Predicting human serum albumin affinity of interleukin-8 (CXCL8) inhibitors by 3D-QSPR approach

Article abstract of DOI:10.1021/jm049227l

A novel class of 2-(R)-phenylpropionamides has been recently reported to inhibit in vitro and in vivo interleukin-8 (CXCL8)-induced biological activities. These CXCL8 inhibitors are derivatives of phenylpropionic nonsteroidal antiinflammatory drugs (NSAID

Full text of DOI:10.1021/jm049227l

J. Med. Chem. 2005, 48, 2469-2479
2469
Predicting Human Serum Albumin Affinity of Interleukin-8 (CXCL8) Inhibitors
by 3D-QSPR Approach
Loretta Aureli,† Gabriele Cruciani,‡ Maria Candida Cesta,§ Roberto Anacardio,| Lucio De Simone,| and
Alessio Moriconi*,§
Molecular Discovery Ltd., 4, Chandos Street, W1A 3BQ, London, United Kingdom, Laboratory for Chemometrics and
Cheminformatics, University of Perugia, Via Elce di Sotto, 10, I-06123 Perugia, Italy, and Chemistry Department and
Analytical Department, Dompe´ Research and Development, Dompe´ S.p.A., via Campo di Pile, 67100, L’Aquila, Italy
Received September 22, 2004
A novel class of 2-(R)-phenylpropionamides has been recently reported to inhibit in vitro and
in vivo interleukin-8 (CXCL8)-induced biological activities. These CXCL8 inhibitors are
derivatives of phenylpropionic nonsteroidal antiinflammatory drugs (NSAIDs), high-affinity
ligands for site II of human serum albumin (HSA). Up to date, only a limited number of in
silico models for the prediction of albumin protein binding are available. A three-dimensional
quantitative structure-property relationship (3D-QSPR) approach was used to model the
experimental affinity constant (Ki) to plasma proteins of 37 structurally related molecules,
using physicochemical and 3D-pharmacophoric descriptors. Molecular docking studies high-
lighted that training set molecules preferentially bind site II of HSA. The obtained model shows
satisfactory statistical parameters both in fitting and predicting validation. External validation
confirmed the statistical significance of the chemometric model, which is a powerful tool for
the prediction of HSA binding in virtual libraries of structurally related compounds.
Introduction
It has been shown that some drugs, such as warfarin,6
aspirine,7 and â-lactam antibiotics,5,8 prefer to bind to
site I,9 while general anaesthetics10 and benzodi-
azepines11 interact with site II. To date, however,
limited and conflicting information about the exact
binding mode of drugs with HSA sites has been reported
in the literature.12-15
HSA is the most abundant protein in human serum
plasma.1 HSA binds a wide range of endogenous and
exogenous compounds.1,2 By virtue of its exceptional
conformational adaptability, HSA strongly binds cations
(Ca2+, Na+, K+), fatty acids, hormones, and many
xenobiotes, including drugs. In addition to its unique
binding capability, HSA has an interesting enzymatic
“esterase-like” activity toward a large variety of sub-
strates.3
Plasma protein binding is a significant factor in the
transport and release of many drugs in the human body,
strongly influencing their distribution and excretion
and, by consequence, their pharmacological and toxico-
logical profile. Since the pharmacological effect is at-
tributed to the unbound drug fraction, plasma protein
binding can affect the pharmacodynamics of many
drugs.
HSA is constituted of a single polypeptidic chain (585
amino acids) with a molecular mass of 65 kDa.4,5 Its
amino acids sequence includes a large number of
charged amino acids: about 100 acidic residues (aspar-
tates and glutamates) and about 100 basic residues
(arginines, lysines and histidines). The 3D structure of
HSA contains three homologous domains (I-III), each
including two subdomains (A and B) with common
structural motifs.4,5 The main drug binding regions are
located in subdomains IIA (site I) and IIIA (site II); both
binding sites are located in hydrophobic cavities able
to accommodate a large range of chemical structures.4,5
Several widely used NSAIDs,16 such as ibuprofen and
ketoprofen, belonging to the chemical class of 2-phenyl-
propionic acids, are highly bound to plasma proteins.
The carboxylic moiety of NSAIDs is believed to play a
key role in the binding to site II of HSA.16
Interleukin-8 (CXCL8) is a proinflammatory chemo-
kine that is a major mediator of PMN recruitment in
several acute and chronic inflammatory disorders.
Several studies suggest a key role of CXCL8 in PMN
recruitment and tissue injury occurring in postischaemic
reperfusion injury.17 In addition, CXCL8 induces kera-
tinocyte proliferation as well as angiogenesis and has
been strongly implicated in melanoma progression and
metastasis.18 CXCL8 exerts its biological effects by
activating two specific receptors, CXCR1 and CXCR2,
belonging to the family of seven transmembrane G-
protein-coupled receptors (7TM-GPCRs).19 CXCL8 is
considered a primary target for a novel therapeutic
approach for various human diseases.20,21 N-[(R)-2-(4-
Isobutylphenyl)propionyl]methanesulfonamide (2), which
belongs to a novel class of small molecular weight
CXCL8 inhibitors, is currently undergoing clinical trials
for the prevention of delayed graft function (DGF)
occurring during organ transplant. 2 has recently been
described as a potent noncompetitive allosteric inhibitor
of CXCL8 acting in the trans-membrane region of
CXCR1.22
* Corresponding author. Phone: +39 0862 338424. Fax: +39 0862
338219. E-mail: alessio.moriconi@dompe.it.
† Molecular Discovery Ltd.
‡ University of Perugia.
In this paper we show that several CXCL8 small
molecular weight inhibitors, structurally related to
§ Chemistry Department, Dompe´ Research and Development.
| Analytical Department, Dompe´ Research and Development.
10.1021/jm049227l CCC: $30.25 © 2005 American Chemical Society
Published on Web 03/08/2005

2470 Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7
Aureli et al.
NSAIDs, are highly bound to HSA, although lacking the
carboxylic moiety. To explore the relationships between
physicochemical (VOLSURF) and specific 3D-pharma-
cophoric descriptors (ALMOND) to HSA, a chemometric
3D-QSPR model has been derived.23 The chemical
interpretation of the final model strongly supports the
hypothesis of a common binding region between NSAIDS
and 2-(R)-phenylpropionamides in site II of HSA. The
3D-QSPR model will prove useful in designing a second
generation of optimized CXCL8 inhibitors with the
desired HSA binding affinity and be helpful to the
prioritization of structurally related chemical libraries.
Due to the lack of structural information, this combined
molecular docking/3D-QSPR approach might help to
provide additional information about drugs binding to
site II of HSA.
descriptors calculation was made using VOLSURF29 and
ALMOND30 software. Finally, the statistical model was
generated using GOLPE software.31
Molecular Docking. Molecular docking was per-
formed on all the training set molecules, using the
flexible protein-ligand option in the GRID-GLUE
procedure.28 GLUE is a docking tool that uses the GRID
force field to locate the ligand inside the protein cavity.
A new algorithm evaluates the protein-ligand interac-
tions and selects the most favorable ligand torsion
angles to maximize protein-ligand interactions. To
study the HSA sites I and II, 3D crystal structures of
HSA complexed with warfarin6 (PDB code 1h9z)32 and
complexed with propofol10 (PDB code 1e7a)32 have been
used as starting models.
Physicochemical Descriptors. VOLSURF is a mo-
lecular modeling software that generates 2D molecular
descriptors from the 3D molecular interaction field
(MIF) on GRID maps.29 The basic idea of VOLSURF is
to compress the information contained in 3D maps
calculated by GRID into a few 2D numerical descriptors,
which show the great advantage of a simple chemical
interpretation. GRID is a computational procedure for
the determination of energetically favorable binding
sites between a probe and all the atoms in a molecule
virtually inserted in a 3D GRID map.28,33 The probe is
chosen in a variety of chemical groups that are supposed
to interact with the molecule. The GRID force field uses
a potential based on the total energy of interaction (the
sum of Leonard-Jones, H-bonding, electrostatic, and
hydrophobic terms) between a target molecule and a
probe to characterize putative polar or hydrophobic
interactions sites around target molecules.28,33 GRID
may be used to study small molecular weight molecules,
such as drugs, or biological macromolecules.33,34 VOL-
SURF descriptors calculate the principal molecular
physicochemical properties and are specifically designed
for the optimization of in silico pharmacokinetic proper-
ties for pharmaceutically relevant compounds.35-37 In
this work, VOLSURF calculation produced 94 descrip-
tors (grid spacing 0.5 Å) using three probes to represent
potentially important groups in the putative binding
site: the water probe (OH2) simulating the solvation-
desolvation process, the hydrophobic probe (DRY) com-
puting the hydrophobic energy, and the carbonyl oxygen
probe (O) representing the hydrogen-bond acceptors.
