Computational discovery of picomolar Q o site inhibitors of cytochrome bc 1 complex

Article abstract of DOI:10.1021/ja3001908

A critical challenge to the fragment-based drug discovery (FBDD) is its low-throughput nature due to the necessity of biophysical method-based fragment screening. Herein, a method of pharmacophore-linked fragment virtual screening (PFVS) was successfully developed. Its application yielded the first picomolar-range Q_o site inhibitors of the cytochrome bc₁ complex, an important membrane protein for drug and fungicide discovery. Compared with the original hit compound 4 (K_i = 881.80 nM, porcine bc₁), the most potent compound 4f displayed 20 507-fold improved binding affinity (K_i = 43.00 pM). Compound 4f was proved to be a noncompetitive inhibitor with respect to the substrate cytochrome c, but a competitive inhibitor with respect to the substrate ubiquinol. Additionally, we determined the crystal structure of compound 4e (K_i = 83.00 pM) bound to the chicken bc₁ at 2.70 A resolution, providing a molecular basis for understanding its ultrapotency. To our knowledge, this study is the first application of the FBDD method in the discovery of picomolar inhibitors of a membrane protein. This work demonstrates that the novel PFVS approach is a high-throughput drug discovery method, independent of biophysical screening techniques.

Full text of DOI:10.1021/ja3001908

Article
pubs.acs.org/JACS
Computational Discovery of Picomolar Qo Site Inhibitors of
Cytochrome bc1 Complex
Ge-Fei Hao,†,∥ Fu Wang,†,∥ Hui Li,†,‡,∥ Xiao-Lei Zhu,† Wen-Chao Yang,† Li-Shar Huang,§ Jia-Wei Wu,*,‡
Edward A. Berry,§ and Guang-Fu Yang*,†
†Key Laboratory of Pesticide & Chemical Biology, Ministry of Education, College of Chemistry, Central China Normal University,
Wuhan 430079, P. R. China
‡MOE Key Laboratory of Protein Sciences, Tsinghua-Peking Center for Life Sciences, School of Life Sciences, Tsinghua University,
Beijing 100084, P. R. China
§Department of Biochemistry and Molecular Biology, SUNY Upstate Medical University, Syracuse, New York 13210, United States
S
* Supporting Information
ABSTRACT: A critical challenge to the fragment-based drug discovery
(FBDD) is its low-throughput nature due to the necessity of biophysical
method-based fragment screening. Herein, a method of pharmacophore-linked
fragment virtual screening (PFVS) was successfully developed. Its application
yielded the first picomolar-range Qo site inhibitors of the cytochrome bc1
complex, an important membrane protein for drug and fungicide discovery.
Compared with the original hit compound 4 (Ki = 881.80 nM, porcine bc1),
the most potent compound 4f displayed 20 507-fold improved binding affinity
(Ki = 43.00 pM). Compound 4f was proved to be a noncompetitive inhibitor
with respect to the substrate cytochrome c, but a competitive inhibitor with respect to the substrate ubiquinol. Additionally, we
determined the crystal structure of compound 4e (Ki = 83.00 pM) bound to the chicken bc1 at 2.70 Å resolution, providing a
molecular basis for understanding its ultrapotency. To our knowledge, this study is the first application of the FBDD method in
the discovery of picomolar inhibitors of a membrane protein. This work demonstrates that the novel PFVS approach is a high-
throughput drug discovery method, independent of biophysical screening techniques.
Based on inhibition patterns and the crystallographically
observed inhibitor binding sites in the bc1, the existing bc1
inhibitors can be divided into three classes (Figure 1,
Supporting Information):2 (1) class P inhibitors bind at the
Qo site and include azoxystrobin, kresoxim-methyl, famox-
adone, stigmatellin, and UHDBT; (2) class N inhibitors target
the Qi site and include antimycin A and diuron; and (3) class
PN inhibitors can bind to both the Qo and Qi sites and include
NQNO and possibly funiculosin. The most potent bc1 inhibitor
that has been identified is antimycin A, a natural Qi site
inhibitor that binds to bovine heart mitochondrial particles with
a dissociation constant of 32 pM.8 Although Qo site inhibitors
have attracted great interest as antifungal agents, no picomolar
range Qo site inhibitor has yet been reported. Therefore,
discovering new ultrapotent Qo site inhibitors is of great
interest, not only for the functional study of bc1 but also for
potential antifungal applications.
INTRODUCTION
■
High potency that should be balanced with other properties is a
sought-after characteristic of drug molecules. The challenge of
discovering high-potency membrane protein inhibitors is of
great interest, since about 60% of currently marketed drugs
target this type of protein.1 The cytochrome bc1 complex (EC
1.10.2.2, bc1) is an essential and central component of the
cellular respiratory chain and of the photosynthetic apparatus in
photosynthetic bacteria; it catalyzes the electron transfer (ET)
from quinol to a soluble cyt c, with concomitant translocation
of protons across the membrane to generate a proton gradient
and membrane potential for ATP synthesis.2,3 Its critical
importance in life processes makes the bc1 a promising action
target for numerous antiparasitic agents, antibiotics, and
fungicides.4 Three subunits containing prosthetic groups are
essential for ET function: the cyt b subunit bearing two b-type
hemes (bL and bH heme), cyt c1 with a c-type heme, and the
iron−sulfur protein (ISP) having a 2Fe−2S cluster.5,6 The
“proton-motive Q-cycle” is a favored mechanism for electron
and proton transfer in this complex,7 suggesting the existence of
two discrete reaction sites: a quinone reduction site near the
negative side of the membrane (Qi or QN) and a quinol
oxidation site close to the positive side of the membrane (Qo or
QP).
