Drug permeation in biomembranes: in vitro and in silico prediction and influence of physicochemical properties

European Journal of Pharmaceutical Sciences 23 (2004) 13–47

Review

Drug permeation in biomembranesIn vitro and in silico prediction and influence

of physicochemical properties

Annika Mälkiäa, Lasse Murtomäkia,∗, Arto Urtti b, Kyösti Kontturia

a Laboratory of Physical Chemistry and Electrochemistry, Helsinki University of Technology, P.O. Box 6100, FIN-02015 HUT, Finlandb Department of Pharmaceutics, University of Kuopio, P.O. Box 1627, FIN-70211 Kuopio, Finland

Received 25 November 2003; received in revised form 13 May 2004; accepted 24 May 2004

Available online 10 July 2004

Abstract

In the past decades, it has become increasingly apparent that in addition to therapeutic effect, drugs need to exhibit favourable absorption,distribution, metabolism and excretion (ADME) characteristics to produce a desirable response in vivo. As the recent progress in drug discoverytechnology enables rapid synthesis of vast numbers of potential drug candidates, robust methods are required for the effective screening ofcompounds synthesized within such programs, so that compounds with poor pharmacokinetic properties can be rejected at an early stageof drug development. Furthermore, a viable in silico method would save resources by enabling virtual screening of drug candidates alreadyprior to synthesis. This review gives a general overview of the approaches aimed at predicting biological permeation, one of the cornerstonesbehind the ADME behaviour of drugs. The most important experimental and computational models are reviewed. Physicochemical factorsunderlying the permeation process are discussed.© 2004 Elsevier B.V. All rights reserved.

Keywords:Biological permeation; Partition; Experimental models; Computational models; Physicochemical properties

1. Introduction

The recent development in drug discovery method-ology, including concepts such as combinatorial chem-istry and high throughput screening, is accompanied bya need to rapidly and effectively evaluate the biophar-maceutical properties of compounds synthesised withinsuch programs in small quantities. It is recognised thatthe absorption, distribution, metabolism and excretion(ADME) characteristics of the drug compound are crucialfor successful therapeutic activity in vivo. Accordingly,the focus in drug development has moved from aimingsolely at maximum drug–receptor interactions to broaderproperty-based design, including also pharmacokineticand pharmaceutical properties (van de Waterbeemd et al.,2001).

∗ Corresponding author. Tel.:+358 9 4512575; fax:+358 9 4512580.E-mail address:[email protected] (L. Murtomäki).

One property of particular importance is the ability ofdrugs to cross biological membranes. The biological per-meability of a drug shapes its pharmacokinetic profile inthe body, affecting its absorption, distribution and elimina-tion. Scientists from various fields including biology, chem-istry, pharmacy, medicine, physics and computer sciencehave approached the task of developing effective and accu-rate methods for predicting biological permeation of drugs.The importance of this work is corroborated by the findingsthat approximately 40% of the failures in drug developmentprograms in clinical phase are due to problems in pharma-cokinetics and drug delivery (Prentis et al., 1988; Kennedy,1997).

The experimental approaches to study drug transport canbe roughly divided into two groups: (i) transport studiesthrough different types of biomimetic layers to model bi-ological permeation; and (ii) two-phase partition studieswhere drug partitioning between water and a lipidic phaseis quantified. Although effective methods to predict biolog-ical permeation are of utmost importance, two-phase parti-

0928-0987/$ – see front matter © 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.ejps.2004.05.009

14 A. Mälkiä et al. / European Journal of Pharmaceutical Sciences 23 (2004) 13–47

tion studies of drugs can provide valuable information ondetails and reasons behind the mechanisms of permeation,such as hydrogen bonding ability and electrostatic interac-tions. Furthermore, due to their longer and more standard-ised experimental traditions, partition coefficients measuredin two-phase bulk solvent systems have so far remained themain descriptors of in vivo drug permeation.

The first calculative approaches related to biological per-meation were semi-empirical models, developed to calculatetwo-phase partition coefficients of drugs. The increase incomputer resources over the past decades has given rise toa variety of computational approaches, devoted to calculat-ing descriptors of biological permeation. The developmentof a reliable theoretical method to predict biological perme-ation would not only save time and resources spent on ex-perimental permeation studies, but also enable screening ofdrug candidates prior to synthesis.

This review discusses the methods used to study drug par-tition and permeation employing experimental model sys-tems as well as computational approaches. The emphasislies on passive transcellular drug transport, and electrochem-ical methods. Solubility, while an equally important ADMEproperty, is not included in the discussion in order to limitthe scope and length of the survey.

2. Basic concepts

2.1. Biological permeation

Lipid solubility and partitioning into lipid phases are cru-cial factors in pharmacokinetics. The transfer of drugs in thehuman body is determined by their ability to move acrossthe lipid bilayer of epithelial and endothelial cell linings.Permeation across the cell membranes takes place by threemain mechanisms: transcellular diffusion, paracellular dif-fusion and active transport (either transport into the cells orefflux out of the cells). Paracellular permeation is mainlygoverned by the size and the number of the pores betweenthe cells, as well as the size and charge of the drug. Activetransport processes require specific binding of the drug tothe transporter protein. Diffusion across the cell membranesand transcellular permeation through the cells constitute themost important mechanisms, by which drugs cross biologi-cal membranes. The rate of transcellular diffusion affects thepharmacokinetics of the drug in the body in various ways.

Oral, transdermal, ocular and pulmonary absorptionrequire that the drug readily passes through biologicalmembranes, as does the distribution of the drug acrossthe blood–brain barrier (BBB) or blood retina barrier, orits displacement to intracellular targets. Cell membranepermeation is also a prerequisite for drug elimination bythe hepatocytes in the liver, but on the other hand, readilypermeating drugs may also undergo reabsorption throughthe tubular membranes of the kidney, thereby reducingtheir excretion into the urine. Since the majority of drugs

are administered via the oral route, the most studied formof biological permeation is human intestinal absorption(HIA). Likewise, a large number of drugs act via the centralnervous system, producing either therapeutic or adverse ef-fects. Therefore, it is important to evaluate or predict drugpermeation of the blood–brain barrier.

The main properties of a drug influencing its per-meation through biological membranes are lipophilicity,hydrogen-bonding capacity, charge and size (Camenischet al., 1996). These will be discussed in the following sec-tions together with properties of the partition or permeationmedium.

2.2. Factors affecting biological permeation

Cellular membranes consist of a lipid bilayer with em-bedded membrane proteins and polysaccharide chains onthe cell membrane surface. For passive transcellular perme-ation to take place, the drug must partition into the lipoidmembrane. Several factors describe and influence partition-ing into cell membranes.

2.2.1. log PThe lipophilicity of a drug is the most used single physico-

chemical property to predict its permeation in biologicalsystems (Testa et al., 1996). Simply stated, the lipophilic-ity of a drug is its tendency to prefer a lipidic, or oil-likeenvironment to an aqueous one. However, behind this prop-erty lies a net of intermolecular interactions such as hydro-gen bonding and dipole effects. Thus, although lipophilicityis a property ascribed to the drug compound, it is highlydependent on the choice of lipidic environment (Kaliszan,1992). The lipophilicity of a drug is traditionally expressedas logP, the logarithm of its partition coefficient between alipidic and an aqueous phase.

The chemical potential of a solute,µi, can be expressedeither on the mole fraction scale or the concentration (ormolality) scale:

µi = µ0i,x + RTln xi = µ0

i,c + RTln(ai/c0) (2.1)

whereµ0i,x/c is the standard chemical potential,xi the mole

fraction andai = γici the activity of a single species of thesolute;c0 is the standard concentration, 1.0 mol dm−3. Themole fraction scale refers to Henry’s law, which is alwaysvalid for the solvent, and for the solute at infinite dilution.The most significant problem when applying Henry’s lawto the solute is that the standard state is odd,xi = 1, i.e.there is no solvent. The standard state of the concentrationscale is also a bit hypothetical: 1.0 mol dm−3, with the ac-tivity coefficient �i equal to one, but at least achievable inprinciple.

The thermodynamic partition coefficientPi is defined asthe ratio of the activity of a single species of solute in twoimmiscible phases at equilibrium, by convention placing the

A. Mälkiä et al. / European Journal of Pharmaceutical Sciences 23 (2004) 13–47 15

aqueous phase activity in the denominator (Reymond, 2001):

Pi = aoi

awi

= γoi co

i

γwi cw

i

= exp

(−µ

o,0i,c − µ

w,0i,c

RT

)

= exp

(−�o

wG0i

RT

)(2.2)

A larger P or more positive logP thus corresponds to ahigher lipophilicity. On basis ofEq. (2.1)the partition coef-ficient can also be defined as the ratio of the mole fractions:

Pi,x = xoi

xwi

(2.3)

As the ratio of the activity coefficients approaches unityrather fast with decreasing concentration, it is easy to showthat the relation between these two quantities simply is:

logPi,c = logPi,x + log

(V̄w

V̄o

)(2.4)

whereV̄w/o is the molar volume of the water or the oil phase.For example, in then-octanol/water system the last term is−0.94 and hence not insignificant. Furthermore, the parti-tion coefficient has been calculated as the ratio of amount ofsolute in the respective phases—it should be noted that thismethodology only yields comparable logP values if the vol-umes of the phases are equal. In the following, the partitioncoefficient refers to the one defined inEq. (2.2).

2.2.2. Hydrogen bondingOrdered lipid layers provide a finite amount of hydrogen

bonding groups. These groups, the majority of which are hy-drogen bond acceptors, are located exclusively in the headgroup region of the lipids. In order to partition into the hy-drocarbon region of the bilayer, the solute must be lipophilicenough to overcome the energy losses that occur in break-ing the hydrogen bonds with water or the lipid head groups.This step can thus present a considerable energy barrier forsolutes, which exhibit strong hydrogen bonding (donor) ten-dencies. Accordingly, biological permeation can be expectedto markedly depend on the hydrogen bonding capacity ofthe solute (El Tayar et al., 1991; Conradi et al., 1991).

Octanol, which is the most commonly used lipophilicphase in two-phase partition experiments, can provide agood hydrogen bonding environment for solutes. A solute,which makes strong hydrogen bonds with water will beable to make similar bonds in octanol. The energy requiredin the desolvation step of the transfer from water to oc-tanol is thus small, and hydrogen bonding will consequentlybe of little effect to the octanol–water partition coefficient(Conradi et al., 1991). An example of the effect of hydro-gen bonding on biological partition is seen inFig. 1. Pep-tide mimetic compounds of similar octanol–water partitioncoefficients show variable transcellular permeation throughCaco-2 monolayers in accordance with a difference in theirhydrogen bonding ability. Compounds in the upper group,

Fig. 1. Transcellular permeation through Caco-2 cell monolayers as afunction of octanol–water partition coefficients for a set of dipeptidemimetic compounds. The division of the data points into two groups is inaccordance with the different hydrogen bonding ability of the peptides.Compound names are given in the Appendix. Reproduced fromGoodwinet al. (2001)with permission from the American Chemical Society.

numbers 7, 8, 9a and 10 are able to form four hydrogenbonds (two donors and two acceptors) while the compoundswith numbers 3a, 4, 5a and 6 are able to form six (threedonors and three acceptors). Hence, higher hydrogen bond-ing capacity has a negative effect on permeation.

The hydrogen bonding capacity of a solute can be ex-pressed as the difference between its partition coefficientinto octanol and a solvent devoid of forming hydrogenbonds, typically an alkane.Young et al. (1988)comparedblood–brain partition data of histamine H2 receptor antago-nists to their octanol–water partition coefficients and foundno correlation. The blood–brain partitioning was, however,successfully predicted by�logPoctanol/alkane, defined aslogP (1-octanol/water) – logP (cyclohexane/water).

In a study byEl Tayar et al. (1991), the solvatochromicapproach was employed to analyse two-phase partition dataof a variety of solutes in five solvent systems. The solva-tochromic methodology (Taft et al., 1985a; Kamlet et al.,1983), also called the linear solvation energy relationshipapproach (LSER), states that solubility of a solute in a givensolvent system can be related to solute–solvent interactionsthrough the linear combinations of three types of terms: thecavity formation term expressed by the intrinsic molecularvolume of the solute (V), the solute polarity/polarisabilityterm (π∗) and the terms indicating the hydrogen bond ac-ceptor (βH) and donor (αH) strength of the solute. The sol-vatochromic equation has also successfully been applied topredict oil–water partition coefficients:

logP = a

(V

100

)+ bπ∗ + cβH + dαH + e (2.5)

El Tayar et al. (1991)correlated the partition coeffi-cients of 121 solutes with their solvatochromic parametersin five water–oil solvent systems (1-octanol,n-heptane,


chloroform, diethyl ether andn-butyl acetate) in order toevaluate the relative contribution of the parameters and toestablish the information content of the descriptor�logP.They found that the solute hydrogen bond donor abil-ity (αH) was significant only in then-heptane–water andchloroform–water systems, meaning that in the other threesystems the organic solvents were as good hydrogen bondacceptors as water. This led the authors to suggest thedifference between the partition coefficients measured inoctanol–water andn-heptane–water (�logPoctanol/heptane)or chloroform–water (�logPoctanol/chloroform), as parametersto describe the hydrogen bond donor capacity of the solutes.

It has been pointed out byLeo (2000)that caution shouldbe taken in the calculation of�logP as the difference be-tween the partition coefficients in two solvent systems toestimate solute hydrogen bond donor capacity. For example,the contribution of the cavity term inEq. (2.4)has been foundto vary between the octanol–water and the chloroform–watersystems (Taft et al., 1996) and should be taken into ac-count. Furthermore, even if the remaining solvatochromicparameters may appear insignificant in a correlation equa-tion, this may result from an opposite effect of the polar-ity/polarisability term (π∗) and the solute hydrogen bondacceptor term (βH). In addition, Leo suggested the inclusionof a parameter for excess alkane affinity (XAA) to accountfor the ability of the alkane chain of, e.g. octanol to accom-modate large alkane portions of a solute, which he foundcapable of explaining outliers in earlier calculations of so-lute hydrogen bond donor strength (Leo, 2000). Naturally,the simplicity of the�logP calculation will suffer from anyadditional parameter that has to be included.

Goodwin et al. (2001)compared the Caco-2 monolayerpermeation of a series of dipeptide mimetics to their bulkpartition coefficients. Neither the octanol–water partition co-efficient (seeFig. 1) nor �logPoctanol/n-heptanewas success-ful in predicting biological permeation alone. A qualitativelybetter correlation was obtained with�logPn-heptane/glycol.Solvatochromic analysis showed that this partition coeffi-cient is a hybrid of the other two, consisting of contributionsfrom both hydrogen bond donor and acceptor potential,as well as a volume term. An improved correlation wasobtained with a linear combination of logPoctanol/water and�logPoctanol/n-heptane, suggesting that a more specific rela-tionship of solute volume and hydrogen bonding capacityis required to properly describe biological permeation.

Lipid-containing partition systems have been consideredto model the hydrogen bonding ability of biological mem-branes better than bulk solvents. Indeed, it was found byVaes et al. (1998)that differences in octanol–water andliposome–water partition coefficients of neutral compoundswere almost exclusively explained by hydrogen bonding.Disappearance of solute from the aqueous bulk phase inliposome–water partition studies does, however, not nec-essarily indicate partition into the bilayer core.Jacobs andWhite (1989) studied the partition of small hydrophobicpeptides into liposomes and found that the binding of

the peptides at the liposome–water interface was mainlydriven by the hydrophobic effect, while insertion into thebilayer interior was strongly dependent on the interfacialhydrogen-bonding.

2.2.3. Solute chargeMost drugs are weak acids or weak bases and exist in solu-

tion at equilibrium between their neutral and ionised forms.When studying two-phase partition of ionisable drugs, onehas to distinguish between the partition coefficient and thedistribution coefficient (also called the apparent partition co-efficient), the latter of which is defined as the ratio of thetotal drug concentrations (both neutral and ionised species)in the lipidic and aqueous phases at a certain pH:

logDi = logao

i,tot

awi,tot

(2.6)

An assumption that has been widely stated in associationwith biological permeation is the pH partition hypothesis,according to which only neutral and non-polar compoundsare able to cross biological membranes. Accordingly, trans-membrane permeation of charged and hydrophilic drugs hasgenerally been assumed to take place through paracellularspace. However, the surface area of the cell membranes isby several orders of magnitude larger than that of the para-cellular channels, which outweighs the lower permeation ofthe transmembrane route, and due to this some of the trans-port of such compounds may in reality take place throughtranscellular diffusion (Artursson et al., 2001).

Observations of transmembrane diffusion of ionic specieswere initially explained by the formation of neutral ion pairs(Scherrer, 1984; Neubert, 1989; Takács-Novák and Szász,1999). While not counteracting the ion pair theory, it haslately been shown that lipophilic ions may partition intomembrane phases on their own account (Kürschner et al.,2000). In addition, it has been pointed out that even if thepassive membrane transport of ions may be much slowerthan that of neutral compounds, the abundance of the ionisedspecies can make it significant (Palm et al., 1999). Further-more, a too lipophilic neutral solute may accumulate in themembrane interior instead of passing trough it. In such cases,ionisation could enhance permeation (Suhonen et al., 1998).

Other studies propose that permeating compounds enterthe membrane in their neutral form despite being ionisedin the bulk aqueous phase, due to a shift in the pKa at themembrane surface caused by the change in polarity and thelocal electrostatic surface potential (Miyazaki et al., 1992;Beschiaschvili and Seelig, 1992). This explanation is sup-ported by observations that compounds that carry at leastone stable charge, i.e. acids with pKa <4 or bases with pKa

>10 do not readily cross the blood–brain barrier. On theother hand, compounds possessing tertiary ammonium moi-eties with a pKa of ∼8 exhibit higher BB permeation thanexpected on basis of the fraction of unionised drug (Fischeret al., 1998).


Yet another explanation for ion permeation is the existenceof an aqueous diffusion layer at the membrane interface towhich the ionised species diffuses from the bulk aqueousphase much faster than the neutral species, whereas the lat-ter enters the membrane much more readily than the ionisedspecies. To maintain equilibrium, an interfacial dissociationreaction takes place in the diffusion layer where the neu-tral species is recovered due to the dissociation equilibrium(Bouchard et al., 2002). Furthermore, it has been shown thation transport can be increased by a transmembrane potential(see next section,Mayer et al., 1985) and that the observedincrease of ion partition in presence of counter ions may infact be caused by an interfacial potential difference insteadof ion pair partition (Bouchard et al., 2001a). At present,however, the effect of charge on biological permeation is notwell understood and more and comprehensive studies on thesubject are required.

The reason for the poor permeability of lipid bilayers to-wards ions is desolvation. Ions transferring through a lipidbilayer must first partition into the interfacial region andthen diffuse through the non-polar hydrocarbon core. Be-fore entering the hydrocarbon region, most ions will give upsome of their hydration water. This process is energeticallyunfavourable and the main reason for the low diffusive per-meation of ions in lipid bilayers (Cevc, 1990). The situationis traditionally appreciated in electrostatic terms by consid-ering the energy cost of transferring an ionic charge fromwater, with a high dielectric constant, to the membrane in-terior with a low dielectric constant. The electrostatic con-tribution to this energy can be estimated from theoreticalmodels, such as the modified Born equations put forth byParsegian (1969)(2.7) andAbraham and Liszi (1978)(2.8):

�owG0

i,es = NAz2i e2

4πε0

[1

2ri

(1

εo− 1

εw

)− 1

εodo ln2εw

εw + εo

](2.7)

�owG0

i,es = NAz2i e2

8πε0

[(1

εo − 1

ε1

)1

boi

−(

1

εw − 1

ε1

)1

bwi

](2.8)

whereεw andεo are the relative permittivities of the aque-ous and lipidic phases, respectively,ε0 is the permittivityof vacuum andri is the ionic radius. InEq. (2.7), the sec-ond term in the square brackets corrects the Born modelfor finite thicknessdo of the lipidic phase. The correc-tion term remains insignificant so long asri � do. In theAbraham–Liszi approach inEq. (2.8), the ion is surroundedby a local solvent layer of thickness (bi − ri) with the rela-tive permittivity of ε1, as shown inFig. 2; εb is the relativepermittivity of the bulk solution.

The traditional estimates above,Eq. (2.7) and (2.8), ap-ply in a dielectric continuum with an appropriate relativepermittivity, which is, of course, an oversimplification ofthe lipid bilayer. For example, the permeability of Na+calculated fromEq. (2.7) is of the order of 10−29 cm s−1,

Fig. 2. Schematic representation of the Abraham–Liszi model.

while measured permeabilities are ca. 14–17 orders ofmagnitude higher (Wilson and Pohorille, 1996). Therefore,molecular dynamic simulations of the transport of Na+ andCl− across a glycerol 1-monooleate (GMO) bilayer werecarried out, and it was found that when the ion approachedthe polar heads of the bilayer, the bilayer withdrew inwards,forming a cavity filled with water and making the bilayerthinner at the crossing site. Accordingly, when the ion leftthe bilayer on the opposite side, a similar disruption wasobserved. The loss of hydration water in the centre of thebilayer was compensated for with solvation by the oxygenatoms of the GMO headgroups, keeping the total solvationnumber of the ion practically constant during the crossing.With these simulations, the authors (Wilson and Pohorille,1996) achieved permeabilities, which were in accordancewith the measured values.

