Institut für Pharmazeutische Chemie,
Heinrich-Heine-Universität Düsseldorf, Germany (F.S.,
H.-D.H.); Institut für Pharmakologie und Klinische Pharmakologie,
Heinrich-Heine-Universität Düsseldorf, Germany (S.L., T.H.,
K.S.); Department of Chemistry, Purdue Unversity, West Lafayette,
Indiana (P.L.F.)
Prostacyclin is an endogenous mediator that shows potent platelet
inhibitory activity and powerful relaxation of peripheral resistance
vessels. Prostacyclin receptor agonists are valuable drugs in the
treatment of various vascular diseases spanning primary pulmonary
hypertension to Raynaud's syndrome. Although agonists from
various structural classes were synthesized, a common pharmacophore was
never defined. Therefore, an attempt was made to integrate the
different agonists into a single model. A dataset of structurally diverse prostacyclin receptor agonists was tested for its affinity to
the human platelet prostacyclin receptor. The dataset included prostanoid and nonprostanoid ligands comprising iloprost, cicaprost, and BMY45778. Extensive conformational analyses were performed for both classes of compounds because of the absence of rigid templates. The search and superimposition procedure yielded a pharmacophore that aligns the essential carboxylate group of the agonists as well as demonstrates that different functional groups in
prostanoid and nonprostanoid agonists can be arranged in a uniform
conformation. A three-dimensional quantitative structure-activity relationship study was performed using the programs GRID and GOLPE. This analysis yielded a cross-validated correlation coefficient of
0.77. With this model, it is possible to predict the affinity of
untested compounds.
 |
Introduction |
Prostacyclin
(PGI2), an endogenous mediator, is synthesized
primarily in the vascular endothelium. It plays an important role in
the regulation of blood flow; it is a potent vasodilator and inhibits
platelet aggregation. Both actions are mediated by a specific G-protein
coupled receptor, the prostacyclin receptor (IP receptor). The binding
of PGI2 to this receptor leads to the coupling of
Gs protein to adenylate cyclase and subsequent
elevation of intracellular cAMP levels.
PGI2 is a clinically useful agent for the precise
control of platelet function. Its use is impaired by its instability:
the enol ether linkage of PGI2 is spontaneously
hydrolyzed with a half-life of 3 min (Stehle, 1982
; Armstrong, 1996),
limiting therapeutic application to parenteral administration. Much
effort was therefore directed toward developing metabolically stable
and orally available IP-receptor agonists. These can be divided into
two groups: the so-called prostanoid agonists preserve the
characteristic structural features found in PGI2,
namely the carboxylate group and hydroxyl functions at C-9 and C-15
(see Fig. 1), whereas the
nonprostanoid agonists do not show any structural similarities with
PGI2 except for the essential carboxylate group.
As can be seen in Fig. 2, the skeletons
of the nonprostanoid IP receptor agonists differ considerably. In some
compounds, putative hydrogen bond acceptors such as heterocycles (i.e.,
oxazole) or oxime groups serve instead of the hydroxyl groups.
Several groups have previously attempted to define a pharmacophore for
prostacyclin receptor agonists (Tsai et al., 1991; Meanwell et al.,
1993a,b), and structure-activity relationships have been described in
great detail for both types of compounds (Skuballa and Vorbruggen,
1983; Nickolson et al., 1985; Armstrong et al., 1986; Tsai and Wu,
1989; Meanwell et al., 1992a,b,c, 1993a,b, 1994a,b; Jones et al., 1993;
Muir et al., 1993), but no pharmacophore that integrates both groups of
compounds has previously appeared.
Correlation of the previous attempts in conjunction with incorporation
of new compounds of different structural classes were united into a
common agonistic pharmacophore. Extensive conformational searching was
required in the absence of a rigid lead compound. The superimposition
of one prostanoid [(S)-iloprost] and one nonprostanoid agonist (BMY45778) formed the basis of the pharmacophore, and all other
compounds were modeled on these two agonists. Human platelet affinity
data served as input for a 3D QSAR study to quantify the structure
affinity relationships. The resulting 3D-QSAR model showed a
cross-validated correlation coefficient of 0.77. The model predicts the
binding affinity for untested compounds and serves as a tool for the
development of new high affinity ligands.
| |
Materials and Methods |
Materials.