Fragment-based drug discovery (FBDD) has recently been
rapidly developed as an alternative to traditional methods of hit
identification, such as high-throughput screening (HTS).9
Compared with HTS, FBDD has several significant advan-
tages.10 First, compared to larger drug-like molecules, fragments
Received: January 13, 2012
Published: June 12, 2012
© 2012 American Chemical Society
11168
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
usually exhibit higher ligand efficiency (LE), which is defined as
the binding free energy (ΔG) divided by the heavy atom count
(HAC).11,12 Comparison based on LE rather than potency
alone could be useful in determining the potential of drug
candidates. Second, the hit rate from fragment screening is
typically much higher than that observed with HTS13 because
the property space covered by fragments is much broader and
more extensive than that covered by drug-like molecules.14
Third, in most cases, there are more opportunities to optimize
fragments to high quality leads with relatively low molecular
weight and better drug-like properties.
However, FBDD also poses challenges. First, FBDD often
relies on the detection of fragment binding via sensitive
biophysical methods, such as NMR spectroscopy, X-ray
diffraction, isothermal calorimetry (ITC), surface plasmon
resonance, and mass spectrometry,15,16 that require specialized
equipment, personnel with specific expertise, and supporting
informatics infrastructure. Second, due to the much weaker
binding affinity of fragments, excellent solubility is required to
enable screening at high concentrations. Furthermore, large
amounts of purified protein (>10 mg) are always essential,
which are very difficult to achieve in most cases, especially for
membrane proteins.17,18 The low-throughput nature of FBDD
has prompted several attempts to develop a computational
screening method for fragment identification from the much
larger commercially available library.19−21 However, fragment
docking is also very challenging; since fragments may be more
promiscuous in the binding pocket than larger drug-like
molecules, it is very difficult to predict the binding mode of a
fragment and accurately estimate its binding affinity.
Based on the X-ray diffraction structures of bc1 bound to
various methoxyacrylate (MOA)-type inhibitors,2,22 we found
that the conformation of the MOA pharmacophore in the
binding pocket was highly conserved. This observation
prompted us to hypothesize that computational screening of
the side-chain fragment without changing the pharmacophore
should be an effective way to discover new high-potency bc1
inhibitors. In the present work, we developed a new approach
named pharmacophore-linked fragment virtual screening
(PFVS) by integrating the advantages of FBDD and docking
methods. Using this new approach, we successfully discovered a
series of new bc1 inhibitors with picomolar potencies. To our
knowledge, this is the first application of the FBDD method in
the computational discovery of ultrapotent inhibitors of
membrane proteins. To verify the simulated binding mode,
we also determined the crystal structure of chicken bc1 in
complex with one of the picomolar inhibitors.
Figure 1. The workflow of PFVS. During the computational virtual
screening, each fragment from a small library containing 1735
fragments was set to link with the MOA pharmacophore via a sulfur
atom. The energies of the complexes were optimized step by step in a
three-step cycle, and candidates were finally determined.
Compound 1 was successfully synthesized but was eventually
abandoned because this compound and its N-methylated
derivatives were determined to be unstable. The synthesis of
compound 5 was unsuccessful, but its analogue, 5a, was
obtained with a Ki of 94.97 nM against porcine bc1. Further
optimization of compound 5a did not significantly improve its
potency. We successfully synthesized compounds 2−4, with Ki
values against porcine bc1 of 31.10, 316.09, and 881.80 nM,
respectively. Initially, we focused on the structural optimization
of the two more prominent compounds, 2 and 3; however, this
did not result in development of a compound with significantly
improved potency. Unexpectedly, structural optimization of
compound 4, which contained a side-chain moiety of
quinoxaline and exhibited relatively lower inhibitory activity,
resulted in several compounds with greatly improved potency
(see the following text and Figure 3).
We previously established that improving the Ar−Ar
interactions between the side chain of an inhibitor and the
hydrophobic residues in the binding pocket (e.g., Phe274) can
effectively improve the potency of bc1 inhibitors.23 Further-
more, the presence of electron-donating and electron-with-
drawing groups can increase the Ar−Ar dimer interaction
energies, as can increasing the number of heavy atoms in the
dimers.24 The quinoxalinyl group of compound 4 was
surrounded by a hydrophobic pocket formed by the side
chains Phe274, Phe127, Ile146, Pro270, Glu271, Ala277,
Leu294, Met124, and Ile298 (Figure 4a, numbered according
to the cyt b subunit of the bovine bc1). First, we introduced a
hydrophobic methyl group onto the 3-position of the
quinoxaline ring, creating compound 4a; this greatly improved
the potency (Ki = 41.00 nM) as expected due to the increased
Ar−Ar interaction energies between the quinoxalinyl group and
its surrounding hydrophobic residues (Figure 4b). Additionally,
the LE of compound 4a was improved to 0.37, higher than that
of compound 4 (LE = 0.32). On the other side, we replaced the
bridge sulfur atom in compound 4 with an oxygen atom; the
resultant compound 4b also exhibited much higher potency
against porcine bc1 with a Ki value of 51.50 nM and LE of 0.38,
which was likely due to the conformational change-induced
improvement of the Ar−Ar interaction between the quinox-
alinyl ring and the phenyl group of Phe274 (Figure 4c).
Combining the above two strategies led to compound 4c
(Figure 4d), whose potency (Ki = 28.00 nM) was further
improved. Interestingly, introduction of an additional methyl
group onto the 6-position of the quinoxaline ring resulted in
RESULTS
■
Hit Identification through PFVS and Hit Optimization.
The workflow of PFVS is shown in Figure 1. During the
computational screening, each fragment from a small library
containing 1735 fragments was set to link with the MOA
pharmacophore via a sulfur atom, since the nucleophilic
substitution reaction between the E-methyl 2-(2-chloromethyl-
phenyl)-3-methoxyacrylate pharmacophore and the mercapto-
containing fragment is highly active and easy to perform. After
performing a 3-step computational screening (see the
Experimental Section), we obtained 10 hits with the most
favorable binding free energies (Table 1, Supporting
Information). By setting the criterion of LE over 0.28, five
candidates (Figure 2) were selected for chemical synthesis and
further evaluation of bc1 complex−inhibition activity.