As solvation increases with charge, bilayers are al-most impermeable to polyions, except for certain divalentcations, which are able to cross membranes by formingneutral complexes with membrane components of oppositecharge (Cevc, 1993). The zwitterions, however, form anexception to this.Pagliara et al. (1997)point out in theirthorough review on the subject that the apparent lipophilic-ity of zwitterions is often higher than expected from theircharged nature due to two reasons: (i) the presence of asignificant portion of the neutral tautomer of the zwitterions(seeFig. 3); or (ii) the partial intramolecular neutralisation

Fig. 3. Dissociation scheme of a zwitterion.KZ is the tautomeric con-stant describing the equilibrium between the globally neutral, zwitterionicspecies and its uncharged tautomer. Reproduced fromBouchard et al.(2002) with permission from Kluwer.


Fig. 4. Theoretical distribution profile of the monobase D with logPD

= 3, logPDH+ = −1 and aqueous pKa = 9.

of the two opposite charges. As discussed by the authors,zwitterions can provide several benefits over compoundscarrying a single ionisable group: (i) they exhibit a practi-cally constant value of lipophilicity around the isoelectricpH (typically overlapping with the physiologically relevantpH region); and (ii) zwitterions in which the neutral tau-tomer is present to a notable extent may exhibit significanttranscellular permeation.

Conclusively, taking into account the possibility of ionpartition, the relationship between pH and partition for amonoprotic compound can be expressed by the followingEq. (2.9)(Reymond, 2001) andFig. 4:

logDD/DH = log(PD + PDH × 10pKwa −pH)

− log(1 + 10pKwa −pH) (2.9)

where D refers to the neutral base or deprotonated acid, andDH stands for the protonated base or neutral acid, respec-tively.

2.2.4. Electrostatics of lipid bilayersThe majority of membrane lipids are comprised of a head

group region with one or more charged units. Since chargedlipids are commonly either zwitterionic or anionic, the netcharge of membrane surfaces is negative. The transversestructure of the bilayer causes the charged and dipolar lipidgroups to be relatively fixed with respect to their orientationand location from the bilayer centre. Consequently, thesecharges and dipoles are only partially compensated by wa-ter dipoles and solution electrolytes, and a complex elec-tric profile is generated over the membrane. This profile iscomposed of two components: the surface, or double-layerpotential, and the dipole potential.

The surface potential arises from charged lipid head-groups at the membrane surface. These fixed charges attractcounterions from the bulk aqueous solution to the interface,giving rise to a so-called electrical double layer, and anelectric potential profile is established at the surface. Thesurface potential is not likely to have a considerable effecton the partition of neutral species into the bilayer, but in-

creases the surface concentration, and thus the probabilityof permeation, of cations, whereas the probability of anionpermeation should decrease.

The membrane dipole potential establishes itself in theregion between the aqueous phase and the hydrocarbon inte-rior of the membrane. The origin of the dipole potential is notwell understood, but the orientation of the lipid head groupsand the polarised water associated with the membrane in-terface are believed to provide an important contribution(Brockman, 1994; Gawrisch et al., 1992). The dipole poten-tial of lipid bilayers is estimated to be hundreds of millivoltspositive in direction of the membrane interior (Brockman,1994), thereby making the bilayer interior more accessibleto anions than cations and counteracting the effect of thesurface potential.

Meijer et al. (1999)employed a self-consistent anisotropicfield approach to model a phospholipid bilayer. They foundthat the plane tilting of the dimyristoyl phosphatidylcholine(DMPC) headgroups generated an electrostatic potentialprofile over the membrane, which, albeit an order of mag-nitude smaller than the estimated dipole potential in realbilayers, caused anions and cations to distribute unevenlyover the bilayer. Both positively and negatively chargedions had difficulties penetrating the hydrocarbon interiorof the bilayer, but due to the potential profile it was moreaccessible to anions (Fig. 5).

Fig. 5. The volume fractionϕ, charge distributionq and electrostaticpotential ψ profiles through a cross section of a free-standing liquidcrystalline DMPC membrane. The director shown in the diagrams isperpendicular to the membrane surface, indicating the hydrophobic corewith its tail and the head group area with its head. The volume fractionof salt solution in the bulk is 0.002. The centre of the membrane is atz = 0. Reproduced fromMeijer et al. (1999)with permission from theAmerican Institute of Physics.


An additional feature of cellular membranes that is absentin bulk partitioning systems, is the potential difference overthe membrane. This so-called transmembrane potential ofapproximately –100 mV (interior to exterior) is maintainedby active transporter proteins, which move ions across themembrane against their concentration gradient.Mayer et al.(1985)studied the effect of a transmembrane potential dif-ference on cation partitioning into lipid vesicles. They foundthat the partition of the positively charged anaesthetic dibu-caine was greatly enhanced in the presence of a potentialgradient of –170 mV compared to the situation where nopotential gradient was present.

The picture presented in this section only concerns theelectric properties of a lipid bilayer. In biological membranesfurther contributions arise from the variety of proteins asso-ciated with the bilayer as well as the negatively charged gly-cocalix surrounding the outer leaflet of the cellular plasmamembrane (Langner and Kubica, 1999).

2.2.5. Solute size and order of the partition phaseThe effect of solute size on two-phase partition depends

significantly on the structure and nature of the lipidic phase.Upon partitioning, in addition to the energy of cavity forma-tion, the energy of reorganization of the solvent moleculesaround the solute needs to be taken into account. When par-tition takes place into a bulk phase, the size effect mainlydepends on the polarity of the solvent compared to that ofwater (Spencer et al., 1979). However, in lipid membranessize effects may be more complex due to steric hindrance.Since an increase in solute size usually induces changes inother properties related to partition, an experimental evalu-ation of the size effect is not straightforward.

Various studies have focused on mapping the structure ofa membrane bilayer.Hubbell and McConnell (1971)usedelectron paramagnetic resonance to show that the hydrocar-bon chains of membrane phospholipids are highly orderedin the vicinity of the polar head groups, but that there existsa gradient of chain disorder from the membrane surface to-ward midbilayer. Similar results have been reported on basisof deuterium magnetic resonance (Seelig and Seelig, 1974)and computational studies (Tu et al., 1995; Meijer et al.,1999).

Marrink and Berendsen (1996)studied the transport ofsmall molecules across a phospholipid membrane usingmolecular dynamics simulations. They described the lipidbilayer with a four-region model: region 1 is characterisedby loosely bound water molecules that are attached to thecholine headgroups; in region 2 the headgroup density ishigh and the water molecules are strongly bound, the majorpart of the glycerol backbones also reside in this region; re-gion 3 contains the more ordered parts of the lipid tails andresembles a soft polymer; and region 4 comprises of themajor parts of the lipid chains, in this region the chains aredisordered and their density is low. According to their sim-ulations the largest resistance for solute permeation occursin region 3, especially for hydrophilic penetrants. Only for

very hydrophobic solutes may diffusion in the water layer(regions 1 and 2) become rate-limiting. Region 4, i.e. thebilayer centre, is characterised by lower resistance to pene-tration than region 3 due to its more disordered nature. How-ever, for highly hydrophobic solutes, which exhibit smallelectrostatic interactions, this region may act as a trap due toits favourable solution environment. An increasing size ofthe solute is expected to further emphasise the rate-limitingeffect of region 3 for polar solutes. The anisotropic structureof the region is furthermore likely to favour permeation ofnon-spherical molecules over spherical ones.

An experimental study supports the importance of phasestructure on permeation and partition. Benzene partition intophospholipid monolayers was found to depend significantlyon the surface density of the lipids (De Young and Dill,1988). Increasing the surface density led to solute exclu-sion: benzene partitioning decreased by an order of magni-tude as the surface density increased from 50 to 90% of itsmaximum value, a range readily accessible in bilayers andbiomembranes under physiological conditions. The effect ofsurface density was attributed to an ordering of the lipidchains, as a similar effect could be observed regardless ofthe method used to alter the surface density: cholesterol ad-dition, as well as alteration of temperature and phospholipidchain length were examined.

Due to its polar hydroxyl groups, octanol is thoughtto form inverted micellar aggregates, particularly in itswater-saturated form (Franks et al., 1993; DeBolt andKollman, 1995), as shown by the molecular dynamics sim-ulation image inFig. 6. This would suggest that the octanolphase is not an isotropic bulk phase, but that it containsstructures similar to those formed by lipids in biological sys-tems. Furthermore, both experimental (Cramb and Wallace,1997) and theoretical (Michael and Benjamin, 1995) stud-ies indicate that at the octanol–water interface the octanolmolecules are ordered with their polar groups towards theaqueous phase. These findings present a probable explana-tion for the success of the octanol–water partition coefficientas a lipophilicity descriptor of biopermeation. However, thestructural organisation of the interfacial octanol layer is notequivalent of the order in lipid bilayers, and thus partitioninto octanol will not reflect steric hindrance properly.

2.2.6. ConclusionsOctanol–water partition coefficients can be expected to

model the affinity of the drug for the membrane interface andhydrocarbon chains, and will successfully predict passivepermeation in biomembranes only when hydrogen-bondingand electrostatic effects are not rate-limiting. Lipid partitionphases provide a more biomimetic environment, but beingan overall value, the lipid–water partition coefficient doesnot specify whether partition takes place into the bilayerinterior or to the lipid–water interface. Biological permeationof charged species has received relatively little attention,however thorough studies of the topic would be warranteddue to the charged nature of many drugs.


Fig. 6. Molecular dynamics simulation of water-saturated 1-octanol. Yellow van der Waals radii represent water oxygens, red represents octanol hydroxylhydrogens, and hydrocarbon tails are shown as black chains. Reproduced fromDeBolt and Kollman (1995)with permission from the American ChemicalSociety.

3. Experimental approaches

3.1. Octanol

The octanol–water partition coefficient is by far the mostextensively used descriptor of the lipophilic character ofdrugs. The partition coefficient is traditionally measured bystirring or shaking an octanol–water mixture in the pres-ence of the solute, followed by analysis of the equilibriumconcentration of solute in one or both phases (Leo et al.,1971). In principle the procedure is very simple, but in realityconventional shake-flask measurements are time-consumingand tedious to make, and require relatively large amountsof solute. Recent literature presents solutions to these prob-lems in the form of microscaling (Morgan et al., 1998) andan automated parallel plate assay (Hitzel et al., 2000). ApH-metric technique based on potentiometric titration hasbeen developed for rapid determination of partition coeffi-cients for ionisable drugs (Avdeef, 1993).

Correlations between octanol–water distribution coeffi-cients and biological permeation have in some cases beenreasonably good (Palm et al., 1996; Krämer, 1999; Wilset al., 1994; Zhao et al., 2002), suggesting the existence of abell-shaped (or sigmoidal) relationship with maximum ab-sorption around logD7.4 = 2–3 (seeFig. 7). However, many

studies also report poor or nonexistent correlations (Figs. 1and 12b; Balon et al., 1999b; Palm et al., 1997; Conradi et al.,1991; Zhu et al., 2002), thus no definite relationship hasbeen established. However, in combination with other de-

Fig. 7. Logarithm of apparent permeation across Caco-2 cell monolayers(Papp/nm s−1) and logD at pH 7.4. Compounds above the dotted line areassumed to be well absorbed. The arrow indicates uncertainty of the resultowing to analytical limit of the determination. Reprinted fromKrämer(1999) with permission from Elsevier.


scriptors, octanol–water partition coefficients have in manycases resulted in good correlations, as discussed inSection 5.

3.2. Liposomes

After the realisation that lipids spontaneously form closedbilayer membranes in aqueous surroundings, these so-calledliposomes were introduced as models for cellular mem-branes in 1965 (Bangham et al., 1965; Bangham, 1993).Owing to their superior biomimetic properties, liposomeshave also been proposed as alternatives to octanol in drugpartition studies.

The abundance of available lipids and preparation tech-niques has resulted in a variety of liposome types beingemployed as partition phases.Balon et al. (1999a)under-took a study in which they compared partition data ob-tained with different liposomes. They found only small dif-ferences between small and large unilamellar liposomes, andbetween small unilamellar liposomes generated by sonica-tion and equilibrium dialysis. The authors suggested the useof sonicated small unilamellar vesicles (S-SUV) as a stan-dard system due to the simple preparation procedure andlarge surface area, which enables shorter equilibrium timesand higher lipid to drug ratios. Liposomes have furthermorebeen used in association with chromatography, as discussedin the following section.

Studies on the effect of lipid composition have indicatedthat the presence of cholesterol generally decreases solutepartition (Balon et al., 1999a; Betageri and Rogers, 1988).Lipid charge has been found to influence partition of ion-isable solutes (Betageri and Rogers, 1988; Krämer andWunderli-Allenspach, 1996; Surewicz and Leyko, 1981).In a study byKrämer et al. (1998), partition into negativelycharged liposomes was in fact larger for the cationic speciesthan for the neutral solute. However, it has been pointedout that partition measured in the liposome–water systemmay not always reflect transmembrane permeation (Palmet al., 1998) as solutes may also associate with the mem-brane interface without entering the bilayer interior (Jacobsand White, 1989). This may also be the reason behind theobserved charge-induced increase in partition. Likewise,this is a possible explanation for the significantly higherionic partition coefficients obtained in liposomes comparedto the octanol–water system (Avdeef et al., 1998; Balonet al., 1999b). Contrasting reports on correlations betweenthe octanol and the liposome systems for neutral solutes(Gobas et al., 1988; Avdeef et al., 1998; Balon et al., 1999b)suggest that a general similarity between the two systemscannot be assumed, even if correlations can be found forstructurally similar compounds. If partition is driven mainlyby hydrophobic interactions, correlations between the twosystems can be expected, but if electrostatics or hydrogenbonding play a significant part in the process, correlationsare likely to be weak.

Relatively few studies report on correlations betweenliposome partition and biological permeation.Lohmann

et al. (2002) reported on neither liposome–water noroctanol–water partition being able to successfully pre-dict blood–brain barrier permeation of drugs.Balon et al.(1999b)found no correlation between liposome distributioncoefficients and human intestinal absorption. A so-calledabsorption potential descriptor was subsequently calculatedfrom the liposomal distribution coefficient, solubility, doseand intestinal volume, and a sigmoidal correlation to HIAcould be established. This suggested that the dataset in-cluded compounds for which passive transcellular diffusionwas not (a) the transport mechanism or (b) the rate-limitingfactor in absorption, but active transport was taking place.Absorption potentials based on octanol–water distributioncoefficients did, however, not yield significant correlationswith HIA.

Liposomes have been captured on biosensor chips forthe determination of drug–membrane interactions (Danelianet al., 2000). The detection method employed was surfaceplasmon resonance, which detects the change in the refrac-tive index at the sensor surface caused by the interaction,and thus does not require the compounds to be radiolabelledor contain chromophores. The majority of drugs with hightranscellular absorption could be identified on basis of thebinding assay, however it was pointed out that the bindingobserved by the sensor may not always be related to trans-cellular transport.

In conclusion, despite offering a more biological partitionenvironment than octanol, the liposome system suffers fromsimilar drawbacks in terms of time and effort. To increaseexperimental efficacy, a fast pH-metric titration method hasbeen developed (Avdeef et al., 1998; Balon et al., 1999a).A high-throughput development of the conventional lipo-some/water system for drug screening does, however, seemunlikely. The partition coefficients measured in the liposomesystem may not always reflect biological permeation, butdrug interactions with the membrane surface. Nevertheless,information of such interactions could have implications fordrug-induced membrane effects (Grinius et al., 2002).

3.3. Chromatographic membrane phases

Reverse phase liquid chromatography (RPLC) was ini-tially introduced to improve the static octanol–water method,with advantages such as increased speed of determination,ease of automation, small sample amount and insensitivitytowards impurities (Pagliara et al., 1995; Kaliszan et al.,1993). The partition coefficient is easily calculated from theretention time of the solute in the RPLC column. In RPLC,drug partition has mainly been studied using silica-boundalkylsilanes (typically octadecylsilica, ODS) as stationaryphases. The stationary phase offers a more ordered organicphase than the octanol–water method, but lacks a polar re-gion and retention is thus expected to reflect only hydropho-bic interactions. Solvatochromic analysis of capacity factorshave confirmed the similarity of the RPLC C18 retentionwith octanol–water partition (Abraham et al., 1997).


The main drawback of the RPLC technique is the vari-ability of commercial columns and sensitivity towards ex-perimental conditions, which have prevented the establish-ment of a universal logP scale. Other shortcomings are alow chain density found in many commercial columns andthe limited pH operating range of silica-bonded phases. Inaddition, despite end-capping, silica-based chromatographicsurfaces still contain free silanol groups that may affect so-lute retention. The use of polymer-based stationary phaseshas been considered, but their influence on the retention timeis not clear and furthermore they are expected to show lessstability towards hydrodynamic pressure (Law et al., 1992;Dorsey and Khaledi, 1993).

In recent years, with efforts to improve the biological pre-dictivity of the octanol–water system, chromatography hasapproached the problem by introducing immobilised artifi-cial membrane (IAM) surfaces (for reviews on the techniqueseeTaillardat-Bertschinger et al., 2003; Stewart and Chan,1998; Yang et al., 1996). IAMs are constructed of phos-pholipid analogues, which are covalently bound by one oftheir alkyl chains to silica particles at high molecular sur-face densities, thereby mimicking fluid phase phospholipidbilayers (seeFigs. 8 and 9). In comparison with fluid lipo-somes, the density of the lipid headgroups is smaller in IAMsurfaces (Ong et al., 1996) and the hydrocarbon chains aremore ordered due to immobilisation. Molecular dynamicssimulations (Sheng et al., 1995) have indicated that the po-lar interfacial region of the IAM surface mimics closely thatof fluid membranes, which has further supported the use ofIAMs as substitutes for liposomes in drug partition studies.The main problems associated with IAM chromatographyare similar to RPLC: column variability, the instability aswell as the silanophilic interactions of the silica support, andthe limited pH-range (Kaliszan et al., 1993; Caldwell et al.,1998; Ottiger and Wunderli-Allenspach, 1999). Guidelineshave been proposed for carrying out proper measurementsof capacity factors on IAM phases (Taillardat-Bertschingeret al., 2002a).

For relatively lipophilic compounds, the use of purelyaqueous mobile phases results in long retention times. It istherefore common to add an organic modifier to the mo-

Fig. 8. Schematic structure of (A) a unilamellar liposome and (B) anIAM particle, in which a phospholipid monolayer is covalently bound toa silica particle. Reproduced fromTaillardat-Bertschinger et al. (2003)with permission from the American Chemical Society.

Fig. 9. Schematic of an IAM phospholipid (esterIAM .PCC10/C3). Repro-duced fromOng and Pidgeon (1995)with permission from the AmericanChemical Society.

bile phase in order to accelerate elution. The most popu-lar co-solvents are methanol and acetonitrile. Methanol isthought to be a more recommendable co-solvent due to itsmore water-like solvent properties. In addition, the presenceof acetonitrile at concentrations above 30 wt.% disrupts thestructure of water (Marcus and Migron, 1991). However, sta-tionary phases end-capped with methyl glycolate are liableto methanolysis when methanol is used in the mobile phase(Rhee et al., 1994). In order to yield comparable partitioncoefficients, the capacity factors obtained in the presence ofan organic modifier has to be extrapolated to 100% aque-ous mobile phase. In this process, it should be kept in mindthat the presence of the modifier affects the pH and ionicstrength of the solution as well as the apparent pKa of thesolute. (Taillardat-Bertschinger et al., 2002a, 2003).

Results from IAM chromatography have been comparedwith octanol–water (Kaliszan et al., 1994; Ong et al., 1996;Barbato et al., 1997; Amato et al., 2000; Valko et al., 2000)and liposome–water partition (Ong et al., 1996; Ottiger andWunderli-Allenspach, 1999; Taillardat-Bertschinger et al.,2002b) for various solutes. Contradictions in the compar-isons suggest that the balance between the various molec-ular interactions that affect partition differs in the threesystems. Studies on the thermodynamics of partition sup-port this conclusion (Ong and Pidgeon, 1995; Betageri andRogers, 1987). Structurally-related solutes, which partitionmainly as a result of hydrophobic interactions, typicallyyield comparable results in the above mentioned partitionsystems. When polar (hydrogen-bonding) and electrostatic


interactions intervene, correlations tend to break down. Par-tition of charged and hydrogen-bonding solutes is typicallylarger into liposomes than IAM phases, which is likely toresult from the higher density and mobility of the polar headgroups in liposomes (Taillardat-Bertschinger et al., 2002b).It is, however, not clear whether partition actually takes placeinto the hydrocarbon region or only to the interfacial regionof the lipid head groups. In the latter case, this would ex-plain the lower partition of charged and hydrogen-bondingsolutes into octanol than liposomes and IAM, as the hydro-carbon interior of lipid mono- and bilayers should not offera more superior environment for charged species, and lesshydrogen-bonding sites than octanol. However, if partitioninto the lipid-containing phases actually takes place into theacyl chain region, a plausible explanation could be foundin the concentrating effect of the headgroup–solute interac-tions, and, in the case of liposomes, a withdrawal effect ofthe lipid bilayer, as described in Section 2.2.3, and subse-quent transfer to the liposome interior.

IAM capacity factors have also been compared withCaco-2 cell monolayer permeation, yielding only weakcorrelations (n = 11, r2 = 0.58, Ong et al., 1996; n = 9,r2 = 0.32,Stewart et al., 1998). In the latter case, correla-tions improved significantly when a molecular weight anda hydrogen-bonding term were included in the linear re-gression. Comparison between rat intestinal absorption andIAM resulted in correlations ofr2 = 0.63 (n = 12; Onget al., 1996) and r2 = 0.77 (n = 12; Genty et al., 2001).Again, correlations improved with the addition of a molarvolume term.

