Optimization of azoles as anti-human immunodeficiency virus agents guided by free-energy calculations

Article abstract of DOI:10.1021/ja8019214

Efficient optimization of an inactive 2-anilinyl-5-benzyloxadiazole core has been guided by free energy perturbation (FEP) calculations to provide potent non-nucleoside inhibitors of human immunodeficiency virus (HIV) reverse transcriptase (NNRTIs). An FEP "chlorine scan" was performed to identify the most promising sites for substitution of aryl hydrogens. This yielded NNRTIs 8 and 10 with activities (EC₅₀) of 820 and 310 nM for protection of human T-cells from infection by wild-type HIV-1. FEP calculations for additional substituent modifications and change of the core heterocycle readily led to oxazoles 28 and 29, which were confirmed as highly potent anti-HIV agents with activities in the 10-20 nM range. The designed compounds were also monitored for possession of desirable pharmacological properties by use of additional computational tools. Overall, the trends predicted by the FEP calculations were well borne out by the assay results. FEP-guided lead optimization is confirmed as a valuable tool for molecular design including drug discovery; chlorine scans are particularly attractive since they are both straightforward to perform and highly informative.

Full text of DOI:10.1021/ja8019214

Published on Web 06/28/2008
Optimization of Azoles as Anti-Human Immunodeficiency
Virus Agents Guided by Free-Energy Calculations
Jacob G. Zeevaart,† Ligong Wang,†,‡ Vinay V. Thakur,† Cheryl S. Leung,†
Julian Tirado-Rives,† Christopher M. Bailey,‡ Robert A. Domaoal,‡
Karen S. Anderson,‡ and William L. Jorgensen*,†
Department of Chemistry, Yale UniVersity, New HaVen, Connecticut 06520-8107, and
Department of Pharmacology, Yale UniVersity School of Medicine,
New HaVen, Connecticut 06520-8066
Received March 14, 2008; E-mail: william.jorgensen@yale.edu
Abstract: Efficient optimization of an inactive 2-anilinyl-5-benzyloxadiazole core has been guided by free
energy perturbation (FEP) calculations to provide potent non-nucleoside inhibitors of human immunode-
ficiency virus (HIV) reverse transcriptase (NNRTIs). An FEP “chlorine scan” was performed to identify the
most promising sites for substitution of aryl hydrogens. This yielded NNRTIs 8 and 10 with activities (EC50)
of 820 and 310 nM for protection of human T-cells from infection by wild-type HIV-1. FEP calculations for
additional substituent modifications and change of the core heterocycle readily led to oxazoles 28 and 29,
which were confirmed as highly potent anti-HIV agents with activities in the 10-20 nM range. The designed
compounds were also monitored for possession of desirable pharmacological properties by use of additional
computational tools. Overall, the trends predicted by the FEP calculations were well borne out by the assay
results. FEP-guided lead optimization is confirmed as a valuable tool for molecular design including drug
discovery; chlorine scans are particularly attractive since they are both straightforward to perform and highly
informative.
Introduction
that have to be synthesized and assayed to yield a drug candidate.
Our approach features creation and evaluation of virtual libraries,
In our pursuit of new anti-human immunodeficiency virus
(HIV) agents, we have focused on non-nucleoside inhibitors of
HIV reverse transcriptase (NNRTIs). Three drugs in this class
arose in the 1990s (nevirapine, delavirdine, and efavirenz) and
in January 2008 a new NNRTI, known as etravirine or TMC125,
was approved.1 NNRTIs are true inhibitors, which bind to an
allosteric pocket in the vicinity of RT’s polymerase active site.2
A well-known problem with anti-HIV chemotherapy is the rapid
mutation of the virus to yield drug-resistant strains. The
occurrence of undesirable side effects is also often problematic
and the onset may be immediate or arise after extended
treatment. In either event, the availability of alternative thera-
peutic options is essential. Thus, continual anti-HIV drug
discovery efforts are required to yield new drugs with alternative
pharmacological characteristics as well as activity against the
current and future spectrum of variants.1
estimation of pharmacological properties, and lead optimization
guided by free-energy perturbation (FEP) calculations to assess
relative protein-ligand binding affinities.3 Potent, structurally
diverse anti-HIV agents have been discovered.3
In a recent report, leads were sought by docking a library of
commercially available compounds into the NNRTI binding
site.4 Though known NNRTIs were retrieved well, purchase and
assaying of representative, top-scoring compounds from the
library failed to yield any active anti-HIV agents. Persisting,
the highest-ranked library compound 1 was pursued computa-
tionally to seek constructive modifications. Specifically, the
substituents were removed to yield the anilinylbenzyloxadiazole
core, 2. A set of small substituents was reintroduced in place
of each hydrogen; scoring with the BOMB program and FEP
results led to synthesis and assaying of several polychloro
analogues with EC50 values as low as 310 nM in an HIV-
Simultaneously, we have sought improved computational meth-
ods of general utility to streamline the discovery of therapeutic
agents that are both potent and have auspicious pharmacological
properties. The basic goal is to minimize the number of compounds
(3) (a) Jorgensen, W. L.; Ruiz-Caro, J.; Tirado-Rives, J.; Basavapathruni,
A.; Anderson, K. S.; Hamilton, A. D. Bioorg. Med. Chem. Lett. 2006,
16, 663–667. (b) Ruiz-Caro, J.; Basavapathruni, A.; Kim, J. T.; Wang,
L.; Bailey, C. M.; Anderson, K. S.; Hamilton, A. D.; Jorgensen, W. L.
Bioorg. Med. Chem. Lett. 2006, 16, 668–671. (c) Thakur, V. V.; Kim,
J. T.; Hamilton, A. D.; Bailey, C. M.; Domaoal, R. A.; Wang, L.;
Anderson, K. S.; Jorgensen, W. L. Bioorg. Med. Chem. Lett. 2006,
16, 5664–5667. (d) Kim, J. T.; Hamilton, A. D.; Bailey, C. M.;
Domaoal, R. A.; Wang, L.; Anderson, K. S.; Jorgensen, W. L. J. Am.
