aminobenzoicacidand(f)anthracenewithEq.(5)usingsegmentsfrom theliteratureasafunctionofexperimentalsolubility.

(14)

Fig.8.NRTL-SACpredictionof(a)ibuprofen,(b)paracetamol,(c)salicylicacid,(d)benzoicacid,(e)4-aminobenzoicacidand(f)anthracenewithEq.(5)usingsegmentsfrom thiswork(thiswork1foribuprofen,2forparacetamol,1forsalicylicacidand1forbenzoicacid)asafunctionofexperimentalsolubility.

donothavethesameweightoneachsegmentnumber.Eveniftheir quadrupletsmayindicateaclosebehaviourtendency,theycanalso bequitedifferent.Inaddition,theuseofalargeexperimentaldata setincreasesthemodelperformanceandreliability.

Inordertoshowtheinfluenceofthesegmentvaluesusedinthis study,fourtypesofibuprofensolubilitypredictionsinethanolare reportedinFig.9.

Toendwiththesolventselectionandinordertogetthemost reliableregression, theidea is toselectthesolventso thatthe weightofeachsegmentisthesame:

X

i Xi=

X

i Y+ i =

X

i Y− i =

X

i Zi (17)

(15)

Fig.9.NRTL-SACpredictionofparacetamolinethanolusingfourdifferentsegment quadrupletsasafunctionofthetemperature.

withithesubscriptforthesolvent.Inthisway,allthebehaviourof thesolutemoleculeswillbeinvestigatedwiththesameweight.

Thismethodisstillinstudy,butwetriedtofindthebest sol-ventsfortheparacetamolregression.Todothat,wetakeallthe solventsinvestigatedby[17]andweforcedthissumtobegreater than1(tohaveeachsegmentnumberrepresentedwithenough weight).Thecalculation,madewithversion23.6.5ofGAMS soft-ware,gaveatotalof10solvents(chloroform,1,2-dichloroethane, 1,4-dioxane, ethanol, formamide, formic acid, isobutylalcohol, nitromethane,1-propanolandwater)toreachamaximum devi-ation of 0.002. If we regressed the segment numbers of these solvents, themeansquare errorwould behigherthan themse obtainedinTable6becauseofthediversityofthesolventused. But,theobtainedquadrupletshouldbemoresuitableasthemean behaviourshouldbebetterrepresented.Furtherstudiesarestill inprogress.

Evenifthepredictiondiffersquantitativelyfromaquadruplet totheother,NRTL-SACgivesgoodordersofmagnitudewhatever thesegmentsused(seeTables7and8).Moreoverthesolubility rankingispreserved.

Asshown in Figs. 7 and8 in thecase of ibuprofen, solubil-itypredictionsfortemperaturesabove25◦Carelessaccurate.In

thismodel,theinfluenceoftemperatureonsolubilityis underes-timated,especiallywhenhydrogenbondsareinvolved.Thiscould beexplainedbythefactthatthesolubilitydependenceon tem-peraturebeingrelatedtothesolidstateproperties,andnotonthe possibleevolutionofmolecularinteractions(hydrogenbond,Van derWaals,dispersionforces,...)betweenmoleculesinsolution.

ToconcludeonNRTL-SAC,thismodelcanbeusefultogivean estimationofthesolubilityofAPIsinvarioussolvents.Butitneeds reliablesegmentvaluesforquantitativepurposeandneeds exper-imentaldata.