IAM chromatography has furthermore been evaluatedfor prediction of drug permeation across the blood–brainbarrier. One group found that BBB partitioning of a set ofstructurally and electrically diverse drugs was only weaklycorrelated with IAM retention or octanol–water partitioncoefficients, correlations were improved when size andionisation was taken into account (Salminen et al., 1997).In another study, IAM chromatography was found to besuperior over octanol–water partition and ODS chromatog-raphy in predicting BBB partition of polar and ionisablecompounds (Reichel and Begley, 1998).

Recently a novel method to construct IAM stationaryphases has been introduced.Krause et al. (1999)report onthe use of noncovalent IAM surfaces, which are constructedby dynamic coating of a reversed-phase C18 column withphospholipids, thereby resulting in a bilayer structure. Im-provements in comparison with the covalent IAM phasesinclude a wide selection of lipid matrices and a morebiomimetic partition phase. Column stability was found to besufficient for multiple column runs in the presence of reason-able eluent concentrations. Excellent correlations (n = 13,r2 = 0.92) were found between peptide binding to phos-pholipids vesicles and noncovalent IAM retention times.

A similar approach was taken byLoidl-Stahlhofen et al.(2001). Their study employed solid-supported lipid mem-branes (SSLM) composed of lipid bilayers noncovalently

immobilised on silica beads. The membrane particleswere incubated with the solutes of interest on microtiterplates for 2 min. Subsequently, the particles were separatedby filtration and the solute concentration in the aqueousphase was analysed with HPLC. Good correlations be-tween the obtained membrane affinity (MA) coefficientsand octanol–water partition coefficients were observed forneutral compounds, whereas membrane affinity of chargedsolutes was found to correlate well with liposome–waterdistribution coefficients. The technique was found to behighly reproducible and automatisable.

The work by Beigi and co-workers (Beigi et al., 1995,1998; Lundahl and Beigi, 1997) combines the IAM ap-proach with liposome partition to immobilised liposome andbiomembrane chromatography (ILC and IBC) wherein vesi-cles are immobilised in gel beads. The approach avoids thestability problems of silica-bonded lipid phases and the useof organic, non-physiological solvents as eluents. In addi-tion, the membrane structure is more biomimetic comparedto IAM lipid phases and allows for the incorporation of pro-teins.

Apparent (both ionised and neutral species) specific ca-pacity factorsKs obtained with immobilised egg PC lipo-some chromatography were found to exhibit good linear cor-relations with octanol–water distribution coefficients (Beigiet al., 1995; Palm et al., 1998). Ks is defined as

Ks = tr − t0

t0(3.1)

wheretr = retention time;t0 = column dead time. Sigmoidalcorrelations between IC capacity factors and human intesti-nal absorption were established (Beigi et al., 1995; Liu et al.,2002) allowing for qualitative identification of drugs withfavourable absorption properties, as shown inFig. 10a. In astudy byPalm et al. (1998), the presence of a large and flex-ible compound significantly weakened the correlation be-tween apparent capacity factors and Caco-2 cell monolayerpermeation (Fig. 10b). It was pointed out that the ILC re-tention time results from a combination of partition into theliposome interior, partition into the lipid bilayer and elec-trostatic solute–lipid surface interactions, and that retentionmay therefore not always reflect transmembrane diffusion.

Conclusively, IAM and IL chromatography offer an au-tomatisable system for rapid screening of drug compounds,although not really in HTS scale. However, similarly to theliposome system, the capacity factors obtained from thesemay in some cases not reflect biological permeation, but druginteractions with the membrane surface. Perhaps as a resultof this, correlations of IAM indices with biological perme-ation have often improved in combination with other descrip-tors. A proper validation of the IAM approach would requiretesting with a large number of compounds. Often, only a lim-ited number of drugs within a fairly narrow structural rangeare studied. Recent studies suggest that chromatographic ca-pacity factors may find use in membrane activity studies ofdrugs (Ollila et al., 2002; Krause et al., 1999).


Fig. 10. Apparent (includes both ionised and neutral species) specific capacity factors from immobilised liposome chromatography vs. (a) human intestinalabsorption. Compound list in Appendix. Reproduced fromBeigi et al. (1995)with permission from Elsevier. (b) Caco-2 cell monolayer permeability.Reproduced fromPalm et al. (1998)with permission from the American Chemical Society.

3.4. Cell culture monolayers

In recent years, the use of cell cultures to predict drug per-meation in biomembranes, such as intestine (Artursson et al.,2001), skin (Suhonen et al., 2003) and cornea (Toropainenet al., 2003) has gained popularity. The most frequentlyused cell cultures for studies of passive drug transport arethe Caco-2 cell cultures, which are derived from humancolon cancer cells (human adenocarcinoma colon cells). TheCaco-2 cells can be cultivated to spontaneously differentiateto form monolayers of polarised cells, with functions simi-lar to intestinal enterocytes. The monolayers are grown onfilter supports and drug passage from the donor to the ac-ceptor compartment is measured. Transport studies throughthe Caco-2 monolayers can provide information on perme-ability coefficients, transport mechanisms and pathways, andmetabolism of drug compounds (Audus et al., 1990). An-other advantage of studying biological permeation with cellmonolayers is that they measure the transport of the drugacross the cell membrane, instead of just its interaction withthe lipid bilayer.

Despite their undeniable benefits as model membranes,also the approach of using cultured cell monolayers comeswith certain limitations. The method is rather laborious andtime-consuming as cells have to be cultured for approxi-mately 3 weeks prior to use and during the experimentssamples are collected at time intervals of hours. To increaseexperimental efficacy alternative cell lines have been intro-duced, requiring less than a week of culture (Irvine et al.,1999; Tavelin et al., 2003a), and an automatic liquid handlingsystem has been reported to increase capacity and signifi-cantly decrease the need for manpower during experiments(Garberg et al., 1999).

Despite the ability of Caco-2 cell cultures to predictthe oral drug absorption, in particular for well absorbed

compounds, the comparisons of Caco-2 data from differ-ent laboratories has revealed discrepancies, which arisefrom differences in cell culturing methods, experimentalconditions (such as temperature, pH-gradient and hydro-dynamics of the system) as well as the source of Caco-2cells (Yee, 1997; Artursson et al., 2001; Walter and Kissel,1995). An example of this is seen inFig. 11, where Caco-2permeation data from different laboratories is plottedagainst absorbed fraction in humans. Although a quali-tatively similar correlation is observed in all cases, theassumption of one common Caco-2 permeation scale couldlead to misunderstandings. To correct for intra- as wellas interlaboratory variability, the cell monolayers can becharacterised with respect to their electrical resistance aswell as permeation of reference compounds (Larger et al.,2002).

In addition to the interlaboratory variability in Caco-2 per-meability, it is also known that the expression of both theparacellular and active transport routes, as well as efflux sys-tems, is different in Caco-2 cell lines compared to the humanintestine (Artursson et al., 2001). Despite differing levels intransporter expression, Caco-2 cell cultures can be used inthe identification of the role of transporters in the permeationof drug candidates. Caco-2 cell permeation has also gainedregulatory acceptance as part of the biowaivers based on theBiopharmaceutical Classification System (BCS) (Lobenbergand Amidon, 2000).

Alternative cell lines that better mimic the human smallintestine are continuously under development. One recentlyintroduced cell line is 2/4/A1, which exhibits similar passive(paracellular and transcellular) absorption characteristics asthe human jejunum (Tavelin et al., 2003a,b). Due to its lackof functional expression of active transport and efflux sys-tems, the 2/4/A1 cell line holds promise for characterisationof the passive absorption of drugs.


Fig. 11. Fraction absorbed (FA) in humans as a function of Caco-2 permeation data from different laboratories. The fitted lines represent the equationFA = (0–100)/(1+(logPapp/a)b )+100, wherea = logPapp at FA = 50%. Reproduced fromArtursson et al. (2001)with permission from Elsevier.

In a commentary,Artursson and Borchardt (1997)list sev-eral technical improvements that are required for cell culturemonolayers to become an integral part of drug screening.These include miniaturisation of the cell culture apparatus;automation of the transport experiments; adaptation of theassays to handle poorly soluble compounds and complexformulations; and connecting them to effective and sensi-tive analytical detection apparatus. In addition, the authorsstress the importance of developing standardised methodsfor quantification, analysis, storage and retrieval of the dataaccumulated in the experiments.

3.5. Artificial membranes

Kansy et al. (1998) have proposed the use of a par-allel artificial membrane permeation assay (PAMPA) asa high-throughput alternative to Caco-2 monolayers forthe prediction of passive drug absorption. In the PAMPAapproach, the aqueous donor and acceptor compartments(96-well microtiter plate) are separated by a hydrophobicpermeability filter (96-well microtiter filterplate). The filteris impregnated with an organic solution of lipid, whichforms bilayer structures in the filter pores. The solute con-centrations in the acceptor well are determined in parallelby UV spectrophotometry using a 96-well microplate pho-tometer. The authors found that the PAMPA flux could besuccessfully used to classify compounds of low, intermedi-ate and high human intestinal absorption.

Avdeef et al. (2001)automated the PAMPA approachfurther into a filter-immobilised artificial membrane(filter-IAM) assay. In this approach both acceptor anddonor well concentrations were analysed att = 0 and t= permeation time with UV spectrophotometry to deter-mine membrane retention of the drugs.

In a study byZhu et al. (2002), an artificial membranepermeability (AMP) assay was constructed on a hydrophilicmembrane support, thereby shortening the permeation timesto 2 h. The artificial membrane permeabilities were com-pared with human intestinal absorption as well as with boththeoretical and experimental models thereof. The compar-ison with AMP and human intestinal absorption for a setof 93 commercial drugs (Fig. 12a) suggested that the arti-ficial membrane system could be used in classifying drugsinto different absorption categories. Correlating logP andlogD7.4 with human intestinal absorption for the same set ofdrugs resulted in a much more scattered plot than for AMP(Fig. 12b). Even if the correlation between the AMP and theCaco-2 monolayer assays was only reasonable (n = 49, r2

= 0.67), the systems were found to be highly comparable inabsorption classification.

The structure and repeatability of the filter-immobilisedlipid–solvent membranes in the above-described assays havebeen questioned (Wohnsland and Faller, 2001). Indeed, elec-trochemical studies of such filter systems have shown atime-dependent thinning of the micromembranes to occur(Ikematsu et al., 1996). In the high-throughput assay pro-posed byWohnsland and Faller (2001)the acceptor anddonor compartments of the 96-well titer plate are sepa-rated by a liquid hexadecane layer immobilised by a fil-ter of 9–10�m thickness. The liquid alkane was chosenfor its ability to model the essential properties for biologi-cal permeation, i.e. lipophilicity and hydrogen bonding po-tential. The concentrations of the acceptor compartmentswere analysed with UV detection after a 5 h permeationtime. Human gastrointestinal absorption was significantlybetter predicted by the liquid membrane permeation than byoctanol–water distribution coefficients. The permeation as-say was also used to construct pH-permeation profiles of


Fig. 12. (a) Artificial membrane permeation (AMPPapp) vs. fraction absorbed in humans for 93 commercial drugs. (b) logP and logD7.4 vs. fractionabsorbed in humans for 86 commercial drugs. Circled compound groups are outliers. Reproduced fromZhu et al. (2002)with permission from Elsevier.

the ionisable compounds, which could in turn be employedto calculate alkane–water partition coefficients and accountfor the unstirred water layer effect in the permeation coeffi-cients for the highly permeable solutes.

In conclusion, the filter-immobilised membrane approachprovides certain benefits over the Caco-system in termsof access to a wider pH-range, ease of automation, highthroughput and lower cost. Although the lack of paracellularand active transport mechanisms found in Caco-2 mono-layers may be considered a drawback, this, on the otherhand, enables studies of exclusively passive transcellulardiffusion without intervening processes. Further studies ofimmobilised membrane permeation are required before def-inite conclusions can be drawn concerning its applicabilityin drug screening.

4. Ion partition—a task for electrochemistry?

4.1. Principles

Partition and permeation of ionic drugs has received rel-atively little attention. In 1992, we presented the use ofliquid–liquid electrochemistry as a fast, convenient and accu-rate means to determine ionic partition coefficients (Kontturiand Murtomäki, 1992) (for more detailed information onliquid–liquid electrochemistry, see for instanceGirault andSchiffrin, 1989; Senda et al., 1991; Vanýsek, 1995). The ba-sic equation is the Nernst equation that relates the activityof a charged speciesi in water and oil phase to the potentialdifference across a phase boundary,�w

o φ:

�wo φ = �w

o φ0i + RT

ziFln

aoi

awi

(4.1)

zi is the charge number ofi and�wo φ0

i its standard transferpotential between the phases, which is related to the standard

Gibbs free energy of transfer,�owG

0,Ii , and therefore to the

partition coefficient:

logP Ii = − �o

wG0,Ii

2.303RT= − zF�w

o φ0i

2.303RT(4.2)

Since it is not possible to experimentally determine ionic freeenergies of transfer, or any thermodynamic ionic quantities,one commonly uses the so-called TATB assumption, whichstates that due to similar structure and electrostatics, thestandard free energies of transfer of the tetraphenylarsoniumcation (TPAs+) and the tetraphenylborate anion (TPB−) areequal (Parker, 1969):

�wo G0

TPAs+ = �wo G0

TPB− = 12�w

o G0TPAsTPB (4.3)

Based on the TATB assumption, a Galvani potentialscale can be established, and a database of standard trans-fer potentials of electrolytes is available on the internet(http://dcwww.epfl.ch/cgi-bin/LE/DB/InterrDB.pl). A po-larisable liquid–liquid interface is formed by placing anorganic solvent containing a hydrophobic electrolyte in con-tact with an aqueous solution of a hydrophilic electrolyte.The solvents should be only sparingly mutually solubleand the electrolytes hydrophobic and hydrophilic enoughso that a potential region, where no ion transfer occurs,is obtained (Kontturi and Murtomäki, 1992). As n-octanoldoes not dissolve electrolytes, other solvents must be used;the most common choices are 1,2-dichloroethane (DCE),nitrobenzene ando-nitrophenyl octyl ether. The choice ofsolvents will be discussed later on.

Upon addition of the drug of interest into the aque-ous phase, the potential difference across the interface isscanned. The drug is now transferred from the aqueous tothe organic phase, which is seen as electric current. Duringthe reverse scan, the drug returns back to the aqueous phase.A voltammogram is obtained (Fig. 13) from which the halfwave potential,�w

o φ01/2, is determined approximately as

http://dcwww.epfl.ch/cgi-bin/LE/DB/InterrDB.pl


Fig. 13. A cyclic voltammogram and the half wave potential.

the midpoint of the forward (Epf ) and backward (Epb) scanpeak potentials. The standard transfer potential of ioni isthen obtained as (Samec et al., 1996):

�wo φ0

i = �wo φ1/2 − RT

ziFln

γoi

γwi

− RT

2ziFln

Dwi

Doi

− RT

ziF

× ln

[1 + Kaαoco(γo)2

(Dip

Doi

)1/2]

(4.4)

where�i and Di are the activity and diffusion coefficientof i in the aqueous or oil phase. The expression within thebrackets refers to ion association between the transferringion and the counter ion in the oil phase:Ka is the ion pairassociation constant,�o is the degree of association of theoil phase base electrolyte,co and�o are its concentration andactivity coefficient, andDip is the diffusion coefficient of theion pair in the oil phase. Association in the oil phase is oftenfound to be negligible so that the last term can be ignored.The activity coefficient and diffusion coefficient terms arealso rather small (∼5 mV) and can usually be ignored.

A typical experimental setup in a liquid–liquid electro-chemical measurement is:

where XCl, XY and MCl are the base electrolytes in the ref-erence, organic and aqueous phases, respectively. The dou-ble line between the aqueous and the organic phases indi-cates the polarisable interface under study. Due to the lack ofsuitable reference electrodes for the organic phase, a sepa-rate reference phase (w′) is often used. The Galvani potentialacross the polarisable interface,�w

o φ, is obtained subtract-ing the potential of the reference phase from the measuredcell potential,E, as described in the literature (Girault andSchiffrin, 1989).

4.2. Ionic versus neutral partition coefficients

In the electrochemical method, the partition coefficientof an ionised species is measured, and the question ariseshow it is related to the partition coefficient of an electricallyneutral drug. The free energy of an ion consists of an elec-trostatic part and a neutral part; the former part is naturallyzero for a non-charged molecule. Assuming that charging amolecule by, for instance, protonation, leaves the molecularstructure practically intact, subtraction of the electrostaticpart,�o

wG0i,es, from the ionic free energy of transfer yields

the free energy of transfer of the neutral molecule,�owG

0,Ni :

�owG

0,Ni = �o

wG0,Ii − �o

wG0i,es (4.5)

One approach to estimate the contribution of electrostaticsis to use theoretical solvation models, such as the onesbased on the Born equation, presented in (2.7) and (2.8).As opposed to these models, where the electrostatic termof the transfer energy describes long-range ion–solvent in-teractions, Osakai et al. (Osakai and Ebina, 1998; Osakaiet al., 1997) proposed a semi-empirical theory, where itis governed by specific short-range interactions such asdonor–acceptor effects or hydrogen bonds. It was con-cluded that the electrostatic part of the Gibbs free energy oftransfer could be described by a quadratic function of thesurface electric field strengthE:

�owG0

i,es

4πri2

= �A′ + �B′E + �C′E2 (4.6)

where

E = zie

4πε0r2i

(4.7)

The coefficients ofEq. (4.6)were determined by regres-sion analysis of the�o

wG0i,es values, which were calculated

from Eq. (4.5)using experimental Gibbs energies of trans-fer and the UhligEq. (4.8)for the cavity formation energy.As ionic radii needed inEqs. (4.6) and (4.7)depend onthe extent of hydration, different data sets were used for

nonhydrated and hydrated cations and anions as well aspolyanions.

�owG

0,Ni = −4πNAr2

i γ (4.8)

whereγ is the interfacial tension between the aqueous andthe organic phase.

The second order polynomial fits of�owG0

i,es as a func-tion of E yielded excellent correlation coefficients between0.931 and 0.999.Osakai and Ebina (1998)concluded that ig-noring long-range electrostatic ion–solvent interactions may


be justifiable if one realises that the Gibbs free energy oftransfer from one phase to another is the difference in thesolvation energies between the two phases. It is thus prob-able that the electrostatic solvation energies in the respec-tive phases are for the most part cancelled out in�o

wG0i,es,

whereas short-range interactions remain. The fits presentedby Osakai et al. are, of course, only valid for transfer be-tween the solvents used in their study, namely water andnitrobenzene. For any other solvents, new fits have to becarried out, which makes the approach a bit tedious unlessdata is already available.

In our previous work (Kontturi and Murtomäki, 1992), westudied protonated drugs, DH+, and calculated the partitioncoefficient of the neutral drug by subtracting the contribu-tion of the proton. A slightly corrected analysis based onthe complete thermodynamic cycle was presented recently(Murtomäki and Kontturi, 2002). Conclusively, the Gibbsfree energy of transfer of the neutral drug can be written as:

�owG0

D = �owG0

DH+ − �owG0

H+ − 2.303× RT

× (pKwa − pKo

a) (4.9)

where the last correction term includes the pKa values ofthe drug in water and oil; values for the free energy oftransfer of the proton,�o

wG0H+ , are available in literature

(http://dcwww.epfl.ch/cgi-bin/LE/DB/InterrDB.pl).

4.3. Potential–pH diagrams

Two co-operating research groups in Lausanne combinedliquid–liquid electrochemistry with conventional bulk par-tition methods and constructed complete potential–pH di-agrams at liquid–liquid interfaces for a number of ionis-able drugs (Reymond et al., 1996a, 1999a; Gobry et al.,2000). These diagrams, which resemble the ones developedby Pourbaix to study corrosion of metals, give the completepicture of the effect of both the pH and an interfacial poten-tial difference on drug partition.

The diagram is constructed on the basis of the ther-modynamic cycle of the drug in a biphasic system. Allof the species appearing in the cycle are bound togetherby three different types of equations, which are used todefine the boundary lines between the predominance ofeach species in their respective phases. These equationsare: the NernstEq. (4.1), which describes the partition ofan ionic species between the organic and aqueous phase;Eq. (2.2), which defines the partition coefficient of a neutralspecies and its relation to the Gibbs energy of transfer; andequations which describe acid–base equilibria, such as theHenderson–Hasselbalch equation.

We have recently studied a series of aminoacridinederivatives with varying lipophilicity (Malkia et al., 2003).The half wave potentials in the water–nitrophenyl octylether system were measured with cyclic voltammetry as afunction of pH of the aqueous phase. As an example, thepotential–pH diagram of the topical antiseptic agent ami-

Fig. 14. Ion partition diagram of aminacrine showing the variation in theformal transfer potential with aqueous pH. The circles indicate experi-mental values, and the solid line represents the fitted equiconcentrationlines between the dominant forms of the drug D.

nacrine (9-aminoacridine) is presented inFig. 14. The pKaof aminacrine is 9.99, thus below pH 10 its dominant formin water is protonated, and above pH 10 electrically neutral.