Chem. Soc. 2006, 128, 15372–15373.
† Department of Chemistry.
‡ Department of Pharmacology.
(1) For reviews, see (a) 2007 AIDS Epidemic Update; UNAIDS: Geneva,
2007. (b) Flexner, C. Nat. ReV. Drug DiscoVery 2007, 6, 959–966.
(c) De Clerq, E. Nat. ReV. Drug DiscoVery 2007, 6, 1001–1018.
(2) (a) Kohlstaedt, L. A.; Wang, J.; Friedman, J. M.; Rice, P. A.; Steitz,
T. A. Science 1992, 256, 1783–1790. (b) Smerdon, S. J.; Ja¨ger, J.;
Wang, J.; Kohlstaedt, L. A.; Chirino, A. J.; Friedman, J. M.; Rice,
P. A.; Steitz, T. A. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 3911–
3915.
(4) Barreiro, G.; Guimara˜es, C. R. W.; Tubert-Brohman, I.; Lyons, T. M.;
Tirado-Rives, J.; Jorgensen, W. L. J. Chem Inf. Model. 2007, 47, 2416–
2428.
9
9492 J. AM. CHEM. SOC. 2008, 130, 9492–9499
10.1021/ja8019214 CCC: $40.75
2008 American Chemical Society

Optimization of Azoles as Anti-HIV Agents
A R T I C L E S
infected T-cell assay.5 The present report documents the FEP-
guided developments that have now led to analogues of 2 with
potencies in the 10-20 nM range. The essentially exhaustive
FEP-guided lead optimization can serve as a model for future
applications.
30M configurations of averaging. For the bound calculations, the
equilibration period was again at least 10M configurations followed
by 20M configurations of averaging for the DW simulations and
10M for the OS alternative. In a recent comprehensive comparison
for perturbations between substituted benzenes in water, the
accuracy and precision from the 14-window DW and 11-window
OS protocols were similar.10b Thus, the intention here was to expand
the testing to protein-ligand cases and to explore the possibility
of halving the averaging period for the bound OS calculations (the
most time-consuming step). If successful, the overall time for the
FEP calculations would also be cut roughly in half, shortening the
total time from about 2 weeks (for 14-DW) to 1 week (for 11-OS)
to obtain a computed change in free energy of binding between
two ligands by use of a 2.4 GHz Pentium processor. With
commitment of 13 processors, one each for the 11 bound windows
and 2 for the unbound calculations, a predicted change in binding
affinity could then be obtained in 1 day.
Computational Details
FEP calculations were carried out in the context of Monte Carlo
(MC) statistical mechanics simulations to predict relative free
energies of binding. The MC/FEP calculations are performed to
interconvert two ligands unbound in water and bound to the protein;
standard protocols were followed.3a Briefly, initial structures were
generated with the molecule growing program BOMB starting from
the PDB file 1s9e;6 the ligand was removed and replaced by cores
such as ammonia or 2 that are used by BOMB to grow the desired
analogues in the binding site.5 A reduced model of the protein was
utilized that consisted of the ca. 175 amino acid residues closest to
the NNRTI binding site; a few remote side chains were neutralized
so that there was no net charge for the protein. The MC/FEP
calculations are executed with MCPRO,7 which also adds 1250
and 2000 water molecules in 25-Å caps for the complexes and
unbound ligands, respectively. The energetics for the systems are
described classically with the OPLS-AA force field for the protein,
OPLS/CM1A for the ligands, and TIP4P for water molecules.8 For
the MC simulations, all degrees of freedom were sampled for the
ligand, while the TIP4P water molecules only translated and rotated,
as usual; bond angles and dihedral angles for protein side chains
were also sampled, while the backbone was kept fixed after
conjugate-gradient relaxation.
The present study included some methodological testing of FEP
protocols. The calculations used either double-wide (DW) sampling9
with 14 windows or simple overlap (OS) sampling with 11
windows.10 A window refers to a MC simulation at one point along
the mutation coordinate λ, which interconverts two ligands as λ
goes from 0 to 1; two free-energy changes are computed at each
window, corresponding to a forward and backward increment. The
spacing between windows, ∆λ, is 0.1 except with 14 windows the
spacing is 0.05 for λ ) 0-0.2 and 0.8-1, which addresses the fact
that the free energy often changes most rapidly in these regions.
Each window for the unbound ligands in water consisted of at least
10 × 106 (10M) configurations of MC equilibration followed by
All MC simulations were run at 298 K. The reported uncertainties
((1σ) for the free energy changes were obtained from the fluctuation
in separate averages over batches of 2M configurations.11 Equation 1
is used, where m is the number of batches, θi is the average of property
θ for the ith batch, and θ is the overall average for θ.
m
σ2 )
(θ - θ )2/m(m - 1)
(1)
∑
i
i
Experimental Details
The synthesized compounds were primarily oxadiazole and
oxazole derivatives. 1,3,4-Oxadiazole-2-amines were prepared either
via cyclization of phenylacetic hydrazides and phenyl isocyanide
dichlorides, as described previously,5 or via cyclization of the
hydrazidecarboxamide by heating with POCl3. Representative
examples are provided in Schemes 1 and 2.
1,3,4-Oxadiazole-2,5-diamines were prepared in a similar manner
via a hydrazine-1,2-dicarboxamide intermediate starting from the
substituted phenylisocyanates (Scheme 3).12
Finally, oxazoles were synthesized as shown in Scheme 4 by minor
modification of Froyen’s procedure.13 Substituted 2-azidoacetophe-
nones, which are readily obtained from corresponding arylacetic acids,
were converted in situ to the iminophosphoranes. Following condensa-
tion with various arylisothiocyanates to yield ꢀ-keto carbodiimides,
the desired oxazoles emerged in good yield. Additional synthetic and
spectral details are provided in Supporting Information.
