Quantitative three dimensional structure linear interaction energy model of 5′-O-[N-(salicyl)sulfamoyl]adenosine and the aryl acid adenylating enzyme MbtA

Article abstract of DOI:10.1021/jm800668u

MbtA (a salicyl AMP ligase) is a key target for the design of new antitubercular agents. On the basis of structure-activity relationship (SAR) data generated in our laboratory, a structure-based model is developed to predict the binding affinities of aryl acid-AMP bisubstrate inhibitors of MbtA. The approach described takes advantage of the linear interaction energy (LIE) technique to derive linear equations relating ligand structure to function. With only two parameters derived from molecular dynamics simulations, good correlation (R² = 0.70) was achieved for a set of 31 inhibitors with binding affinities spanning 6 orders of magnitude. The results were applied to understand the effect of steric and heteroatom substitutions on bisubstrate ligand binding and to predict second generation inhibitors of MbtA. The resulting model was further validated by chemical synthesis of a novel inhibitor with a predicted LIE binding affinity of 1.6 nM and a subsequently determined experimental K_i^app of 0.7 nM.

Full text of DOI:10.1021/jm800668u

7154
J. Med. Chem. 2008, 51, 7154–7160
Quantitative Three Dimensional Structure Linear Interaction Energy Model of
5′-O-[N-(Salicyl)sulfamoyl]adenosine and the Aryl Acid Adenylating Enzyme MbtA
Nicholas P. Labello,† Eric M. Bennett,† David M. Ferguson,*,†,‡ and Courtney C. Aldrich†
Center for Drug Design, Academic Health Center, UniVersity of Minnesota, Minneapolis, Minnesota 55455, and
Department of Medicinal Chemistry, UniVersity of Minnesota, Minneapolis, Minnesota 55455
ReceiVed June 3, 2008
MbtA (a salicyl AMP ligase) is a key target for the design of new antitubercular agents. On the basis of
structure-activity relationship (SAR) data generated in our laboratory, a structure-based model is developed
to predict the binding affinities of aryl acid-AMP bisubstrate inhibitors of MbtA. The approach described
takes advantage of the linear interaction energy (LIE) technique to derive linear equations relating ligand
structure to function. With only two parameters derived from molecular dynamics simulations, good correlation
(R2 ) 0.70) was achieved for a set of 31 inhibitors with binding affinities spanning 6 orders of magnitude.
The results were applied to understand the effect of steric and heteroatom substitutions on bisubstrate ligand
binding and to predict second generation inhibitors of MbtA. The resulting model was further validated by
chemical synthesis of a novel inhibitor with a predicted LIE binding affinity of 1.6 nM and a subsequently
determined experimental Kiapp of 0.7 nM.
Introduction
Iron is an essential nutrient for almost all known organisms;
however, its concentration in serum and human body fluids is
approximately 10-24 M, which is far too low to support bacterial
colonization and growth.1 In order to overcome this iron
limitation, many bacterial pathogens such as Mycobacterium
tuberculosis, the causative agent of tuberculosis (TBa), synthe-
size small molecule iron chelators termed siderophores for iron
acquisition.2 Consequently, siderophore biosynthesis has emerged
as a promising biochemical pathway for antibiotic development.3
The availability of detailed structural and biochemical informa-
tion for aryl acid adenylation enzymes (AAAEs), which catalyze
the first step in biosynthesis of aryl-capped siderophores, has
made these enzymes attractive targets.4-6 5′-O-[N-(Salicyl)sul-
famoyl]adenosine (1, Sal-AMS, Figure 1), a rationally designed
bisubstrate analogue, is the prototypical AAAE inhibitor that
has been shown to potently inhibit AAAEs from numerous
organisms.7-9 As shown in Figure 1, Sal-AMS comprises four
modules: aryl, linker, glycosyl, and nucleobase. We have
systematically and methodically examined the impact of each
module on enzyme inhibition of the AAAE known as MbtA
from M. tuberculosis as well as antitubercular activity to provide
comprehensive structure-activity-relationships (SAR).10-15
The ability to predict the binding affinity of new compounds
can be of substantial benefit during the optimization phase of
drug development. While the difference in free energy of binding
could be calculated exactly for two related molecules it is in
practice an intractable problem to consider large numbers of
Figure 1. Structure of Sal-AMS (1).
ligands in this manner. The free energy perturbation method
(FEP), for instance, requires dozens of converged molecular
dynamics (MD) simulations for each ligand. On the other hand,
the literature contains a number of reports where a predictive
binding model is constructed using experimental binding
affinities to weight theoretical interaction energies.16-19 Within
such linear interaction energy (LIE) approximations, binding
affinities are estimated after only one ligand-receptor and one
ligand-solvent simulation for each additional compound. The
resulting models often display high correlation and, in contrast
to some activity relationship models, have the advantage of
being structure-based and therefore serve as interpreters and
guides for rational drug design. While most such studies utilize
crystal structures, there are examples in the literature where
homology models have been substituted with good results.20,21
We present a structure based analysis and linear LIE model
for the Sal-AMS scaffold with emphasis on providing a
quantitative model for predicting binding affinities and a
grounded physical interpretation of the SAR to guide future
synthesis. Modifications of the nucleobase are of particular
interest, as this moiety represents the best opportunity for
improving potency and increasing specificity and lipophilicty.