The DRY probe is a specific probe to compute the
hydrophobic energy that can be estimated by means the
computation of three terms: the Lennard-Jones poten-
tial (that includes stacking, induction, and dispersion
interactions with a maximum around -2.0 kcal/mol),
the entropic (about -0.8 kcal/mol) and hydrogen-bond-
ing contributions.28
Materials and Methods
Database. The training set includes 37 drug and
druglike compounds extracted from three different
chemical families. More specifically, it is formed by 10
novel CXCL8 inhibitor compounds (1-10 in Chart 1),
17 structural analogues of diflunisal24 (11-27 in Chart
1), and 10 classical NSAIDs25 (28-37 in Chart 1). The
test set includes seven compounds (38-44 in Chart 2)
with different biological activity but with a close struc-
tural relationship to the compounds of the training set.25
Protein Binding Determination. The human pro-
tein binding for 10 CXCL8 inhibitors (1-10 in Chart 1)
was experimentally determined as described in the
Experimental Section. The HSA binding constant af-
finity Ki was calculated for 1-10 and 28-44 according
to eq 114
fb
1 - fb
log Ki ) log
- log [HSA]
(1)
where Ki is a measure of binding affinity to HSA, under
the assumption that the binding to plasma proteins
occurs mostly to HSA, fb represents the drug fraction
bound to plasma proteins; the concentration of albumin
([HSA]) is fixed to 0.6 mM. The fb values for compounds
1-10 were experimentally determined, whereas fb for
28-44 were obtained from Goodman and Gilman’s
handbook and from the literature.14,25 Moreover, the
binding constant affinity for compounds 11-27 was
derived from literature data.24
The 2D chemical structures of the training set and
test set compounds are respectively shown in Charts 1
and 2, whereas the corresponding log Ki values are listed
in Tables 1 and 2.
Chemistry. 2-(R)-Phenylpropionamides (1-10 in
Chart 1) were prepared starting from optically pure
2-(R)-phenylpropionic acids following well-established
procedures for the preparation of amides (see Experi-
mental Section).
Computational Approach. All calculations were
accomplished using a Linux workstation, under the
Fedora operating system. The molecular structures were
drawn and geometrically optimized using SYBYL soft-
ware with the Powell method, within the standard
TRIPOS force-field.26 The conformational analysis was
performed by the CONFORT procedure fixing the
energy level to 20 kcal/mol.27 Molecular docking was
performed by GRID-GLUE software.28 The molecular
3D-Pharmacophoric Descriptors. ALMOND is a
program specifically developed for generating and han-
dling a novel type of molecular descriptor called GRIND
(grid independent descriptors), based on 3D MIF (cal-
culated by GRID) around a ligand and based on phar-
macophoric concepts.30,38 GRIND are derived in such a
way as to be highly relevant for describing biological
properties of compounds while being alignment-in-
dependent, chemically interpretable, and easy to com-
pute.38 ALMOND is a program developed for the com-
putation, analysis, and interpretation of GRIND and for

Predicting Affinity of CXCL8 Inhibitors by 3D-QSPR
Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7 2471
Chart 1. Chemical Structures of the Training Set (n ) 37).
the generation of the resulting model.30 ALMOND is
able to re-elaborate the information contained in the
GRIND in 2D descriptors strictly correlated to the
molecular pharmacophoric properties.30 In this work,
ALMOND computation produced 510 descriptors (grid-
spacing 0.5 Å) using the DRY probe, the O probe, the
molecular shape probe (TIP), and the amide nitrogen
probe (N1) representing the hydrogen-bond donors.
Global Model: Data Matrix. GOLPE is a powerful
tool for handling 3D-QSPR problems and improving the
quality of the results of a multivariate regression
analysis.31 GOLPE contains an advanced variable selec-
tion procedure, aimed to obtain a PLS model with
highest prediction ability.39,40 In our chemometric study,
GOLPE was used to obtain a global statistical model
able to correlate the physicochemical and pharmaco-

2472 Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7
Aureli et al.
Chart 2. Chemical Structures of the Test Set Compounds (n ) 7).
Table 1. log Ki Values for Dataset Compounds
Table 3. PLS Statistical Results
compd log Ki SDa PB SDb compd log Ki SDa PB SDb
PC
r2
q2 a
q2 b
PC
r2
q2 a
q2 b
1
3
5.92 0.09 99.8
4.28 0.05 92.0
3.43 0.03 62.0
3.70 0.03 75.1
4.60 0.01 96.0
4.77 0.10
2
4
6.22 0.03 99.9
3.59 0.02 70.0
3.98 0.02 85.1
3.91 0.08 83.1
3.59 0.02 70.2
5.00 0.09
1
2
3
0.69
0.84
0.90
0.54
0.65
0.67
0.51
0.61
0.63
4
5
0.94
0.96
0.62
0.58
0.59
0.57
5
6
7
9
8
a Leave one out (LOO). b Random groups.
10
12
14
16
18
20
22
24
26
28
30
32
34
36
11
13
15
17
19
21
23
25
27
29
31
33
35
37
3.77 0.13
4.42 0.19
demonstrated that its protein binding is mainly due to
the high affinity for HSA, whereas only a very low
fraction binds R1-acid glycoprotein (data not shown).
HSA being the most abundant protein in human
plasma, it has been assumed that the binding of drugs
occurs mostly to HSA.
4.31 0.12
4.33 0.16
3.98 0.12
5.40 0.33
5.30 0.09
3.79 0.07
3.89 0.04
3.00
3.00
3.76 0.06
3.49 0.11
4.38 0.02
3.44
5.22
5.52
4.18
5.32
5.62
99
Molecular docking analysis supports the hypothesis
that site II of HSA is a preferential binding site for most
of the compounds in the data set. As assessed by direct
binding energy comparison (data not shown), the bind-
ing preference to site II is strongly accentuated for the
molecules in the data set with an high value of log Ki
(Table 1). In agreement with the literature data, all
NSAIDs were selective ligands for site II. Similarly,
most of diflunisal analogues were selective ligands for
site II, excepting the weak binders 20, 21, 22, 23, and
27, which show some propensity to bind site I. Finally,
the 10 CXCL8 inhibitors were preferential ligands for
site II with poor affinity for site I.
According to the proposed models, all the compounds
in the training set engage extensive interactions with
hydrophobic residues in the site II pocket (Leu453,
Ile388, Asn391, Leu407). Furthermore, electrostatic and
hydrogen-bond interactions with specific amino acids of
site II, including the backbone of Leu430 and several
residues (Tyr411, Arg410, and Ser489), seem crucial in
determining the HSA site II binding affinity.
Chemometric analysis of the X-matrix data gave a
statistically significant model without any outlier be-
havior; Table 3 shows the statistical values of the PLS
model obtained with five principal components.
The PLS T-U plot in Figure 1 (panel a) represents
the trend of all the modeled objects. The points in the
upper part of the plot represent the compounds with
major HSA binding affinity, whereas those in the lower
part represent the compounds with lower affinity.
5.32
5.52
4.91
4.42
5.04
99.2 0.1
99.5
98
94
98.5
99.5
90
99.2 0.1
99.6 0.1
0.04
1
a Standard deviation (SD) associated with log Ki (log Ki ( SD).
b Standard deviation (SD) of the percentage plasma protein binding
data (PB ( SD), from refs 14 and 25.
