Evaluation of dynamic polar molecular surface area as predictor of drug absorption: Comparison with other computational and experimental predictors

Article abstract of DOI:10.1021/jm980313t

The relationship between various molecular descriptors and transport of drugs across the intestinal epithelium was evaluated. The monolayer permeability (P(c)) of human intestinal Caco-2 cells to a series of nine β- receptor-blocking agents was investigated in vitro. The dynamic polar molecular surface area (PSA(d)) of the compounds was calculated from all low- energy conformations identified in molecular mechanics calculations in vacuum and in simulated chloroform and water environments. For most of the investigated drugs, the effects of the different environments on PSA(d) were small. The exception was H 216/44, which is a large flexible compound containing several functional groups capable of hydrogen bonding (PSA(d,chloroform) = 70.8 A?² and PSA(d,water) = 116.6 A?²). The relationship between P(c) and PSA(d) was stronger than those between Pc and the calculated octanol/water distribution coefficients (log D(calc)) or the experimentally determined immobilized liposome chromatography (ILC) retention. P(c) values for two new practolol analogues and H 216/44 were predicted from the structure-permeability relationships of a subset of the nine compounds and compared with experimental values. The P(c) values of the two practolol analogues were predicted well from both PSA(d) calculations and ILC retention studies. The P(c) value of H 216/44 was reasonably well- predicted only from the PSA(d) of conformations preferred in vacuum and in water. The other descriptors overestimated the Pc of H 216/44 100-500-fold.

Full text of DOI:10.1021/jm980313t

5382
J . Med. Chem. 1998, 41, 5382-5392
Eva lu a tion of Dyn a m ic P ola r Molecu la r Su r fa ce Ar ea a s P r ed ictor of Dr u g
Absor p tion : Com p a r ison w ith Oth er Com p u ta tion a l a n d Exp er im en ta l
P r ed ictor s
Katrin Palm, Kristina Luthman,*,† Anna-Lena Ungell,‡ Gert Strandlund,§ Farideh Beigi,| Per Lundahl,| and
Per Artursson
Department of Pharmacy (Box 580), Department of Organic Pharmaceutical Chemistry (Box 574), and Department of
Biochemistry (Box 576), Uppsala University, BMC, SE-751 23 Uppsala, Sweden, and Pharmacokinetics and Drug Metabolism
and Medicinal Chemistry, Astra Ha¨ssle AB, SE-431 83 Mo¨lndal, Sweden
Received May 18, 1998
The relationship between various molecular descriptors and transport of drugs across the
intestinal epithelium was evaluated. The monolayer permeability (Pc) of human intestinal
Caco-2 cells to a series of nine â-receptor-blocking agents was investigated in vitro. The dynamic
polar molecular surface area (PSAd) of the compounds was calculated from all low-energy
conformations identified in molecular mechanics calculations in vacuum and in simulated
chloroform and water environments. For most of the investigated drugs, the effects of the
different environments on PSAd were small. The exception was H 216/44, which is a large
flexible compound containing several functional groups capable of hydrogen bonding
(PSAd,chloroform ) 70.8 Å2 and PSAd,water ) 116.6 Å2). The relationship between Pc and PSAd was
stronger than those between Pc and the calculated octanol/water distribution coefficients (log
D
calc) or the experimentally determined immobilized liposome chromatography (ILC) retention.
Pc values for two new practolol analogues and H 216/44 were predicted from the structure-
permeability relationships of a subset of the nine compounds and compared with experimental
values. The Pc values of the two practolol analogues were predicted well from both PSAd
calculations and ILC retention studies. The Pc value of H 216/44 was reasonably well-predicted
only from the PSAd of conformations preferred in vacuum and in water. The other descriptors
overestimated the Pc of H 216/44 100-500-fold.
In tr od u ction
experimental methods or based on calculated molecular
properties. Chromatographic methods are attractive for
screening purposes, since they require only small
amounts of solute, and it is also possible to automate
the process. The use of stationary phases that mimic
the lipid bilayer of cell membranes is expected to result
in stronger relationships with epithelial permeability
than seen when conventional octadecylsilyl silica (ODS)
columns are used. One approach is to use stationary
phases consisting of covalently bound phosphatidyl
moieties (immobilized artificial membranes, IAM).9,10
Alternatively, immobilized liposome chromatography
(ILC), which also incorporates the bilayer structure of
the cell membranes, may offer some advantages.11
The use of combinatorial chemistry and high through-
put screening in the field of drug discovery has increased
the demand for biopharmaceutical screening methods,
e.g., for prediction of oral absorption. Caco-2 cell
monolayers have been successfully used to model pas-
sive drug absorption,1,2 but the model has not yet been
fully adapted for screening purposes. The octanol/buffer
distribution coefficient (log D) therefore remains a
common tool for estimating the extent of drug perme-
ation across epithelia, but with varying results.1,3-5
Some of the shortcomings of using log D in modeling
epithelial permeability have been attributed to the
ability of the solute to form intermolecular hydrogen
bonds in both water and octanol.6 Therefore, the
hydrogen-bonding capacity of the molecule has been
introduced as an alternative predictor of passive drug
absorption.4,7,8 In addition to the relatively weak rela-
tionship between epithelial permeability and log D, the
traditional methods of measuring log D are time-
consuming and require relatively large amounts of the
substance under investigation.
Computational methods used to calculate various
physicochemical properties, such as log D, often account
for the three-dimensional shape of the molecule inad-
equately. Steric or conformational effects resulting from
the formation of intramolecular hydrogen bonds are not
taken fully into consideration with these methods. The
calculated molecular surface properties of a drug have
been shown to be related to many physicochemical
properties believed to be relevant to epithelial permea-
bility12-15 and uptake across the blood-brain barrier.16
Interestingly, such calculations could be extended to
include also the shape and flexibility of the molecule.17
More efficient drug permeability screening methods
are needed; these should be either less time-consuming
* To whom correspondence should be addressed. Present address:
Institute of Pharmacy, University of Tromsø, N-9037 Tromsø, Norway.
† Department of Organic Pharmaceutical Chemistry.
‡ Pharmacokinetics and Drug Metabolism.
In a previous study, we investigated the relationship
between molecular surface properties and epithelial
permeability for a series of six â-receptor-blocking
§ Medicinal Chemistry.
| Department of Biochemistry.
10.1021/jm980313t CCC: $15.00 © 1998 American Chemical Society
Published on Web 12/05/1998

Evaluation of Dynamic Polar Molecular Surface Area
J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27 5383
Sch em e 1. Synthesis of H 95/71 and H 244/45a
agents.18 These compounds were selected as model
compounds since their absorption after oral administra-
tion in humans and their lipophilicity displayed a wide
variability. Further, they have similar molecular weights
and pKa values, which minimized the influence of these
parameters on the results. The surface properties were
found to vary with the molecular conformation. There-
fore, to account for the flexibility of the molecules,
dynamic averages of molecular surface properties were
calculated from all low-energy conformations identified
in conformational analyses. We found promising in-
verse linear relationships between the polar part of the
dynamic molecular surface area (PSAd) and the intes-
tinal epithelial permeability to the drugs in two in vitro
models. These relationships were stronger than those
between calculated log D values (log Dcalc) and perme-
ability.
a
(a) Propionic anhydride or formic acid; (b) epichlorohydrin,
KOH; (c) 2-propylamine.
Resu lts a n d Discu ssion
Further studies, on 20 structurally diverse but con-
formationally rigid compounds, suggested that intesti-
nal drug absorption in humans after oral administration
could be predicted from PSAd.19 PSAd was also shown
to be related to the hydrogen-bonding capacity of the
compounds. Other theoretical approaches have been
used to explore models based on a large number of
calculated structural properties using PLS methodol-
ogy20,21 or considerably simpler but still powerful clas-
sification systems.22 In these studies the hydrogen-
bonding/polar component has generally been the dom-
inating factor in predicting epithelial permeability to
drugs.
Ch em istr y. Two practolol analogues, H 95/71 and
H 244/45, were synthesized using the synthetic pathway
described in Scheme 1. Treatment of 4-aminophenol
with formic acid or propionic anhydride afforded the
acylated derivatives 1 and 2, respectively. Alkylation
of 1 or 2 using epichlorohydrin in the presence of base
followed by an epoxide ring opening using isopropyl-
amine produced the target compounds.