The linker and glycosyl regions are also examined, as variation
of these moieties may be required to modify the number of
hydrogen bond donors and acceptors or otherwise tune phar-
macokinetic properties.
* To whom correspondence should be addressed. Address: Department
of Medicinal Chemistry, University of Minnesota, 308 Harvard Street SE,
8-101 WDH, Minneapolis, MN 55455. Telephone: 612-626-2601. Fax: 612-
624-0139. E-mail: ferguson@umn.edu.
†
Center for Drug Design, University of Minnesota.
Department of Medicinal Chemistry, University of Minnesota.
‡
a Abbreviations: AAAE, aryl acid adenylating enzyme; ATP, adenosine
triphosphate; FEP, free energy perturbation; Kiapp, apparent inhibition
constant; LIE, linear interaction energy; MD, molecular dynamics; MM,
molecular mechanics; PDB, Protein Data Bank; QM, quantum mechanics;
QSAR, quantitative structure-activity relationship; rmsd, root-mean-square
deviation; SAR, structure-activity relationships; Sal-AMS, 5′-O-[N-(sali-
cyl)sulfamoyl]adenosine; TB, tuberculosis.
Computational Methods
Receptor and Ligand Starting Structures. Although a crystal
structure for MbtA is not yet available, we have detailed the
10.1021/jm800668u CCC: $40.75
2008 American Chemical Society
Published on Web 10/30/2008

LIE Model of an Adenosine and MbtA
Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22 7155
Table 1. LIE Coefficients for Various Receptor and Ligand Groupsa
A ) R∆GvdW + ꢀ∆GEl + γ
model
R
ꢀ
γ
no. of compds
R2
I
X
6.49 × 10-3 -8.22
31
25
6
31
21
0.66
0.71
0.99
0.70
0.81
N I
L I
II
1.04 × 10-2 1.31 × 10-2 -8.36
1.58 × 10-2 4.98 × 10-3 -8.23
1.37 × 10-2 7.48 × 10-3 -8.48
1.64 × 10-2 1.14 × 10-2 -8.66
BII
a N, L, and B refer to nucleoside, linker, and base compound sets,
respectively. I and II refer to truncation procedures 1 and 2, respectively.
R and ꢀ have units of 1/(kJ/mol) so that multiplication by an interaction
energy in kJ/mol results in a pure number. γ is unitless. A is an estimation
of log(Kiapp). X indicates that the coefficient was statistically insignificant.
that the core of the ligand scaffold can be considered anchored by
these contacts and has little room for adjustment. For substituents
added to the core to have the desired effect of improving potency,
they must readily access the available binding pockets from their
attachment to the core without disrupting the structure apparent in
Figure 2. Compounds 16 and 28 are representative ligands that do
exactly that, accessing the N-6 and C-2 binding clefts, respectively.
While the model was found to be somewhat sensitive to starting
structures (because of the relatively rigid conformation of the core
scaffold discussed above), most ligands have only one conformation
of interest. The active conformation of the ligand Sal-AMS was
constructed in the homology model by mutation of DhbE’s
adenylated natural substrate 2,3-DHB and subsequent minimization.
Ligands with large adenine C-2 substituents result in significant
changes of nearby residues such as Lys332 and Val448. A second
homology model differing from the DhbE crystal structure but
allowing for the exceptional potency of compounds 25-31 was
used to model these ligands. Briefly, the largest difference involves
the loss of a hydrogen bond between the two residues mentioned
above and relaxation of the receptor’s tertiary structure around the
bulkier ligands. MD simulations indicate that the C-2 position,
despite having little obvious room for substitution in the DhbE
crystal structure, is located in an extremely flexible region of the
protein at the seam of the two domains and linker and thus very
amenable to substitution.10 As this work progressed, it was found
that ligands with relatively small modifications of 1 could be studied
effectively with the C-2 pocket residues optimized in either
conformation.