Table 2. log Ki Values for Test Set Compounds
compd log Ki
PB
SDa
compd log Ki
PB
SDa
38
39
40
41
4.60
5.14
3.36
4.18
96.0
98.8
58
0.6
0.2
17
42
43
44
2.55
4.73
4.60
17.5
97
96
1
90
a Standard deviation (SD) of the percentage plasma protein.
binding data (PB ( SD) found in the literature (from refs 14 and
25).
phoric descriptors of the analyzed compounds with HSA
binding affinity. PLS analysis was performed in the
global X-matrix data represented from the VOLSURF
and the ALMOND descriptors; the Y-matrix includes
the log Ki value for each object of the data set.41,42 The
two blocks of variables were scaled by means of the
BUW (block unscaled weights) scaling procedure as
implemented in GOLPE software.31 This operation
separately scales each variable block, whereas the
relative scales of single variables within each block
remain unchanged. The effect of the BUW procedure is
to normalize the scaling of the blocks, thus giving the
same importance to each of them. The obtained model
was validated by random groups and leave-one-out
procedures.31
The PLS score plot in Figure 1 (panel b), representing
the positions of X-matrix objects in the multidimen-
sional descriptor space, shows the correlation between
the X descriptors and the Y-dependent variable for each
component. The first component is, by far, the most
informative and reports the main correlation between
HSA binding affinity and structural descriptors. The
three lines drawn in Figure 1 (panel b) graphically
identify three regions of HSA binding affinity: high
affinity, intermediate affinity, and low affinity.
Results and Discussion
The plasma protein binding has been experimentally
determined for the CXCL8 inhibitors (shown in Table
1). Binding of 2 to plasma protein has been thoroughly
investigated in different animal species. At the concen-
tration of 50 µM, the unbound drug fraction in human
plasma is extremely low (0.1%); further studies have

Predicting Affinity of CXCL8 Inhibitors by 3D-QSPR
Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7 2473
Figure 1. (a) PLS T-U plot for compounds reported in Table 1. Circles depicted in the upper part of the plot represent compounds
with high HSA binding affinity; circles in the central part of the plot represent the compounds with intermediate affinity; circles
in the lower part represent compounds with low HSA binding affinity. (b) PLS score plot (first versus second component) for the
global model. The three lines graphically identify three distinct regions of HSA binding affinity as described above. The model
offers a good discrimination between compounds with high HSA binding affinity (black circles on the right of the plot), medium
affinity (dark gray circles in the center), and low affinity (light gray circles on the left of the plot). Compounds 2, 35, and 22 are
visualized respectively as high, intermediate, and low HSA binders. (c) PLS loading plot for the global model. Open circles represent
the physicochemical molecular descriptors (VOLSURF); filled circles represent the 3D-pharmacophoric descriptors (ALMOND).
HSA binding affinity increases with the relative increase of the hydrophobic integy moment at five energy levels (IDDRY), hydrogen-
bonding capability (HBO), and amphiphilic moment (A); conversely, it decreases when capacity factor values (CW) at two energy
levels increase and hydrophilic regions at four different energy levels (W) become larger. The right-lower ellipse includes the
VOLSURF descriptors IWOH2, DIFF, and BV, and the left-upper ellipse includes the VOLSURF descriptors VOH2, SOH2, GOH2,
POL, MW, and EEFR as described in the text. Moreover, the HSA binding increases when in the target hydrogen donor groups
are present at about 12.5 Å distance; conversely it decreases when hydrogen acceptor/donor groups are present at about 13.5 Å.
The high affinity ligand (right region) cluster contains
three CXCL8 inhibitors (1, 2, 9), three diflunisal ana-
logues (12, 18, 19), and most of the NSAIDs compounds
(28-31, 33, 34, 36, 37).
The intermediate affinity ligand (central region)
cluster contains one CXCL8 inhibitor (3), some di-
flunisal analogues (11, 14-17, 26), and two NSAIDs (32,
35).
The low affinity ligand (left region) cluster comprises
some CXCL8 inhibitors (4-8, 10) and most of the
diflunisal analogues (13, 20-25, 27).
The loading plot of Figure 1 (panel c) represents the
contribution of original variables to the latent variables.
VOLSURF and ALMOND descriptors are reported with
different symbols.
It is clear that the physicochemical descriptors (VOL-
SURF) are more directly related to the HSA binding
than the 3D-pharmacophoric ones (ALMOND). The
statistical comparison shows that the physicochemical
descriptors account for 50% of the HSA binding whereas
the 3D-pharmacophoric ones account for 30% of the HSA
affinity.43 It is important to stress that it is the balance
of all descriptors which controls HSA binding affinity.
Table 4 lists some of the most important VOLSURF
descriptors and their definitions.29
A careful analysis of the loading plot shows that HSA
binding affinity increases with the relative increase of
the hydrophobic integy moments, calculated at five
different energy levels (IDDRY), hydrogen-bonding ca-
pability (HBO), and amphiphilic moment (A); conversely

2474 Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7
Aureli et al.
Table 4. Definition of Several VOLSURF Physicochemical Descriptors
Volsurf 2D descriptors
IDDRY (hydrophobic integy moment) measures the unbalance between the center of mass of molecules and the position
of the hydrophobic regions around them, calculated at eight energy levels by hydrophobic probe DRY. Integy moments are
vectors pointing from the center of the mass to the center of the hydrophobic regions.
HBO (hydrogen bonding) measures of the hydrogen-bonding capability calculated by O probe.
A (amphiphilic moment) is defined as a vector pointing from the center of the hydrophobic domain to the center
of the hydrophilic domain.
CWOH2 (capacity factors) represents the ratio between the hydrophilic regions and the molecular surface calculated
by water probe.
W (hydrophilic regions at eight energy levels) may be defined as the molecular envelope accessible by solvent water molecules.
IWOH2 (hydrophilic integy moment) measures the unbalance between the center of the mass of molecules and the position of
the hydrophilic regions around them calculated by water probe at eight energy levels.
DIFF (diffusivity) measures the molecular diffusivity in water at 25 °C.
BV (best volume) identifies the largest volumes of hydrophilic (BVOH2) or hydrophobic regions (BVDRY) in the molecular structures.
VOH2 (molecular volume) is defined when a water probe is interacting with a target solute molecule. It represents the water
solvent excluded volume (in Å3) or the volume contained within the water accessible surface computed at 0.20 kcal/mol.
SOH2 (molecular surface) is defined when a water probe is interacting with a target solute molecule. It represents the
accessible surface (in Å2) traced out by a water probe interacting at 0.20 kcal/mol when it rolls over the target molecule.
GOH2 (molecular globularity) defined by the ratio Surface/Sequiv, in which Sequiv is the surface area of a sphere of volume V.
POL (molecular polarizability) is an estimation of the average molecular polarizability.
MW (molecular weight).
EEFR (elongation/fixed elongation) refers to the maximum extension a molecule could reach if properly streched. The two
descriptors are
elongation, the most probable extension of the molecule;
fixed elongation, the portion of the extension given by the rigid part of a molecule.
it decreases when best volume (BV) values increase and
hydrophilic regions at four different energy levels (W)
become larger. The following conclusions can be drawn
by the analysis of the results: VOLSURF descriptors
of hydrophilic interactions, such as hydrophilic regions
(W), are inversely correlated with HSA binding; on the
contrary, VOLSURF descriptors of hydrophobic interac-
tions, such as hydrophobic integy moment (IDDRY), are
directly correlated with HSA binding affinity. Hence,
hydrophilic regions are detrimental for drug interaction
with albumin. However, the presence of concentrated
hydrophobic regions, located opposite to regions with
high capability to engage hydrogen bonds with the
target, strongly favors the binding to site II of HSA.
In the loading plot of Figure 1 (panel c), two additional
blocks of VOLSURF descriptors (Table 4) directly
related to the HSA binding are highlighted. The first
group is located in the lower right ellipse of the loading
plot and includes molecular physicochemical descriptors
with a positive absolute value on the first component
and negative absolute value on the second component
of the model (DIFF, IWOH2, BV). The second group is
located in the upper left ellipse of Figure 1 (panel c) and
includes molecular physicochemical descriptors with
negative coefficients on the first component and positive
on the second component of the PLS model (VOH2,
SOH2, POL, GOH2, MW, EEFR). Although both these
groups of descriptors clearly influence HSA binding
affinity, their orthogonal arrangement in the 3D-space
makes the global impact on HSA binding weak.