11169
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
Figure 2. Chemical structures of five final candidates. The fragments determined by PFVS are shown in blue boxes. 5a is a synthetic analogue of
compound 5. Ligand efficiency (LE) value is defined as the calculated binding free energy (ΔGcal) divided by the HAC, LE = −ΔGcal/HAC.
Figure 3. The process of structural optimization of compound 4. Under the guidance of computational simulation, different fragments were
introduced as side-chain moieties onto positions 3, 6, and 7 of quinoxaline; the bridged sulfur atom was also replaced with an oxygen atom. As a
result, the potency was improved step by step. Compound 4a was designed by introducing a hydrophobic methyl group onto the 3-position of the
quinoxalinyl ring of compound 4. Compound 4b was designed by replacing the bridged sulfur atom with an oxygen atom. Compound 4c was
designed by the combination of the above two strategies. Compound 4d was designed by introducing an additional methyl group onto the 6-position
of compound 4c. Compound 4e was designed by replacing the methyl group at the 3-position of compound 4d with a trifluoromethyl group.
Compound 4f and its isomer 4g were designed by introducing trifluoromethyl groups to positions 6 and 7 of the quinoxaline ring. The
experimentally determined Ki values are marked, and LE values are calculated according to the binding free energies (ΔGexp) derived from Ki values.
compound 4d (Figure 4e) with a much improved potency (Ki =
0.75 nM).
group with a trifluoromethyl group introduced several C−F···H
bonds between the trifluoromethyl and the above-mentioned
residues, which is believed to greatly improve the binding free
energy.25 As expected, the resultant compound 4e formed two
C−F···H bonds with Pro270, three with Ile146, and two each
with Tyr278 and Phe274; distances between the F and H atoms
ranged from 2.45 to 2.98 Å (Figure 4f). The Ki value of
These results led us to carefully analyze the interactions
between the substituted quinoxalinyl ring and its surrounding
residues. The methyl group at the 3-position of the quinoxaline
ring penetrated into a groove lined by hydrophobic side chains
of Pro270, Ile146, Tyr278, and Phe274. Replacing this methyl
11170
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
Figure 4. The simulated binding modes of eight different compounds. (a) Side view of the quinoxalinyl group of compound 4 surrounded by a
hydrophobic pocket formed by the side chains Phe274, Phe127, Ile146, Pro270, Glu271, Ala277, Leu294, Met124, and Ile298. (b) The potency of
compound 4a is improved greatly due to the improvement of the Ar−Ar interaction energies between the quinoxalinyl group and its surrounding
hydrophobic residues. (c) The potency of compound 4b is improved by the conformational change-induced improvement of the Ar−Ar interaction
between the quinoxalinyl ring and the phenyl group of Phe274. (d) The Ar−Ar interaction between the quinoxalinyl ring of compound 4c and the
phenyl group of Phe274 is further improved. (e) The hydrophobic interaction energies between the quinoxalinyl group of compound 4d and its
surrounding hydrophobic residues are further improved. (f) Compound 4e was designed by replacing the methyl group at position 3 with a
trifluoromethyl group, which resulted in two C−F···H bonds with Pro270, three with Ile146, and two each with Tyr278 and Phe274. (g) Additional
trifluoromethyl at position 6 of compound 4f resulted in more C−F···H bonds with F121, M124, and I298. (h) The additional trifluoromethyl at
position 7 of compound 4g resulted in more C−F···H bonds with M124 and A125. The pink lines show the hydrogen and the C−F···H bonds. The
F···H distances range from 2.38 to 3.00 Å.
Table 1. Binding Free Energies (kcal/mol) of Compounds 2−4, 4a−g, and 5a
a
b
no.
ΔH
−TΔS
ΔGcal
ΔGexp
Ki (nM)
LE
MW
2
−43.94
−40.57
−45.74
−49.68
−45.48
−47.45
−48.76
−49.92
−52.55
−51.12
33.89
31.82
37.46
39.63
36.20
36.37
36.47
37.22
39.30
37.96
35.39
−10.05
−8.75
−8.28
−10.05
−9.28
−11.08
−12.29
−12.70
−13.25
−13.16
−9.77
−10.26
−8.89
−8.28
−10.10
−9.97
−10.33
−12.48
−13.78
−14.18
−13.93
−9.60
31.10 0.90
316.09 10.71
881.80 33.92
41.00 3.60
51.50 1.30
28.00 0.53
0.75 0.23
0.41
0.32
0.32
0.37
0.38
0.38
0.45
0.44
0.42
0.41
0.36
371.48
411.50
366.44
380.46
350.37
364.39
378.42
432.39
486.36
486.36
381.46
3
4
4a
4b
4c
4d
4e
4f
4g
5a
0.083 0.013
0.043 0.006
0.065 0.018
94.97 3.82
−45.16
b
a
ΔGexp = −RTln Ki. LE is defined as the experimental binding free energy (ΔGexp) divided by the HAC, LE = −ΔGexp/HAC.
compound 4e was 0.083 nM, improved 10 624-fold compared
to compound 4; its LE was also greatly increased to 0.44. We
further designed two ditrifluoromethyl derivatives, compound
4f and its isomer, 4g. As shown in Figure 4g,h, the 6- or 7-
position CF3 group formed several additional C−F···H bonds
with the surrounding residues Phe121, Met124, Ile298, and
Ala125; the F···H distances ranged from 2.66 to 2.93 Å. The
potencies of compounds 4f and 4g were further improved 20
507- and 13 566-fold, respectively, compared to the original hit
compound 4; their LE values were still over 0.4. The synthetic
routes for compounds 2−4, 4a−g, and 5a are summarized in
Scheme 1, Supporting Information. Their chemical structures
It should be noted that the optimization of compound 4
summarized in Figure 3 is based on the above-mentioned
computational strategy. The experimental and calculated
binding free energies (ΔG) of compounds 2−4, 4a−g, and
5a (Table 1) showed a good linear correlation with a
correlation coefficient of r2 = 0.95, further confirming the
reliability of this computational strategy.