Effects of Differ en t Solva tion Mod els Used in
Con for m a t ion a l An a lyses. A. Con for m a t ion a l
P r efer en ces. A large number of low-energy conforma-
tions (∆Es e 2.5 kcal/mol) were identified for all
compounds except H 216/44 in the Monte Carlo confor-
mational searches. In general, slightly fewer conform-
ers were identified in the simulated chloroform envi-
ronment compared to vacuum, and most compounds
proved to be most flexible in simulated water. (For
detailed information about the number of low-energy
conformations identified in each environment, see the
Supporting Information.) Surprisingly, only one low-
energy conformation resulted from an extensive confor-
mational search of H 216/44 in the water environment
(see Experimental Section), while 36 and 6 conforma-
tions were found within 2.5 kcal/mol of the global
minimum in the vacuum and chloroform environments,
respectively. H 216/44 has many rotatable bonds, and
the use of continuous solvent simulations dramatically
increased the amount of CPU time needed to obtain
reliable conformational information. However, all con-
formations obtained from the vacuum and chloroform
Monte Carlo searches which were within the search
window (7 kcal/mol) were energy-minimized in the
water environment to further validate the result. None
of these conformations were found to be within 2.5 kcal/
mol of the global minimum conformation identified in
the Monte Carlo searches of H 216/44 in simulated
water.
In this paper we extend our evaluation of PSAd as a
predictor of passive epithelial permeability. The first
objective was to investigate whether the use of simu-
lated solvent environments instead of vacuum in the
Monte Carlo-based conformational searches would in-
fluence the calculated dynamic surface properties of
nine model drugs and subsequently improve the cor-
relations with epithelial permeability. The molecular
mechanics calculations were therefore performed in
simulated chloroform or water environments and com-
pared to the results from calculations in vacuum.
The second objective was to compare more closely the
two computational predictors PSAd and log Dcalc. The
descriptors were used in predictions of the permeability
of Caco-2 cell monolayers (Pc) to two synthesized ana-
logues of practolol (H 95/71 and H 244/45) from the
structure-permeability relationships of the previously
studied model drugs (in ref 18). H 95/71 and H 244/45
were designed to display similar polar surface area to
that of practolol but to differ in their distribution in
octanol/water. The predicted permeabilities were then
compared with experimental results. In addition, a
structurally different â-receptor antagonist (H 216/44)
previously shown to be an outlier in the relationship
between log D and cell monolayer permeability23 was
investigated.
The conformational preferences of atenolol, practolol,
pindolol, metoprolol, oxprenolol, and alprenolol identi-
fied by the Monte Carlo search in vacuum corresponded
well with those obtained previously by a stepwise
conformational search strategy.18 The conformational
preferences of the 2-hydroxy-3-(isopropylamino)propoxy
substituent of H 95/71 and H 244/45 were similar to
those of practolol. Many conformations were stabilized
by intramolecular hydrogen bonds between both the
amine and alcohol, and the alcohol and ether function-
The third objective was to compare the computational
PSAd method with a new experimental chromatographic
screening method, using immobilized liposome chroma-
tography.11 In ILC, the stationary phase consists of
liposomes immobilized in porous gel beads. Retention
of solute on these columns is assumed to reflect an
interaction of the solute with a lipid bilayer.

5384 J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27
Ta ble 1. Structure of the â-Adrenoreceptor-Blocking Agents
Palm et al.
a
From ref 46.
alities. However, amide bond conformations differed;
in H 95/71 a cis-amide conformation was preferred,
whereas trans-conformations were favored in practolol
and H 244/45.24
B. Dyn a m ic Su r fa ce P r op er ties. The low-energy
conformations identified for each compound varied in
their surface properties (see Figures 2 and 3 and
Supporting Information). The variations in polar sur-
face area (differences between the largest and smallest
PSA calculated for each compound) ranged from 11 to
47 Å2 and corresponded to 6-54% of the dynamic
averages (Table 2).26 This was not a random variability
caused by poor precision in the surface area calculations
but was directly attributable to the influence of confor-
mation.27 The dynamic averages of the polar surface
area resulting from conformational analyses in vacuum
(PSAd,vacuum), simulated chloroform (PSAd,chloroform), and
simulated water (PSAd,water) environments are presented
in Table 2.28 In general, the PSAd increased in the order
vacuum < chloroform < water as a result of changes in
intramolecular hydrogen-bonding patterns. Although
clear differences were observed in conformational pref-
erences using different simulated solvent environments,
the effect on the dynamic average of the molecular
surface properties was relatively small for all com-
pounds but H 216/44 (Table 2). The dynamic polar
Only small differences in the conformational prefer-
ences of compounds B-I (Table 1) could be observed
between results from conformational analyses per-
formed in vacuum and those performed in simulated
chloroform. However, low-energy conformations identi-
fied in simulated water showed marked differences in
the intramolecular hydrogen-bonding pattern from those
seen in the other solvation models. Fewer intramolecu-
lar hydrogen bonds were observed in water, indicating
an increased interaction of polar groups with the
solvent. Interestingly, a preference for conformations
stabilized by aliphatic-aromatic interactions was also
noted in water. The conformational preferences of H
216/44 clearly differed between vacuum and the two
solvents. With this molecule, folded conformations
stabilized by intramolecular hydrogen bonds were iden-
tified in vacuum and chloroform, whereas more stretched
conformations were preferred in water (Figure 1).25

Evaluation of Dynamic Polar Molecular Surface Area
J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27 5385
F igu r e 3. Relationship between PSAd,water of the nine com-
pounds and their Caco-2 cell monolayer permeability coef-
ficients. The insert shows the correlation between log Pc and
PSAd,water. The dashed line shows the resulting linear correla-
tion when H 216/44 (open circle) is excluded from the data
set. Error bars represent SD of Pc and range in PSAd,water
.
F igu r e 1. Global minimum conformations of H 216/44 found
in Monte Carlo-based conformational searches in (a) vacuum
and (b) simulated water environments. Polar areas are dis-
played in darker colors.
functional groups participating in inter- and intra-
molecular hydrogen bonds, and the PSA of H 216/44 is
therefore sensitive to conformational influences (Figure
1).
Rela tion sh ip s betw een Ep ith elia l P er m ea bility
a n d P SAd . There was wide variability in the cell
monolayer permeability coefficients of the nine inves-
tigated compounds (0.104 × 10-6 to 242 × 10-6 cm/s;
Table 2). The resulting relationship between epithelial
permeability and PSAd,vacuum for the compounds is
shown in Figure 2. The insert shows the correlation
between log Pc and PSAd,vacuum (r2 ) 0.907, q2 ) 0.810,
n ) 9). If H 216/44 is excluded from the data set, the
correlation is slightly improved (r2 ) 0.968, q2 ) 0.942,
n ) 8; the dashed line in the Figure 2 insert), indicating
a stronger relationship for compounds that are more
similar structurally. For compounds B-I, the relation-
ships between Pc and PSAd resulting from the different
solvation models were not qualitatively different; they
appeared rather to be only slightly shifted along the
PSAd axis (Figures 2 and 3, Table 4). When H 216/44
was included in the data set, however, a weaker
relationship was obtained between PSAd,chloroform and Pc
than between PSAd,vacuum and Pc. In contrast, the
relationship between PSAd,water and Pc was unaffected
by the inclusion of H 216/44 (Table 4, Figure 3).
P SAd a n d log Dca lc in P r ed iction s of Ca co-2 Cell
Mon ola yer P er m ea bility. As expected, the three
analogues, practolol, H 95/71, and H 244/45, displayed
comparable polar surface areas: PSAd,vacuum ranged
from 71.6 to 77.0 Å2 (Table 2). The calculated octanol/
water distribution coefficients were similar for H 95/71
and practolol, but a higher lipophilicity, comparable to
that of pindolol and metoprolol, was predicted for
H 244/45 from these calculations. However, the meas-
ured log D values of the three analogues were of the
same magnitude (Table 2). Thus, in this case, the
F igu r e 2. Relationship between PSAd,vacuum of the nine
compounds and their Caco-2 cell monolayer permeability
coefficients. The insert shows the correlation between log Pc
and PSAd,vacuum. The dashed line shows the resulting linear
correlation when H 216/44 (open circle) is excluded from the
data set. Error bars represent SD of Pc and range in
PSAd,vacuum
.
surface area of H 216/44 varied extensively with the
solvation model used: 88.7, 70.8, and 116.6 Å2 in the
vacuum, chloroform, and water environments, respec-
tively. These results reflected the larger size and
flexibility of H 216/44. The molecule contains many

5386 J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27
Palm et al.
