3-D QSAR Investigations of the Inhibition of Leishmania major Farnesyl Pyrophosphate Synthase by Bisphosphonates

Article abstract of DOI:10.1021/jm0302344

We report the activities of 62 bisphosphonatesas inhibitors of the Leishmania major mevalonate/isoprene biosynthesis pathway enzyme, farnesyl pyrophosphate synthase. The compounds investigated exhibit activities (IC ₅₀ values) ranging from ～100 nM to ～80 μM (corresponding to K_i values as low as 10 nM). The most active compounds were found to be zoledronate (whose single-crystal X-ray structure is reported), pyridinyl-ethane-1-hydroxy-1,1-bisphosphonates or picolyl aminomethylene bisphosphonates. However, N-alicyclic amino-methylene bisphosphonates, such as incadronate (N-cycloheptyl aminomethylene bisphosphonate), as well as aliphatic aminomethylene bisphosphonates containing short (n = 4, 5) alkyl chains, were also active, with 1C₅₀ values in the 200-1700 nM range (corresponding to K_i values of ～20-170 nM). Bisphosphonates containing longer or multiple (N,N-) alkyl substitutions were inactive, as were aromatic species lacking an o- or m-nitrogen atom in the ring, or possessing multiple halogen substitutions or a p-amino group. To put these observations on a more quantitative structural basis, we used three-dimensional quantitative structure-activity relationship techniques: comparative molecular field analysis (CoMFA) and comparative molecular similarity index analysis (CoMSIA), to investigate which structural features correlated with high activity. Training set results (N = 62 compounds) yielded good correlations with each technique (R² = 0.87 and 0.88, respectively), and were further validated by using a training/test set approach. Test set results (N = 24 compounds) indicated that IC₅₀ values could be predicted within factors of 2.9 and 2.7 for the CoMFA and CoMSIA methods, respectively. The CoMSIA fields indicated that a positive charge in the bisphosphonate side chain and a hydrophobic feature contributed significantly to activity. Overall, these results are of general interest since they represent the first detailed quantitative structure-activity relationship study of the inhibition of an expressed farnesyl pyrophosphate synthase enzyme by bisphosphonate inhibitors and that the activity of these inhibitors can be predicted within about a factor of 3 by using 3D-QSAR techniques.

Full text of DOI:10.1021/jm0302344

J . Med. Chem. 2003, 46, 5171-5183
5171
3-D QSAR In vestiga tion s of th e In h ibition of Leish m a n ia m a jor F a r n esyl
P yr op h osp h a te Syn th a se by Bisp h osp h on a tes
J ohn M. Sanders,† Aurora Ortiz Go´mez,‡ J unhong Mao,† Gary A. Meints,† Erin M. Van Brussel,†
Agnieszka Burzynska,§ Pawel Kafarski,§ Dolores Gonza´lez-Pacanowska,‡ and Eric Oldfield*,†,|
Department of Chemistry, University of Illinois at Urbana-Champaign, 600 South Mathews Avenue, Urbana, Illinois 61801;
Lo´pez-Neyra Institute of Parasitology and Biomedicine, CSIC, Granada, Spain; Institute of Organic Chemistry, Biochemistry
and Biotechnology, Wroclaw University of Technology, Wroclaw, Poland; and Center for Biophysics and Computational
Biology, University of Illinois at Urbana-Champaign, 600 South Mathews Avenue, Urbana, Illinois 61801
Received May 15, 2003
We report the activities of 62 bisphosphonates as inhibitors of the Leishmania major
mevalonate/isoprene biosynthesis pathway enzyme, farnesyl pyrophosphate synthase. The
compounds investigated exhibit activities (IC50 values) ranging from ∼100 nM to ∼80 µM
(corresponding to Ki values as low as 10 nM). The most active compounds were found to be
zoledronate (whose single-crystal X-ray structure is reported), pyridinyl-ethane-1-hydroxy-1,1-
bisphosphonates or picolyl aminomethylene bisphosphonates. However, N-alicyclic amino-
methylene bisphosphonates, such as incadronate (N-cycloheptyl aminomethylene bisphospho-
nate), as well as aliphatic aminomethylene bisphosphonates containing short (n ) 4, 5) alkyl
chains, were also active, with IC50 values in the 200-1700 nM range (corresponding to Ki values
of ∼20-170 nM). Bisphosphonates containing longer or multiple (N,N-) alkyl substitutions
were inactive, as were aromatic species lacking an o- or m-nitrogen atom in the ring, or
possessing multiple halogen substitutions or a p-amino group. To put these observations on a
more quantitative structural basis, we used three-dimensional quantitative structure-activity
relationship techniques: comparative molecular field analysis (CoMFA) and comparative
molecular similarity index analysis (CoMSIA), to investigate which structural features
correlated with high activity. Training set results (N ) 62 compounds) yielded good correlations
with each technique (R2 ) 0.87 and 0.88, respectively), and were further validated by using a
training/test set approach. Test set results (N ) 24 compounds) indicated that IC50 values
could be predicted within factors of 2.9 and 2.7 for the CoMFA and CoMSIA methods,
respectively. The CoMSIA fields indicated that a positive charge in the bisphosphonate side
chain and a hydrophobic feature contributed significantly to activity. Overall, these results
are of general interest since they represent the first detailed quantitative structure-activity
relationship study of the inhibition of an expressed farnesyl pyrophosphate synthase enzyme
by bisphosphonate inhibitors and that the activity of these inhibitors can be predicted within
about a factor of 3 by using 3D-QSAR techniques.
In tr od u ction
in some areas resistance can be as high as 40%.5
A
second line of defense is the use of AmBisome, a less
toxic liposomal formulation of amphotericin B6 which,
because of its cost, is not well suited for use in less
developed countries. More recently, the drug Miltefosine
has been introduced for the treatment of visceral
leishmaniasis,7,8 and there are promising results with
long-term fluconazole treatments for L. major,9 but
there is still a need for additional, inexpensive, and
effective therapies.
The leishmaniases are a series of diseases caused by
Leishmania species.1 The most lethal form is visceral
leishmaniasis (also known as Kala Azar), which is
caused by L. donovani, with untreated cases reaching
a ∼90% mortality rate within 6-24 months.2 Cutaneous
and mucocutaneous leishmaniasis are caused primarily
by L. major and L. mexicana, respectively, and cause
severe skin lesions. There are approximately 1.5 million
cases of leishmaniasis each year and some 350 million
individuals are at risk of infection.1 The drugs which
have been used most frequently to treat the leishma-
niases are the pentavalent antimonials Pentostam and
Glucantime.3,4 However, these drugs are quite toxic and
In recent work, it has been found that bisphospho-
nates of the type currently in use in bone resorption
therapy, such as risedronate (Actonel),10 and in treating
hypercalcemia due to malignancy, such as pamidronate
(Aredia),11 have considerable in vitro activity against a
variety of trypanosomatid parasites, including L. dono-
vani, Trypanosoma cruzi, and Trypanosoma brucei.12
Moreover, risedronate effected a parasitological cure of
visceral leishmaniasis (caused by L. donovani), and
pamidronate effected a parasitological cure of cutaneous
leishmanisis (caused by L. mexicana), in Balb/c mice.13,14
* To whom correspondence should be addressed. Telephone: 217-
333-3374. Fax: 217-244-0997. E-mail: eo@chad.scs.uiuc.edu.
† Department of Chemistry, University of Illinois at Urbana-
Champaign.
‡ Lo´pez-Neyra Institute of Parasitology and Biomedicine.
§ Wroclaw University of Technology.
| Center for Biophysics and Computational Biology, University of
Illinois at Urbana-Champaign.
10.1021/jm0302344 CCC: $25.00 © 2003 American Chemical Society
Published on Web 10/21/2003