Inhibitory Kinetics of Compounds 4 and 4e. Kinetic
properties are of great importance for understanding the
molecular mechanism of bc1 function. We therefore examined
the inhibitory effects of compounds 4 and 4e on porcine
succinate-cytochrome c reductase (SCR, mixture of respiratory
complex II and bc1 complex). As previously described,23 we
measured the complex II activity of SCR using succinate and
DCIP as substrates, we used decylubiquinol (DBH2) and cyt c
as substrates to measure the bc1 complex activity, and we used
1
were characterized by H and 13C NMR and HRMS. The
crystal structure of compound 4f (CCDC 831447) was further
confirmed by X-ray diffraction analyses (Figure 2, Supporting
Information).
11171
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
succinate and cyt c as substrates to measure SCR (both
complex II and bc1 complex) activity. Both compounds 4 and
4e significantly inhibited bc1 complex activity as well as the SCR
activity. Neither compound exhibited any effect on the activity
of complex II, even with inhibitor concentrations as high as 20
μM. These results indicated that compounds 4 and 4e are
effective bc1 complex inhibitors.
Compound 4 displayed a linear product versus time
relationship (Figure 3, Supporting Information), similar to
that of the classical reversible inhibitor azoxystrobin. However,
compound 4e exhibited typical characteristics of a slow, tight-
binding inhibitor, with the product formation versus time
showing curvilinear functions (Figure 4, Supporting Informa-
tion). In the presence of compound 4e and saturated substrate
concentrations, the product formation curves started linearly in
the initial phase, but the slopes decreased with increasing time,
approaching steady states; more dramatic slope reduction was
observed with increasing compound 4e concentrations (Figure
5a). We calculated the observed first-order rate constant (kobs
)
and found it to be proportional to the concentrations of the
inhibitor 4e (Supporting Information).
Slow-tight binding inhibitors can be further classified as
competitive, noncompetitive, or uncompetitive, as ascertained
by studying the effect of substrate concentration on kobs. With
an increase in the substrate concentration, kobs decreases for a
competitive inhibitor, increases for a uncompetitive inhibitor,
or is constant for a noncompetitive inhibitor. We next
monitored the time courses of bc1 complex inhibition with
different cyt c concentrations and a fixed compound 4e
concentration (Figure 5b). The independence of kobs on [cyt c]
clearly indicated that compound 4e was a noncompetitive
inhibitor with respect to cyt c; the inhibition constant can
therefore be calculated as Ki = k−0/k+0 = 0.083 0.013 nM
(Supporting Information).
To assess the effect of the substrate ubiquinol on compound
4e, we conducted inhibitory kinetic studies of SCR, using
DBH2 and cyt c as substrates. In the presence of compound 4e,
the progress curves appeared similarly curvilinear (Figure 6).
However, in contrast to the previous observation for cyt c, kobs
decreased with increasing DBH2 concentration at a fixed
compound 4e concentration, clearly demonstrating that
compound 4e was a competitive inhibitor with respect to the
substrate ubiquinol. To further unravel the inhibitory
mechanism, we performed different sets of inhibitory experi-
ments with various concentrations of substrate DBH2 and
inhibitor compound 4e (Figure 5, Supporting Information).
Through detailed kinetic analyses, the inhibition constant Ki
was calculated to be 0.974 0.024 nM, approximately 12-fold
higher than that derived from the succinate-cyt c system. We
believe that this discrepancy is largely owing to the presence of
the nonionic detergent lauryl maltoside in the assays using
DBH2 as substrate.23
Figure 5. Inhibitory kinetics of bc1 complex with cyt c as substrate by
compound 4e. The enzyme activity was measured using succinate and
cyt c as substrates. Each reaction mixture contained 100 mM PBS (pH
7.4), 0.3 mM EDTA, 20 mM succinate, and various concentrations of
cyt c and compound 4e. The reaction was initiated by adding 0.1 nM
SCR to the reaction mixture, and the time course of the absorbance
change at 550 nm was recorded continuously for cyt c reduction.
Experimental data are shown as dots, and theoretical values as lines.
(a) Effect of the concentration of 4e on the inhibition of bc1 complex.
The assays were carried out in the presence of 60 μM of cyt c and
various concentrations of compound 4e (1, 0 nM; 2, 0.5 nM; 3, 1 nM;
4, 1.5 nM; and 5, 2.5nM). Inset: Secondary plot of kobs against
concentrations of compound 4e. (b) Effect of cyt c concentration on
the inhibition of bc1 complex by compound 4e. The assays were
carried out in the presence of 2 nM of compound 4e and various
concentrations of cyt c (1, 20 μM; 2, 28 μM; and 3, 60 μM). Inset:
Secondary plot of kobs against concentrations of cyt c.
oxygen of the methoxyacrylate and the backbone N of Glu271.