Ta ble 2. Physicochemical Properties, Structural Descriptors, and Caco-2 Cell Monolayer Permeability Coefficients of the
â-Adrenoreceptor Antagonists
b
c
PSAd,vacuum
(Å2)
PSAd,chloroform
(Å2)
PSAd,water
(Å2)
SAd,vacuum
(Å2)
log Ks
Pc
compound
log Dcalc
log Dm
(M-1
)
(10-6 cm/s)
A. H216/44
B. atenolol
C. H95/71
D. practolol
E. H244/45
F. pindolol
G. metoprolol
H. oxprenolol
I. alprenolol
88.7
90.9
77.0
73.4
71.6
56.5
53.2
46.7
37.1
70.8
91.6
77.7
74.3
72.7
57.4
53.9
49.1
38.6
116.6
92.6
77.7
74.1
72.9
57.5
53.9
49.6
39.1
574.8
370.4
347.8
372.3
396.9
340.7
395.4
373.4
365.4
-0.61
-2.31
-1.34
-1.35
-0.82
-0.63
-1.11
-0.41
0.40
-0.10a
-2.04a
-1.56
-1.31a
-1.08
-0.55a
-0.42a
0.08a
1.09
0.37
0.62
0.72
0.68
1.61
1.15
1.63
2.44
0.104 ( 0.016
1.02 ( 0.10
3.75 ( 0.34
3.46 ( 0.53
6.03 ( 0.26
54.7 ( 0.6
91.9 ( 4.0d
119.6 ( 6.7
242 ( 14d
0.85a
a
b
d
From ref 46. n ) 3; relative SD e 10%. c n ) 6-8. From ref 18.
Ta ble 3. Predicted Caco-2 Cell Monolayer Permeability Coefficients of the Two Practolol Analogues, H 95/71 and H 244/45, and H
216/44 from Relationships between Structural Descriptors and log Pc Derived from Data of the Six Original Model Drugsa
predicted Pc from linear model (10-6 cm/s)
(predicted Pc/experimental Pc)
predicted Pc from sigmoidal model (10-6 cm/s)
(predicted Pc/experimental Pc)
H 95/71
H 244/45
H 216/44
H 95/71
H 244/45
H 216/44
experimental Pc
PSAd,vacuum
3.75
4.39
(1.2)
4.41
(1.2)
4.66
(1.2)
11.6
(3.1)
4.61
(1.2)
6.03
7.87
(1.3)
7.67
(1.3)
7.93
(1.3)
34.0
(5.6)
5.40
(0.9)
0.104
1.23
(12)
9.54
(92)
0.065
(0.6)
53.3
(512)
15.8
(152)
3.75
2.40
(0.64)
2.42
(0.65)
2.39
(0.64)
4.24
(1.1)
1.70
6.03
4.56
(0.76)
4.35
(0.72)
4.16
(0.69)
116
(19)
2.52
(0.42)
0.104
1.09
(10)
5.74
(55)
0.74
(7)
117
b
c
PSAd,chloroform
d
PSAd,water
e
log Dcalc
(1125)
66.6
(640)
f
log Ks
(0.45)
a
b
Original model drugs: atenolol, practolol, pindolol, metoprolol, oxprenolol, and alprenolol.18 log Pc ) 4.274-0.047 × PSAd,vacuum (r2
) 0.964, n ) 6). c log Pc ) 4.390-0.048 × PSAd,chloroform (r2 ) 0.967, n ) 6). log Pc ) 4.385-0.048 × PSAd,water (r2 ) 0.964, n ) 6). e log
d
Pc ) 2.283 + 0.912 × log Dcalc (r2 ) 0.805, n ) 6). f log Pc ) -0.042 + 1.139 × log Ks (r2 ) 0.808, n ) 6).
Ta ble 4. Resulting r2, q2, and RMSE Values from Fitting Linear or Sigmoidal Models to the Structure-log Pc Dataa
linear model
sigmoidal modelb
descriptor
r2
q2
RMSE
r2
q2
RMSE
PSAd,vacuum
PSAd,chloroform
PSAd,water
log Dcalc
0.968/0.907
0.971/0.650
0.967/0.979
0.725/0.269
0.840/0.485
0.942/0.810
0.944/0.495
0.936/0.965
0.613/< 0
0.17/0.36
0.16/0.70
0.17/0.17
0.49/1.01
0.38/0.85
0.993/0.926
0.995/0.787
0.995/0.988
0.746/0.286
0.948/0.774
0.966/0.722
0.980/0.475
0.980/0.861
<0/<0
0.09ns/0.38ns
0.08ns/0.65ns
0.08ns/0.15ns
0.58ns/1.18ns
0.26ns/0.67ns
log Ks
0.644/0.324
0.768/< 0
a
b
Data from regression analyses excluding/including H 216/44 in the data set are presented. ns, not significantly improved fit compared
to linear model (p > 0.05, F-test for equality of variances).
fragmental method used to calculate log D exaggerated
the difference in lipophilicity between the analogues.
From the linear relationship between PSAd,vacuum and
log Pc of the original model drugs (atenolol, practolol,
pindolol, metoprolol, oxprenolol, and alprenolol in ref
18), the cell monolayer permeability to H 95/71 and H
244/45 was predicted to be 4.39 × 10-6 and 7.87 × 10-6
cm/s, respectively (Table 3). This was in good agree-
ment with the observed Pc values of 3.75 × 10-6 and
6.03 × 10-6 cm/s, respectively (Table 2). The Pc values
predicted from the PSAd,chloroform and PSAd,water of
H 95/71 and H 244/45 were similar (Table 3). How-
ever, when the corresponding relationship between log
Dcalc and log Pc was used, epithelial permeabilities to
H 95/71 and H 244/45 were overestimated, with pre-
dicted Pc values of 11.6 × 10-6 and 34.0 × 10-6 cm/s,
respectively (Table 3).
( 0.016 × 10-6 cm/s). In contrast to H 95/71 and H
244/45, the Pc value of H 216/44 predicted from the data
of the original model drugs was dependent on the
solvation environment of the conformational analysis.
Thus, the predicted Pc of H 216/44 from PSAd,chloroform
was 9.54 × 10-6 cm/s as compared to 0.065 × 10-6 cm/s
from PSAd,water (Table 3). This clearly demonstrates the
influence of conformational preferences on the polar
surface properties of large, flexible molecules with
several functional groups that can participate in hydro-
gen bonding. However, further studies are needed
before it can be stated that the molecular surface
properties in a water environment better describe the
epithelial permeability.
log Dcalc was inferior to PSAd (independent of solvation
treatment) as a descriptor of the epithelial transport of
H 216/44, predicting a Pc of 53.3 × 10-6 cm/s, or 500-
fold higher than that observed (Table 3). The relation-
ship between log Dcalc and Caco-2 permeability is shown
in Figure 4. Inclusion of H 216/44 abolished the
correlation between log Dcalc and log Pc (from r2 ) 0.725,
q2 ) 0.613, n ) 8 to r2 ) 0.269, q2 < 0, n ) 9). This
result cannot be attributed to the theoretical approach
The cell monolayer permeability to H 216/44 was not
as well-predicted from the log Pc vs PSAd,vacuum relation-
ship as those of H 95/71 and H 244/45. From PSAd,vacuum
) 88.7 Å2 for H 216/44, a Pc value of 1.23 × 10-6 cm/s
was predicted (Table 3). This is about 12-fold higher
than the observed cell monolayer permeability (0.104

Evaluation of Dynamic Polar Molecular Surface Area
J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27 5387
F igu r e 4. Relationship between log Dcalc of the nine com-
pounds and their Caco-2 cell monolayer permeability coef-
ficients. The insert shows the correlation between log Pc and
log Dcalc. The dashed line shows the resulting linear correlation
when H 216/44 (open circle) is excluded from the data set.
Error bars represent SD of Pc.
F igu r e 5. Relationship between log Ks of the nine compounds
and their Caco-2 cell monolayer coefficients. The insert shows
the correlation between log Pc and log Ks. The dashed line
shows the resulting linear correlation when H 216/44 (open
circle) is excluded from the data set. Error bars represent SD
of Pc.
of calculating the distribution coefficient since the
measured log D value (log Dm) of H 216/44 at pH 7.4 is
close to the calculated value. The log Dm value of H
216/44 is similar to those of metoprolol and oxprenolol
(Table 2). However, these compounds permeate Caco-2
cell monolayers approximately 1000 times more easily
than H 216/44. This illustrates that single physico-
chemical predictors are often insufficient to describe the
structure-permeability relationship of a more hetero-
geneous set of compounds (see below).
The predictions of epithelial permeability cited above
have been made assuming a linear relationship between
the structural descriptors and log Pc. The relationships
between the logarithm of the transport rate and struc-
tural descriptors such as lipophilicity are, however, often
reported to be sigmoidal1,29 or, when very lipophilic
compounds are included, parabolic.30 In the sigmoidal
relationship, the lower plateau represents the paracel-
lular permeability, which is an aqueous pathway. In
the present study, no clear evidence of a lower plateau
was observed in the investigated permeability range.
