C2-symmetric inhibitors of Plasmodium falciparum plasmepsin II: synthesis and theoretical predictions.

Article abstract of DOI:10.1016/S0968-0896(03)00339-0

A series of C(2)-symmetric compounds with a mannitol-based scaffold has been investigated, both theoretically and experimentally, as Plm II inhibitors. Four different stereoisomers with either benzyloxy or allyloxy P1/P1' side chains were studied. Computational ranking of the binding affinities of the eight compounds was carried out using the linear interaction energy (LIE) method relying on a complex previously determined by crystallography. Within both series of isomers the theoretical binding energies were in agreement with the enzymatic measurements, illustrating the power of the LIE method for the prediction of ligand affinities prior to synthesis. The structural models of the enzyme-inhibitor complexes obtained from the MD simulations provided a basis for interpretation of further structure-activity relationships. Hence, the affinity of a structurally similar ligand, but with a different P2/P2' substituent was examined using the same procedure. The predicted improvement in binding constant agreed well with experimental results.

Full text of DOI:10.1016/S0968-0896(03)00339-0

Bioorganic & Medicinal Chemistry 11 (2003) 3723–3733
C2-Symmetric Inhibitors of Plasmodium falciparum Plasmepsin II:
Synthesis and Theoretical Predictions
Karolina Ersmark,a Isabella Feierberg,b Sinisa Bjelic,b Johan Hulten,a
a,
Bertil Samuelsson,c Johan Aqvistb andAnders Hallberg
*
aDepartment of Medicinal Chemistry, Uppsala University, BMC, Box 574, SE-751 23 Uppsala, Sweden
bDepartment of Cell and Molecular Biology, Uppsala University, BMC, Box 596, SE-751 24 Uppsala, Sweden
cMedivir AB, Lunastigen 7, SE-141 44 Huddinge, Sweden
Received18 March 2003; accepted14 May 2003
Abstract—A series of C2-symmetric compounds with a mannitol-based scaffold has been investigated, both theoretically and
experimentally, as Plm II inhibitors. Four different stereoisomers with either benzyloxy or allyloxy P1/P10 side chains were studied.
Computational ranking of the binding affinities of the eight compounds was carried out using the linear interaction energy (LIE)
method relying on a complex previously determined by crystallography. Within both series of isomers the theoretical binding
energies were in agreement with the enzymatic measurements, illustrating the power of the LIE methodfor the prediction of ligand
affinities prior to synthesis. The structural models of the enzyme–inhibitor complexes obtained from the MD simulations provided a
basis for interpretation of further structure–activity relationships. Hence, the affinity of a structurally similar ligand, but with a
different P2/P20 substituent was examined using the same procedure. The predicted improvement in binding constant agreed well
with experimental results.
# 2003 Elsevier Ltd. All rights reserved.
Introduction
Recently, Plm II inhibitors with goodpotencies in
enzyme assays andwith fair selectivity versus the human
enzyme cathepsin D have been identified.3ꢀ14 In addi-
tion, a 3-D structure of Plm II complexedwith pep-
statin A has been determined (1SME) and utilized for
modeling studies.8ꢀ10 Unfortunately, in cases where
efforts are directed towards new classes of inhibitors,
that is those comprising scaffolds that are structurally
significantly different from those in 3-D determined
ligand–enzyme complexes, the design process is fre-
quently more troublesome.
Malaria, one of the major public health problems in the
world, is caused by parasites of the genus Plasmodium.
Of these, Plasmodium falciparum is most virulent and
responsible for the majority of all malaria deaths.
Hundreds of millions of people are afflicted and almost
3 million people die from the disease each year. Since
resistance to current antimalarial drugs is spreading
rapidly, these numbers are increasing, highlighting an
urgent needfor new rdugs acting at new parasitic
targets.1 In the intraerythrocytic phase, the parasite
uses host hemoglobin as its primary source of amino
acids, and the enzymes involved in the hemoglobin
degradation process have now emerged as promising
targets. Among these enzymes, the aspartic proteases
plasmepsin I (Plm I), plasmepsin II (Plm II), andthe
cystein protease falcipain are of particular interest to
the medicinal chemists.2
We aimedat examining new linear C2-symmetric com-
pounds, with the symmetrical 1,2-dihydroxyethylene
transition state mimetic, as potential Plm II inhibitors.
We previously usedthis type of symmetric scaffol,d
easily available from mannitol, successfully for the
inhibition of both Plm I, Plm II,13 andthe C2-symmetric
HIV-1 protease.15 However, before embarking on a
more extensive medicinal chemistry program, we were
strongly encouragedto examine computational tools
that could guide our selection of scaffolds with the
optimal stereochemical configuration for efficient inhi-
bition of Plm II.
*Corresponding author. Tel.: +46-18-471-4284; fax: +46-18-471-44-
74; e-mail: andersh@orgfarm.uu.se
0968-0896/03/$ - see front matter # 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0968-0896(03)00339-0

3724
K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
It has been shown that the S10 pocket is collapsedin the
pepstatin A/Plm II crystal structure,3 but still Ellman’s
group identified potent ligands with large P10 side
chains,8 demonstrating the difficulties in predicting
activity when one 3-D structure is available solely. For
our purpose, a methodthat allows enzyme flexibility is
needed, since it appeared that the ligands studied here
do not fit readily into the available crystal structure if
the protein is kept fixed. Some methods have been
developed that consider receptor flexibility and this
may be the most promising route to more accurate
docking.16 In this case, we choose to treat the protein
flexibility by molecular dynamics (MD) simulations.
Although this is not a general solution to flexible dock-
ing, it can be expectedto yieldgoodresults if the initial
model is reasonably reliable. Combined with simulated
annealing the scope of the MD conformational search-
ing may, however, be further expanded.17ꢀ19 The dock-
ing procedure used here is based on the similarities
between the present ligands and pepstatin A. Although
there are significant differences between the chemical
structures of pepstatin A andthe ligands examinedin
this report, the similarities suggest a reasonable posi-
tioning of the ligandside chains anda large part of the
backbone. Some guidance can also be obtained by con-
sidering structures of HIV-1 protease complexes.20
Aqvist et al. have described an approach, in which
absolute binding free energies are estimated by a linear
response approximation from thermal averaging of the
structure andinteraction energies by MD simulations,
referredto as the linear interaction energy (LIE)
method.21ꢀ23 The molecular dynamics conformational
sampling allows relaxation, or induced fit, of the ligand–
receptor complex andsolvation is treatedmicro-
scopically. The LIE methodhas in several cases, with
different enzymes, proven to be a useful tool for the
prediction of binding affinities.23ꢀ26 We have now
employed this procedure in an attempt to predict the
activities of mannitol-basedcompounsd with the
1,2-dihydroxyethylene scaffold as inhibitors of the flex-
Results
Chemistry
The synthetic route to the benzyloxy andallyloxy
stereoisomers 3a–h is outlinedin Scheme 1. The com-
pounds were prepared by a similar procedure to that
previously reportedfor the synthesis of 3a.20 Thus, oxi-
dation of l-mannonic g-lactone or d-mannitol with
aqueous nitric acidprovidedthe bislactones
1a20 and
1b.28 Subsequent benzylation using commercially avail-
able benzyl trichloroacetimidate and a catalytic amount
of trifluoromethanesulfonic acidgave the dibenzylated
enantiomers 2a and 2b.20 The allylation of the bis-
lactones 1a and 1b was carriedout using allyl tri-
chloroacetimidate.29 Trifluoromethanesulfonic acidwas
first employed as a catalyst and afforded the diallylated
enantiomers 2c and 2d, although in moderate yields. A
cleaner reaction occurredwith boron trifluoride etherate
as the catalyst, affording the diallylated bislactone 2c in
76% yield. An excess of either d- or l-valine methyl-
amide, which was used to open the bislactone rings,
delivered the final stereoisomers 3a–h in 47–86% yields
(Scheme 1). Ring opening of bislactone 2b with (1S,2R)-
1-amino-2-indanol afforded the diamide 4 in 62% yield
(Scheme 2).
Biological data
The inhibitory activities of the benzyloxy andallyloxy
isomers 3a–h, and the allyloxy indan derivative 4 were
measured experimentally in a standardized Plm II
enzyme assay8 andthe results are presentedas Ki-values
in Table 1. The role of stereochemistry provedto be
crucial. Thus, only the SRRRRS isomers 3a and 3c,
originating from l-mannonic g-lactone and l-valine
methylamide exhibited any inhibitory effect at con-
centrations up to 10 mM. A specific configuration at the
four stereogenic centers of the mannitol core (RRRR),
carrying two hydroxyl groups, of which at least one is
ible Plm II. The ability to predict the stereochemical
0
requirements andthe character of the P1/P1
sub-
stituents prior to synthesis seemedparticularly attrac-
tive. Since it has been demonstrated previously that a
b-branchedcarbon is preferredin the P2 position,
valine methylamide was selected as a suitable P2/P20
residue.5,27 The computational results were compared
to experimental data from a standardized Plm II
assay.