For the biology, the primary assay determined activities against
the wild-type IIIB strain of HIV-114 by use of MT-2 human
T-cells15 at a multiplicity of infection (MOI) of 0.1; EC50 values
are obtained as the dose required to achieve 50% protection of the
infected cells by the MTT colorimetric method. CC50 for inhibition
of MT-2 cell growth by 50% is obtained simultaneously.16
Analogous assays were performed with variant strains of the virus
(5) Barreiro, G.; Kim, J. T.; Guimara˜es, C. R. W.; Bailey, C. M.; Domaoal,
R. A.; Wang, L.; Anderson, K. S.; Jorgensen, W. L. J. Med. Chem.
2007, 50, 5324–5329.
(6) Himmel, D. M.; Das, K.; Clark, A. D., Jr.; Hughes, S. H.; Benjahad,
A.; Oumouch, S.; Guillemont, J.; Coupa, S.; Poncelet, A.; Csoka, I.;
Meyer, C.; Andries, K.; Nguyen, C. H.; Grierson, D. S.; Arnold, E.
J. Med. Chem. 2005, 48, 7582–7591.
(11) Allen, M. P.; Tildesley, D. J. Computer Simulations of Liquids;
Clarendon: Oxford, 1987.
(12) Gehlen, H.; Moeckel, K. Liebigs Ann. Chem. 1965, 685, 176–180.
(13) Froyen, P. Phosphorus, Sulfur Silicon Relat. Elements 1991, 60, 81–
84.
(14) (a) Popovic, M.; Read-Connole, E.; Gallo, R. C. Lancet 1984, 2, 1472–
1473. (b) Popovic, M.; Sarngadharan, M. G.; Read, E.; Gallo, R. C.
Science 1984, 224, 497–500. (c) Ratner, L.; et al. Nature 1985, 313,
277–284.
(15) (a) Haertle, T.; Carrera, C. J.; Wasson, D. B.; Sowers, L. C.; Richmann,
D. D.; Carson, D. A. J. Biol. Chem. 1988, 263, 5870–5875. (b) Harada,
S.; Koyanagi, Y.; Yamamoto, N. Science 1985, 229, 563–566.
(16) (a) Lin, T. S.; Luo, M. Z.; Liu, M. C.; Pai, S. B.; Dutschman, G. E.;
Cheng, Y. C. Biochem. Pharmacol. 1994, 47, 171–174. (b) Ray, A. S.;
Yang, Z.; Chu, C. K.; Anderson, K. S. Antimicrob. Agents Chemother.
2002, 46, 887–891.
(7) Jorgensen, W. L.; Tirado-Rives, J. J. Comput. Chem. 2005, 26, 1689–
1700.
(8) (a) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. J. Am. Chem.
Soc. 1996, 118, 11225–11236. (b) Jorgensen, W. L.; Tirado-Rives, J.
Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 6665–6670. (c) Jorgensen,
W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L.
J. Chem. Phys. 1983, 79, 926–935.
(9) Jorgensen, W. L.; Ravimohan, C. J. Chem. Phys. 1985, 83, 3050–
3054.
(10) For recent reviews, see (a) Chipot, C.; Pohorille, A. In Free Energy
Calculations: Theory and Applications in Chemistry and Biology;
Chipot, C. , Pohorille, A., Eds.; Springer Series in Chemical Physics,
Vol. 86; Springer-Verlag: Berlin, 2007; pp 33-75. (b) Jorgensen,
W. L.; Thomas, L. T. J. Chem. Theory Comput. 2008, 4, 869-876.
9
J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008 9493

A R T I C L E S
Zeevaart et al.
Scheme 1. Synthesis of 1,3,4-Oxadiazole-2-amines
Scheme 2. Alternative Preparation of 1,3,4-Oxadiazole-2-amines
Scheme 3. Synthesis of 1,3,4-Oxadiazole-2,5-diamines
Scheme 4. Synthesis of 2,5-Oxazole Derivatives
that encode the Tyr181Cys (Y181C)17 and Lys103Asn/Y181C
(K103N/Y181C)18 mutant forms of HIV-RT.
hydrogen conversion. A positive value indicates that chlorine
is preferred over hydrogen. The structure in Table 1 also
corresponds to the structure in Figure 1; in the FEP calculations
C2 and C6 are not equivalent as they do not interconvert during
the MC simulations.
Results
Chlorine Scan. Given the false positive 1, a chlorine scan
was performed on the core 2 to seek auspicious sites for
substitution.5 Thus, FEP calculations were performed for the
interconversion of each hydrogen in the phenyl rings of 2 with
a chlorine. The predicted structure of the core bound to RT is
illustrated in Figure 1; it was built by use of BOMB and then
subjected to conjugate gradient optimization by MCPRO.
Confidence in the computed structure comes from past experi-
ence with complexes for other NNRTIs, including numerous
crystal structures.2–6 Key features are the placement of the
benzyl ring in the π-box, formed by Tyr181, Tyr188, Phe227,
and Trp229, and the hydrogen bond between the oxygen of
Lys101 and the amino group of 2. The 10 complexes for the
monochloro derivatives of 2 were also built with BOMB and
each was converted to 2 in the FEP calculations. The FEP-DW
results are listed in Table 1 with ∆∆Gb representing the
predicted change in free energy of binding for the chlorine to
(17) Richman, D.; Shih, C.-K.; Lowy, I.; Rose, J.; Prodanovic, P.; Goff,
S.; Griffin, J. Proc. Natl. Acad. Sci. U.S.A. 1991, 88, 11241–11245.
(18) Nunberg, J. H.; Schleif, W. A.; Boots, E. J.; O’Brien, J. A.; Quintero,
J. C.; Hoffman, A. M.; Emini, E. A.; Goldman, M. E. J. Virol. 1991,
65, 4887–4892.
Figure 1. Some key residues in the NNRTI binding site with the core 2
bound, as created by BOMB and optimized by MCPRO. The full model
contains ca. 175 protein residues and the ligand.
9
9494 J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008

Optimization of Azoles as Anti-HIV Agents
A R T I C L E S
Table 1. FEP-DW Results for Replacements of Chlorine by
Hydrogen
ring for the 2′,6′-dichloro analogue of 2. In this case, adding a
third chlorine at C4 should still be highly favorable and adding a
chlorine at C3 (C5) also remains promising. A chlorine at C2 (C6)
is not likely to be beneficial.