Molecular Dynamics. Two molecular dynamics simulations are
necessary to estimate the binding affinity of a single ligand (ligand
in receptor, ligand in solvent). Two truncation procedures were
applied to the homology model before MD simulation. In the first,
all residues within 10 Å of the ligand were included in the
calculation, and 21 complete residues, those with any atom 3 Å or
closer to Sal-AMS (1) after a frozen-protein minimization, were
allowed to move freely through the course of the simulation. The
outer shell was frozen. After minimization and a 25 ps equilibrium
run, data were sampled every 25 fs and averaged over the entire
simulation. In the second procedure, all residues within 14 Å of
the ligand were included in the simulation. The 33 residues with
any atom within 4 Å of 27 were allowed to move freely while the
outer shell was frozen. To eliminate possible differences due to
the conformation of side chains near the C-2 pocket, the second
truncation procedure was only applied to the more permissive
homology model, which was used with all of the ligands. After an
identical equilibrium run data averaged over the entire simulation
were collected. In every case a GB/SA23 implicit solvation model
was used to approximate the effects of solvent. The running average
of the force field terms was collected and examined at 10, 25, and
50 ns for each ligand. As the terms changed very little between
the 25 and 50 ns simulations, no longer calculations were completed.
LIE Terms. The MD-LIE terms collected in this work differ
only by which truncation procedure was used and length of
simulation. All of the MD models reported utilize values from the
50 ns simulation of the ligand in solution (implicit solvation) and
the electrostatic and van der Waals terms from increasingly lengthy
simulations of the complexes to yield the ∆GvdW and ∆GEl terms
Figure 2. Sal-AMS (1) displayed in tube representation bound in a
homology model of the MbtA binding site.10 The N-terminal domain
residues are presented in red ribbon, and the C-terminal domain residues
are in blue. Residues that make important electrostatic contacts with
the ligand are presented in tube representation. The orange dotted lines
are possible hydrogen bonds. Hydrogens are not shown for clarity.
construction of a homology model in a previous publication.10 Our
homology model is based on the cocrystal structure of DhbE with
an adenylated 2,3-dihydroxybenzoic acid (2,3-DHB).22 DhbE shares
42% sequence identity with MbtA but almost absolute conservation
of active site residues. Thus, 16 of the 21 residues within 4 Å of
the adenylated ligand are identical in MbtA and the remaining 5
changes are conservative. Three of these variations in MbtA
(Y236F, S240C, V337L) map to the aryl acid substrate binding
pocket and are responsible for conferring selectivity to the native
substrate salicylic acid (Sal) over 2,3-DHB. The predicted binding
conformation of Sal-AMS within the homology model is shown in
Figure 2.
The ligand Sal-AMS (1) assumes a relatively compact form when
bound to the receptor compared to the extended conformations that
are possible in solution phase. Sal-AMS forms an internal hydrogen
bond between the 2-hydroxy and negatively charged nitrogen atom
of the acylsulfamate linker, which enforces a coplanar arrangement
of the aryl moiety and linker carbonyl. The aryl binding pocket is
largely nonpolar, and only a single hydrogen bond between the
carboxamide side chain of Asn258 and the aryl hydroxyl group of
1 is predicted. The linker moiety of 1 interacts with conserved
Lys542 (protonated) via a network of three hydrogen bonds and
electrostatically with the negatively charged sulfamate as shown
in Figure 2. The glycosyl moiety adopts a 3′-endo pucker, and both
2′- and 3′- hydroxyl groups are predicted to form a bidentate
hydrogen bond with Asp436. Deletion of either hydroxyl group in
8 and 9 (Table 1) does not disturb binding affinity, while deletion
of both oxygen atoms in analogue 10 results in an approximate
10-fold loss in binding affinity. The interactions of the nucleobase
are largely hydrophobic with a single predicted hydrogen bond
between the exocyclic N-6 amino group and Val352, which is
further supported by the 238-loss in binding affinity when the N-6
amino group is replaced with an oxygen atom. Significantly, no
electrostatic interactions of the purine N-1 and N-3 atoms are
observed, and the N-7 shows poor hydrogen bonding geometry with
Gly330 and Gly354. A small pocket vicinal to the N-6 amino group
enables binding of no more than three carbon atoms, while the
interdomain region of the receptor adjacent to the C-2 purine
position is flexible as predicted by molecular dynamics simulations
and can accommodate bulky substituents. A major consequence of
the significant and relatively rigid hydrogen bonding structure is

7156 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22
Labello et al.
of eq 1. MD calculations were completed with MacroModel24 9.1
using the OPLS force field.25
Results and Discussion
The most basic LIE model is given below in eq 1 where A is
a free energy, IC50, KD, or some other experimentally measured
indication of binding affinity, ∆GvdW and ∆GEl are the force
field determined van der Waals and electrostatic interaction
energies of the receptor-ligand complex, and R, ꢀ, and γ are
constants usually derived through multivariate regression to best
fit the available data.
A ) R∆GvdW + ꢀ∆GEl + γ
(1)
The ∆G vdW term approximates nonpolar binding contributions,
while ∆GEl accounts for electrostatic free energy differences
of the ligand-solvent and ligand-receptor complexes. Each
represents the difference in interaction energy of the ligand in
(a) the receptor environment and (b) the solvent environment.