4.5. Therightmeltingenthalpyortemperatureandtheuseof 1Cp

In the case of SLE, equilibrium equation can be Eq. (4) or Eq.(5).Eq.(4)isthemostrigorous,but,usually, the1Cp term

Fig.10.UNIFACpredictionofparacetamolusingdifferentthermodynamic prop-erties[23,34,35,37]inethanol(upperside)andethylacetate(downside)with (a)1Hm=27,100J/mol,Tm=443.6K,(b)1Hm=27,100J/mol,Tm=443.15K,(c)

1Hm=28,100J/mol,Tm=441.15K,(d)sameas’b’with1Cp=99.8J/molK,(e)same

as’c’with1Cp=99.8J/molK,(f)sameas’a’with1Cp=75J/molK.

is neglected and Eq. (5) is used. Evenif Gracin et al. [5] have found a small influence of the 1Cp term with UNIFAC model,

the two equations willbe compared usingall the models pre-sented previously.In addition, thesignificance of thevalues of thepurecompound propertieshastobeunderlined.Measuring accuratelythemeltingtemperatureandenthalpymaybedifficult, sinceitreliesonproductcrystallinity,purityandonthe analyt-icalmethodused.AspresentedinTable1manydifferentvalues havebeenreportedintheliterature.InFigs.4,5,7and8, stan-darddeviations ofthemodelscausedbypropertieserrors have been presented. So, when a model is used for predicting solu-bility,itserrorhastobetakenintoaccount,and themeasured Tm,1Hm and1Cp havetobeasaccurateaspossible.

Neverthe-less,prediction errorsinducedbythemodelsareusually larger then the ones induced by the solid state property uncertain-ties.

As shown in Fig. 10, it can be seen that a higher melting enthalpyimpliesahigherpredictedsolubility.Thesamebehaviour isobservedforhigher1Cp.Moreover,inthecaseofpolymorphism

(likeparacetamol),theuseofthewrongmeltingenthalpy, tem-peratureor1Cpcancausebadpredictionsormisinterpretations (paracetamolpresentsatleasttwopolymorphswithclosemelting temperatureandenthalpy).

(16)

Table7

NRTL-SACsolubilityrankingat30◦Candordersofmagnitude(insolubilitylogarithm)foribuprofen,paracetamol,salicylicacidandbenzoicacidinvarioussolventsusing

segmentsfromtheliterature.

Ibuprofen Paracetamol

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−2] Dichloromethane [−8;−7]

Acetone [−2;−1.5] Chloroform

Cyclohexane Ethylacetate [−6;−5]

Toluene Methylethylketone

Octanol Acetonitrile

Ethanol Acetone [−5;−4]

Ethylacetate Dioxane

Isopropanol [−1.5;−1] Propanol [−3;−2.5]

Chloroform [3×10−1to4×10−1] Methanol

Butanol Ethanol

Dimethylsulfoxide [−2.5;−2]

Benzoicacid Salicylicacid

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−5;−4] Hexane [−8;−7]

Hexane Cyclohexane

Cyclohexane Carbontetrachloride [−7;−6]

Carbontetrachloride −3 Xylene [−6;−5]

Benzene [−2.5;−2] Acetonitrile [−4;−3]

Acetonitrile Ethylmethylketone

Butanol [−2;−1.5] Ethylacetate

Octanol Acetone [−4;−3]

Isopropanol Octanol

Acetone Methanol

Dioxane [−1.5;−1] Aceticacid [−3;−2]

N-Methylpyrrolidone Ethanol

Dimethylsulfoxide [>−1] Chloroform

Table8

NRTL-SACsolubilityrankingat30◦Candordersofmagnitude(insolubilitylogarithm)foribuprofen,paracetamol,salicylicacidandbenzoicacidinvarioussolventsusing

segmentscalculatedinthisstudy(thiswork1foribuprofen,2forparacetamol,1forsalicylicacidand1forbenzoicacid).