Cicaprost, iloprost, and nileprost were provided
by Dr. F. M. McDonald (Schering, Berlin, Germany). Prostaglandin
E1 was from Dr. P. Ney (Schwarz Pharma, Monheim,
Germany). All BMY compounds were from Dr. N. A. Meanwell (Bristol
Myers Squibb, Wallingford, CT). ONO-1301 was obtained from Dr. K. Kondo
(ONO Pharmaceuticals, Osaka, Japan). Dr. R. A. Armstrong
(University of Edinburgh, Edinburgh, UK) provided EP 157.
Binding Assay.
Human platelet membranes were prepared as
described previously (Kaczmarek et al., 1993) and suspended in 100 mM
NaCl, 20 mM Tris-HCl, 5 mM CaCl2, and 10 mM
glucose, with pH adjusted to 7.4. For ligand binding analysis, 200-µl
aliquots of platelet membrane suspension were incubated with 10 nM
[3H]iloprost (Amersham Biosciences,
Braunschweig, Germany) and 1 nM to 1 µM concentrations of the
respective compounds. Nonspecific binding was measured in the presence
of 10 µM iloprost. Equilibration was allowed for 90 min at 15°C.
Thereafter, bound radioactivity was separated from free by rapid
filtration (GF/C filters; Whatman, Maidstone, UK) and 3 washes with 4 ml of suspension buffer (4°C). Radioactivity was determined with
standard liquid scintillation techniques.
Platelet PGI2 receptor affinity
(KI) for iloprost was determined by
nonlinear fitting analysis of the data obtained from the displacement
of [3H]iloprost by iloprost. The
KI values of all other compounds were calculated by the formula: KI = EC50/[1 + (concentration
radioligand/KI radioligand)], where
EC50 is the concentration of the ligand under investigation required to displace 50% of radioligand
([3H]iloprost) from specific binding. The
KI values obtained for the individual
compounds (Figs. 1 and 2) are means from at least three independent measurements.
Molecular Modeling.
All structures were generated using the
SYBYL software package (SYBYL 6.5; Tripos Inc., St. Louis, MO). The
carboxylate group was always deprotonated to imitate physiological
conditions. Partial atomic charges were calculated with the
Gasteiger-Hückel method (Gasteiger and Marsili, 1980). Energy
minimizations and conformational analyses also employed the SYBYL
software. Energy minimizations were performed using first the steepest
descent method (500 steps) and then refined using conjugate gradient to
a gradient of 0.05 kcal/mol Å.
For the conformational analysis no rigid template was available;
therefore, 16S-iloprost and BMY45778 were chosen
because of their structural diversity. A comparison of the
conformational possibilities of the compounds was used to outline the
conformational space both can occupy. By this procedure, a template for
the fitting of the other agonists was created. Extensive conformational
analyses were performed to determine the putative binding conformations of the two ligands.
The conformational space of BMY45778 was studied using systematic
conformational analysis. Because of the huge number of possible conformations, not all rotatable bonds were treated at once. Initially only the bonds between phenyl and oxazole rings were rotated using a
10° increment to assure high accuracy; the side chain remained in an
extended conformation. More than 300,000 conformations were obtained in
this way. They could be classified into 45 families. The lowest energy
conformer was chosen as representative of each family and minimized.
Approximately half of the family representatives could be eliminated
because their carboxylate group was near one of the phenyl rings.
Because this carboxylate group is essential for high affinity (Tsai and
Wu, 1989) it has to be accessible to the interacting amino acid
(Arg279). The lowest energy conformer of each of the eight remaining
collections was selected for further consideration.
In preparation for systematic analysis of the side chain, a preliminary
fit with the molecular features of 16S-iloprost (i.e., the
bicyclooctane ring and the omega chain) was carried out (for nomenclature, see Fig. 3). Because a
systematic conformational search is not an effective method to explore
cyclic systems, Random Search, a Monte Carlo method, was chosen (SYBYL
6.5). The Random Search was performed with a 16S-iloprost
fragment (Fig. 4) and was repeated seven
times with different starting conformations to insure that all
low-energy conformations were found.
View larger version (21K):
[in this window]
[in a new window]
|
Fig. 3.
Conformational analysis of
16S-iloprost and BMY45778. Numbers, superimposed atoms;
dark arrows, increment 10°; lighter arrows, increment 30°; LP, lone
pair.
|
|
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 4.