The bridging phenyl ring was nearly at right angles to the plane
of the methoxyacrylate and was inserted between residues
Pro270 and Gly142 (Figure 7c,d). The side chain extended
from the bridging ring past the ring of Phe274, in loose contact
with Met124, toward an opening to the bulk lipid phase
between helices αC and αF. This was presumably the entry
path for lipophilic substrates and inhibitors from the lipid
phase. The quinoxaline ring stacked with Phe274, and this
interaction, predicted by the modeling, has been shown to be
important for tight binding. As also predicted by the modeling
studies, the trifluoromethyl substituent inserted into a space
between Ile146, Phe274, Ala277, and Leu294, capped at the
end by Tyr278 (Figure 7c). The fit was quite close, with 10
contacts having less than 3.5 Å between the fluorine atoms and
these residues, 4 of them closer than 3.3 Å. While the rotational
position of the trifluoromethyl group may not be accurately
determined by the data at this resolution, it is clear from the
dimensions of the pocket that any rotamer would make
numerous contacts.
Crystal Structure of Chicken bc1 in Complex with
Compound 4e. We determined the crystal structure of the
representative compound 4e bound to chicken bc1 complex at a
resolution of 2.70 Å and found it to be similar to that seen with
other MOA-type inhibitors, with the Rieske iron−sulfur protein
in the c1 position.2 Electron density in the Qo site (Figure 7a)
promoted unambiguous positioning of the inhibitor. As shown
in Figure 7b, the pharmacophore of this new inhibitor bound in
the fashion of typical MOA inhibitors; the planar methox-
yacrylate was inserted into a slot bounded by Phe128, Tyr131,
Phe274, and Glu271, with an H-bond between the carbonyl
11172
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
nature of fragment screening continues to raise critical
questions about both fundamental and practical aspects of
this approach. The present work successfully developed a new
PFVS approach, providing a solution to this shortfall of FBDD.
When using computational docking for fragment screening, the
biggest challenge lies in predicting the binding mode and
accurately estimating the binding affinity, due to the
promiscuous conformation of a fragment in the binding pocket.
However, in our study, when the fragment was linked with the
pharmacophore, it became a “drug-like” molecule, making it
easy to predict the binding mode and accurately calculate its
binding affinity. Using the rational design of bc1 inhibitors to
test our approach, herein, we determined the Ki value of the
pharmacophore E-methyl-2-(2-methylphenyl)-3-methoxyacry-
Figure 6. Effect of DBH2 concentration on the inhibition of bc1
complex by compound 4e. The enzyme activity was measured using
DBH2 and cyt c as substrates. Each reaction mixture contained 100
mM PBS (pH 6.5), 2 mM EDTA, 750 μM lauryl maltoside, 100 μM
oxidized cyt c, 20 nM compound 4e, and various concentrations of
DBH2 (1, 20 μM; 2, 40 μM; 3, 60 μM; 4, 80 μM and 5, 120 μM). Each
reaction was initiated by adding 0.05 nM SCR, and the time course of
the absorbance change at 550 nm was recorded continuously for cyt c
reduction. Inset: Secondary plot of kobs against concentration of DBH2.
late to be 4065.12
206 nM, corresponding to a ΔG of
−7.37 kcal/mol. Therefore, the contribution of the fragments
to the binding free energies of compounds 4a−g varied from
0.91 to 6.81 kcal/mol. We also estimated the binding energies
of the hit fragments of compounds 4a−g. When keeping the
bridge atom invariant except for compounds 4f and 4g, the
calculated binding energies (ΔH) of these hit fragments were
qualitatively consistent with the ΔG values of their correspond-
ing pharmacophore-linked virtual ligands (Figure 8, Supporting
Information). Thus, it is practicable to identify the fragment
through evaluation of the pharmacophore-linked virtual ligand.
Of course, the pharmacophore used in the present work has a
highly conserved conformation, which made it much easier to
determine the binding mode of the pharmacophore-linked
fragments. When using the PFVS approach to design inhibitors
of proteins with a conformationally flexible pharmacophore,
sufficient energy minimization might be needed to ensure that a
rational binding mode will be obtained for the pharmacophore-
linked fragments.
Four of the 10 contacts were with Tyr278, and of these, three
involved the aromatic side chain. This residue is known to
undergo conformational changes depending on the position of
the iron−sulfur protein extrinsic domain.2,26 It is involved in
fixing the iron−sulfur protein in the “b” position or
famoxadone-induced position,27 and it may be involved in
capturing the iron−sulfur protein from its mobile state. In the
presence of MOA inhibitors, Tyr278 formed H-bonds to Ile268
(Figure 7a), partially closing and covering the outward-facing
mouth of the Qo pocket.26,27 It seems likely that the
interactions of Tyr278 with the trifluoromethyl group of the
inhibitor stabilized it in this position, preventing it from being
released to capture the iron−sulfur protein and thus accounting
for the ultrahigh activities of compounds 4e−g.
The structural similarity between the predicted and the X-ray
crystal model was found to be 0.65 Å by assessing the values of
the root-mean-square deviation (rmsd) of the atomic positions
(Figure 7, Supporting Information), which again confirmed the
reliability of our computational protocol. Furthermore, the
inhibitor binding of the predicted model was very close to that
of the crystal, except for an H-bond in the crystal structure; in
the predicted structure, the methoxy rather than carbonyl
oxygen of the methoxyacrylate formed an H-bond with the
backbone N of Glu271. While this could be a difference
between the chicken and the porcine heart bc1 complexes,
structure−activity relationship studies showed that the methoxy
oxygen could be replaced by carbon (giving a ketone instead of
ester) with little loss of affinity, whereas the carbonyl oxygen
was definitely required for any significant affinity.28 Our
previous models may have been unduly influenced by the
structure 1SQB used as the starting model; revisiting the
modeling results23 showed that the energetic difference was
small. As expected from the modeling studies, the N-containing
ring of quinoxaline in the inhibitor side chain stacked with the
phenyl ring of Phe274 (Figure 7b).