The related experimental data for the chlorine scan are provided
by compounds 2-10 in Table 4. The results for 2-5, 8, and 10
were reported previously.5 Perfect correlation between the whole-
cell assay results and the computed changes in free energies of
binding is not expected. Direct correlation with the FEP results
would require an unavailable binding assay yielding Kd values.
However, good correlations have been observed consistently for
NNRTI series in assays for inhibition of reverse transcription and
in cell-based anti-HIV analyses.19 For lead optimization, the critical
element is that the computations need to be qualitatively correct
in providing guidance toward enhanced activity. The message from
the FEP results was that it should be most beneficial to pursue
chlorination at C4, C2′, C6′, and possibly C3.
The false positive 1, the core 2, and 3-5, which do not have
the 2′,6′-dichloro substitution, are all inactive (Table 4).
However, the results for the series 6-8 nicely conform to the
pattern in Tables 1–3; the 2,2′,6′-trichloride 6 is still inactive,
while moving the 2-chloro substituent to C3 and C4 does yield
active compounds with EC50 values of 4.3 and 0.82 µM. The
accord is upheld for the tetrachlorides 9 and 10, where again
addition of chlorine at C3 is more beneficial than at C2. Thus,
the recommendations from the FEP results in Tables 1–3 for
optimal addition of three and four chlorines directly lead to
NNRTIs 8 and 10 with activities of 820 and 310 nM. Compound
11 with a methyl group at C3 was also prepared and found to
be less potent than the chlorine analogue 10.
Cl f H
∆Gbound (kcal/mol)
∆Gun (kcal/mol)
∆∆Gba (kcal/mol)
σb
C2
C3
C4
C5
C6
C2′
C3′
C4′
C5′
C6′
2.71
1.33
6.62
-0.54
-6.86
4.86
-4.01
1.01
1.71
4.89
0.51
-2.00
2.42
-1.71
-4.26
1.43
0.69
1.81
0.80
1.12
2.21
3.33
4.20
1.17
-2.60
3.43
-4.70
-0.80
0.91
0.10
0.11
0.09
0.10
0.12
0.08
0.11
0.09
0.08
0.08
3.77
a Computed change in free energy of binding from FEP-DW
calculations. b Statistical uncertainties computed from eq 1.
The FEP-DW results indicate that the most promising places
for chlorines are at C3, C4, C2′, and C6′. The most desirable
situation is when equivalent positions, for example, C2′ and
C6′, are both favorable. If only one is favorable, then a penalty
of RT ln 2 (0.6 kcal/mol) can be expected for loss of one
rotameric state upon binding. This is not an issue at a
symmetrical position, such as C4 and C4′. Thus, the message
from the FEP-DW results is that the placement of chlorines at
C4, C2′, and C6′ should be particularly favorable and substitu-
tion at C3 (C5) would be the next most favorable.
Why a chlorine scan? Addition of a single non-hydrogen atom
is desirable since it minimizes potential steric conflicts. The other
obvious possibilities are then a fluorine or methyl scan. The
methyl scan is a good idea since the electronic character is
different, electron-donating instead of electron-withdrawing.
Consequently, there will be cases in which methyl is more
activating than chlorine, and it is generally desirable to perform
both FEP scans. If predicted free energies of binding for both
series are obtained relative to the unsubstituted core, then ideal
sites for introduction of both groups would be revealed.
However, if only one scan is to be performed, chlorine has the
technical advantage of being the most straightforward since
the methyl to hydrogen perturbation requires the creation of
three dummy atoms. Other considerations that can favor chlorine
are that aryl chlorides are less prone to metabolism than methyl
analogues and, synthetically, chlorine can provide a handle for
coupling reactions. The problem with fluorine is that it is
smaller; computed solvent-accessible surface areas with the
OPLS-AA force field are 266, 275, 290, and 298 Å2 for benzene,
fluorobenzene, chlorobenzene, and toluene. Thus, a chlorine and
a methyl group have similar size, while fluorine is more like a
hydrogen in size and its successful introduction does not
guarantee that a chlorine or methyl group could be accommodated.
The calculations were repeated with the FEP-OS methodol-
ogy, as summarized in Table 2. The results agree well with
those in Table 1, so this time-saving alternative is promising,
and the message for the most favorable substitutions remains
the same. The largest discrepancy occurs for C5, with the FEP-
DW prediction being slightly favorable while the FEP-OS result
is slightly unfavorable. With the FEP-OS calculations, the
perturbations were also carried out in the gas phase. This allows
computation of the change in free energy of hydration as ∆Gun
- ∆Ggas in Table 2. Each monochloro analogue is found to be
more hydrophobic than 2 by 1-2 kcal/mol; in both the gas-
phase and aqueous MC simulations, the average dipole moments
are typically about 1 D less for the monochloro analogues than
the parent. The accord for equivalent positions now reflects
interconversion by rotation in the unbound aqueous and gas-
phase simulations. It should be noted that the results in Tables
1 and 2 are far from obvious if one just displays the computed
structures for the 10 monochloro analogues of 2 bound to RT.
All of the structures appear reasonable with the possible
exception of the C6 analogue, which seems to have the chlorine
and benzyl ring crowded together. One might then execute
conjugate gradient optimizations on the 10 complexes and the
core 2. With a dielectric constant of 2, the resultant protein-ligand
interaction energies are -46 (C5), -47 (C3′), -49 (C6), -50
(C3), -51 (C5′), -52 (C2′ and C4′), -54 (C2 and C6′), and
-55 (C4) kcal/mol for the monochlorides and -50 kcal/mol
for the unsubstituted core. The correlation of these results with
those in Tables 1 and 2 is weak with the exception of the
optimism for chlorination at C4 and C6′.
