Design, synthesis, and evaluation of potential inhibitors of nitric oxide synthase

Article abstract of DOI:10.1016/j.bmc.2008.04.036

Selective inhibitors of neuronal nitric oxide synthase (nNOS) were shown to protect brain and may be useful in the treatment of neurodegenerative diseases. In this context, our purpose has been to design and synthesize a new family of derivatives of thiad

Full text of DOI:10.1016/j.bmc.2008.04.036

Available online at www.sciencedirect.com
Bioorganic & Medicinal Chemistry 16 (2008) 6193–6206
Design, synthesis, and evaluation of potential inhibitors
of nitric oxide synthase
ꢀ
´
´
Tania Castano, Arantxa Encinas, Concepcion Perez, Ana Castro,
˜
Nuria E. Campillo* and Carmen Gil*
Instituto de Quı´mica Me´dica, (CSIC), Juan de la Cierva 3, 28006 Madrid, Spain
Received 29 January 2008; revised 8 April 2008; accepted 16 April 2008
Available online 18 April 2008
Abstract—Selective inhibitors of neuronal nitric oxide synthase (nNOS) were shown to protect brain and may be useful in the treat-
ment of neurodegenerative diseases. In this context, our purpose has been to design and synthesize a new family of derivatives of
thiadiazoles as possible inhibitors of nNOS. To achieve it a supervised artificial neural network model has been developed for the
prediction of inhibition of Nitric Oxide Synthase using a dataset of 119 nNOS inhibitors. The definition of the molecules was
achieved from a not-supervised neural network using a home made program named CODES. Also, thiadiazole-based heterocycles,
previously predicted, were prepared as conformationally restricted analogues of
N-phenylisothiourea.
a selective nNOS inhibitor, S-ethyl
ꢀ 2008 Elsevier Ltd. All rights reserved.
1. Introduction
in the origin and in the development of neurodegenera-
tive diseases such as Alzheimer’s disease (AD).5 The fact
that the oxidative damage found in some biomolecules is
an event that precedes the characteristic lesions of AD
namely, senile plaques and neurofibrillary tangles, sup-
ports the idea that oxidative stress is an early event in
neurodegeneration.6,7 Thus, drugs that directly scavenge
free radicals or inhibit enzymes implicated in their pro-
duction could be therapeutically useful in AD.8 One of
these enzymes is nitric oxide synthase (NOS) which pro-
duces nitric oxide (NO). Excessive production of NO by
activated glial cells contributes to inflammation-related
neurodegenerative processes. In fact, expression and
activity of NOS play an important role in the mainte-
nance and increase of NO liberation in neurons which
makes it an attractive therapeutic objective in the search
of neuronal protective agents.9
Neuroprotection is one of the major challenges of mod-
ern medicine for the treatment of chronic, progressive
and neurodegenerative disorders such as Alzheimer’s
disease, Parkinson’s disease, and multiple sclerosis.
The protection of neurons and their synapses against
damage and death, and the preservation of their func-
tions could provide effective cure for these patholo-
gies.1–3
The brain is particularly vulnerable to oxidative damage
because of the high rate of oxygen utilization, the high
content of oxidizable polyunsaturated fatty acids, and
of redox-active transition metal ions that can generate
free radicals. Moreover, the relative dearth of antioxi-
dant protective defenses renders the central nervous sys-
tem (CNS) particularly prone to oxidative damage.4 It is
widely accepted that oxidative stress increases during
aging and recent evidences suggest its important role
There are three isoenzymes of NOS which include a neu-
ronal enzyme (nNOS), an endothelial enzyme (eNOS),
and an inducible enzyme (iNOS). Although these three
enzymes have cofactor requirements in common and
similar mechanism of action, they are structurally dis-
tinct from one another.10 The research and clinical util-
ity of NOS inhibitors are dependent on both their
specificity and their potency of inhibition. Selective
NOS isoform inhibition to regulate NO synthesis has re-
ceived much attention. Inhibitors with high isoform
Keywords: NOS; Artificial neural network; CODES; QSAR;
Thiadiazole.
*
Corresponding authors. Tel.: +34 915622900; fax: +34 915644853
(NEC; CG); e-mail addresses: nuria@suricata.iqm.csic.es; cgil@iqm.
csic.es
ꢀ
Present address: NeuroPharma, S.A., Avda. de la Industria 52, 28760
Tres Cantos, Madrid, Spain.
0968-0896/$ - see front matter ꢀ 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmc.2008.04.036

6194
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
selectivity are required to further define the role of each
NOS isoform in various biological processes.11 The
majority of described inhibitors to date are analogues
of the endogenous substrate L-arginine.12
tion Line S.L. CODES program is available at http://
www.iqm.csic.es.
This singular way to encode chemical entities converts
CODES into an excellent tool to carry out QSAR and
QSPR studies.37–40 Thus, the numeric descriptors gener-
ated by CODES have been used as input data in feed-
forward back-propagation network to obtain a predict
model (see Section 4).
Due to the fact that nNOS is the primary NO regulator
in neurons, selective inhibitors of this isoenzyme were
shown to protect brain and may be useful in the treat-
ment of neurodegenerative diseases.13,14 The aim of this
article is the study of new-design compounds based on
thiadiazole framework, conformationally restricted ana-
logues of a selective nNOS inhibitor S-ethyl N-phen-
yliosthiourea, as possible nNOS inhibitors. To achieve
this goal we developed a neural network model based
on known nNOS inhibitor to predict their activity.
The synthesis and evaluation of nNOS inhibitors of
thiadiazole derivatives with aryl substituents are also
presented here.
Using the strategy gathered in Figure 2, we have devel-
oped different neural network models for the prediction
of nNOS inhibition.
The first step in this procedure is the codification of the
whole dataset using CODES program, to define the mol-
ecules from a topological point of view. CODES gener-
ates for each molecule a dynamic matrix (A · R), where
A is the number of atoms of each structure and R is the
number of iterations necessary to achieve convergence in
the codification process. In order to have the same num-
ber of descriptors for each structure we developed two
different strategies (see Section 4). One of these two
strategies is using the whole matrix of each compound
and following of the data reduction process to obtain
the same descriptor number for each chemical structure
without losting any structural data (strategy 1). The
other strategy is used to choose the parameters from
the last step of the codification process generated by
CODES. We decided to employ the descriptors of the
four atoms more conserved in all the structures (Fig.
3) (see supporting information, Tables S1 and S2).
2. Results and discussions
2.1. Neural network model
The quantitative structure–activity relationship was per-
formed by means of artificial neural network system using
NOS inhibitor families previously described (Table 1).
This family is formed by isothioureas,15–17 bisisothioure-
as,15–17 arylisothioureas,18 aminoguanidines,19,20 nitro-
guanidines,21–24 arylamidines,25 ornitine derivatives,26
thiazole and thiazoline,17 imidazole,23,27–29 and pyra-
zole30 (see Fig. 1).
As a theoretical tool, the Artificial Neural Networks
(ANNs) are a modeling methodology whose application
in some areas of Medicinal Chemistry such as quantita-
tive structure–property relationship (QSPR), quantita-
tive structure–activity relationship (QSAR) and
prediction of pharmacokinetic properties has increased
spectacularly in recent years.31–35
It is well-known that the quality of QSAR models is af-
fected by the dataset partition. Thus, the original data-
base was randomly divided into two sets, the training
set and the test set. Furthermore, to account for sam-
pling error, models were generated from different train-
ing/test set partitions. Thus, the training set of different
sizes (50–80 compounds) was generated randomly.
Regarding the architecture of the models, specially the
number of neurons in hidden layer was optimized in
each network. Unfortunately, the models developed
with quantitative data using both the strategies were
not statistically acceptable. The results in the training
process using strategy 1 were r values between 0.78
and 0.98, r2cv values between 0.57 and 0.95, and the s val-
ues between 0.17 and 0.58. However, the prediction abil-
ity of the models was poor in the external validation
process. Regarding strategy 2, the results were analo-
gous (s = 0.33–0.46, r = 0.77–0.91, and r2cv ¼ 0:59–0:80).
The development of nonlinear modeling approaches,
such as artificial intelligence-based algorithms, opened
up the field to the concurrent analysis of a wider variety
of structures with potentially varying modes of action
and noncongeneric chemicals. These artificial systems
emulate the function of the brain, where a very high
number of information-processing neurons are intercon-
nected and are known for their ability to model a wide
set of functions, including linear and nonlinear, without
knowing the analytic forms in advance.
To overcome the first step, we used an original home-
made program called CODESꢀ.36 CODES is an efficient
and easy-to-use program to encode chemical structures
by means of neural computing. The molecular descrip-
tors obtained from this method contain all the underly-
ing information of their chemical structure. The original
CODES program was created and developed by Prof.
To overcome these results, we re-addressed the problem
from a qualitative point of view. Thus, the compounds
were clustered into three groups, active (1), moderate
(0), or inactive (ꢀ1) (Section 4). The statistical results to-
gether with the results of the internal validation are
gathered in Table 2 for different training sets and both
the strategies.
´
´
Manfred Stud in Instituto de Quımica Medica (IQM,
Madrid, Spain). At present, CODES program continues
being further developed and optimized by our research
group in collaboration with Advanced Software Produc-
At the first glance, these results indicate that the best
models are model 2 (strategy 1) and model 6 (strategy
2), which show the highest values of r and r2cv and