22; this characteristic, joined with long integy moment
vectors, accounts for the high affinity of 2. 35, which is
not characterized by diffused hydrophobic regions and
whose integy moment vectors are intermediate long, is
a moderate site II binder. Hydrophobic regions in 22
are very restricted and its integy moment vector is very
short; for these reasons 22 is a weak HSA site II
substrate.
The loading plot in Figure 1 (panel c) also shows that
binding to HSA increases when specific pharmacophoric
groups are present at certain distances. When two
hydrogen-acceptor groups of the ligand interacting with
the hydrogen-donor groups in the target are at about
12.5 Å distance, the binding to HSA increases. This
requirement is satisfied in all the best HSA ligands in
the database. Conversely, binding affinity decreases
when hydrogen-acceptor-donor groups in the target at
about 13.5 Å are present. The strong effect of acceptor
groups on the ligands suggests that the binding with
site II occurs through hydrogen bonds with Leu430
(backbone), Arg410, Tyr411, and Ser489 residues in the
binding cavity of HSA site II. The precise orientation
of hydrogen-bond-donor groups on the target resulting
a crucial factor for an optimal HSA site II targeting.
Indeed, docking calculations highlight the presence at
12.5 Å distance, in the binding cavity of HSA site II, of
Arg410 and Leu430. These specific residues engage
hydrogen bonds with carbonylic or sulfonylic oxygen
acceptor atoms, present in high-affinity ligands for te
site II of HSA.
Figure 2 (panels d-f) shows a visual comparison for
the MIFs obtained using the N1 probe for 2, 35, and
22. Only 2 features the two hydrogen-bond acceptor
(connected by the red line) at 12.5 Å distance, whereas
the same features in 35 and 22 are clearly more distant.
Summarizing, the evaluation of hydrophobic regions
and integy moment values, combined with the analysis
of specific pharmacophoric points, allows the identifica-
tion of high-affinity ligands for HSA site II.
To clarify the model interpretation, Figure 2 (panels
a-c) shows a visual comparison of the 3D MIFs calcu-
lated with the DRY probe at about -0.4 kcal/mol for
three compounds (2, 35, 22) respectively selected as
strong, intermediate, and weak HSA binders. The
contoured zones around the molecules represent the
hydrophobic regions interacting with the DRY probe;
the red vectors represent the hydrophobic integy mo-
ments (IDDRY) calculated at eight energy levels as
reported in Table 4. The size and volume of the
hydrophobic regions of 2 are larger than those of 35 and
Validation Model. The evaluation of the predictive
capability of the developed PLS model has been carried

Predicting Affinity of CXCL8 Inhibitors by 3D-QSPR
Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7 2475
Figure 2. (a, b, c) Comparison of the 3D MIFs calculated by use of the DRY probe among 2, 35, and 22. The areas surrounding
these molecules represent the hydrophobic regions interacting with the DRY probe contoured at -0.4 kcal/mol; the red vectors
represent the hydrophobic integy moment (IDDRY) calculated at eight energy levels. (d, e, f) Comparison of the 3D MIFs calculated
by using N1 probe among 2, 35, and 22. The areas surrounding the molecules represent regions interacting with the N1 probe
and able to accept hydrogen bonds. The red line emerging in 2 represents the pharmacophoric distance of 12.5 Å, which is essential
to enhance the HSA affinity.
shows that prediction of the test set is in good agree-
ment with the experimental data (see Table 2). In fact,
43 is located in the high binding affinity region; 38, 39,
41, 44 are predicted in the intermediate affinity region,
whereas 40 and 42 are projected into the low binding
affinity region (Figure 3). The prediction ability of the
developed PLS model was demonstrated by the correct
prediction for this external test set, confirming the
quality and the statistical significance of the PLS model.
Conclusions
Plasma protein binding dramatically influences both
the pharmacokinetic and the pharmacodynamic profile
of potential drug candidates but, to date, only a limited
number of in silico models for the evaluation of drug-
binding affinity to HSA have been reported. A novel
class of CXCL8 inhibitors showed an extremely high
affinity toward HSA. With the aim of guiding the design
Figure 3. PLS score prediction plot: the shown external test
of second-generation leads with reduced protein binding,
set molecules are projected on the global PLS model of Figure
a chemometric 3D-QSPR approach was used to model
1b. From a visual inspection of the plot, all the external
HSA binding affinity. Due to the structural similarity
between CXCL8 inhibitors and classical NSAIDs, a
preferential binding to site II was assumed.
compounds are correctly predicted by the model (see text).
out for seven test set molecules selected from literature
data.14,25 The molecular structures and the experimental
binding data are respectively reported in Chart 2 and
Table 2. The quality of the external prediction is
graphically shown in the PLS score plot of Figure 3 in
which these compounds are projected into the model.
The three lines graphically highlight the three regions
of HSA binding affinity: high (right), intermediate
(central), and low (left) binding affinity regions. This plot
A preliminary molecular docking was performed on
all the training set molecules to estimate their binding
affinity for sites I and II of HSA. This approach
confirmed that the most of the molecules in the training
set are preferential ligands for site II. On the basis of
this assumption, a good statistical model was developed
that was able to describe the drug HSA site II interac-
tion in a quantitative manner.

2476 Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7
Aureli et al.
area of filtered sample
area of unfiltered sample
The global PLS model examined the impact on HSA
binding of physicochemical molecular properties as well
as of pharmacophoric properties. The quantitative
relationship between 3D-structure and HSA binding
affinity to site II was discussed. Hydrophilic regions are
detrimental for drug interaction with HSA, whereas the
presence of concentrated hydrophobic regions, located
opposite to the hydrogen-bond-acceptor groups, strongly
favor HSA binding. Moreover, specific pharmacophoric
points distancing about 12.5 Å are important to enhance
the affinity for site II of HSA. Finally, the good predic-
tion ability of the PLS model was demonstrated by the
correct prediction for external test molecules. Accord-
ingly, on the basis of the model, novel optimized CXCL8
inhibitors could be designed with the desired HSA
binding affinity. All these results suggest that the
chemometric strategies described in this paper might
be a powerful tool in pharmaceutical industries to
optimize ADME properties of lead compounds, during
the early phase of drug discovery.
% free fraction )
× 100
% protein binding ) 100 - % free fraction
Chromatographic separations were performed at room tem-
perature. Analytical columns, mobile phases, detection condi-
tions, injected volumes, and flow rates together with typical
retention times are reported in Table 5. The selected chro-
matographic conditions were suitable for good resolution
between the analytes of interest and the plasma ultrafiltered
endogenous peaks.
Chemistry. Synthesis of (R)-2-Arylpropionic Acids
Intermediates. (R)-2-(4-Isobutylphenyl)propionic Acid.
The optical resolution of the racemic acid was performed as
described.47 (R)-2-(4-Isobutylphenyl)propionic acid (50 g, 35%)
was obtained as a white solid: mp 48-49 °C; [R]25 (c ) 2,
D
EtOH) -53°; ee ) 98%; HPLC-UV, Chiralcel OD column, 500:
5:0.5 n-hexane:iPrOH:AcOH, tR ) 13.8 min; 1H NMR (CDCl3)
δ 7.15 (d, 2H, J ) 7 Hz), 6.95 (d, 2H, J ) 7 Hz), 3.58 (q, 1H,
J ) 7 Hz), 2.30 (d, 2H, J ) 7 Hz), 1.75 (m, 1H), 1.35 (d, 3H, J
) 7 Hz), 0.85 (d, 6H, J ) 7 Hz). Anal. (C13H18O2) C, H, N.
(R)-2-(4-Hydroxyphenyl)propionic Acid. The optical
resolution of the racemic acid was performed as described.47
(R)-2-(4-Hydroxyphenyl)propionic acid (34.7 g, 29%) was ob-
tained as a white solid: mp 156-158 °C; [R]25D (c ) 1, MeOH)
-72°; ee ) 97%; HPLC-UV, Chiralcel OJ column, 75:15:10
n-hexane:EtOH:iPrOH and 0.5% trifluoroacetic acid, tR ) 10.95
min; 1H NMR (CDCl3) δ 7.12 (d, 2H, J ) 7 Hz), 6.75 (d, 2H, J
) 7 Hz), 5.45 (s, 1H, OH), 3.62 (q, 1H, J ) 7 Hz), 1.45 (d, 3H,
J ) 7 Hz). Anal. (C9H10O3) C, H, N.