Compounds 4e−g, discovered by PFVS, are the first
reported picomolar inhibitors of the bc1 Qo site. Compared to
that of the original hit compound 4, the potencies of
compounds 4e−g were improved 10 624-, 20 507-, and 13
566-fold, respectively. The inhibitory kinetics studies showed
that these ultrapotent inhibitors exhibited slow-tight binding
characteristics, different from those of classical MOA-type
inhibitors, such as azoxystrobin and kresoxim-methyl. X-ray
crystal diffraction analysis indicated that, although the binding
mode of compound 4e was very similar to that of other MOA-
type inhibitors, it involved a unique interaction with residue
Tyr278, which constituted a part of the ISP docking crater and
was involved in fixing the ISP. It appears that loss of this
interaction between Tyr278 and the ISP would induce a more
loose state of the ISP extramembrane domain, and it has been
established that appropriate ISP mobility is crucial for the
catalytic activity of the bc1 complex. Therefore, a possible
explanation of the ultrapotencies of compounds 4e−g is that,
apart from the formation of C−F···H bonds, they enhanced the
mobility of the ISP by interacting with residue Tyr278.
In summary, the present promising study shows the newly
developed PFVS approach to be a useful tool for the rational
design of bc1 inhibitors. We have demonstrated the use of this
approach to yield the first picomolar-range Qo site inhibitors of
the cytochrome bc1 complex. Furthermore, this high-
throughput method is generally applicable to the identification
of other types of fragments for drug discovery.
DISCUSSION
■
Although FBDD has been widely used in developing new
inhibitors against important protein targets, the low-throughput
11173
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
Figure 7. X-ray structure of 4e bound to chicken bc1 complex. Residues are numbered according to the bovine/swine sequence. (a) Overall shape of
the inhibitor and quality of the electron density (2Fo−Fc map contoured at 1.5 σ). Multiple van der Walls contacts of the trifluoromethyl group with
residues 278 and 294 are shown as brown and H-bonds between the residues as white, dotted lines. (b) Space-filling model of the “bottom” of the
binding pocket, viewed from above. The inhibitor is shown as a stick figure with green carbons. The pharmacophore, at right angles to the rest of the
molecule, fits into a slot at upper right and contacts Glu271, Phe274, and Ala143. The side-chain quinoxaline ring stacks with Phe274. (c) Stick
figure model, detailing interactions of the pharmacophore and trifluoromethyl group with the protein. Distances are described in Table 2, Supporting
Information. Stereo views of these figures are available in Figure 6, Supporting Information. (d) Space-filling model of the top of the binding pocket,
viewed from below. Tyr278 (magenta, unlabeled) blocks the hole in the “top” through which the ISP accesses the site. The phenyl “bridging” ring of
the inhibitor inserts between Pro270 and the cd1 helix at 142−243. The trifluoromethyl group protrudes between Ile146, Phe274, Ala277, and
Leu294. Dotted lines show the C−F···H interactions. Both the trifluoromethyl group and the bridging ring make multiple contacts with Tyr278.
Stacking of the quinoxaline ring with Phe274 and contacts of the pharmacophore with Glu270, Tyr273, and the cd1 helix can also be seen.
MM/PBSA method31 for the enthalpy and an empirical method for
EXPERIMENTAL SECTION
■
the entropy.32
Computational Protocol. The previously established homology
Fragments were identified by three sequential steps ( Figure 9,
23
model of the porcine bc1 and a small library of 1735 fragments
Supporting Information): (1) All fragments were preliminarily
(Supporting Information) were used herein. Because all of the
screened. The energy minimization of each newly produced complex
fragments were derived from commercial products in practical use,
was achieved in four steps by using the Sander module of Amber 8.0.
they should possess good druggable properties that are favorable for de
First, the fragment was minimized with the pharmacophore and the
novo design. Based on the modification and combination of AutoGrow
protein fixed. Then, the ligand was minimized with the protein fixed.
Subsequently, the backbone atoms of the protein were fixed, and other
and the Amber 8.0 program,29,30 the PFVS protocol was designed to
automatically perform molecule generation, energy minimization, MD
simulation, and binding affinity evaluation. The detailed workflow was
as follows: (1) The structure of the pharmacophore-binding bc1 was
prepared, and the graft point on the pharmacophore was defined. (2)
The fragment was linked to the pharmacophore via a sulfur atom. The
orientation of the fragment was minimized to make the minimum
steric repulsion with the surrounding residues. (3) The ff99 and gaff
parameters for amino acids and ligand were created automatically.31
(4) Energy minimization and MD simulation were performed on the
resultant complex to obtain a reasonable binding conformation. (5)
ΔG was calculated as previously described by the combination of the
atoms were relaxed. The final minimization was performed with both
the ligand and the protein relaxed. In each step, the energy
minimization was executed by using the steepest descent method for
the first 2000 cycles, and the conjugated gradient method for the
subsequent 3000 cycles with a convergence criterion of 0.1 kcal mol−1
Å−1. Then, the ΔH calculation was performed on the minimized
complex. To reduce the computational cost, each newly produced
ligand was charged by the Gasteiger method, and only the ΔH
calculation was considered in the first step. The top ∼10% of hits
(170) with the most favorable ΔH were selected. (2) These 170
fragments were further estimated by taking the entropy effect into
11174
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
consideration. According to the calculated ΔG, the top-ranked 17
fragments were identified. (3) These 17 virtual ligands were recharged
by the restrained electrostatic potential method,33 then subjected to
energy minimization as described in step 1, and additional 20 ps MD
simulation. For temperature regulation, the Langevin thermostat was
used to maintain a temperature of 300 K.34 The atomic coordinates
were saved per ps. Subsequently, the last snapshot of the MD
simulation was minimized to a convergence criterion of 0.1 kcal mol−1
Å−1. Finally, the ΔG was calculated, and the top-ranked 10 candidates
were selected for synthetic evaluation.
ASSOCIATED CONTENT
* Supporting Information
Chemical compound information. This material is available free
of charge via the Internet at http://pubs.acs.org.