We herein report that the ranking of the C2-symmetric
stereoisomers obtainedby our calculations agrees with
the experimental results. Furthermore, the calculations
provide useful models for the 3-D structures of Plm II
complexed with 1,2-dihydroxyethylene-based inhibitors.
One inhibitor, comprising the optimal core configur-
ation (RRRR) and a (1S,2R)-1-amino-2-indanol P2/P20
substituent, was employedas another test compound
for the computational model. An increase of inhibitor
activity of greater than 40-foldcomparedto the corres-
ponding valine methylamide analogue was predicted
which was foundto be in agreement with the experi-
mental data.
Scheme 1. Reagents and conditions: (a) benzyl trichloroacetimidate or
.
allyl trichloroacetimidate, Tf-OH or BF3 Et2O, dioxane; (b)
ValNHMe, CH2Cl2.
Scheme 2. Reagents andconditions: (a) (1 S,2R)-1-amino-2-indanol,
CH2Cl2.

K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
Table 1. Experimental andcalculatedenergetics of inhibition
3725
CompdConfig.
Plm II
Expt
Ki (mM)
LIE
ꢀGbind (kcal/mol)
Scorea
Ligand-surrounding interactions (kcal/mol)b
ꢀ
ꢁ
p
ꢀ
ꢁ
ꢀ
ꢁ
ꢀ
vdW
ꢁ
el
lꢀs
vdW
lꢀs
el
lꢀs
Expt
LIE
V
V
V
V
lꢀs
p
w
w
3a
3b
3e
3f
3c
3d
3g
3h
4
SRRRRS
3.8
>10
>10
>10
0.52
160
.
ꢀ12.8ꢁ0.7 (ꢀ12.7) ꢀ7.4 ꢀ8.6ꢁ0.4 ꢀ75.6ꢁ0.1 ꢀ93.9ꢁ0.2 ꢀ71.8ꢁ0.8 ꢀ53.2ꢁ0.1
RRRRRR
SSSSSS
RSSSSR
SRRRRS
RRRRRR
SSSSSS
RSSSSR
-RRRR-
ꢀ10.8ꢁ1.3 (ꢀ12.0)
ꢀ11.6ꢁ0.8 (ꢀ9.7)
ꢀ10.3ꢁ0.8 (ꢀ11.0)
ꢀ6.8ꢁ0.7 (ꢀ6.5)
ꢀ4.5ꢁ1.2 (ꢀ7.4)
ꢀ4.5ꢁ0.9 (ꢀ4.6)
ꢀ4.8ꢁ0.9 (ꢀ4.8)
ꢀ8.5ꢁ0.9 (ꢀ9.4)
nd
nd
nd
ꢀ5.2ꢁ0.4 ꢀ62.5ꢁ0.6 ꢀ94.8ꢁ1.0 ꢀ69.8ꢁ0.0 ꢀ52.7ꢁ0.0
ꢀ2.9ꢁ0.4 ꢀ54.7ꢁ0.2 ꢀ96.2ꢁ1.5 ꢀ69.8ꢁ0.0 ꢀ52.7ꢁ0.0
ꢀ2.5ꢁ0.6 ꢀ55.4ꢁ0.5 ꢀ96.9ꢁ0.7 ꢀ71.8ꢁ0.8 ꢀ53.2ꢁ0.1
3
4
7.8 10
.
1.5 10
11
330
4.1
ꢀ7.4 ꢀ6.8ꢁ0.4 ꢀ73.2ꢁ0.3 ꢀ79.6ꢁ0.0 ꢀ71.9ꢁ0.8 ꢀ44.3ꢁ0.0
>10
>10
>10
nd
nd2.9
nd0.0
ꢀ4.8ꢁ0.5 ꢀ71.1ꢁ0.9 ꢀ80.8ꢁ0.3 ꢀ77.0ꢁ0.1 ꢀ43.6ꢁ0.3
ꢁ0.7 ꢀ47.0ꢁ1.5 ꢀ82.2ꢁ0.4 ꢀ77.0ꢁ0.1 ꢀ43.6ꢁ0.3
ꢁ0.7 ꢀ51.1ꢁ1.0 ꢀ82.4ꢁ0.3 ꢀ71.9ꢁ0.8 ꢀ44.3ꢁ0.0
8
.
1.3 10
5
.
9.4 10
0.24
0.096
ꢀ9.6 ꢀ9.1ꢁ0.4 ꢀ67.2ꢁ0.1 ꢀ85.9ꢁ0.1 ꢀ64.5ꢁ0.9 ꢀ40.6ꢁ0.1
aThe calculatedbinding affinity using an empirical scoring function, 34 where the average value of 100 energy minimizedsnapshots andtheir standard
deviation is given. In parentheses is the score of the initial minimized structures of the complexes after MD equilibration.
ꢀ
ꢁ
ꢀ
ꢁ
el
l-s
vdW
l-s
bThe calculatedaverage polar
V
andnonpolar
V
interactions between the ligandandits surrounding. The subscripts p andw describe the
ligandin complex with the protein andfree in water, respectively.
postulatedto be important for cooridnation to the
aspartic acidresidues Asp34 andAsp214, is essential.
strongest overall binding affinities of ꢀ8.6ꢁ0.4 and
ꢀ6.8ꢁ0.4 kcal/mol, respectively. It is noteworthy that
the simulations predict the absolute binding free ener-
gies quite accurately, without any reparametrization of
the LIE equation as usedin ref 31. From Table 1 and
the LIE equation (eq 1) it is evident that the nonpolar
interactions (obtainedfrom the difference in interaction
energy between the solvatedprotein–ligandcomplex
andthe free ligandin water) are favorable for the bind-
ing of all eight ligands with a contribution to the binding
free energy of 6.4–7.9 kcal/mol. In absolute terms, all four
stereoisomers of both compounds have similar nonpolar
interactions with the protein, the least favorable one
being, in fact, with the SRRRRS isomers. The electrostatic
interactions display at least a 2 kcal/mol repulsive con-
tribution to the binding affinity for all compounds except
3a and 3c. Thus, as apparent from Table 1, the hyd ro-
phobic interactions seem to give the largest contributions
to the absolute affinity, while electrostatic interactions
(mainly H-bonding) are important for the specificities. In
addition, it is notable that the more hydrophobic com-
pounds 3a, b, e and f contribute with only ꢂ1 kcal/mol
extra to the nonpolar term of the binding free energy than
the smaller allyl compounds 3c, d, g and h.
30
Furthermore, despite the RRRR configuration of the
central part of the molecules, the compounds 3b and 3d
with d-valine rather than l-valine in P2/P20 were devoid
of any activity. A comparison between benzyloxy and
allyloxy side chains in the P1/P10 positions, revealedno
significant difference in activity, Ki=3.8 and4.1 mM
respectively (c.f. 3a and 3c).
A replacement of valine methylamide by (1S,2R)-1-
amino-2-indanol with the RRRR configuration of the
scaffoldimprovedthe inhibitory activity 40 times (c.f.
and 3c). Unfortunately, as deduced from preliminary
experiments the most potent inhibitor in the series,
compound 4, exhibitedno activity at 5 mM against P.
falciparum (clone 3D7) blood-stage parasites.
4
Molecular dynamics simulations and binding affinity
calculations
Since no X-ray structures of the complexes of Plm II
with 3a–h are available, a theoretical prediction of the
binding affinities and binding conformations of the
complexes was performedusing computer simulations.
Automated docking of the flexible ligands into the rigid,
partially collapsedbinding site of the X-ray structure
(1SME) of Plm II did not yield useful results. The rea-
son for this is the steric hindrance imposed by the rigid
receptor model. Instead, we employed manual docking
andsubsequent MD relaxation of the protein–ligand
complex andthe average MD structures of the com-
plexes predict that the protein undergoes very little
structural changes upon binding of the ligands, except
for rotation of certain side chains in order to accom-
modate the bulky P10 groups. Thus, the position of the
Met75-Val82 loop was not disrupted in the simulations.
The binding free energies of 3a–h were estimatedusing
the linear interaction energy (LIE) methodin combi-
nation with molecular dynamics simulations.22 In Table
1, the calculated average ligand-surrounding interaction
energies are shown together with the calculatedand
experimental inhibition constants. The calculations pre-
dicted that the SRRRRS isomers 3a and 3c have the
It is necessary that the conformational space is suffi-
ciently well sampledin the MD simulations, since the
convergence of the calculatedbinidng free energies
clearly depends on the extent of conformational sam-
pling. This can be problematic even for moderately
large ligands if they carry net charges, so that high
absolute values of the interaction energies are involved.24
Although the present inhibitors are uncharged, to deal
with this problem, MD equilibration runs of up to 1.35
ns were performed for the ligands in water before data
was collectedandthe averages were calculatedon at
least 1.2 ns trajectories after equilibration in order to
ensure that the conformational space hadindeedbeen
adequately sampled. For the simulations of the ligands
in complex with protein, data collected over 300–600 ps
cover a significant amount of the conformational space
aroundeach equilibratedstarting conformation, which
was assessedby comparing the average energies for the
first andsecondhalf of each trajectory.