While the constants are typically unique for each receptor and
force field, once R, ꢀ, and γ are established, an affinity
prediction for a new compound can be determined with minimal
effort, requiring only the relevant interaction energy calculation.
We chose to transform the interaction energies calculated from
the MD simulations (∆GvdW and ∆GEl) into values relative to 1
(the parent compound) before further processing. As a result,
the ideal value of γ is the experimental log(Kiapp) of compound
1 (-8.3). Variation in the binding affinity estimation, A, for
each compound is determined by the first two terms. For
example, desolvating an additional hydrophobic group relative
to 1, as in the case of 28, results in a negative ∆GvdW and
correspondingly lower (better binding) estimation of A. An
additional term fitting the difference in solvent accessible surface
area (SASA) is sometimes included to further improve correla-
tion; however, we found no significant improvements when
including a SASA term possibly because of the polar nature of
the ligands (a dipole moment of 16.8 D, for instance, was
calculated for a compact solution phase conformation of 1 at
the B3LYP26,27 level of theory and a basis set28,29 designed for
calculating dipole moments) or the van der Waals issues with
some ligands growing too large for the pocket discussed below.
More recently, specific receptors have been studied with linear
response methods that replace the fitted van der Waals and
electrostatic terms with a QM/MM interaction energy.30,31 The
QM/MM interaction energy was calculated for each ligand at
both minimized and time averaged single points by treating the
protein with the OPLS force field and the ligand at the B3LYP/
6-31G*32 level of theory. While trends were noted for specific
ligand sets, the QM/MM interaction energy was incapable of
describing all or even most of the ligand binding in even a
qualitative sense. We concluded that in this case evaluation of
a relatively large ensemble is more important for estimating
binding affinities than highly accurate studies on a single or
smaller number of structures. Furthermore, attempts to correlate
binding affinity with short 10-50 ps MD simulations featuring
the entire unrestricted MbtA protein resulted in worse correlation
than longer simulations (150-300 ps) in a truncated receptor.
Initially, a poor correlation was noted when including all
active ligands in a single model, and we opted to separate the
inhibitors based on the modular structure of Sal-AMS. The
analogues were separated into groups corresponding to modi-
fications in the aryl, linker, glycosyl, and base moieties. Most
of the error in a unified model resulted from compounds
featuring modifications of the aryl ring. As a result of the
conjugated nature of the salicyl moiety, much of the modulation
Figure 3. Experimental vs estimated binding affinity of MbtA
inhibitors, with LIE model of all compounds within the small receptor
model. The boxed ligands all feature modifications of the linker such
that the charged nitrogen is removed.
in potency caused by substitution of the aryl ring may be
indescribable by classical physics, instead requiring electronic
structure and polarization effects. For example, substituting the
para-position of the aryl ring with the trifluoromethyl group
results in a 380-fold difference in potency, suggesting force
fields based methods may be incapable of describing the
phenomena without highly specific reparameterization. Regard-
less, almost every modification (>20) on the ring has resulted
in a significant loss of potency and the focus of this work is on
the adenosine binding site.22 The linker is a critically important
region, as it must be metabolically stabile and correctly position
the aryl and nucleoside moieties in their respective binding
pockets.13 A systematic series of six compounds differing only
in the linker feature widely varying potencies ranging from 3.8
nM to 143 µM.9,11,13-15 The complete final set of compounds
that we investigated additionally includes 25 inhibitors with
modifications of the glycosyl and nucleobase groups. All Kiapp
values were determined using an ATP pyrophosphate exchange
assay under identical assay conditions, and tight-binding inhibi-
tion observed for many of the most potent compounds was
accounted for using the analysis of Morrison.14 Therefore,
differences in binding affinity as measured by the assay are
representative of differences in free energy. The binding
affinities of the inhibitors span 6 orders of magnitude with Kiapp
values ranging from 270 pM to 143 µM.
The smaller of the two receptor models (truncation procedure
1) has the distinct advantage of being much more computa-
tionally efficient allowing longer simulation times while simul-
taneously converging in fewer time steps because of the fewer
degrees of freedom. Evaluation of eq 1 using 150 ps simulations
of all ligand-receptor complexes results in an R2 value of 0.66.
Experimental log(Kiapp) are plotted against those predicted by
the equation in Figure 3. The resulting coefficients necessary
to predict binding affinity are presented in Table 1. The
structures of these compounds are reported in Table 2 along
with the Kiapp and log(Kiapp) used to construct the model.