Ibuprofen(thiswork2) Paracetamol(thiswork2)

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane −2 Dichloromethane [−7;−6]

Cyclohexane Ethylacetate [−6;−5]

Ethanol [−1.5;−1] Ethylmethylketone

Toluene Acetonitrile [−5;−4]

Acetone Chloroform

Isopropanol Acetone [−4;−3]

Ethylacetate Dioxane

Chloroform Propanol −3

Octanol Butanol

Ethanol [−3;−2.5]

Methanol

DMSO [−2.5;−2]

Benzoicacid(thiswork2) Salicylicacid

Solvent Orderofmagnitude Solvent Orderofmagnitude

Heptane [−4.5;−4] Hexane [−9;−8]

Cyclohexane Cyclohexane [−8;−7]

Hexane Carbontetrachloride [−6;−5]

Carbontetrachloride [−4;−3] Xylene [−5;−4]

Benzene [−3;−2] Acetonitrile [−4;−3]

Acetonitrile Ethylacetate [−3;−2]

Acetone [−2;−1.5] Acetone

Octanol Ethylmethylketone

Dioxane Methanol [−2;−1.5]

Butanol Octanol

Isopropanol Aceticacid

N-methylpyrrolidone [−1.5;−1] Ethanol

(17)

Fig.11. CoefficientBasafunctionofTm/T.

Toevaluatetheimpactofneglectingthe1CptermintheSLEon

solubilityprediction,relativevalueofthetwoparts,constituting Eq.(4),canbecomparedusingthefollowingequation:

1hm(Tm) T1m− 1 T



1Cp(Tm) ln(TTm)−TTm+1



(18)

Then,wewillrewrite: 1Hm 1CpTm × 1− Tm T



ln(Tm T )− Tm T +1



=A×B (19)

If|AB|≪1,thenthesecondpartofEq.(4)cannotbeneglected, andiftheratioishighenough,itmaybeneglected(seeexample inFig.11).TheAtermisconstantandonlydependsonthe ther-modynamicpropertiesofthesolidstate.TheBtermisafunctionof temperatureasshowninFig.11.

Considering that for a standard crystallization operation, the runningtemperature is around 290K. At this temperature, B=−11.33andA=1.48foribuprofen,and B=−5.10,A=0.71 for paracetamol.Thedefinitionofarelativeerrorenablestoestimate theeffectbyneglectingthe1Cpterm:

relativeerror(%) = 1

AB−1 (20)

Inthecaseofthesetwomolecules,thiserrorisabout6%forthe ibuprofenand22%fortheparacetamol.Wecanconcludethatthe 1Cptermisnegligibleinthecaseofibuprofen,andnotinthecase

ofparacetamol. 5. Conclusion

Toconclude,noneofthemodelspresentedinthispaperisableto predictpreciselythesolubilityofthestudiedmolecules(containing variouscommonfunctionalgroupslikealcohols,ketones,amines) invarioussolvents.Evenconsideringthemodelerrorsmostofthe timestakenbytheimprecisiononthepuresolidproperties,the predictedsolubilityaregenerallyinagreementwithexperimental data(IbuprofenwithOriginalUNIFAC,somesolutes with NRTL-SAC).COSMO-SACpredictionsusuallyproducehighervaluesthan theexperimental,especiallywhenhydrogenbondsmayoccur.

UNI-FACismoreaccuratethanCOSMO-SACfortheAPIinthisstudy andgivesgoodresultsforsimplemolecules(anthracene). How-ever,evenifthesolubilityordersofmagnitudearepreserved,it isnotgoodenoughforperformingquantitativepredictions.Inthe caseofNRTL-SAC,predictionsareverydependentonthe parame-tersused(segmentvalues).Iftheseparametersdonotrepresentthe correctbehaviouroftheAPIinthesolventchosen,theresultswill notbesatisfying.Finally,UNIFACandNRTL-SAC,whichpreservethe solubilityordersofmagnitudecanbeagoodcomplementof exper-imentaltechniquestoguidethechoiceofacrystallizationsolvent. Butintheend,experimentsarestillrequiredtoobtainquantitative results.

Inaddition,allthemodelstestedfailtocorrectlydescribethe solubilitydependencewithtemperature.Theuseofanequationof stateseemstobemoreappropriateforpredictingtheevolutionof solubilitywithtemperature[57,58].