3726
K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
group. The P10 side chain is also in close contact with
side chain atoms of Val78, Tyr192, Ile212, Thr298 and
Ile300, andthe P1 0 phenyl group of 3a also has stacking
interactions with the aromatic ring of Tyr192. The P20
valine group is positionedvery similarly to the Ala
group on pepstatin A, pointing into S20. The P1 hydro-
phobic group is positionedin the S1 pocket andturns
the phenyl (allyl in 3c) group towards S3 in the con-
tinuous S1–S3 cleft, while the P2 valine group points
into S2, albeit slightly displaced in the S3–S4 direction
comparedto the P2 group of pepstatin A. The protein–
ligandbackbone H-bondpattern obtainedfrom the
simulations is shown in Figure 1 andis similar to that of
the initial X-ray structure for the P0 part of the ligand,
where the 3a andpepstatin A backbones are virtually 0
superimposedon each other. A slight rotation of the P
methylamide group and the loss of H-bond interaction
with the Asn76 carbonyl oxygen in the simulatedstruc-
ture is the largest difference between the simulated and
the experimental structure. Although the amide group
has rotated, the H-bond between the hydroxyl group of
Figure 1. Stereo view of the predicted (MD average) binding con-
formation of inhibitor 3a (green) in complex with Plm II (yellow),
superimposedon the X-ray structure of pepstatin A (purple) com-
plexedwith Plm II (light blue). 3 ‘Wat’ denotes the water molecule that
was found to interact with the P hydroxyl group as well as the side
chain of Thr217 in the MD simulations.
To demonstrate the predicted binding conformation of
the most potent methylamide inhibitor 3a, in Figure 1
an average MD structure from 50 ps simulation is
shown superimposedon the X-ray structure of Plm II in
complex with pepstatin A. For clarity reasons, all water
molecules not in direct contact with the ligand and
many residues that have nonpolar interactions with the
ligandare omittedfrom the picture. Figure 2 shows a
scheme of the interactions in the active site of the same
complex together with a similar schematic picture of
compound 4. The compounds 3a and 3c display the
same binding conformation and the protein structure is,
in both cases, very similar to the X-ray structure. In
order to make space for the benzyloxy P10 group, which
was docked into the partially collapsed S10 pocket of the
X-ray structure, the side chain of Phe294 is rotated
towards Leu292, which in turn rotates towards Ile290.
The Phe294 rotation is the only major side chain dis-
0
Tyr192 andthe P2 carbonyl oxygen is maintainedas in
the X-ray structure of the Plm II–pepstatin A complex.
0
Furthermore, in the simulatedstructure the P2 amide
nitrogen interacts with the carbonyl of Gly36 andthe
nitrogen of the loop residue Val78 donates its H-bond
to the P10 carbonyl. The P10 hydroxyl is oriented in the
same fashion as in the crystal structure: it is located
between the carboxylates of Asp34 andAsp214, and
donates its H-bondto the negatively chargedAsp214.
Since the direction of the amide bonds is reversed for
the P part of the ligand, the H-bonding interaction pat-
tern at these sites differs from that of the structure of the
pepstatin A complex. Here, the P1 hydroxyl donates its
H-bondto Asp214 andthis hydroxyl also accepts an
0
placement observedwhen accommodating the bulky P1
Figure 2. Schematic picture showing the predicted (MD average) interactions of inhibitor 3a and 4 with the active site.

K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
3727
H-bondfrom a water molecule. This water molecule
donates an H-bond to the sidechain of Thr217 and
accepts one from a secondwater molecule in the binding
site (not shown). The MD simulations further suggest
that the P1 carbonyl oxygen accepts an H-bondfrom
Ser79 N, andthe P2 nitrogen andcarbonyl oxygen make
H-bonds with Gly216 O and Ser218 N, respectively.
other hand, the P2 group is more flexible, which is in
contrast to the simulation with 3c, where the P2 group
andmethylamide remain essentially in the same place
during the whole simulation. In the simulation with 4
the P2 indan ring hydroxyl group interacts with solvent
andalso forms an intramolecular H-bond, similar to the
P20 group. The indan rings exhibit nonpolar interactions
with the side chains of Thr114, Ile290, Ser79 and the P1
group, while one face of the indan moiety is exposed to
the solvent.
Since the present ligands exhibit only moderate inhibi-
tion of Plm II, we aimedat modifying them to create
more potent inhibitors. 2-Indanol has been used pre-
viously as an effective P2 substituent in HIV-1 protease
inhibitors andthis substituent was therefore also con-
sidered here.20,32 Thus, we decided to predict the binding
As a comparison to the LIE results the binding affinities
were also estimatedusing an empirical scoring func-
tion.33 This function takes into account H-bonding and
lipophilic interactions, as well as the loss of rotational
degrees of freedom upon ligand binding (see the Meth-
ods section). While scoring functions in general are
intended for rapid binding affinity estimates of a single
conformation, we decided also to examine the effects of
averaging over several snapshots from the MD trajec-
tories since this has been shown previously to improve
the results.34 Hence, the initial equilibratedcomplexes
of Plm II with 3a–h and 4 were first minimizedand
scored. Then, 100 snapshots from the MD trajectory of
each complex were minimizedandscoredandthe results
were averaged. In Table 1, the average values with
standard deviations are shown together with the single
snapshot scores. Within the isomer series of the allyloxy
compounds, a correct affinity ranking was obtained
when averaging over snapshots was performed, while
scoring only the initial minimizedstructures didnot
yieldthe correct result. The compound 3c was thus
predicted to be the most active inhibitor with a score
within 0.6 kcal/mol from the experimental value, using
the averaging approach. In addition, the indan com-
pound 4 is reasonably well predicted by the scoring
function. The affinities within the benzyloxy series were
also correctly ranked. However, all of them have pre-
dicted affinities to Plm II that are too great when com-
paredto the experimental values andthe average affinity
of 3a is overpredicted by 5.4 kcal/mol. The over-
estimatedhigh affinity for the benzyloxy series can be
attributedto the lipophilic term of the scoring function,
which clearly overestimates the hydrophobic binding
contribution (predicted to about ꢀ10 kcal/mol). This
may indicate a more general problem with scoring
functions that have a built-in size-dependency of the
hydrophobic term through surface area or similar size
measures.
constant for 4 where the terminal methylamide had been
0
removedandthe P2/P2
valines hadbeen substituted
with (1S,2R)-1-amino-2-indanol. The result, shown in
Table 1, is a 2.3 kcal/mol more negative binding free
energy (or a 47 times stronger inhibition constant) than
for 3c. An experimental measurement of the potency of
4 gave a corresponding increase in the binding affinity
of 2.2 kcal/mol (or 43 times greater inhibition) relative
to 3c, which essentially confirms the computational
result. The simulatedstructure of the complex with
4
(Fig. 3) reveals a binding mode similar to that of 3a and
3c, with the central parts (both hydroxyls, P1/P10 ally-
loxy chains, P1/P10 carbonyls) in essentially the same
position as in the valine residue compounds. In addi-
tion, the water molecule that donates H-bonds to the P1
andThr217 hydroxyl oxygen atoms in the simulations
of the complexes with 3a and 3c was also foundin the
complex with 4. The resulting hydrogen bonding pat-
terns of the P2/P20 groups differ, however, from the case
with 3a and 3c. Figure 3 shows the H-bonds that are
present over at least ꢂ50% of the MD simulation
superimposedon the X-ray structure of Plm II in com-
plex with pepstatin A. The P20 indan ring system in 4 is
situatedalong the position of the 3c backbone, with its
five-memberedring at the valine side-chain site andits
aromatic ring along the methylamide group of 3c. The
indan hydroxyl alternates b0etween donating an intra-
molecular H-bondto the P1 carbonyl oxygen andone
to the Asn76 carbonyl oxygen. Occasionally, a water
molecule and the Ser37 side chain also donate H-bonds
to this hydroxyl group. The position of the P20 indan
ring system remains the same throughout the whole
simulation, as the corresponding part of 3c. On the
Discussion
The ranking of the predicted binding affinities, in addi-
tion to the absolute values, agree well with the experi-
mental finding that only the SRRRRS isomers 3a and 3c
display inhibitory effects in the Plm II assay. In addi-
tion, among these isomers the same stereo configuration
(SRRRRS) is the only one that produces active inhibi-
tors in a standardized HIV-1 protease assay.35,36 An
X-ray structure of 3c with Val substitutedfor Ile in
complex with HIV-1 protease shows that both hydroxyl
groups point towards one of the catalytic aspartates.20
Figure 3. Stereo view of the predicted (MD average) binding con-
formation of inhibitor 4 (green) in complex with Plm II (yellow), super-
imposedon the X-ray structure of pepstatin A (purple) complexedwith
Plm II (light blue).3 ‘Wat’ denotes the water molecule that was found to
interact with the P hydroxyl as well as the side chain of Thr217 in the
MD simulations.