Examination of the coefficients reveals that the intercept
constant, γ, maintains a very reasonable value of -8.22. As
the interaction energies of eq 1, ∆GvdW, and ∆GEl, were
transformed into values relative to Sal-AMS (1) before further
statistical analysis, the ideal value of γ is the log(Kiapp) of Sal-
AMS (the relative ∆G and ∆GEl terms are always zero for this
compound). The Kiapp values of the additional compounds are

LIE Model of an Adenosine and MbtA
Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22 7157
Table 2. Experimental and Calculated log(Kiapp) Valuesa
a Calculated using interaction energies from 150 ps MD simulations and truncation procedure II (R ) 1.37 × 10-2, ꢀ ) 7.48 × 10-3, γ ) -8.48)
described in text. b Reference 9. c Reference 12. d Reference 15. e Reference 11. f Reference 10.
then modulated by the ∆GvdW and ∆GEl terms. The coefficients
R and ꢀ are both always positive indicating that the model
responds intuitively to the ∆G terms rewarding compounds that
are predicted to interact more strongly with the receptor with
greater binding affinities. Closer inspection, however, reveals
that R is nearly statistically insignificant and that most of the
correlation is due to the electrostatic term. The most important
reason for this phenomenon appears to be due to compounds
15-24, where a series of aliphatic and aromatic groups are
appended onto the N-6 position of the purine. The removal of

7158 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22
Labello et al.
Figure 4. Plots of the experimental versus estimated binding affinities
for MbtA inhibitors using model I after 150 ps of molecular dynamics
simulations. The linker set (6 compounds, hollow red squares) and
nucleoside set (25 compounds, filled green circles) are treated separately
(see Table 1, LI and NI, respectively). The solid identity line represents
the point where prediction is equal to the experimental value.
Figure 5. Experimental vs estimated binding affinity of MbtA inhitors:
experimental log(Kiapp) vs predicted activity. All compounds are
included in a single set (hollow red squares), though improvement is
still noted by clustering ligands into groups based on which region of
the scaffold was modified (linker, filled green circles, and base
modifications separately as hollow blue triangles). R2 (all compounds,
log(Kiapp) ) 0.0137∆GvdW + 0.00748∆GEl - 8.48) ) 0.70.
these hydrophobic groups from solution results in increasingly
favorable ∆GvdW contributions. The binding affinity, however,
shows a correlated decrease as the substituents grow larger than
three carbon atoms. The isobutyl, cyclobutyl, cyclopentyl, and
benzyl substituents are larger than the volume available in the
N-6 pocket and significantly displace the nucleoside (rms of
heavy base atoms 1.6-1.9 Å) making room for the large
hydrophobic groups. As the more polar glycosyl and linker
regions are displaced, an energetic penalty is partially expressed
in the electrostatic term, recovering correlation though in a
nonphysical manner resulting from unrealistic binding modes.
Further examination of Figure 3 reveals that even for
compounds without large substituents the electrostatic term alone
is not an adequate predictor of activity. Four ligands included
in the set modify the linker such that the negatively charged
nitrogen is either removed from the scaffold or has neighboring
atoms modified such that it is neutral within the receptor.
Removing the compounds with linker modifications increases
the R2 to 0.71 for the remaining set (Figure 4). While most of
the correlation is still due to the electrostatic term, R increases
to a meaningful value of 1.04 × 10-2. Increasing simulation
time to 300 ps resulted in no significant changes to the model
and only very slightly improved R2 to 0.72. Simulations longer
than 300 ps did not improve correlation. The six compounds
with varying linker groups (1-6), when treated separately from
the nucleoside modifications, demonstrate an extremely high
R2 of 0.99. While the correlation would undoubtedly decrease
with a larger set of compounds, the physically sensible values
of the coefficients (R ) 1.58 × 10-2, ꢀ ) 4.98 × 10-3, γ )
-8.23) and individual R2 values resulting from ∆GvdW and ∆GEl
of 0.39 and 0.82, respectively, indicate that this small receptor
model should be a reasonable starting point for interpretation
of any similar future modifications. It also implies that differ-
ences in charge or protonation state may require reparameter-
izing the electrostatic coefficient or properly accounting for this
effect by electrostatic continuum calculations.
are to be included in a single model. Additionally, the relatively
poor correlation of the van der Waals term and observation that
several N-6 and C-2 substituents are large enough to extend
well beyond the volume occupied by 1 indicate that it may be
important to extend the shell of residues allowed to move. To
address these deficiencies, a study similar to that described above
was completed, though modeling all residues with any atom
closer than 14 Å and allowing 33 residues in an expanded inner
shell to move freely. Interestingly, many aspects of the model
were improved (Figure 5). The van der Waals term is consider-
ably more significant, demonstrating an R2 of 0.33. The
electrostatic term continues to correlate well, R2 ) 0.56, and
remains independent of the van der Waals term demonstrated
by the cross correlation between the two of only 0.08.