The upper plateau is often attributed to the limiting
effect of the aqueous boundary layer on drug diffusion
in vitro.31 In this study, cell monolayer permeabilities
unbiased by the aqueous boundary layer were used, and
an upper plateau was therefore not evident. Fitting a
four-parameter sigmoidal model to the data did not
significantly improve any of the studied structure-
absorption relationships (Table 4). Similarly, no im-
provement was obtained in the prediction of monolayer
permeability to H 95/71, H 244/45, and H 216/44 from
relationships obtained with the original set of model
drugs18 (Table 3).
account for the charge of the compounds. The unex-
pectedly low permeability of H 216/44 cannot be ex-
plained by charge, since this compound is predicted to
be charged to a lesser extent at pH 7.4 than the other
compounds (pKa 8.5 compared to pKa 9.5-9.7 for all
other compounds). Instead, the larger size of H 216/44
(MW 479 compared to MW 248-267; SAd 575 Å2
compared to SAd 341-397 Å2) could in part explain its
low epithelial transport.32 First, it has been proposed
that the permeability of a membrane to molecules is
directly proportional to the diffusion coefficient of the
molecules33 and subsequently to 1/(molecular volu-
me)n.34 Second, assuming that molecular weight is
proportional to molecular volume, epithelial perme-
ability has been corrected for molecular size using
factors such as MW1/3 35 MW,1,10 and MW1.4 36
. Accord-
,
ingly, the larger size of H 216/44 compared to the other
compounds would account for a 1.3-2.5-fold size-de-
pendent reduction in Pc. Thus, the molecular size of H
216/44 only provides a marginal contribution to the
1000-fold lower cell monolayer permeability to
H 216/44 compared to the other compounds of compa-
rable lipophilicity. Instead, it is more likely that the
hydrogen-bonding/polar properties of H 216/44 explain
most of its limited epithelial transport.
Im m obilized Lip osom e Ch r om a togr a p h y. The
relationship between specific capacity factor on ILC
columns (log Ks) and Caco-2 cell monolayer permeability
was not as strong as those between the PSAd descriptors
and Pc (Figure 5). Practolol and the two analogues, H
95/71 and H 244/45, displayed comparable retention
properties on the ILC columns (Table 2). As for PSAd,
the predicted cell monolayer permeabilities of H 95/71
and H 244/45 (obtained from the correlation between
log Ks and log Pc of the original model drugs) were in
good agreement with the Pc values determined experi-
mentally (Table 3). However, log Ks of H 216/44
Apart from hydrogen bonding and lipophilicity, the
charge and the size of molecules can also influence their
transepithelial transport. These factors are not in-
cluded in the PSAd parameter. log D, however, does

5388 J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27
Palm et al.
predicted a Pc value of 15.8 × 10-6 cm/s, which is about
150-fold higher than that observed. When H 216/44 was
included in the data set, the relationship between log
Pc and log Ks weakened significantly (from r2 ) 0.840,
q2 ) 0.644, n ) 8 to r2 ) 0.485, q2 ) 0.324, n ) 9; Figure
5 and Table 4). Similar results were obtained by fitting
a sigmoidal model to the data (Table 4). It is possible
that better relationships can be obtained after optimiza-
tion of the phospholipid composition of the immobilized
liposomes.40 The retention on membrane mimicking
chromatographic columns has previously been shown
to be closely correlated to log D.37 Also in this study,
log Ks was strongly correlated to experimentally deter-
mined log D values (log Dm, Table 2; r2 ) 0.92).
However, the relationship between log Ks and log Pc was
somewhat stronger than that between log Dm and log
Pc when H 216/44 was included.
the Pc of H 216/44 92-fold. Including H 216/44 in the
data set weakened the correlation between PSAd,chloroform
and Pc dramatically. These results indicate that while
single molecular descriptors such as PSAd may be good
predictors of epithelial permeability to series of drug
molecules with limited structural differences, they may
be insufficient for drugs of greater structural diversity.
However, in all cases, PSAd was a better predictor of
drug transport across the intestinal epithelium than
calculated log D values or experimentally determined
log Ks values obtained using immobilized liposome
chromatography. These results motivate further stud-
ies on how PSAd and other molecular surface properties
determine membrane permeability for structurally dif-
ferent drugs.
Exp er im en ta l Section
Syn th esis of H 95/71. N-(4-Hyd r oxyp h en yl)for m a m id e
(1). 4-Aminophenol (49.1 g, 0.45 mol) was dissolved in formic
acid (337.5 mL). The solution was heated to reflux for 1 h
and was then allowed to cool to room temperature. The formic
acid was evaporated off. The residue was partitioned between
EtOAc (500 mL) and 2 M HCl (250 mL). The organic phase
was washed with 2 M HCl (250 mL), H2O (250 mL), and brine
(250 mL) and dried (MgSO4). Evaporation of the volatiles gave
40 g (64.8%) of 1 as a solid. Mp: 138-139 °C. Compound 1
was used in the next step without further purification.
N-[4-(2,3-Ep oxyp r op oxy)p h en yl]for m a m id e (3). KOH
(5.8 g, 90.7 mmol) was dissolved in H2O (150 mL), and
compound 1 (15 g, 90.7 mmol) was added. Epichlorohydrin
(8.81 g, 95.2 mmol) was thereafter added dropwise to the dark
solution. The mixture was stirred at room temperature for 6
h. EtOAc (300 mL) was added, and the phases were separated.
The organic phase was washed with H2O (200 mL) and brine
(100 mL). After drying (MgSO4) the volatiles were evaporated
off. The residual brownish oil was purified by column chro-
matography on silica gel using heptane/EtOAc (3:2) as eluent
to afford 8.0 g (45.6%) of 3. Compound 3 was used in the next
step without further purification.
In contrast to PSAd, log Ks is expected to account for
the distribution of charged and uncharged species since
it is measured at the pH of the transport experiments.
However, the influence of charge on the retention on
membrane-like columns is unclear. It has been shown
that retention on IAM columns decreases with ioniza-
tion of acidic groups but is not affected by the ionization
of basic groups.38 Further, distribution of ion pairs into
liposomes has been suggested to decrease the effects of
molecular charge on liposome partitioning.39
The ILC retention of a solute results from a combina-
tion of electrostatic interactions with the bilayer surface,
partitioning into the lipid bilayer and partitioning into
the interior of the immobilized liposomes. Although
electrostatic interactions are weak for small solutes at
physiological ionic strength,40 a large hydrogen-bonding
capacity will not necessarily prohibit an interaction with
the lipid bilayer surface and subsequent retention on
the ILC column. For example, peptides have been
shown to be able to remain solvated at the membrane
interface.41 Thus, retention on an ILC column will not
always reflect transport across a cell membrane. This
might explain the discrepancy between the retention on
ILC columns and cell monolayer permeability found in
this study, especially for H 216/44, and the stronger
relationships obtained with the PSAd descriptors.
N-[4-(2-Hyd r oxy-3-(isop r op yla m in o)p r op oxy)p h en yl]-
for m a m id e (H 95/71). Compound 3 (8.0 g, 41.4 mmol) and
2-propylamine (3.55 mL, 41.4 mmol) were dissolved in EtOH
(30 mL) in a sealed glass ampule. The mixture was stirred at
100 °C for 2 h and was then allowed to cool to room
temperature. The volatiles were evaporated off. The residue
was purified by column chromatography on silica gel using
CH2Cl2/MeOH/NH3(aq) (10:1.5:0.15) as eluent to afford 2.9 g
(27.8%) of pure product. Mp: 121-122 °C. 1H NMR (270
MHz) (CDCl3) (1:1 mixture of amide rotamers): δ 8.49 (br s,
0.5H), 8.34 (d, 0.5H, J ) 1.7 Hz), 7.43 (d, 1H, J ) 9.0 Hz),
7.01 (d, 1H, J ) 9.0 Hz), 6.91 (d, 1H, J ) 8.8 Hz), 6.88 (d, 1H,
J ) 9.0 Hz), 4.12-4.01 (m, 1H), 3.96-3.98 (m, 2H), 2.96-2.71
(m, 5H), 1.13 (dd, 6H, J ) 6.37, 1.64 Hz). Anal. Calcd. for
Con clu sion s
The results of this study provide further support for
the hypothesis that molecular surface properties deter-
mine cell membrane permeability and, hence, epithelial
permeability to drug molecules. More specifically, the
molecular descriptor PSAd, which is related to the
hydrogen-bonding capacity, was found to be a good
predictor of the intestinal epithelial permeability to an
analogous series of â-receptor-blocking agents. In gen-
eral, the influence on PSAd of using simulated solvent
environments in conformational analyses was small.