Optimization of the C4 Substituent. At this point, further
refinements were probed by FEP calculations, starting with the
substituent at C4. Analogous calculations had been carried out for
the NNRTI series in Scheme 5 to optimize the substituent X.3a–c
The present case is reminiscent of the thiazole-containing example,
though the presence of the dimethylallyloxy substituent at C5 in
A final potential concern for design is addressed in Table 3.
Perhaps the predictions would change as the core is progressively
chlorinated. Table 3 records the results for substitution in the phenyl
(19) Rizzo, R. C.; Udier-Blagovic, M.; Wang, D. P.; Watkins, E. K.;
Kroeger Smith, M. B.; Smith, R. H., Jr.; Tirado-Rives, J.; Jorgensen,
W. L. J. Med. Chem. 2002, 45, 2970–2987.
9
J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008 9495

A R T I C L E S
Zeevaart et al.
Table 2. Overlap Sampling Results for Replacements of Chlorine by Hydrogen for Core 2
a
Cl f H
∆Gbound (kcal/mol)
∆Gun (kcal/mol)
∆Ggas (kcal/mol)
∆∆Gaq (kcal/mol)
∆∆Gbb (kcal/mol)
σc
C2
C3
C4
C5
C6
C2′
C3′
C4′
C5′
C6′
3.37
1.06
6.47
-2.04
-6.67
3.98
-3.80
0.15
1.73
5.08
0.62
-1.94
2.20
-1.47
-4.26
1.01
0.67
1.94
0.67
0.88
2.57
-0.57
3.53
-0.50
-1.88
2.05
1.44
2.75
1.62
1.71
-1.95
-1.37
-1.33
-0.97
-2.38
-1.04
-0.77
-0.81
-0.95
-0.83
2.75
3.00
4.27
-0.57
-2.41
2.97
-4.47
-1.79
1.06
0.13
0.14
0.11
0.13
0.14
0.11
0.13
0.12
0.10
0.09
4.20
a Computed change in free energy of hydration from FEP-OS calculations. b Computed change in free energy of binding from FEP-OS calculations.
c Statistical uncertainties computed from eq 1.
Table 3. FEP-DW Results for Replacements of Chlorine by
Hydrogen
Scheme 5. Thiazole- and Pyrimidine-Containing NNRTIs3
Table 5. FEP-DW Results for Optimization of the 4-Substituent X
Cl f H
∆Gbound (kcal/mol)
∆Gun (kcal/mol)
∆∆Gba (kcal/mol)
σb
C2
C3
C4
C5
C6
2.51
1.42
6.88
1.15
-8.77
0.95
-1.38
2.38
-1.23
-2.60
1.57
2.81
4.50
2.38
-6.17
0.11
0.10
0.08
0.13
0.13
a Computed change in free energy of binding from FEP-DW
calculations. b Statistical uncertainties computed from eq 1.
X f
Y
∆Gbound (kcal/mol) ∆Gun (kcal/mol) ∆∆Gba (kcal/mol) σb
Table 4. Oxadiazoles: Anti-HIV-1 Activity (EC50) and Cytotoxicity
CN
Cl
F
Cl
Cl
F
H
H
H
1.53
1.22
5.56
6.99
5.49
0.34
-0.54
3.27
2.23
3.88
-1.93
-4.91
-10.63
-0.63
1.19
1.75
2.29
4.76
1.61
-1.34
-1.55
0.62
0.13
0.05
0.05
0.06
0.12
0.13
0.16
0.08
0.17
(CC50
)
CH3
CH3CH2 CH3
CH3CH2 OCH3
CF3
-3.27
-6.46
-10.01
0.79
CH3
CH3OCH2 CH3CH2
1.41
a Computed change in free energy of binding from FEP-DW
calculations. b Statistical uncertainties computed from eq 1.
a
b
EC50
(µM)
CC50
(µM)
compd
R1
R2
R3
R4
1
2
3
4
5
6
7
8
4-CH3
H
4-Cl
4-Cl
4-Cl
2-Cl
3-Cl
4-Cl
4-Cl
4-Cl
4-Cl
4-CH2OCH3
4-CN
4-CN
H
H
H
H
H
H
H
H
3′-OCH3 4′-OCH3 NA
61
>100
31
>100
32
>100
71
20
45
>100
>100
100
40
90
>100
>10
short, the only substituent that is predicted to be preferred over
chlorine is a cyano group, and the relative free energies of binding
(in kilocalories per mole) are H (0.0) > Et (-0.3) > Me (-1.6) >
methoxymethyl (-1.7) > OMe (-1.8) > CF3 (-2.2) > F (-2.3)
> Cl (-4.0) > CN (-5.2). As shown in Table 4, the methoxym-
ethyl (MOM) and cyano derivatives 12 and 13 were prepared and
bracket the chloro analogue 8 in activity, as predicted. The activity
has progressed to 130 nM with 13 from 820 nM for 8. For
comparison, it is noted in Table 4 that the nucleoside d4T is a 1.4
µM inhibitor in the MT-2 assay, while nevirapine and efavirenz
are 110-nM and 2-nM NNRTIs.
The 2′,6′-difluoro compound 14 was also prepared and is
somewhat less potent at 230 nM than 13. Our interest in the
difluoro analogue arose from prediction of relatively poor
solubility for 13 by QikProp.20 The predicted intrinsic aqueous
solubility increases from 10-5.8 M for 13 to 10-5.3 M for 14.
As noted previously, analysis of 1700 oral drugs with QikProp
shows that 90% have predicted solubilities above 10-5.7 M.3c
H
H
H
H
H
H
6′-Cl
6′-Cl
6′-Cl
6′-Cl
6′-Cl
6′-Cl
6′-Cl
6′-Cl
6′-F
NA
NA
NA
>10
NA
4.3
0.82
2.4
0.31
1.3
4′-Cl
5′-Cl
2′-Cl
2′-Cl
2′-Cl
2′-Cl
2′-Cl
9
2-Cl
3-Cl
3-CH3 2′-Cl
H
H
H
10
11
12
13
14
2′-Cl
2′-Cl
2′-F
4.3
0.13
0.23
1.4
d4T
nevirapine
efavirenz
0.11
0.002 >0.1
a For 50% protection in MT-2 cells; antiviral curves used triplicate
samples at each concentration. NA for EC50
CC50. b For 50%
inhibition of MT-2 cell growth; toxicity curves also used triplicate
samples.