Furthermore, the improvement from using both terms over the
entire set of 31 ligands is clear with an R2 of 0.70 and standard
deviation of 0.85 log units. The leave-group-out (or leave-one-
out) cross-validation method was run for 100 cycles with group
sizes of 1, 2, 4, and 9. That is, 100 (or 31 for a group size
of 1) randomly selected groups of molecules of the appropriate
size were treated as a test set with the remainder serving as a
training set. The average correlation coefficient (q2) of the test
sets was generally near and did not fall below 0.60 in any of
the validation tests. Furthermore, manually breaking the ligands
into a trainer and test set of 21 and 10 ligands (chosen to
represent modifications of the linker, glycosyl, and base
subunits) yields R2 values of 0.69 and 0.59, respectively, in line
with the leave-group-out analysis.
Correlation can be increased further if the ligands are broken
into sets that split away the four compounds with very different
electrostatic properties. While all parameters are allowed to
reoptimize freely, the van der Waals term encouragingly varies
little between groups while the electrostatic coefficient adjusts
significantly accounting for the difference in electrostatic
interaction energies between charged and noncharged species.
Finally, issues apparent in the smaller, less flexible receptor
model have been addressed. The N-6 pocket appears to be more
realistically modeled, as the substituents that crowd the small
cleft no longer displace a large part of the ligand resulting in
electrostatic penalties but instead see van der Waals energetic
Several issues were noted that indicate that a larger, more
flexible model may be important to provide a fully rational
interpretation of the binding affinity data. Charge-charge
interactions in particular are long-range, suggesting that a larger
protein shell may be necessary if ligands with different charges

LIE Model of an Adenosine and MbtA
Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22 7159
Figure 6. Sal-AMS (1) docked in the MbtA homology model. The protein is represented as a ribbon diagram: N-terminal, red; C-terminal, blue.
The ligand is shown in a tube representation and colored by element. The green surfaces expose volumes that, when filled with hydrophobic groups,
improve potency and specificity. The larger of the two is the C-2 cleft (accessed by compounds 25-31), while the smaller is the N-6 pocket
(accessed by compounds 15-24). Filling the yellow volumes results in a reduction in potency.
penalties consistent with steric crowding. The electrostatic term
of the four neutral ligands does not fit well with the remainder
of the set, but reduced significance of ∆GEl with respect to
∆GvdW resulting from the larger, more flexible receptor allows
for the full set to be included in a unified model and still exhibit
quite reasonable predictive power. For affinity predictions of
new molecules, however, a model trained with only the most
closely related compounds appears to be the wisest course of
action. For example, to investigate dual modifications of the
scaffold at the purine N-6 and C-2 positions a priori, we would
consider only the 21 compound subset representing single
modifications at these positions.
potency and specificity. As a final check of the applicability of
the model, we have applied the LIE analysis to predict the
binding affinity of a compound not included in the initial training
set that includes modifications to both the N-6 and C-2 positions.
The binding affinity of N-6-cyclopropyl-2-phenyl-Sal-AMS was
estimated using the LIE model to be 1.6 nM, indicating the
activity would fall between that of 23 and 28. This compound,
the first dual modification of the scaffold, has been synthesized
and found to have a Kiapp of 0.7 nM.33 The analogue is
approximately twice as potent as 23 but 3-fold less active than
28, in concert with the LIE predicted result. Additional
applications of the LIE model are currently underway to guide
the synthesis of even more diverse Sal-AMS analogues, and
on the basis of the results presented here, significant gains in
potency is expected.
Conclusions
We have presented the first computational analysis of MbtA
and analogues based on 5′-O-[N-(salicyl)sulfamoyl]adenosine,
a promising class of new antibacterial compounds with a novel
mechanism of action. The LIE model provides a clear frame-
work for both qualitatively predicting trends and quantitatively
estimating binding affinities of related compounds within MbtA
using only two short MD simulations. The improvements in
predictive power realized with a larger, more flexible protein
model speaks well for the applicability of structure based design
and interpretation for MbtA inhibitors within our homology
model. Moreover, it seems reasonable to expect that similar
methods should be applicable to a number of homologous aryl
acid adenylating enzymes such as BasE, YbtE, VibE, PchD,
and EntE (from the pathogens Acinetobacter baumannii, Yers-
inia pestis, Vibrio cholerae, Pseudomonas aeruginosa, and
Escherichia coli, respectively).
Acknowledgment. This research was supported by a grant
from the NIH (Grant R01AI070219) and funding from the
Center for Drug Design, Academic Health Center, University
of Minnesota to C.C.A. We thank the Minnesota Supercom-
puting Institute for computing time.
Supporting Information Available: Complete details describing
the synthesis, compound characterization, and HPLC purity of N6-
cyclopropyl-5′-O-[N-(2-hydroxybenzoyl)sulfamoyl]-2-phenylade-
nosine sodium salt. This material is available free of charge via
the Internet at http://pubs.acs.org.