At low pH, the protonated form DH+ is in equilibriumbetween the phases, and the horizontal line represents theirequiconcentration in the two phases. At high pH, the neutralform D is in equilibrium between the phases, but as thepartition of the neutral species is independent of potentialand pH, only the neutral species in one of the phases canappear in the diagram. Typically, the phase for which thecompound expresses the higher affinity is chosen. The risingline 60 mV/pH unit corresponds to the proton transfer acrossthe phase boundary:

D(o) + H+(w) ↔ DH+(o)

Similar assisted proton transfer by the anti-ischemic drugtrimetazidine (Reymond et al., 1999b), the antihistamine ce-tirizine (Bouchard et al., 2001b) and piroxicam (Reymondet al., 1996b) has been reported in literature.

The pH-potential diagram can be used to measure thelogP of the neutral species, as indicated inFig. 14. The crosspoint of the extrapolated horizontal and rising lines differsfrom the aqueous pKw

a value by the quantity log(1+ PD/ξ)whereξ = (Dw/Do)1/2, the square root of the diffusion coeffi-cient ratio in the two phases. For aminacrine, log(PD,NPOE) =2.37 as measured fromFig. 14, while the calculated valuefrom the Born model gives 2.56 (Malkia et al., 2003). Themeasured value in the water–octanol system is log(PD,oct) =2.74 (Drayton, 1990).

Recently (Gobry et al., 2001), potential–pH diagramswere generalized to apply to an arbitrary phase volume ratio,r = Vo/Vw, using rather than equiconcentration lines, lineswith an equal number of moles in the adjacent phases. Forvery lipophilic solutes the ratior should be large, and forvery hydrophilic solutes small. This way, the applicabilityof voltammetry can be expanded significantly. The Nernst

http://dcwww.epfl.ch/cgi-bin/LE/DB/InterrDB.pl


Fig. 15. The relation of logPoctanol and logPDCE. Data from Kontturiand Murtomäki (1992).

Eq. (4.1)also implies that the interfacial potential differencecan be adjusted by adding salt with a common ion into thewater and oil phase. The Lausanne groups (Bouchard et al.,2001a) demonstrated that the apparent increased partition ofionised drugs due to addition of lipophilic salt is not due tothe formation of lipophilic ion-pairs, but the shift of the Gal-vani potential difference caused by the added electrolytes.The electrochemical pH–potential approach was also foundto be a useful tool to study the partition behaviour of zwit-terionic drugs (Bouchard et al., 2002; see Section 2.2.3).

4.4. Solvent problem

In bulk partition systems,n-octanol has the status of astandard. Unfortunately, electrochemistry is very trouble-some inn-octanol, as it does not dissolve electrolytes andthus its conductivity is practically zero. The only reportknown to us is byGulaboski et al. (2002)who attached adecamethyl ferrocene (DmFc) containg droplet ofn-octanolon the graphite electrode and immersed it in an aqueoussolution. Upon oxidation of DmFc on graphite, anionswere transferred inton-octanol at the graphite-oil–waterthree-phase junction to maintain electroneutrality. In theelectrochemical literature, there are signs of this approachbecoming more common.

In our previous work (Kontturi and Murtomäki, 1992),the partition coefficients obtained at the water-1,2-dichloroe-thane interface were plotted against the water–n-octanolones (seeFig. 15). It appeared that the drugs fell in twocategories, depending on their hydrogen bonding ability.Molecules able to form hydrogen bonds, viz.β-blockers inFig. 15, fell on their own line, while the three others with-out hydrogen bonding capacity formed another straight line.Taking this factor into account, the order of lipophilicity ex-pressed as logPoctanol was retained in DCE for this set ofdrugs.

The Lausanne groups (Caron et al., 1999) investigatedthe effect of solvent properties on the partition of a series

of β-adrenergic receptor agents (β-blockers) in both theirneutral and ionised forms. The obtained�logP values forthe different solvents were found to mainly reflect differ-ences in hydrogen bonding. The majority of the neutralβ-blockers expressed poor hydrogen bond donor characterin 1,2-dichloroethane, which was attributed to the formationof an internal hydrogen bond in these compounds. In bothdibutyl ether (DBE) andn-octanol, internal hydrogen bond-ing was not favoured among the neutral species of the drugs,whereas cationicβ-blockers bearing anortho-O atom wereobserved to form internal H-bonds also in the octanol–watersystem.

The difference between partition coefficients of neutraland charged species in each solvent, diff(logPN–I), was in-troduced to describe the effects of charge localisation onion partition. Thus, the significantly larger diff-values ob-tained in the dichloroethane–water system compared to theoctanol–water system, were suggested to reflect the localisedcharge of these monocations, which causes the ions to retainmore water molecules when transferring into octanol thanto DCE. This, on the other hand, indicates the greater wa-ter solubilisation and hydrogen bonding capacity of octanolcompared to DCE (Caron et al., 1999). In another study(Steyaert et al., 1997), a solvatochromic analysis of a set of44 solutes was carried out, and it was suggested that DCEcould actually be a better solvent thann-octanol to predictbiodistribution of solutes.

4.5. Biomimetic liquid–liquid interfaces

In our recent work (Liljeroth et al., 2000; Mälkiä et al.,2001a,b; Malkia et al., 2003), we introduced a novel ap-proach for using liquid–liquid electrochemistry to studymembrane activity of ionised drugs. We employed theLangmuir–Blodgett technique (Petty, 1996; Petty andBarlow, 1990) to deposit a lipid monolayer at the interfacebetween an aqueous solution and a NPOE–PVC gel, thusimproving the biomimetic properties of the conventionalliquid–liquid interface. Although the presence of a mono-layer was not found to inhibit ion transfer, a distinct effectcould be seen in the apparent standard rate constant uponintroducing the monolayer. Furthermore, combining the ex-perimental results with a theoretical model, it was possibleto estimate the strength and site of ion adsorption on themonolayer. Differences in adsorption behaviour were foundeven between structurally similar drugs (Malkia et al.,2003), which was attributed to small variabilities in theirhydrogen bonding ability.

4.6. Conclusions

Liquid–liquid electrochemistry presents an important ad-vantage compared to conventional bulk experiments in thepossibility to study the effect of a potential difference onion transfer, thereby mimicking the situation in biologicalsystems. In addition, liquid–liquid electrochemistry can be


employed to investigate the effect of charge on lipophilic-ity and to determine the hydrogen bonding characteristicsof the compounds under study. An additional benefit of themethodology is the ability to distinguish between the trans-fer of differently charged species of the same compound(Reymond et al., 1996a). Furthermore, it is possible to mod-ify the liquid–liquid interface to resemble one half of a lipidbilayer and conduct transport studies through this modelmembrane.

An obvious disadvantage is that electrochemistry is nota very widely known discipline and, therefore, the thresh-old to take it into use might be difficult to overcome. Apractical problem is the need for reasonable conductivity inboth phases which requires the addition of supporting elec-trolytes. The effect of these supporting electrolytes needs tobe subtracted in the analysis. Furthermore, electric currentis an integral value over the contributions of all the trans-ferring species, i.e. there is also a selectivity problem to betaken care of. For screening purposes, the method shouldbe expandable for the parallel measurement of thousands ofcandidates, which may require rather a complicated instru-mentation.

5. Computational approaches

In addition to the experimental approaches discussedabove, an increasing number of studies focus on predictingmembrane permeation of drugs from calculated molecu-lar properties. The aim of the computational approachesis to save resources by enabling virtual screening of drugcandidates and the identification of drug candidates withpossible absorption problems at an early stage of drug de-velopment. In addition to saving time and effort, this canalso help reduce the amount of animals used in testing. Themain factor enabling the rapid growth of computational ap-proaches has been the development in computer resourcesand calculation algorithms.

The basis of computational prediction of drug partitionor permeation lies in quantitative structure–property rela-tionships (QSPR). The concept of QSPR is to transformthe chemical intuition and experience used in the search forcompounds with desired properties into a mathematicallyquantified and computerised form. In other words, QSPRmethods for predicting biological permeation seek to trans-form the chemical structure of a compound into a set of nu-merical descriptors of the properties relevant to its perme-ation, and to establish the quantitative relationship betweenthese descriptors and the permeation of the compound. Oncea correlation between structure and property is found, thisknowledge can be used to screen other compounds, includ-ing ones not yet synthesised, on the computer in order toselect structures with the desired properties for further trials(Karelson, 2000).

To obtain a significant correlation, it is essential to useappropriate data in building and testing the model. As dis-

cussed byEgan and Lauri (2002), many existing models arebuilt with data that suffers from one or several of the follow-ing problems: too small sample size; bias of data towardshighly permeable compounds; use of data on actively trans-ported compounds when modelling passive transport; use ofdata for which other factors such as solubility, and not pas-sive transmembrane diffusion, is the absorption rate-limitingstep; and a general lack of data quality, meaning poorly re-peatable experimental results and large standard errors. Fur-thermore, as discussed inSection 3.4in relation with Caco-2cell cultures: owing to variable experimental conditions, pro-cedures and materials, some interlaboratory variability willalways be present in experimental data. This should be keptin mind when building QSPR models, particularly whencompiling datasets from widely different sources.

Furthermore, when constructing the model, it is importantthat appropriate descriptors are employed, whether theoret-ical or experimental (Livingstone, 2000). For drug designpurposes, it is desirable that the descriptors reflect simplemolecular properties and provide easily interpreted informa-tion. When a compound is flagged by a computational screenas likely to exhibit poor intestinal absorption, this will savethe time and effort of synthesising and experimentally ex-amining that compound. But in order to provide feedbackfor the drug design chemists, the important question to ad-dress is “what structural properties caused the compound tohave poor intestinal absorption?”

Further attention needs to be given to the obtained model:a low standard error, a restricted number of descriptors, com-patibility with common scientific knowledge, and the abil-ity to correctly predict properties of compounds that werenot employed in the construction of the model, are some ofthe qualities of a proper QSPR model (Nirmalakhandan andSpeece, 1988). Several approaches exist to ensure model sta-bility and predictive ability, including cross-validation andvariable selection schemes (Wold, 1979; Norinder et al.,1997).

The first part of the section reviews some of the ap-proaches to calculate the octanol–water partition coefficient(a thorough review can be found byLeo, 1993), whereas thesecond part looks into models of biological permeation.

5.1. Calculating log P from molecular structures

5.1.1. Background and developmentThe oil–water partition coefficient was introduced by

Hansch and co-workers in the 1960s (Hansch et al., 1963;Hansch and Fujita, 1964; Fujita et al., 1964) and it is stillthe most popular descriptor linking drug structure to bio-logical permeation. In relation to the partition coefficient,Hansch et al. (1963)defined theπ-parameter, which allowsfor quantitative estimates of partition coefficients on basis ofwell-characterised “parent” compounds. Theπx-parameterrepresents the difference in logP between anx-substitutedand an unsubstituted compound:


logPx = logPH + πx (5.1)

The fragment method was introduced by Rekker andco-workers (Nys and Rekker, 1973). Instead of using par-ent compounds, they divided the compounds into simple,additive fragments as shown inEq. (5.2). The fragment con-tributions were determined on basis of statistical analysisof a large library of experimental logP values.

logP =∑

anfn +∑

bmFm (5.2)

wherean is the number of occurrences of fragmentf of typen, andbm is the number of occurrences of correction factorF of type m. The major corrections are related to whethera polar fragment is connected to an aliphatic or aromaticcarbon atom, and to the effect of two polar fragments situatedat close proximity.

A more general approach to predict the oil–water partitioncoefficient is the solvatochromic methodology (Taft et al.,1985a; Kamlet et al., 1983), also called the linear solvationenergy relationship approach, which states that solubility canbe related to solute–solvent interactions through the linearcombinations of three types of terms: the cavity formationterm V, the solute polarity/polarisability termπ∗ and theterms indicating the hydrogen bond acceptor (βH) and donor(αH) strength of the solute.

logP = aV+ bπ∗ + cβH + dαH + e (5.3)

The solvatochromic approach has proved successful incorrelations and predictions of various physicochemicalproperties that involve solute–solvent interactions (Kamletet al., 1986; Taft et al., 1985b) and the parameters containphysical meaning that can be directly related to the molec-ular structure. The inconvenience of the method is the needto assign fragment values to all the parameters and accountfor intramolecular fragment interactions, even if tabulatedparameter values and rules to calculate them exist (Kamletet al., 1983, 1988).

The above methods generally perform well for small,rigid molecules, but for larger and flexible compounds in-tramolecular interactions such as shielding and folding areoften not properly accounted for. The concept of molecularlipophilicity potential (MLP) was first introduced in the late1980s (Audry et al., 1986; Furet et al., 1988). The MLP isa structure–property descriptor that visualises the lipophilicproperties of the molecule on its three-dimensional (3D)surface. The most commonly used MLP methodology todayis the one developed byGaillard et al. (1994). In their ap-proach, the MLP is generated on the solvent-accessible sur-face (SAS) of the molecule and calculated on basis of a frag-mental description of lipophilicity and a distance function:

MLPk =N∑

i=1

fifct(dik) (5.4)

wherek is the index of a given point in space,i the index ofa molecular fragment,N the total number of fragments in

the molecule,fi the lipophilic constant of fragmenti, fct is adistance function anddik is the distance between fragmenti and pointk.

The MLP calculation begins with the generation of thethree-dimensional structure of the molecule. Selection ofseveral low energy conformations for the calculations canthus shed light on the relationship between conformationand lipophilicity. Such information may be particularlyuseful when modelling ligand–receptor interactions. Tothis end, the MLP has been successfully implemented inComparative Molecular Field Analysis (CoMFA) (Gaillardet al., 1996; Testa et al., 1996). A back-calculation to theone-dimensional logP value is also possible. After obtain-ing the MLP value at each point on the SAS of a molecule,a numerical integration over this space is performed. Theintegration yields two quantities: the total of positive MLPvalues, representing the lipophilic part of the molecule; andthe total of negative MPL values, standing for the polarpart of the molecule. These quantities are then used as in-dependent parameters in multilinear regression for a set ofcompounds with known experimental logP values in orderto establish the relationship between MLP and logP. Thistraining set equation can subsequently be employed forcompounds with unknown partition coefficients.

5.1.2. log P programsThe first, and still most popular commercial computer

program to calculate octanol–water partition coefficientsfrom molecular structure is ClogP, developed by thePomona MedChem Project (Chou and Jurs, 1979; Leo andHoekman, 2000; available from Daylight Corp., CA, USAand BioByte Corp., Claremont, CA, USA). The methodfollows the RekkerEq. (5.2), but the approach used toderive the constants is different. Whereas Rekker used a‘reductionist’ approach and derived the fragment valuesfrom statistical treatment of logP data without really spec-ifying what constitutes a fragment, ClogP is based on a‘constructionist’ approach. The fragments were first definedand then evaluated from as simple structures as possible,such as hydrogen gas, methane and ethane, to minimiseobscuring interactions. In cases where the use of thesefragment constants was incapable of yielding the measuredlogP value, the difference was defined in terms of universalcorrection factors. Many improvements have been madeto ClogP since its introduction, the most significant beingthe ability to estimate an unknown polar fragment valuewhich has not appeared in a solute for which a measuredlogP(oct) exists.

Many computerised algorithms for calculating partitioncoefficients have been developed since the introduction ofClogP, most of which are discussed in the recent reviews byCarrupt et al. (1997)andMannhold and van de Waterbeemd(2001). The methodologies can roughly be divided into (i)fragmental approaches such as ClogP, logKow (Meylan andHoward, 1995; available from Syracuse ResearchCorporation, Syracuse, NY, USA), ACD/logPDB (Spessard,


1998; available from Advanced Chemistry Development,Toronto, Canada) and KlogP (Klopman et al., 1994); (ii)atom contribution approaches, such as AlogP98 (Ghoseet al., 1998; available from Accelrys Ltd., Cambridge, UK);(iii) topological approaches such as MlogP (Moriguchiet al., 1992), VlogP (Gombar and Enslein, 1996; avail-able from Accelrys Ltd.); and (iv) a neural network study(Huuskonen et al., 2000).

An essential prerequisite for the use of computerised logPcalculators is the existence of a simple yet unambiguousway to feed the structural information into the programs.The SMILES notation (Weininger, 1988) is accepted bymany logP programs. SMILES strings are easily written di-rectly from knowledge of chemical structure. In addition,certain chemical structure drawing programs are able to cre-ate SMILES notations on basis of drawn structures. Anotherpopular input format is the MDL mol-file, described morethoroughly byDalby et al. (1992)and used by a variety ofother computational applications in chemistry.

5.2. Computational approaches to predict biologicalpermeation

Although the octanol–water partition coefficient has beenfound to correlate fairly well with the biological permeationof many drug compounds (Palm et al., 1996; Krämer, 1999)several studies also report on its failure (Young et al., 1988;Conradi et al., 1991; Palm et al., 1997). Good correlationsare generally obtained for homologous series of solutes,but when comparing structurally diverse compounds, vari-able contributions of hydrogen bonding and electrostaticsdisrupt the trend (Artursson and Karlsson, 1991; Goodwinet al., 2001). In an attempt to improve correlations, severalphysicochemical descriptors have been incorporated intoone model using multiple linear regression (MLR) analy-sis (Sugawara et al., 1998). Lobell et al. (2003)recentlyreported on a 34-descriptor model for the prediction oflogBB, which compared favourably with other correspond-ing models (ntrain = 48, r2

train = 0.84, q2train = 0.79, strain

= 0.19,ntest = 17, r2test = 0.68; wheren = number of com-

pounds, r2 = correlation coefficient,q2 = cross-validatedcorrelation coefficient,s = standard deviation). However,caution needs to be taken in the choice of variables, sincecorrelations often improve when the number of descrip-tors grows. Furthermore, many physicochemical propertiesare interrelated, which makes them unsuitable for multiplelinear regression analysis (Artursson et al., 2001).

One approach to circumvent the problem of collinearityof the descriptors has been the use of partial least-squaresmethods (PLS).Luco (1999)employed the PLS approachto model the blood–brain concentration ratio of 58 drugsusing topological and constitutional descriptors calcu-lated from the molecular structures of the compounds. Athree-component model comprising 18 descriptors was ableto successfully describe the blood–brain concentration ratio

of the training set (n = 58, r2 = 0.85,q2 = 0.75,s = 0.318).The most important descriptors reflected molecular polarity,size and hydrogen bonding potential.

Another approach to build multidescriptor models is theuse of neural networks.Wessel et al. (1998)employed aneural network routine in combination with a genetic algo-rithm to find the best descriptor subsets and subsequentlyto build a nonlinear 6-descriptor model for the prediction ofhuman intestinal absorption (ntrain = 67, RMSEtrain = 9.4%HIA; ncv = 9, RMSEcv = 19.7% HIA; ntest = 10, RMSEtest= 16.0% HIA). Although a functioning tool in model con-struction, the authors admitted that the neural network ap-proach suffers from difficulties in result interpretation in thatit is difficult to assess the individual contribution of eachdescriptor.

Computational studies have focused on developing im-proved descriptors of biological permeation. The approacheshave emphasised variable things, such as accuracy of corre-lation with in vivo results, speed of calculation, low numberof descriptors or to keep descriptors simple and informativein order to enable feedback to drug design. A brief reviewof the most common methods follows below. For each ap-proach, a few representative example studies are discussedwith the purpose of eliciting some aspects of the methodin question. However, these should not be taken as a com-prehensive coverage of all the studies carried out with thatparticular method. More detailed descriptions of computa-tional studies on biological permeation can be found in thereviews byNorinder and Haeberlein (2002), Egan and Lauri(2002)andClark and Grootenhuis (2003).

5.2.1. Univariate approachesSeveral mnemonics for rapid and preliminary identifi-

cation of drugs with permeation problems have been de-veloped on basis of analysis of vast databases of druglikecompounds. The “rule of 5” developed byLipinski et al.(1997) is a popular atom count method for rapid and ap-proximate identification of drugs with potential absorptionproblems. The rule states that if a compound satisfies anytwo of the following rules, it is likely to exhibit poor intesti-nal absorption: (1) MW > 500; (2) the number of hydro-gen bond donors > 5 (O–H or N–H groups); (3) the numberof hydrogen bond acceptors > 10 (any N or O atom, in-cluding donors); (4) CLOGP > 5.0 (or MlogP (Moriguchiet al., 1992) > 4.15). The authors emphasised that the alertis primarily designed to weed out compounds with absorp-tion and permeation problems at an early stage of drug de-velopment – compounds that pass the “rule of 5” withoutalert may still prove troublesome in later trials. A simi-lar model proposes a three-dimensional box, composed oflipophilicity, hydrogen-bonding potential and size, to iden-tify well-absorbed drugs (van de Waterbeemd, 2000).

A 2-rule model has been proposed for blood–brain par-titioning by Norinder and Haeberlein (2002). It states that(1) if N + O (the sum of the number of nitrogen and oxy-gen atoms) in a molecule is less than or equal to five, the


molecule has a good chance of entering the brain; and (2) iflogP – (N + O) is positive, then log BB is positive.

The main reasons for the approximate results obtainedwith univariate approaches is that they are unable to dealwith interactions between the properties. Furthermore, con-formational effects are unaccounted for when hydrogenbonding atoms are counted. However, due to their simpleand rapid format, such methods can be effective in prelim-inary drug screening and their readily interpretable rulesprovide guidelines for future drug design (Lipinski et al.,1997).