Resultspresentedinthispaperalsoshowthattheuseofthe mostpreciseequilibriumequationdoesnotsignificantlyinfluence thequalityoftheresults.Asthemodelsarenotaccurateenough, theuseofthemostaccuratethermodynamicpropertiesvaluesis notnecessary.Moreover,the1Cptermcanbeignoredaslongas

thegeneratederrorisnegligibleregardingthemodelerror.But,in thehypotheticalcaseofaperfectmodel,themostprecisevalues shouldbeused.

Acknowledgements

Theauthorswant toacknowledge ProSimCompanyfor their collaborationintheuseofSimulisThermodynamicssoftware,and Pr.FlorentBourgeoisforenlighteningdiscussions.

References

[1]S.Morissete,O.Almarsson,M.Peterson,J.Remenar,M.Read,A.Lemmo,S. Ellis,M.Cima,C.Gardner,High-throughputcrystallization:polymorphs,salts, co-crystalsandsolvatesofpharmaceuticalsolids,AdvancedDrugDelivery Reviews56(2003)275–300.

[2]P.Laval,N.Lisai,J.-B.Salmon,M.Joanicot,Amicrofluidicdevicebasedondroplet storageforscreeningsolubilitydiagrams,LabonaChip7(2007)829–834. [3]A.A.Noyes,W.R.Whitney,PhysicalChemistry23(1909)689.

[4]J.G.Gmehling,T.F.Anderson,J.M.Prausnitz,Solid–liquidequilibriausing UNI-FAC,Industrial&EngineeringChemistryFundamentals17(1978)269–273. [5]S.Gracin,T.Brinck,A.Rasmuson,Predictionofsolubilityofsolidorganic

com-poundsinsolventsbyUNIFAC,Industrial&EngineeringChemistryResearch 41(2002)5114–5124.

[6]E.Mullins,Y.Liu,A.Ghaderi,S.D.Fast,Sigmaprofiledatabaseforpredicting solidsolubilityinpureandmixedsolventmixturesfororganic pharmacolog-icalcompoundswithCOSMO-basedthermodynamicmethods,Industrial& EngineeringChemistryResearch47(2008)1707–1725.

[7]F.L.Mota,A.P.Carneiro,S.P.Pinho,E.A.Macedo,Temperatureandsolventeffect insolubilityofsomepharmaceuticalcompounds:measurementsand mod-elling,EuropeanJournalofPharmaceuticalScience37(2009)499–507. [8]I.Hahnenkamp,G.Graubner,J.Gmehling,Measurementandpredictionof

solubilitiesofactivepharmaceuticalingredients,InternationalJournalof Phar-maceutics388(2010)73–81.

[9]J.Mullin,in:Crystallization—3Rev,Oxford,1993.

[10]M.Brown,B.Glass,Decompositionofsolidsaccompaniedbymelting-Bawn kinetics,InternationalJournalofPharmaceutics254(2003)255–261. [11]A.Fredenslund,R.Jones,J.Prausnitz,Group-contributionestimationof

activity-coefficientsinnonidealliquid-mixtures,AIChEJournal21(1975)1086–1099. [12]U.Weidlich,J.Gmehling,AmodifiedUNIFACmodel.1.PredictionofVLE,HE, andgamma-infinity,Industrial&EngineeringChemistryResearch26(1987) 1372–1381.

[13]A.Klamt,G.Schuurmann,COSMO—anewapproachtodielectricscreeningin solventswithexplicitexpressionsforthescreeningenergyanditsgradient, JournaloftheChemicalSociety-PerkinTransactions2(1993)799–805. [14]A.Klamt,V.Jonas,T.Burger,J.Lohrenz,Refinementandparametrizationof

COSMO-RS,JournalofPhysicalChemistryA102(1998)5074–5085. [15]S.Lin,S.Sandler,Aprioriphaseequilibriumpredictionfromasegment

con-tributionsolvationmodel,Industrial&EngineeringChemistryResearch41 (2002)899–913.