5172 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
The site of action of these drugs has been shown to be
the mevalonate/isoprene biosynthesis pathway enzyme
farnesyl pyrophosphate (FPP) synthase,14-19 and some
bisphosphonates have been shown to be potent inhibi-
tors of FPP synthase from the trypanosomatids T. cruzi
and T. brucei.20,21 Here, we describe the activity of a
wide variety of bisphosphonates against a recombinant
FPP synthase from L. major, the organism responsible
for cutaneous leishmaniasis in Europe and Asia. We
then use three-dimensional quantitative structure-
activity relationship (3D-QSAR) techniques to analyze
the results and demonstrate that the QSAR models are
predictive. These results may help facilitate the devel-
opment of novel antileishmanial drugs for treatment of
cutaneous and other forms of leishmaniasis in both
immunocompetent and immunosuppressed22 individuals
and represent the first comprehensive QSAR investiga-
tion of the inhibition of any FPP synthase enzyme by
bisphosphonates.
concentration of the inhibitor (µM), and IC50 is the
concentration (µM) for 50% enzyme inhibition. The Ki
values were determined from the relation:24
IC50
Ki )
(3)
[GPP]
1 +
Km
where [GPP] ) 100 µM and Km ) 10 µM.
Com p a r a tive Molecu la r F ield An a lysis. To carry
out our CoMFA25 investigations, we basically followed
the approaches described previously for investigating
the activity of bisphosphonates against the growth of
T. brucei bloodstream form trypomastigotes,26 in inhib-
iting the growth of Dictyostelium discoideum, and in
bone resorption.27,28 Bisphosphonate structures were
built and then geometry optimized by using a three-
step protocol, consisting of steepest-descent, Powell,
then BFGS algorithms, using the Tripos force field in
the Sybyl 6.9 program.29 Each energy minimized struc-
ture was then aligned to the lowest energy conformation
of compound 4 by using the Database Align dialogue in
Sybyl 6.929 to fit each compound to the (H)O-P-C-P-
O(H) atoms in the bisphosphonate backbone common
to all of the bisphosphonates investigated. The set of
aligned structures is shown in Figure 2. We used default
settings to automatically build a 3-D rectangular grid
with 2 Å spacing surrounding the alignment shown in
Figure 2 and then used hydrophobic, electrostatic, and
steric probes to calculate descriptors at these gridpoints.
In all cases, we used singly deprotonated phosphonate
groups, while the nitrogen-containing side chains were
in most cases singly protonated, basically as described
before.26-28 Since the structure and, more particularly,
the protonation state of the most active compound,
zoledronate (4), had not been reported previously, we
also carried out a single-crystal X-ray crystallographic
study of 4 (crystallized at pH ) 7.4) and verified that
indeed, the imidazole ring was protonated, as shown in
Figure 3. Alkyl chains were constrained to the all-trans
geometry. For chiral compounds, the enantiomers which
best fit the initial alignment were used in the QSAR
calculations. In the cases of compounds 24 and 33, the
chiralities were the same as those used in our previous
T. brucei QSAR investigation.26
Resu lts a n d Discu ssion
We investigated the activity of the 62 bisphosphonates
shown in Figure 1 in inhibiting the activity of an
expressed L. major FPP synthase enzyme, or more
specifically, the activity of bisphosphonates in inhibiting
the condensation of the homoallylic substrate, isopen-
tenyl pyrophosphate (1, IPP), with the allylic substrate,
geranyl pyrophosphate (2, GPP):
The resulting farnesyl pyrophosphate (3, FPP) is used
extensively in the biosynthesis of isoprenoids, such as
dolichols, in the biosynthesis of sterols, such as ergos-
terol, as well as in protein prenylation, and is essential
for parasite survival. The FPP synthase from L. major23
has a 63% identity with the FPP synthase from T.
brucei, where RNAi experiments have demonstrated
that FPPS is the target for bisphosphonate drugs.21
The activities of the compounds investigated (Figure
1) are shown in Table 1 as IC50 (µM), Ki (µM) and pIC50
( ) -log IC50 (M)) values, where the IC50 values
represent the concentrations of bisphosphonates re-
quired for 50% enzyme inhibition. Enzyme inhibition
was determined basically as described elsewhere for the
T. cruzi enzyme,20 by measuring 14C incorporation into
FPP from [14C]-IPP (the reaction shown in eq 1). IC50
values were obtained from least-squares fits to the
following equation:
We first performed a CoMFA analysis on all 62
compounds using a partial least squares (PLS) approach
and a Gasteiger-Marsili charge set.30 The optimal
number of components in the final PLS model was
determined to be 6 by using q2 values, as obtained from
the SAMPLS leave-one-out cross-validation technique
implemented in Sybyl 6.9.29 This six component CoMFA
model gave a correlation coefficient R2 ) 0.88 and a q2
of 0.50. These and other statistical parameters are given
in Table 1, together with the experimental and predicted
pIC50 values. The training set results for these 62
compounds are shown graphically in Figure 4A.
To validate this CoMFA analysis, we next carried out
a series of calculations to evaluate to what extent the
CoMFA model had predictive value. To do this, we
deleted eight points at random from the training set and
computed a new CoMFA model, using this model to
predict the activities of the eight excluded compounds.
This process was repeated two additional times, giving
I
maxC
I )
(2)
IC50 + C
where I is the inhibition fraction, Imax ) 1, C is the

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5173
F igu r e 1. Structures of bisphosphonates investigated, ordered by decreasing activity.
a total of 24 predicted values. The results for the three
sets of calculations (24 points) are given in Table 1, in
which the predicted values are shown as bold entries,
and Figure 4B shows the results of the 24 predictions,
graphically. In all cases, the CoMFA models so obtained
were statistically significant (Table 1) and the predicted
test set values were in generally good agreement with
those of the training set, indicating that the training
set model is robust with respect to the training set
composition. The average pIC50 error for the 24 pre-
dicted values from the three sets of calculations was
0.44, corresponding to a factor of 2.8 uncertainty in the
IC50 predictions.
We next used a cross-validation analysis to investi-
gate the stability of the CoMFA analysis with respect
to training set composition.31 The training set model was

5174 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
Ta ble 1. Experimental (IC50, Ki and pIC50) and CoMFA Predicted (pIC50) Values for Bisphosphonates against L. major FPPS and
Statistical Results for the CoMFA Models
a
a
experimental activity
predicted pIC50 Gasteiger-Marsili charges
predicted pIC50 MSK charges
IC50
(µM)
Ki
(µM)
CoMFA
training set
CoMFA
test set
CoMFA
test set
CoMFA
test set
CoMFA
training set
CoMFA
test set
CoMFA
test set
CoMFA
test set
compound
pIC50
4
5
6
7
8
9
0.11
0.11
0.14
0.16
0.16
0.17
0.17
0.18
0.21
0.21
0.23
0.24
0.25
0.31
0.33
0.35
0.36
0.42
0.43
0.45
0.48
0.49
0.49
0.50
0.51
0.54
0.78
0.90
0.95
1.0
1.1
1.3
1.3
1.4
1.7
1.7
1.9
1.9
2.4
2.4
3.2
3.3
3.4
4.4
4.6
5.4
0.010
0.011
0.014
0.016
0.015
0.016
0.016
0.017
0.020
0.020
0.022
0.023
0.024
0.030
0.032
0.033
0.034
0.040
0.041
0.043
0.046
0.046
0.047
0.047
0.049
0.052
0.075
0.086
0.091
0.10
0.11
0.12
0.13
0.14
0.16
0.17
0.18
0.18
0.23
0.23
0.30
0.31
0.33
0.42
0.44
0.51
0.56
0.62
0.64
0.65
0.74
0.90
0.92
1.0
6.96
6.95
6.85
6.80
6.80
6.77
6.77
6.74
6.68
6.68
6.64
6.62
6.60
6.51
6.48
6.46
6.44
6.38
6.37
6.35
6.32
6.31
6.31
6.31
6.29
6.27
6.11
6.05
6.02
5.98
5.94
5.89
5.87
5.85
5.78
5.76
5.73
5.72
5.63
5.61
5.49
5.49
5.47
5.36
5.34
5.27
5.23
5.19
5.17
5.17
5.11
5.03
5.02
4.97
4.82
4.78
4.74
4.70
4.39
4.37
4.36
4.12
6.68
6.74
6.56
6.47
6.61
6.70
6.66
6.45
6.20
6.55
6.28
6.43
6.67
6.41
6.94
6.03
6.71
6.20
6.66
6.35
6.37
6.21
6.21
6.39
5.98
6.24
6.46
6.28
6.21
5.89
5.92
5.90
6.42
6.40
6.01
5.67
5.54
5.74
5.30
6.08
5.28
5.38
5.42
5.65
5.30
5.31
5.12
5.27
5.26
5.28
5.31
5.02
5.17
4.77
5.03
4.56
4.46
4.76
4.89
4.32
4.18
4.98
0.88
64.41
0.50
6
6.72
6.71
6.67
6.58
6.65
6.66
6.65
6.47
6.28
6.50
6.44
6.38
6.65
6.63
6.82
6.17
6.69
6.17
6.60
6.43
6.44
6.37
6.27
6.28
6.19
6.26
6.35
6.23
6.10
5.89
5.99
5.92
6.30
6.35
5.98
5.76
5.51
5.69
5.06
6.22
5.62
5.35
5.32
5.62
5.27
5.32
5.26
5.01
5.01
5.31
5.19
4.93
5.22
4.65
5.71
4.67
4.61
4.83
4.70
4.54
4.31
5.40
0.91
80.64
0.50
6
6.66
6.41
6.62
6.66
6.67
6.67
6.75
6.56
6.24
6.45
6.64
6.33
6.69
6.46
6.91
6.14
6.77
6.28
6.61
6.46
6.51
6.17
6.20
6.38
6.03
6.34
6.35
6.26
6.11
5.88
5.87
5.86
6.39
6.31
5.91
5.81
5.47
5.88
5.33
6.62
5.26
5.41
5.39
5.63
5.27
5.33
5.27
5.25
5.26
5.23
5.25
4.92
5.25
4.87
4.99
4.55
4.40
4.96
4.89
5.16
4.12
4.96
0.89
61.04
0.46
6
6.08
6.26
6.30
5.91
6.14
6.36
6.44
6.38
5.86
6.37
5.92
6.48
6.54
6.55
6.25
5.95
5.93
6.33
6.32
6.53
6.62
6.76
6.01
6.64
5.40
5.67
5.79
6.19
6.08
5.76
5.77
6.27
6.36
6.50
5.74
5.95
5.77
5.94
5.22
5.98
5.24
5.46
5.25
6.11
6.15
5.70
5.03
5.15
5.12
4.63
5.21
5.02
4.91
5.68
5.20
5.20
5.02
5.04
5.41
4.88
5.26
5.16
0.59
36.88
0.40
2
6.69
6.79
6.39
6.49
6.48
6.59
6.68
6.56
6.24
6.64
6.29
6.50
6.52
6.39
6.87
6.06
6.73
6.46
6.69
6.27
6.30
6.12
6.32
6.24
5.91
6.26
6.45
6.15
6.27
5.93
5.88
6.09
6.47
6.49
6.03
5.69
5.46
5.65
5.34
6.10
5.25
5.35
5.43
5.71
5.36
5.32
5.30
5.18
5.16
5.16
5.29
4.93
5.35
4.85
4.80
4.59
4.41
4.93
4.85
4.48
4.17
4.89
0.87
61.02
0.49
6
6.72
6.78
6.54
6.50
6.43
6.61
6.71
6.56
6.34
6.58
6.36
6.49
6.54
6.50
6.77
6.23
6.70
6.44
6.64
6.29
6.32
6.14
6.41
6.21
6.15
6.19
6.37
6.25
6.16
5.84
5.87
6.12
6.47
6.47
6.03
5.85
5.40
5.61
5.11
6.19
5.64
5.30
5.36
5.70
5.38
5.24
5.35
5.09
5.09
5.13
5.26
4.97
5.32
4.76
5.63
4.63
4.71
4.93
4.74
4.53
4.23
5.46
0.89
64.35
0.46
6
6.71
6.44
6.51
6.67
6.54
6.61
6.78
6.63
6.24
6.49
6.62
6.35
6.64
6.43
6.84
6.15
6.79
6.52
6.66
6.50
6.48
6.13
6.28
6.27
6.03
6.33
6.34
6.23
6.12
5.89
5.85
6.00
6.50
6.35
5.90
5.85
5.38
5.81
5.38
6.62
5.26
5.33
5.42
5.55
5.21
5.33
5.35
5.22
5.22
5.19
5.24
4.95
5.32
4.97
5.01
4.60
4.41
5.06
4.80
5.35
4.14
4.90
0.89
60.62
0.46
6
6.15
6.13
6.25
5.95
5.99
6.28
6.37
6.27
5.89
6.25
5.96
6.36
6.53
6.59
6.27
5.97
5.96
6.35
6.18
6.57
6.64
6.78
6.05
6.66
5.40
5.70
5.83
6.20
6.08
5.78
5.78
6.23
6.28
6.38
5.76
5.96
5.76
6.00
5.23
6.01
5.24
5.48
5.26
6.15
6.21
5.73
5.04
5.16
5.14
4.59
5.22
5.01
4.92
5.70
5.21
5.20
5.01
5.05
5.46
4.79
5.28
5.15
0.58
35.40
0.39
2
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
R2 b
Ftest
q2d
Ne
Nf
q2
5.9
6.5
6.7
6.8
7.8
9.3
9.6
11
15
17
18
20
41
43
43
76
1.5
1.6
1.7
1.9
3.9
4.1
4.2
7.3
c
62
54
54
54
62
54
54
54
g
0.41
0.08
0.47
0.05
-0.15
0.08
0.39
0.09
0.45
0.05
-0.16
0.11
2CV
SD (σ) q2
h
2CV
5CV
q2
g
5CV
SD (σ) q2
ave. q2
h
i
random
max. q2
j
random
field contributions
steric
electrostatic
0.670
0.330
0.666
0.334
0.676
0.324
0.612
0.388
0.666
0.334
0.647
0.353
0.670
0.330
0.631
0.369
a
b
Bold values represent predicted activities of compounds that were not included in the training set. Correlation coefficient. c Ratio of
R2 explained to unexplained ) R2/(1-R2). Cross-validated correlation coefficient after leave-one-out procedure. e Optimal number of principal
d
components. f Number of compounds. Average cross-validated correlation coefficient for 100 trials using the indicated number of cross-
g
validation groups. Standard deviation of average cross-validated correlation coefficient for 100 trials. i Average cross-validated correlation
h
j
coefficient obtained for random pIC50 runs. Maximum cross-validated correlation coefficient obtained for random pIC50 runs.