5.2.2. Fragment count methods

5.2.2.1. The Abraham descriptors.The Abraham grouphas recently developed models of biological permeation withparticular attention on the size and quality of their datasets.In 2001, they undertook an extensive study in which theyevaluated the human intestinal absorption data of 241 drugs(Zhao et al., 2001). Reliable and diffusion rate-limited datawas found for 169 compounds. A model was then con-structed based on the general linear solvation energy equa-tion:

%Abs= c + eE+ sS+ aA+ bB+ vV (5.5)

where E is the excess molar refraction,S the dipolar-ity/dipolarisability,A andB the hydrogen bond acidity andbasicity, respectively, andV is the McGowan characteristicvolume. The equation was initially applied using experi-mentally derived parameter values (Abraham et al., 1999,1995, 1994; Abraham and McGowan, 1987) but recently, afragment-based method was developed enabling calculationdirectly from structure (Platts et al., 1999, 2000; Abrahamand Platts, 2001). The lower-case coefficients are obtainedby multiple linear regression analysis. Two models wereconstructed on basis of training set data, and were sub-sequently employed to predict the absorption of test set

Fig. 16. Observed vs. calculated log BB usingEq. (5.6)for the complete set of compounds (n = 148). The corresponding statistics are given in the textand in Table 1, method 12. Reproduced fromPlatts et al. (2001)with permission from Elsevier.

compounds. The better model gave statistics of (n = 38, r2

= 0.83,q2 = 0.75, RMSE = 14) and (n = 131, RMSE = 14)for the training set and the test set, respectively. Step-wiseregression was carried out to identify the most significantdescriptors. Hydrogen bond acidity and basicity were foundto dominate absorption, both with a negative impact, butalso the volume-term was found to be a significant descrip-tor (positive impact). The effect of charge was investigatedby inclusion of an additional parameterI in the equation,but a negligible effect was found.

In a subsequent publication (Zhao et al., 2002), thesolvation-energy approach was refined. Instead of a linearrelationship, the intestinal absorption was taken to dependon a linear combination of the descriptors through first or-der kinetics, which led to improved correlations. Removalof 20 compounds, which contained both an acid and a basegroup, further improved correlations. No explanation for thenegative impact of these compounds on the predictabilitywas presented. The majority of them were well-absorbed,while their charged nature at the measurement pH was notestablished in the study. The importance of identifying andremoving dissolution rate-limited drugs from datasets whenstudying diffusion rate-limited absorption was emphasised.

In another study, the Abraham group (Platts et al., 2001)compiled log BB data and corresponding descriptor valuesfor 157 compounds. After removal of nine outliers, and theinclusion of a carboxylic acid indicator,I1, the followingmodel was obtained (Fig. 16):

log BB= 0.021+ 0.463E − 0.864S − 0.564A

− 0.731B + 0.933V − 0.567I1 (5.6)

n = 148, r2 = 0.745, q2 = 0.711, s = 0.343

The predictivity of the model was assessed by splittingthe full set of compounds into training and test sets of vari-able size and composition. Subsequently, equations of thetype of (5.6) were developed from the training sets to pre-


dict the test sets. The soundness of the method was con-firmed by that the statistics of the different test sets var-ied between 0.643< r2 < 0.775 and 0.319< s < 0.418.Larger size was found to enhance brain uptake whereas po-larity/polarisability, hydrogen bond acidity and basicity, andthe presence of carboxylic acid groups had an inhibitingeffect.

The approach of using linear solvation energy relation-ship descriptors has been successful in prediction of variousphysicochemical properties related to solute–solvent inter-actions (Kamlet et al., 1986; Taft et al., 1985b; Abrahamet al., 1999). Furthermore, the Abraham descriptors containphysical meaning that can be directly related to the structureof the compounds. The initial problem of obtaining exper-imentally derived values for all the descriptors and ensurethat intra-molecular interactions between the fragments areaccounted for has been solved, which renders the methodavailable for rapid screening. The use of linear regressionschemes to adjust the model parameters, however, raises thequestion of descriptor interrelations.

5.2.2.2. The electrotopological state.One set of molecu-lar structure descriptors to have gained popularity in predic-tion of biological permeation are the electrotopological stateindices (E-state) (Kier and Hall, 1999). The E-state indicesare numerical values, which contain information about thetopological and electronic environment of each atom in amolecule. Several E-state formalisms have been developed.The atom level E-state index,S(i), is calculated for eachskeletal atom or hydride groupi in a molecule as the sum ofthe intrinsic state of that atom,Ii, and the perturbations fromall the other skeletal atoms/hydride groups in the molecule,�Iij:

S(i) = Ii +∑

j

�Iij (j �= i) (5.7)

where the intrinsic state value for an atom is obtained fromits valence state electronegativity and its local topology, andthe perturbation term is calculated as the difference betweenthe intrinsic state values of the atoms divided by their dis-tance squared.

While the atom level E-state indices may be useful whencomparing structurally similar compounds, that is, whenatom-by-atom superposition is possible, they are not veryfeasible in themselves when studying structurally diversecompounds. For quantitative structure–property relationshipof molecules with no similarly located atoms, the atom typeE-state indices have proven more useful. The atom typeE-state index is defined as the sum of the individual atomlevel E-state values for a particular atom type (such as –OH,=O or –NH-) in the molecule (Hall and Kier, 1995). In par-allel with the atom level and atom type E-state indices, al-gorithms for calculating E-state indices for hydrogen atomshave been developed. Accordingly, the hydrogen atom typeE-state index is defined as the sum of the individual atom

level hydrogen E-state values for all atoms of a particularatom type (Rose et al., 2002) Hydrogen bond acceptor anddonor E-state indices have furthermore been developed bysumming the atom level hydrogen E-state values for all atomtypes classified as hydrogen bond donors/acceptors.

Norinder and Österberg (2001)investigated the use ofelectrotopological state indices in combination with calcu-lated molecular refractivity and octanol–water partition co-efficients for predicting human intestinal absorption, Caco-2cell permeation, blood–brain partition and IAM chromatog-raphy data. Principal component analysis and partial leastsquares methodology were employed in building the mod-els, and the importance of the variables was assessed with aleave-one-out approach. The E-state descriptors used in themodels were SumN (the sum of all atom-level E-state in-dices of nitrogen atoms), SumO, SumH (the sum of all hy-drogen E-state indices for hydrogen atoms attached to oxy-gen, nitrogen or sulphur atoms) and SHother (sum or all hy-drogen E-state indices for remaining hydrogen atoms). Theauthors pointed out that more detailed atom type E-state in-dices led to models with degraded predictivity due to ab-sence of these substructures from part of the compounds.

All models showed good statistics (r2: 0.75–0.93,q2:0.70–0.89) and predictivity when used on test set com-pounds. Permeation/partition was favoured by positive con-tributions from the octanol–water partition coefficient aswell as polarisability as described by SHother. The mod-els contained negative contributions from the descriptorsSumO, SumN and SumH indicating the effect of hydro-gen bonding.

In a recent study,Rose et al. (2002)employed E-state,molecular connectivity and shape indices to modelblood–brain partitioning of 106 compounds. The variablesfor the model were selected by eliminating intercorrelationsand subsequently by ranking the remaining descriptorsby statistical analysis. Three descriptors were selected forthe final model: the hydrogen bond donor E-state indexHST(HBd), the aromatic hydrogen E-state index squared[HST(arom)]2 and the second order difference valencemolecular connectivity index squared [d2�v]2. After somepreliminary validation a final model was developed (n= 102, r2 = 0.66, q2 = 0.62, MAE = 0.38). On basis ofthe model, blood–brain partition was found to be higherfor compounds with aromatic CH groups, less branchingand fewer or weaker hydrogen bond donor groups. Themodel was also used to predict blood–brain partition ofa large set of 20,039 drugs and druglike compounds. Asno experimental data was available for these compounds,the predictivity of the model on external data was notestablished.

The use of electrotopological state indices holds poten-tial for rapid screening of large virtual structure libraries,the model ofRose et al. (2002)was reported to compute5000–6000 molecules per minute. However, due to the abun-dance of available E-state descriptors, care needs to be takenwhen selecting variables.


5.2.3. Polar surface areaThe polar surface area (PSA) is commonly computed as

the van der Waals surface area of all nitrogen and oxygenatoms, plus the area of the hydrogen atoms attached to these.van de Waterbeemd and Kansy (1992)were the first to cor-relate biological permeation with polar surface area. Theyfound a strong correlation between brain uptake and the hy-drophilic part of the van der Waals surface calculated forsingle drug conformations for a small dataset of six com-pounds (n = 6, r2 = 0.945,s = 0.294). A poorer correlationwas obtained with a larger set of 20 compounds (n = 20, r2

= 0.610).Palm et al. (1996)argued that as many molecules can

adopt several low-energy conformations, the dynamic polarsurface area (PSAd) should be a more appropriate pre-dictor. In the dynamic polar surface area approach, allthree-dimensional conformations of the compounds are firstconstructed using molecular mechanics calculations. Thedynamic surface area of a molecule is then obtained as theBoltzmann-weighted average of the van der Waal’s surfaceareas calculated for all low-energy conformations of a com-pound. Excellent correlations were obtained between thedynamic polar surface areas of sixβ-adrenoreceptor antag-onists and their permeation in the Caco-2 monolayer andrat ileum models. The permeation could also be reasonablywell predicted by the calculated octanol–water distributioncoefficients (logDoct,7.4) of the drugs, but unlike PSAd, thisdescriptor failed to rank the permeation of the compoundsin the correct order.

In a subsequent study,Palm et al. (1997)demonstratedthat the dynamic polar surface area descriptor is also capa-ble of predicting human intestinal absorption of structurallymore diverse solutes. A strong sigmoidal relationship be-tween PSAd and fraction absorbed (FA) was observed for thetwenty drugs under study (n = 20,r2 = 0.94, RMSE = 9.2%).Drugs with a dynamic polar surface area≤ 60 Å2 were com-pletely absorbed (FA > 90%) whereas drugs with a PSAd≥ 140 Å2 exhibited poor intestinal absorption (FA< 10%).A weaker sigmoidal correlation was observed between FAand the total number of hydrogen bonds (r2 = 0.87, RMSE= 13.9%) and a poor result was obtained when correlatingClogP to FA (r2 = 0.34, RMSE = 31.6%).

Clark (1999a)used the same 20-compound data set asPalm et al. (1997)and showed that the polar molecularsurface areas calculated from a single conformation (PSA)were as successful in predicting intestinal absorption asthe Boltzmann-averaged areas (n = 20, r2 = 0.94, RMSE= 9.1%). A similar result was reached byStenberg et al.(2001) for a set of 27 structurally diverse drugs. It wasconcluded that although the use of dynamic PSA may bemore exact for large, flexible molecules, the computation-ally much faster single-conformation PSA suffices for rapiddrug screening purposes.

The effect of varying the simulated solvation environmenthas also been investigated (Palm et al., 1998; Stenberg et al.,1999, 2001). The influence on PSA of the simulated solvent

was small for relatively small and rigid solute molecules.For larger and more flexible compounds, the conformationalsearches in simulated water yielded more open conforma-tions with larger PSA values than in vacuum. However, inmost cases, the correlation between PSA and biological per-meation was not weakened by the lack of solvent, and there-fore it was postulated that for rapid screening, solvent-freesimulations could be used (Stenberg et al., 2001).

A downside of the polar surface area descriptor is itsinability to distinguish between nonpolar compounds oraccount for nonpolar atom groups, thereby failing to dis-criminate between molecules with identical polar surfacearea but different size and lipophilicity. Thus, the polar sur-face area will reflect the ability of the solutes to leave thehydrogen bonding environment provided by the aqueousphase and the polar head group region of the lipids, butnot the affinity of the molecules for the membrane interioror size-related effects. Consequently, the PSA has beencombined with other descriptors.

Stenberg et al. (1999)studied the correlation betweenCaco-2 monolayer permeation and surface area propertiesin oligopeptide derivatives. The dynamic polar surface areadescriptor area was able to distinguish between compoundsof variable hydrogen bonding ability, but compounds withsimilar hydrogen bonding characteristics and variable size orlipophilicity were poorly described. However, a linear com-bination of polar and nonpolar dynamic surface area yieldeda strong sigmoidal relationship with Caco-2 monolayer per-meation. Other descriptors that have successfully been usedin combination with polar surface area to predict biologicalpermeation include molar volume (van de Waterbeemd andKansy, 1992), the octanol–water partition/distribution coef-ficient (Clark, 1999b; Egan et al., 2000; Winiwarter et al.,1998) and number of hydrogen bond donors (Winiwarteret al., 1998).

In another study byStenberg et al. (2001), the polarsurface area was reasonably correlated with intestinal ab-sorption (n = 21, r2 = 0.81), but poorer correlations werefound for Caco-2 permeation (n = 27, r2 = 0.63), as shownin Fig. 17a and b. The weaker correlations compared tothose ofPalm et al. (1997)were believed to result fromthe greater structural diversity of the data (peptide deriva-tives, highly lipophilic drugs). Combination with nonpolarsurface area resulted in only slightly improved correlations.This led the authors to introduce a so-called partitioned to-tal surface area (PTSA) model in which the total molecu-lar surface area was fragmented into atom type areas. PLSand principal components analysis was employed to build amodel from various surface area descriptors. A significantimprovement in explanatory power was achieved with thismodel (r2 = 0.95, Fig. 17c), whereas external predictivitywas somewhat poorer. PSA remained the most important de-scriptor of the 9-descriptor model. The high interpretabilityof the PTSA approach was considered an additional bene-fit. Recently PTSA models were successfully employed inrapid construction of absorption profiles of drugs, taking into


Fig. 17. Relationship between dynamic polar surface area (PSAd) and (a)fraction absorbed after oral drug administration to humans; (b) Caco-2 cellmonolayer permeability coefficient. (c) Observed Caco-2 cell monolayerpermeabilities vs. calculated permeabilities from partitioned molecularsurface properties (PSTA). Squares and crosses denote training and testset compounds, respectively. Statistics can be found inTable 1, method6. All figures are fromStenberg et al. (2001), with permission from theAmerican Chemical Society.

account both solubility and permeation (Bergström et al.,2003).

While it is obvious that hydrogen bonding is one of theproperties that lie behind the polar surface area, a closer ex-amination of this descriptor is desirable for two reasons: (i)to facilitate its interpretation so that it can be used as feed-back in drug design; (ii) to investigate whether it is possi-ble to calculate the PSA in a simplified and faster manner.Stenberg et al. (2001)undertook a study in which the PSA(single conformation) of 128 structurally diverse compoundswas deconvoluted into various easily interpreted physico-

chemical properties using PLS methodology and principalcomponents analysis. It was found that the number, ratherthan the strength of hydrogen bonding atoms constituted themain property behind PSA and that three descriptors, corre-sponding to the number of hydrogen bond donors, hydrogenbond acceptor oxygens and hydrogen bond acceptor nitro-gens accounted for 93% of the variance in PSA.

Even if PSA can be reliably computed from a singleconformation and without simulated solvent, simpler andfaster methods are continuously sought.Pickett et al. (2000)reported having speeded up the calculation process em-ployed byClark (1999a)by omitting the optimisation step,which led to calculation times of over 10 compounds persecond while retaining good correlations to human intesti-nal absorption. In a study byErtl et al. (2000), the polarsurface areas of 34,810 drug-like molecules were used toderive fragment contributions for a variety of polar frag-ments using least-squares fitting. Topological polar surfaceareas (TPSA) were then computed by simple summationof the obtained fragment values. Results were compared tothree-dimensional PSA values for several test sets and ex-cellent correlations (r2 = 0.94) were established. Thus, themethod was proposed as a replacement of the conventionalPSA in rapid screening of drug molecules, having a capacityof processing over 8000 molecules per minute on a standard450 MHz PC.

An even simpler approach correlates PSA with atom typehydrogen bond counts.Österberg and Norinder (2000)foundexcellent correlations between PSA and a PLS model con-sisting of hydrogen bond count descriptors. Prediction ofvarious types of biological permeation (Fig. 18) using acombination of hydrogen bond descriptors and the calcu-lated octanol–water partition coefficient varied from mod-erate (r2 = 0.65) to good (r2 = 0.92). Cheng et al. (2002)likewise found a strong correlation between PSA and aleast squares regression model consisting of 18 E-state atomtype count descriptors for oxygen and nitrogen atoms (n= 438, r2 = 0.992). However,Stenberg et al. (2001)reporton calculation times of the order of milliseconds for gen-eration of static PSA and PTSA descriptors, which is notinferior to the above-presented 2D methods and retains the3D nature of the approach. When large and flexible com-pounds which exhibit significant internal hydrogen-bondingare modelled, the dynamic polar surface area is expected toprovide superior accuracy to the static, and especially the 2D,PSA.

In conclusion, several studies have proved the polar sur-face area a useful descriptor of biological permeation. Drugswith a PSA greater than 140 Å2 have been found to exhibitpoor intestinal absorption (Palm et al., 1997; Clark, 1999a;Kelder et al., 1999; Veber et al., 2002), whereas an upperlimit of 60–90 Å2 has been found for blood–brain partition(van de Waterbeemd et al., 1998; Kelder et al., 1999). Recentdevelopment has focused on the fast estimation of PSA forrapid drug screening as well as on the use of PTSA modelsfor combined assessment of drug permeation and solubility.


Fig. 18. Experimental log BB vs. log BB calculated (training set)and predicted (test set) using hydrogen bond count descriptors and theoctanol–water partition coefficient. Statistic parameters can be found inTable 1, method 17. Compound structures are available inClark (1999b).Fig. reproduced fromÖsterberg and Norinder (2000)with permissionfrom the American Chemical Society.

5.2.4. 3D molecular field modelsAnother approach that relates three-dimensional molec-

ular structure with physicochemical properties of relevanceto biological permeation is the use of 3D molecular fields.These fields are three-dimensional maps of the repul-sive/attractive energies between a molecule and an interact-ing substance. The molecular electrostatic potential (MEP),describing the interaction energy between a point charge ordipole and the molecule, has found significant use in stud-ies of ligand–receptor interactions (e.g.Boer et al., 2001).The molecular lipophilicity potential (MLP, Section 5.1.1)was developed to determine the dependence of lipophilicityon conformation and has found use in the calculation oflogP and characterisation of ligand–receptor interactions(Gaillard et al., 1994, 1996). Recently, the same group re-ported on their version of the molecular hydrogen-bondingpotential (MHBP) and its use in structure–permeation rela-tionships (Rey et al., 2001).

The GRID program (available from Molecular DiscoveryLtd., Middlesex, UK) is a commonly used tool in compu-tational modelling of molecular surfaces (Goodford, 1985;Boobbyer et al., 1989). Its energy function is based onthe sum of three interaction potentials between the targetmolecule and a probe (atom or group): the Lennard–Jones,the electrostatic and the hydrogen bonding potentials. GRIDhas been applied successfully in structure-based drug de-sign (von Itzstein et al., 1993) and as descriptor input forquantitative structure–activity relationship (QSAR) models(Cruciani and Watson, 1994). Comparative molecular fieldanalysis (CoMFA) is one of the most frequently employed

procedures in 3D-QSAR (Kim et al., 1998). CoMFA, likeGRID, uses probes to construct a three-dimensional map ofthe interaction energies of the target molecule. Whereas theGRID approach simply calculates the interaction fields thatcan subsequently be visualised using a molecular graphicsprogram, an integral part of the CoMFA procedure is theconstruction of a PLS model on basis of the correlation be-tween biological response and the computed interaction en-ergies (Livingstone, 2000).

Despite giving rise to QSAR models with good predic-tivity (e.g. Cruciani and Watson, 1994), the informationcontained in the above-described 3D field maps can be cum-bersome to interpret for drug design purposes. The problembecomes critical when the aim is quantification and compar-ison of a set of molecules. To this end, specialised tools forthe transformation of 3D molecular interaction fields intoeasily interpretable descriptors are required. The VolSurfprocedure (available from Molecular Discovery Ltd.) wasdeveloped byCruciani et al. (2000). The basic concept ofVolSurf is to convert the information present in 3D molec-ular field maps into a limited number of quantitative nu-merical descriptors. Molecular recognition is achieved withimage analysis software, while its conversion into descrip-tors involves the use of external chemical knowledge. Oncethe descriptors have been calculated, they are correlatedwith the desired experimental property using chemometrictools, and turned into a model. The procedure is com-pletely automatic once the 3D structures of the compoundshave been generated, and reportedly takes approximately2 min for 100 compounds at low resolution (Crivori et al.,2000).

Several models have been built employing VolSurf tech-nology. Crivori et al. (2000)report on the use of VolSurffor qualitative (±) prediction of blood–brain barrier perme-ation. Subjecting a total of 229 compounds and 72 VolSurfdescriptors to PLS discriminant analysis they arrived at amodel that assigned a correct BBB permeation profile toover 90% of the compounds. The score plots for the discrim-inant PLS model are shown inFig. 19. The most importantdescriptors affecting BBB permeation were the polarity de-scriptors, which, as expected, were inversely correlated withpermeation. Overall, it appeared that permeation was gov-erned by the balance of all descriptors, rather than a singledescriptor type.