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6195
Table 1. Ki values of nNOS dataset
Table 1 (continued)
Inhibitor (I)
R
nNOS
(Ki, lM)
Lit.
Inhibitor (I)
R
nNOS
(Ki, lM)
Lit.
I–1
I–2
CH3
0.16
0.029
0.037
0.62
0.63
14
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
15–17
18
I–60
I–61
I–62
I–63
I–64
I–65
I–66
I–67
I–68
I–69
I–70
I–71
I–72
I–73
I–74
I–75
I–76
I–77
I–78
I–79
I–80
I–81
I–82
I–83
I–84
I–85
I–86
I–87
I–88
I–89
I–90
I–91
I–92
I–93
I–94
I–95
I–96
I–97
I–98
I–99
I–100
I–101
I–I02
I–103
I–104
I–105
I–106
I–107
I–108
I–109
I–110
I–111
I–112
I–113
I–114
I–115
I–116
I–117
I–118
I–119
3-CH2SC(NH)NH2
3-SO2NH2
3-NHNH2
0.36
29
25
25
CH2CH3
CH(CH3)2
C(CH3)3
I–3
I–4
1.1
25
25
3-C(NH)NH2
3-NHC(NH)Me
R1 = CH2NH2, R2 = Me
0.57
1.5
I–5
I–6
CH2CH2CH3
CH2Ph
25
25
0.04
R1 = CH2NH2, R2 = CH2NH2 0.19
R1 = CH2NH2, R2 = CH2SMe 0.011
I–7
I–8
CH2CH2Ph
CH2CH2CH2Ph
CH2CH2NH2
CH2CH@C(CH3)2
–(CH2)2-(1,3-Ph)–(CH2)2–
–(CH2)2-(1,4-Ph)–(CH2)2–
–(CH2)-(1,4-Ph)–(CH2)–
–(CH2)3-(1,3-Ph)–(CH2)3–
H
0.80
14
25
25
I–9
1.80
2.00
0.25
0.016
0.1
R1 = CH2NH2, R2 = CH2F
R1 = CH2OH, R2 = CH2F
0.011
0.30
25
25
I–10
I–11
I–12
I–13
I–14
I–15
I–16
I–17
I–18
I–19
I–20
I–21
I–22
I–23
I–24
I–25
I–26
I–27
I–28
I–29
I–30
I–31
I–32
I–33
I–34
I–35
I–36
I–37
I–38
I–39
I–40
I–41
I–42
I–43
I–44
I–45
I–46
I–47
I–48
I–49
I–50
I–51
I–52
I–53
I–54
I–55
R1 = CH2NH2, R2 = 2-pyridiyl 0.33
R1 = CH2NH2, R2 = 2-furanyl 0.0063
25
25
R1 = CH2NH2, R2 = 2-thienyl
H
0.0087
1.7
0.1
3
25
26
4.8
0.12
0.25
0.17
0.45
0.14
0.17
0.56
0.29
0.70
0.19
0.21
0.34
1.6
–CH@CH2
Me
–CH2CH3
26
26
2-Br
2-Cl
18
18
5.3
12
26
17
3-Cl
4-Cl
18
18
R1, R2 = H
R1 = Me, R2 = H
R1 = H, R2 = Me
R1, R2 = Me
R1 = –CH2CH2Ph, R2 = H
—
10
17
17
2-OCH3
3-OCH3
18
18
1.1
0.38
0.89
0.41
125
32
17
17
4-OCH3
4-OCF3
18
18
17
23
4-OPh
4-OCH2Ph
4-OH
18
18
n = 3, R = 4-Br
n = 4, R = 4-Br
n = 6, R = 4-Br
n = 3, R = 4-CF3
n = 4, R = 4-CF3
n = 6, R = 4-CF3
n = 4, R = 2-Br
n = 4, R = 3-Br
n = 2, R = 4-Br
n = 2, R = 4-CF3
n = 0, R = H
n = 1, R = H
n = 2, R = H
n = 3, R = H
n = 4, R = H
n = 1, R = Ph
n = 2, R = Ph
n = 3, R = Ph
—
23
23
18
18
57
75
4-COOEt
3-COOH
4-CH3
23
23
11
18
18
94
35
0.18
1.1
23
23
2-iPr
4-iPr
18
18
180
60
0.33
1.5
23
23
4-c-C6H11
2-CF3
3-CF3
18
18
125
45
1.4
1.1
23
27
18
18
170
2
4-CF3
4-NO2
0.32
0.66
1.4
27
27
18
18
65
2
4-N(CH3)2
Ar = 2-pyridil
Ar = 3-pyridil
Ar = 4-pyridil
H
27
27
4.8
1.0
18
18
150
80
27
27
0.8
3.3
18
100
70
19,20
19,20
19,20
19,20
27
CH3
14
1.0
74
28,29
28,29
28,29
28,29
28,29
28,29
28,29
30
CH2CH3
OH
Ph
—
—
105
375
175
125
175
430
0.4
3
0.68
4.0
19,20
21–24
—
—
—
—
0.10
1.1
21–24
21–24
21–24
—
—
—
—
0.13
0.5
H
Propyl
3-CH2NH2
4-CH2NH2
3-NH2
0.04
1.5
25
25
25
25
25
25
30
30
Butyl
Pentyl
14
60
14
1
30
30
3-CH2CH2NH2
3-CH2NHMe
3-CH2NMe2
Cyclopropyl
Cyclopropylmethyl
Isobutyl
50
10
0.056
0.17
30
30
45
R1 = CH3, R2 = H, R3 = H
R1 = H, R2 = CH3, R3 = H
R1 = CH3, R2 = H, R3 = CH3
—
8
12
30
30
O
I–56
2.8
25
350
6
30
30
N
3-
—
2
30
I–57
I–58
3-CH2NHOH
3-CH2NHC(NH)Me
2.1
0.37
25
25
NH
regarding internal validation, in both cases, the predic-
tion is 100% (4/4). However, according to the statistical
indexes (Table 3), the models obtained with strategy 1
N
H
I–59
0.042
25
3-

6196
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
Et
NHNO2
NH
HN
O
HN NH2
S
H2N
S
S
NH
HN
NH
SR
NH
R
H
N
H2N
HN
HN
2
NH
R
N
H
CONH2
R
HN
NH2
NH
I-46
I-1 to I-10
I-11 to I-14
I-15 to I-40
I-41 to I-45
NHNO2
NHNO2
NH
NHNO2
HN
HN
NH
NH
NH
H
N
R
H
N
CONH2
OH
H2N
H2N
NH2
Me
N
H
CONH2
NH
O
O
O
L-NNA
NH2
I-48
I-47
I-49
I-50 to I-64
HN
NH
R
R1
R2
N
N
NH
N
N
O
n
H2N
H2N
S
S
R2
N
R1
R
H
H
H2N
COOH
I-73 (L-NIO)
I-65 to I-72
I-82
I-83 to I-92
I-77 to I-81
I-74 (L-VNIO)
I-75 (Methyl-L-NIO)
I-76 (Ethyl-L-NIO)
NO2
R
Me
NO2
N
COOH
NH2
N
N
N
N
N
N
n
N
O
NO2
O
O
I-93 to I-100
I-101
I-102
I-103
NH
Ph
N
Ph
N
N
R N
H
N
N
N
N
Ph
N
H
N
N
OH
I-104
I-105
I-108 to I-114
I-106
I-107
NH
NH
NH
N
R1
R2
N
N
H2N
N
NH2
NH2
N
N
N
H
N
H
R3
I-118
I-119
I-115 to I-117
Figure 1. NOS inhibitor families.
(models 1, 2, 3, and 4) show lower values of index frac-
tion correct (FC) the probability of detection (POD) and
higher values of false alarm rate (FAR) than the models
obtained with strategy 2 (models 5, 6, and 7) pointing
out that there is a high percentage of misclassified inac-
tive compounds (classified as active compounds) and
therefore, these models seem not to detect real inhibitors
(see Section 4).
one, showing the highest values of FC and POD and the
lowest value of FAR.
From these results, the best net is model 7, which uses
the static vector as definition of the structures (strategy
2). The model shows a good overall classification per-
centage of 91 in the training set. A total of 89% of
the compounds in the external prediction set were
correctly identified by the model. The results of the
training and test sets are gathered in Tables 4 and 5,
respectively.
The models obtained using the strategy 2 show better
values of FC, FAR, and POD, with the model 7, the best

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6197
Figure 2. General methodology to develop an inhibition NOS model.
Table 3. Test statistical indices
N
C
Models
FC (%)
FAR (%)
POD (%)
R2
N
X
R1
1
2
3
4
5
6
7
8
43.33
43.33
62.06
62.06
70.58
76.47
89.47
72.00
84.61
92.30
33.33
27.77
25.00
15.38
5.88
72.72
66.66
72.22
75.00
81.81
81.81
92.85
78.57
X = C, N, S
Figure 3. Atom template to static set descriptor.
The percentages of correct classifications for the three
categories in the training set and test set are shown in
Figures 4 and 5. Analysis of the classification by cate-
gory shows that the model correctly evaluates 100%
for active compounds (1), 33% for moderates and 91%
for inactive (ꢀ1) in training sets and 92% for active com-
pounds (1), 50% for moderates and 100% for inactives
(ꢀ1) in test sets.
22.22
Once the neural model was developed, we used it to pre-
dict the activity of new design compounds. Griffith41
and Furfine18 have reported that L-thiocitrulline (1)
and S-ethyl N-phenylisothiourea (2) are potent NOS
inhibitors with selectivity for the neuronal isoform.
Table 2. nNOS model statistical parameters
Models
Strategy
Architecture
ntraining
s
r
r2cv
Internal validationa
1
2
3
4
5
6
7
8
1
1
1
1
2
2
2
2
4:7:1
4:7:1
4:7:1
4:7:1
4:5:1
4:5:1
4:5:1
4:7:1
77
68
69
71
59
57
55
68
0.22
0.08
0.25
0.37
0.26
0.20
0.36
0.36
0.96
0.99
0.95
0.90
0.93
0.96
0.89
0.89
0.92
0.98
0.91
0.80
0.87
0.92
0.78
0.79
3/4
4/4
3/4
3/4
4/4
4/4
4/4
3/4
a Correct prediction of the randomly excluded data.