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5175
F igu r e 3. Single-crystal X-ray structures of zoledronate (4)
crystallized from water. Displacement ellipsoids are drawn at
the 35% probability level and H atoms are shown as small
spheres of arbitrary radii. Detailed structural information is
available in the Supporting Information.
yielding a q2 ) 0.08 and an average q2 < 0 (Table 1).
This suggests that the CoMFA model obtained by using
the experimental activities does not result from any
chance correlation.
In almost all of the molecules investigated, as noted
above, we used protonated side chains since it is thought
that N-containing bisphosphonates act as carbocation
transition state/reactive intermediate analogues.32 How-
ever, using charged side chains for 63 and 64 gave
predicted activities considerably higher than those
observed experimentally. This suggested that these
species might in fact be nonprotonated. This is consis-
tent with our observation28 that bisphosphonate activity
(in bone resorption) is highly correlated with the pKa of
the parent (heterocycle) base with pKas in the ∼5-9
range being optimum for activity.28 For 63, the dichlo-
rosubstitution yields a computed pKa for the free base
of ∼2.48 while that for aniline is ∼4.61.33 Thus, 63 is
unlikely to be protonated while 64 might be. Likewise,
the halogenated species 15 and 37 have pKa values of
∼4.6833 and might be protonated or nonprotonated. In
earlier work,26 we found that 15 was only a very weak
inhibitor of T. brucei growth, implying that it is not
protonated. To investigate this topic in more detail, we
therefore next carried out a single-crystal X-ray crystal-
lographic study of 15, grown from pH ∼5.5 and pH ∼2.5
solutions, in addition to exploring the effects of removing
15, 37, 63, and 64 from the training set and using the
reduced training set to predict the activities of both
protonated and nonprotonated species.
F igu r e 2. CoMFA (and CoMSIA) structure alignment; su-
perposition of 62 compounds.
cross-validated using two and five cross-validation
groups 100 times each: the average and standard
deviation (σ) of q2 are shown in Table 1. When two cross-
validation groups were used, the average q2 value was
0.41 with σ ) 0.08. The use of two cross-validation
groups leaves 31 of the 62 training set molecules in the
model construction group and predicts the activities of
the remaining 31 compounds. These values were im-
proved, at q2 ) 0.47 and σ ) 0.05 with the use of five
cross-validation groups, which results in a more con-
sistent cross-validation training set composition for each
run.
We also used a random number generator to generate
random pIC50 values between 4.12 (the activity of the
least active compound investigated) and 6.96 (the activ-
ity of most active compound investigated). We replaced
the experimentally observed pIC50 values with the
random values and generated new QSAR models using
all 62 (random) training set pIC50 values. This process
was repeated 100 times with the best CoMFA model
We show in Figure 5 the single-crystal X-ray struc-
tures of 15 grown from pH 5.5 (A) and pH 2.5 (B)
solutions. At the more basic pH value, we found that
the ring was nonprotonated, Figure 5A. This is unprec-
edented for N-containing bisphosphonates but is con-
sistent with NMR pH titrations of this compound.34 On
the other hand, 15 was found to be ring protonated
when crystallized from the more acidic solution, Figure
5B. While the actual protonation state of the bisphos-
phonates in FPP synthase is clearly not known for
certain, these results, together with the observation that

5176 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
F igu r e 4. Plots of experimental pIC50 versus predicted pIC50 values for the CoMFA models of bisphosphonates inhibiting
Leishmania major FPP synthase. A, 62 compound training set, Gasteiger-Marsili charges; B, results of the three, eight-compound
test set prediction studies, Gasteiger-Marsili charges; C, as A but Merz-Singh-Kollman charges; D, as B but MSK charges.
The straight lines represent the ideal 45° slopes.
both 15 and 37 have relatively high activity, strongly
supports the idea that both bind in their protonated
form.
the “local pH” value is close to the pKa value of the
compound of interest since both protonated and non-
protonated species, perhaps in exchange, may be present
in the experimental sample. Nevertheless, our results
do suggest that 15 and 37 are most likely protonated
when bound while 63 and 64 are nonprotonated. This
lack of activity of bisphosphonates containing weakly
basic side chains is also reflected in recent studies on
bone resorption35 in which thiazole-, triazole-, and
pyrazole-containing bisphosphonates had little or no
activity as bone resorption agents, which can be cor-
related with parent base pKa values of <3.28 On the
other hand, the pKas of aminopyridines and pyridines,
the precursors of highly active species such as 5, 6, 9,
and 11, are all in the range 5.52-7.38,33 and the
protonated forms of these species are expected to be good
inhibitors, as found experimentally, and as computed
using QSAR techniques when using ring-protonated
species.
This sensitivity to charge states in several species led
us next to consider whether improved results might be
obtained by using the Merz-Singh-Kollman (MSK)
approach to deduce partial atomic charges,36 as opposed
to our initial use of Gasteiger-Marsili charges.30 We
thus carried out Hartree-Fock calculations using a
6-31G* basis set for each compound using the Gaussian
98 program37 and used the resulting wave functions to
evaluate the MSK charges.36 The CoMFA test and
training set calculations discussed above were then
repeated using the MSK charge set. The results are
To investigate this question in more detail, we next
carried out a series of calculations using a reduced
training set of 58 compounds in which 15, 37, 63, and
64 were not included in the set and then predicted the
activities of both the protonated and nonprotonated
forms of each of these four compounds. Results are
shown in Table 2. From Table 2, it can be seen that the
nonprotonated nitrogen predictions for 63 and 64 are
much closer to experiment than are the protonated
nitrogen predictions. For example, the pIC50 errors for
63 are 1.20 (charged) and 0.54 (uncharged) while those
for 64 are 1.91 (charged) and 0.11 (uncharged), sug-
gesting that the side chains of 63 and 64 bind in an
uncharged form to the FPPS enzyme. This is consistent
with the weak basicity of anilines and the very weak
basicity predicted for the dichloroaniline species, 63. For
the protonated ring form of 15, the deviation between
the experimental pIC50 and the predicted pIC50 was only
0.16, Table 2, while for the unprotonated form, the error
is 1.46. That is, the experimental result can only be
reconciled with the predictions by invoking ring proto-
nation in the enzyme. For 37, the situation is less clear
since the computed errors were 0.94 (charged) and 0.78
(uncharged), Table 2. The CoMSIA result however,
discussed below, implies that 37 should most likely be
treated as a ring protonated species. This clearly il-
lustrates a difficulty with any QSAR analysis in which