Ekins et al. (2001)compared different 3D-based quanti-tative structure–permeation relationship models, includingCoMFA and VolSurf (using GRID), in the prediction ofCaco-2 permeability of 2-aminobenzimidazoles. Ninteencompounds were used as a training set to construct themodel. Both approaches yielded a three-component modelwith the statistics: (CoMFA:n = 19, r2 = 0.96,q2 = 0.30;VolSurf: n = 19, r2 = 0.76,q2 = 0.54). Despite the superiorcorrelation coefficient of the CoMFA model for the trainingset compounds, both models gave similar statistics for a testset of nine compounds: (CoMFA:n = 9, r2 = 0.84, VolSurf:n = 9, r2 = 0.83). The best test set statistics were obtained


Fig. 19. The score plot for the two significant latent variables of the discriminant PLS model based on 229 compounds and 72 VolSurf descriptors. Acorrect BBB profile was assigned to over 90% of the compounds. Based on the 0.6 unit prediction error of the discriminant PLS, a confidence intervalwas drawn between the BBB+ and BBB− regions where prediction can be doubtful. Reproduced fromCrivori et al. (2000)with permission from theAmerican Chemical Society.

with the pharmacophore generation program Catalyst (Ac-celrys): (n = 9, r2 = 0.94).

5.2.5. Quantum chemical approachesRecent progress in computational hardware and algo-

rithms has also assisted the development of molecularquantum-mechanical calculations. In addition to moleculargeometry, quantum mechanics approaches take into accountelectronic effects, such as charge distribution, which are notinherently incorporated into molecular mechanics calcula-tions. The main benefit of quantum-chemical descriptors istheir accuracy and that they can be derived solely from thetheoretical structure of a molecule (Karelson et al., 1996).

Along with many benefits, the quantum-chemical ap-proach does contain certain downsides. Despite the con-tinuous improvement of computational resources, quantumchemical calculations are very time-consuming. As a re-sult, the extremely computer costly ab initio methods haveoften been replaced by more rapid, semiempirical quantumchemical models. An additional drawback of the quantumchemical approach is that the calculations are performedfor a single structure at an unrealistic energetic minimum,corresponding to the hypothetical physical state of thegas at 0 K and infinitely low pressure. Thus, when study-ing flexible molecules some type of averaging over thelowest energy conformations may be required. Quantumchemical approaches are also not suitable for modelling

solute–solvent interactions. When studying specific effectssuch as hydrogen bonding, the supermolecule approach hasbeen adopted, where the solute molecule is modelled to-gether with its first solvent coordination shell. Solvent bulkeffects are commonly accounted for with dielectric reactionfield models (Karelson et al., 1996).

In a recent study,Norinder et al. (1997, 1999)investigatedthe use of MolSurf (available from Qemist AB, Karlskoga,Sweden) technology in combination with multivariate statis-tics to predict biological permeation. The computationalprotocol involves performing a conformational analysis ofeach compound using molecular mechanics, followed bya semi-empirical geometry optimisation of the lowest en-ergy conformation. The wave function from an ab initiocalculation of this conformation is then used by the Mol-Surf program to compute various properties related to thevalence region of the molecule. These descriptors includebasicity/acidity, lipophilicity, hydrogen bonding, polarity,polarisability and charge-transfer characteristics. A partialleast squares method is used to determine the relationshipsbetween the calculated MolSurf descriptors and the biolog-ical permeation of a selected training set of compounds. Asnot all calculated descriptors will be of importance to themodel, a leave-one-out procedure is carried out, whereinthe importance of each variable for the predictivity of themodel is assessed. Descriptors, which decrease the predic-tivity are permanently deleted. Subsequently, the obtained


Fig. 20. Experimental human intestinal absorption vs. computed HIAusing MolSurf parametrisation. Calculated = training set; Predicted = testset. Statistics can be found inTable 1, method 3. Chemical names ofcompounds are given in the Appendix. Reproduced fromNorinder et al.(1999) with permission from Elsevier.

model can be used to predict permeation of test compoundson basis of their MolSurf parameters.

When relating the computed MolSurf descriptors of aset of compounds to their Caco-2 permeation (Norinderet al., 1997) and human intestinal absorption (Norinder et al.,1999), good correlations were obtained both in the trainingset and the test set (r2 = 0.90,q2 = 0.69,Fig. 20). The mostimportant factors affecting the intestinal absorption of thedrugs proved to be related to hydrogen bonding, which had anegative impact on permeation. The number of possible hy-drogen bonding atoms was found to have a greater effect onpermeation than the actual strength of the bonds. Curiously,hydrogen bond acceptor characteristics related to nitrogenatoms appeared to exert a stronger (negative) effect on in-testinal absorption than those related to oxygen (Norinderet al., 1999). This was suggested to be due to the tendency ofnitrogen atoms to be ionised at physiological pH. Assumingthat the drugs need to be neutral in order to transfer throughthe membrane, the PLS modelling will thus incorporate inthe MolSurf descriptors of the nitrogen atoms also the effectsof deprotonation and desolvation. In both studies (Norinderet al., 1997, 1999), high polarisability and charge-transfercharacteristics were factors that promoted permeation.

Quantum-chemical descriptors have also found use inoriginally empirical models. As mentioned earlier, the linearsolvation energy relationship approach (see Sections 5.1.1and 5.2.2) suffers from difficulties in encountering values forthe various descriptors.Wilson and Famini (1991)developeda similar approach based on computationally-derived de-scriptors, called theoretical linear solvation energy relation-

ships (TLSER), where the parameters are calculated fromsemi-empirical quantum chemistry methods.

5.3. Conclusions

Comparison of the various models developed to predictbiological permeation reveals that relatively few descriptorsseem to be significant for biological permeation. Accord-ingly, the combination of a hydrogen bond donor descriptor,a general hydrogen bonding descriptor and a lipophilicitydescriptor have been found to be a working recipe for pre-diction of human intestinal absorption (Winiwarter et al.,2003).

When comparing the various computational methods topredict biological permeation of drugs, one should keepin mind the target application of the method. Preliminaryscreening of large libraries of drug candidates requires arapid and simple method. However, for detailed and accuratestudies of the mechanisms relating structure to property, theuse of computationally more demanding approaches may befounded. For most applications high interpretability of re-sults is desirable, i.e. a direct and comprehensible relationbetween descriptors and compound structure.

The statistics of the models discussed in this section arecompiled inTable 1. As a direct comparison between themodels is not feasible due to the variable size and composi-tion (differences in structural diversity) of the datasets theyare based on, the table is intended more as a summary. Onething that becomes evident upon examining the table is thatwhile there exists an abundance of computational approachesto model biological permeation, most of them share the sameproblem: they are constructed/validated on too small datasets to allow for any real conclusions to be drawn. Manydatasets are furthermore biased towards highly permeablecompounds and contain drugs that are transported by otherpathways than the passive transcellular route. As pointed outby several researchers (Bergström et al., 2003; Clark, 2003;Egan and Lauri, 2002), large and well-balanced datasetsof reliable and structurally diverse data are a prerequisitefor the development of computational models that can beimplemented as routine screens in the drug developmentprocess.

As most of the computational approaches developed areaimed at enabling high-throughput virtual screening of vastcompound libraries, the recent trend has shifted towardsfaster methods and more easily interpreted descriptors. Fur-thermore, recent studies report on combined computationalscreening of drug permeation and other biopharmaceuticalproperties, such as solubility (Bergström et al., 2003) andactivity (Egan and Lauri, 2002; Pickett et al., 2000).

6. Overall conclusions

The emphasis in passive drug permeation modelling isshifting from experimental to computational methods. How-


Table 1Summary of computational approaches to predict biological permeation

Method Predicted quantity ntrain r2train q2

train strain ntest Test set statistics Reference

1 Abraham % Abs 38 0.83 0.75 14a 131 RMSE = 14, MAE = 11 Zhao et al. (2001)2 E-state Logit (Abs) 13 0.90 0.82 0.63 7 RMSE = 0.413 Norinder and Österberg (2001)3 MolSurf Logit (Abs) 13 0.90 0.69 0.63 7 RMSE = 0.49,Fig. 20 Norinder et al. (1999)4 E-state log (PappCaco-2) 9 0.93 0.79 0.37 8 RMSE = 0.41 Norinder and Österberg (2001)5 E-state, ClogP log (PCaco-2) 17 0.71 0.67 0.84a 10 RMSE = 0.86 Stenberg et al. (2001)6 PSTA log (PCaco-2) 17 0.95 0.86 0.33a 10 RMSE = 1.15 Stenberg et al. (2001)7 MolSurf log (PCaco-2) 17 0.87 0.78 0.56a 10 RMSE = 0.83 Stenberg et al. (2001)8 MolSurf log (PappCaco-2) 9 0.94 0.85 0.33 8 RMSE = 0.41 Norinder et al. (1997)9 CoMFA log (PappCaco-2) 19 0.96 0.30 0.08b 9 r2 = 0.84 Ekins et al. (2001)10 VolSurf log (PappCaco-2) 19 0.76 0.54 – 9 r2 = 0.83 Ekins et al. (2001)11 Various molecular

propertieslog (BB) 48 0.84 0.79 0.19 17 r2 = 0.68, MAE = 0.41 Lobell et al. (2003)

12 Abraham log (BB) 148 0.75 0.71 0.34 –Fig. 16 Platts et al. (2001)13 Abraham log (BB) 74 0.77/0.76 – 0.33/0.34 74r2 = 0.74/0.70,

newline s = 0.38/0.39Platts et al. (2001)

14 E-state log (BB) 28 0.75 0.70 0.39 31 RMSE = 0.44 Norinder and Österberg (2001)15 E-state log (BB) 102 0.66 0.62 0.45 – – Rose et al. (2002)16 H-bonds log (BB) 35 0.71 0.68 0.42a 34 RMSE = 0.37 Österberg and Norinder (2000)17 H-bonds log (BB) 23 0.72 0.68 0.52a 22 RMSE = 0.50,Fig. 18 Österberg and Norinder (2000)

ntrain: number of compounds in the training set;r2train and q2

train: the r2 value and the leave one out cross-validatedr2 value of the training set;strain:standard deviation of the training set. Exceptions, values marked with lettersa are RMSE, andb are standard error of the estimate,ntest: number ofcompounds in the test set, RMSE: root mean square error; MAE: mean absolute error.

ever, this by no means diminishes the importance of ex-perimental approaches. More than ever, large amounts ofhigh quality data on biological permeation are needed forthe construction of reliable models, which can be applied toa wide region in chemical space. One aspect that remainsunclear, is the effect of charge on permeation. Most experi-mental and computational methods do not properly accountfor solute charge, nor do there exist comprehensive stud-ies on the subject. Liquid–liquid electrochemistry provides

Appendix A. Explanations to figure legends

Figure Legend Chemical name of compound Reference

1 3a N-Acetyl-d-valyl-d-phenylalanineN-methylamide Goodwin et al. (2001)4 N-Acetyl-d-leucyl-d-phenylalanineN-methylamide5a N-Acetyl-d-phenylalanyl-d-phenylalanineN-methylamide6 N-Acetyl-d-cyclohexylalanyl-d-phenylalanineN-methylamide7 N-Acetylglycine phenethylamide8 N-Acetyl-d-alanine phenethylamide9a N-Acetyl-d-valine phenethylamide10 N-Acetyl-d-leucine phenethylamide

10(a) a Peg (polyethylene glycol) Beigi et al. (1995)b Atenololc Salicylic acidd Aspirine Terbutalinef Warfaring Metoprololh Hydrocortisonei Alprenolol

a system for exclusive characterisation of properties con-tributing to biological permeation of charged species. How-ever, well-characterised reference data on ion permeation isneeded for correlations.

Acknowledgements

A.M. thanks the Academy of Finland for financial support.


Appendix A (Continued)

Figure Legend Chemical name of compound Reference

j Corticosteronek Chlorpromazinel dl-Propranolol

20 1 Alprenolol Norinder et al. (1999).2 Atenolol3 Ciprofloxacin4 Diazepam5 Foscarnet6 Lactulose7 Mannitol8 Metolazone9 Metoprolol10 Nordiazepam11 Olsalazine12 Oxazepam13 Oxprenolol14 Phenazone15 Pindolol16 Practolol17 Raffinose18 Sulpiride19 Sulphasalazine20 Tranexamic acid

References

Abraham, M.H., Liszi, J., 1978. Calculations on ionic solvation. Part I.Free energies of solvation of gaseous univalent ions using a one-layercontinuum model. J. Chem. Soc., Faraday Trans. I 74, 1604–1614.

Abraham, M.H., McGowan, J.C., 1987. The use of characteristic volumesto measure cavity terms in reversed phase liquid chromatography.Chromatographia 23, 243–246.

Abraham, M.H., Platts, J.A., 2001. Hydrogen bond structural group con-stants. J. Org. Chem. 66, 3484–3491.

Abraham, M.H., Martins, F., Mitchell, R.C., Salter, C.J., 1999. Hydrogenbonding. 47. Characterization of the ethylene glycol–heptane partitionsystem: hydrogen bond acidity and basicity of peptides. J. Pharm. Sci.88, 241–247.

Abraham, M.H., Chadha, H.S., Leitao, R.A.E., Mitchell, R.C., Lam-bert, W.J., Kaliszan, R., Nasal, A., Haber, P., 1997. Determina-tion of solute lipophilicity, as logP(octanol) and logP(alkane) usingpoly(styrene–divinylbenzene) and immobilised artificial membrane sta-tionary phases in reversed-phase high-performance liquid chromatog-raphy. J. Chromatogr. A 766, 35–47.

Abraham, M.H., Chadha, H.S., Mitchell, R.C., 1995. The factors thatinfluence skin penetration of solutes. J. Pharm. Pharmacol. 47, 8–16.

Abraham, M.H., Chadha, H.S., Mitchell, R.C., 1994. Hydrogen-bonding.33. Factors that influence the distribution of solutes between bloodand brain. J. Pharm. Sci. 83, 1257–1268.

Amato, M., Barbato, F., Morrica, P., Quaglia, F., La Rotonda, M.I., 2000.Interactions between amines and phospholipids: a chromatographicstudy on immobilised artificial membrana (IAM) stationary phases atvarious pH values. Helv. Chim. Acta 83, 2836–2847.

Artursson, P., Borchardt, R.T., 1997. Intestinal drug absorption andmetabolism in cell cultures: Caco-2 and beyond. Pharm. Res. 14,1655–1658.

Artursson, P., Karlsson, J., 1991. Correlation between oral drug absorp-tion in humans and apparent drug permeability coefficients in humanintestinal epithelial (Caco-2) cells. Biochem. Biophys. Res. Commun.175, 880–885.

Artursson, P., Palm, K., Luthman, K., 2001. Caco-2 monolayers in ex-perimental and theoretical predictions of drug transport. Adv. DrugDeliv. Rev. 46, 27–43.

Audry, E., Dubost, J.-P., Colleter, J.-C., Dallet, P., 1986. A new approachto structure–activity relations: the molecular lipophilicity potential.Eur. J. Med. Chem. 21, 71–72.

Audus, K.L., Bartel, R.L., Hidalgo, I.J., Borchardt, R.T., 1990. The useof cultured epithelial and endothelial cells for drug transport andmetabolism studies. Pharm. Res. 7, 435–451.

Avdeef, A., 1993. pH-metric log-P. 2. Refinement of partition-coefficientsand ionization-constants of multiprotic substances. J. Pharm. Sci. 82,183–190.

Avdeef, A., Strafford, M., Block, E., Balogh, M.P., Chambliss, W., Khan,I., 2001. Drug absorption in vitro model: filter immobilized artificialmembranes. 2. Studies of the permeability properties of lactones inPiper methysticum. Forst. Eur. J. Pharm. Sci. 14, 271–280.

Avdeef, A., Box, K.J., Comer, J.E.A., Hibbert, C., Tam, K.Y., 1998.pH-metric logP 10. Determination of liposomal membrane-water par-tition coefficients of ionizable drugs. Pharm. Res. 15, 209–215.

Balon, K., Riebesehl, B.U., Müller, B.W., 1999a. Determination of lipo-some partitioning of ionisable drugs by titration. J. Pharm. Sci. 88,802–806.

Balon, K., Riebesehl, B.U., Müller, B.W., 1999b. Drug liposome partition-ing as a tool for the prediction of human passive intestinal absorption.Pharm. Res. 16, 882–888.

Bangham, A.D., 1993. Liposomes: the Babraham connection. Chem. Phys.Lipids 64, 275–285.

Bangham, A.D., Standish, M.M., Watkins, J.C., 1965. Diffusion of univa-lent ions across the lamellae of swollen phospholipids. J. Mol. Biol.13, 238–252.


Barbato, F., La Rotonda, M.I., Quaglia, F., 1997. Interactions of nons-teroidal anti-inflammatory drugs with phospholipids: comparison be-tween octanol/buffer partition coefficients and chromatographic in-dexes on immobilized artificial membranes. J. Pharm. Sci. 86, 225–229.

Beigi, F., Gottschalk, I., Lagerquist Hägglund, C., Haneskog, L., Brekkan,E., Zhang, Y., Österberg, T., Lundahl, P., 1998. Immobilized liposomeand biomembrane partitioning chromatography of drugs for predictionof drug transport. Int. J. Pharm. 164, 129–137.

Beigi, F., Yang, Q., Lundahl, P., 1995. Immobilized-liposome chromato-graphic analysis of drug partitioning into lipid bilayers. J. Chromatogr.A 704, 315–321.

Bergström, C.A.S., Strafford, M., Lazorova, L., Avdeef, A., Luthman, K.,Artursson, P., 2003. Absorption classification of oral drugs based onmolecular surface properties. J. Med. Chem. 46, 558–570.

Beschiaschvili, G., Seelig, J., 1992. Peptide binding to lipid bilayers.Nonclassical hydrophobic effect and membrane-induced pK shifts.Biochemistry 31, 10044–10053.

Betageri, G.V., Rogers, J.A., 1988. The liposome as a distribution modelin QSAR studies. Int. J. Pharm. 46, 95–102.

Betageri, G.V., Rogers, J.A., 1987. Thermodynamics of partitioning ofb-blockers in then-octanol-buffer and liposome systems. Int. J. Pharm.36, 165–173.

Boer, D.R., Kooijman, H., van der Louw, J., Groen, M., Kelder, J., Kroon,J., 2001. Relation between the molecular electrostatic potential andactivity of some FF-MAS related sterol compounds. Bioorg. Med.Chem. 9, 2653–2659.

Boobbyer, D.N.A., Goodford, P.J., McWhinnie, P.M., Wade, R.C., 1989.New hydrogen-bond potentials for use in determining energeticallyfavorable binding sites on molecules of known structure. J. Med.Chem. 32, 1083–1094.

Bouchard, G., Pagliara, A., Carrupt, P.-A., Testa, B., Gobry, V., Gi-rault, H.H., 2002. Theoretical and experimental exploration of thelipophilicity of zwitterionic drugs in the 1,2-dichloroethane/water sys-tem. Pharm. Res. 19, 1150–1159.

Bouchard, G., Carrupt, P.-A., Testa, B., Gobry, V., Girault, H.H., 2001a.The apparent lipophilicity of quaternary ammonium ions is influencedby Galvani potential difference, not ion-pairing: a cyclic voltammetrystudy. Pharm. Res. 18, 702–708.

Bouchard, G., Pagliara, A., van Palen, G.P., Carrupt, P.-A., Testa, B.,Gobry, V., Girault, H.H., Caron, G., Ermondi, G., Fruttero, R., 2001b.Ionic partition diagram of the zwitterionic antihistamine Cetirizine.Helv. Chim. Acta 84, 375–387.

Brockman, H., 1994. Dipole potential of lipid membranes. Chem. Phys.Lipids 73, 57–79.

Caldwell, G.W., Masucci, J.A., Evangelisto, M., White, R., 1998. Eval-uation of the immobilized artificial membrane phosphatidylcholine.Drug discovery column for high-performance liquid chromatographicscreening of drug–membrane interactions. J. Chromatogr. A 800, 161–169.

Camenisch, G., Folkers, G., van de Waterbeemd, H., 1996. Review oftheoretical passive drug absorption models: historical background,recent developments and limitations. Pharm. Acta Helv. 71, 309–327.

Caron, G., Steyaert, G., Pagliara, A., Reymond, F., Crivori, P., Gaillard, P.,Carrupt, P.-A., Avdeef, A., Comer, J., Box, K.J., Girault, H.H., Testa,B., 1999. Structure-lipophilicity relationships of neutral and protonatedb-blockers. Part I. Intra- and intermolecular effects in isotropic solventsystems. Helv. Chim. Acta 82, 1211–1222.

Carrupt, P.-A., Testa, B., Gaillard, P., 1997. Computational approachesto lipophilicity: methods and applications. In: Lipkowitz, K.B., Boyd,D.B. (Eds.), Reviews in Computational Chemistry. Wiley, New York,vol. 11, Chapter 5, pp. 241–315.

Cevc, G., 1993. Solute transport across bilayers. In: Cevc, G. (Ed.),Phospholipids Handbook. Marcel Decker, New York, pp. 639–661.

Cevc, G., 1990. The molecular mechanism of interaction between mono-valent ions and polar surfaces, such as lipid bilayer membranes. Chem.Phys. Lett. 170, 283–288.

Cheng, A., Diller, D.J., Dixon, S.L., Egan, W.J., Lauri, G., Merz Jr.,K.M., 2002. Computation of the physio-chemical properties and datamining of large molecular collections. J. Comput. Chem. 23, 172–183.

Chou, J.T., Jurs, P.C., 1979. Computer-assisted computation of partitioncoefficients from molecular structures using fragment constants. J.Chem. Inf. Comput. Sci. 19, 172–178.

Clark, D.E., 2003. In silico prediction of blood–brain barrier permeation.Drug Discov. Today 8, 927–933.

Clark, D.E., 1999a. Rapid calculation of polar molecular surface area andits application to the prediction of transport phenomena. 1. Predictionof intestinal absorption. J. Pharm. Sci. 88, 807–814.