6198
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
Table 4. Training set of model 7
Table 5. Test set of model 7
Compound
Ki, (lM)
Groupa
Predicted group
Compound
Ki, (lM)
Groupa
Predicted group
I–3
I–4
0.037
0.62
0.63
14
1
1
1
1
I–1
I–2
0.16
0.029
0.17
0.14
0.21
1
1
1
1
1
I–5
I–6
1
1
I–17
I–19
I–25
I–39
I–43
I–46
I–47
I–48
I–52
I–54
I–56
I–70
I–71
I–80
I–112
I–113
I–115
1
1
ꢀ1
1
ꢀ1
1
1
1
I–7
I–8
0.8
14
1
1
ꢀ1
0
1
1
1
I–10
I–11
I–12
I–13
I–15
I–16
I–18
I–20
I–21
I–22
I–23
I–24
I–26
I–29
I–31
I–35
I–36
I–40
I–41
I–49
I–50
I–51
I–53
I–55
I–58
I–59
I–60
I–62
I–63
I–65
I–66
I–67
I–68
I–69
I–72
I–74
I–77
I–78
I–81
I–101
I–108
I–109
I–110
I–111
I–114
I–116
I–117
I–118
I–119
2
0
1
0.1
1
1
0.25
0.016
0.1
1
1
1
1
1
1
1.1
0.13
14
0
1
1
1
1
1
0.12
0.25
0.45
0.17
0.56
0.29
0.7
1
1
ꢀ1
1
ꢀ1
1
1
1
0.056
2.8
1
1
0
0
1
1
0.33
0.0063
0.38
50
1
1
1
1
1
1
1
1
1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
1
1
ꢀ1
ꢀ1
ꢀ1
0.19
0.34
0.18
0.33
0.32
0.66
0.8
1
1
10
8
1
1
1
1
a Active (1), moderate (0) or inactive (ꢀ1).
1
1
1
1
1
1
1
1
3.3
0.5
0
1
1
9%
-
50%
33%
17%
100%
1
1
0.04
1.5
1
1
0
1
Predicted
-
0
1
1
1
0.17
0.37
0.042
0.36
1.1
1
1
-1
91%
-1
-
1
1
1
1
1
1
0
1
0
1
0.57
0.04
0.19
0.011
0.011
0.3
1
1
Real
1
1
Figure 4. Analysis of the classification by category in training set of
model 7.
1
1
1
1
1
1
1
1
0.0087
0.1
12
1
1
1
1
1
-
-
50%
50%
-
92%
-
ꢀ1
ꢀ1
1
ꢀ1
ꢀ1
1
10
0.89
74
Predicted
0
ꢀ1
1
ꢀ1
1
0.4
3
14
0
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
0
-1
100%
-1
8%
1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
0
60
45
0
Real
12
350
6
Figure 5. Analysis of the classification by category in test set of model
7.
2
a Active (1), moderate (0) or inactive (ꢀ1).
S
R''
S
Taking into account these results we proposed the syn-
thesis of thiadiazole derivatives with aryl substituents
(3), which can be considered as conformationally re-
stricted analogues of S-ethyl N-phenylisothiourea (2)
(Fig. 6). This scaffold should not only be able to donat
hydrogen bonds (assuming that the ring nitrogen re-
mains protonated at physiological pH), but also make
non-polar contacts. These two interactions have been
N
COOH
NH2
HN
NH
S
H2N
N
H
O
N
HN
R'
L-Thiocitrulline (1)
2
3
Figure 6. Structure of S-ethyl N-phenylisothiourea.

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6199
O
R
R
S
O
O
S NH
N
a
O
S
N
O
S N
N
a
N
H
N
H
NH2
N
H
O
N
H
O
N
O
N
7
4
CN
4 R=H
4 R=H
14 R=H
15 R=CH2CN
16 R=Me
5 R=Me
Me
S
O
S N
N
Scheme 3. Reagents and condition: (a) ClCH2CN, K2CO3, DMF, 60–
80 ꢁC, 2.5–22 h (compound 14: 60%, 15: 75%, 16: 97%).
a
R
R
Me
N
H
O
N
H
N
N
H
H
5 R=Bz
8
hydrogen bonds in the NOS active site (Scheme 2). Ben-
zoylamino derivatives 4 and 5 did not react in these con-
ditions due to the coordination between the hypervalent
sulfur and the carbonyl group, which has been postu-
lated in similar heterocycles,48 and that would imply
the blockade of the atom of sulfur for the addition
reaction.
R=Bz
6
R=Bn
9 R=Bn
Scheme 1. Reagents and condition: (a) NBS, MeOH, reflux, 12 h
(compound 4: 76%, 5: 86%, 6: 80%).
shown as important requirements for the NOS inhibi-
tory activity of isothioureas.42 Departing from the
framework 2 (Fig. 6) we proposed several aminothia-
diazoles in order to predict their activity as nNOS inhib-
itors. All the considered compounds were predicted as
active except when the amino at position 5 is not substi-
tuted, pointing out that it is necessary for the activity
that the amino in this position will be substituted. In
addition, these initial structures could be functionalized
subsequently in order to obtain an extensive family of
compounds.
In order to introduce a cyano functionality in the het-
erocycles, the alkylation of compounds 4 and 5 with
chloroacetonitrile was performed following the de-
scribed procedures for related compounds.45 The best
results were achieved by using potassium carbonate as
a base in anhydrous dimethylformamide followed by
condensation with chloroacetonitrile. Starting from 5-
benzoylamino-3-oxo-2,3-dihydro-1,2,4-thiadiazole (4),
a mixture of mono- and dialkylated derivatives (14
and 15, respectively) was obtained. Obviously, when
the starting material was 5-benzoylamino-2-methyl-3-
oxo-2,3-dihydro-1,2,4-thiadiazole (5), only the mono-
alkylated product 16 was formed (Scheme 3).
2.2. Synthesis and NOS evaluation
The 5-amino-1,2,4-thiadiazole-3-ones 4, 5, and 6 de-
scribed as GSK-3b inhibitors43 were obtained by oxida-
tive cyclization of thiobiurets 7, 8, and 9 via N–S bond
formation with N-bromosuccinimide44 (Scheme 1). Previ-
ously, the thiobiurets 7, 8 and 9 were prepared by conden-
sation of isothiocyanates with urea45 or methylurea.46
Taking into account that amidine group appears to be a
common pharmacophore of NOS inhibitors,12 we took
advantage of the nitrile group reactivity present in the
thiadiazol framework to prepare different related ana-
logues as amidoxime 17, imido ester 18, or amidines
20 and 21 from 5-benzoylimino-4-cyanomethyl-2-
methyl-3-oxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (16).
In order to generate amidoxime 17, hydroxylamine
was added to the cyano group of 16 according to a pre-
viously described procedure.49 On the other hand, com-
pound 16 was transformed into the imido esther 18 by
the Pinner synthesis,50,51 where also traces of the hydro-
lysis by-product 19 were also identified. Finally, amidine
The diversification of 5-benzylamino-2-methyl-3-oxo-2,3-
dihydro-1,2,4-thiadiazole (6) by addition of isocyanates
and isothiocyanates in the presence of triethylamine
yielded the imino derivatives, 10–13, the structures of
which were elucidated on the basis of NMR experi-
ments. These thiadiazoles were obtained after an addi-
tion–rearrangement reaction,47 and the presence of an
amide group should allow the molecule to establish
Me
R
Me
R
Me
HN
N
NH
N
S
N
S
N
S
N
a
O
X
N
H
O
X
O
N
N
N
6
X=O R=Et
X=S R=Et
10
11
12
X=
NO2
S
S
R=
R=
MeO
X=
13
Scheme 2. Reagents and condition: (a) RNCX, Et3N, DMF, 80 ꢁC, 5–48 h (compound 10: 79%, 11: 46%, 12: 26%, 13: 20%).