References
(1) Raymond, K. N.; Dertz, E. A.; Kim, S. S. Enterobactin: an archetype
for microbial iron transport. Proc. Natl. Acad. Sci. U.S.A. 2003, 100,
3584–3588.
(2) Crosa, J. H.; Walsh, C. T. Genetics and assembly line enzymology of
siderophore biosynthesis in bacteria. Microbiol. Mol. Biol. ReV. 2002,
66, 223–249.
Figure 6 presents an amalgamation of experimental and
theoretical work, summarizing the structure-activity relationship
data used in the development of the LIE model. The schematic
shows where the scaffold can be most easily modified to
favorably access additional volume within MbtA and improve
(3) Monfeli, R. R.; Beeson, C. Targeting iron acquisition by Mycobac-
terium tuberculosis. Infect. Disord.: Drug Targets 2007, 7, 213–220.

7160 Journal of Medicinal Chemistry, 2008, Vol. 51, No. 22
Labello et al.
(4) Drake, E. J.; Nicolai, D. A.; Gulick, A. M. Structure of the EntB
multidomain nonribosomal peptide synthetase and functional analysis
of its interaction with the EntE adenylation domain. Chem. Biol. 2006,
13, 409–419.
(18) Sham, Y. Y.; Chu, Z. T.; Tao, H.; Warshel, A. Examining methods
for calculations of binding free energies: LRA, LIE, PDLD-LRA, and
PDLD/S-LRA calculations of ligands binding to an HIV-protease.
Proteins 2000, 39, 393–407.
(5) Ehmann, D. E.; Shaw-Reid, C. A.; Losey, H. C.; Walsh, C. T. The
EntF and EntE adenylation domains of Escherichia coli enterobactin
synthetase: sequestration and selectivity in acyl-AMP transfers to
thiolation domain cosubstrates. Proc. Natl. Acad. Sci. U.S.A. 2000,
97, 2509–2514.
(6) May, J. J.; Kessler, N.; Marahiel, M. A.; Stubbs, M. T. Crystal structure
of DhbE, an archetype for aryl acid activating domains of modular
nonribosomal peptide synthetases. Proc. Natl. Acad. Sci. U.S.A. 2002,
99, 12120–12125.
(7) Ferreras, J. A.; Ryu, J. S.; Di Lello, F.; Tan, D. S.; Quadri, L. E.
Small-molecule inhibition of siderophore biosynthesis in Mycobac-
terium tuberculosis and Yersinia pestis. Nat. Chem. Biol. 2005, 1, 29–
32.
(8) Miethke, M.; Bisseret, P.; Beckering, C. L.; Vignard, D.; Eustache,
J.; Marahiel, M. A. Inhibition of aryl acid adenylation domains
involved in bacterial siderophore synthesis. FEBS J. 2006, 273, 409–
419.
(9) Somu, R. V.; Wilson, D. J.; Bennett, E. M.; Boshoff, H. I.; Celia, L.;
Beck, B. J.; Barry, C. E.; Aldrich, C. C. Antitubercular nucleosides
that inhibit siderophore biosynthesis: SAR of the glycosyl domain.
J. Med. Chem. 2006, 49, 7623–7635.
(10) Neres, J.; Labello, N. P.; Somu, R. V.; Boshoff, H. I.; Wilson, D. J.;
Vannada, J.; Chen, L.; Barry, C. E., 3rd; Bennett, E. M.; Aldrich,
C. C. Inhibition of siderophore biosynthesis in Mycobacterium
tuberculosis with nucleoside bisubstrate analogues: SAR of the
nucleobase domain of 5′-O-[N-(salicyl)sulfamoyl]adenosine. J. Med.
Chem. 2008, 51, 5349–5370.
(11) Qiao, C. H.; Gupte, A.; Boshoff, H. I.; Wilson, D. J.; Bennett, E. M.;
Somu, R. V.; Barry, C. E.; Aldrich, C. C. 5′-O-[(N-Acyl)sulfamoy-
l]adenosines as antitubercular agents that inhibit MbtA: an adenylation
enzyme required for siderophore biosynthesis of the mycobactins.
J. Med. Chem. 2007, 50, 6080–6094.
(12) Qiao, C. H.; Wilson, D. J.; Bennett, E. M.; Aldrich, C. C. A
mechanism-based aryl carrier protein/thiolation domain affinity probe.
J. Am. Chem. Soc. 2007, 129, 6350–6351.
(19) Su, Y.; Gallicchio, E.; Das, K.; Arnold, E.; Levy, R. M. Linear
interaction energy (LIE) models for ligand binding in implicit solvent:
theory and application to the binding of NNRTIs to HIV-1 reverse
transcriptase. J. Chem. Theory Comput. 2007, 3, 256–277.