Clark, D.E., 1999b. Rapid calculation of polar molecular surface area andits application to the prediction of transport phenomena. 2. Predictionof blood–brain barrier penetration. J. Pharm. Sci. 88, 815–821.

Clark, D.E., Grootenhuis, P.D.J., 2003. Predicting passive transport insilico: history, hype, hope. Curr. Top. Med. Chem. 3, 1193–1203.

Conradi, R.A., Hilgers, A.R., Ho, N.F.H., Burton, P.S., 1991. The influenceof peptide structure on transport across Caco-2 cells. Pharm. Res. 8,1453–1460.

Cramb, D.T., Wallace, S.C., 1997. Structure and biomembrane mimeticbehavior of the water–octanol interface. J. Phys. Chem. B 101, 2741–2744.

Crivori, P., Cruciani, G., Carrupt, P.-A., Testa, B., 2000. Predictingblood–brain barrier permeation from three-dimensional molecularstructure. J. Med. Chem. 43, 2204–2216.

Cruciani, G., Watson, K.A., 1994. Comparative molecular field analysisusing GRID force-field and GOLPE variable selection methods in astudy of inhibitors of glycogen phosphorylaseb. J. Med. Chem. 37,2589–2601.

Cruciani, G., Crivori, P., Carrupt, P.-A., Testa, B., 2000. Molecular fields inquantitative structure–permeation relationships: the VolSurf approach.J. Mol. Struct. (Theochem.) 503, 17–30.

Dalby, A., Nourse, J.G., Hounshell, W.D., Gushurst, A.K.I., Grier, D.L.,Lealand, B.A., Laufer, J., 1992. Description of several chemical struc-ture file formats used by computer programs developed at moleculardesign limited. J. Chem. Inf. Comput. Sci. 32, 244–255.

Danelian, E., Karlén, A., Karlsson, R., Winiwarter, S., Hansson, A., Löfås,S., Lennernäs, H., Hämäläinen, M.D., 2000. SPR biosensor studiesof the direct interaction between 27 drugs and a liposome surface:correlation with fraction absorbed in humans. J. Med. Chem. 43,2083–2086.

DeBolt, S.E., Kollman, P.A., 1995. Investigation of structure, dynamics,and solvation in 1-octanol and its water-saturated solution: moleculardynamics and free-energy perturbation studies. J. Am. Chem. Soc.117, 5316–5340.

De Young, L.R., Dill, K.A., 1988. Solute partitioning into lipid bilayermembranes. Biochemistry 27, 5281–5289.

Dorsey, J.G., Khaledi, M.G., 1993. Review: hydrophobicity estimationsby reversed-phase liquid chromatography–implications for biologicalpartitioning processes. J. Chromatogr. A 656, 485–499.

Drayton, C.J., 1990. Cumulative subject index and drug compendium.In: Hansch, C., Sammes, P.G., Taylor J.B. (Eds.), ComprehensiveMedicinal Chemistry, vol. 6. Pergamon Press, Oxford.

Egan, W.J., Lauri, G., 2002. Prediction of intestinal permeability. Adv.Drug Deliv. Rev. 54, 273–289.

Egan, W.J., Merz Jr., K.M., Baldwin, J.J., 2000. Prediction of drugabsorption using multivariate statistics. J. Med. Chem. 43, 3867–3877.

Ekins, S., Durst, G.L., Stratford, R.E., Thorner, D.A., Lewis, R.,Loncharich, R.J., Wikel, J.H., 2001. Three-dimensional quantitativestructure-permeability relationship analysis for a series of inhibitorsof rhinovirus replication. J. Chem. Inf. Comput. Sci. 41, 1578–1586.

El Tayar, N., Tsai, R.-S., Testa, B., Carrupt, P.-A., Leo, A., 1991. Par-titioning of solutes in different solvent systems: the contribution ofhydrogen bonding capacity and polarity. J. Pharm. Sci. 80, 590–598.

Ertl, P., Rohde, B., Selzer, P., 2000. Fast calculation of molecular polarsurface area as a sum of fragment-based contributions and its appli-


cations to the prediction of drug transport properties. J. Med. Chem.43, 3714–3717.

Fischer, H., Gottschlich, R., Seelig, A., 1998. Blood–brain barrier perme-ation: molecular parameters governing passive diffusion. J. Membr.Biol. 165, 201–211.

Franks, N.P., Abraham, M.H., Lieb, W.R., 1993. Molecular organizationof liquid n-octanol: an X-ray diffraction analysis. J. Pharm. Sci. 82,466–470.

Fujita, T., Iwasa, J., Hansch, C., 1964. A new substituent constant�

derived from partition coefficients. J. Am. Chem. Soc. 86, 5175–5180.Furet, P., Sele, A., Cohen, N.C., 1988. 3D molecular lipophilicity potential

profiles: a new tool in molecular modeling. J. Mol. Graph. 6, 182–189.Gaillard, P., Carrupt, P.-A., Testa, B., Schambel, P., 1996. Binding

of arylpiperazines, (aryloxy)propanolamines, and tetrahydropyridylin-doles to the 5-HT1A receptor: contribution of the molecular lipophilic-ity potential to three-dimensional quantitative structure-affinity rela-tionship models. J. Med. Chem. 39, 126–134.

Gaillard, P., Carrupt, P.-A., Testa, B., Boudon, A., 1994. Molecularlipophilicity potential, a tool in 3D QSAR: method and applications.J. Comput. Aided Mol. Des. 8, 83–96.

Garberg, P., Eriksson, P., Schipper, N., Sjöström, B., 1999. Automatedabsorption assessment using Caco-2 cells cultured on both sides ofpolycarbonate membranes. Pharm. Res. 16, 441–445.

Gawrisch, K., Ruston, D., Zimmerberg, J., Parsegian, V.A., Rand, R.P.,Fuller, N., 1992. Membrane dipole potentials, hydration forces, and theordering of water at membrane surfaces. Biophys. J. 61, 1213–1223.

Genty, M., González, G., Clere, C., Desangle-Gouty, V., Legendre, J.-Y.,2001. Determination of the passive absorption through the rat intestineusing chromatographic indices and molar volume. Eur. J. Pharm. Sci.12, 223–229.

Ghose, A.K., Viswanadhan, V.N., Wendoloski, J.J., 1998. Prediction ofhydrophobic (lipophilic) properties of small organic molecules usingfragmental methods: an analysis of ALOGP and CLOGP methods. J.Phys. Chem. A 102, 3762–3772.

Girault, H.H.J., Schiffrin, D.J., 1989. Electrochemistry of liquid–liquidinterfaces. In: Bard, A.J. (Ed.), Electroanalytical Chemistry, vol. 15.Marcel Dekker, New York, pp. 1–141.

Gobas, F.A.P.C., Lahittete, J.M., Garofalo, G., Shiu, W.Y., MacKay,D., 1988. A novel method for measuring membrane-water parti-tion coefficients of hydrophobic organic chemicals: comparison with1-octanol–water partitioning. J. Pharm. Sci. 77, 265–272.

Gobry, V., Ulmenau, S., Reymond, F., Bouchard, G., Carrupt, P.-A., Testa,B., Girault, H.H., 2001. Generalization of ionic partition diagramsto lipophilic componds and to biphasic systems with variable phasevolume ratios. J. Am. Chem. Soc. 123, 10684–10690.

Gobry, V., Bouchard, G., Carrupt, P.-A., Testa, B., Girault, H.H., 2000.Physicochemical characterization of Sildenafil: ionisation, lipophilicitybehaviour, and ionic-partition diagram studied by two-phase titrationand electrochemistry. Helv. Chim. Acta 83, 1465–1474.

Gombar, V.K., Enslein, K., 1996. Assessment ofn-octanol/water partitioncoefficient: when is the assessment reliable? J. Chem. Inf. Comput.Sci. 36, 1127–1134.

Goodford, P.J., 1985. A computational procedure for determining en-ergetically favorable binding sites on biologically important macro-molecules. J. Med. Chem. 28, 849–857.

Goodwin, J.T., Conradi, R.A., Ho, N.F.H., Burton, P.S., 2001. Physic-ochemical determinants of passive membrane permeability: role ofsolute hydrogen-bonding potential and volume. J. Med. Chem. 44,3721–3729.

Grinius, L., Stanton, D.T., Morris, C.M., Howard, J.M., Curnow, A.W.,2002. Profiling of drugs for membrane activity using liposomes as anin vitro model system. Drug Dev. Ind. Pharm. 28, 193–202.

Gulaboski, R., Mirèeski, V., Scholz, F., 2002. An electrochemical methodfor determination of the standard Gibbs energy of anion trans-fer between water andn-octanol. Electrochem. Comm. 4, 277–283.

Hall, L.H., Kier, L.B., 1995. Electrotopological state indices for atomtypes: a novel combination of electronic, topological, and valencestate information. J. Chem. Inf. Comput. Sci. 35, 1039–1045.

Hansch, C., Fujita, T., 1964. r-s-p analysis. A method for the correlationof biological activity and chemical structure. J. Am. Chem. Soc. 86,1616–1626.

Hansch, C., Muir, R.M., Fujita, T., Maloney, P.P., Geiger, F., Streich, M.,1963. The correlation of biological activity of plant growth regulatorsand chloromycetin derivatives with Hammett constants and partitioncoefficients. J. Am. Chem. Soc. 85, 2817–2824.

Hitzel, L., Watt, A.P., Locker, K.L., 2000. An increased throughput methodfor the determination of partition coefficients. Pharm. Res. 17, 1389–1395.

Hubbell, W.L., McConnell, H.M., 1971. Molecular motion in spin-labeledphospholipids and membranes. J. Am. Chem. Soc. 93, 314–326.

Huuskonen, J.J., Livingstone, D.J., Tetko, I.V., 2000. Neural networkmodeling for estimation of partition coefficient based on atom-typeelectrotopological state indices. J. Chem. Inf. Comput. Sci. 40, 947–955.

Ikematsu, M., Iseki, M., Sugiyama, Y., Mizukami, A., 1996. Lipid bi-layer formation in a microporous membrane filter monitored by acimpedance analysis and purple membrane photoresponses. J. Elec-troanal. Chem. 403, 61–68.

Irvine, J.D., Takahashi, L., Lockhart, K., Cheong, J., Tolan, J.W., Selick,H.E., Grove, J.R., 1999. MDCK (Madin–Darby Canine Kidney) cells:a tool for membrane permeability screening. J. Pharm. Sci. 88, 28–33.

Jacobs, R.E., White, S.H., 1989. The nature of the hydrophobic bindingof small peptides at the bilayer interface: implications for the insertionof transbilayer helices. Biochemistry 28, 3421–3437.

Kaliszan, R., 1992. Quantitative structure–retention relationships. Anal.Chem. 64, 619A–631A.

Kaliszan, R., Nasal, A., Bucinski, A., 1994. Chromatographic hydropho-bicity parameter determined on an immobilized artificial membranecolumn: relationships to standard measures of hydrophobicity andbioactivity. Eur. J. Med. Chem. 29, 163–170.

Kaliszan, R., Kaliszan, A., Wainer, I.W., 1993. Deactivated hydrocarbona-ceous silica and immobilized artificial membrane stationary phases inhigh-performance liquid chromatographic determination of hydropho-bicities of organic bases: relationship to logP and CLOGP. J. Pharm.Biomed. Anal. 11, 505–511.

Kamlet, M.J., Doherty, R.M., Carr, P.W., Mackay, D., Abraham, M.H.,Taft, R.W., 1988. Linear solvation energy relationships. 44. Parameterestimation rules that allow accurate prediction of octanol/water parti-tion coefficients and other solubility and toxicity properties of poly-chlorinated biphenyls and polycyclic aromatic hydrocarbons. Environ.Sci. Technol. 22, 503–509.

Kamlet, M.J., Abraham, D.J., Doherty, R.M., Taft, R.W., Abraham, M.H.,1986. Solubility properties in polymers and biological media. 6. Anequation for correlation and prediction of solubilities of liquid organicnonelectrolytes in blood. J. Pharm. Sci. 75, 350–355.

Kamlet, M.J., Abboud, J.L.M., Abraham, M.H., Taft, R.W., 1983. Linearsolvation energy relationships. 23. A comprehensive collection ofthe solvatochromic parameters,�, �, and �, and some methods forsimplifying the generalized solvatochromic equation. J. Org. Chem.48, 2877–2887.

Karelson, M., 2000. Molecular Descriptors in QSAR/QSPR. Wiley, NewYork.

Karelson, M., Lobanov, V.S., Katritzky, A.R., 1996. Quantum-chemicaldescriptors in QSAR/QSPR studies. Chem. Rev. 96, 1027–1043.

Kelder, J., Grootenhuis, P.D.J., Bayada, D.M., Delbressine, L.P.C., Ploe-men, J.-P., 1999. Polar molecular surface as a dominating determinantfor oral absorption and brain penetration of drugs. Pharm. Res. 16,1514–1519.

Kennedy, T., 1997. Managing the drug discovery/development interface.Drug Discov. Today 2, 436–444.

Kier, L.B., Hall, L.H., 1999. Molecular Structure Description: The Elec-trotopological State. Academic Press, San Diego.


Kim, K.H., Greco, G., Novellino, E., 1998. A critical review of recentCoMFA applications. In: Kubinyi, H., Folkers, G., Martin, Y.C. (Eds.),3D QSAR in Drug Design. Kluwer Academic, Doldrecht, The Nether-lands, pp. 257–316.

Klopman, G., Li, J.-Y., Wang, S., Dimayuga, M., 1994. Computer au-tomated logP calculations based on an extended group contributionapproach. J. Chem. Inf. Comput. Sci. 34, 752–781.

Kontturi, K., Murtomäki, L., 1992. Electrochemical determination of par-tition coefficients of drugs. J. Pharm. Sci. 81, 970–975.

Krause, E., Dathe, M., Wieprecht, T., Bienert, M., 1999. Noncovalent im-mobilized artificial membrane chromatography, and improved methodfor describing peptide-lipid bilayer interactions. J. Chromatogr. A 849,125–133.

Krämer, S.D., 1999. Absorption prediction from physicochemical param-eters. Pharm. Sci. Technol. Today 2, 373–380.

Krämer, S.D., Wunderli-Allenspach, H., 1996. The pH-dependence in thepartitioning behaviour of (RS)-[3H] propranolol between MDCK celllipid vesicles and buffer. Pharm. Res. 13, 1851–1855.

Krämer, S.D., Braun, A., Jakits-Deiser, C., Wunderli-Allenspach, H., 1998.Towards the predictability of drug-lipid membrane interactions: thepH-dependent affinity of propranolol to phosphatidylionsitol contain-ing liposomes. Pharm. Res. 15, 739–744.

Kürschner, M., Nielsen, K., von Langen, J.R.G., Schenk, W.A., Zimmer-mann, U., Sukhorukov, V.L., 2000. Effect of fluorine substitution onthe interaction of lipophilic ions with the plasma membrane of mam-malian cells. Biophys. J. 79, 1490–1497.

Langner, M., Kubica, K., 1999. The electrostatics of lipid surfaces. Chem.Phys. Lipids 101, 3–35.

Larger, P., Altamura, M., Catalioto, R.-M., Giuliani, S., Maggi, C.A.,Valenti, C., Triolo, A., 2002. Simultaneous LC-MS/MS determinationof reference pharmaceuticals as a method for the characterization ofthe Caco-2 cell monolayer absorption properties. Anal. Chem. 74,5273–5281.

Law, B., Weir, S., Ward, N.A., 1992. Fundamental studies inreversed-phase liquid–solid extraction of basic drugs I: ionic interac-tions. J. Pharm. Biomed. Anal. 10, 167–179.

Leo, A.J., 2000. Evaluating hydrogen-bond donor strength. J. Pharm. Sci.89, 1567–1578.

Leo, A.J., 1993. Calculating logPoct from structures. Chem. Rev. 93,1281–1306.

Leo, A.J., Hoekman, D., 2000. Calculating logP(oct) with no missingfragments the problem of estimating new interaction parameters. Per-spect. Drug Discov. Des. 18, 19–38.

Leo, A., Hansch, C., Elkins, D., 1971. Partition coefficients and theiruses. Chem. Rev. 71, 525–616.

Liljeroth, P., Mälkiä, A., Cunnane, V.J., Kontturi, A.-K., Kontturi, K., 2000.Langmuir–Blodgett monolayers at a liquid–liquid interface. Langmuir16, 6667–6673.

Lipinski, C.A., Lombardo, F., Dominy, B.W., Feeney, P.J., 1997. Ex-perimental and computational approaches to estimate solubility andpermeability in drug discovery and development settings. Adv. DrugDeliv. Rev. 23, 3–25.

Liu, X.-Y., Nakamura, C., Yang, Q., Kamo, N., Miyake, J., 2002. Immobi-lized liposome chromatography to study drug–membrane interactions.Correlation with drug absorption in humans. J. Chromatogr. A 961,113–118.

Livingstone, D.J., 2000. The characterization of chemical structures usingmolecular properties. A survey. J. Chem. Inf. Comput. Sci. 40, 195–209.

Lobenberg, R., Amidon, G.L., 2000. Modern bioavailability, bioequiva-lence and biopharmaceutics classification system. New scientific ap-proaches to international regulatory standards. Eur. J. Pharm. Bio-pharm. 50, 3–12.

Lobell, M., Molnár, L., Keserü, G.M., 2003. Recent advances in theprediction of blood–brain partitioning from molecular structure. J.Pharm. Sci. 92, 360–370.

Lohmann, C., Huwel, S., Galla, H.J., 2002. Predicting blood–brain barrierpermeability of drugs: evaluation of different in vitro assays. J. DrugTarget 10, 263–276.

Loidl-Stahlhofen, A., Eckert, A., Hartmann, T., Schöttner, M., 2001.Solid-supported lipid membranes as a tool for determination of mem-brane affinity: high-throughput screening of a physicochemical param-eter. J. Pharm. Sci. 90, 599–606.

Luco, J.M., 1999. Prediction of the brain–blood distribution of a largeset of drugs from structurally derived descriptors using partialleast-squares (PLS) modeling. J. Chem. Inf. Comput. Sci. 39, 396–404.

Lundahl, P., Beigi, F., 1997. Immobilized liposome chromatography ofdrugs for model analysis of drug–membrane interactions. Adv. DrugDeliv. Rev. 23, 221–227.

Malkia, A., Liljeroth, P., Kontturi, K., 2003. Membrane activity of ion-isable drugs: a task for liquid–liquid electrochemistry? Electrochem.Commun. 5, 473–479.

Mannhold, R., van de Waterbeemd, H., 2001. Substructure and wholemolecule approaches for calculating logP. J. Comput. Aided Mol.Des. 15, 337–354.

Marcus, Y., Migron, Y., 1991. Polarity, hydrogen bonding, and structureof mixtures of water and cyanomethane. J. Phys. Chem. 95, 400–406.

Marrink, S.J., Berendsen, H.J.C., 1996. Permeation process of smallmolecules across lipid membranes studied by molecular dynamicssimulations. J. Phys. Chem. 100, 16729–16738.

Mayer, L.D., Bally, M.B., Hope, M.J., Cullis, P.R., 1985. Uptake ofdibucaine into large unilamellar vesicles in response to a membranepotential. J. Biol. Chem. 260, 802–808.

Meijer, L.A., Leermakers, F.A.M., Lyklema, J., 1999. Self-consistent-fieldmodelling of complex molecules with united atom detail in inho-mogenous systems. Cyclic and branched foreign molecules in dimyris-toylphosphatidylcholine membranes. J. Chem. Phys. 110, 6560–6579.

Meylan, W.M., Howard, P.H., 1995. Atom/fragment contribution methodfor estimating octanol–water partition coefficients. J. Pharm. Sci. 84,83–92.

Michael, D., Benjamin, I., 1995. Proposed experimental probe of theliquid/liquid interface structure: molecular dynamics of charge transferat the water/octanol interface. J. Phys. Chem. 99, 16810–16813.

Miyazaki, J., Hideg, K., Marsh, D., 1992. Interfacial ionization and par-titioning of membrane-bound local anaesthetics. Biochim. Biophys.Acta 1103, 62–68.

Morgan, M.E., Liu, K., Anderson, B.D., 1998. Microscale titrimetric andspectrophotometric methods for determination of ionization constantsand partition coefficients of new drug candidates. J. Pharm. Sci. 87,238–245.

Moriguchi, I., Hirono, S., Liu, Q., Nakagome, I., Matsushita, Y., 1992.Simple method of calculating octanol/water partition coefficient.Chem. Pharm. Bull. 40, 127–130.

Murtomäki, L., Kontturi, K., 2002. Correction to electrochemical deter-mination of partition coefficients of drugs. J Pharm. Sci. 91, 900–901[J. Pharm Sci 81 (1992) 970–974].

Mälkiä, A., Liljeroth, P., Kontturi, A.-K., Kontturi, K., 2001a. Electro-chemistry at lipid monolayer-modified liquid–liquid interfaces as animprovement to drug partitioning studies. J. Phys. Chem. B 105,10884–10892.

Mälkiä, A., Liljeroth, P., Kontturi, K., 2001b. Drug transfer throughbiomimetic Langmuir–Blodgett monolayers at a liquid–liquid inter-face. Anal. Sci. 17 (Suppl.), i345–i348.

Neubert, R., 1989. Ion-pair transport across membranes. Pharm. Res. 6,743–747.