[16] E.Mullins,R.Oldland,Y.Liu,S.Wang,S.I.Sandler,C.-C.Chen,M.Zwolak,K.C. Seavey,Sigma-profiledatabaseforusingCOSMO-basedthermodynamic meth-ods,Industrial&EngineeringChemistryResearch45(2006)4389–4415.

(18)

[17]C.-C.Chen,Y.Song,Solubilitymodelingwithanonrandomtwo-liquidsegment activitycoefficientmodel,Industrial&EngineeringChemistryResearch43 (2004)8354–8362.

[18]H.Renon,J.Prausnitz,Localcompositioninthermodynamicexcessfunctions forliquidmixtures,AIChEJournal14(1968)135–144.

[19]C.-C.Chen,Asegment-basedlocalcompositionmodelfortheGibbsenergyof polymersolutions,FluidPhaseEquilibria83(1993)301.

[20] L.C.Garzon,F.Martinez,Temperaturedependenceofsolubilityforibuprofenin someorganicandaqueoussolvents,JournalofSolutionChemistry33(2004) 1379–1395.

[21]S.Gracin,A.Rasmuson,Solubilityofphenylaceticacid,p-hydroxyphenylacetic acid,p-aminophenylaceticacid,p-hydroxybenzoicacid,andibuprofeninpure solvents,JournalofChemicalandEngineeringData47(2002)1379–1383. [22] R.Granberg,A.Rasmuson,Solubilityofparacetamolinpuresolvents,Journal

ofChemicalandEngineeringData44(1999)1391–1395.

[23] K.DeFina,T.Sharp,L.Roy,W.Acree,Solubilityof8-hydroxybenzoicacidin selectorganicsolventsat298.15K,JournalofChemicalandEngineeringData 44(1999)1262–1264.

[24]H.Fung,T.Higuchi,Molecularinteractionsandsolubilityofpolar nonelec-trolytesinnonpolarsolvents,JournalofPharmaceuticalSciences60(1971) 1782–1788.

[25]F.L.Nordstrom,A.C.Rasmuson,Solubilityandmeltingpropertiesofsalicylic acid,JournalofChemicalandEngineeringData51(2006)1668–1671. [26]A.Paruta,B.Sciarrone,N.Lordi,Solubilityofsalicylicacidasafunctionof

dielectricconstant,JournalofPharmaceuticalSciences53(1964)1349–1353. [27]A.Shalmashi,A.Eliassi,Solubilityofsalicylicacidinwater,ethanol,carbon tetrachloride,ethylacetate,andxylene,JournalofChemicalandEngineering Data53(2008)199–200.

[28]W.Acree,G.Bertrand,Thermochemicalinvestigationsofnearlyidealbinary solvents.7.Monomeranddimermodelsforsolubilityofbenzoic-acidin sim-plebinaryandternarysolvents,JournalofPharmaceuticalSciences70(1981) 1033–1036.

[29]A.Beerbower,P.Wu,A.Martin,Expandedsolubilityparameterapproach.1. Naphthaleneandbenzoic-acidinindividualsolvents,Journalof Pharmaceuti-calSciences73(1984)179–188.

[30] D.Li,D.Liu,F.Wang,Solubilityof4-methylbenzoicacidbetween288Kand 370K,JournalofChemicalandEngineeringData46(2001)234–236. [31]H.Lin,R.Nash,AnexperimentalmethodfordeterminingtheHildebrand

solu-bilityparameteroforganicnonelectrolytes,JournalofPharmaceuticalSciences 82(1993)1018–1025.

[32]S.Yalkowski,S.Valvani,T.Roseman,Solubilityandpartitioningvi:octanol solubilityandoctanol–waterpartitioncoefficients,JournalofPharmaceutical Science72(1983)866–870.