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5177
both yielding about a 67% steric and a 33% contribution
to the CoMFA field, Table 1. With the CoMSIA analysis
(described below) however, there is a definite improve-
ment when using the ab initio computed charges.
Com p a r a tive Molecu la r Sim ila r ity In d ex An a ly-
sis. Next, we investigated the use of the comparative
molecular similarity index analysis38 approach to in-
vestigate structure-activity relationships with the bis-
phosphonates. In contrast to CoMFA, which uses stan-
dard Lennard-J ones and Coulombic terms, CoMSIA
indices are calculated using Gaussian-type functions.
The CoMSIA indices vary less rapidly at short gridpoint-
molecule distances than do the CoMFA energies, plus
the lattice itself is used to elucidate similarity, and
CoMSIA maps are thought by some workers to be more
useful since they focus on areas that are actually
occupied by ligand atoms.38 CoMSIA analyses are also
typically less sensitive to small alignment errors than
are CoMFA analyses, since there are no singularities
at ligand atom sites. The CoMSIA analysis was per-
formed basically following the procedure outlined above
for the CoMFA analysis and as described in more detail
in the Experimental Section. The CoMSIA alignment
and grid were the same as those used in the CoMFA
analysis discussed above (Figure 2), and the training
and test set results are given in Table 3 and graphically
in Figure 6A-D. The resulting CoMSIA field maps are
shown in Figure 7. 15 and 37 were taken to have
protonated rings while 63 and 64 were nonprotonated,
in the initial CoMSIA analysis. We also carried out an
additional set of calculations in which the activities of
these species were predicted using a reduced training
set, basically as discussed above. These results are
shown in Table 2 and are generally similar to the
CoMFA results, with the exception that the CoMSIA
analysis with MSK charges more clearly supports the
use of a protonated structure for 37.
For the training set results, we obtain an R2 ) 0.86,
an F-test value of 54.2, and a q2 of 0.57 for N ) 62
compounds using six components, Table 3, using
Gasteiger-Marsili charges. The quality of this training
set was investigated using two and five cross-validation
groups, as described above for the CoMFA analysis. For
two cross-validation groups we obtained an average q2
) 0.40 with σ ) 0.13 while with the five cross-validation
groups we obtained an average q2 ) 0.52 with σ ) 0.06.
We also repeated the use of the randomized data
analysis discussed above, which also resulted in a
negative average q2 for 100 runs and a maximum q2 of
0.09 (Table 3), again supporting the validity of the
CoMSIA analysis.
F igu r e 5. Single-crystal X-ray structures of 15 crystallized
at A, pH ) 5.5; B, pH ) 2.5. Displacement ellipsoids are drawn
at the 35% probability level and H atoms are shown as small
spheres of arbitrary radii. Detailed structural information is
available in the Supporting Information.
Ta ble 2. Experimental and Predicted CoMFA and CoMSIA
pIC50 Values and Statistical Results of Corresponding Models
for Selected Compounds
predicted pIC50
Gasteiger-Marsili
predicted pIC50
MSK Charges
charges
exptl CoMFA CoMSIA CoMFA CoMSIA
compound
pIC50
pIC50
pIC50
pIC50
pIC50
15 - charged-N
6.62
6.78
5.00
6.77
4.90
5.73
4.73
5.97
4.51
6.28
3.99
6.32
4.01
5.39
4.54
5.53
4.27
6.78
5.16
6.79
5.07
5.57
4.91
6.27
4.47
6.43
4.20
6.46
4.24
5.12
4.28
5.77
4.07
15 - uncharged-N 6.62
37 - charged-N
37 - uncharged-N 5.85
63 - charged-N 4.37
63 - uncharged-N 4.37
4.36
5.85
64 - charged-N
As with the CoMFA approach, we also investigated
to what extent the CoMSIA method, as applied here to
bisphosphonates, is predictive. We again used three 54-
compound training sets to predict the activities of three
sets of eight compounds. Typical 24 component test set
results are shown in Figure 6B. On average, the error
in prediction is about a factor of 2.6, slightly better than
that obtained with the CoMFA analysis. The training/
test set calculations were then repeated using the MSK
charge set: results are shown in Table 3 and Figure
6C,D. For CoMSIA, use of the MSK charge set produced
slightly improved statistics in all cases (for R2, Ftest, q2,
Table 3). The average errors in the CoMSIA test set
64 - uncharged-N 4.36
statistics
R2a
0.88
61.18
0.45
6
0.85
46.88
0.47
6
0.87
59.18
0.44
6
0.88
60.21
0.55
6
b
Ftest
q2c
Nd
Ne
58
58
58
58
a
b
Correlation coefficient. Ratio of R2 explained to unexplained
) R2/(1 - R2). c Cross-validated correlation coefficient after leave-
d
one-out procedure. Optimal number of principal components.
e Number of compounds.
shown in Tables 1 and 2 and Figures 4C,D but (at least
with the CoMFA analysis) are essentially the same, with