■
S
AUTHOR INFORMATION
Corresponding Author
E-mail: gfyang@mail.ccnu.edu.cn and jiaweiwu@mail.tsinghua.
edu.cn
■
Kinetic Assays. The porcine SCR, the mixture of complex II and
bc1, was prepared essentially according to the previously reported
method.35 The enzyme concentration was estimated using an
Author Contributions
∥These authors contributed equally.
552
540
extinction coefficient of 17.5 mM−1 cm−1 for A −A , which was
Notes
red
red
The authors declare no competing financial interest.
derived from the cyt c1 difference spectra between the reduced and
oxidized SCR.36 The three redox reactions are summarized as follows:
ACKNOWLEDGMENTS
■
SCR
Assay1: succinate + cyt c3+ ⎯⎯⎯→ fumarate + cyt c2+
The research was supported in part by the National Basic
Research Program of China (no. 2010CB126103) and the
NSFC (nos. 20925206, 20932005, and 31070643). We
acknowledge Dr. Z. X. Wang for critical discussions and
reading of the manuscript. We are also very thankful for the
comments and suggestions by anonymous reviewers.
complexII
Assay2: succinate + DCIP ⎯⎯⎯⎯⎯⎯⎯⎯→⎯ fumarate + DCIPH2
bc1complex
Assay3: DBH2 + cyt c3+ ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯→ DB + cyt c2+
The enzymatic activities of SCR, complex II, and the bc1 complex
were analyzed in separate reaction mixtures as reported previ-
ously.37−39 The reactions were initiated by adding a catalytic amount
of enzyme to each reaction mixture. The time course of the absorbance
change was recorded continuously at 550 nm for cyt c reduction
REFERENCES
■
(1) Zheng, C.; Han, L.; Yap, C. W.; Xie, B.; Chen, Y. Drug Discovery
Today 2006, 11, 412−420.
(2) Esser, L.; Quinn, B.; Li, Y. F.; Zhang, M.; Elberry, M.; Yu, L.; Yu,
C. A.; Xia, D. J. Mol. Biol. 2004, 341, 281−302.
(3) Kim, H.; Xia, D.; Yu, C. A.; Xia, J. Z.; Kachurin, A. M.; Zhang, L.;
Yu, L.; Deisenhofer, J. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 8026−
8033.
550
600
(A
= 18.5 mM−1 cm−1) or 600 nm for DCIP reduction (A
=
red−ox
red−ox
21 mM−1 cm−1). Initial rates were determined from the linear slope of
the obtained progress curves, and the experimental data were analyzed
using a nonlinear regression analysis program.
Crystallization and Structure Determination. Orthorhombic
crystals of chicken bc1 in the space group P212121, containing a
complete dimer in the asymmetric unit, were prepared by sitting-drop
vapor diffusion at 273 K, under optimized initial crystallization
conditions with 50 mm cacodylate; 9.4 mM TrisHCl; 30 mM K-MES,
pH 6.8; 1.8 mM K-MOPS, pH 7.2; 30 mM NaCl; 31 mM KCl; 10 mM
MgCl2; 91 g/L glycerol; 30 g/L PEG 4 kDa; 0.9 mM NaN3; 0.05 mM
EDTA; 0.47 g/L undecyl maltoside; and 31 mM octyl glucoside, pH
6.77. Crystals were grown from chicken bc1 treated with a two-fold
excess of 4e. Diffraction data were collected at beamline A1 of the
Cornell High Energy Synchrotron Source (CHESS) at an X-ray
wavelength of 0.9770 Å. Data from one crystal extending to 2.70 Å
were used to refine the previously determined structure of the protein
(3L71), and an electron density map (2Fo−Fc) was calculated. A
model of 4e was placed in the density in the Qo site map, and the
structure including inhibitor was subjected to further rounds of manual
rebuilding in O40 against 2Fo−Fc maps (CCP4)41 and to automated
refinement (atomic positional and individual isotropic ADP) in CNS
1.1.42 Noncrystallographic symmetry was restrained during positional
but not ADP refinement, dividing the monomer into 33 NCS groups
and releasing NCS restraints for numerous residues that did not seem
to conform to the NCS. As reported in Table 3, Supporting
Information, the overall Rfree value was 0.289 with 0.404 in the last
shell justifying the resolution cutoff. Rms deviation from ideal bond
lengths (0.008 Å) and angles (1.3°) was reasonable. A total of 85.0%
of the residues were within the most favorable region of the
Ramachandran plot as defined in the Procheck program,43 and only
0.3% (nine residues) was in the disallowed region. Four of these (two
in each monomer) were known to be “true” outliers from high-
resolution structures from a number of species,44 and three were in
poorly ordered regions of subunit 9. The inhibitor and surrounding
protein were relatively well ordered, especially in the first monomer
(residue C2001), which had an average isotropic ADP of 42.1, well
below the average for the protein. The structure and diffraction data
have been deposited in the PDB with ID code 3TGU.
(4) Barton, V.; Fisher, N.; Biagini, G. A.; Ward, S. A.; O’Neill, P. M.
Curr. Opin. Chem. Biol. 2010, 14, 440−446.
(5) Trumpower, B. L. J. Bioenerg. Biomembr. 1991, 23, 241−255.
(6) Yang, X. H.; Trumpower, B. L. Methods Enzymol. 1986, 126,
316−325.
(7) Brandt, U.; Trumpower, B. Crit. Rev. Biochem. Mol. Biol. 1994, 29,
165−197.
(8) Esser, L.; Elberry, M.; Zhou, F.; Yu, C. A.; Yu, L.; Xia, D. J. Biol.
Chem. 2008, 283, 2846−2857.
(9) Masaya, O.; Masaichi, W.; Yasushi, A.; Kazuki, O.; Tatsuya, N.
Expert Opin. Drug Discovery 2009, 4, 1125−1144.