3728
K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
Also, in the X-ray structure 1SME, the configurations
of the pepstatin A P10 hydroxyl and P20 Ala correspond
to the SRRRRS stereoisomers. The LIE model for this
class of ligands uses a=0.181, b=0.33 and g=0 and
was parameterizedin ref 23. It provides an excellent
estimation of the absolute binding free energies of the
methylamide compounds and also captures the effect of
the P2/P20 modified compound 4. Overall, this LIE
model has indeed proven successful in several other
cases with the existing parametrization.24,26,37,38 The
simulatedstructure of Plm II in complex with the
ligands 3a–h and 4 remains essentially as in the X-ray
structure, although movement of some side chains was
Apart from the LIE method, there are several other
techniques available for calculation of protein–ligand
binding affinities. The free energy perturbation (FEP)
approach41 has often been usedas a tool to calculate
differential binding affinities between chemically similar
ligands. The method is usually reliable for calculating
the effects on binding affinity resulting from small per-
turbations, for example the addition of a functional
group or change of atom type. However, in the present
case FEP wouldnot be a realistic choice since the size of
the perturbations involved, that is the addition of a
phenyl or indan ring, would cause severe convergence
problems. Another simplifiedmethodfor calculating
protein–ligandbinidng affinities is the MM-PBSA
approach,42 which has been successful in predicting
absolute binding free energies in several cases.43 An
important difference between the LIE method and MM-
PBSA is that the latter employs a continuum solvent
description in the final binding energy calculation.
While this may improve the convergence, the energetic
details of solvation are lost. For instance, the water
molecule that was foundto bridge between the ligand
andPlm II wouldnot give any contribution. Further-
more, the usual implementation of MM-PBSA assumes
that the conformations exploredby the ligandare the
same in the boundandfree states andonly the complex
is simulated. This assumption does not appear warranted
in the present case since our ligands are considerably
more conformationally restrictedin the complexes than
in solution. Empirical scoring functions, such as that of
Bohm44 and Eldridge et al.,33 may be usedfor estima-
tion of binding free energies on docked complexes. Here,
we employedEldridge’s function for scoring snaphots
of the different complexes. The results correctly rank the
affinities within the isomer series, provided that con-
formational averaging is carriedout, andthe affinity of
4 is also reasonably well predicted. The scoring func-
tion, however, largely overpredicts the affinities of all
benzyloxy ligands, especially those having no experi-
mentally detectable affinity. The large lipophilic term
that dominates the scores of the benzyloxy compounds
is the main source of the systematic over prediction.
Regardless of which method is used for binding energy
calculations it seems that conformational averaging
tends to improve the result.34 This is, of course, parti-
cularly true for force-fieldbasedscoring approaches,
such as LIE, since the energy fluctuations due to ther-
mal motion can be substantial. While such methods are
more time-consuming than empirical scoring functions
their predictive power may still justify the extra effort,
as the present calculations attest.
0
observedin the S1 pocket upon accommodation of the
P10 group. In addition, a water molecule not present in
the X-ray structure of the Plm II/pepstatin A complex
was found to bridge between the ligand hydroxyl group
anda threonine residue in the simulatedcomplexes of
3a, 3c, and 4 as well as in those with 3d and 3e. It is
noteworthy that the water molecule is not introduced
into the binding site in the initial solvation of the com-
plex, but enters during MD equilibration and then
remains at its location. In the Plm II/pepstatin A crystal
structure, the corresponding position is occupied by the
hydrophobic P2 side chain. It may be of interest to note
that the intramolecular H-bonds found in compound 4
are neither observedin HIV-1 protease complexes with
similar ligands nor with the existing anti-HIV drug
Indinavir.20,39
In a paper by Haque et al.8 a crystal structure is descri-
bedof Plm II in complex with a hydrox0yethylamine-
basedinhibitor that contains a bulky P1 group. In this
structure, which has very recently been made available
under the PDB code 1LF3,40 the authors observe that
the distance between the side chain of Val78 and the
Phe294 phenyl is 3 A longer than in the X-ray structure
with pepstatin A. It is the backbones of the Met75-
Val82 andLeu292-Pro197 loops that move when incor-
0
porating the ligandwith the bulky P1 group. In the
present study, however, the bulky P10 groups did not
induce any substantial conformational changes of the
loops, but the S10 pocket was expanded to accom-
modate the P10 group mainly by movement of the
Phe294 side chain. Other very recently published struc-
tures are a Plm II/pepstatin A complex, 1M43 solvedat
2.4 A,40 andtwo complexes of Met205Ser mutants of
14
Plm II with inhibitors (1LEE and1LF2).
Also a
structure of uncomplexedPlm II has been solved(1LF4)
at 1.90 A resolution has become available.40
Since the inhibitors 3a–h cause only moderate inhibition
of Plm II, we investigatedthe effect on the binding
0
constant of a scaffoldmodification where the P2/P2
Conclusion
valines of inhibitor 3a were substitutedwith (1 S,2R)-1-
amino-2-indanol and the terminal methylamide groups
were removed. This change introduced large, hydro-
phobic groups andreducedthe peptide-like properties
as well as the number of torsional degrees of freedom.
The computational result predicts that such a modifi-
cation yields a 47 times more potent inhibitor than the
reference compound 3c, an effect that was also observed
experimentally.
In summary, C2-symmetric compounds encompassing a
mannitol-basedscaffold, previously usedsuccessfully in
HIV-1 protease inhibitors, have been theoretically and
experimentally examinedas core structures in Plm II
inhibitors. Four different stereoisomers with either ben-
zyloxy or allyloxy P1/P10 side chains were studied.
Computational ranking of the binding affinities of the
eight compounds was performed and the isomers were

K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
3729
synthesizedandevaluatedin a Plm II enzyme assay.
Since the ligandbinding site in the X-ray structure of
the protein in complex with pepstatin A is partially col-
lapsed, manual docking of the ligands with subsequent
relaxation of the complex was employedrather than
automated docking into a rigid protein. Binding affi-
nities were estimatedby the linear interaction energy
(LIE) method. Within both series of isomers the theo-
retical binding energies were in agreement with the
enzymatic measurements, illustrating the usefulness of
the LIE methodfor the prediction of ligandaffinities
prior to synthesis. Furthermore, the structural models
of the enzyme–inhibitor complexes obtainedfrom the
MD simulations provide a framework for the inter-
pretation of further structure–activity relationships.
Hence, the affinity of a similar ligandwith a different
P2/P20 substituent was estimatedusing the same pro-
cedure, and the predicted improvement in binding con-
stant agrees well with experimental results. In the
future, we aim to further improve this diol type of Plm
II inhibitors.
AA0XX0, 2H), 4.80 (d, J=11.75 Hz, 2H), 4.73 (d,
J=11.75 Hz, 2H); 13C NMR (DMSO-d6) d 171.8, 136.9,
128.4, 128.1, 128.0, 74.8, 74.4, 72.1. Anal. calcdfor
C20H18O6: C, 67.79; H, 5.12. Found: C, 67.4; H, 5.4.
2,5-Di-O-allyl-L -mannaro-1,4:3,6-di-ꢀ-lactone (2b).
l-1,4:3,6-Mannarodilactone 1a (1.1 g, 6.0 mmol) was
dissolved in dry dioxane (50 mL) under a nitrogen
atmosphere. The mixture was heateduntil the dilactone
was completely dissolved, and the solution was then
allowedto attain room temperature. To the stirred
solution was added allyl trichloroacetimidate (3.7 g, 18
.
mmol) anda catalytic amount of BF 3 Et2O (200 mL).
The solution was stirredat room temperature for 3 h.
After filtration through a silica padthe filtrate was
concentrated and dried under vacuum overnight. The
crude product was washed with diethyl ether (3ꢃ50 mL)
to give pure product 2b (1.2 g, 76%) as a white solid:
ꢄ
1
[a]2D2 ꢀ213.15 (c 1.034,CH3CN); H NMR (DMSO-d6)
d 5.93 (dddd, J=2.83, 5.63, 10.39, 17.25 Hz, 2H), 5.36
(ddd, J=2.01, 3.98, 17.25 Hz, 2H), 5.28 (AA0 part of
AA0XX0,02H), 5.25 (ddd, 0J=1.74, 3.48, 10.39 Hz, 2H),
4.81 (XX part of AA0XX , 2H), 4.31–3.15 (m, 4H); 13C
NMR (DMSO-d6) d 171.8, 133.9, 118.1, 74.8, 74.4, 71.2.
Experimental
Chemistry
.
Anal. calcdfor C 12H14O6 1/4H2O: C, 55.70; H, 5.65.
Found: C, 55.7; H, 5.5.
General information. 1H and 13C NMR spectra were
recorded on a Jeol JNM-EX 270 spectrometer at 270.2
and67.8 MHz, respectively, or on a JEOL JNM-EX400
spectrometer at 399.8 and100.5 MHz, respectively.