Good correlations between PSAd and Pc were established
in all environments (vacuum, water, and chloroform).
The exception was H 216/44 which is an outlier with
regard to molecular properties. H 216/44 is a larger and
more flexible molecule capable of forming several intra-
and intermolecular hydrogen bonds. The Pc of H 216/
44 was excellently predicted by the PSA of the extended
conformation preferred in the simulated water environ-
ment. The PSAd obtained in chloroform overestimated
C
13H20N2O3: C, 61.8; H, 8.0; N, 11.2. Found: C, 61.6; H, 7.7;
N, 11.0.
Syn th esis of H 244/45. N-(4-Hyd r oxyp h en yl)p r op ion -
a m id e (2). Propionic anhydride (33.1 g, 0.254 mol) was added
to a suspension of 4-aminophenol (21.8 g, 0.2 mol) in H2O (60
mL). A homogeneous solution was obtained after heating to
reflux for 10 min. A precipitate was formed when the solution
was allowed to cool to room temperature. The precipitate was
filtered off and washed with 4 × 75 mL of H2O. The white
solid was dried in vacuo over P2O5 to obtain 29.8 g (90.2%) of
2. Compound 2 was used in the next step without further
purification.
N-[4-(2,3-Epoxypr opoxy)ph en yl]pr opion am ide (4). KOH
(11.5 g, 0.18 mol) was dissolved in H2O (300 mL). Compound
2 (29.8 g, 0.18 mol) was added, and a clear solution was
obtained after a few minutes. Epichlorohydrin was added
dropwise. A white precipitate was observed after ca. 30 min.
The mixture was stirred overnight at room temperature. The
solid material was filtered off, washed with 4 × 100 mL H2O,

Evaluation of Dynamic Polar Molecular Surface Area
J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27 5389
and dried in vacuo over P2O5 to obtain 38.2 g (95.9%) of 4.
Compound 4 was used in the next step without further
purification.
calculated according to the method by Hansch and co-workers44
using CLOGP v 4.42 (Daylight C.I.S. Inc., Irvine, CA). The
distribution coefficients at pH 7.4 (log D) were computed from
the CLOGP values and experimentally determined pKa val-
ues.45,46
N-[4-(2-Hyd r oxy-3-(isop r op yla m in o)p r op oxy)p h en yl]-
p r op ion a m id e (H 244/45). Compound 4 (15.0 g, 67.8 mmol)
and 2-propylamine (5.8 mL, 67.8 mmol) were dissolved in
EtOH (40 mL) in a sealed glass ampule. The mixture was
stirred at 100 °C for 2 h and was then allowed to cool to room
temperature. The volatiles were evaporated off. Dry diethyl
ether (300 mL) was added to the residue to obtain a white
solid which was filtered off. Purification by column chroma-
tography on silica gel using CH2Cl2/MeOH/NH3(aq) (10:1:0.1)
as eluent afforded 9.5 g (50.0%) of pure product. Mp: 135-
138 °C. 1H NMR (270 MHz) (CDCl3): δ 7.39 (d, 2H, J ) 8.9
Hz, H2′, H2′′), 7.11 (br s, 1H, amide NH), 6.86 (d, 2H, J ) 8.9
Hz, H3′, H3′′), 4.02 (m, 1H, H2), 3.95 (app d, 2H, H1), 2.91
(dd, 1H, J ) 3.66, -11.96 Hz, H3a), 2.87 (hept, 1H, J ) 6.35
Hz, NCH), 2.74 (dd, 1H, J ) 7.90, -11.96 Hz, H3b), 2.54 (br
s, 2H, NH, OH), 2.37 (q, 2H, J ) 7.57 Hz, CH3CH2), 1.24 (t,
3H, J ) 7.57 Hz, CH3CH2), 1.12 (d, 6H, J ) 6.35, CHCH3).
Anal. Calcd. for C15H24N2O3‚0.1H2O: C, 63.8; H, 8.6; N, 9.9.
Found: C, 63.8; H, 8.5; N, 9.8.
Ca lcu la tion of Molecu la r Su r fa ce P r op er ties. A. Con -
for m a tion a l An a lysis. Conformational analyses of all nine
â-receptor-blocking agents (see Table 1) were performed by
molecular mechanics calculations, using the MM2 force field
as implemented in the MacroModel program (v 4.5).42 Sys-
tematic pseudo-Monte Carlo conformational search procedures
were used for each compound in vacuum, simulated chloro-
form, and simulated water environments.
D. p Ka a n d Oct a n ol/Wa t er Dist r ib u t ion Coefficien t
Mea su r em en ts. The octanol/water partition coefficients of
H 95/71 and H 244/45 were determined experimentally using
a titrimetric method. For all other compounds experimentally
determined values of log D (log Dm) were taken from litera-
ture.46 The apparatus, a Sirius PC101 pKa and log P analyzer
(Sirius Analytical Instruments Ltd., Forest Row, England),
was equipped with an Orion Ross semimicro combination pH
electrode (model 8103SC, ATI Orion, Boston, MA). The
titration vessel was a closed system containing an electrode,
a stirrer, a temperature probe, a precision dispenser, and a
tube for introduction of nitrogen gas. Separate titrations were
performed to estimate pKa in water and the apparent pKa in
the presence of n-octanol (pKa′) in a two-phase system of water
and n-octanol. The difference between pKa and pKa′ is related
to log P.47 Weighed samples of H 95/71 and H 244/45 (1-2
mg), and 1.0 mL of n-octanol in the pKa′ titrations, were added
to a vial before it was placed into the temperature-controlled
vessel (25 °C). Water adjusted for ionic strength (0.15 M KCl;
10-15 mL) was added, and the solutions were acidified with
0.5 M HCl to a pH of 2. The mixture was stirred until all of
the sample was dissolved. Titrations were performed under
a nitrogen blanket with 0.5 M KOH as titrant up to pH 11. To
obtain the equilibrium constants, nonlinear regression analysis
of the difference curves47 was performed on the data sets using
the pKa LOGP software (v 4.0; Sirius Analytical Instruments).
Two to four 5000-step conformational searches were carried
out in each environment. Starting structures with two dif-
ferent configurations of the amino group and when appropriate
cis- and trans-conformations of amide groups were used. The
energy window was set to 7 kcal/mol. Energy minimizations
were made using the truncated Newton algorithm, a maximum
of 3000 iterations, and a convergence criterion of 0.01 kJ /Å‚
mol. The conformational analysis of H 216/44 in vacuum was
performed using 100 000 steps in order to account for the
extensive conformational freedom of this molecule. Since the
calculations took considerably longer in simulated chloroform
and water environments than in vacuum, it was not possible
to analyze the conformational space of H 216/44 as exhaus-
tively in these environments. Six different starting conforma-
tions were used in 5000-step searches in simulated chloroform.
A similar procedure was used for calculations in simulated
water, but because the calculations were even slower in this
environment, nine different starting conformations and 1000
steps were used.
Dr u g Tr a n sp or t Stu d ies. A. Cell Cu ltu r e. Caco-2 cells
were obtained from American Type Culture Collection (Rock-
ville, MD). The cells were maintained in Dulbecco’s modified
Eagle’s medium (DMEM), containing 10% heat-inactivated
fetal calf serum (FCS) and 1% nonessential amino acids, in
an atmosphere of 90% air and 10% CO2 as described else-
where;48 5 × 105 cells (passage number 93-105) were seeded
on polycarbonate filter inserts (Transwell Costar, Badhoeve-
dorp, The Netherlands; mean pore size 0.45 µm, diameter 12
mm) and cultivated in DMEM supplemented with 10% FCS,
1% nonessential amino acids, penicillin (100 units/mL), and
streptomycin (100 µg/mL).23 All tissue culture media were
obtained from Gibco through Laboratorie Design AB, Lidingo¨,
Sweden. The cells were allowed to grow and differentiate for
22-32 days before the monolayers were used in drug transport
experiments.
B. Dr u g Tr a n sp or t Exp er im en ts. In general, drugs
were dissolved in Hanks’ balanced salt solution containing 25
mM Hepes buffer (HBSS; pH 7.4) to a final concentration of
1-2 mM. In addition, H 216/44 was used also at 5 mM in
order to saturate active transport processes (see below). The
drug solutions (pH 7.4) were added to the donor side of the
monolayers, and HBSS without drug was added to the receiver
side. The monolayers were incubated at 37 °C in a humidified
atmosphere. At regular time intervals (4-40 min depending
on transport rate) samples were withdrawn from the receiving
For all compounds, an additional energy minimization with
a lower convergence criterion (0.001 kJ /Å‚mol) was performed
on the resulting conformations from all searches. The con-
formations of each compound were then combined, and dupli-
cate conformers were removed. Only the (S)-enantiomers were
considered, since enantiomers have identical surface proper-
ties.