Nirmalakhandan, N., Speece, R.E., 1988. Structure–activity relationships.Quantitative techniques for predicting the behaviour of chemicals inthe ecosystem. Environ. Sci. Technol. 22, 606–615.

Norinder, U., Haeberlein, M., 2002. Computational approaches to theprediction of the blood–brain distribution. Adv. Drug Deliv. Rev. 54,291–313.

Norinder, U., Österberg, T., 2001. Theoretical calculation and predictionof drug transport processes using simple parameters and partial least


squares projections to latent structures (PLS) statistics. The use ofelectrotopological state indices. J. Pharm. Sci. 90, 1076–1085.

Norinder, U., Österberg, T., Artursson, P., 1999. Theoretical calculationand prediction of intestinal absorption of drugs in humans usingMolSurf parametrization and PLS statistics. Eur. J. Pharm. Sci. 8,49–56.

Norinder, U., Österberg, T., Artursson, P., 1997. Theoretical calculationand prediction of Caco-2 cell permeability using MolSurf parametriza-tion and PLS statistics. Pharm. Res. 14, 1786–1791.

Nys, G.G., Rekker, R.F., 1973. Statistical analysis of a series of partitioncoefficients with special reference to the predictability of folding ofdrug molecules. Introduction of hydrophobic fragmental constants (fvalues). Chim. Ther. 8, 521–535.

Ollila, F., Halling, K., Vuorela, P., Vuorela, H., Slotte, J.P., 2002. Char-acterization of flavonoid-biomembrane interactions. Arch. Biochem.Biophys. 399, 103–108.

Österberg, T., Norinder, U., 2000. Prediction of polar surface area anddrug transport processes using simple parameters and PLS statistics.J. Chem. Inf. Comput. Sci. 40, 1408–1411.

Ong, S., Pidgeon, C., 1995. Thermodynamics of solute partitioning intoimmobilized artificial membranes. Anal. Chem. 67, 2119–2128.

Ong, S., Liu, H., Pidgeon, C., 1996. Immobilized-artificial-membranechromatography: measurements of membrane partition coefficient andpredicting drug membrane permeability. J. Chromatogr. A 728, 113–128.

Osakai, T., Ebina, K., 1998. Non-Bornian theory of the Gibbs energy ofion transfer between two immiscible liquids. J. Phys. Chem. B 102,5691–5698.

Osakai, T., Ogata, A., Ebina, K., 1997. Hydration of ions in organicsolvent and its significance in the Gibbs energy of ion transfer betweentwo immiscible liquids. J. Phys. Chem. B 101, 8341–8348.

Ottiger, C., Wunderli-Allenspach, H., 1999. Immobilized artificial mem-brane (IAM)-HPLC for partition studies of neutral and ionised acidsand bases in comparison with the liposomal partition system. Pharm.Res. 16, 643–650.

Pagliara, A., Carrupt, P.-A., Caron, G., Gaillard, P., Testa, B.,1997. Lipophilicity profiles of ampholytes. Chem. Rev. 97, 3385–3400.

Pagliara, A., Khamis, E., Trinh, A., Carrupt, P.-A., Tsai, R.-S., Testa,B., 1995. Structural properties governing retention mechanisms onRP-HPLC stationary phases used for lipophilicity measurements. J.Liq. Chromatogr. 18, 1721–1745.

Palm, K., Luthman, K., Ros, J., Gråsjö, J., Artursson, P., 1999. Effect ofmolecular charge on intestinal epithelial drug transport: pH-dependenttransport of cationic drugs. J. Pharmacol. Exp. Ther. 291, 435–443.

Palm, K., Luthman, K., Ungell, A.-L., Strandlund, G., Beigi, F., Lundahl,P., Artursson, P., 1998. Evaluation of dynamic polar molecular surfacearea as predictor of drug absorption: comparison with other compu-tational and experimental predictors. J. Med. Chem. 41, 5382–5392.

Palm, K., Stenberg, P., Luthman, K., Artursson, P., 1997. Polar molecularsurface properties predict the intenstinal absorption of drugs in humans.Pharm. Res. 14, 568–571.

Palm, K., Luthman, K., Ungell, A.-L., Strandlund, G., Artursson, P., 1996.Correlation of drug absorption with molecular surface properties. J.Pharm. Sci. 85, 32–39.

Parker, A.J., 1969. Protic-dipolar aprotic solvent effects on rates of bi-molecular reactions. Chem. Rev. 69, 1–32.

Parsegian, A., 1969. Energy of an ion crossing a low dielectric membrane:solutions to four relevant electrostatic problems. Nature 221, 844–846.

Petty, M.C., 1996. Langmuir–Blodgett Films: An Introduction. UniversityPress, Cambridge, p. 234.

Petty, M.C., Barlow, W.A., 1990. Film deposition. In: Roberts, G.G. (Ed.),Langmuir–Blodgett Films. Plenum Press, New York, pp. 93–132.

Pickett, S.D., McLay, I.M., Clark, D.E., 2000. Enhancing the hit-to-leadproperties of lead optimization libraries. J. Chem. Inf. Comput. Sci.40, 263–272.

Platts, J.A., Abraham, M.H., Zhao, Y.H., Hersey, A., Ijaz, L., Butina, D.,2001. Correlation and prediction of a large blood—brain distributiondata set—an LFER study. Eur. J. Med. Chem. 36, 719–730.

Platts, J.A., Abraham, M.H., Butina, D., Hersey, A., 2000. Estimationof molecular linear free energy relationship descriptors by a groupcontribution approach. 2. Prediction of partition coefficients. J. Chem.Inf. Comput. Sci. 40, 71–80.

Platts, J.A., Butina, D., Abraham, M.H., Hersey, A., 1999. Estimationof molecular linear free energy relation descriptors using a groupcontribution approach. J. Chem. Inf. Comput. Sci. 39, 835–845.

Prentis, R.A., Lis, Y., Walker, S.R., 1988. Pharmaceutical innovation bythe seven UK-owned pharmaceutical companies (1964–1985). Br. J.Clin. Pharmacol. 25, 387–396.

Reichel, A., Begley, D.J., 1998. Potential of immobilized artificial mem-branes for predicting drug penetration across the blood–brain barrier.Pharm. Res. 15, 1270–1274.

Rey, S., Caron, G., Ermondi, G., Gaillard, P., Pagliara, A., Carrupt, P.-A.,Testa, B., 2001. Development of molecular hydrogen-bonding poten-tials (MHBPs) and their application to structure–permeation relations.J. Mol. Graph. Model. 19, 521–535.

Reymond, F., 2001. Transfer mechanisms and lipophilicity of ionizabledrugs. In: Volkov, A.G. (Ed.), Liquid Interfaces in Chemical, Biologi-cal, and Pharmaceutical Applications. Marcel Dekker, New York, pp.729–773.

Reymond, F., Chopineaux-Courtois, V., Steyaert, G., Bouchard, G., Car-rupt, P-A., Testa, B., Girault, H.H., 1999a. Ionic partition diagramsof ionisable drugs: pH-lipophilicity profiles, transfer mechanisms andcharge effects on solvation. J. Electroanal. Chem. 462, 235–250.

Reymond, F., Steyaert, G., Carrupt, P.-A., Morin, D., Tillement, J.-P.,Girault, H.H., Testa, B., 1999b. The pH-partition profile of theanti-ischemic drug trimetazidine may explain its reduction of intra-cellular acidosis. Pharm. Res. 16, 616–624.

Reymond, F., Steyaert, G., Carrupt, P.-A., Testa, B., Girault, H., 1996a.Ionic partition diagrams: a potential–pH representation. J. Am. Chem.Soc. 118, 11951–11957.

Reymond, F., Steyaert, G., Pagliara, A., Carrupt, P.-A., Testa, B., Girault,H.H., 1996b. Transfer mechanism of ionic drugs: piroxicam as anagent facilitating proton transfer. Helv. Chim. Acta 79, 1651–1669.

Rhee, D., Markovich, R., Chae, W.G., Qiu, X., Pidgeon, C., 1994. Chro-matographic surfaces prepared from lyso phosphatidylcholine ligands.Anal. Chim. Acta 297, 377–386.

Rose, K., Hall, L.H., Kier, L.B., 2002. Modeling blood–brain barrierpartitioning using the electrotopological state. J. Chem. Inf. Comput.Sci. 42, 651–666.

Salminen, T., Pulli, A., Taskinen, J., 1997. Relationship between immo-bilized artificial membrane chromatographic retention and the brainpenetration of structurally diverse drugs. J. Pharm. Biomed. Anal. 15,469–477.

Samec, Z., Langmaier, J., Trojánek, A., 1996. Polarization phenomenaat the water-nitrophenyl octyl ether interface. Part 1. Evaluation ofthe standard Gibbs energies of ion transfer from the solubility andvoltammetric measurements. J. Electroanal. Chem. 409, 1–7.

Scherrer, R.A., 1984. The treatment of ionizable compounds in quantitativestructure-activity studies with special consideration to ion partitioning.ACS Symp. Ser. 255 (Pestic. Synth. Ration. Approaches), 225–246.

Seelig, A., Seelig, J., 1974. The dynamic structure of fatty acyl chains ina phospholipid bilayer measured by deuterium magnetic resonance.Biochemistry 13, 4839–4845.

Senda, M., Kakiuchi, T., Osakai, T., 1991. Electrochemistry at the interfacebetween two immiscible electrolyte solutions. Electrochim. Acta 36,253–262.

Sheng, Q., Schulten, K., Pidgeon, C., 1995. Molecular dynamics simula-tion of immobilized artificial membranes. J. Phys. Chem. 99, 11018–11027.

Spencer, J.N., Gleim, J.E., Blevins, C.H., Garrett, R.C., Mayer, F.J., 1979.Enthalpies of solution and transfer enthalpies. An analysis of the pure


base calorimetric method for the determination of hydrogen bondenthalpies. J. Phys. Chem. 83, 1249–1255.

Spessard, G.O., 1998. ACD Labs/LogP dB 3.5 and ChemSketch 3.5. J.Chem. Inf. Comput. Sci. 38, 1250–1253.

Stenberg, P., Norinder, U., Luthman, K., Artursson, P., 2001. Experimentaland computational screening models for the prediction of intestinaldrug absorption. J. Med. Chem. 44, 1927–1937.

Stenberg, P., Luthman, K., Artursson, P., 1999. Prediction of membranepermeability to peptides from calculated dynamic molecular surfaceproperties. Pharm. Res. 16, 205–212.

Stewart, B.H., Chan, O.H., 1998. Use of immobilized artificial membranechromatography for drug transport applications. J. Pharm. Sci. 87,1471–1478.

Stewart, B.H., Chung, F.Y., Tait, B., John, C., Chan, O.H., 1998. Hy-drophobicity of HIV protease inhibitors by immobilized artificial mem-brane chromatography: applications and significance to drug transport.Pharm. Res. 15, 1401–1406.

Steyaert, G., Lisa, G., Gaillard, P., Boss, G., Reymond, F., Girault, H.H.,Carrupt, P.-A., Testa, B., 1997. Intermolecular forces expressed in1,2-dichloroethane-water partition coefficients. A solvatochromic anal-ysis. J. Chem. Soc., Faraday Trans. 93, 401–406.

Sugawara, M., Takekuma, Y., Yamada, H., Kobayashi, M., Iseki, K.,Miyazaki, K., 1998. A general approach for the prediction of theintestinal absorption of drugs: regression analysis using the physic-ochemical properties and drug–membrane electrostatic interaction. J.Pharm. Sci. 87, 960–966.

Suhonen, T.M., Pasonen-Seppänen, S., Kirjavainen, M., Tammi, M.,Tammi, R., Urtti, A., 2003. The permeability barrier in organotypicculture model derived from rat epidermal keratinocytes. Eur. J. Pharm.Sci. 20, 107–113.

Suhonen, P., Järvinen, T., Koivisto, S., Urtti, A., 1998. Different effects ofpH on the permeation of pilocarpine and pilocarpine prodrugs acrossthe isolated rabbit cornea. Eur. J. Pharm. Sci. 6, 169–176.

Surewicz, W.K., Leyko, W., 1981. Interaction of propranolol with modelphospholipid membranes. Monolayer, spin label and fluorescence spec-troscopy studies. Biochim. Biophys. Acta 643, 387–397.

Syracuse Research Corporation, 301 Plainfield Road, Suite 350, Syracuse,New York 13212-2510,http://www.syrres.com/.

Taft, R.W., Berthelot, M., Laurence, C., Leo, A., 1996. Hydrogen bondsand molecular structure, CHEMTECH, 20–29 July.

Taft, R.W., Abraham, M.H., Famini, G.R., Doherty, R.M., Abboud, J.L.,Kamlet, M.J., 1985a. Solubility properties in polymers and biologicalmedia 5: an analysis of the physicochemical properties which influenceoctanol–water partition coefficients of aliphatic and aromatic solutes.J. Pharm. Sci. 74, 807–814.

Taft, R.W., Abraham, M.H., Doherty, R.M., Kamlet, M.J., 1985b. Lin-ear solvation energy relationships. 29. Solution properties of sometetraalkylammonium halide ion pairs and dissociated ions. J. Am.Chem. Soc. 107, 3105–3110.

Taillardat-Bertschinger, A., Carrupt, P.-A., Barbato, F., Testa, B., 2003.Immobilized artificial membrane HPLC in drug research. J. Med.Chem. 46, 655–665.

Taillardat-Bertschinger, A., Galland, A., Carrupt, P.-A., Testa, B., 2002a.Immobilized artificial membrane liquid chromatography: proposedguidelines for technical optimization of retention measurements. J.Chromatogr. A 953, 39–53.

Taillardat-Bertschinger, A., Marca Martinet, C.A., Carrupt, P.-A., Reist,M., Caron, G., Fruttero, R., Testa, B., 2002b. Molecular factors influ-encing retention on immobilized artificial membranes (IAM) comparedto partitioning in liposomes andn-octanol. Pharm. Res. 19, 729–737.

Takács-Novák, K., Szász, G., 1999. Ion-pair partition of quaternary am-monium drugs: the influence of counter ions of different lipophilicity,size and flexibility. Pharm. Res. 16, 1633–1638.

Tavelin, S., Taipalensuu, J., Hallböök, F., Vellonen, K.-S., Moore, V.,Artursson, P., 2003a. An improved cell culture model based on 2/4/A1cell monolayers for studies of intestinal drug transport: characterizationof transport routes. Pharm. Res. 20, 373–381.

Tavelin, S., Taipalensuu, J., Söderberg, L., Morrison, R., Chong, S., Arturs-son, P., 2003b. Prediction of the oral absorption of low-permeabilitydrugs using small intestine-like 2/4/A1 cell monolayers. Pharm. Res.20, 397–405.

Testa, B., Carrupt, P.-A., Gaillard, P., Billois, F., Weber, P., 1996.Lipophilicity in molecular modelling. Pharm. Res. 13, 335–343.

Toropainen, E., Ranta, V.P., Palmgren, J., Vellonen, K.S., Talvitie, A.,Suhonen, P., Hämäläinen, K.M., Auriola, S., Urtti, A., 2003. Paracel-lular and transcellular permeability in human corneal epithelial cellculture model. Eur. J. Pharm. Sci. 20, 99–106.

Tu, K., Tobias, D.J., Klein, M.L., 1995. Constant pressure and temperaturemolecular dynamics simulation of a fully hydrated liquid crystal phasedipalmitoylphosphatidylcholine bilayer. Biophys. J. 69, 2558–2562.

Vaes, W.H.J., Urrestarazu Ramos, E., Verhaar, H.J.M., Cramer, C.J., Her-mens, J.L.M., 1998. Understanding and estimating membrane/waterpartition coefficients: approaches to derive quantitative structure prop-erty relationships. Chem. Res. Toxicol. 11, 847–854.

Valko, K., Du, C.M., Bevan, C.D., Reynolds, D.P., Abraham, M.H., 2000.Rapid-gradient HPLC method for measuring drug interactions withimmobilized artificial membrane: comparison with other lipophilicitymeasures. J. Pharm. Sci. 89, 1085–1096.

van de Waterbeemd, H., 2000. Intestinal permeability: prediction fromtheory. Drugs Pharm. Sci. 106, 31–49.

van de Waterbeemd, H., Kansy, M., 1992. Hydrogen-bonding capacityand brain penetration. Chimia 46, 299–303.

van de Waterbeemd, H., Smith, D.A., Beaumont, K., Walker, D.K., 2001.Property-based design: optimization of drug absorption and pharma-cokinetics. J. Med. Chem. 44, 1313–1333.

van de Waterbeemd, H., Camenisch, G., Folkers, G., Chretien, J.R.,Raevsky, O.A., 1998. Estimation of blood–brain barrier crossing ofdrugs using molecular size and shape, and H-bonding descriptors. J.Drug Target. 6, 151–165.

Vanýsek, P., 1995. Charge transfer processes on liquid/liquid interfaces:the first century. Electrochim. Acta 40, 2841–2847.

Veber, D.F., Johnson, S.R., Cheng, H.-Y., Smith, B.R., Ward, K.W., Kop-ple, K.D., 2002. Molecular properties that influence the oral bioavail-ability of drug candidates. J. Med. Chem. 45, 2615–2623.

von Itzstein, M., Wu, W.-Y., Kok, G.B., Pegg, M.S., Dyason, J.C., Jin, B.,Phan, T.V., Smythe, M.L., White, H.F., Oliver, S.W., Colman, P.M.,Varghese, J.N., Ryan, D.M., Woods, J.M., Bethell, R.C., Hotham,V.J., Cameron, J.M., Penn, C.R., 1993. Rational design of potentsialidase-based inhibitors of influenza virus replication. Nature 363,418–423.

Walter, E., Kissel, T., 1995. Heterogeneity in the human intestinal cellline Caco-2 leads to differences in transepithelial transport. Eur. J.Pharm. Sci. 3, 215–230.

Weininger, D., 1988. SMILES, a chemical language and informationsystem. 1. Introduction to methodology and encoding rules. J. Chem.Inf. Comput. Sci. 28, 31–36.

Wessel, M.D., Jurs, P.C., Tolan, J.W., Muskal, S.M., 1998. Predictionof human intestinal absorption of drug compounds from molecularstructure. J. Chem. Inf. Comput. Sci. 38, 726–735.

Wils, P., Warnery, A., Phung-ba, V., Legrain, S., Scherman, D., 1994.High lipophilicity decreases drug transport across intestinal epithelialcells. J. Pharmacol. Exp. Ther. 269, 654–658.

Wilson, L.Y., Famini, G.R., 1991. Using theoretical descriptors in quan-titative structure-activity relationships: some toxicological indices. J.Med. Chem. 34, 1668–1674.

Wilson, M.A., Pohorille, A., 1996. Mechanism of unassisted ion transportacross membrane bilayers. J. Am. Chem. Soc. 118, 6580–6587.

Winiwarter, S., Ax, F., Lennernäs, H., Hallberg, A., Pettersson, C., Karlén,A., 2003. Hydrogen bonding descriptors in the prediction of humanin vivo intestinal permeability. J. Mol. Graph. Model. 21, 273–287.

Winiwarter, S., Bonham, N.M., Ax, F., Hallberg, A., Lennernäs, H.,Karlén, A., 1998. Correlation of human jejunal permeability (in vivo)of drugs with experimentally and theoretically derived parameters. Amultivariate data analysis approach. J. Med. Chem. 41, 4939–4949.

http://www.syrres.com/


Wohnsland, F., Faller, B., 2001. High-throughput permeability pH profileand high-throughput alkane/water logP with artificial membranes. J.Med. Chem. 44, 923–930.

Wold, S., 1979. Cross-validatory estimation of the number of componentsin factor and principal components models. Technometrics 20, 379–405.

Yang, C.Y., Cai, S.J., Liu, H., Pidgeon, C., 1996. Immobilized artificialmembranes: screens for drug membrane interactions. Adv. Drug Deliv.Rev. 23, 229–256.

Yee, S., 1997. In vitro permeability across Caco-2 cells (colonic) canpredict in vivo (small intestinal) absorption in man: fact or myth.Pharm. Res. 14, 763–766.

Young, R.C., Mitchell, R.C., Brown, T.H., Ganellin, C.R., Griffiths, R.,Jones, M., Rana, K.K., Saunders, D., Smith, I.R., Sore, N.E., Wilks,

T.J., 1988. Development of a new physicochemical model for brainpenetration and its application to the design of centrally acting H2receptor histamine antagonists. J. Med. Chem. 31, 656–671.

Zhao, Y.H., Abraham, M.H., Le, J., Hersey, A., Luscombe, C.N., Beck,G., Sherborne, B., Cooper, I., 2002. Rate-limited steps of human oralabsorption and QSAR studies. Pharm. Res. 19, 1446–1457.

Zhao, Y.H., Le, J., Abraham, M.H., Hersey, A., Eddershaw, P.J., Luscombe,C.N., Boutina, D., Beck, G., Sherborne, B., Cooper, I., Platts, J.A.,2001. Evaluation of human intestinal absorption data and subsequentderivation of a quantitative structure-activity relationship (QSAR) withthe Abraham descriptors. J. Pharm. Sci. 90, 749–784.

Zhu, C., Jiang, L., Chen, T.-M., Hwang, K.-K., 2002. A comparative studyof artificial membrane permeability assay for high throughput profilingof drug absorption potential. Eur. J. Med. Chem. 37, 399–407.

Drug permeation in biomembranes: in vitro and in silico prediction and influence of physicochemical properties

Documents