Experimental Section
General Experimental Procedures. Optical rotations
were measured on a Perkin-Elmer 241 polarimeter and the
1
[R]25 values are given in 10-1 deg cm2 g-1. H NMR spectra
D
were recorded on a Bruker ARX 300 spectrometer. Melting
points were determined using a Bu¨chi capillary melting point
apparatus and are uncorrected. Elemental analyses were
within (0.4% of the theoretical values calculated for C, H, and
N and are reported only with symbols.
(R)-2-(4-Hydroxyphenyl)propionic Acid Methyl Ester.
HCl (37%, 5 mL) was added to a solution of (R)-2-(4-hydroxy-
phenyl)propionic acid (34.7 g, 0.21 mol) in methanol (450 mL)
and, after stirring at rt for 18 h, the solvent was removed under
vacuum and the residue taken up with CH2Cl2 (250 mL). The
organic phase was washed with water and with a saturated
solution of NaHCO3 and dried over Na2SO4, to give, after
solvent evaporation, pure (R)-2-(4-hydroxyphenyl)propionic
acid methyl ester (35.2 g, 93%) as a slightly yellow oil: 1H
NMR (CDCl3) δ 7.18 (d, 2H, J ) 7 Hz), 6.82 (d, 2H, J ) 7 Hz),
5.7 (bs, 1H, CONH), 3.70 (s, 3H), 3.65 (q, 1H, J ) 7 Hz), 1.52
(d, 3H, J ) 7 Hz). Anal. (C10H12O3) C, H, N.
(R)-2-(4-Trifluoromethanesulfonyloxyphenyl)pro-
pionic Acid Methyl Ester. A solution of (R)-2-(4-hydroxy-
phenyl)propionic acid methyl ester (35 g, 0.19 mol) and N,N-
diisopropylamine (39 mL, 0.23 mol) in CH2Cl2 (170 mL) at -15
°C was treated, under inert atmosphere, with trifluoromethane-
sulfonic anhydride (38.7 mL, 0.23 mol). At the end of the
adding, the mixture was left stirring for 3 h at rt. The organic
solution was washed with water (3 × 250 mL) and, after drying
with Na2SO4, solvent was removed under vacuum to give (R)-
2-(4-trifluoromethanesulfonyloxyphenyl)propionic acid methyl
ester (56.4 g, 95%) as a brown oil: 1H NMR (CDCl3) δ 7.45 (d,
2H, J ) 7 Hz), 7.21 (d, 2H, J ) 7 Hz), 3.85 (q, 1H, J ) 7 Hz),
3.65 (s, 3H), 1.50 (d, 3H, J ) 7 Hz). Anal. (C11H11F3O5S) C, H,
N.
All reagents and solvents were purchased from Sigma-
Aldrich or Lancaster and used without further purification.
Reaction courses and product mixtures were monitored by
thin-layer chromatography on silica gel (precoated F254 Ma-
cherey-Nagel plates); the spots were examined with UV light
and visualized with I2.
The HPLC-UV analysis was carried out using a chromato-
graphic system composed by a Model 2690 pump, a Model 486
UV-vis detector (Waters, Milford, MA), and a Model 7725i
sample injector (Rheodyne, Cotati, CA) equipped with a 20-
µL loop. Chromatographic data management was automated
using Millennium32 software (Waters, Milford, MA).
The HPLC-fluorescence analysis was carried out using the
same chromatographic system above-described equipped with
a Model 474 scanning fluorescence detector (Waters, Milford,
MA).
The HPLC-MS/MS analysis was carried out using a Model
Surveyor MS pump, a Model Surveyor AS sample injector, and
a Model LCQ Deca XP Plus MSn ion trap mass spectrometer
(Thermo Finnigan, San Jose, CA). Chromatographic data
management was automated using a software Xcalibur Instru-
ment Control, (Thermo Finnigan, San Jose, CA).
Protein Binding Determination in Human Plasma.
The in vitro binding of CXCL8 inhibitors to human plasma
proteins was investigated by ultrafiltration using the Centri-
free micropartition system. Separation of free compound from
protein-bound material was achieved by filtration of the
sample through an ultrafiltration membrane.44-46
Duplicate drug-free plasma samples were mixed with each
CXCL8 inhibitor under investigation at the nominal concen-
tration of 20 µg/mL and incubated at 37 °C for 20 min. The
spiked plasmas were subsequently transferred to ultrafiltra-
tion units, which were centrifuged at 1500 rpm for about 15
min. The degree of test article binding to the apparatus, or
nonspecific binding, was assessed by investigating the ultra-
filtration of sodium phosphate buffered saline spiked with the
analytes at 20 µg/mL, in the same manner.
The unbound test article in the obtained ultrafiltrates was
quantified by reversed phase HPLC analysis with UV, fluo-
rescence, or MS/MS detection, depending on the nature of the
molecule, using the following expressions:
(R)-2-(4-Trifluoromethanesulfonyloxyphenyl)pro-
pionic Acid. (R)-2-(4-Trifluoromethanesulfonyloxyphenyl)-
propionic acid methyl ester (56 g, 0.18 mol) was dissolved in a
solution of glacial acetic acid (330 mL) and 37% HCl (57 mL)
and refluxed for 3 h. After cooling at rt the residue was
dissolved in CH2Cl2 (200 mL) and washed with water (3 × 250
mL). After drying over Na2SO4, solvent was removed under
vacuum to give (R)-2-(4-trifluoromethanesulfonyloxyphenyl)-
propionic acid (44.5 g, 83%) as a white solid: mp 57-60 °C;
1
[R]25 (c ) 1, MeOH) -32°; H NMR (CDCl3) δ 7.54 (d, 2H, J
D
) 7 Hz), 7.25 (d, 2H, J ) 7 Hz), 3.85 (q, 1H, J ) 7 Hz), 1.58 (d,
3H, J ) 7 Hz). Anal. (C10H9F3O5S) C, H, N.