[33] C.Hancock,J.Pawlowski,J.Idoux,JournalofOrganicChemistry31(1966)3801. [34] C.Daniels,A.Charlton,R.Wold,R.Moreno,W.Acree,M.Abraham, Mathemat-icalcorrelationof4-aminobenzoicacidsolubilitiesinorganicsolventswith theAbrahamsolvationparametermodel,PhysicsandChemistryofLiquids42 (2004)633–641.

[35] E.Cepeda,B.Gomez,Solubilityofanthraceneandanthraquinoneinsome pureandmixedsolvents,JournalofChemicalEngineeringData34(1989) 273–275.

[36] E.Cepeda,M.Diaz,Solubilityofanthraceneandanthraquinonein acetoni-trile,methylethylketone,isopropylalcoholandtheirmixtures,FluidPhase Equilibria121(1996)267–272.

[37]G.Al-Sharrah,S.Ali,M.Fahim,Solubilityofanthraceneintwobinarysolvents containingtoluene,FluidPhaseEquilibria193(2002)191–201.

[38]J.Powell,W.Acree,Solubilityofanthraceneinbinaryalcohol+dibutylether solventmixtures,JournalofChemicalEngineeringData40(1995)914–916.

[39]A.Zvaigzne,W.Acree,Thermochemicalinvestigationsofmolecular complex-ation.Estimationofanthracene–ethylacetateandanthracene–diethyladipate associationparametersfrommeasuredsolubilitydata,Physics&Chemistryof Liquids24(1991)31–42.

[40]D.Andrews,G.Lynn,J.Johnston,Theheatcapacitiesandheatofcrystallization ofsomeisomericaromaticcompounds,JournalofAmericanChemicalSociety 48(1926)1274–1287.

[41] D.David,Determinationofspecificheatandheatoffusionbydifferential ther-malanalysis.studyoftheoryandoperatingparameters,AnalysisChemistry36 (1964)2162–2166.

[42]D.Ginnings,G.Furukawa,Heatcapacitystandardsfortherange14to1200, JournalofAmericanChemicalSociety75(1953)522–527.

[43]D. Grant, M.Mehdizadeh, A. Chow,J. Fairbrother, Nonlinear van’tHoff solubility–temperatureplotsandtheirpharmaceuticalinterpretation, Inter-nationalJournalofPharmaceutics18(1984)25–38.

[44] D.Li,J.Liu,D.Liu,F.Wang,Solubilitiesofterephthalaldehydic,p-toluic,benzoic, terephthalicandisophthalicacidsinn,n-dimethylformamidefrom294.75to 370.45K,FluidPhaseEquilibria200(2002)69–74.

[45]R.Manzo,A.Ahumada,Effectsofsolventmediumonsolubility.5.Enthalpicand entropiccontributionstothefree-energychangesofdisubstituted benzene-derivativesinethanolwaterandethanolcyclohexanemixtures,Journalof PharmaceuticalSciences79(1990)1109–1115.

[46]S.Neau,S.Bhandarkar,E.Hellmuth,Differentialmolarheatcapacitiestotest idealsolubilityestimations,PharmaceuticalResearch14(1997)601–605. [47]T.Nielsen,J.Abildskov,P.Harper,I.Papaeconomou,R.Gani,TheCAPECdatabase

,JournalofChemicalandEngineeringData46(2001)1041–1044.

[48]J.Qingzhu,M.Peisheng,Y.Shouzhi,W.Qiang,W.Chan,L.Guiju,Solubilitiesof benzoicacid,p-methylbenzoicacid,m-methylbenzoicacid,o-methylbenzoic acid,p-hydrobenzoicacid,ando-nitrobenzoicacidin1-octanol,Journalof ChemicalandEngineeringData53(2008)1278–1282.

[49]S.Romero,A.Reillo,B.Escalera,P.Bustamante,Thebehaviourofparacetamol inmixturesofamphiproticandamphiprotic–aproticsolvents.Relationshipof solubilitycurvestospecificandnonspecificinteractions,Chemical& Pharma-ceuticalBulletin44(1996)1061–1064.