6200
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
Me
O
S N
N
N
O
H
N
20
HN
c
Me
Me
Me
O
O
S N
N
S N
N
O
S
N
b
N
O
N
O
+
N
O
N
OH
OMe
CN
18
16
19
NH
NH
d
a
Me
Me
O
S
N
O
S N
N
N
O
N
O
N
H
N
NOH
NH2
NH2
HN
17
21
Scheme 4. Reagents and conditions: (a) Et3N, NH2OHÆHCl, DMSO, 75 ꢁC, 8 h (85%); (b) NaOMe/MeOH, rt, 4 h (18: 92%); (c) NH2Ph, 40 ꢁC, 32 h
(6%); (d) NH2NH2ÆHCl, 45 ꢁC, 72 h (48%).
derivatives 20 and 21 were obtained by reaction with
aniline and hydrazine, respectively51 (Scheme 4), but
for moderates, and 75% for inactives (ꢀ1) in test
sets.
unfortunately
characterization.
they
decomposed
after
their
The results of the training and test sets are gathered in
Tables 6 and 7 respectively. This model correctly predicts
our compounds according to the experimental data.
The inhibition of nNOS isoform was determined by
monitoring the conversion of L-[3H]arginine to L-[3H]cit-
rulline as previously described.52
3. Conclusions
All the compounds reported here, with the exception of
20 and 21 which were unstable, were tested but none of
them showed any significant activity (IC50 > 100 lM).
In summary, we have performed a QSAR study based
on different inhibitor families that are able to predict
the activity of new compounds. It is interesting to
emphasize that CODES is an efficient and easy way to
encode structures and it does not need 3D information,
thus avoiding the risky choice of the appropriate physi-
cochemical descriptors and problems associated with the
conformation.
2.3. Refined neural network model
The low-capacity prediction of the model 7 regarding
the new derivatives would be due to a lack of data of
new compounds during the learning process. Trying to
correctly address the model, we decided to refine it. Fol-
lowing the same strategy as for model 7 (strategy 2), we
developed model 8 with the 119 inhibitors previously
used, plus the seven new synthesized derivatives. This
new dataset of compounds was randomly divided in to
two sets. According to this distribution, compounds 5,
6, 14, 17 and 18 were included in the training set and
compounds 10 and 19 in the test set during model vali-
dation. The new model shows acceptable statistical
parameters (Tables 2 and 3, entry 8).
On the other hand, in relation to the strategies used in
the dimension-reduction process, we can conclude that
the best results were obtained when the static vector of
the four atoms more conserved in all the structures
was considered (strategy 2).
Regarding the new-design family, we can conclude that
the conformational restriction of the isothiourea group
causes a lack of activity, although the synthesized mole-
cules provide nitrogen-containing functional groups as
hydrogen-bond donators and alkyl groups for interac-
tion with a hydrophobic site. Probably, the rigidity of
the molecules prevents their assuming the appropriate
discriminatory-binding orientations.
Analysis of the classification by category shows that the
model 8 correctly evaluates 92.5% for active compounds
(1), 50% for moderates, and 93% for inactives (ꢀ1) in
training sets and 82% for active compounds (1), 40%

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6201
Table 6. Training set of model 8
Table 6 (continued)
Compound
Ki (lM)
Groupa
Predicted group
Compound
Ki (lM)
Groupa
Predicted group
I–1
I–3
0.16
0.037
0.62
0.63
14
1
1
1
1
14
17
18
—
—
—
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
I–4
I–5
1
1
1
1
a Active (1), moderate (0) or inactive (ꢀ1).
I–6
I–7
ꢀ1
1
ꢀ1
1
0.8
I–11
I–12
I–13
I–14
I–15
I–16
I–18
I–20
I–21
I–22
I–23
I–24
I–25
I–26
I–29
I–30
I–33
I–34
I–35
I–37
I–40
I–44
I–45
I–49
I–50
I–51
I–53
I–55
I–56
I–57
I–58
I–59
I–60
I–62
I–63
I–64
I–65
I–66
I–67
I–68
I–69
I–70
I–72
I–73
I–74
I–77
I–79
I–81
I–82
I–101
I–110
I–111
I–114
I–116
I–117
I–118
I–119
0.25
0.016
0.1
1
1
1
1
Table 7. Test set of model 8
1
1
Compound
Ki (lM)
Groupa
Predicted group
4.8
0
1
0.12
0.25
0.45
0.17
0.56
0.29
0.7
1
1
I–2
I–8
0.029
14
1
ꢀ1
0
1
1
1
1
1
1
I–10
I–17
I–19
I–38
I–39
I–41
I–43
I–46
I–47
I–48
I–52
I–54
I–71
I–78
I–80
I–108
I–109
I–112
I–113
I–115
2
0
1
1
0.17
0.14
4.8
1
1
1
1
1
1
1
1
1
0
1
1
1
1
1
0.19
0.21
0.34
0.18
1.1
1
1
3.3
1
0
-0
1
1
1
1
1
1
0.1
1.1
0.13
14
1
1
1
1
0
1
0
0
1
1
1.4
1.1
0
1
ꢀ1
1
1
0
1
0.056
0.0063
10
0
0.32
1.4
1
1
1
1
0
0
ꢀ1
1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
0.8
1
1
0.38
0.4
3
0.68
4
1
1
1
0
1
0
0.5
1
1
50
10
ꢀ1
ꢀ1
ꢀ1
0.04
1.5
1
0
0
0
8
1
1
1
0.17
2.8
1
1
10
19
—
—
ꢀ1
ꢀ1
0
1
ꢀ1
ꢀ1
2.1
0
1
a Active (1), moderate (0) or inactive (ꢀ1).
0.37
0.042
0.36
1.1
1
1
1
1
1
1
0
1
4. Experimental
0.57
1.5
1
1
0
1
4.1. Development of neural models
0.04
0.19
0.011
0.011
0.3
1
0
1
1
The following strategy was pursued to develop a neuro-
nal network model to predict nNOS inhibition:
1
1
1
1
1
1
4.1.1. Database. A series of 119 inhibitors were collected
from the literature and their biological activity is ex-
pressed as Ki (lM) (see Fig. 1 and Table 1). This data-
base is formed by different families, which show Ki
values in nNOS. The original database was divided ran-
domly into training sets and test sets of different sizes.
0.33
0.0087
1.7
1
1
1
1
0
0
0.1
12
1
1
ꢀ1
0
0
1.1
0
0.89
0.41
74
1
1
1
0
Regarding the output (values of the activity) we used for
the training of the neural net: quantitative values to
measure the inhibition of logKi and qualitative values
that refer to assess whether the compounds are active
(1), moderate (0) or inactive (ꢀ1). Active compounds
(1) represent the compounds the activity of which com-
prises the values until 1 lM, moderate compounds (0)
represent the compounds with activity between 1.1 and
5 lM and inactive compounds (ꢀ1) represent the com-
pounds with activity higher than 5 lM.
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
0
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
ꢀ1
0
14
60
45
12
350
6
2
5
6
—
—
ꢀ1
ꢀ1
ꢀ1
ꢀ1