Synthesis of CXCL8 Inhibitors. (R)-2-(4-Trifluoro-
methanesulfonyloxyphenyl)-N-methanesulfonylpro-
pionamide (1). 1,1′-Carbonyldiimidazole (7.21 g, 44.5 mmol)
was added to a solution of (R)-2-(4-trifluoromethanesulfonyl-
oxyphenyl)propionic acid (13.28 g, 44.5 mmol) in dry CH2Cl2
(130 mL), and the resulting mixture was left stirring at rt for

Predicting Affinity of CXCL8 Inhibitors by 3D-QSPR
Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7 2477
Table 5. Chromatographic Conditions for Human Protein Binding Determination
injected
vol
(µL)
flow
rate
tR
analyte
mobile phase
detection
analytical column
(mL/min) (min)
1
0.05 M KH2PO4 (pH 3): CH3CN:CH3OH, UV, 215 nm
40:25:35
20
Hypersil BDS C18, 250 × 4 mm,
5 µm (Thermo Hypersil)
Luna C18 150 × 2 mm,
5 µm (Phenomenex)
1.0
6.3
2
0.02 M NH4HCOO (pH 4.3):MeOH
gradient elution (linear), from 30/70 to
20/80 in 3 min
0.05 M KH2PO4 (pH 3):CH3CN:CH3OH,
40:25:35
0.05 M KH2PO4 (pH 3):CH3CN, 66:34
MS/MS, ESI+,
m/z: parent 301,
daughter 284
UV, 215 nm
10
0.2
5.8
3
4
5
6
7
8
20
10
10
20
20
20
Hypersil BDS C18, 250 × 4 mm,
5 µm (Thermo Hypersil)
Luna C18, 250 × 4.6 mm
(Phenomenex)
1.0
1.0
1.0
1.0
1.0
1.5
5.5
6.5
9.5
6.0
5.4
6.9
UV, 215 nm
UV, 215 nm
UV, 223 nm
UV, 223 nm
0.05M KH2PO4 (pH 3):CH3CN, 66:34
0.05 M KH2PO4 (pH 2.5):CH3OH, 32:68
0.05 M KH2PO4 (pH 2.5):CH3OH, 32:68
0.05M KH2PO4 (pH 3):CH3CN, 60:40
Luna C18, 250 × 4.6 mm
(Phenomenex)
Hypersil BDS C18, 250 × 4 mm,
5 µm (Thermo Hypersil)
Hypersil BDS C18, 250 × 4 mm,
5 µm (Thermo Hypersil)
Lichrospher RP 60 Select B,
250 × 4 mm (Merck)
fluorescence:
ex 223 nm,
em 288 nm
fluorescence:
ex 223 nm,
em 287 nm
UV, 223 nm
9
0.05 M KH2PO4 (pH 3):CH3CN, 70:30
0.05 M KH2PO4 (pH 3):CH3OH, 50:50
20
20
Lichrospher RP 60 Select B,
250 × 4 mm (Merck)
1.5
1.0
9.7
5.7
10
Hypersil BDS C18, 250 × 4 mm,
5 µm (Thermo Hypersil)
90 min. Methanesulfonamide (4.23 g, 44.5 mmol) and DBU
(6.65 mL, 44.5 mmol) were added, and the mixture was left
stirring 16 h further at rt. The organic phase was washed with
0.5 N HCl (2 × 50 mL), 5% NaH2PO4 (3 × 50 mL), and water
(2 × 50 mL). After drying with Na2SO4, solvent was removed
under vacuum, and the residue was taken up with n-hexane
(50 mL) and pulped at rt for 16 h. The formed precipitate was
filtered under vacuum and dried in an oven at 50 °C to give
synthesized (70%) using the same procedure described for 3
starting from (R)-2-(4-trifluoromethanesulfonyloxyphenyl)-
propionic acid and 3-(pyrrolidin-1-yl)propylamine. [R]25 (c )
D
1, MeOH) -41°; 1H NMR (CDCl3) δ 8.01 (bs, 1H, CONH), 7.62
(d, 2H, J ) 7 Hz), 7.15 (d, 2H, J ) 7 Hz), 3.80 (q, 1H, J ) 7
Hz), 3.52 (m, 2H), 3.31 (m, 2H), 2.95 (m, 2H), 2.78 (m, 2H),
2.15-1.90 (m, 6H), 1.55 (d, 3H, J ) 7 Hz). Anal. (C17H23F3-
N2O4S) C, H, N.
pure 1 as a white powder (13.2 g, 79%): mp 98-100 °C; [R]25
(R)-2-(4-Isobutylphenyl)-N-(1-methylpiperidin-4-yl)-
propionamide (6). Compound 6 was synthesized (70%) using
the same procedure described for 3 starting from (R)-2-(4-
isobutylphenyl)propionic acid and 1-methyl-4-aminopiperi-
D
1
(c ) 0.5, MeOH) -49°; H NMR (CDCl3) δ 7.40 (d, 2H, J ) 7
Hz), 7.23 (d, 2H, J ) 7 Hz), 3.68 (q, 1H, J ) 7 Hz), 3.15 (s,
3H), 1.42 (d, 3H, J ) 7 Hz). Anal. (C11H12F3NO6S2) C, H, N.
N-[(R)-2-(4-Isobutylphenyl)propionyl]methanesul-
fonamide (2). Compound 2 was synthesized (79%) using the
same procedure described for 1 starting from (R)-2-(4-isobu-
tylphenyl)propionic acid: mp 103-105 °C; [R]25D (c ) 1, MeOH)
dine: [R]25 (c ) 0.5, MeOH) -29°; 1H NMR (CDCl3) δ 7.65
D
(bs, 1H, CONH), 7.32 (m, 4H), 3.85 (q, 1H, J ) 7 Hz), 3.75 (m,
1H), 3.52 (m, 2H), 3.15 (m, 2H), 2.91 (s, 3H), 2.50 (d, 2H, J )
7 Hz), 2.21-2.05 (m, 2H), 1.95 (m, 1H), 1.80-1.68 (m, 2H),
1.50 (d, 3H, J ) 7 Hz), 0.95 (d, 6H, J ) 7 Hz). Anal.
(C19H30N2O) C, H, N.
1
-68°; H NMR (DMSO-d6) δ 7.30 (d, 2H, J ) 7 Hz), 7.09 (d,
2H, J ) 7 Hz), 3.42 (q, 1H, J ) 7 Hz), 2.81 (s, 3H), 2.45 (d,
2H, J ) 7 Hz), 1.55 (m, 1H), 1.32 (d, 3H, J ) 7 Hz), 0.95 (d,
3H, J ) 7 Hz). Anal. (C14H21NO3S) C, H, N.
(R)-2-(4-Isobutylphenyl)-N-(3-dimethylaminopropyl)-
propionamide (7). Compound 7 was synthesized (72%) using
the same procedure described for 3 starting from (R)-2-(4-
isobutylphenyl)propionic acid and 3-(dimethylamino)propy-
lamine: [R]25D (c ) 0.5, MeOH) -14°; 1H NMR (CDCl3) δ 7.32
(d, 2H, J ) 7 Hz), 7.15 (d, 2H, J ) 7 Hz), 6.77 (bs, 1H, CONH),
3.52 (q, 1H, J ) 7 Hz), 3.35-3.20 (m, 2H), 2.55 (d, 2H, J ) 7
Hz), 2.22 (t, 2H, J ) 7 Hz), 2.05 (s, 6H), 1.95 (m, 1H), 1.50 (m,
5H), 1.05 (d, 6H, J ) 7 Hz). Anal. (C18H30N2O) C, H, N.
(R)-2-(4-Isobutylphenyl)-N-[2-(2-hydroxyethoxy)ethyl]-
propionamide (8). Compound 8 was synthesized (65%) using
the same procedure described for 3 starting from (R)-2-(4-
(R)-2-(4-Trifluoromethanesulfonyloxyphenyl)pro-
pionamide (3). Thionyl chloride (4.8 mL, 67 mmol) was added
to a solution of (R)-2-(4-trifluoromethanesulfonyloxyphenyl)-
propionic acid (10 g, 33.5 mmol) in dry CH2Cl2 (130 mL), and
the resulting solution was refluxed for 2 h. After cooling at rt,
toluene and thionyl chloride were removed under vacuum, and
the residue was dissolved in CH2Cl2 (25 mL); gaseous ammonia
was bubbled into the solution for 1 h. The organic solution
was washed with water (2 × 50 mL), and after drying with
Na2SO4, solvent was removed under vacuum and the crude
was suspended in n-hexane. After stirring at rt for 16 h, pure
isobutylphenyl)propionic acid and 2-(2-aminoethoxy)ethanol.
1
3 was isolated by filtration as a white powder (8.1 g, 27.2
[R]25 (c ) 3, EtOH) -3.5°; H NMR (CDCl3) δ 7.23 (d, 2H, J
D
1
mmol, 81%) mp 67-69 °C; [R]25 (c ) 0.5, MeOH): -12°; H
) 7 Hz), 7.13 (d, 2H, J ) 7 Hz), 5.77 (bs, 1H, CONH), 3.75-
3.33 (m, 9H), 2.47 (d, 2H, J ) 7 Hz), 1.85 (m, 1H), 1.63 (d, 3H,
J ) 7 Hz), 0.94 (d, 6H, J ) 7 Hz). Anal. (C17H27NO3) C, H, N.