Chemical shifts are reportedas d values (ppm) indirectly
referencedto TMS by the solvent residual signal. Opti-
cal rotations were obtainedon a Perkin-Elmer 241
polarimeter. Specific rotations (aD) are reportedin deg/
dm, and the concentration (c) is given in g/100 mL in
the specifiedsolvent. Elemental analysis was performed
by Mikro Kemi AB, Uppsala, Sweden. Flash column
chromatography was performedon Merck silica gel 60,
0.04–0.063 mm. Thin-layer chromatography was per-
formedusing aluminium sheets precoatedwith silica gel
60 F254 (0.2 mm; E. Merck) andvisualizedwith UV
light, permanganate or ninhydrin. RP-FPLC was per-
formedon a PepRPC 15 mm HR 16/10 column (Phar-
macia Biotech AB) at a flow rate of 3.5 mL/min. SP-
HPLC was performedon a Dynamax 60-A 21.4 ꢃ250
mm, 8 mm silica gel column (Rainin Instrument co. inc)
at a flow rate of 20 mL/min.
2,5-Di-O-allyl-D -mannaro-1,4:3,6-di-ꢀ-lactone (2d).
d-1,4:3,6-Mannarodilactone 1b (101 mg, 0.58 mmol)
was dissolved in dry dioxane (6 mL) under a nitrogen
atmosphere. The mixture was heateduntil the dilactone
was completely dissolved, and the solution was then
allowedto attain room temperature. To the stirred
solution was added allyl trichloroacetimidate (444 mg,
2.3 mmol) anda catalytic amount of trifluoro-
methanesulfonic acid(20 mL). The solution was stirred
at room temperature for 7 h. After filtration through a
padof Na CO3 (2 cm) on silica (1 cm) the filtrate was
2
concentrated and dried under vacuum overnight. The
crude product was purified on RP-FPLC (15 min gra-
dient of 30–45% CH3CN in 0.1% aqueous TFA) to give
pure product 2d (76 mg, 51%) as a white solid: [a]D22
+213.8ꢄ (c 1.034, CH3CN); 1H NMR (DMSO-d6) d
5.93 (dddd, J=2.85, 5.69, 10.31, 17.24, 2H), 5.36 (ddd,
J=1.98, 3.92, 17.24 Hz, 2H), 5.28 (AA0 part of AA0XX0,
2H), 5.25 (ddd, J=1.73, 3.51, 10.31 Hz, 2H), 5.81 (XX0
part of AA0XX0, 2H), 4.31–4.15 (m, 4H); 13C NMR
(DMSO-d6) d 171.8, 133.9, 118.1, 74.8, 74.4, 71.2. Anal.
calcdfor C 12H14O6: C, 56.69; H, 5.55. Found: C, 56.3;
H, 5.7.
2,5-Di-O-benzyl-D-mannaro-1,4:3,6-di-ꢀ-lactone (2c).
d-1,4:3,6-Mannarodilactone 1b (303 mg, 1.7 mmol) was
dissolved in dry dioxane (40 mL) under a nitrogen
atmosphere. To the stirred solution was added benzyl
trichloroacetimidate (1.4 g, 5.3 mmol) and a catalytic
amount of trifluoromethanesulfonic acid(250 mL). The
solution was stirredat room temperature for 5 h. After
General procedure for preparation of the compounds 3b
and 3e–f
Method I. Bislactone 2a or 2c was dissolved in dry
CH2Cl2 (2 mL) under a nitrogen atmosphere, and the
solution was cooledto 0 ꢄC. Four equivalents of d- or
l-valine methylamide in dry CH2Cl2 (2 mL) was added
andthe reaction mixture was allowedto attain room
temperature. Stirring was continuedfor 9–21 h at room
temperature andfor another 11–15 h at reflux. Puri-
fication by flash chromatography or SP-HPLC gave the
products 3b and 3e–f.
filtration through a padof K CO3 (2 cm) on silica (2
2
cm) the filtrate was concentrated and dried under
vacuum overnight. The crude product was washed with
diethyl ether (2ꢃ50 mL) to give pure product 2c (420
mg, 68%) as white crystals. [a]2D3 +129.7ꢄ (c 1.04,
1
CH2Cl2); H NMR (DMSO-d6) d 7.42–7.31 (m, 10H),
5.27 (AA0 part of AA0XX0, 2H), 4.88 (XX0 part of

3730
K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
(2R,3R,4R,5R)-N1,N6-Bis[(1R)-2-methyl-1-(methylcarb-
amoyl)propyl]- 2,5- di(benzyloxy) -3,4-dihydroxyhexane-
1,6-diamide (3b). The title compoundwas prepared
according to Method I, using 57 mg (0.16 mmol) of
bislactone 2a (stirring 9 h at room temperature and15 h
at reflux). Purification by flash chromatography
(CH2Cl2/CH3OH, 20:1–10:1) gave the title compound
3b (52 mg, 53%) as a white solid: [a]2D2 +48.1ꢄ (c 0.53,
(2R,3R,4R,5R)-N1,N6-Bis[(1S)-2-methyl-1-(methylcarb-
amoyl)propyl]-2,5-di(allyloxy)-3,4-dihydroxyhexane-1,6-
diamide (3c). The title compoundwas preparedaccord-
ing to MethodII, using 179 mg (0.71 mmol) of bislac-
tone 2b andtotally 9 mL dry CH 2Cl2 (reflux 21 h). The
crude product was suspended in H2O (5 mL), filtered,
dissolved in CHCl3/CH3OH, 9:1 (20 mL) andfiltered
through a padof silica (1.5 cm). The padwas further
washedwith CHCl 3/MeOH, 9:1 (30 mL). Concentration
of the filtrate under reduced pressure gave product 3c
(243 mg, 67%) as a white solid: [a]2D2 +8.4ꢄ (c 0.44,
1
CHCl3/CH3OH, 10:1); H NMR (CDCl3/CD3OD, 4:2)
d 7.36–7.22 (m, 10H), 4.62 (d, J=11.13 Hz, 2H), 4.49
(d, J=11.13 Hz, 2H), 4.23 (d, J=6.06 Hz, 2H), 4.13
(app s, 4H), 2.71 (s, 6H), 2.14 (dh, J=6.06, 6.89 Hz,
2H), 0.92 (d, J=6.89 Hz, 6H), 0.86 (d, J=6.89 Hz, 6H);
13C NMR (CDCl3/CD3OD, 4:2) d 172.9, 172.8, 137.4,
129.1, 128.0, 128.7, 80.0, 73.4, 71.8, 58.9, 31.1, 26.3,
1
CHCl3/CH3OH, 10:1); H NMR (CDCl3/CD3OD, 4:2)
d 5.91 (dddd, J=2.80, 5.61, 10.35, 17.24 Hz, 2H), 5.31
(ddd, J=1.94, 3.67, 17.24 Hz, 2H), 5.21 (ddd, J=1.32,
2.80, 10.35 Hz, 2H), 4.22 (d, J=5.40 Hz, 2H), 4.18–4.01
(m, 6H), 3.95 (XX0 part of AA0XX0, 2H), 2.71 (s, 6H), 2.27
(dh, J=5.40, 6.89 Hz, 2H), 0.94 (d, J=6.89 Hz, 6H), 0.87
(d, J=6.89 Hz, 6H); 13C NMR (CDCl3/CD3OD, 4:2) d
172.7, 172.5, 134.0, 118.6, 80.5, 72.4, 72.3, 58.7, 30.4,
26.3,19.8, 17.6. Anal. calcdfor C 24H42N4O8: C, 56.01; H,
8.23; N, 10.89. Found: C, 56.0; H, 8.3; N, 10.7.
.
19.6, 17.9. Anal. calcdfor C 32H46N4O8 1/2H2O: C,
61.62; H, 7.60; N, 8.98. Found: C, 61.3; H, 7.4; N, 8.7.
(2S,3S,4S,5S)-N1,N6-Bis[(1S)-2-methyl-1-(methylcarb-
amoyl)propyl] - 2,5- di(benzyloxy) -3,4-dihydroxyhexane-
1,6-diamide (3e). The title compoundwas prepared
according to Method I, using 51 mg (0.14 mmol) of
bislactone 2c (stirring 10 h at room temperature and13
h at reflux). Purification by flash chromatography
(EtOAc/CH3OH, 50:1–10:1) provided product 3e (49
mg, 55%) as a white solid: [a]2D2 ꢀ47.4ꢄ (c 0.53, CHCl3/
(2R,3R,4R,5R)-N1,N6-Bis[(1R)-2-methyl-1-(methylcarb-
amoyl)propyl]-2,5-di(allyloxy)-3,4-dihydroxyhexane-1,6-
diamide (3d). The title compoundwas preparedaccord-
ing to MethodII, using 39 mg (0.71 mmol) of bislactone
2b andtotally 4 mL dry CH 2Cl2 (reflux 30 h). Purifica-
tion by SP-HPLC (CHCl3) gave ꢄproduct 3d (68 mg,
1
CH3OH, 10:1); H NMR (CDCl3/CD3OD, 4:2) d 7.36–
7.23 (m, 10H), 4.62 (d, J=11.22 Hz, 2H), 4.49 (d,
J=11.22 Hz, 2H), 4.23 (d, J=6.02 Hz, 2H), 4.01 (app s,
4H), 2.71 (s, 6H), 2.15 (dh, J=6.02, 6.89 Hz, 2H), 0.92
(d, J=6.89 Hz, 6H), 0.87 (d, J=6.89 Hz, 6H); 13C
NMR (CDCl3/CD3OD, 4:2) d 172.9, 172.7, 137.3,
129.1, 128.9, 128.7, 80.0, 73.4, 71.7, 58.9, 31.1, 26.3,
1
86%) as a white solid: [a]2D1 +73.1 (c 0.50, CHCl3); H
NMR (CDCL3/CD3OD, 4:2) d 5.89 (dddd, J=2.89,
5.86, 10.35, 17.24 Hz, 2H), 5.28 (ddd, J=1.57, 3.13,
17.24 Hz, 2H), 5.19 (ddd, J=1.32, 2.80, 10.35 Hz, 2H),
4.24 (d, J=6.02 Hz, 2H), 4.14–3.97 (m, 4H), 3.95 (app s,
4H), 2.72 (s, 6H), 2.16 (dh, J=6.02, 6.85 Hz, 2H), 0.93
(d, J=6.85 Hz, 6H), 0.88 (d, J=6.85 Hz, 6H); 13C
NMR (CDCL3/CD3OD, 4:2) d 172.9, 172.7, 134.0,
118.9, 79.8, 72.4, 71.6, 58.8, 31.0, 26.2, 19.6, 17.8. Anal.
calcdfor C 24H42N4O8: C, 56.01; H, 8.23; N, 10.89.