6202
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
4.1.2. Input data. The first step is to draw the selected
compounds using CHEMDRAW software (v8.0) and
encode them using SMILES system (see Table S1 in sup-
porting information) that denotes a molecular structure
as a graph.53 This is included in CHEMDRAW software
(v8.0). Subsequently, the molecular descriptors are ob-
tained using CODESꢀ software (v1.0, revision 3).
CODES encodes each molecule into a dynamic matrix.
CODES consists of two levels, topological and neural,
and its philosophy relies on a Gestalt isomorphism54 be-
tween both the levels. While the topological space is the
chemical structure by itself, the neural one consists of an
interactive and competitive network. Each point or
atom of the topological space corresponds with each
unit or neuron of the neural space, and each type of
atom takes a different initial value based on the atom
nature, the number of atom bonds, the connectivity with
the rest of the molecule, and chirality (if applicable).
Attending to connectivity, CODES considers both
bonding and not bonding interactions between atoms.
If atoms are not bonded in the topological space, it
means an inhibitory connection in the neural level (value
ꢀ1), otherwise the neural space considers an excitatory
connection and the value depends on bond type (values:
1 for X–X; 2 for X@X; 3 for X„X; 1 + 1/2 for aromatic
bonds). The stereochemistry is also taken into account
during the codification process and R or S configuration
is expressed by a corrective non-linear function.
Non-linear Dimension Reduction). The network consists
of five layers of neurons: an input layer, a first hidden
layer (coding), a central layer of two or three neurons,
a third hidden layer (decoding) and the output layer.
The input and output layers contain the same informa-
tion and therefore the same number of neurons, and
the coding and decoding layers are of the same size and
the central layer contains a number of neurons equal to
the dimensionality of the ReNDer plot required (usually
2 or 3).55 In the developed model, the process of dimen-
sion-reduction is carried out in order to compress the dy-
namic matrix data to a set of four numeric codes for each
molecule (hidden neurons: A, B, C, and D; see support-
ing information, Table S1). RD process is carried out
using TSARꢀ program55 which applies Monte Carlo
algorithm. Convergence parameters are 0.005 RMS
(Root Mean Square) of convergence, 1000 cycles past
best, 3000 iterations/cycle, and data excluding of 1%.
The process is finished when Best RMS and Test RMS
are constant and their values are not higher than 0.02.
If the values were higher, the reductions settings would
be 0.0005 RMS of convergence, 4000 cycles past best,
20000 iterations/cycle and 1% of excluded data for test-
ing. The neural network is considered trained when the
lines diagrams of the convergence plot are unchanging.
The second one is choice of four parameters per mole-
cule from the last step of the codification process
generated by CODES, which correspond to descriptors
of the four atoms more conserved in all the structures
(Fig. 3). In this way, a static set of four descriptors is
obtained (see Supporting information, Table S2).
The neural network employs a sigmoideal function in
the codification process and the network is characterized
by a non-supervised learning. In the learning process,
CODES records all the activities reached in every cycle
(or iteration) of the network. This process finishes when
the equilibrium state is reached, so, all activity values of
each atom of the structure during each cycle or iteration
are gathered in a matrix from the initial to the final steps
forming the dynamic matrix, which contains the whole
codification process.
4.1.3. Development of neural network model. This proce-
dure is carried out by a standard feed-forward network
with back-propagation using TSAR software (v3.0)55
with an architecture 4 ꢀ n ꢀ 1, where four is the param-
eter above described, n is the hidden neurons, and one is
the output value (Ki).
It is interesting to emphasize that CODES does not need
three-dimensional information because the topological
space and its conversion to a neural space only need de-
tails about points (atoms) and the relationships between
them (bonds); that is, the chemical structure by itself.
Thus, CODES avoids the risky choice of appropriate
physicochemical descriptors and problems associated
with the conformation.
In each training set established, we performed a system-
atic study of the neural network learning process. In the
first approach, we evaluated the suitable number of hid-
den neurons attending to q factor, thus, several initial
trainings were carried out with the appropriate architec-
ture. Models were evaluated using statistical parameters
and internal validation method (described below) and
the best ones were retrained in the second learning stage.
We used two different approaches in this second learning
phase (see Fig. 2). Via 1 consists in retraining the se-
lected model several times while the second approach,
via 2, consists also in several retrainings but in this case,
each retraining is used as the previous model in the fol-
lowing one.
Based on the topological matrix, we have developed two
different strategies for obtaining the molecular descriptors:
One of these is using the whole previous matrix of each
compound. The next step is reduction of dimension of
matrices of each chemical in order to have the same
number of descriptors or variables.
4.1.4. Validation of the models. All models were evalu-
ated using some statistical indices: as fraction correct
(FC), false alarm rate (FAR), and probability of detec-
tion (POD). Before calculating such indices, TA (num-
ber of cases that were truly classified as active), TM
(number of cases that were truly classified as moderate),
TI (number of cases that were truly classified as inac-
tive), FA (number of cases that were wrongly recognized
Reduction of dimension (RD) philosophy resides in
reducing the complexity of any system without loss of
any intrinsic characteristics or information about the
chemical nature. High-dimensional data can be con-
verted to low-dimensional codes by training a supervized
multilayer neural network namely ReNDeR (Reversible

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6203
as active), and FI (number of cases that were wrongly
recognized as inactive) should have been counted by
each final model.
643 (0.20 g, 0.90 mmol) in DMF (9.0 mL), triethylamine
(0.2 mL, 1.53 mmol) and ethylisocyanate (0.1 mL,
1.42 mmol) were added and the mixture was heated at
80 ꢁC. After 5 h, the reaction mixture was cooled to
room temperature and the solvent was removed under
reduced pressure. The residue was purified by silica gel
column chromatography using dichloromethane/metha-
nol (90:1) as eluent to give a colorless oil. Yield: 0.20 g
Fraction correct (FC) is the fraction of compounds that
were classified correctly.
TA þ TM þ TI
1
FC ¼
ꢁ 100
Nototal
(79%). Purity >99% (by HPLC). H NMR (200 MHz,
CDCl3): d 7.43–7.38 (m, 2H, Ar-H), 7.32–7.23 (m, 3H,
Ar-H), 5.54–5.52 (br m, 1H, NH), 5.00 (s, 2H, CH2Ph),
3.58 (c, 2H, J = 7.2 Hz, CH2), 2.97 (d, 0.3H, J = 5.2 Hz,
anti-MeNH), 2.85 (d, 2.7H, J = 5.2 Hz, syn-MeNH),
1.23 (t, 3H, J = 7.2 Hz, Me). 13C NMR (50 MHz,
CDCl3): d 165.8 (C-5), 165.2 (CONH), 152.5 (C-3),
135.6 (Ar-C), 128.4 (2C, Ar-C), 128.4 (2C, Ar-C),
127.8 (Ar-C), 47.4 (CH2Ph), 38.9 (CH2), 27.2 (MeNH),
14.0 (Me). MS (ESI): m/z 293 (M+H)+. Anal. Calcd
for C13H16N4O2S: C, 53.41; H, 5.52; N, 19.16; S,
10.97. Found: C, 53.36; H, 5.73; N, 19.24; S, 10.68.
False alarm rate (FAR) represents the fraction of inac-
tive compounds that were wrongly classified.
FA
FAR ¼
ꢁ 100
TA þ TM þ TI
Probability of detection (POD) is another index by
which we evaluated our models. It represents the frac-
tion of active cases that were truly classified.56
TA þ TM
POD ¼
ꢁ 100
4.2.2. 4-Benzyl-2-ethyl-5-methylcarbamoylimino-3-thioxo-
2,3,4,5-tetrahydro-1,2,4-thiadiazole (11). To a solution of
643 (0.20 g, 0.90 mmol) in DMF (9.0 mL), triethylamine
(0.2 mL, 1.50 mmol) and ethylisothiocyanate (0.1 mL,
1.42 mmol) were added and the mixture was heated at
80 ꢁC. After 15 h, the reaction mixture was cooled to
room temperature and the solvent was removed under re-
duced pressure. The residue was purified by silica gel col-
umn chromatography using hexane/ethyl acetate (3:1) as
eluent to give a colorless oil. Yield: 0.13 g (46%). Purity
TA þ TM þ FI
Moreover, we carried out an internal validation by regres-
sion data like s, r, cross-validation expressed by r2cv data
that is carried out by leave-group-out method and inter-
nal prediction of four randomly excluded structures.
On the other hand, we carried out an external validation
by test-set prediction.
1
4.2. Chemical procedures
>99% (by HPLC). H NMR (400 MHz, DMSO-d6): d
8.15 (c, 1H, J = 4.8 Hz, NH), 7.38–7.24 (m, 5H, Ph),
5.46 (s, 2H, CH2Ph), 3.98 (c, 2H, J = 7.2 Hz, CH2),
2.79 (d, 0.3H, J = 4.8 Hz, anti-MeNH), 2.67 (d, 2.7H,
J = 4.8 Hz, syn-MeNH), 1.24 (t, 3H, J = 7.2 Hz, Me).
13C NMR (100 MHz, DMSO-d6): d 172.1 (C-3), 165.8
(C-5), 164.3 (CONH), 135.4 (Ar-C), 128.3 (2C, Ar-C),
127.4 (2C, Ar-C), 127.3 (Ar-C), 50.7 (CH2Ph), 43.0
(CH2), 27.9 (anti-MeNH), 26.9 (syn-MeNH), 12.9 (Me).
MS (ESI): m/z 309 (M+H)+. Anal. Calcd for
C13H16N4OS2: C, 50.63; H, 5.23; N, 18.17; S, 20.79.
Found: C, 50.37; H, 5.15; N, 18.26; S, 20.86. IR: t
Substrates were either purchased from commercial
sources or used without further purification. Melting
points were determined with a Reichert–Jung Thermovar
apparatus and are uncorrected. Flash column chroma-
tography was carried out at medium pressure using silica
gel (E. Merck, Grade 60, particle size 0.040–0.063 mm,
230–240 mesh ASTM) with the indicated solvents as elu-
ents. Compounds were detected with UV light (254 nm).
1HNMR spectra were obtained on Varian INOVA-400,
Bruker AVANCE-300, and Varian-Gemini-200 spec-
trometers working at 400, 300 and 200 MHz, respec-
tively. Typical spectral parameters: spectral width
10 ppm, pulse width 9 ls (57ꢁ), and data size 32 K. 13C
NMR experiments were carried out on the Varian INO-
VA-400, Bruker AVANCE-300, and Varian-Gemini-200
spectrometers operating at 100, 75, and 50 MHz, respec-
tively. The acquisition parameters: spectral width
16 kHz, acquisition time 0.99 s, pulse width 9 ls (57ꢁ),
and data size 32 K. Chemical shifts are reported in values
(ppm) relative to internal Me4Si and J values are re-
ported in Hertz. IR (infrared-spectroscopy) was mea-
sured on a Perkin-Elmer Spectrum One Spectrometer.
EI (electronic ionization mass spectroscopy), MSD
5973 Hewlett Packard and ESI (electrospray ionization
mass spectroscopy), and LC/MSD-Serie 100 Hewlett
Packard. Elemental analyses were performed by the ana-
lytical department at CENQUIOR (CSIC).
3338 (m, NH), 1621 (s, C@O), 1213 (s, C@S) cmꢀ1
.
4.2.3. 4-Benzyl-2-(2-methoxy-4-nitrophenyl)-5-methylcarba-
moylimino-3-thioxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (12).
To a solution of 643 (50.0 mg, 0.22 mmol) in DMF
(2.2 mL), triethylamine (53.5 lL, 0.38 mmol) and 4-nitro-2-
methoxyphenylisothiocyanate (75.0 mg, 0.35 mmol) were
added and the mixture was heated at 80 ꢁC. After 48 h, the
reaction mixture was cooled to room temperature and the
solvent was removed under reduced pressure. The residue
was purified by silica gel column chromatography using hex-
ane/ethyl acetate (3:1) as eluent to give a yellow solid. Yield:
25.2 mg (26%). Mp: 180-182 ꢁC. Purity: 97% (by HPLC). 1H
NMR (300 MHz, CDCl3): d 7.86 (dd, 1H, J = 8.5, 2.1 Hz,
Ar-H), 7.79 (d, 1H, J = 2.1 Hz, Ar-H), 7.52 (dd, 2H,
J = 7.3, 2.1 Hz, Ar-H), 7.34–7.32 (br m, 3H, Ar-H), 6.93
(d, 1H, J = 8.5 Hz, Ar-H), 5.56–5.54 (br m, 1H, NH), 5.46
(s, 2H, CH2Ph), 3.89 (s, 3H, MeO), 3.00 (d, 0.3H,
J = 4.9 Hz, anti-MeNH), 2.90 (d, 2.7H, J = 4.9 Hz, syn-
MeNH). 13C NMR (75 MHz, CDCl3): d 167.9 (C-3), 163.8
4.2.1. 4-Benzyl-2-ethyl-5-methylcarbamoylimino-3-oxo-
2,3,4,5-tetrahydro-1,2,4-thiadiazole (10). To a solution of