B. Dyn a m ic Su r fa ce Ar ea Ca lcu la tion s. The van der
chamber and frozen (-20 °C) pending HPLC analysis.
A
Waals’ surface areas of all low-energy conformations (∆Es
e
calibrated plate shaker was used to agitate the monolayers,
and transport experiments were carried out at both low- and
high-stirring rates.31 The integrity of the monolayers was
routinely checked by determination of their permeability to
the hydrophilic marker [14C] mannitol.49 The baseline appar-
ent permeability of mannitol was 0.12 ( 0.04 × 10-6 cm/s (n
) 67). No significant effect on integrity was observed at the
drug concentrations and stirring rates used, compared to
control experiments. The transport rates were also deter-
mined in both apical-to-basolateral (a-b) and basolateral-to-
apical (b-a) directions in order to detect any significant active
transport of the compounds.
2.5 kcal/mol) of the nine investigated compounds were calcu-
lated using an in-house program. The analytical algorithm
was developed to reduce the effect of spatial orientation found
in numerical calculations and to increase speed and precision
in the surface area calculations. The program calculates the
free surface area of each atom bounded by the arcs of the
overlapping atoms using the Gauss-Bonnet theorem.43 The
molecular surface area was divided into a polar fraction
(nitrogen atoms, oxygen atoms, and hydrogen atoms attached
to these heteroatoms) and a nonpolar fraction. The dynamic
averages of the surface properties were then calculated from
all low-energy conformations according to a Boltzmann dis-
tribution at 37 °C.17,18
In general, the Pc ratio (b-a)/(a-b) was in the vicinity of 1,
indicating that the prevailing transport mechanism was
passive diffusion. However, a (b-a)/(a-b) ratio of approxi-
mately 30 was observed for H 216/44 at 1 mM,50 which was
C. Ca lcu la tion of Octa n ol/Wa ter Distr ibu tion Coef-
ficien ts. The logarithms of the octanol/water partition coef-
ficients (log P) for the nine investigated compounds were

5390 J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27
Palm et al.
of an interval and the concentration on the donor side at the
beginning of that interval. The cell monolayer permeability
(Pc), unbiased by the aqueous boundary layer, was calculated
from the relationship between Papp determined at the different
stirring rates (V) and V, as described previously:31
V/Papp ) 1/K + (1/Pc + 1/Pf)‚V
(2)
where Pf is the calculated permeability coefficient of the filter
support and K is a constant.
Im m obilized Lip osom e Ch r om a togr a p h y. A. P r ep a -
r a tion a n d Im m obiliza tion of Lip osom es. Superdex 200
(preparative grade) and a glass column (HR 5/5) were bought
from Pharmacia Biotech (Uppsala, Sweden) and phosphati-
dylcholine (PC; hens’ egg, 95% purity) was obtained from
Avanti Polar Lipids (Alabaster, AL). Chemicals were of
analytical grade. The procedures for liposome immobilization
and drug retention analysis were essentially as described
earlier.11 In brief, liposomes (1.5 mL, 150 µmol of phospho-
lipid), prepared by rehydration of a PC film in 150 mM NaCl,
1 mM Na2EDTA, and 10 mM Tris-HCl, pH 7.4, were mixed
with dried Superdex 200 gel beads (110 mg) and were
immobilized by gel bead swelling followed by freezing and
thawing to effect liposome fusion.54,55 The material was packed
in the 5-mm (i.d.) glass column to a 1.0-mL gel bed. Phos-
phorus analysis showed 47.6 µmol of phospholipid before
packing and 43.9 µmol after the runs.55 The average value
was used in eq 3 below.
F igu r e 6. Transport of H 216/44 across Caco-2 cell monolay-
ers in the apical-to-basolateral and basolateral-to-apical direc-
tions in the absence or presence of the P-glycoprotein substrate
verapamil (0.5 mM). Error bars represent SD of Pc.
taken as an indication of a significant active secretory com-
ponent.51,52 To obtain a better estimate of the passive absorp-
tion component, Caco-2 permeability to H 216/44 was also
measured at a higher drug concentration (5 mM), with or
without addition of 0.5 mM verapamil, an inhibitor of P-
glycoprotein-mediated secretion52 (Figure 6). Increasing the
concentration of H 216/44 to 5 mM without verapamil did not
reduce the (b-a)/(a-b) ratio.53 However, after addition of
verapamil, the (b-a)/(a-b) ratio decreased from 32 to 2.5,
which was comparable to that of practolol. Consequently, the
apical-to-basolateral Caco-2 cell monolayer permeability coef-
ficient of H 216/44 measured at 5 mM in the presence of 0.5
mM verapamil apically was used in the analysis of structure-
absorption relationships.
B. Ch r om a togr a p h ic P r oced u r e. The mobile phase
consisted of 150 mM NaCl in 10 mM sodium phosphate buffer,
pH 7.4. Aliquots of the nine investigated drugs in the eluent
(10-50 µL, 0.1 mg/mL) were analyzed at 0.80 ( 0.02 mL/min
at 23 °C with detection at 220 nm.
C. Sp ecific Ca p a city F a ctor (Ks). The capacity factor,
Ks, for a drug was calculated using
Ks ) (VR - V0)/A
(3)
C. An a lytica l Meth od s. The samples from the drug
transport studies were analyzed using a reversed-phase HPLC
system. The system consisted of a Perkin-Elmer isocratic LC
pump 250, a Perkin-Elmer advanced LC sample processor ISS-
200, and a Spectra Physics UV100 detector. The integration
software Chromatography Station for Windows was used in
the analysis of H 216/44. For all other compounds a Spectra
Physics SP4270 integrator was used. A Hichrom Partisil
ODS3 (100 × 4.6 mm) was used as analytical column and a
Hichrom Partisil ODS3 (10 × 3.2 mm) was used as guard
column, both with a mean particle size of 5 µm, for all
compounds but H 216/44. The analytical column used in the
analysis of H 216/44 was a Beckman Ultrasphere ODS (250
× 4.6 mm) with a particle size of 5 µm. The mobile phase was
composed of phosphate buffer (pH 3.0; 60 mM KH2PO4, 8 mM
H3PO4) and acetonitrile; 10-30% acetonitrile was used to
obtain retention times of 3-12 min at flow rates of 0.5-1.0
mL/min. Injection volumes were 10 to 200 µL. Samples
containing [14C]mannitol were counted in a liquid scintillation
spectrometer (Tricarb 1900 CA, Packard Instruments).
D. P er m ea bility Coefficien ts. All rate constants were
obtained under “sink” conditions (i.e., the concentration of the
drug on the receiver side never exceeded 10% of the concentra-
tion of the drug on the donor side), and the mass balance was
checked at the end of all transport experiments. The observed
mass balance was in general >95% and never <85%. The
apparent permeability coefficient (Papp, cm/s) was calculated
from:
where VR is the retention volume of the drug which was read
at the peak maximum, V0 is the retention volume of the
reference substance (K2Cr2O7), and A is the amount of im-
mobilized phospholipid (see above). The units were chosen to
obtain Ks in M-1
.
D. Sta tistica l An a lysis. Values are given as mean ( one
standard deviation (SD). Unpaired Student’s t-test or ANOVA,
when appropriate, was used to determine statistical differences
between mean values. Variances were compared using Schef-
fe’s F-test. P < 0.05 was considered as significant. The
coefficient of determination (r2), the (leave-one-out) cross-
validated r2 (q2), and the root-mean-square error (RMSE) were
used as measures of goodness-of-fit in regression analyses. The
sum of squared residuals was minimized in nonlinear regres-
sion analyses to fit the four-parameter sigmoidal equation:
ycalc ) (ymax - ymin)/(1 + (x/x50)γ) + ymin
(4)
to the data, where x is the structural property (PSAd, log D,
or log Ks), x50 is the value of x at ymin + (ymax - ymin)/2, and γ
is a slope factor. When the permeability data were not
logarithmically transformed, a weighting scheme of 1/ycalc2 was
used in the regression analyses.