(S)-2-[(R)-2-(4-Isobutylphenyl)propionylamino]pro-
pionic Acid (9). To a solution of (R)-2-(4-isobutylphenyl)-
propionic acid (5 g, 24.24 mmol) in dioxane (10 mL) was added
thionyl chloride (2.71 mL, 36.36 mmol), and the resulting
solution was heated at reflux for 3.5 h. After cooling at rt the
acyl chloride was dissolved in DMF (20 mL) at T ) 0 °C, and
DCC (5 g, 24.24 mmol) and HOBT (3 g, 22.2 mmol) were added
under stirring. After 30 min a solution of L-alanine methyl
ester hydrochloride (3.2 g, 22.2 mmol) and triethylamine (3
mL) in DMF (5 mL) was added. The resulting mixture was
left stirring for 2 h at 0 °C and overnight at rt. The precipitated
DCU was filtered off; the filtrate was diluted with EtOAc (50
D
NMR (CDCl3) δ 7.48 (d, 2H, J ) 7 Hz), 7.22 (d, 2H, J ) 7 Hz),
5.38 (bs, 2H, CONH2), 3.65 (q, 1H, J ) 7 Hz), 1.55 (d, 3H, J )
7 Hz). Anal. (C10H10F3NO4S) C, H, N.
(R)-2-(4-Trifluoromethanesulfonyloxyphenyl)-N-[2-(pyr-
rolidin-1-yl)ethyl]propionamide (4). Compound 4 was
synthesized (75%) using the same procedure described for 3
starting from (R)-2-(4-trifluoromethanesulfonyloxyphenyl)-
propionic acid and 2-(pyrrolidin-1-yl)ethylamine: [R]25 (c )
D
1, MeOH) -34°; 1H NMR (CDCl3) δ 8.65 (bs, 1H, CONH), 7.75
(d, 2H, J ) 7 Hz), 7.22 (d, 2H, J ) 7 Hz), 4.02 (m, 2H), 3.85-
3.74 (m, 3H), 3.31 (m, 2H), 3.0-2.85 (m, 2H), 2.41-2.12 (m,
4H), 1.65 (d, 3H, J ) 7 Hz). Anal. (C16H21F3N2O4S) C, H, N.
(R)-2-(4-Trifluoromethanesulfonyloxyphenyl)-N-[3-(pyr-
rolidin-1-yl)propyl]propionamide (5). Compound 5 was

2478 Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7
Aureli et al.
(11) Domenici, E.; Bertucci, C.; Salvadori, P.; Wainer, I. W. Use of a
human serum albumin-based chiral stationary phase for high
performance liquid chromatography for the investigation of
protein binding: Detection of the allosteric interaction between
warfarin and benzodiazepine binding sites. J. Pharm. Sci. 1991,
80, 164-169.
(12) Colmenarejo, G.; Alvarez-Pedraglio, A.; Lavandera, J. L. Chem-
informatic models to predict binding affinities to human serum
albumin. J. Med. Chem. 2001, 44, 4370-4378.
(13) Deschamps-Labat, L.; Pe´hourcq, F.; Jagou, M.; Bannwarth, B.
Relationship between lipophilicity and binding to human serum
albumin of arylpropionic acid nonsteroidal and antiinflammatory
drugs. J. Pharm. Biomed. Anal. 1997, 16, 223-229.
(14) Kratochwil, N. A.; Huber, W.; Mu¨ller, F.; Kansy, M.; Gerber, P.
R. Predicting plasma protein binding of drugs: A new approach.
Biochem. Pharmacol. 2002, 64, 1355-1374.
(15) Ermondi, G.; Lorenti, M.; Caron, G. Contribution of ionization
and lipophilicity to drug binding to albumin: A preliminary step
toward biodistribution prediction. J. Med. Chem. 2004, 47,
3949-3961.
(16) Rahim, S.; Aubry, A. F. Location of binding sites on immobilized
human serum albumin for some nonsteroidal antiinflammatory
drugs. J. Pharm. Sci. 1995, 84, 949-952.
(17) Allegretti, M.; Cesta, M. C.; Bizzarri, C.; Sabbatini, V.; Bertini,
R.; Caselli, G.; Gandolfi, C. Amides of R-2-(aminoaryl)-pro-
pionic acids for use in the prevention of leucocyte activation.
WO0179189, 2001.
(18) Schadendorf, D.; Moller, T.; Algermissen, B.; Worm, M.; Sticher-
ling, M.; Czarnetzki, B. M. IL-8 produced by human malignant
melanoma cells in vitro is an essential autocrine growth factor
J. Immunol. 1993, 151, 2667-2675.
(19) Lee, J.; Horuk, R.; Rice, G. C.; Bennett, G. L.; Camerato, T.;
Wood, W. I. Characterization of two high affinity human
interleukin-8 receptors. J. Biol. Chem. 1992, 15, 16283-16287.
(20) Bizzarri, C.; Allegretti, M.; Di Bitondo, R.; Cervellera, M. N.;
Colotta, F.; Bestini, R. Pharmacological inhibition of Interleu-
kin-8 (CXCL8) as a new approach for the prevention and
treatment of several human diseases. Curr. Med Chem.-Anti-
inflammatory Anti-Allergy Agents 2003, 2, 67-79.
mL), and the organic phase was washed with 10% buffer citric
acid (2 × 20 mL), with a saturated solution of NaHCO3 (2 ×
20 mL), and then with brine (20 mL). After drying with Na2-
SO4, solvent was evaporated to give a crude product (3.86 g)
that was suspended in n-hexane (60 mL) and left stirring
overnight at rt. (S)-2-[(R)-2-(4-isobutylphenyl)propionylamino]-
propionic acid methyl ester was isolated by filtration as a white
powder (4.9 g, 16.84 mmol). To a solution of the methyl ester
in dioxane (9 mL) was added 1 N NaOH (17 mL), and the
mixture was left stirring overnight at rt. An ice/water mixture
(130 mL) was added and the resulting mixture was acidified
with H2SO4 concentrated to pH ) 2. The aqueous phase was
extracted with CH2Cl2 (4 × 20 mL), and the collected organic
extracts were washed with brine (20 mL), dried over Na2SO4,
and evaporated under vacuum to give an oily residue. 9 was
isolated by crystallization from ethyl ether (30 mL) as a white
solid (2.55 g, 38%): mp 125-128 °C; [R]25D (c ) 1, MeOH) -44°;
1H NMR (CDCl3) δ 7.21 (d, 2H, J ) 7 Hz), 7.12 (d, 2H, J ) 7
Hz), 5.95 (bs, 1H, CONH), 4.54 (q, 1H, J ) 7 Hz), 3.62 (q, 1H,
J ) 7 Hz), 2.47 (d, 2H, J ) 7 Hz), 1.85 (m, 1H), 1.53 (d, 3H, J
) 7 Hz), 1.35 (d, 3H, J ) 7 Hz), 0.93 (d, 6H, J ) 7 Hz). Anal.
(C16H23NO3) C, H, N.
3-[4-((R)-2-(Methylsulfonylamino)-1-methyl-2-oxoeth-
yl)phenyl]-2-methylpropionic Acid (10). Compound 10 was
isolated by extraction from the collected urine of three rats
treated with compound 2 (1 g/die) po for a single day. The
collected urine (about 30 mL) was acidified with 37% HCl and
extracted with EtOAc (2 × 15 mL). The collected organic
extracts were dried over Na2SO4, and the solvent was evapo-
rated under vacuum. Pure 10, as mixture of diastereoisomers,
was isolated by silica gel column chromatography (eluting
mixture CHCl3:MeOH, 9:1) as a white solid (0.65 g, 2.07
mmol): mp 127-140 °C (diastereoisomers mixture); 1H NMR
(CDCl3) δ 8.20 (bs, 1H, CONH), 7.12 (s, 4H), 3.66 (q, 1H, J )
7 Hz), 3.21 (s, 3H), 3.01 (m, 1H), 2.78 (m, 2H), 1.51 (d, 3H, J
) 7 Hz), 1.23 (dd, 3H, J1 ) 7 Hz, J2 ) 5 Hz). Anal. (C14H19-
NO5S) C, H, N.
(21) Liu, Z.; Giudice, G. J.; Zhou, X.; Swartz, S. J.; Troy, J. L.; Fairley,
J. A.; Till, G. O.; Diaz, L. A. A major role for neutrophils in
experimental bullous pemphigoid. J. Clin Invest. 1997, 100,
1256-1263.
(22) Bertini, R.; Allegretti, M.; Bizzarri, C.; Moriconi, A.; Locati, M.;
Zampella, G.; Cervellera, M.; Di Ciccio, V.; Cesta, M.; Galliera,
E.; Martinez, F.; Di Bitondo, R.; Troiani, G.; Sabbatini, V.;
D’Annibale, G.; Anacardio, R.; Cutrin, J.; Cavalieri, B.; Mainiero,
F.; Strippoli, R.; Villa, P.; Di Girolamo, M.; Martin, F.; Gentile,
M.; Santoni, A.; Corda, D.; Poli, G.; Mantovani, A.; Ghezzi, P.;
Colotta, F. A new class of non-competitive allosteric inhibitors
of the inflammatory chemokine receptors CXCR1 and CXCR2:
Prevention of reperfusion injury. Proc. Natl. Acad. Sci. U.S.A.