5178 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
Ta ble 3. Experimental (IC50, Ki and pIC50) and CoMSIA Predicted (pIC50) Values for Bisphosphonates against L. major FPPS and
Statistical Results for the CoMSIA Models
a
a
experimental activity
predicted pIC50 Gasteiger-Marsili charges
predicted pIC50 MSK charges
IC50
(µM)
Ki
(µM)
CoMSIA
training set
CoMSIA
test set
CoMSIA
test set
CoMSIA
test set
CoMSIA
training set
CoMSIA CoMSIA CoMSIA
compound
pIC50
test set
test set
test set
4
0.11
0.11
0.14
0.16
0.16
0.17
0.17
0.18
0.21
0.21
0.23
0.24
0.25
0.31
0.33
0.35
0.36
0.42
0.43
0.45
0.48
0.49
0.49
0.50
0.51
0.54
0.78
0.90
0.95
1.0
0.010
0.011
0.014
0.016
0.015
0.016
0.016
0.017
0.020
0.020
0.022
0.023
0.024
0.030
0.032
0.033
0.034
0.040
0.041
0.043
0.046
0.046
0.047
0.047
0.049
0.052
0.075
0.086
0.091
0.10
0.11
0.12
0.13
0.14
0.16
0.17
0.18
0.18
0.23
0.23
0.30
0.31
0.33
0.42
0.44
0.51
0.56
0.62
0.64
0.65
0.74
0.90
0.92
1.0
6.96
6.95
6.85
6.80
6.80
6.77
6.77
6.74
6.68
6.68
6.64
6.62
6.60
6.51
6.48
6.46
6.44
6.38
6.37
6.35
6.32
6.31
6.31
6.31
6.29
6.27
6.11
6.05
6.02
5.98
5.94
5.89
5.87
5.85
5.78
5.76
5.73
5.72
5.63
5.61
5.49
5.49
5.47
5.36
5.34
5.27
5.23
5.19
5.17
5.17
5.11
5.03
5.02
4.97
4.82
4.78
4.74
4.70
4.39
4.37
4.36
4.12
6.62
6.98
6.45
6.40
6.87
6.72
6.79
6.83
6.15
6.77
6.38
6.20
6.56
6.50
6.41
6.02
6.53
6.23
6.81
6.13
6.48
6.31
6.41
6.51
5.79
6.04
6.40
6.09
5.80
5.86
5.81
6.06
6.54
6.18
6.01
5.68
5.56
5.21
5.31
6.37
5.27
5.51
5.45
5.45
5.30
5.15
5.28
5.17
5.11
4.90
5.39
5.10
5.40
4.87
4.93
4.90
4.70
4.99
5.17
4.41
4.33
4.66
0.86
54.15
0.57
6
6.53
6.82
6.68
6.16
6.85
6.87
6.81
6.71
6.18
6.73
6.07
6.19
6.37
6.52
6.56
6.08
6.36
6.00
6.69
6.22
6.36
6.41
6.42
6.36
5.97
5.91
6.31
5.85
5.89
5.87
5.73
6.06
6.69
6.15
6.09
5.75
5.73
5.54
5.11
6.06
5.63
5.62
5.21
5.64
5.63
5.25
5.34
5.26
5.27
4.66
5.20
5.31
5.42
5.18
5.25
5.06
5.25
5.05
5.49
4.18
4.24
5.30
0.80
49.19
0.45
4
6.63
6.84
6.51
6.64
6.96
6.77
6.78
6.85
6.23
6.64
6.77
6.13
6.49
6.51
6.39
6.14
6.63
6.29
6.79
5.96
6.53
6.20
6.53
6.44
5.85
6.19
6.38
6.03
5.88
5.85
5.84
6.04
6.58
6.09
5.97
5.85
5.52
5.32
5.29
6.88
5.29
5.56
5.36
5.45
5.44
5.15
5.38
5.31
5.33
4.81
5.30
5.04
5.46
4.95
4.97
4.88
4.62
5.14
5.17
5.18
4.14
4.69
0.87
50.64
0.51
6
6.75
7.01
6.69
6.33
6.19
6.70
6.64
6.77
6.09
6.61
6.28
6.12
6.61
6.48
6.50
5.95
6.57
6.23
7.11
6.24
6.48
6.31
6.40
6.55
5.86
5.94
6.46
6.09
5.78
5.98
5.80
5.86
6.58
6.10
5.95
5.63
5.67
5.39
5.28
6.31
5.28
5.43
5.42
5.49
5.29
5.16
5.20
5.20
5.16
4.87
5.37
5.14
5.33
4.98
5.01
5.17
4.74
4.93
5.14
4.09
4.73
4.61
0.86
46.42
0.50
6
6.74
6.81
6.53
6.43
6.62
6.61
6.87
6.64
6.26
6.67
6.36
6.09
6.57
6.55
6.58
6.07
6.62
6.31
6.70
6.29
6.45
6.31
6.35
6.45
5.82
6.15
6.45
6.04
5.87
5.81
5.91
5.86
6.64
6.03
6.19
5.75
5.75
5.38
5.29
6.38
5.36
5.59
5.36
5.38
5.41
5.28
5.27
5.15
5.12
4.99
5.29
5.22
5.42
4.94
4.88
4.84
4.89
5.00
4.94
4.11
4.16
4.45
0.88
65.69
0.61
6
6.59
6.74
6.53
6.40
6.62
6.60
6.87
6.59
6.27
6.62
6.31
6.23
6.53
6.58
6.56
6.18
6.51
6.21
6.71
6.22
6.45
6.47
6.46
6.42
6.02
6.12
6.38
6.03
5.71
5.68
5.91
6.03
6.60
6.21
6.18
5.83
5.70
5.42
5.19
6.35
5.66
5.64
5.30
5.17
5.21
5.16
5.30
5.36
5.37
4.96
5.29
5.41
5.36
5.00
5.53
4.82
5.35
5.01
5.13
4.39
4.10
5.20
0.85
52.81
0.50
5
6.77
6.65
6.54
6.66
6.72
6.64
6.92
6.71
6.32
6.57
6.74
6.12
6.59
6.47
6.64
6.16
6.67
6.38
6.66
5.91
6.48
6.12
6.42
6.33
5.79
6.28
6.35
6.06
6.00
5.80
5.91
5.97
6.65
6.06
6.11
5.93
5.72
5.44
5.38
6.89
5.31
5.61
5.40
5.30
5.44
5.25
5.38
5.24
5.24
4.82
5.31
5.07
5.50
4.98
4.81
4.76
4.75
5.22
4.92
5.00
4.17
4.38
0.90
67.84
0.57
6
6.75
6.90
6.70
6.39
5.91
6.49
6.80
6.68
6.19
6.55
6.34
6.22
6.61
6.47
6.48
6.02
6.67
6.49
7.09
6.21
6.44
6.22
6.34
6.42
5.82
6.06
6.48
6.06
5.87
5.87
5.78
5.86
6.57
6.20
6.08
5.73
5.81
5.40
5.32
6.37
5.31
5.45
5.40
5.42
5.36
5.34
5.25
5.16
5.13
4.92
5.32
5.16
5.39
4.96
4.83
5.12
4.86
5.00
4.76
4.19
4.71
4.31
0.88
57.76
0.56
6
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1.1
1.3
1.3
1.4
1.7
1.7
1.9
1.9
2.4
2.4
3.2
3.3
3.4
4.4
4.6
5.4
5.9
6.5
6.7
6.8
7.8
9.3
9.6
11
15
1.5
17
1.6
18
1.7
20
1.9
41
3.9
43
4.1
43
4.2
76
7.3
R2 b
Ftest
c
R2
d
cv
Ne
Nf
q2
62
54
54
54
62
54
54
54
g
0.40
0.13
0.52
0.06
-0.14
0.09
0.44
0.11
0.57
0.05
-0.14
0.12
2CV
SD (σ) q2
h
h
2CV
5CV
q2
g
5CV
SD (σ) q2
ave. q2
i
Random
max. q2
j
Random
field contributions
hydrophobic
electrostatic
steric
0.449
0.311
0.240
0.422
0.366
0.212
0.432
0.323
0.245
0.478
0.288
0.234
0.418
0.342
0.240
0.411
0.350
0.239
0.389
0.365
0.246
0.450
0.311
0.239
a
b
Bold values represent predicted activities of compounds that were not included in the training set. Correlation coefficient. c Ratio of
R2 explained to unexplained ) R2/(1-R2). Cross-validated correlation coefficient after leave-one-out procedure. e Optimal number of principal
d
components. f Number of compounds. Average cross-validated correlation coefficient for 100 trials using the indicated number of cross-
g
validation groups. Standard deviation of average cross-validated correlation coefficient for 100 trials. i Average cross-validated correlation
h
j
coefficient obtained for random pIC50 runs. Maximum cross-validated correlation coefficient obtained for random pIC50 runs.

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5179
F igu r e 6. Plots of experimental pIC50 versus predicted pIC50 values for the CoMSIA models of bisphosphonates inhibiting
Leishmania major FPP synthase. A, 62 compound training set, Gasteiger charges; B, results of the three, eight-compound test
set prediction studies, Gasteiger charges; C, as A but MSK charges; D, as B but MSK charges. The straight lines represent the
ideal 45° slopes.
predictions were 0.42 for the Gasteiger-Marsili charge
set, 0.43 for the MSK charge set. Overall, a slight
improvement over the CoMFA results, Table 1.
using hydrophobic, electrostatic, and steric descriptors
as opposed to the use of only electrostatic and steric
descriptors as in the CoMFA analysis, although the
electrostatic contributions were the same in both analy-
ses, contributing ∼1/3 of the overall field.
The molecular field map results from the training set
(N ) 62) equations are shown in Figure 7. The hydro-
phobic field contributions are shown in Figure 7A
(yellow-favored; white-disfavored) while the electrostatic
contributions are shown in Figure 7B (blue-positive
charge region; red-negative charge region) and the steric
contributions are shown in Figure 7C (green-favored;
yellow-disfavored). The individual contributions of each
type of descriptor are given in Table 3. The hydrophobic
field contributions make the largest contribution (42%)
to the overall field and the contour map suggests that
enhanced activity might be obtained by further ring
substitution. Figure 7B shows a pronounced positive
charge field feature (shown in blue), which encompasses
the N-substituted imidazole feature of 4, (zoledronate:
the most active species; Figure 3). This positive charge
feature (blue region) accounts for the high activities of
4, 5, 6, 7, and 9 (and structurally related compounds)
since it almost completely encloses the positive charge
region identified in the Merz-Singh-Kollman charge
calculations in each of these compounds. Figure 7C
shows the steric fields, with the green (attractive)
hydrophobic feature clearly encompassing several car-
bons in the aromatic ring, which can be expected to
contribute to the high overall activity of the aryl
bisphosphonates versus the alkyl bisphosphonates. With
the CoMSIA analysis, we obtained better results by
Activities of In d ivid u a l Com p ou n d s. Finally, we
discuss some of the more obvious chemical structure-
activity relationships which can be made for the com-
pounds investigated, and make additional correlations
with bisphosphonate activity in the literature. Com-
pounds 4-32 all have IC50 values <1 µM (corresponding
to Ki values <100 nM) and are, therefore, quite potent
inhibitors of L. major FPP synthase. The majority of
the most active compounds (4, 5, 6, 8, 9, 10, 11) contain
both a positive charge feature and an aromatic ring
feature, which can presumably help delocalize the
charge, although charge delocalization per se is not
essential, since this will not occur in other active species,
such as 7, 12, 14, and 19. The actual spatial localization
of the positive charge also appears to be relatively
flexible (varying from R to γ and δ positions) and
presumably reflects a relatively diffuse electrostatic field
stabilization by the protein of the putative carbocation.
That is, the carbocation is not expected to be stabilized
by a specific anionic group (which would likely be
reactive), but rather by a more delocalized interaction,
such as a cation-π interaction or a more global charge
field (electrostatic field) effect. The preferred location
of the positive charge feature is clearly identified in the
CoMSIA electrostatic fields. Interestingly, some of the