(10) Murray, C. W.; Rees, D. C. Nat. Chem. 2009, 1, 187−192.
(11) Hopkins, A. L.; Groom, C. R.; Alex, A. Drug Discovery Today
2004, 9, 430−431.
(12) Abad-Zapatero, C.; Metz, J. T. Drug Discovery Today 2005, 10,
464−469.
(13) Hann, M. M.; Leach, A. R.; Harper, G. J. Chem. Inf. Comput. Sci.
2001, 41, 856−864.
(14) Fink, T.; Bruggesser, H.; Reymond, J. L. Angew. Chem., Int. Ed.
2005, 44, 1504−1508.
(15) Erlanson, D. A.; McDowell, R. S.; O’Brien, T. J. Med. Chem.
2004, 47, 3463−3482.
(16) Rees, D. C.; Congreve, M.; Murray, C. W.; Carr, R. Nat. Rev.
Drug Discovery 2004, 3, 660−672.
(17) Fruh, V.; Zhou, Y.; Chen, D.; Loch, C.; Ab, E.; Grinkova, Y. N.;
Verheij, H.; Sligar, S. G.; Bushweller, J. H.; Siegal, G. Chem. Biol. 2010,
17, 881−891.
(18) Congreve, M.; Rich, R. L.; Myszka, D. G.; Figaroa, F.; Siegal, G.;
Marshall, F. H. Methods Enzymol. 2011, 493, 115−36.
(19) Villar, H. O.; Hansen, M. R. Curr. Top. Med. Chem. 2007, 7,
1509−1513.
(20) Chen, Y.; Shoichet, B. K. Nat. Chem. Biol. 2009, 5, 358−364.
(21) Lovinge, K.; Alberts, I.; Sherman, W. Curr. Top. Med. Chem.
2010, 10, 14−32.
(22) Berry, E. A.; Zhang, Z.; Bellamy, H. D.; Huang, L. Biochim.
Biophys. Acta 2000, 1459, 440−448.
11175
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176

Journal of the American Chemical Society
Article
(23) Zhao, P. L.; Wang, L.; Zhu, X. L.; Huang, X.; Zhan, C. G.; Wu, J.
W.; Yang, G. F. J. Am. Chem. Soc. 2010, 132, 185−194.
(24) Sanders, J. M. J. Phys. Chem. A 2010, 114, 9205−9211.
(25) Li, X. Z.; Walker, B.; Michaelides, A. Proc. Natl. Acad. Sci. U.S.A.
2011, 108, 6369−6373.
(26) Crofts, A. R.; Guergova-Kuras, M.; Kuras, R.; Ugulava, N.; Li, J.;
Hong, S. Biochim. Biophys. Acta, Bioenerg. 2000, 1459, 456−466.
(27) Berry, E. A.; Huang, L.-S. Biochim. Biophys. Acta, Bioenerg. 2011,
1807, 1349−1363.
(28) Sauter, H.; Ammermann, E.; Roehl, F. Crop Protection Agents
from Nature; The Royal Society of Chemistry, Thomas Graham
House: Cambridge, U.K., 1996.
(29) Durrant, J. D.; Amaro, R. E.; McCammon, J. A. Chem. Biol. Drug
Des. 2009, 73, 168−178.
(30) Case, D. A. et al. AMBER 8; University of California: San
Francisco, CA, 2004.
(31) Hornak, V.; Abel, R.; Okur, A.; Strockbine, B.; Roitberg, A.;
Simmerling, C. Proteins 2006, 65, 712−725.
(32) Pan, Y.; Gao, D.; Zhan, C. G. J. Am. Chem. Soc. 2008, 130,
5140−5149.
(33) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Kollman, P. A. J. Am.
Chem. Soc. 1993, 115, 9620−9631.
(34) Adelman, S. A.; Doll, J. D. J. Chem. Phys. 1976, 64, 2375−2388.
(35) Yu, L.; Yu, C. A. J. Biol. Chem. 1982, 257, 2016−2021.
(36) King, T. E. Methods Enzymol. 1967, 10, 216−225.
(37) Chretien, D.; Slama, A.; Briere, J. J.; Munnich, A.; Rotig, A.;
Rustin, P. Curr. Med. Chem. 2004, 11, 233−239.
(38) Fisher, N.; Bourges, I.; Hill, P.; Brasseur, G.; Meunier, B. Eur. J.
Biochem. 2004, 271, 1292−1298.
(39) Fisher, N.; Brown, A. C.; Sexton, G.; Cook, A.; Windass, J.;
Meunier, B. Eur. J. Biochem. 2004, 271, 2264−2271.
(40) Jones, T. A.; Zou, J. Y.; Cowan, S. W.; Kjeldgaard, M. Acta
Crystallogr. A 1991, 47, 110−119.
(41) Collaborative Computational Project, Number 4. Acta
Crystallogr. Sect. D: Biol. Crystallogr. 1994, 50, 760-763.
(42) Brunger, A. T.; Adams, P. D.; Clore, G. M.; DeLano, W. L.;
̈
Gros, P.; Grosse-Kunstleve, R. W.; Jiang, J. S.; Kuszewski, J.; Nilges,
M.; Pannu, N. S.; Read, R. J.; Rice, L. M.; Simonson, T.; Warren, G. L.
Acta Crystallogr., Sect. D: Biol. Crystallogr. 1998, 54, 905−921.
(43) Vaguine, A. A.; Richelle, J.; Wodak, S. J. Acta Crystallogr., Sect. D:
Biol. Crystallogr. 1999, 55, 191−205.
(44) Huang, L. S.; Cobessi, D.; Tung, E. Y.; Berry, E. A. J. Mol. Biol.
2005, 351, 573−597.
11176
dx.doi.org/10.1021/ja3001908 | J. Am. Chem. Soc. 2012, 134, 11168−11176