Ack n ow led gm en t. We would like to thank Magnus
Ko¨ping-Ho¨ggård for skillful assistance with the trans-
port experiments and J ohan Gråsjo¨ for valuable math-
ematical assistance. This work was supported by Grant
97-46 from Centrala Fo¨rso¨ksdjursna¨mnden, Grant 9478
from The Swedish Medical Research Council, Grant
K11163-302 from The Swedish Natural Science Re-
search Council, and The Swedish Fund for Scientific
Research without Animals.
Papp ) (k‚VR)/(A‚60)
(1)
where k is the slope obtained by linear regression of cumula-
tive fraction absorbed as a function of time (min), VR is the
volume in the receiver chamber (mL), and A is the area of the
membrane (cm2). The fraction absorbed was defined as the
ratio between the concentration on the receiver side at the end

Evaluation of Dynamic Polar Molecular Surface Area
J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27 5391
(24) This is in contrast with results obtained by NMR spectroscopy
in CDCl3 which indicate a 1:1 mixture of amide rotamers in H
95/71. This prevented a more detailed analysis of the spectra
due to overlapping peaks. NMR spectra of H 244/45 were more
informative indicating preferences for conformations stabilized
by intramolecular hydrogen bonding between the amine and
alcohol functionalities. No hydrogen bond stabilization involving
the ether oxygen was identified. This is in contrast with the
results from the Monte Carlo conformational search which gave
low-energy conformations including both these hydrogen bonds
to about the same extent.
(25) The use of a folded conformation as a starting geometry in a
conformational search in simulated water resulted in stretched
conformations. The opposite was also observed: using a stretched
conformation as starting geometry in simulated chloroform
resulted in folded conformations.
Su p p or tin g In for m a tion Ava ila ble: Table of results
from the conformational analyses and surface area calculations
(1 page). Ordering information is given on any current
masthead page.
Refer en ces
(1) Artursson, P.; Karlsson, J . Correlation between oral drug
absorption in humans and apparent drug permeability coef-
ficients in human intestinal epithelial (Caco-2) cells. Biochem.
Biophys. Res. Commun. 1991, 175, 880-885.
(2) Lennerna¨s, H.; Palm, K.; Fagerholm, U.; Artursson, P. Com-
parison between active and passive drug transport in the human
intestinal epithelial Caco-2 cells in vitro and human jejunum in
vivo. Int. J . Pharm. 1996, 127, 103-107.
(3) Martin, Y. C. A practitioner’s perspective of the role of quantita-
tive structure-activity analysis in medicinal chemistry. J . Med.
Chem. 1981, 24, 229-237.
(4) Young, R. C.; Mitchell, R. C.; Brown, T. H.; Ganellin, C. R.;
Griffiths, R.; J ones, M.; Rana, K. K.; Saunders, D.; Smith, I. R.;
Sore, N. E.; Wilks, T. J . Development of a new physicochemical
model for brain penetration and its application to the design of
centrally acting H2 receptor histamine antagonists. J . Med.
Chem. 1988, 31, 656-671.
(5) Schoenwald, R. D.; Huang, H.-S. Corneal penetration behavior
of beta-blocking agents I: Physicochemical factors. J . Pharm.
Sci. 1983, 72, 1266-1271.
(6) El Tayar, N.; Tsai, R.-S.; Testa, B.; Carrupt, P.-A.; Leo, A.
Partitioning of solutes in different solvent systems: The con-
tribution of hydrogen-bonding capacity and polarity. J . Pharm.
Sci. 1991, 80, 590-598.
(7) Conradi, R. A.; Hilgers, A. R.; Ho, N. F. H.; Burton, P. S. The
influence of peptide structure on transport across Caco-2 cells.
Pharm. Res. 1991, 8, 1453-1460.
(8) Conradi, R. A.; Hilgers, A. R.; Ho, N. F. H.; Burton, P. S. The
Influence of Peptide Structure on Transport Across Caco-2 Cells.
II. Peptide Bond Modification Which Results in Improved
Permeability. Pharm. Res. 1992, 9, 435-439.
(9) Ong, S.; Qiu, X.; Pidgeon, C. Solute interactions with immobilized
artificial membranes. J . Phys. Chem. 1994, 98, 10189-10199.
(10) Pidgeon, C.; Ong, S.; Liu, H.; Qiu, X.; Pidgeon, M.; Dantzig, A.
H.; Munroe, J .; Hornback, W. J .; Kasher, J . S.; Glunz, L.;
Szczerba, T. IAM chromatography: an in vitro screen for
predicting drug membrane permeability. J . Med. Chem. 1995,
38, 590-594.
(11) Beigi, F.; Yang, Q.; Lundahl, P. Immobilized-liposome chromato-
graphic analysis of drug partitioning into lipid bilayers. J .
Chromatogr. A 1995, 704, 315-321.
(12) Amidon, G. L.; Yalkowsky, S. H.; Anik, S. T.; Valvani, S. C.
Solubility of nonelectrolytes in polar solvents. V. Estimation of
the solubility of aliphatic monofunctional compounds in water
using a molecular surface area approach. J . Phys. Chem. 1975,
79, 2239-2246.
(13) Dunn, W. J ., III; Koehler, M. G.; Grigoras, S. The role of solvent-
accessible surface area in determining partition coefficients. J .
Med. Chem. 1987, 30, 1121-1126.
(14) Barlow, D.; Satoh, T. The design of peptide analogues for
improved absorption. J . Contr. Relat. 1994, 29, 283-291.
(15) Ooi, T.; Oobatake, M.; Ne´methy, G.; Scheraga, H. A. Accessible
surface areas as a measure of the thermodynamic parameters
of hydration of peptides. Proc. Natl. Acad. Sci. U.S.A. 1987, 84,
3086-3090.
(16) van de Waterbeemd, H.; Kansy, M. Hydrogen-bonding capacity
and brain penetration. Chimia 1992, 46, 299-303.
(17) Lipkowitz, K. B.; Baker, B.; Larter, R. Dynamic molecular
surface areas. J . Am. Chem. Soc. 1989, 111, 7750-7753.
(18) Palm, K.; Luthman, K.; Ungell, A.-L.; Strandlund, G.; Artursson,
P. Correlation of drug absorption with molecular surface proper-
ties. J . Pharm. Sci. 1996, 85, 32-39.
(26) The smallest variation in PSA of all low-energy conformations,
11.5 Å2, was found for pindolol (largest PSA ) 61.8 Å2; smallest
PSA ) 50.3 Å2) and the largest variation, 47 Å2, was displayed
by H 216/44 (largest PSA ) 116.6 Å2; smallest PSA ) 69.2 Å2).
The low-energy conformations of atenolol identified in chloroform
displayed the smallest relative variation in PSA (∆PSA/PSAd )
5.4 Å2/91.6 Å2 ) 6%), whereas the largest relative variation was
found for alprenolol in water (∆PSA/PSAd ) 20.9 Å2/39.2 Å2
54%).
)
(27) The low-energy conformations were divided into groups accord-
ing to ref 18. The groups displayed different polar surface
properties (p < 0.001; ANOVA), and the variation within groups
was small.
(28) The polar surface areas reported in this paper differ slightly from
those reported in ref 18. In the earlier paper we used the Lee
and Richard-based numerical algorithm57 included in the PC-
MODEL program (v 4.0), whereas in this paper (as well as in
our more recent publication, ref 19) we used an in-house
computer program based on an analytical algorithm. In general
the polar surface areas calculated with the analytical algorithm
were larger (2-4 Å2) than the numerically calculated areas. This
is in agreement with the grid-based numerical algorithm in
PCMODEL and the observation that the computed surface areas
get larger when a smaller grid size is used in PCMODEL.
However, for atenolol and practolol the polar surface areas
calculated according to the analytical algorithm were smaller
(4-5 Å2) than the polar areas calculated by PCMODEL. After
closer inspection this could be attributed to the fact that the
carbon atoms in the carbonyl groups (accounting for approxi-
mately 7 Å2 of the total area) were included in the polar surface
area obtained from PCMODEL.
(29) van de Waterbeemd, H.; Camenisch, G.; Folkers, G.; Raevsky,
O. A. Estimation of Caco-2 cell permeability using calculated
molecular descriptors. Quant. Struct.-Act. Relat. 1996, 15, 480-
490.
(30) Wils, P.; Warnery, A.; Phung-Ba, V.; Legrain, S.; Scherman, D.
High lipophilicity decreases drug transport across intestinal
epithelial cells. J . Pharmacol. Exp. Ther. 1994, 269, 654-658.
(31) Karlsson, J .; Artursson, P. A method for the determination of
cellular permeability coefficients and aqueous boundary layer
thickness in monolayers of intestinal epithelial (Caco-2) cells
grown in permeable filter chambers. Int. J . Pharm. 1991, 71,
55-64.