5180 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
the positive charge feature too distal to the bisphospho-
nate backbone to mimic a GPP carbocation intermedi-
ate. Compounds which lack a steric feature (like a ring,
as indicated by the green region in Figure 7C), are
expected to have lower activity than species where this
feature is present. The inactivity of 49, which has both
a carbon ring feature and a positive charge feature, is
expected, since the positive charge feature is ꢀ to the
bisphosphonate backbone19 and falls outside the region
identified by the CoMSIA electrostatic field (Figure 7B).
However, this does not mean that bisphosphonates
containing ꢀ nitrogens will always be inactive. For
example, in the context of aminomethylene bisphospho-
nates, the p-pyridyl bisphosphonate (66) shown below:
has been shown to be extremely active as an herbicide,39
most likely due to the possibility of strong quinonoid
resonance stabilization which delocalizes the positive
charge (as in the amidinium-like species 5, 8, 10, etc.).
Similarly, 21 is also capable of this type of resonance
stabilization and is quite active in our assay. Notably,
while a meta nitrogen does not enable either an ami-
dinium or quinonoid-like resonance stabilization, species
such as risedronate (9) are of course very active, perhaps
due to the positive charge feature being “ideally” located
for interaction with the protein. The presence of a polar
functionality at the para position, e.g. in the following
para-methoxy, p-hydroxy and p-nitro species (67-69):
F igu r e 7. CoMSIA fields for L. major FPP synthase inhibition
by bisphosphonates. A, Hydrophobic fields. Yellow regions
indicate where hydrophobic groups are expected to enhance
activity; white regions indicate where hydrophobic groups are
expected to reduce activity. B, Electrostatic fields. Blue regions
indicate where positive charge is expected to enhance activity;
red regions indicate where negative charge is expected to
enhance activity. C, Steric fields. Green regions indicate where
steric bulk is expected to enhance activity; yellow regions
indicate where steric bulk is expected to reduce activity.
general patterns of activity seen in L. major FPP
synthase inhibition have also been observed in very
early patent literature from Nissan where 5, 11, 15, and
35 were identified as herbicides.39 In that work, the
effects of ring substitution were investigated, and it was
found that the 3′ (5), 4′ (11), and 5′ species were highly
active, while the 6′ (35) and 4′,6′ dimethyl species were
very much less active, completely consistent with our
observations vs L. major FPP synthase. Since Zeneca
have shown that 5 is a ∼25 nM inhibitor of a farnesyl
pyrophosphate synthase from daffodils,15 it seems that
the herbicidal activity of such bisphosphonates, as well
as their antiparasitic12 and bone resorption activity,19
may all be attributable to FPP synthase inhibition.
The second set of compounds (33-53) are less active
and have IC50 values in the range ∼1-10 µM (Table 1).
Eight compounds (42, 46, 50, 51, 52, and 53) lack a
positive charge (nitrogen) feature, which accounts for
their lower activity, while five (32, 36, 47, 48, 49) have
removes essentially all herbicidal activity,39 consistent
with our observation that the p-amino species (64) has
low activity vs L. major FPP synthase. The origins of
the lower activities of 35 (versus e.g. 5 or 11) and 37
(versus 15) suggest a steric effect may also be operative
in both compounds. A purely electronic effect of Br in
37 on the pKa of the nitrogen seems unlikely since the
Cl in 15 is expected to be a better electron-withdrawing
group. Consequently, it appears likely that there may
be steric crowding contributing to the decreasing activity
seen on moving from 5 (3′ ring position), to 11 (4′ ring
position), to 15 and 37 (5′ ring position) and finally to
35 (6′ ring position). This is consistent with the steric
and hydrophobic field results shown in Figure 7 and,
indeed, ring methylation in compounds 70 and 71 has
been shown to contribute a factor of 100 to inactivity,19
clearly indicating that a single methyl group can have

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5181
pH adjustment to 4.3 with a 50% NaOH solution, followed by
recrystallization from H2O:12,26
a large repulsive interaction. The activity of 22 is to be
expected, since the additional carbons are in the 3′,4′
positions of the pyridine ring, where methylations do
not compromise activity, consistent with the favorable
hydrophobic field result shown in Figure 7A.
Zoledronate (4) was made from the corresponding imidazol-
1-ylacetic acid,40 which was prepared from imidazole using a
BBDE Cl41 phase-transfer catalyst:
The least active compounds (54-64) have IC50 values
in the ∼11-76 µM range (Table 1). 59 lacks the positive
charge feature, 60 lacks the ring or aliphatic carbon
feature, 61 lacks the positive charge feature, and 64 has
a ú-N, distal from the bisphosphonate backbone. In
addition, as noted above, the theoretical results suggest
that 63 and 64 are not very basic. While these comments
are clearly only descriptive, they do give a potentially
useful qualitative description of the key features which
contribute to bisphosphonate activity and help put the
quantitative CoMFA and CoMSIA modeling approaches
on a perhaps more readily understood structural basis,
which should facilitate future drug design.
The aminomethylene bisphosphonates were also synthesized
basically as described before,12 by reacting stoichiometric
amounts of the corresponding amines, triethyl orthoformate
and diethyl phosphite,12,26 followed by acid hydrolysis, for
example:
Con clu sion s
The results we have described above are of interest
for a number of reasons. First, we have made the first
detailed investigation of the inhibition of a recombinant
farnesyl pyrophosphate synthase enzyme by 62 1-hy-
droxy-1,1-bisphosphonates and aminomethylene bis-
phosphonates. Second, we have made the first detailed
CoMFA and CoMSIA investigations of the inhibition of
an expressed farnesyl pyrophosphate synthase by bis-
phosphonates. The CoMFA and CoMSIA training set
results have theory-versus-experiment R2 correlations
of ∼0.87 and ∼0.88, respectively, with F-test values of
∼61 (CoMFA) and ∼66 (CoMSIA). The CoMFA and
CoMSIA results were validated by using two cross-
validation and data randomization techniques. Third,
we have used a training/test set approach to predict the
activities of 24 test set compounds: pIC50 errors were
∼0.46 (CoMFA) and ∼0.43 (CoMSIA), corresponding to
average prediction errors of ∼2.9 and ∼2.7. Fourth, we
report CoMSIA field maps which help describe some of
the key features of the bisphosphonates important in
enzyme inhibition. Fifth, we have provided a descriptive
analysis of the activities of the compounds investigated,
and made comparisons with the activities of these and
other bisphosphonate FPP synthase inhibitors active as
herbicides. These comparisons show a number of inter-
esting trends related to ring methyl substitution and
activity, the position of ring nitrogens and activity, the
likely effects of resonance stabilization, and in some
cases the deleterious effects of incorporation of polar
substitutions on the aryl rings, which should be of use
in the development of other FPP synthase inhibitors.
The purity of all samples was verified by microchemical
analysis (H/C/N) and in some cases via quantitative H NMR
1
spectroscopy. Compounds 5-7, 9, 11, 15, 18, 19, 24, 26, 32,
33, 36, 37, 39, 40, 42, 46-50, 54, 56, 59-61, and 64 were
available from previous work12,26,42. All compounds not previ-
ously reported, with the exception of the compounds discussed
below, had experimental H/C/N analyses that agreed within
0.4% of the calculated values. 4: Anal. (C5H10N2O7P2‚H2O) C,
H; N: calcd, 9.66; found, 9.25. 13: Anal. (C6H10N2O6P2‚0.5H2O)
C, N; H: calcd, 4.00; found, 3.56. 17: Anal. (C12H19NO7P2Na2‚
0.5 H2O) C, N; H: calcd, 4.96; found, 5.37. 21: Anal.
(C7H11N2O7P2Na) C, N; H: calcd, 3.65; found, 4.19. 22: Anal.
(C10H13N2O6P2) C, N; H: calcd, 4.11; found, 3.53. 28: Anal.
(C7H17NO6P2) H, N; C: calcd, 30.78; found, 30.35. 38: Anal.
(C4H13NO6P2‚H2O) C, N; H: calcd, 6.02; found, 5.55. 43: Anal.
(C9H21NO6P2) H, N; C: calcd, 35.89; found, 35.44. 58: Anal.
(C13H27NO6P2) C, N; H: calcd, 7.66; found, 8.32. Full analytical
data for all compounds not previously reported is provided in
the Supporting Information.
Cr ysta llogr a p h ic Asp ects. For zoledronate (4), crystals
were grown from water in their zwitterionic form. For com-
pound 15, crystals of the nonprotonated form were obtained
by vapor diffusion of ethanol into an aqueous solution (con-
taining Na+, pH ) 5.5), while the crystals of the protonated
form were grown from water. Single-crystal data were collected
on a Siemens (Madison, WI) Platform/CCD diffractometer
using the SHELXTL system.43 Five frame series were filtered
for statistical outliers then corrected for absorption by integra-
tion using SHELXTL/ XPREP. Crystal structures were solved
by direct methods.44 Donor-H atom positions were refined
under restraints to idealized bond distances, with an effective
standard deviation of 0.03 Å. The remaining H atoms were
included as riding idealized contributors [Uiso ) 1.2 or 1.5
Exp er im en ta l Section
Syn th etic Asp ects. The bisphosphonates investigated were
synthesized and characterized basically according to the
methods described in refs 12 and 26. In short, 1-hydroxyal-
kylidene-1,1 bisphosphonates were prepared by reaction of the
appropriate carboxylic acid with phosphorous acid and phos-
phorus trichloride, followed by hydrolysis and in most cases