Found: C, 55.7; H, 8.1; N, 10.8.
.
19.6, 17.9. Anal. calcdfor C 32H46N4O8 11/2H2O: C,
59.89; H, 7.70; N, 8.73. Found: C, 59.6; H, 7.3; N, 8.4.
(2S,3S,4S,5S)-N1,N6-Bis[(1R)-2-methyl-1-(methylcarb-
amoyl)propyl] - 2,5- di(benzyloxy) -3,4-dihydroxyhexane-
1,6-diamide (3f). The title compoundwas prepared
according to Method I, using 51 mg (0.15 mmol) of
bislactone 2c (stirring 21 h at room temperature and11
h at reflux). Purification by SP-HPLC (CHCl3 to
CHCl3/CH3OH, 50:1) gave product 3f (42 mg, 47%) as
a white solid: [a]D22 +3.9ꢄ (c 0.58, CHCl3/CH3OH, 10:1);
1H NMR (CDCl3/CD3OD, 4:2) d 7.37–7.24 (m, 10H),
4.65 (d, J=11.55 Hz, 2H), 4.58 (d, J=11.55 Hz, 2H), 4.21
(d, J=5.57 Hz, 2H), 4.14 (AA0 part of AA0XX0, 2H), 4.05
(XX0 part of AA0XX0, 2H), 2.69 (s, 6H), 2.24 (dh, J=5.57,
6.85 Hz, 2H), 0.90 (d, J=6.85 Hz, 6H), 0.82 (d, J=6.85
Hz, 6H); 13C NMR (CDCl3/CD3OD, 4:2) d 172.6, 172.4,
137.4, 129.1, 128.8, 128.6, 80.7, 73.4, 72.3, 58.7, 30.5, 26.3,
(2S,3S,4S,5S)-N1,N6-Bis[(1S)-2-methyl-1-(methylcarb-
amoyl)propyl]-2,5-di(allyloxy)-3,4-dihydroxyhexane-1,6-
diamide (3g). The title compoundwas preparedaccord-
ing to MethodII, using 17 mg (0.067 mmol) of bislac-
tone 2dandtotally 4 mL dry CH
2Cl2 (reflux 50 h).
Purification by SP-HPLC (CHCl3 to CHCl3/CH3OH,
50:1) gave product 3g (26 mg, 75%) as a white solid:
[a]2D1 ꢀ73.1ꢄ (c 0.50, CHCl3); 1H NMR (CDCl3/
CD3OD, 4:2) d 5.89 (dddd, J=2.89, 5.90, 10.35, 17.24
Hz, 2H), 5.28 (ddd, J=1.53, 3.09, 17.24 Hz, 2H), 5.19
(ddd, J=1.32, 2.80, 10.35 Hz, 2H), 4.24 (d, J=6.02 Hz,
2H), 4.14–3.98 (m, 4H), 3.96 (app s, 4H), 2.72 (s, 6H),
2.16 (dh, J=6.06, 6.85 Hz, 2H), 0.93 (d, J=6.85 Hz,
6H), 0.89 (d, J=6.85 Hz, 6H); 13C NMR (CDCl3/
CD3OD, 4:2) d 173.0, 172.8, 134.1, 118.9, 79.8, 72.4,
71.7, 58.8, 31.1, 26.3, 19.6, 17.8. Anal. calcdfor
.
19.8, 17.7. Anal. calcdfor C 32H46N4O8 1/2H2O: C, 61.62;
H, 7.60; N, 8.98. Found: C, 61.3; H, 7.4; N, 8.7.
General procedure for preparation of the compounds 3c–d,
3g–h
.
C24H42N4O8 1/4H2O: C, 55.53; H, 8.25; N, 10.79.
Found: C, 55.3; H, 8.1; N, 10.5.
Method II. Bislactone 2b or 2d was dissolved in dry
CH2Cl2 under a nitrogen atmosphere. Four equivalents
of d- or l-valine methylamide in dry CH2Cl2 was added
andthe reaction mixture was stirredat reflux for 21–50
h. Concentration under reduced pressure and different
workup procedures gave the pure products 3c–d, 3g–h.
(2S,3S,4S,5S)-N1,N6-Bis[(1R)-2-methyl-1-(methylcarb-
amoyl)propyl]-2,5-di(allyloxy)-3,4-dihydroxyhexane-1,6-
diamide (3h). The title compoundwas preparedaccord-
ing to MethodII, using 47 mg (0.19 mmol) of bislactone

K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
3731
2d andtotally 4 mL dry CH 2Cl2 (reflux 24 h). Purifica-
tion by SP-HPLC (CHCl3 to CHCl3/CH3OH, 50:1)
gave product 3h (45 mg, 47%) as a white solid: [a]D22
ꢀ8.58ꢄ (c 0.44, CHCl3/CH3OH, 10:1); 1H NMR
(CDCl3/CD3OD, 7:3) d 5.90 (dddd, J=2.85, 5.61, 10.31,
17.24 Hz, 2H), 5.31 (ddd, J=1.94, 3.67, 17.24 Hz, 2H),
5.21 (ddd, J=1.32, 2.85, 10.31 Hz, 2H), 4.22 (d, J=5.40
Hz, 2H), 4.18–4.01 (m, 6H), 3.95 (XX0 part of AA0XX0,
2H), 2.70 (s, 6H), 2.28 (dh, J=5.40, 6.89 Hz, 2H), 0.94
(d, J=6.89 Hz, 6H), 0.87 (d, J=6.89 Hz, 6H); 13C
NMR (CDCl3/CD3OD, 4:2) d 172.5, 172.3, 133.8,
118.6, 80.4, 72.4, 72.2, 58.6, 30.2, 26.2, 19.8, 17.5. Anal.
tration.45 The Ki was subsequently calculatedby using
Ki=IC50/(1+[S]/Km)46 anda Km value determined
according to Michaelis–Menten.
MD simulations and free energy calculations
The linear interaction energy method(LIE), described
in detail elsewhere,21,22 can be usedto predict absolute
binding free energies of receptor–ligand complexes. A
general version of the LIE formula can be written as
ꢀ
ꢁ
ꢀ
vdWꢁ
ꢀ
ꢁ
ꢀ
ꢁ
el
lÀs
el
lÀs
DGbind ¼að VvlÀdsW À VlÀs Þþbð V
À V
Þþg ð1Þ
p
w
p
w
.