>
place of the benzyl group in the oxadiazole series could affect the
preferences for X. The FEP results are recorded in Table 5. In
(20) Jorgensen, W. L. QikProp, v 3.0; Schro¨dinger LLC: New York,
2006.
9
9496 J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008

Optimization of Azoles as Anti-HIV Agents
A R T I C L E S
Table 6. Summary of FEP-DW Results for Optimization of the
Linker X
Table 7. Benzylic Analogues: Anti-HIV-1 Activity (EC50) and
Cytotoxicity (CC50
)
a
a
X
∆∆Gb (kcal/mol)
σ
compd
R
R′
X
Y
EC50 (µM)
CC50 (µM)
CH2
(R)-CHCH3
(S)-CHCH3
NH
NCH3
O
0
0
8
Cl
Cl
Cl
Cl
Cl
Cl
Cl
F
Cl
Cl
Cl
CN
CN
CN
CN
CN
CH2
NH
CH2
CH2
NH
NH
CH2
NH
NH
NH
NH
NH
0.82
>100
18
0.13
9.8
0.27
1.20
0.13
20
>100
43
0.50
2.80
1.20
-1.60
1.37
-1.40
0.21
0.18
0.21
0.29
0.16
0.19
15
16
13
17
18
19b
20b
40
NH
>100
>100
54
NCH3
CHCH3
CHCH3
S
8.5
a See footnotes in Table 4. b Racemate.
On the other hand, the predicted octanol/water partition coef-
ficients, QP log P, of 3.5 and 3.1 for 13 and 14 are both well
below the 90% limit of 5.0.3c
17 compared to 13. Replacement of the pro-R hydrogen with a
methyl group is predicted to be 2.3 kcal/mol more favorable
than the pro-S substitution but still 0.5 kcal/mol less favorable
than for the methylene version, 8. Indeed, 19, though racemic,
loses a factor of 9 in potency compared to 13, while curiously
the difluoro analogue 20 is a little more potent than 14 (Table
4) and nearly 10-fold more potent than 19. The FEP calculations
were run only for the 2′,6′-dichlorides, so there may be some
differences in conformational details that are responsible for
the varying effects of the addition of the benzylic methyl group.
The principal hope was for the methylamino analogue, which
was predicted to be 2.1 kcal/mol better bound than the (R)-
CHCH3 isostere and 1.6 kcal/mol better bound than the
methylene analogue. However, the one related compound that
was prepared, 18, turned out to be a factor of 2 less active than
its methylene analogue, 13. In view of this and the results in
Table 6, enthusiasm could not be mustered for pursuing the
oxa and thia (X ) O and S) analogues.
The preference for a cyano group at C4 was also found for
the thiazole and pyrimidines series in Scheme 5, with typical
reductions of a factor of 3-10 in EC50 values over the
corresponding chlorides.3a–c As discussed previously,3b other
NNRTIs also feature a chlorine or cyano group in a similar
position including efavirenz, TMC125, 8-chloro-TIBO, and
UC781; however, the activity data are variable in these series
with no consistent preference. For our series, we attribute the
preference for cyano and chloro over the alternatives above to
(a) their strengthening of the hydrogen bond with Lys101 by
acidification of the hydrogen of the para-amino group and (b)
constructive interaction of the C-Cl and C-CN dipoles with
the electric field at the binding site. Concerning (a), MP2/
6-311+G(d,p) results for hydrogen-bonded complexes of a
water molecule with para-substituted phenols (p-HOPhY) show
striking hydrogen-bond strengthening in the order Y ) NO2 >
CN > CF3 > Cl > F > H > OH > CH3 > NH2.21 For the
inhibitors, the binding to the amide carbonyl of Lys101 is
expected to be more affected by this substituent change than
the binding to a water molecule in the unbound state. Concerning
(b), His235 was included in Figure 1 to illustrate its proximity
to C4 of the inhibitor. Upon conjugate-gradient minimization
of the complex of RT with 13, the distance between the cyano
nitrogen and CR of His235 is only 3.3 Å. Consequently, it is
highly favorable to have a δ+-δ- dipole as provided by the
4-chloro or cyano substituent directed at His235.
As a curiosity, 16 was also prepared to obtain an estimate
for the importance of the putative hydrogen bond between the
amino group of 8 and Lys101; the answer is that there is a 22-
fold loss in activity for deletion of the hydrogen bond. In
summary, modifications were considered for the methylene
linker between the oxadiazole ring and the 2,6-dihalophenyl
group, but none provided significant gains in potency.
Optimization of the Heterocycle. The final target was the
heterocycle. The oxadiazole arose from the docking study,4 but
there was no reason to assume that it was the optimal five-
membered ring heterocycle. Thus, MC/FEP calculations were
performed to convert the thiophene analogue of 8 to 10 other
common possibilities. The structures were all created by BOMB,
and thiophene was taken as the reference because it has the
maximum number of atoms; the FEP calculations are set up
such that atoms are deleted, if appropriate, in progressing from
λ ) 0 to 1. The results are summarized in Table 8, and the
order of predicted increasing activity is for the 1,3,4-thiadiazole,
thiazole, imidazole, furan, isoxazole, thiophene, 1,2,3-triazole,
isothiazole, pyrazole, 1,3,4-oxadiazole, and oxazole. Remark-
ably, only the oxazole analogue was predicted to be more potent
than the oxadiazole. It is expected that the predicted range of
free energies of binding, ∆∆Gb, in Table 8 is exaggerated. This
may arise from the lack of sampling of the protein backbone in
the MC simulations, which can be expected to disfavor the
Optimization of the Benzylic Linker. Attention next turned
to the benzylic CH2 group, designated X in the structure in Table
6. FEP calculations were performed starting from the trichloride
8 for X being CH2, NH, O, S, NCH3, and (R)- and (S)-CHCH3.