5182 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24
Sanders et al.
Ueq(C)]. The space group choices were confirmed by successful
convergence of the full-matrix least-squares refinement on F2.44
Final analyses of variance between observed and calculated
structure factors showed no dependence on amplitude or
resolution. The crystallographic data, space group, and other
information related to the crystal structure determination are
summarized in the Supporting Information.
tables). This information is available free of charge via the
Internet at http://pubs.acs.org.
Refer en ces
(1) http://www.who.int/inf-fs/en/fact116.html. The Leishmaniases
and Leishmania/HIV Co-Infections. Fact Sheet No. 116. Revised
May 2000.
L. m a jor F P P Syn th a se Exp r ession . The coding sequence
of the Leishmania major farnesyl diphosphate synthase gene
was cloned in the expression vector pET11c to give pETLm-
FPPS. After induction with 1 mM IPTG for 4 h at 37 °C, FPPS
from soluble BL21(DE3)/pETLmFPPS extracts was purified
using a combination of ammonium sulfate precipitation and
HAP chromatography.
(2) Bora, D. Epidemiology of visceral leishmaniasis in India. Natl.
Med. J . India 1999, 12, 62-68.
(3) Croft, S. L.; Yardley, V. Chemotherapy of leishmaniasis. Curr.
Pharm. Des. 2002, 8, 319-342.
(4) Roberts, W. L.; McMurray, W. J .; Rainey, P. M. Characterization
of the antimonial antileishmanial agent meglumine antimonate
(Glucantime). Antimicrob. Agents Chemother. 1998, 42, 1076-
1082.
En zym e In h ibition by Bisp h osp h on a tes. Geranyl py-
rophosphate and isopentenyl pyrophosphate were from Sigma
(St. Louis, MO) and [14C] isopentenyl pyrophosphate (40-60
mCi/mmol) from NEN Life Sciences (Boston, MA). To deter-
mine the effects of BPs on FPP synthase activity, drugs were
tested at a range of concentrations, from 0.1 to 100 µM. The
analyses were carried out in 100 µL reaction mixtures contain-
ing 850 nmol of potassium phosphate buffer (pH 7.0), 85 nmol
of MgCl2, 10 nmol of geranyl pyrophosphate, and 5 nmol of
isopentenyl pyrophosphate (specific activity 6.34 mCi/mmol).
The enzyme reaction was initiated by the addition of 10 ng of
pure recombinant farnesyl diphosphate synthase, incubated
for 15 min at 30 °C and terminated by addition of 1 mL of an
aqueous saturated NaCl solution. Radiolabeled FPP was
extracted with 1 mL of 1-butanol. After thorough mixing and
brief centrifugation, 0.5 mL of the alcohol phase was removed
and radioactivity determined by using a scintillation counter.
Enzyme inhibition results are expressed as IC50 values, which
represent the drug concentrations required to reduce enzyme
activity by 50%, as pIC50 values ( ) -log IC50(M)) and as Ki
values and are collected in Table 1.
3D-QSAR/CoMF A/CoMSIA. CoMFA and CoMSIA analy-
ses were performed by using the Sybyl 6.9 program29 with
default settings. Molecular mechanics calculations were per-
formed by using the Tripos force field, with a convergence
criterion requiring a minimum RMS gradient of 0.01 kcal/mol‚
Å at the steepest descent step and an RMS gradient of 0.001
kcal/mol‚Å at the Powell and BFGS steps. Structures were
optimized to convergence at each minimization step. Atomic
charge calculations for CoMFA and CoMSIA analyses were
performed on the minimized structures using the Gateiger-
Marsili method30 in Sybyl 6.9 and also by using Hartree-Fock
theory with a 6-31G* basis set, by using the Merz-Singh-
Kollman method36 in the Gaussian 98 program.37 CoMFA
energies and CoMSIA indices were calculated on a rectangular
grid containing the aligned molecules by using steric and
electrostatic probes and hydrophobic, steric and electrostatic
probes, respectively. The atomic coordinates of the models were
used to compute field values at each point of a 3D grid using
a grid spacing of 2.00 Å. To obtain a quantitative analysis of
the dependece of enzymatic activity on CoMFA and CoMSIA
parameters, PLS analysis was applied, leading to the results
shown in Tables 1-3.
(5) Grogl, M.; Thomason, T. N.; Franke, E. D. Drug resistance in
leishmaniasis: its implication in systemic chemotherapy of
cutaneous and mucocutaneous disease. Am. J . Trop. Med. Hyg.
1992, 47, 117-126.
(6) Adler-Moore, J .; Proffitt, R. T. AmBisome: liposomal formula-
tion, structure, mechanism of action and pre-clinical experience.
J . Antimicrob. Chemother. 2002, 49, 21-30.
(7) Unger, C.; Maniera, T.; Kaufmann-Kolle, P.; Eibl, H. In vivo
antileishmanial activity of hexadecylphosphocholine and other
alkylphosphocholines. Drugs Today 1998, 34, 133-140.
(8) Escobar, P.; Yardley, V.; Croft, S. L. Activities of hexadecylphos-
phocholine (Miltefosine), AmBisome, and sodium stibogluconate
(Pentostam) against Leishmania donovani and immunodeficient
scid mice. Antimicrob. Agents Chemother. 2001, 45, 1872-1875.
(9) Alrajhi, A. A.; Ibrahim, E. A.; De Vol, E. B.; Khairat, M.; Faris,
R. M.; Maguire, J . H. Fluconazole for the treatment of cutaneous
leishmaniasis caused by Leishmania major. N. Engl. J . Med.
2002, 346, 891-895.
(10) Geusens, P.; McClung, M. Review of risedronate in the treatment
of osteoporosis. Expert Opin. Pharmacother. 2001, 2, 2011-2025.
(11) Pistevou-Gombaki, K.; Eleftheriadis, N.; Sofroniadis, I.; Makris,
P.; Kouloulias, V. Palliative treatment of painful bone me-
tastases from non-Hodgkin lymphoma with disodium pamidr-
onate. J . Exp. Clin. Cancer Res. 2002, 21, 429-432.
(12) Martin, M. B.; Grimley, J . S.; Lewis, J . C.; Heath, H. T. III;
Bailey, B. N.; Kendrick, H.; Yardley, V.; Caldera, A.; Lira, R.;
Urbina, J . A.; Moreno, S. N.; Docampo, R.; Croft, S. L.; Oldfield,
E. Bisphosphonates inhibit the growth of Trypanosoma brucei,
Trypanosoma cruzi, Leishmania donovani, Toxoplasma gondii,
and Plasmodium falciparum: a potential route to chemotherapy.
J . Med. Chem. 2001, 44, 909-916.
(13) Yardley, V.; Khan, A. A.; Martin, M. B.; Slifer, T. R.; Araujo, F.
G.; Moreno, S. N.; Docampo, R.; Croft, S. L.; Oldfield, E. In vivo
activities of farnesyl pyrophosphate synthase inhibitors against
Leishmania donovani and Toxoplasma gondii. Antimicrob.
Agents Chemother. 2002, 46, 929-931.
(14) Rodriguez, N.; Bailey, B. N.; Martin, M. B.; Oldfield, E.; Urbina,
J . A.; Docampo, R. Radical cure of experimental cutaneous
leishmaniasis by the bisphosphonate pamidronate. J . Infect. Dis.
2002, 186, 138-140.
(15) Cromartie, T. H.; Fisher, K. J .; Grossman, J . N. The discovery
of a novel site of action for herbicidal bisphosphonates. Pesticide
Biochem. Physiol. 1999, 63, 114-126.
(16) van Beek, E.; Pieterman, E.; Cohen, L.; Lo¨wik, C.; Papapoulos,
S. Farnesyl pyrophosphate synthase is the molecular target of
nitrogen-containing bisphosphonates. Biochem. Biophys. Res.
Commun. 1999, 264, 108-111.
(17) Keller, R. K.; Fliesler, S. J . Mechanism of aminobisphosphonate
action: characterization of alendronate inhibition of the iso-
prenoid pathway. Biochem. Biophys. Res. Commun. 1999, 266,
560-563.
Ack n ow led gm en t. We thank Scott Wilson for his
help with the crystallographic aspects of the paper. This
work was supported in part by the United States Public
Health Service (National Institutes of Health Grant
GM-50694), the National Computational Science Alli-
ance (grant MCB000020), an EC INCO-DEV CT2001-
10074 contract and by the Plan Andaluz de Investiga-
cion (Cod. CVI-199). J .M. and J .M.S. are American
Heart Association, Midwest Affiliate Predoctoral Fel-
lows. G.M. is a USPHS NRSA Postdoctoral Fellow (NIH
grant GM65782). A.O. is a Spanish Ministry of Science
and Technology Predoctoral fellow.
(18) Grove, J . E.; Brown, R. J .; Watts, D. J . The intracellular target
for the antiresorptive aminobisphosphonate drugs in Dictyoste-
lium discoideum is the enzyme farnesyl diphosphate synthase.
J . Bone Miner. Res. 2000, 15, 971-981.
(19) Dunford, J . E.; Thompson, K.; Coxon, F. P.; Luckman, S. P.;
Hahn, F. M.; Poulter, C. D.; Ebetino, F. H.; Rogers, M. J .
Structure-activity relationships for inhibition of farnesyl diphos-
phate synthase in vitro and inhibition of bone resorption in vivo
by nitrogen-containing bisphosphonates. J . Pharmacol. Exp.
Ther. 2001, 296, 235-242.
(20) Montalvetti, A.; Bailey, B. N.; Martin, M.; Severin, G.; Oldfield,
E.; Docampo, R. Bisphosphonates are potent inhibitors of
Trypanosoma cruzi farnesyl pyrophosphate synthase. J . Biol.
Chem. 2001, 276, 33930-33937.
(21) Montalvetti, A.; Fernandez, A.; Sanders, J . M.; Ghosh, S.; Van
Brussel, E.; Oldfield, E.; Docampo, R. Farnesyl pyrophosphate
synthase is an essential enzyme in Trypanosoma brucei: in vitro
RNA interference and in vivo inhibition studies. J . Biol. Chem.
2003, 278, 17075-17083.
Su p p or tin g In for m a tion Ava ila ble: Crystallographic
data for 4 and 15 (nonprotonated and protonated forms, 19