The six conformations for the bicyclooctane ring
system of iloprost received with Random Search.
|
|
Random Search was also used for analysis of cicaprost and
isotetralynaprost (the side chains were truncated to speed up the calculation). Seven runs with different starting structures were done
for each compound, each run consisting of 1000 cycles. In a cycle, the
bonds defined as rotatable were set to random torsion angles and the
compounds minimized subsequently. The resulting conformers were
compared with those already found. Side chains were then reintroduced
to the conformers chosen for superimposition with BMY45778.
A systematic search was performed with the omega chains of both
iloprost conformations that were chosen. Using an increment of 30°,
it was possible to cover the entire conformational space with
reasonable accuracy. For both 16S-iloprost conformers more than 10,000 side chain conformations were classified into 89 (85) families using IXGROS (Sippl, 1997). Again, all family representatives were minimized to a gradient of 0.05 kcal/mol Å using the Conjugate Gradient method.
Further reduction could not be achieved without taking into account the
conformational space of BMY45778. In this study, family representatives
of both compounds were superimposed with each other using the fit
points 1 to 5 depicted in Fig. 3; the root-mean-square deviation was
restricted to 
1.5. This preliminary superimposition reduced the
number of conformers to be further considered to 4 for BMY45778 and 24 (23) for 16S-iloprost. Each of these conformers was then
subjected to an incremental (30°) systematic search of the (
) side
chain. The conformers obtained were again divided into families and
each of the 14 family representatives of BMY45778 was superimposed with
each of the 21 + 20 representatives of 16S-iloprost. For
this final superimposition, all six fit points were considered and a
tighter root-mean-square deviation of 1.1 was applied.
The chosen fit points considered both steric correspondences as well as
the existence of a similar hydrogen bond pattern. For the fit points 1 and 2, the ends of the corresponding lone pairs were used to insure
that both functional groups could interact with the same amino acid in
the receptor.
For the other ligands, a shorter procedure was used, only cicaprost was
treated as extensively as 16S-iloprost to test the proceeding. The diphenyloxazole derivatives and
PGE1 were integrated into the pharmacophore using
the Multifit option of SYBYL.
Multifit is a flexible fitting method in which pairs of atoms are
forced onto each other by a force constant during a minimization procedure. Because the pairing atoms have to be chosen explicitly, this
method is applicable only for very similar compounds. The reference
ligand was always treated as a rigid entity; the force constants used
were high (50 kcal/mol Å2) for the very
important fit points chosen in analogy to Fig. 3 and more relaxed for
all others (20 kcal/mol Å2).
For isotetralynaprost (Jakubowski et al., 1994), Random Search
was applied to find favorable ring conformations. The conformation that
is similar to the one used for 16S-iloprost was chosen and side chains were adjusted using Multifit. To integrate
16R-iloprost into the pharmacophore, the conformation
of 16S-iloprost was adopted, position 16 was inverted, and
the epimer was minimized subsequently.
For EP 157 and ONO-1301, a Multifit was not possible because their
structural homology with other ligands is insufficient. Inclusion of
these latter two structures was accomplished using the pharmacophoric
superimposition program FLEXS (Lemmen et al., 1998a). The validity of
this approach was verified by fitting (S)- and
(R)-nileprost on the reference ligand
16S-iloprost. FLEXS (Lemmen and Lengauer, 1997; Lemmen,
1998a,b) was designed to flexibly superimpose pairs of molecules. One
of the molecules, the reference ligand, is kept rigid. The other one is
split into several fragments that are fitted incrementally on the
reference ligand, starting with the base fragment. Many different
solutions are obtained and are ranked by a scoring function. All
compounds had Gasteiger-Hückel charges and the carboxylate group
was chosen as the base fragment. The best 20 solutions were inspected
visually, and the selected solution was energy minimized.
To ensure that the conformation obtained by Multifit or FLEXS was not
dramatically changed, parts of the ligand were fixed in the beginning
of the minimization and only relaxed step-by-step using Steepest
Descent and Conjugate Gradient.
MEPs.
Molecular electrostatic potentials (MEPs) were
computed with the HF3-21G* basis set using SPARTAN 5.1.1 (Wavefunction
Inc., Irvine, CA).