Multiple FEP cycles were considered and the final results for
predicted free energies of binding relative to 8 are summarized
in Table 6. In brief, no modification was predicted to be
particularly constructive in comparison to the results from the
previous substituent scans. Some related compounds were
synthesized and assayed, as summarized in Table 7.
Conversion to the amino analogue was computed to be
unfavorable by 1.2 kcal/mol, which was confirmed by the
observed inactivity of 15 compared to 8 and weak activity of
(21) Jorgensen, W. L.; Jensen, K. P.; Alexandrova, A. A. J. Chem. Theory
Comput. 2007, 3, 1987–1992.
9
J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008 9497

A R T I C L E S
Zeevaart et al.
Table 8. FEP-DW Results for Optimization of the Heterocycle
W
X
Y
Z
∆Gbound (kcal/mol) ∆Gun (kcal/mol) ∆∆Gba (kcal/mol)
σb
S
S
S
CH
O
CH
CH
CH
CH
O
C
C
C
N
C
C
N
C
C
C
C
CH CH
0
2.23
0
0
4.91
4.58
1.39
0.69
0.42
-0.15
-0.64
-1.29
-3.47
-6.04
0
N
CH
CH
N
N
N
-2.68
0.47
0.32
0.47
0.22
0.25
0.43
0.52
0.29
0.29
0.22
-7.18
-19.18
-7.55
-22.10
-12.21
-16.80
-29.37
-24.92
-29.65
-11.76
-20.57
-8.24
CH CH
Figure 2. Image constructed from the last configuration of an MC/FEP
calculation for 29 bound to HIV-RT. Carbon atoms of 29 are in yellow.
Fragments of some residues in the binding site are shown along with one
of the 1250 water molecules that were included.
O
N
S
NH
N
N
N
N
N
N
N
-22.52
-12.06
-16.16
-28.08
-21.45
-23.61
O
CH
10-5.6 M, 3.6, and 1054 nm/s. The compound has two predicted
primary metabolites arising from oxidation at the benzylic
methylene group and epoxidation of the oxazole.
a Computed change in free energy of binding from FEP-DW
calculations. b Statistical uncertainties computed from eq 1.
An illustration of the computed structure for 29 bound to RT
is provided in Figure 2. The preference for the oxazole over
the oxadiazole can be attributed to improved protein-ligand
electrostatic interactions. The oxazole nitrogen is more basic
than N3 in the oxadiazole, which should improve the interaction
with the backbone NH of Lys101. In addition, the proximity of
N4 of the oxadiazole and the side-chain carboxylate of Glu138
is expected to be unfavorable. It is also noted that, in the MC
simulations, one or two water molecules are often present in
the binding site situated in the region between the two phenyl
rings of the inhibitors. One is seen in the snapshot in Figure 2;
the water molecules are typically forming π-type hydrogen
bonds with aryl rings of the inhibitor and protein, and when
more than one is present, they are also hydrogen-bonded to each
other. On the other hand, a water molecule is not found
hydrogen-bonded to the nitrogen of any C4-cyano groups in
the bound inhibitors; the key stabilizing interaction in this case
is the charge-dipole interaction with His235 (Figure 1).
Returning to Table 9, it may be noted that the difluoro (29)
over dichloro (28) preference here repeats the pattern for 19
and 20 but not for 13 and 14, again suggesting some confor-
mational subtleties. In addition, to confirm further the structure-
activity predictions for the substituent at C4, the ethyl and fluoro
derivatives 25 and 26 were also synthesized. The observed
activity order of CN > Cl > F > Et for 25-28 exactly mirrors
the predictions in Table 5. As a last thrust, FEP calculations
were performed for possible replacement of the oxazole C4
hydrogen of 28 or 29 by F, Et, Me, CF3, and CH2OH. The
calculations were performed for 29, and the five analogues were
predicted to be less well bound than 29 by 0.8, 1.5, 1.8, 2.2,
and 3.9 kcal/mol, respectively. This is another example where
expectations based on visual inspection of modeled structures,
as in Figure 2, are ambiguous. The qualitative FEP result was
confirmed experimentally for the C4-methyl derivative of 28,
which was found to be 7-fold less potent than 28. Another option
that had been considered was replacement of the aminophenyl
ring in 8 with aminopyridinyl. The phenyl to 2-pyridinyl
perturbation was unfavorable by 0.9 kcal/mol, while 4-pyridinyl
and 3-pyridinyl were favorable, but by only 0.4 and 0.9 kcal/
Table 9. Heterocycle Analogues: Anti-HIV-1 Activity (EC50) and
Cytotoxicity (CC50
)
a
a
compd
R
R′
R′′
W
Y
EC50 (µM)
CC50 (µM)
21
22
23
24
25
26
27
28
29
Cl
Cl
Cl
Cl
Cl
Cl
Cl
Cl
F
Cl
Cl
Cl
CN
Et
F
Cl
CN
CN
H
H
Cl
H
H
H
H
H
H
S
S
S
S
O
O
O
O
O
N
NA
3.1
1.4
0.43
2.8
0.43
0.11
0.022
0.013
>100
29
9.8
>100
>100
12
2.8
2.1
7.4
CH
CH
CH
CH
CH
CH
CH
CH
a See footnotes in Table 4. NA for EC50> CC50.
larger, sulfur-containing heterocycles. The changes in hetero-
cycles are also accompanied by significant changes in the
electrostatic interactions with the ligands; this would benefit
from inclusion of polarization effects, which are not explicitly
represented by fixed charged force fields such as OPLS.