6204
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
(CONH), 156.2 (C-5), 150.8 (NO2-C), 145.1 (MeO-C), 144.4
(Ar-C), 135.3 (Ar-C), 128.5 (2C, Ar-C), 128.3 (2C, Ar-C),
127.9 (Ar-C), 121.3 (Ar-C), 117.2 (Ar-C), 107.2 (Ar-C),
56.1 (MeO), 51.7 (CH2Ph), 29.6 (anti-MeNH), 27.3 (syn-
MeNH). MS (ESI): m/z 432 (M+H)+, 885 (2M+Na)+.
NMR (300 MHz, DMSO-d6):
d
8.28 (d, 2H,
J = 7.1 Hz, Ar-H), 7.72 (t, 1H, J = 7.3 Hz, Ar-H), 7.61
(dd, 2H, J = 7.3, 7.1 Hz, Ar-H), 5.19 (s, 2H, CH2-4),
4.89 (s, 2H, CH2-2). 13C NMR (75 MHz, DMSO-d6):
d 178.1 (PhCO), 169.4 (C-5), 150.8 (C-3), 134.0 (Ar-C),
132.2 (Ar-C), 129.5 (2C, Ar-C), 129.0 (2C, Ar-C),
115.6 (CN-2), 114.5 (CN-4), 32.2 (CH2-4), 31.5 (CH2-
2). MS (EI): m/z 299 (M+, 8). Anal. Calcd for
C13H9N5O2S: C, 52.17; H, 3.03; N, 23.40; S, 10.71.
Found: C, 52.23; H, 3.17; N, 23.76; S, 10.50.
4.2.4. 4-Benzyl-2-tert -butyl-5-methylcarbamoylimino-3-
thioxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (13). To a
solution of 643 (50.0 mg, 0.22 mmol) in DMF (2.2 mL),
triethylamine (53.5 lL, 0.38 mmol) and tert-butylisothi-
ocyanate (45.2 lL, 0.35 mmol) were added and the mix-
ture was heated at 80 ꢁC. After 48 h, the reaction
mixture was cooled to room temperature and the solvent
was removed under reduced pressure. The residue was
purified by silica gel column chromatography using hex-
ane/ethyl acetate (3:1) as eluent to give a colorless oil.
Yield: 15.0 mg (20%). Purity >99% (by HPLC). 1H
NMR (200 MHz, CDCl3): d 7.45 (dd, 2H, J=7.4,
2.2 Hz, Ar-H), 7.32–7.25 (m, 3H, Ar-H), 5.57 (s, 2H,
CH2Ph), 5.51–5.49 (br m, 1H, NH), 2.96 (d, 0.3H,
J = 5.0 Hz, anti-MeNH), 2.87 (d, 2.7H, J = 5.0 Hz,
syn-MeNH), 1.81 (s, 9H, Me3). 13C NMR (75 MHz,
CDCl3): d 172.2 (C-3), 166.0 (C-5), 165.0 (CONH),
135.6 (Ar-C), 128.2 (2C, Ar-C), 128.2 (2C, Ar-C),
127.5 (Ar-C), 50.4 (CH2Ph), 29.5 (C-tBu), 28.0 (3C,
Me3), 27.4 (MeNH). MS (EI): m/z 336 (M+, 30).
4.2.7. 5-Benzoylimino-4-cyanomethyl-2-methyl-3-oxo-
2,3,4,5-tetrahydro-1,2,4-thiadiazole (16). To a solution
of 543 (0.15 g, 0.63 mmol) in anhydrous DMF
(6.3 mL), K2CO3 (54.6 mg, 0.39 mmol) was added
and the mixture was stirred at room temperature.
After 1 h, chloroacetonitrile (24.9 lL, 0.39 mmol)
was added and the reaction mixture was heated at
80 ꢁC. After 2.5 h, the reaction mixture was cooled
to room temperature and cold distilled water
(10.0 mL) was added. The resulting precipitate was
isolated by filtration to give a white solid. Yield:
1
0.17 g (97%). Mp: 219–221 ꢁC. H NMR (300 MHz,
DMSO-d6): d 8.27 (d, 2H, J = 7.1 Hz, Ar-H), 7.69 (t,
1H, J = 7.3 Hz, Ar-H), 7.59 (dd, 2H, J = 7.3, 7.1 Hz,
Ar-H), 5.16 (s, 2H, CH2), 3.17 (s, 3H, Me). 13C
NMR (75 MHz, DMSO-d6): d 177.4 (PhCO), 168.7
(C-5), 150.8 (C-3), 133.7 (Ar-C), 132.8 (Ar-C), 129.4
(2C, Ar-C), 128.9 (2C, Ar-C), 114.8 (CN), 32.1
(CH2), 30.2 (Me). MS (EI): m/z 274 (M+, 14). Anal.
Calcd for C12H10N4O2S: C, 52.54; H, 3.67; N, 20.43;
S, 11.69. Found: C, 52.31; H, 3.74; N, 20.64; S. 11.88.
4.2.5. 5-Benzoylimino-4-cyanomethyl-3-oxo-2,3,4,5-tetra-
hydro-1,2,4-thiadiazole (14). To a solution of 443 (0.15 g,
0.67 mmol) in anhydrous DMF (6.7 mL), K2CO3
(58.1 mg, 0.42 mmol) was added and the mixture was
stirred at room temperature. After 1 h, chloroacetonitri-
le (26.5 lL, 0.42 mmol) was added and the reaction mix-
ture was heated at 60 ꢁC. After 63 h, the reaction
mixture was cooled to room temperature and cold dis-
tilled water (10.0 mL) was added. The resulting precipi-
tate was isolated by filtration and the solid was purified
by silica gel column chromatography using hexane/ethyl
acetate (3:2) as eluent to give the monoalkylated product
as a white solid. Yield: 0.10 g (60%). Mp: 129–131 ꢁC.
1H NMR (300 MHz, DMSO-d6): d 8.22 (d, 2H,
J = 7.1 Hz, Ar-H), 7.62–7.49 (m, 3H, Ar-H), 4.99 (s,
2H, CH2), 3.50 (br s, 1H, NH). 13C NMR (75 MHz,
DMSO-d6): d 175.3 (PhCO), 173.5 (C-5), 154.3 (C-3),
135.0 (Ar-C), 132.4 (Ar-C), 129.1 (2C, Ar-C), 128.5
(2C, Ar-C), 115.6 (CN), 31.3 (CH2). MS (EI): m/z 260
(M+, 9). Anal. Calcd for C11H8N4O2S: C, 50.76; H,
3.10; N, 21.53; S, 12.32. Found: C, 50.39; H, 3.17; N,
21.28; S, 12.47.
4.2.8. 4-Aminocarbohydroxymoylmethyl-5-benzoylimino-
2-methyl-3-oxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (17).
To a suspension of hydroxylamine hydrochloride
(0.12 g, 1.82 mmol) in DMSO (2.0 mL), triethylamine
(0.2 mL, 1.82 mmol) was added and the resulting salts
were filtered off and washed with THF. The filtrate
was concentrated in vacuo to remove THF and 16
(0.10 g, 0.36 mmol) was added to the DMSO solution
of hydroxylamine. The reaction mixture was stirred at
75 ꢁC for 8 h and distilled water (4.0 mL) was added.
The resulting precipitate was isolated by filtration to
give a white solid. Yield: 95.7 mg (85%). Mp: 244–
246 ꢁC. 1H NMR (300 MHz, DMSO-d6): d 9.23 (s,
1 H, OH), 8.18 (dd, 2H, J = 7.1, 1.4 Hz, Ar-H), 7.65
(t, 1H, J = 7.4 Hz, Ar-H), 7.55 (dd, 2H, J = 7.4,
7.1 Hz, Ar-H), 5.63 (br s, 2H, NH2), 4.65 (s, 2H,
CH2), 3.15 (s, 3H, Me). 13C NMR (75 MHz, DMSO-
d6): d 177.3 (PhCO), 169.6 (C-5), 151.9 (C-3), 146.7
(C@NOH), 133.3 (Ar-C), 133.2 (Ar-C), 129.2 (2C, Ar-
C), 128.8 (2C, Ar-C), 43.6 (CH2), 29.8 (Me). MS
(ESI): m/z 308 (M+H)+, 637 (2M+Na)+. Anal. Calcd
for C12H13N5O3S: C, 46.90; H, 4.26; N, 22.79; S,
10.43. Found: C, 47.03; H, 3.99; N, 22.58; S, 10.78.
4.2.6.5-Benzoylimino-2,4-bis(cyanomethyl)-3-oxo-2,3,4,5-
tetrahydro-1,2,4-thiadiazole (15). To a solution of 443
(0.15 g, 0.67 mmol) in anhydrous DMF (6.7 mL),
K2CO3 (58.1 mg, 0.42 mmol) was added and the mixture
was stirred at room temperature. After 1 h, chloroaceto-
nitrile (26.5 lL, 0.42 mmol) was added and the reaction
mixture was heated at 80 ꢁC. After 22 h, the reaction
mixture was cooled to room temperature and cold dis-
tilled water (10.0 mL) was added. The resulting precipi-
tate was isolated by filtration and the solid was purified
by silica gel column chromatography using hexane/ethyl
acetate (3:2) as eluent to give the dialkylated product as
4.2.9. 5-Benzoylimino-4-methoxycarbonimidoylmethyl-2-
methyl-3-oxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (18)
and 5-benzoylimino-4-hydroxycarbonimidoylmethyl-2-
methyl-3-oxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (19).
To a solution of 16 (75.0 mg, 0.27 mmol) in MeOH
(2.7 mL), MeONa was added (1.4 mg, 0.02 mmol) and
1
a white solid. Yield: 0.15 g (75%). Mp: 150–152 ꢁC. H