Using the program GRID 16 (Molecular Discovery Ltd., London, Great
Britain; Goodford, 1985), interaction energies between a compound and a
probe can be calculated. A wide variety of probes with very different
chemical features are available. They are designed to cover the
different qualities of atoms in a protein-binding site so that the
surroundings can be mapped in an indirect way. The probe is placed on
each point of a grid that is created around the ligand and the
interaction energy is then calculated. The types of noncovalent
interaction accounted for in the GRID program are steric,
electrostatic, and hydrogen-bonding energies. The results can be
visualized by contouring isoenergy levels.
For each compound, a grid with a spacing of 1 Å was generated and GRID
fields were calculated using an amidic NH probe (N1, simulating a
hydrogen bond donor), the carbonyl probe (a hydrogen bond acceptor),
and the amidine probe. With the latter, interactions between the
essential carboxylate group of the ligands and an arginine of the
binding pocket (Arg279) were simulated. Best results for hydrophobic
interactions were received with the dry probe (a "dry" water molecule).
The Program GOLPE 4.0 (Multivariate Infometric Analysis, Perugia,
Italy) is used to statistically analyze three-dimensional molecular
fields and to correlate the important data points with biological data.
For this 3D-QSAR analysis, interaction fields calculated with GRID were
used as input. Several probes were tested in order to determine which
was best suited for the description of the differences between the
compounds. Principal component analysis showed that of all probes used
the NH = probe (sp2 hybridized NH with a
lone pair) could distinguish best between the compounds.
The interaction field between the NH = probe and the ligands was
calculated as described before: the pure enantiomers were placed into a
standardized grid (grid spacing 1 Å) and interactions were computed
with different GRID probes. The interaction energies obtained between
each compound and the probe as well as the affinity to the prostacyclin
receptor served as input for GOLPE. The preliminary model calculated
with this probe contained 10,098 x variables for each
compound (interaction values, x variables; affinity, y variable). Most of these variables are not meaningful for
the explanation of the differences in affinity and introduced noise into the statistical PLS analysis. This noise was eliminated during the
data pretreatment procedure for variable selection.
Data Pretreatment.
GRID points with interaction energies
near to zero (0.03) as well as those with a very low standard
deviation of 0.02 were eliminated. Grid points where all but one
compound have the same interaction value (2-level variables) as well as
3- and 4-level variables were discarded. By this procedure the number
of x variables was reduced from 10098 per ligand to 4473.
The approach using D-optimal preselection of variables
searches for the most informative variables since most grid points do
not contain information with relevance for biological data. Applying
this method enabled a further reduction of x-variables to 2236. The
"smart region" definition (Pastor et al., 1997) was subsequently
employed to reduce the number of groups from 1009 to 740. In the last
step, a fractional factorial design procedure was employed to optimize
the predictability of the model. The final model contained 959 x variables, with three principal components being used.
Cross validation of the model was done using the leave-one-out method
and the leave-20%-out method (five random groups). For the first
method, one compound is not used for the generation of a new model and
its affinity is predicted using that new model. The model building and
prediction cycle is repeated until each compound was left out once. A
correlation coefficient q2 is calculated from the
correlation between experimental and predicted pKI values. The second method works
the same way but the compounds are distributed randomly into five
groups and each group is left out once.
The remaining compounds were used as an external data set. The
isomers/enantiomers were treated separately and the average of the
predictions for both isomers was calculated and compared with the
binding data.
| |
Results |
Binding Assay.
The experiments showed clearly that all
compounds displaced [3H]iloprost in a
competitive way from the binding site. This is exemplified in Fig.
5 for isotetralynaprost, iloprost, BMY45778, and BMY43450.
Pharmacophore Development.
To define a common pharmacophore
for prostanoid and nonprostanoid agonists, a set of structurally
diverse agonists was chosen and their affinity to the prostacyclin
receptor was measured using human platelet membranes.
The 21 compounds used (five of which are mixtures of
isomers) are listed in Figs. 1 and 2 together with the
corresponding affinity data.
The conformational analysis of the BMY45778 (leaving out the side
chain) resulted in 45 family representatives. A superimposition of all
45 conformations is shown in Fig. 6. The
phenyl rings of the biphenyl oxazole moiety show two distinct
arrangements, both leading to maximum planarity.