calcdfor C 24H42N4O8 1/4H2O: C, 55.53; H, 8.25; N,
10.79. Found: C, 55.2; H, 8.1; N, 10.6.
where nonbonded ligand-surrounding (lꢀs) polar (electro-
static) andnonpolar (van der Waals) components of the
interaction energies are denoted by the superscripts el
andvdW, respectively, while the subscripts p andw
describe the ligand in complex with the solvated protein
andfree in water, respectively. Thus, in each binding
calculation an MD simulation of the complex is per-
formedas well as one of the free ligandin water. The
averages are calculatedas time averages from molecular
dynamics trajectories. The previously determined values
of a, b and g from ref 23 are usedalso here. In this
model the b parameter takes into account the deviation
from linear response and depends on the chemical nat-
ure of the ligand. For all ligands considered in this work
b=0.33. The values for a and g are 0.181 and0,
respectively.31
(2R,3R,4R,5R)-N1,N6-Bis[(1S,2R)-2-hydroxy-1-indanyl]-
2,5-di(allyloxy)-3,4-dihydroxyhexane-1,6-diamide (4). A
mixture of 2,5-di-O-allyl-l-mannaro-1,4:3,6-di-g-lac-
tone (1.2 g, 4.5 mmol) and(1 S,2R)-1-amino-2-indanol
(2.7 g, 18.2 mmol) in dry CH2Cl2 (115 mL) was heated
to reflux for 16 h under a nitrogen atmosphere. The
solvent was removedunder reducedpressure andpur-
ification by flash chromatography (CHCl3/CH3OH,
100:3) gave product 4 (1.6 g, 62%) as a white solid:
[a]3D0 ꢀ22.6ꢄ (c 1.00, CHCl3) H NMR (CDCl3) d 7.36
1
(d, J=8.87 Hz, 2H), 7.27–7.18 (m, 8H), 5.91 (ddt,
J=5.77, 10.35, 17.28 Hz, 2H), 5.34 (ddd, J=1.52, 3.05,
17.28 Hz, 2H), 5.32–5.24 (m, 2H), 5.23 (ddd, J=1.32,
2.76, 10.35 Hz, 2H), 4.62–4.55 (m, 2H), 4.41 (app s, 4H),
4.21–4.16 (m, 8H), 3.06 (dd, J=5.40, 16.66 Hz, 2H), 2.86
(dd, J=1.94, 16.66 Hz, 2H); 13C NMR (CDCl3) d 171.6,
140.8, 139.8, 133.2, 128.2, 126.9, 125.3, 123.8, 118.6, 80.9,
Automated docking with Autodock347 of the ligands
with a rigidprotein didnot yieldany useful results,
which may be due to steric reasons if the binding pocket
in the X-ray structure is partially collapsed. The X-ray
structure 1SME3 of plasmepsin II in complex with pep-
statin A was insteadusedas a starting point for all
simulations of the protein–ligandcomplexes andfive
internal water molecules were kept (804, 820, 826, 851
and892 as numberedin the PDB entry). In order to get
reliable results from binding free energy calculations it is
necessary to use a reasonably well-docked starting con-
formation of the protein–ligandcomplex. Since the
hydrogen bonding donors and acceptors of the present
inhibitor backbone are identical to those of the corres-
ponding pepstatin A P10–P30 moieties, it is likely that
the binding conformation of the backbone of 3a–h is
similar to that of pepstatin A at these positions. In
addition, the P2, P1, P10 andP2 0 side chains are likely to
.
72.4, 72.2, 71.5, 57.7, 39.2. Anal. calcdfor C 30H36N2O8 1/
4H2O: C, 64.68; H, 6.60; N, 5.03. Found: C, 64.6; H, 6.5;
N, 5.1.
Plasmepsin assay and Ki determination
Pro-Plm II was a generous gift from Helena Danielson
(Department of Biochemistry, Uppsala University,
Uppsala, Sweden). The activity of Plm II was measured
essentially as described previously,8 using a total reac-
tion volume of 100 mL. The concentration of pro-Plm II
was 3 nM. The pro-sequence of Plm II was cleavedoff
by preincubation in assay reaction buffer (100 mM
sodium acetate buffer (pH 4.5), 10% glycerol and
0.01% Tween 20) at room temperature for 40 min. The
reaction was initiatedby the addition of 3 mM substrate
(DABCYL-Glu-Arg-Nle-Phe-Leu-Ser-Phe-Pro-EDANS,
AnaSpec Inc, San Jose, CA, USA) andhydrolysis was
recorded as the increase in fluorescence intensity over a
10-min time period, during which the rate increased
linearly with time.
0
reside in S2, S1, S10 andS2 , respectively. Therefore the
natural docking approach was to superimpose the entire
ligandbackbone onto that of pepstatin A in the X-ray
structure and fit the side chains into the corresponding
subsites. Here, nonpolar interaction energies rose sig-
nificantly due to the repulsive 1/r12 nonbonded interac-
tion potential term but decreased in the following MD
simulations, where the complex was allowedto relax.
Stock solutions of inhibitors in DMSO were serially
diluted in DMSO and added directly before addition of
substrate, giving a final DMSO concentration of 1%.
0
Since superposition of the ligandbackbone, the P2 side
0
chain andthe P hydroxyl on the corresponding groups
IC50 values were obtainedby assuming competitive inhi-
bition andfitting a Langmuir isotherm [ vi/vo=1/(1+[I]/
IC50)] to the dose response data (Grafit), where vi and vo
are the initial velocities for the inhibitedanduninhibited
reaction respectively and[I] is the inhibitor concen-
of pepstatin A yieldthe SRRRRS isomer, all other iso-
mers were generatedby subjecting the SRRRRS isomer
to the appropriate potential function andthus changing
its stereochemistry. During this process protein and
ligand were kept under restraints in order to prevent

3732
K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
sudden structural changes due to high energies. All MD
simulations were done using the program Q48 andthe
force fieldparameters of GROMOS87. AM1 calcula-
calculations described above. The estimated affinity
using this scoring function has the form
49
score
bind
DG
¼ DGH-bondNH-bond þ DGmetalNmetal þ DGlipoNlipo
þ DGrotNrot þ DG0
tions were performedto obtain partial charges for the
P1/P10 side chains of the inhibitors while the backbone
and hydroxyl groups were assigned standard force field
charges. Asp34, which is involvedin a hydrogen bond
with Asp214 in the X-ray structure, was protonatedat
the oxygen interacting with Asp214. (The groups sur-
rounding the aspartic acids yield virtually equivalent
environments for the two residues and Asp34 was cho-
sen as the one to be protonated.) The Asp303 side chain
was also protonatedbecause of its hydrogen bondwith
the backbone oxygen of Ser215, while Asp214 was
assigneda negative charge. A simulation sphere for the
molecular dynamics was defined with 18 A radius and a
where NH-bond denotes the number of H-bonds, based
on distance and angle criteria for the atoms involved,
Nmetal is the number of interactions between the ligand
andmetal ions in the receptor, Nlipo, is the number of
interactions between lipophilic groups in the ligandand
receptor and Nrot is the number of rotatable bonds in
the ligand that are considered frozen due to interactions
with the receptor. The constants DGH-bond, DGlipo
DGmetal, DGrot, and DG0 have been assignedempiri-
cally.34
,
2
1.5 A restrained(50 kcal/mol A ) outer shell andthe
center of simulation was chosen at the P10 hydroxyl
oxygen of the ligand. The system within the simulation
sphere interactedwith the system across the boundary
only by bonds and angles. The simulation sphere was
solvatedin an SPC water gridwith radius 18 A , using a
minimum solute–solvent distance of 2.4 A. Long-range
electrostatics of the system, except interactions involving
Acknowledgements
We thank the Swedish Foundation for Strategic
Research (SSF), the National Graduate School of Sci-
entific Computing (NGSSC) andthe Swedish Research
Council (VR) for financial support.
the ligand, were treated with the LRF method50 andthe
51
water surface was subjectedto SCAAS restraints
in
order to mimic bulk water at the sphere boundary. MD
simulations of the free ligands in water were set up using
the same simulation sphere parameters anda central
atom of the ligandwas restrainedto the center of the
sphere by 100 kcal/mol A2 to keep the ligandfrom
drifting towards the sphere boundary. All MD simula-
tions useda 1 fs timestep andenergies were recorded
every 10 fs. Each solvatedprotein–ligandsystem was
gradually heated from 1 to 300 K. During the heating
process restraints were imposedon the protein–ligand
amide H-bonds and on those between the hydroxyl
groups andthe catalytic aspartates. Subsequently, the
system was equilibratedafter which energies were
recorded every 50 fs during 300–600 ps. In addition,
simulations of the ligands in aqueous solution were
performed under the same conditions as the protein
simulation. Here, a protocol was usedwhere the water
molecules were first equilibratedaroundthe restrained
ligandwhile changing stereochemistry ude to the
appliedpotential function to ensure an extendedstart-
ing conformation, andthe whole system was then
relaxedandequilibrated. Data collection was performed
for 1.2–2.35 ns until stable ligand-surrounding interac-
tions energy averages were obtained. Since the
SRRRRS and RSSSSR compounds are mirror images
of each other, only one of the stereoisomers needed to
be simulatedin water. This is also true for the RRRRRR
and SSSSSS compounds, so in total two simulations
were performedfor the allyl andtwo for the benzyl
ligands.
References and Notes
1. Bremen, J. Am. J. Trop. Med. Hyg. 2001, 64, 1.
2. Werbovetz, K. A. Curr. Med. Chem. 2000, 7, 835.
3. Silva, A. M.; Lee, A. Y.; Gulnik, S. V.; Maier, P.; Collins,
J.; Bhat, T. N.; Collins, P. J.; Cachau, R. E.; Luker, K. E.;
Gluzman, I. Y.; Francis, S. E.; Oksman, A.; Goldberg, D. E.;
Erickson, J. W. Proc. Natl. Acad. Sci. U.S.A. 1996, 93, 10034.
4. Moon, R. P.; Tyas, L.; Certa, U.; Rupp, K.; Bur, D.; Jac-
quet, C.; Matile, H.; Loetscher, H.; Grueninger-Leitch, F.;
Kay, J.; Dunn, B. M.; Berry, C.; Ridley, R. G. Eur. J. Bio-
chem. 1997, 244, 552.
5. Carroll, C. D.; Patel, H.; Johnson, T. O.; Guo, T.;
Orlowski, M.; He, Z.-M.; Cavallaro, C. L.; Guo, J.; Oksman,
A.; Gluzman, I. Y.; Connelly, J.; Chelsky, D.; Goldberg, D. E.;
Dolle, R. E. Bioorg. Med. Chem. Lett. 1998, 8, 2315.