T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
6205
the mixture was stirred at room temperature. After 4 h,
the product 18 was obtained as a precipitate, being iso-
lated by filtration to give a white solid. Sometimes,
traces of the hydrolysis by-product, 19, were identified.
Then, the purification employing silica gel column chro-
matography with ethyl acetate/hexane (2:1) as eluent,
produced the hydrolysis of 18 into 19 and allowed the
isolation of compound 19 as a white solid. First fraction,
yield of compound 18, 77.0 mg (92%). Mp: 182–184 ꢁC.
(br s, 1H, C@NH), 5.75 (s, 2H, CH2), 5.40 (br s, 1H,
NH2), 4.59 (br s, 1H, NH2), 3.15 (s, 3H, Me). 13C
NMR (75 MHz, CDCl3): d 177.4 (PhCO), 169.5 (C-5),
150.2 (C-3), 149.7 (C@NH), 133.4 (Ar-C), 133.0 (Ar-
C), 129.2 (2C, Ar-C), 128.7 (2C, Ar-C), 55.1 (CH2),
30.1 (Me). MS (EI): m/z 306 (M+, 1).
Acknowledgments
1
Purity >99% (by HPLC). H NMR (300 MHz, CDCl3):
d 8.22 (d, 2H, J = 7.1 Hz, Ar-H), 7.56 (t, 1H, J = 7.3 Hz,
Ar-H), 7.46 (dd, 2H, J = 7.3, 7.1 Hz, Ar-H), 7.32 (br s,
1H, NH), 4.78 (s, 2H, CH2), 3.80 (s, 3H, OMe), 3.22
(s, 3H, Me). 13C NMR (75 MHz, CDCl3): d 178.6
(PhCO), 169.2 (C-5), 165.8 (C@NH), 151.8 (C-3),
133.3 (Ar-C), 133.0 (Ar-C), 129.7 (2C, Ar-C), 128.4
(2C, Ar-C), 53.8 (OMe), 46.1 (CH2), 30.0 (Me). MS
(ESI): m/z 307 (M+H)+, 329 (M+Na)+, 635
(2M+Na)+. Anal. Calcd for C13H14N4O3S: C, 50.97;
H, 4.61; N, 18.29; S, 10.47. Found: C, 50.69; H, 4.48;
N, 18.07; S, 10.75. Second fraction, yield compound
19, Mp: 277–279 ꢁC. Purity >99% (by HPLC). 1H
NMR (300 MHz, DMSO-d6): d 8.17 (dd, 2H, J = 7.1,
1.3 Hz, Ar-H), 7.81 (br s, 1H, OH), 7.65 (t, 1H,
J = 7.3 Hz, Ar-H), 7.54 (dd, 2H, J = 7.3, 7.1 Hz, Ar-
H), 7.35 (br s, 1H, NH), 4.60 (s, 2H, CH2), 3.17 (s,
3H, Me). 13C NMR (75 MHz, DMSO-d6): d 177.2
(PhCO), 169.5 (C-5), 167.1 (C@NH), 152.0 (C-3),
133.4 (Ar-C), 133.1 (Ar-C), 129.2 (2C, Ar-C), 128.8
(2C, Ar-C), 46.4 (CH2), 29.9 (Me). MS (ESI): m/z 293
(M+H)+, 315 (M + Na)+, 607 (2M+Na)+.
We gratefully acknowledge the financial support, for
this project, of CICYT (SAF2006/01249). Thanks are
´
also due to Dr. Juan Antonio Paez for his helpful discus-
sions on the manuscript. Tania Castan˜o and Arantxa
Encinas acknowledge a predoctoral fellowship from
the Spanish Ministry of Education and Science.
Supplementary data
Supplementary data associated with this article can be
found, in the online version, at doi:10.1016/j.bmc.2008.
04.036.
References and notes
1. Kapoor, R. Curr. Opin. Neurol. 2006, 19, 255–259.
2. Poewe, W. Neurology 2006, 66, S2–S9.
3. Thatcher, G. R.; Bennett, B. M.; Reynolds, J. N. Curr.
Alzheimer Res. 2006, 3, 237–245.
4. Casetta, I.; Govoni, V.; Granieri, E. Curr. Pharm. Des.
2005, 11, 2033–2052.
4.2.10. 5-Benzoylimino-2-methyl-3-oxo-4-(N1-phenylamidi-
nomethyl)-2,3,4,5-tetrahydro-1,2,4-thiadiazole hydrochlo-
ride (20). To a solution of 18 (50.0 mg, 0.16 mmol) in
MeOH (1.6 mL), distilled aniline (15.0 lL, 0.17 mmol)
was added and the mixture was stirred at 40 ꢁC. After
32 h, the reaction mixture was cooled to room tempera-
ture, saturated with dry hydrogen chloride gas and stirred
for a few minutes to remove the excess of hydrogen chlo-
ride gas. The solvent was evaporated under reduced pres-
sure and dichloromethane was added to the residue. The
resulting precipitate was isolated by filtration to give a
5. Zhu, X.; Raina, A. K.; Perry, G.; Smith, M. A. Lancet
Neurol. 2004, 3, 219–226.
6. Nunomura, A.; Perry, G.; Aliev, G.; Hirai, K.; Takeda, A.;
Balraj, E. K.; Jones, P. K.; Ghanbari, H.; Wataya, T.;
Shimohama, S.; Chiba, S.; Atwood, C. S.; Petersen, R. B.;
Smith, M. A. J. Neuropathol. Exp. Neurol. 2001, 60, 759–767.
7. Nunomura, A.; Perry, G.; Pappolla, M. A.; Friedland, R.
P.; Hirai, K.; Chiba, S.; Smith, M. A. J. Neuropathol. Exp.
Neurol. 2000, 59, 1011–1017.
8. Moreira, P. I.; Honda, K.; Liu, Q.; Santos, M. S.; Oliveira,
C. R.; Aliev, G.; Nunomura, A.; Zhu, X.; Smith, M. A.;
Perry, G. Curr. Alzheimer Res. 2005, 2, 403–408.
9. Heneka, M. T.; Feinstein, D. L. J. Neuroimmunol. 2001,
114, 8–18.
1
white solid. Yield: 4.0 mg (6%). H NMR (400 MHz,
DMSO-d6): d 11.73 (br s, 1H, PhNH), 9.77 (br s, 1H,
C@NH2Cl), 8.97 (br s, 1H, C@NH2Cl), 8.20 (d, 2H,
J = 7.2 Hz, Ar-H), 7.66 (t, 1H, J = 7.4 Hz, Ar-H), 7.57–
7.50 (m, 4H, Ar-H), 7.41 (t, 1H, J = 7.2 Hz, Ar-H),
7.27 (d, 2H, J = 7.6 Hz, Ar-H), 5.18 (s, 2H, CH2), 3.19
(s, 3H, Me). MS (EI): m/z 367 (M+ꢀHCl, 12).
10. Alderton, W. K.; Cooper, C. E.; Knowles, R. G. Biochem.
J. 2001, 357, 593–615.
11. Low, S. Y. Mol. Aspects Med. 2005, 26, 97–138.
12. Salerno, L.; Sorrenti, V.; Di Giacomo, C.; Romeo, G.;
Siracusa, M. A. Curr. Pharm. Des. 2002, 8, 177–200.
13. Fan, J.-S.; Zhang, Q.; Li, M.; Tochio, H.; Yamazaki, T.;
Shimizu, M.; Zhang, M. J. Biol. Chem. 1998,273, 33472–33481.
14. Schulz, J. B.; Matthews, R. T.; Klockgether, T.; Dichgans,
J.; Flint Beal, M. Mol. Cell. Biochem. 1997, 174, 193–197.
15. Macdonald, J. E. Ann. Rep. Med. Chem. 1996, 221–230.
16. Babu, B. R.; Griffith, O. W. Curr. Opin. Chem. Biol. 1998,
2, 491–500.
17. Garvey, E. P.; Oplinger, J. A.; Tanoury, G. J.; Sherman,
P. A.; Fowler, M.; Marshall, S.; Harmon, M. F.; Paith, J.
E.; Furfine, E. S. J. Biol. Chem. 1994, 269, 26669–26676.
18. Shearer, B. G.; Lee, S. L.; Oplinger, J. A.; Frick, L. W.;
Garvey, E. P.; Furfine, E. S. J. Med. Chem. 1997, 40,
1901–1905.
4.2.11. 5-Benzoylimino-4-hydrazinocarbonimidoylmethyl-
2-methyl-3-oxo-2,3,4,5-tetrahydro-1,2,4-thiadiazole (21).
To a solution of 16 (50.0 mg, 0.18 mmol) in MeOH
(1.8 mL), MeONa (9.8 mg, 0.18 mmol) was added and
the mixture was stirred for 2 h until the imido ester 18
was obtained. Hydrazine Hydrochloride (13.7 mg,
0.20 mmol) was added to the reaction mixture and stir-
red at 45 ꢁC. After 72 h, the reaction mixture was cooled
to room temperature and the resulting precipitate was
1
filtered to give a white solid. Yield: 27.0 mg (48%). H
NMR (200 MHz, DMSO-d6): d 8.36 (br s, 1H, NH),
8.12 (br m, 2H, Ar-H), 7.64–7.49 (m, 3H, Ar-H), 6.27
19. Wolff, D. J.; Gauld, D. S.; Neulander, M. J.; Southan, G.
J. Pharmacol. Exp. Ther. 1997, 283, 265–273.