Bisphosphonates as FPP Synthase Inhibitors
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 24 5183
(22) Rosenthal, E.; Marty, P.; del Giudice, P.; Pradier, C.; Ceppi, C.;
Gastaut, J . A.; Le Fichoux, Y.; Cassuto, J . P. HIV and Leish-
mania coinfection: a review of 91 cases with focus on atypical
locations of Leishmania. Clin. Infect. Dis. 2000, 31, 1093-1095.
(23) Go´mez, A. O.; Gonza´lez-Pacanowska, D., unpublished results.
(24) Cheng, Y.; Prusoff, W. H. Relationship between the inhibition
constant (KI) and the concentration of inhibitor which causes
50% inhibition (I50) of an enzymatic reaction. Biochem. Phar-
macol. 1973, 22, 3099-3108.
(25) Cramer, R. D., III; Patterson, D. E.; Bunce, J . D. Comparative
molecular field analysis (CoMFA). 1. Effect of shape on binding
of steroids to carrier proteins. J . Am. Chem. Soc. 1988, 110,
5959-5967.
(26) Martin, M. B.; Sanders, J . M.; Kendrick, H.; de Luca-Fradley,
K.; Lewis, J . C.; Grimley, J . S.; Van Brussel, E. M.; Olsen, J . R.;
Meints, G. A.; Burzynska, A.; Kafarski, P.; Croft, S. L.; Oldfield,
E. Activity of bisphosphonates against Trypanosoma brucei
rhodesiense. J . Med. Chem. 2002, 45, 2904-2914.
(27) Szabo, C. M.; Martin, M. B.; Oldfield, E. An investigation of bone
resorption and Dictyostelium discoideum growth inhibition by
bisphosphonate drugs. J . Med. Chem. 2002, 45, 2894-2903.
(28) Kotsikorou, E.; Oldfield, E. A quantitative structure-activity
relationship and pharmacophore modeling investigation of aryl-X
and heterocyclic bisphosphonates as bone resorption agents. J .
Med. Chem. 2003, 46, 2932-2944.
(29) SYBYL 6.9, Tripos Inc., 1699 South Hanley Rd., St. Louis,
Missouri, 63144.
(30) Gasteiger, J .; Marsili, M. Iterative partial equalization of orbital
electronegativity - a rapid access to atomic charges. Tetrahedron
1980, 36, 3219-3228.
(31) Matter, H.; Kotsonis, P.; Klingler, O.; Strobel, H.; Frohlich, L.
G.; Frey, A.; Pfleiderer. W.; Schmidt, H. H. Structural require-
ments for inhibition of the neuronal nitric oxide synthase (NOS-
I): 3D-QSAR analysis of 4-oxo- and 4-amino-pteridine-based
inhibitors. J . Med. Chem. 2002, 45, 2923-2941.
(32) Martin, M. B.; Arnold, W.; Heath, H. T. III; Urbina, J . A.;
Oldfield, E. Nitrogen-containing bisphosphonates as carbocation
transition state analogues for isoprenoid biosynthesis. Biochem.
Biophys. Res. Commun. 1999, 263, 754-758.
(33) Advanced Chemistry Development (ACD) Software Solaris
V4.67.
pamidronate disodium (Aredia) to zoledronic acid (Zometa). J .
Med. Chem. 2002, 45, 3721-3738.
(36) Besler, B. H.; Merz, K. M.; Kollman, P. A. 1990 Atomic charges
derived from semiempirical methods. J . Comput. Chem. 1990,
11, 431-439.
(37) Frisch, M. J .; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.;
Robb, M. A.; Cheeseman, J . R.; Zakrzewski, V. G.; Montgomery,
J . A., J r.; Stratmann, R. E.; Burant, J . C.; Dapprich, S.; Millam,
J . M.; Daniels, A. D.; Kudin, K. N.; Strain, M. C.; Farkas, O.;
Tomasi, J .; Barone, V.; Cossi, M.; Cammi, R.; Mennucci, B.;
Pomelli, C.; Adamo, C.; Clifford, S.; Ochterski, J .; Petersson, G.
A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, D. K.; Rabuck,
A. D.; Raghavachari, K.; Foresman, J . B.; Cioslowski, J .; Ortiz,
J . V.; Baboul, A. G.; Stefanov, B. B.; Liu, G.; Liashenko, A.;
Piskorz, P.; Komaromi, I.; Gomperts, R.; Martin, R. L.; Fox, D.
J .; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.;
Challacombe, M.; Gill, P. M. W.; J ohnson, B.; Chen, W.; Wong,
M. W.; Andres, J . L.; Gonzalez, C.; Head-Gordon, M.; Replogle,
E. S.; Pople, J . A. Gaussian 98, Revision A.9, Gaussian, Inc.,
Pittsburgh, PA, 1998.
(38) Klebe, G.; Abraham, U.; Mietzner, T. Molecular similarity indices
in
a comparative analysis (CoMSIA) of drug molecules to
correlate and predict their biological activity. J . Med. Chem.
1994, 37, 4130-4146.
(39) Suzuki, F.; Fujikawa, Y.; Yamamoto, S.; Mizutani, H.; Ohya, T.;
Ikai, T.; Oguchi, T. N-Pyridylaminomethylene-diphosphonic and
compounds. UK Patent Application, 2 004 888 A, 1978.
(40) Gil, S.; Zaderenzo, P.; Cruz, F.; Cerda´n, S.; Ballesteros, P.
Imidazol-1-ylalkanoic acids as extrinsic 1H NMR probes for the
determination of intracellular pH, extracellular pH and cell
volume. Bioorg. Med. Chem. 1994, 2, 305-314.
(41) Alvarez-Builla, J .; Vaquero, J . J .; Garcia Navio, J . L.; Cabello,
J . F.; Sunkel, C.; Fau de Casa-J uana, M.; Dorrego, F.; Santos,
L. 1,5-bis-(N-benzyl-N, N-dimethylammonium) diethyl ether,
dichloride (BBDE Cl). A novel bis-ammonium salt as phase
transfer catalyst. Tetrahedron 1990, 46, 967-978.
(42) Szabo, C. M.; Matsumura, Y.; Fukura, S.; Martin, M. B.; Sanders,
J . M.; Sengupta, S.; Cieslak, J . A.; Loftus, T. C.; Lea, C. R.; Lee,
H. J .; Koohang, A.; Coates, R. M.; Sagami, H.; Oldfield, E.
Inhibition of geranylgeranyl diphosphate synthase by bisphos-
phonates and diphosphates:
a potential route to new bone
(34) Matczak-J on, E.; Sawka-Dobrowolska, W.; Kafarski, P.; Vide-
nova-Adrabinska, V. Molecular organization and solution prop-
erties of N-substituted aminomethane-1,1-diphosphonic acids.
New J . Chem. 2001, 11, 1447-1457.
(35) Widler, L.; J aeggi, K. A.; Glatt, M.; Muller, K.; Bachmann, R.;
Bisping, M.; Born, A. R.; Cortesi, R.; Guiglia, G.; J eker, H.; Klein,
R.; Ramseier, U.; Schmid, J .; Schreiber, G.; Seltenmeyer, Y.;
Green, J . R. Highly potent geminal bisphosphonates. From
antiresorption and antiparasitic agents. J . Med. Chem. 2002,
45, 2185-2196.
(43) Bruker 2001. SMART (Version 5.625), SAINT (Version 6.22) and
SHELXTL (Version 6.12). Bruker AXS Inc., Madison, WI.
(44) Sheldrick, G. M. 2001. SHELXL97-2. University of Go¨ttingen,
Germany.
J M0302344