(32) Depending on whether the epithelial transport is transcellular
and/or paracellular, the larger molecular size decreases cell
monolayer permeability by either diffusional restriction and/or
the sieving effect of the paracellular pathway. The Caco-2 cell
monolayer is a relatively tight epithelium with a low paracellular
permeability.2,56 In addition, the active secretion of H 216/44 in
the absence of verapamil suggests efflux of H 216/44 across the
apical cell membrane of the epithelial cells52 (see Experimental
Section). Thus, one can assume that H 216/44 is to a significant
extent transported transcellularly and that the size effect would
result from a higher diffusional restriction.
(19) Palm, K.; Stenberg, P.; Luthman, K.; Artursson, P. Polar
molecular surface properties predict the intestinal absorption
of drugs in humans. Pharm. Res. 1997, 14, 568-571.
(20) The multivariate partial least-squares projections to latent
structures (PLS) method was used in: Norinder, U.; O¨ sterberg,
T.; Artursson, P. Theoretical calculation and prediction of Caco-2
cell permeability using MolSurf parametrization and PLS
statistics. Pharm. Res. 1997, 14, 1785-1790.
(21) Basak, S. C.; Gute, B. D.; Drewes, L. R. Predicting blood-brain
transport of drugs: A computational approach. Pharm. Res.
1996, 13, 775-778.
(22) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J .
Experimental and computational approaches to estimate solubil-
ity and permeability in drug discovery and development settings.
Adv. Drug Deliv. Rev. 1996, 23, 3-25.
(33) Stein, W. D. The molecular basis of diffusion across cell
membranes. The movement of molecules across cell membranes;
Academic Press: New York, 1967.
(34) The value of n is dependent on the environment. Stokes-
Einstein stated a value of 1/3 for spherical molecules. Later
experimental values of 0.6-0.7 have been described. For poly-
mers and in lipid bilayers a steeper size dependence has been
observed, and values of 1-3 have been assigned to n.36
(35) Rubas, W.; Cromwell, M.; Gadek, T.; Narindray, D.; Mrsny, R.
Structural elements which govern the resistance of intestinal
tissues to compound transport. Mater. Res. Soc. Symp. Proc.
1994, 331, 179-185.
(36) Xiang, T.-X.; Anderson, B. D. The relationship between permeant
size and permeability in lipid bilayer membranes. J . Membr.
Biol. 1994, 140, 111-122.
(23) Artursson, P. Epithelial transport of drugs in cell culture. I: A
model for studying the passive diffusion of drugs over intestinal
absorptive (Caco-2) cells. J . Pharm. Sci. 1990, 79, 476-482.

5392 J ournal of Medicinal Chemistry, 1998, Vol. 41, No. 27
Palm et al.
(37) Barbato, F.; La Rotonda, M. I.; Quaglia, F. Chromatographic
indexes on immobilized artificial membranes for local anesthe-
tics: relationships with activity data on closed sodium channels.
Pharm. Res. 1997, 14, 1699-1705.
(38) Salminen, T.; Pulli, A.; Taskinen, J . Relationship between
immobilised artificial membrane chromatographic retention and
the brain penetration of structurally diverse drugs. J . Pharm.
Biomed. Anal. 1997, 15, 469-477.
(39) Avdeef, A.; Box, K. J .; Comer, J . E. A.; Hibbert, C.; Tam, K. Y.
pH-metric logP 10. Determination of liposomal membrane-water
partition coefficients of ionizable drugs. Pharm. Res. 1998, 15,
209-215.
(40) Beigi, F.; Gottschalk, I.; Lagerquist Ha¨gglund, C.; Haneskog,
L.; Brekkan, E.; Zhang, Y.; O¨ sterberg, T.; Lundahl, P. Im-
mobilized liposome and biomembrane partitioning chromatog-
raphy of drugs for prediction of drug transport. Int. J . Pharm.
1998, 164, 129-137.
(41) J acobs, R. E.; White, S. H. The nature of the hydrophobic binding
of small peptides at the bilayer interfaces: implications for the
insertion of transbilayer helices. Biochemistry 1989, 28, 3421-
3437.
(42) Mohamadi, F.; Richards, N. G. J .; Guida, W. C.; Liskamp, R.;
Lipton, M.; Caufield, C.; Chang, G.; Hendrickson, T.; Still, W.
C. MacroModelsAn integrated software system for modelling
organic and bioorganic molecules using molecular mechanics.
J . Comput. Chem. 1990, 11, 440-467.
(43) The program M-AREA v 1.1 used for surface area calculations
is available upon request from the authors (contact johan.grasjo@
galenik.uu.se). The program is available free of charge for
academic users.
(44) Leo, A.; J ow, P. Y. C.; Silipo, C.; Hansch, C. Calculation of
hydrophobic constant (log P) from π and f constants. J . Med.
Chem. 1975, 18, 865-868.
(45) Mannhold, R.; Dross, K. P.; Rekker, R. F. Drug lipophilicity in
QSAR practice: I. A comparison of experimental with calculative
approaches. Quant. Struct.-Act. Relat. 1990, 9, 21-28.
(46) Experimentally determined values of pKa and log Dm were taken
from: Craig, P. N. Drug Compendium. In Cumulative Subject
Index & Drug Compendium, 1st ed.; Drayton, C. J ., Ed.;
Pergamon Press: Oxford, 1990; pp 237-991; with the exception
of pKa and log Dm of H 216/44: personal communication with
Kurt-J o¨rgen Hoffman, Astra Ha¨ssle AB, Sweden.
(47) Avdeef, A. pH-Metric log P. Part 1. Difference plots for determin-
ing ion-pair octanol-water partition coefficients of multiprotic
substances. Quant. Struct.-Act. Relat. 1992, 11, 510-517.
(48) Artursson, P.; Karlsson, J .; Ocklind, G.; Schipper, N. Studying
transport processes in absorptive epithelia. In Cell Culture
Models of Epithelial Tissues - a Practical Approach; Shaw, A.
J ., Ed.; Oxford University Press: New York, 1996; pp 111-133.
(49) Anderberg, E. K.; Nystro¨m, C.; Artursson, P. Epithelial transport
of drugs in cell culture. VII: Effects of pharmaceutical surfactant
excipients and bile acids on transepithelial permeability in
monolayers of human intestinal epithelial (Caco-2) cells. J .
Pharm. Sci. 1992, 81, 879-887.
(50) The receiver concentrations in the a-b direction were close to
or below the limit of detection resulting in only approximate
permeability values.
(51) Karlsson, J .; Kuo, S.-M.; Ziemniak, J .; Artursson, P. Transport
of celiprolol across human intestinal epithelial (Caco-2) cells:
mediation of secretion by multiple transporters including P-
glycoprotein. Br. J . Pharmacol. 1993, 110, 1009-1016.
(52) Burton, P. S.; Conradi, R. A.; Ho, N. F. H.; Hilgers, A. R.;
Borchardt, R. T. How structural features influence the biomem-
brane permeability of peptides. J . Pharm. Sci. 1996, 85, 1336-
1340.
(53) Both a-b and b-a epithelial permeability to H 216/44 was
affected by the addition of verapamil to the apical side. Using a
higher concentration of H 216/44 (5 mM) and addition of
verapamil increased Pc of H 216/44 from 0.047 ( 0.003 × 10-6
to 0.104 ( 0.016 × 10-6 cm/s in the a-b direction and decreased
Pc from 0.859 ( 0.093 × 10-6 to 0.265 ( 0.052 × 10-6 cm/s in
the b-a direction. Thus the largest effect was seen on the b-a
transport.
(54) Lundqvist, A.; Ocklind, G.; Haneskog, L.; Lundahl, P. Freeze-
thaw-immobilization of liposomes in chromatographic gel
beads: Evaluation by confocal microscopy and effects of freezing
rate. In press.
(55) Brekkan, E.; Yang, Q.; Viel, G.; Lundahl, P. Immobilization of
liposomes and proteoliposomes in gel beads. In Methods in
Biotechnology, Vol 1: Immobilization of Enzymes and Cells;
Bickerstaff, G. F., Ed.; Human Press Inc.: Totowa, NJ , 1997;
pp 193-206.
(56) Artursson, P.; Ungell, A.-L.; Lo¨froth, J .-E. Selective paracellular
permeability in two models of intestinal absorption: cultured
monolayers of human intestinal epithelial cells and rat intestinal
segments. Pharm. Res. 1993, 10, 1123-1129.
(57) Lee, B.; Richards, F. M. The interpretation of protein struc-
tures: estimation of static accessibility. J . Mol. Biol. 1971, 55,
379-400.
J M980313T