6206
T. Castan˜o et al. / Bioorg. Med. Chem. 16 (2008) 6193–6206
´
38. Martınez, A.; Castro, A.; Stud, M.; Rodriguez, J.;
20. Ruetten, H.; Southan, G. J.; Abate, A.; Thiemermann, C.
Br. J. Pharmacol. 1996, 118, 261–270.
´
Cardelu´s, I.; Llenas, J.; Fernandez, A.; Palacios, J. M.
Med. Chem. Res. 1998, 8, 171–180.
´
21. Gomez-Vidal, J. A.; Martasek, P.; Roman, L. J.; Silver-
man, R. B. J. Med. Chem. 2004, 47, 703–710.
22. Furfine, E. S.; Harmon, M. F.; Paith, J. E.; Garvey, E. P.
Biochemistry 1993, 32, 8512–8517.
39. Dorronsoro, I.; Chana, A.; Abasolo, I.; Castro, A.; Gil,
C.; Stud, M. QSAR Comb. Sci. 2004, 23, 89–98.
´
´
40. Quin˜ones, C.; Caceres, J.; Stud, M.; Martınez, A. Quant.
23. Di Giacomo, C.; Sorrenti, V.; Salerno, L.; Cardile, V.;
Guerrera, F.; Siracusa, M. A.; Avitabile, M.; Vanella, A.
Exp. Biol. Med. 2003, 228, 486–490.
Struct. Act. Relat. 2000, 19, 448–454.
41. Narayanan, K.; Griffith, O. W. J. Med. Chem. 1994, 37,
885–887.
24. Hagen, T. J.; Bergmanis, A. A.; Kramer, S. W.; Fok, K.
F.; Schmelzer, A. E.; Pitzele, B. S.; Swenton, L.; Jerome,
G. M.; Kornmeier, C. M.; Moore, W. M.; Branson, L. F.;
Connor, J. R.; Manning, P. T.; Currie, M. G.; Hallinan, E.
A. J. Med. Chem. 1998, 41, 3675–3683.
25. Collins, J. L.; Shearer, B. G.; Oplinger, J. A.; Lee, S. L.;
Garvey, E. P.; Salter, M.; Duffy, C.; Burnette, T. C.;
Furfine, E. S. J. Med. Chem. 1998, 41, 2858–2871.
26. Babu, B. R.; Griffith, O. W. J. Biol. Chem. 1998, 273,
8882–8889.
27. Lee, Y.; Martasek, P.; Roman, L. J.; Masters, B. S.;
Silverman, R. B. Bioorg. Med. Chem. 1999, 7, 1941–1951.
28. Sorrenti, V.; Di Giacomo, C.; Salerno, L.; Siracusa, M. A.;
Guerrera, F.; Vanella, A. Nitric oxide 2001, 5, 32–38.
29. Salerno, L.; Sorrenti, V.; Guerrera, F.; Sarva, M. C.;
Siracusa, M. A.; Di Giacomo, C.; Vanella, A. Pharmazie
1999, 54, 685–690.
30. Lee, Y.; Martasek, P.; Roman, L. J.; Silverman, R. B.
Bioorg. Med. Chem. Lett. 2000, 10, 2771–2774.
31. Lisboa, P. J.; Taktak, A. F. Neural Netw. 2006, 19, 408–415.
32. Durisova, M.; Dedik, L. Basic Clin. Pharmacol. Toxicol.
2005, 96, 335–342.
42. Li, H. Y.; Raman, C. S.; Martasek, P.; Kral, V.; Masters,
B. S. S.; Poulos, T. L. J. Inorg. Biochem. 2000, 81, 133–
139.
43. Castro, A.; Encinas, A.; Gil, C.; Brase, S.; Porcal, W.;
Perez, C.; Moreno, F. J.; Martinez, A. Bioorg. Med. Chem.
2008, 16, 495–510.
44. Singh, R.; Choubey, K.; Bhattacharya, A. J. Indian Chem.
Soc. 1998, 75, 430–431.
´
´
45. Parkanyi, C.; Yuan, H. L.; Cho, N. S.; Jaw, J.-H. J.;
Woodhouse, T. E.; Aung, T. L. J. Het. Chem. 1989, 26,
1331–1334.
´ ´
46. Cho, N. S.; Shon, H. I.; Parkanyi, C. J. Het. Chem. 1991,
28, 1645–1649.
´
47. L’abbe, G.; Albrecht, E. J. Het. Chem. 1992, 29, 451–454.
48. Guard, J. A. M.; Steel, P. J. Aust. J. Chem. 1995, 48,
1609–1615.
49. Kohara, Y.; Kubo, K.; Imamiya, E.; Wada, T.; Inada, Y.;
Naka, T. J. Med. Chem. 1996, 39, 5228–5235.
50. Pinner, A. Die Imidoa¨ther und ihre Derivate; Oppenheim:
Berlin, 1892.
51. Bolhofer, W. A.; Habecker, C. N.; Pietruszkiewicz, A. M.;
Torchiana, M. L.; Jacoby, H. I.; Stone, C. A. J. Med.
Chem. 1979, 22, 295–301.
33. Winkler, D. A. Mol. Biotechnol. 2004, 27, 139–168.
34. Kovesdi, I.; Dominguez-Rodriguez, M. F.; Orfi, L.;
Naray-Szabo, G.; Varro, A.; Papp, J. G.; Matyus, P.
Med. Res. Rev. 1999, 19, 249–269.
35. Csermely, P.; Agoston, V.; Pongor, S. Trends Pharmacol.
Sci. 2005, 26, 178–182.
52. Bredt, D. S.; Snyder, S. H. Proc. Natl. Acad. Sci. U.S.A.
1989, 86, 9030–9033.
53. Weininger, D. J. Chem. Inf. Comput. Sci. 1988, 28, 31–36.
54. Kohler, W. The Task of Gestalt Psychology; Princeton
University Press (Princeton): New Jersey, 1969.
55. TSARꢂ v3.3. Oxford Moleculars Ltd.
56. Sadat Hayatshahi, S. H.; Abdolmaleki, P.; Safarian, S.;
Khajeh, K. Biochem. Biophys. Res. Commun. 2005, 338,
1137–1142.
´
36. Stud, M. CODESꢂ v1.0 (revision 3).
´
37. Ochoa, C.; Rodrıguez, J.; Rodrıguez, M.; Chana, A.;
Stud, M.; Alonso-Villalobos, P.; Martınez-Grueiro, M.
´
´
M. Med. Chem. Res. 1998, 7, 530–545.