(20) Zhou, Z.; Bates, M.; Madura, J. D. Structure modeling, ligand binding,
and binding affinity calculation (LR-MM-PBSA) of human heparanase
for inhibition and drug design. Proteins 2006, 65, 580–592.
(21) Ander, M.; Luzhkov, V. B.; Aqvist, J. Ligand binding to the voltage-
gated Kv1.5 potassium channel in the open statesdocking and
computer simulations of a homology model. Biophys. J. 2008, 94,
820–831.
(22) May, J. J.; Kessler, N.; Marahiel, M. A.; Stubbs, M. T. Crystal structure
of DhbE, an archetype for aryl acid activating domains of modular
nonribosomal peptide synthetases. Proc. Natl. Acad. Sci. U.S.A. 2002,
99, 12120–12125.
(23) Still, W. C.; Tempczyk, A.; Hawley, R. C.; Hendrickson, T. Semi-
analytical treatment of solvation for molecular mechanics and dynam-
ics. J. Am. Chem. Soc. 1990, 112, 6127–6129.
(24) MacroModel, version 9.1; Schro¨dinger LLC:, New York, 2005.
(25) Kaminski, G. A.; Friesner, R. A.; Tirado-Rives, J.; Jorgensen, W. L.
Evaluation and reparametrization of the OPLS-AA force field for
proteins via comparison with accurate quantum chemical calculations
on peptides. J. Phys. Chem. B 2001, 105, 6474–6487.
(26) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Ab
initio calculation of vibrational absorption and circular dichroism
spectra using density functional force fields. J. Phys. Chem. 1994,
98, 11623–11627.
(27) Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti
correlation-energy formula into a functional of the electron density.
Phys. ReV. B 1988, 37, 785–789.
(28) Labello, N. P.; Ferreira, A. M.; Kurtz, H. A. An augmented effective
core potential basis set for the calculation of molecular polarizabilities.
J. Comput. Chem. 2005, 26, 1464–1471.
(29) Labello, N. P.; Ferreira, A. M.; Kurtz, H. A. Correlated, relativistic,
and basis set limit molecular polarizability calculations to evaluate
an augmented effective core potential basis set. Int. J. Quantum Chem.
2006, 106, 3140–3148.
(30) Khandelwal, A.; Balaz, S. QM/MM linear response method distin-
guishes ligand affinities for closely related metalloproteins. Proteins
2007, 69, 326–339.
(31) Khandelwal, A.; Lukacova, V.; Comez, D.; Kroll, D. M.; Raha, S.;
Balaz, S. A combination of docking, QM/MM methods, and MD
simulation for binding affinity estimation of metalloprotein ligands.
J. Med. Chem. 2005, 48, 5437–5447.
(32) Hehre, W. J.; Ditchfield, R.; Pople, J. A. Self consistent molecular
orbital methods. XII. Further extensions of Gaussian-type basis sets
for use in molecular orbital studies of organic molecules. J. Chem.
Phys. 1972, 56, 2257–2261.
(13) Somu, R. V.; Boshoff, H.; Qiao, C.; Bennett, E. M.; Barry, C. E.,
3rd; Aldrich, C. C. Rationally designed nucleoside antibiotics that
inhibit siderophore biosynthesis of Mycobacterium tuberculosis.
J. Med. Chem. 2006, 49, 31–34.
(14) Gupte, A.; Boshoff, H. I.; Wilson, D. J.; Neres, J.; Labello, N. P.;
Somu, R. V.; Xing, C.; Barry, C. E., 3rd; Aldrich, C. C. Inhibition of
siderophore biosynthesis by 2-triazole substituted analogues of 5′-O-
[N-(salicyl)sulfamoyl]adenosine: antitubercular nucleosides effective
against Mycobacterium tuberculosis. J. Med. Chem., in press.
(15) Vannada, J.; Bennett, E. M.; Wilson, D. J.; Boshoff, H. I.; Barry, C. E.,
3rd; Aldrich, C. C. Design, synthesis, and biological evaluation of
beta-ketosulfonamide adenylation inhibitors as potential antitubercular
agents. Org. Lett. 2006, 8, 4707–4710.
(16) Aqvist, J.; Luzhkov, V. B.; Brandsdal, B. O. Ligand binding affinities
from MD simulations. Acc. Chem. Res. 2002, 35, 358–365.
(17) Brandsdal, B. O.; Osterberg, F.; Almlof, M.; Feierberg, I.; Luzhkov,
V. B.; Aqvist, J. Free energy calculations and ligand binding. AdV.
Protein Chem. 2003, 66, 123–158.
(33) The synthesis, characterization, and biological activity of this com-
pound are given in the Supporting Information.
JM800668U