2004, 101, 11791-11796.
(23) Oprea, T. I.; Waller, C. L. Theoretical and Practical aspects of
three-dimensional Quantitative Structure-Activity Relation-
ships. In Reviews in Computational Chemistry; Lipkowitz, K.
B., Boyd D. B., Eds.; Wiley-VCH: New York, 1997; 11, pp 127-
182.
Diastereisomeric purity was assayed by HPLC-UV (Chiral-
cel OJ column, eluting mixture 75:15:10 n-hexane:EtOH:iPrOH
and 0.5% trifluoroacetic acid), and tR values for [2R,2′(R,S)]
were 8.4 and 9.4 min, the ratio being 55:45.
Supporting Information Available: Elemental analyses
of the described compounds. This material is available free of
charge via the Internet at http://pubs.acs.org.
References
(1) Peters, T. All about Albumin: Biochemistry, Genetics and
Medical Applications; Academic Press: San Diego, CA, 1996.
(2) Carter, D. C.; Ho, J. X. Structure of serum albumin. Adv. Protein
Chem. 1994, 45, 153-203.
(3) Watanabe, H.; Tanase, S.; Nakajou, K.; Maruyama, T.; Kragh-
Hansen, U.; Otagiri, M. Role of Arg-410 and Tyr-411 in human
serum albumin for ligand binding and esterase-like activity.
Biochem. J. 2000, 349, 813-819.
(4) He, X. M.; Carter, D. C. Atomic structure and chemistry of
human serum albumin. Nature 1992, 358, 209-215.
(5) Sugio, S.; Mochizuki, S.; Noda, M.; Kashima, A. Crystal structure
of human serum albumin at 2.5 Å resolution. Protein Eng. 1999,
12, 439-446.
(6) Petitpas, I.; Bhattacharya, A. A.; Twine, S.; East, M.; Curry, S.
Crystal structure analysis of warfarin binding to human serum
albumin: Anatomy of drug site I. J. Biol. Chem. 2001, 276,
22804-22809.
(7) Gerig, J. T.; Katz, R. E.; Reinheimer, J. D.; Sullivan, J. R.;
Roberts, J. D. Examination of the aspirin acetylation site of HSA
by 13C NMR spectroscopy. Org. Magn. Reson. 1981, 15, 158-
161.
(8) Diaz, N.; Suarez, D.; Sordo, T. L. Theoretical study of the water-
assisted aminolysis of â-lactams: Implications for the reaction
between human serum albumins and penicillins. J. Am. Chem.
Soc. 2000, 122, 6710-6719.
(24) Mao, H.; Hajduk, P. J.; Craig, R.; Bell, R.; Borre, T.; Fesik, S.
W. Rational design of diflunisal analogues with reduced affinity
for human serum albumin. J. Am. Chem. Soc. 2001, 123, 10429-
10435.
(25) Benet, L. Z.; Øie, S.; Schwartz, J. B. Design and optimization of
dosage regimens; pharmacokinetic data. In The Pharmacological
Basis of Therapeutics; 9th ed.; Gilman’s, A. G., Goodman, L. S.,
Eds.; McGraw-Hill: New York, 1998; pp 1707-1792.
(26) SYBYL, Molecular Modelling Software, version. 6.9.2; Tripos
Assoc.: St. Louis, Missouri.
(27) CONFORT, Molecular Modelling Software, version 6.0; Tripos
Assoc.: St. Louis, Missouri.
(28) GRID, version 22; available from Molecular Discovery Ltd.:
London, U.K. (www.moldiscovery.com).
(29) VOLSURF, version 4.0; available from Molecular Discovery
Ltd.: London, U.K. (www.moldiscovery.com).
(30) ALMOND, version 3.2.0; available from Molecular Discovery
Ltd.: London, U.K. (www.moldiscovery.com).
(31) GOLPE, version 4.5; available from Multivariate Infometric
Analysis srl: Perugia, Italy. (www.miasrl.com).
(32) http://www.rcsb.org/pdb.
(33) Goodford, P. J. A computational procedure for determining
energetically favorable binding sites on biologically important
macromolecules. J. Med. Chem. 1985, 28, 849-857.
(34) Wade, R. C.; Clark, K. J.; Goodford, P. J. Further development
of hydrogen bond functions for use in determining energetically
favorable binding sites on molecules of know structure. 1. Ligand
probe groups with the ability to form two hydrogen bonds. J.
Med. Chem. 1993, 36, 140-147.
(9) Diaz, N.; Suarez, D.; Sordo, T. L.; Merz, K. M. Molecular
dynamics study of the binding site in human serum albumin:
Influence of the protonation state of Lys195 and Lys199. J. Med.
Chem. 2001, 44, 250-260.
(10) Bhattacharya, A. A.; Curry, S.; Franks, N. P. Binding of the
general anesthetics propofol and halothane to human serum
albumin. High-resolution crystal structures. J. Biol. Chem. 2000,
275, 38731-38738.

Predicting Affinity of CXCL8 Inhibitors by 3D-QSPR
Journal of Medicinal Chemistry, 2005, Vol. 48, No. 7 2479
(35) Cruciani, G.; Crivori, P.; Carrupt, P. A.; Testa, B. Molecular fields
in quantitative structure-permeation relationships: The Volsurf
approach. J. Mol. Struct. 2000, 503, 17-30.
(36) Crivori, P.; Cruciani, G.; Carrupt, P. A.; Testa, B. Predicting
blood-brain barrier permeation from three-dimensional molec-
ular structure. J. Med. Chem. 2000, 43, 2204-2216.
(37) Zamora, I.; Oprea, T.; Cruciani, G.; Pastor, M.; Ungell, A. L.
Surface descriptors for protein-ligand affinity prediction. J.
Med. Chem. 2003, 46, 25-33.
(38) Pastor, M.; Cruciani, G.; McLay, I.; Pickett, S.; Clementi, S.
GRid-INdependent descriptors (GRIND): A novel class of align-
ment-independent three-dimensional molecular descriptors. J.
Med. Chem. 2000, 43, 3233-3243.
(39) Baroni, M.; Costantino, G.; Cruciani, G.; Riganelli, D.; Valigi,
R.; Clementi, S. Generating Optimal Linear PLS Estimations
(GOLPE): An advanced chemometric tool for handling 3D-QSAR
problems. Quant. Struct.-Act. Relat. 1993, 12, 9-20.
(40) Cruciani, G.; Clementi, S. GOLPE: Philosophy and Applications
in 3D-QSAR. In Advanced Computer-Assisted Techniques in
Drug Discovery; van de Wateerbremd, H., Ed.; VCH: Weinheim,
1994; pp 61-88.
(41) Cruciani, G.; Clementi, S.; Baroni, M. Variable Selection in PLS
Analysis. In 3D QSAR in Drug Design, Theory, Methods and
Applications; Kubinyi, H., Ed.; ESCOM: Leiden, 1993; pp 551-
564.
(42) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Wold, S. PLS.
In Multi and Megavariate Data Analysis using Projection
Methods (PCA & PLS); Umetrics AB: Umea, 1999; pp 69-112.
(43) Kastenholtz, M. A.; Pastor, M.; Cruciani, G.; Haaksma, E. E.;
Fox, T. GRID/CPCA: A new computational tool to design
selective ligands. J. Med. Chem. 2000, 43, 3033-3044.
(44) Whitlam, J. B.; Brown, K. F. Ultrafiltration in serum protein
binding determinations. J. Pharm. Sci. 1981, 70, 146-50.
(45) Bowers, W. F.; Fulton, S.; Thompson, J. Ultrafiltration vs
equilibrium dialysis for determination of free fraction. Clin.
Pharmacokinet. 1984, 9, 49-60.
(46) Kwong, T. C. Free drug measurements: Methodology and clinical
significance. Clin. Chim. Acta 1985, 151, 193-216.
(47) Akgun, H.; Tozkoparan, B.; Ertan, M.; Aksu, F.; Inan, S. Y.
Synthesis of some 2-arylpropionic acid amides as prodrugs.
Arzneimittelforschung 1996, 46, 891-894.
JM049227L