[50]M.Sacchetti,ThermodynamicanalysisofDSCdataforacetaminophen poly-morphs,JournalofThermalAnalysisandCalorimetry63(2001)345–350. [51]R.Weast,M.Astle,in:CRCHandbookofDataonOrganicCompounds,CRCPress,

BocaRaton,FL,1985.

[52]M.Wu,S.Yalkowsky,Estimationofthemolarheatcapacitychangeonmelting oforganiccompounds,Industrial&EngineeringChemicalResearch48(2009) 1063–1066.

[53]F.Xu,L.Sun,Z.Tan,J.Liang,R.Li,Thermodynamicstudyofibuprofenby adia-baticcalorimetryandthermalanalysis,ThermochimicaActa412(2004)33–37. [54] R.VanDuyne,Taylor,Self-associationandhydrationofbenzoicacidinbenzene

,JournalofPhysicsChemistry48(2009)1063–1066.

[55]C.-C.Chen,Y.Song,Extensionofnonrandomtwo-liquidsegmentactivity coef-ficientmodelforelectrolytes,Industrial&EngineeringChemistryResearch44 (2005)8909–8921.

[56] C.-C.Chen,P.A.Crafts,Correlationandpredictionofdrugmoleculesolubility inmixedsolventsystemswiththenonrandomtwo-liquidsegmentactivity coefficient(NRTL-SAC)model,Industrial&EngineeringChemistryResearch 45(2006)4816–4824.

[57] I.Tsivintzelis,G.Dritsas,C.Panayiotou,Analternativeapproachto nonrandom-nessinsolutionthermodynamics,Industrial&EngineeringChemistryResearch (2006)7264–7274.

[58]J.Gross,G.Sadowski,Perturbed-chainsaft:anequationofstatebasedona perturbationtheoryforchainmolecules,Industrial&EngineeringChemistry Research40(2001)1244–1260.

Figure

Fig. 1. (a) Ibuprofen, (b) paracetamol, (c) salicylic acid, (d) benzoic acid, (e) 4- 4-aminobenzoic acid and (f) anthracene.
Fig. 2. Melting enthalpy of (a) ibuprofen, (b) paracetamol, (c) salicylic acid and (d) benzoic acid as a function of the melting temperature.
Fig. 3. Experimental solubility of (a) ibuprofen [20,21], (b) paracetamol [22], (c) salicylic acid [23–27], (d) benzoic acid [28–30], (e) 4-aminobenzoic acid [31–34] and (f) anthracene [35–39] in various solvents as a function of the reverse temperature (v
Fig. 4. UNIFAC predictions of (a) ibuprofen, (b) paracetamol, (c) salicylic acid, (d) benzoic acid, (e) 4-aminobenzoic acid and (f) anthracene with Eq
+7

Références

Documents relatifs

Usually, amines are not corrosive but during operation of amine units, degradation of the solvent may occur due to the reaction with contaminants like oxygen.. As a result,

This measurement, combined with a theoretical solubility model, was used to assess the probable order of magnitude of the NH 4 -dawsonite Gibbs free energy value at 25

In this report, we demonstrated that treatment with CamKII inhibitors (KN62 or KN93) resulted in a partial rescue of the cel- lular phenotype (abnormal centrosome

Ce travail présente une étude sur l’intégrité des joints de colle entre des essences de bois tropicaux différents, les éprouvettes comprenant deux différents

Open porosity measurements are confirmed by the reduction of the wave velocity by ultrasonic measurement and microscopic observations revealing smaller pores for shell

Based on the results of this study, it can be concluded that light curing with an energy density of 4,000 mJ/cm 2 gen- erated the lowest shrinkage force and linear displacement but

A travers l’analyse de la mise en œuvre du Schéma d’action sociale de proximité ressort le fait qu’il n’y a pas une stratégie unique de conduite du changement,

At first pick, yield was lower in WW than in the other two water regimes, but at third pick, the results were opposite (fig. These results indicated that WS and 2xWW