6. Carroll, C. D.; Johnson, T. O.; Tao, S.; Lauri, G.;
Orlowski, M.; Gluzman, I. Y.; Goldberg, D. E.; Dolle, R. E.
Bioorg. Med. Chem. Lett. 1998, 8, 3203.
7. Dolle, R. E.; Guo, J.; O’Brien, L.; Jin, Y.; Piznik, M.;
Bowman, K. J.; Li, W.; Egan, W. J.; Cavallaro, C. L.;
Roughton, A. L.; Zhao, W.; Reader, J. C.; Orlowski, M.;
Jacob-Samuel, B.; DiIanni Carroll, C. J. Comb. Chem. 2000, 2,
716.
8. Haque, T. S.; Skillman, A. G.; Lee, C. E.; Habashita, H.;
Gluzman, I. Y.; Ewing, T. J. A.; Goldberg, D. E.; Kuntz, I. D.;
Ellman, J. A. J. Med. Chem. 1999, 42, 1428.
9. Jiang, S.; Prigge, S. T.; Wei, L.; Gao, Y.-E.; Hudson, T. H.;
Gerena, L.; Dame, J. B.; Kyle, D. E. Antimicrob. Agents
Chemother. 2001, 45, 2577.
10. Nezami, A.; Luque, I.; Kimura, T.; Kiso, Y.; Freire, E.
Biochemistry 2002, 41, 2273.
11. Noteberg, D.; Hamelink, E.; Hulten, J.; Wahlgren, M.;
Vrang, L.; Samuelsson, B.; Hallberg, A. J. Med. Chem. 2003,
46, 734.
12. Dahlgren, A.; Kvarnstrom, I.; Vrang, L.; Hamelink, E.;
Hallberg, A.; Rosenquist, A.; Samuelsson, B. Bioorg. Med.
Chem. 2003, 11, 827.
The scoring function of Eldridge et al.33 was appliedto
each complex after minimization of an initial MD-equi-
libratedstructure, as well as for 100 energy-minimized
trajectory snapshots. The minimization of the com-
plexes was performedby running MD at 1 K tempera-
ture for 4000 steps using the same setup as for the LIE
13. Oscarsson, K.; Oscarson, S.; Vrang, L.; Hamelink, E.;

K. Ersmark et al. / Bioorg. Med. Chem. 11 (2003) 3723–3733
3733
Hallberg, A.; Samuelsson, B. Bioorg. Med. Chem. 2003, 11,
1235.
14. Asojo, O. A.; Afonina, E.; Gulnik, S. V.; Yu, B.; Erick-
son, J. W.; Randad, R.; Medjahed, D.; Silva, A. M. Acta
Cryst. 2002, D58, 2001.
15. Alterman, M.; Andersson, H. O.; Garg, N.; Ahlsen, G.;
Lovgren, S.; Classon, B.; Danielson, U. H.; Kvarnstrom, I.;
Vrang, L.; Unge, T.; Samuelsson, B.; Hallberg, A. J. Med.
Chem. 1999, 42, 3835.
16. Halperin, I.; Ma, B.; Wolfson, H.; Nussinov, R. Proteins
2002, 47, 409.
17. Schaffer, L.; Verkhivker, G. M. Proteins 1998, 33, 295.
18. Wang, J.; Kollman, P. A.; Kuntz, I. D. Proteins 1999, 36, 1.
19. Zabell, A. P.; Post, C. B. Proteins 2002, 46, 295.
20. Alterman, M.; Bjoersne, M.; Muehlman, A.; Classon, B.;
Kvarnstroem, I.; Danielson, H.; Markgren, P.-O.; Nill-
roth, U.; Unge, T.; Hallberg, A.; Samuelsson, B. J. Med.
Chem. 1998, 41, 3782.
21. Aqvist, J.; Medina, C.; Samuelsson, J. E. Protein Eng.
1994, 7, 385.
22. Aqvist, J.; Luzhkov, V. B.; Brandsdal, B. O. Acc. Chem.
Res. 2002, 35, 358.
23. Hansson, T.; Aqvist, J. Protein Eng. 1995, 8, 1137.
24. Marelius, J.; Graffner-Nordberg, M.; Hansson, T.; Hall-
berg, A.; Aqvist, J. J. Comput. Aided Mol. Des. 1998, 12, 119.
25. Graffner-Nordberg, M.; Marelius, J.; Ohlsson, S.; Pers-
son, A.; Swedberg, G.; Andersson, P.; Andersson, S. E.;
Aqvist, J.; Hallberg, A. J. Med. Chem. 2000, 43, 3852.
26. Graffner-Nordberg, M.; Kolmodin, K.; Aqvist, J.;
Queener, S. F.; Hallberg, A. J. Med. Chem. 2001, 44, 2391.
27. Westling, J.; Cipullo, P.; Hung, S. H.; Saft, H.; Dame,
J. B.; Dunn, B. M. Protein Sci. 1999, 8, 2001.
28. Linstead, R. P.; Owen, L. N.; Webb, R. F. J. Chem. Soc
1953, 1225.
29. Patil, V. J. Tetrahedron Lett. 1996, 37, 1481.
30. Bursavich Matthew, G.; Rich Daniel, H. J. Med. Chem.
45, 541.
31. Hansson, T.; Marelius, J.; Aqvist, J. J. Comput. Aided
Mol. Des. 1998, 12, 27.
Lin, J.; Chen, I.-W.; Vastag, K.; Ostovic, D.; Anderson, P. S.;
Emini, E. A.; Huff, J. R. Proc. Natl. Acad. Sci. U.S.A. 1994,
91, 4096.
33. Eldridge, M. D.; Murray, C. W.; Auton, T. R.; Paolini,
G. V.; Mee, R. P. J. Comput. Aided Mol. Des 1997, 11, 425.
34. Marelius, J.; Ljungberg, K. B.; Aqvist, J. Eur. J. Pharm.
Sci. 2001, 14, 87.
35. The inhibitory effect of the isomers 3a–h was determined
with purified HIV-1 protease in a standardized assay as
described in ref 37. Only the SRRRRS isomers 3a and 3c
showedany inhibitory effect, Ki=0.0002 and1.4 mM, respec-
tively.
36. Nillroth, U.; Besidsky, Y.; Classon, B.; Chattopadhyaya,
J.; Ugi, I.; Danielson, U. H. Drug Des. Discov. 1995, 13, 43.
37. Brandsdal, B. O.; Aqvist, J.; Smalas, A. O. Protein Sci.
2001, 10, 1584.
38. Luzhkov, V. B.; Aqvist, J. FEBS Lett. 2001, 495, 191.
39. King, N. M.; Melnick, L.; Prabu-Jeyabalan, M.; Nali-
vaika, E. A.; Yang, S.-S.; Gao, Y.; Nie, X.; Zepp, C.; Heefner,
D. L.; Schiffer, C. A. Protein Sci. 2002, 11, 418.
40. Asojo, O. A.; Gulnik, S. V.; Afonina, E.; Yu, B.; Ellman,
J. A.; Haque, T. S.; Silva, A. M. J. Mol. Biol. 2003, 327, 173.
41. Kollman, P. Chem. Rev. 1993, 93, 2395.
42. Srinivasan, J.; Miller, J.; Kollman, P. A.; Case, D. A. J.
Biomol. Struct. Dyn. 1998, 16, 671.
43. Wang, J.; Morin, P.; Wang, W.; Kollman, P. A. J. Am.
Chem. Soc. 2001, 123, 5221.
44. Bohm, H. J. J. Comput. Aided Mol. Des 1998, 12, 309.
45. Copeland, R. A. (Ed.). Enzymes: A Practical Introduction
to Structure, Mechanism, and Data Analysis; VCH, New York,
N.Y, 1996.
46. Cheng, Y.; Prusoff, W. H. Biochem. Pharmacol. 1973, 22,
3099.
47. Morris, G. M.; Goodsell, D. S.; Halliday, R. S.; Huey, R.;
Hart, W. E.; Belew, R. K.; Olson, A. J. J. Comput. Chem.
1998, 19, 1639.
48. Marelius, J.; Kolmodin, K.; Feierberg, I.; Aqvist, J. J.
Mol. Graph. Model. 1998, 16, 213.
49. van Gunsteren, W. F.; Brendsen, H. J. C. Groningen
Molecular Simulation (GROMOS) Library Manual; Biomos:
Nijenborgh, Groningen, The Netherlands, 1987.
50. Lee, F. S.; Warshel, A. J. Chem. Phys. 1992, 97, 3100.
51. King, G.; Warshel, A. J. Chem. Phys. 1989, 91, 3647.
32. Vacca, J. P.; Dorsey, B. D.; Schleif, W. A.; Levin, R. B.;
McDaniel, S. L.; Darke, P. L.; Zugay, J.; Quintero, J. C.;
Blahy, O. M.; Roth, E.; Sardana, V. V.; Schlabach, A. J.;
Graham, P. I.; Condra, J. H.; Gotlib, L.; Holloway, M. K.;