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 6.
Superimposition of 45 family representatives of
BMY45778. To ensure clearness only a sphere indicates the position of
the carboxylate group for most structures. Left, top view; right, side
view.
|
|
For the two remaining torsion angles (one between the oxazole rings and
one between the second oxazole ring and the side chain phenyl ring),
only a small number of low energy conformations was observed (Fig. 6).
The two oxazole rings always lie in the same plane but the nitrogen
atoms point in the same or in opposite directions. The side-chain
phenyl ring also arranges itself in a pseudoplanar fashion but real
planarity is not possible because of steric repulsion. Therefore, the
side chain is found pointing toward the biphenyl oxazole ("closed
conformation") or outward ("elongated conformation").
Initially, the conformational analysis of 16S-iloprost was
restricted to the central bicyclooctane ring. Six conformations were
detected (Fig. 4). To reduce this number further, a comparison with
crystal structures for similar fragments was done. Of 16 hits found in
the Cambridge Structural Database (Allen and Kennard, 1993; Bruno et
al., 1997), 10 could be assigned to the conformations 3 and
4, whereas two hits were found for conformations 1, 2, and 6. There were no hits for
conformation 5; therefore, this conformation was eliminated.
As mentioned above, all conformers of BMY45778 showed an arrangement
that was as planar as possible. 16S-iloprost has to fit into
the same binding pocket as BMY45778 and should therefore adopt a
conformation that presents the same overall shape. Because of these
restrictions, side chains were added only to the more planar
bicyclooctane conformations 1 and 3. After the
conformational search of the side chains, all conformers of BMY45778
and 16S-iloprost were superimposed with each other. The superimposition
with the best match is shown in Fig. 7.
The conformation of all other compounds was compared to either BMY45778
or to 16S-iloprost and a superimposition procedure was performed. The
superimposition of all compounds is shown in Fig.
8.
View larger version (48K):
[in this window]
[in a new window]
|
Fig. 8.
Superimposition of all 25 prostacyclin receptor
agonists (15 compounds and 5 isomeric mixtures). Dark gray, BMY 45778;
light gray, 16S-iloprost. A, essential carboxylate
groups; B and C, hydrogen bonding groups; D and E, lipophilic areas.
|
|
To complement these efforts, the MEPs of all compounds were computed
(data not shown). The electronic properties of BMY45778 and
16S-iloprost are quite similar: Both MEPs are dominated by the strong negative potential of the carboxylate group, with the areas
near the oxazole nitrogens and around the hydroxyl groups also
presenting negative character.
To verify the superimposition of the ligands, GRID calculations were
performed. The results obtained with the amidine probe confirmed good
superimposition of the carboxylate groups; very similar interaction
fields were found for all ligands. More diverging results were obtained
with the N1 Probe (amidic nitrogen, hydrogen bond donor). Figure
9 shows that here, again, the interaction fields around the carboxylate groups are very similar but differences can be observed in other areas. The high affinity agonist Cicaprost possesses two distinctive fields around its hydroxylate groups; in
contrast to BMY45778, which shows only one field located in the
prolongation of the nitrogen lone pairs. With the other hydrophilic probes, similar results were obtained, whereas the fields produced by
the hydrophobic probe did not present any interpretable variations (data not shown).
To quantify the results, a 3D-QSAR study was carried out using GOLPE.
Because some of the compounds listed in Figs. 1 and 2 are mixtures of
isomers, only the 15 pure enantiomers could be included in the data
set. For details regarding the experimental data, see Materials
and Methods. Examples of four displacement experiments are shown
in Fig. 5. The model calculated uses three principal components and
achieves a correlation coefficient r2 of 0.96. The correlation after cross-validation with the leave-one-out method is
shown in Fig. 10; the cross-validated
correlation coefficient q2 obtained with this
method was 0.77 (SD of error prediction = 0.37). The
cross-validation with the leave-20%-out method yielded a correlation
coefficient q2 of 0.68.
View larger version (14K):
[in this window]
[in a new window]
|
Fig. 10.