For some testing of the predictions, the thiadiazole 21 was
prepared and indeed found to be inactive (Table 9). The thiazole
22 was also predicted to be inactive. It too was prepared and,
though less active than 8, it did yield an EC50 value of 3.1 µM
and the trichloro version 23 came in at 1.4 µM. However, the
synthetic focus naturally turned to oxazoles, and it was gratifying
that 27 emerged as a 110-nM NNRTI, showing a nice boost
over its oxadiazole counterpart 8 at 820 nM. Modification of
the 4-chloro substituent of 27 to cyano in 28 then provided the
expected additional increment to 22 nM. And, finally, the 2,6-
difluorobenzyloxazole 29 was prepared and found to be a very
potent NNRTI at 13 nM. The predicted properties for 29 from
QikProp are also all in the acceptable ranges for a potential
oral drug;3c the molecular weight is 311, and the predicted
aqueous solubility, log P, and Caco-2 cell permeability are
9
9498 J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008

Optimization of Azoles as Anti-HIV Agents
A R T I C L E S
Table 10. Activity (EC50) and Cytotoxicity (CC50) for Wild-Type and
oxazoles in Table 10, the new compounds exhibit no significant
activity against the two mutant strains of the virus. In fact, when
one looks closely at the data for TMC125 analogues, one sees that
the results for the mutant panel vary substantially for small
substituent changes in compounds with no change to the core
structure or number of rotatable bonds.24 For example, compound
30 is a 0.5 nM NNRTI toward wild-type HIV-1 but loses 150-
fold to the Y181C variant.24 Thus, some flexibility is undoubtedly
needed for resilience, but there are additional subtleties that affect
the resistance profiles. Further analyses and design efforts are in
progress on these issues.
Variant Strains of HIV-1a
WT
Y181C
K103N/Y181C
compd
EC50 (µM) CC50 (µM) EC50 (µM) CC50 (µM) EC50 (µM) CC50 (µM)
13
0.13
0.11
0.022
0.013
1.4
0.002
0.001
40
2.8
2.1
7.4
>100
>0.1
>0.1
21
NA
NA
NA
0.68
0.013
0.014
72
6
4
8
>100
>0.1
ND
24
NA
NA
ND
0.28
0.050
0.005
>100
15
5
ND
>100
>0.1
ND
27
28
29
d4T
efavirenz
TMC125
a See footnotes in Table 4. NA for EC50 > CC50; ND ) not
determined.
Conclusion
mol before correction for the RT ln 2 rotamer loss. So, these
possibilities were also abandoned.
MC/FEP calculations have been used to guide efficient
optimization of the inactive core 2 to yield 28 and 29 as 10-20
nM inhibitors of HIV reverse transcriptase. Key milestones were
provided by the chlorine scan, which proposed 8, optimization
of the C4 substituent to yield 13, and prediction of the preference
for the oxazole 28. Additional modifications for the benzylic
linker, replacement of the aminophenyl ring, and C4-substitution
of the oxazole were also considered and were deemed less
promising. The study illustrates the benefits for inhibitor design
of closely coupled computer modeling, synthetic organic
chemistry, and biological assaying. Future work will focus on
application of the technology to other cores and simultaneous
optimization for wild-type and mutant viral targets.
Viral Mutations
In pursuing optimization for the core 2, potential performance
against mutant strains of HIV-1 was considered. Resistance
typically arises from protein mutations that (a) introduce unfavor-
able interactions with the inhibitor, (b) reduce favorable interactions
with the inhibitor, or (c) stabilize the unliganded (apo) form of the
protein. In response to (a) and (b), it is viewed as desirable to keep
the inhibitors relatively small and to incorporate several rotatable
bonds.22 This is also expressed in the wiggle and jiggle idea that
has been claimed as responsible for the excellent resistance profiles
for the NNRTIs TMC125 (etravirine) and TMC278 (rilpivirine).23
These molecules and the progeny of 2 all feature four rotatable
bonds connecting the three rings.
The Y181C and K103N/Y181C mutants have been particularly
problematic for NNRTI therapy,1 so several of the more potent
NNRTIs from the present work were tested against corresponding
strains of HIV in the infected T-cell assay. The results are
summarized in Table 10 and include reference data for d4T,
efavirenz, and TMC125. The findings for efavirenz and TMC125
are consistent with prior reports using similar assays.24 Here,
efavirenz loses factors of 6.5 and 25 in potency to the Y181C and
K103N/Y181C variants and has a 2-nM EC50 for wild-type HIV-
1, while 2- and 40-fold losses and a 1-nM EC50 for the wild type
were previously reported.24 For TMC125, the prior EC50 values
are 1.4, 7, and 4.3 nM for WT, Y181C, and the double mutant,
while our results are 1, 14, and 5 nM. However, in spite of the
similar expected flexibility for TMC125 and the oxadiazole and
Acknowledgment. Gratitude is expressed to the National
Institutes of Health (AI44616, GM32136, GM49551) for support.
Receipt of the following reagents through the NIH AIDS Research
and Reference Reagent Program, Division of AIDS, NIAID, NIH,
is also greatly appreciated: MT-2 cells, catalog no. 237, and
nevirapine-resistant HIV-1 (N119), catalog no. 1392, from Dr.
Douglas Richman; HTLV-IIIB/H9, no. 398, from Dr. Robert Gallo;
and HIV-1IIIB (A17 variant) from Dr. Emilio Emini.
Supporting Information Available: Synthetic details and
NMR and HRMS spectral data for compounds in Tables 4, 6,
and 8 except 1-5, 8, 10, and 11, for which the data were
previously reported;5 also full citations for refs 14c, 23a, and
24. These materials are available free of charge via the Internet
at http://pubs.acs.org.
(22) Wang, D.-P.; Rizzo, R. C.; Tirado-Rives, J.; Jorgensen, W. L. Bioorg.
Med. Chem. Lett. 2001, 11, 2799–2802.
(23) (a) Das, K.; et al. J. Med. Chem. 2004, 47, 2550–2560. (b) Das, K.;
Bauman, J. D.; Clark, A. D., Jr.; Frenkel, Y. V.; Lewi, P. J.; Shatkin,
A. J.; Hughes, S. H.; Arnold, E. Proc. Natl. Acad. Sci. U.S.A. 2008,
105, 1466–1471.
(24) Ludovici, D. W.; et al. Bioorg. Med. Chem. Lett. 2001, 11, 2235–
2239.
JA8019214
9
J. AM. CHEM. SOC. VOL. 130, NO. 29, 2008 9499