Correlation of 15 prostacyclin receptor agonists
(q2 = 0.77, validation method: leave-one-out).
|
|
Because the data set was rather small, it could not be divided into a
training set and a test set. To prove the predictive power of the
model, the binding affinities of the four racemic compounds and the
E/Z-mixture not included in the QSAR analysis were also calculated. The
results obtained for the isomers are shown in Table
1. For iloprost, BMY44521, BMY43675, and
BMY181331, deviations between the average of the predicted
pKI values and the experimental value
of 0.12 to 0.86 were found. Only nileprost shows a greater deviation of
1.89.
| |
Discussion |
A pharmacophore for prostacyclin receptor agonists was defined on
the basis of a data set of 15 structurally diverse compounds. The
superimposition was based on structural analyses of two agonists: the
prostanoid compound 16S-iloprost was chosen because of its high affinity and the nonprostanoid BMY45778 because of its noticeable structural dissimilarity and its relative rigidity. The conformations found for BMY45778 (excluding the side chain) showed a relatively planar arrangement, which is also found in the crystal structure of
BMY45778 (Meanwell et al., 1993a). Using these results, the conformational space of 16S-iloprost could be restricted so
that only very few ring conformations for the central bicyclooctane ring were possible. After adding the side-chains to both compounds and
exploration of their conformational space, it became clear that the
elongated conformations were preferred. The conformations of BMY45778
and 16S-iloprost that agreed best were used as the basis for
the fitting of all other compounds. The carboxylate groups of both
ligands could be overlaid very well. The C-11 hydroxyl group of
16S-iloprost is in a similar position as one of the oxazole nitrogens of BMY45778. For this nitrogen, an important function as
hydrogen-bond acceptor was described in the structure-activity relationships published earlier (Meanwell et al., 1993a).
The other compounds were superimposed with either BMY45778 or
16S-iloprost and good results were also obtained for them.
Figure 8 shows the pharmacophore deduced from the superimposition: the carboxylate group (region A) is an essential feature of all agonists. In a distance of 8 to 11 Å, a hydrogen bond accepting and/or donating group is important (region B). This group is a hydroxyl group in the
prostanoid agonists and can be a heterocyclic nitrogen, an oxime
nitrogen, or an ester in the nonprostanoid compounds.
For the second hydroxyl group (at C-15) of the prostanoid
agonists, no clear superimposition with other hydrogen bond accepting (or donating) groups could be found because the distance between both
hydroxyl groups is larger than that between the heterocyclic nitrogens
present in some of the nonprostanoid compounds (region C). D indicates
an extended lipophilic area that is formed by aromatic or aliphatic
side chains. Region E is occupied only by side chain phenyl rings of
those nonprostanoid ligands that contain an E-configurated double bond
(BMY44046, BMY44495, E-BMY44521; see Fig. 2).
To complement this information, the environment of the agonists was
scanned. The results obtained with GRID can be used to deduct some
information about the binding pocket in the IP receptor. It is well
known that in many prostaglandin receptors, an arginine of the seventh
transmembrane helix (position 7.40; for nomenclature, see Ballesteros
and Weinstein, 1995) is important for the binding of the ligands
(Negishi et al., 1995; Huang and Tai, 1995; Audoly and Breyer, 1997;
Chang et al., 1997; Kedzie et al., 1998). Because all prostaglandin
receptor ligands possess a carboxylate group, it can be assumed that a
charged hydrogen bond between this carboxylate group and the arginine
is formed. Because no experimental results regarding the IP receptor
were available, the amidine GRID probe was used to mimic this
interaction. Not surprisingly, the fields that were found for the
different ligands were very similar. This supports the assumption that
Arg279 (7.40) of the IP receptor is the binding partner of the
carboxylate group.
The results obtained with the hydrogen bond donating group are very
similar in the area around the carboxylate group, but important
differences can be found in the interaction fields induced by the
central parts of the compounds. The high-affinity agonists produce two
distinctive interaction fields around their hydroxyl groups (which are
rotated by GRID). These two fields allow the conclusion that two amino
acids in the binding pocket interact with these compounds and lead to
their good binding properties. The compounds with medium affinity
(i.e., BMY45778) are able to form only one interaction by accepting a
hydrogen bond. Some low-affinity agonists do not show any interaction
fields in this area because no suitable functional groups are found.
Based on the GRID results, a hypothesis can thus be established: the
higher affinity that is found for most prostanoid compounds can be
explained by the interaction with two additional binding partners in
the receptor. In the nonprostanoid compounds, only one or even none
additional interaction field can be found, and their affinity is much
lower than that of most prostanoid agonists. The low affinity measured for nileprost does not fit into this scheme, but because the cyano group protrudes from the common pharmacophoric volume, it can be
assumed that this compound cannot be accommodated easily by the
ligand-binding pocket.
Because there are no mutagenesis studies published for the IP receptor
itself, it is not easy to phrase a hypothesis regarding the interacting
amino acids in the binding pocket. Kedzie et al. (1998) stated that
iloprost did not activate the EP2 receptor wild-type, whereas activity
was measured for the L304Y mutation. This leads to the conclusion that
the corresponding Tyr281 in the prostacyclin receptor could be
important for agonist binding and/or the activation of the receptor.
In all prostaglandin receptors, a conserved Ser/Thr can be found in the
second extracellular loop. For the EP2 and EP4 receptors, it has been
shown that the mutation of this amino acid to Ala decreased the ligand
binding affinity considerably, whereas a mutation from Thr to Ser had
no effect (Stillman et al., 1998). In the IP receptor, a Ser (Ser168)
can be found at the corresponding position.
Both amino acids are plausible binding partners for the
prostanoid and nonprostanoid compounds (hydroxyl groups, heterocycles, oxime functions; see Fig. 8, region B and C). The GRID probes used to
calculate the fields shown in Fig. 9 correspond well with these
experimental results. A more detailed analysis of ligand receptor
interactions will only be possible after the construction of a
prostacyclin receptor model, but it has to be mentioned that even then,
far-reaching conclusions about the stimulus transfer will not be conceivable.
It is achievable, however, to quantify the structure-activity
relationships and predict the affinity of untested compounds using QSAR
methods. The basis of the 3D-QSAR approach used here (program GOLPE) is
a statistical analysis of GRID fields. It could be shown in many cases
that areas found to be important for the explanation of affinity
differences can be superimposed with amino acids of the binding site
when the structure of the protein is known (for example, see Sippl and
Höltje, 2000).
The results of the 3D-QSAR study can be regarded as very satisfactory
for a data set that is as heterogeneous as the one used here. A good
correlation of r2 = 0.96 could be obtained and
the cross-validated correlation coefficient (q2)
of 0.77 (leave-one-out) documents the predictive power and significance of the model. A more-demanding cross-validation using the five random
groups method still gave very satisfactory results with a
q2 of 0.68. It has to be admitted that
enlargement of the training set would improve the reliability of the
model. Especially interesting would be the incorporation of compounds
that cover the gap in affinities between the two existing clusters.
Unfortunately, such compounds were not available to us.
Unfortunately, the data set was too small to extract an independent
test set. Because of this, predictions were made for racemates and
isomeric mixtures. The average of these predictions was compared with
the binding data. This was possible because the difference in
affinities between the two predicted isomers amounts in all cases only
to about 0.2 pKI units. The results
obtained still have to be treated with some caution although they give
an indication of the predictability of the model.
Very good predictions could be obtained for most compounds. The
deviations of iloprost, BMY181331, BMY43675, and BMY44521 can be
considered well within experimental tolerance. Nileprost, however, does
not fit into the 3D-QSAR model. Because the most remarkable structural
variation compared with iloprost or cicaprost is the introduction of
the cyano moiety, it can be assumed that this group causes steric
repulsion in the binding pocket of the protein, which may lead to the
poor affinity measured for nileprost.
It was possible for the first time to define a common
pharmacophore for prostanoid and nonprostanoid agonists of the
prostacyclin receptor. This pharmacophore was supplemented by molecular
electrostatic potentials showing that the compounds studied possess
similar steric and electronic properties. It can be deduced from these findings that both classes of agonists show a similar binding mode.
This result was complemented by the computation of GRID fields with a
number of probes that scanned the different properties of the
compounds. A possible explanation for the higher affinity of most
prostanoid agonists probably results from their ability to form an
additional hydrogen bond to the receptor. A 3D-QSAR study quantified
the results. The final model had a q2 of 0.77, which is considered a significant correlation for the structurally
heterogeneous data set investigated in this study.
![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
Top
Abstract
Introduction
a rapid access to atomic charges.
Tetrahedron
36:
3219-3228[CrossRef].
5 kcal/mol). A, BMY45778; B, cicaprost.