Quantitative structure-activity relationship study and directed synthesis of Thieno[2,3-d]pyrimidine-2,4-diones as monocarboxylate transporter 1 inhibitors

Article abstract of DOI:10.1007/s00044-011-9748-4

The aim of any quantitative structure-activity relationship (QSAR) study is not only to reveal relationships between structure of molecules and their biological activity, but also to explain it within the bounds of theoretical conceptions and to use the obtained model for prediction of properties of new compounds. That provides possibility to execute directed synthesis of new compounds with required biological activities. Monocarboxylate transporter 1 (MCT1) is one of the targets in a search for new immune response modulating and antitumor agents. In the present study, QSAR model for MCT1 binding affinity is developed. Decisive influence of relative negative partial charge, solvation energy, and radius of gyration on MCT1 inhibition has been detected. Theoretical explanation of the obtained model is given, and biological activity prediction for a series of N-vinyl derivatives of thieno[2,3-d]pyrimidine-2,4-dione is made. Directed synthesis of three leading compounds has been executed according to prediction results.

Full text of DOI:10.1007/s00044-011-9748-4

MEDICINAL
CHEMISTRY
RESEARCH
Med Chem Res (2012) 21:2263–2272
DOI 10.1007/s00044-011-9748-4
ORIGINAL RESEARCH
Quantitative structure-activity relationship study and directed
synthesis of Thieno[2,3-d]pyrimidine-2,4-diones
as monocarboxylate transporter 1 inhibitors
•
O. T. Devinyak Mikh. V. Slivka Mar. V. Slivka
•
•
•
V. M. Vais V. G. Lendel
Received: 8 March 2010 / Accepted: 8 July 2011 / Published online: 23 July 2011
ꢀ Springer Science+Business Media, LLC 2011
Abstract The aim of any quantitative structure-activity
relationship (QSAR) study is not only to reveal relation-
ships between structure of molecules and their biological
activity, but also to explain it within the bounds of theo-
retical conceptions and to use the obtained model for pre-
diction of properties of new compounds. That provides
possibility to execute directed synthesis of new compounds
with required biological activities. Monocarboxylate
transporter 1 (MCT1) is one of the targets in a search for
new immune response modulating and antitumor agents. In
the present study, QSAR model for MCT1 binding affinity
is developed. Decisive influence of relative negative partial
charge, solvation energy, and radius of gyration on MCT1
inhibition has been detected. Theoretical explanation of the
obtained model is given, and biological activity prediction
for a series of N-vinyl derivatives of thieno[2,3-d]pyrimi-
dine-2,4-dione is made. Directed synthesis of three leading
compounds has been executed according to prediction
results.
Introduction
Activation of T-cells following antigen challenge is an
important component of the immune response. However,
excessive activation of T-cells can lead to plenty of auto-
immune diseases and a reaction of transplant rejection.
Immunosuppressive agents, which are used to prevent the
rejection of transplanted organs or tissues and to treat
autoimmune diseases or diseases that are most likely of
autoimmune origin (e.g., rheumatoid arthritis, multiple
sclerosis, myasthenia gravis, systemic lupus erythematosus,
Crohn’s disease, pemphigus, ulcerative colitis, etc.), have
numerous side-effects and risks (Berchtold and Seitz, 1996;
Niethammer et al., 1999; Rifai et al., 2006). Therefore, the
search for new compounds with immunosuppressive
activity is important and actual.
The monocarboxylate transporters are a family of pro-
teins which transport lactate and other small monocarb-
oxylates. It was proved that the monocarboxylate
transporter 1 (MCT1) expression grows rapidly after
T-lymphocyte activation in order to eliminate lactate from
the cell, the quantity of which grows as a result of
increased glycolytic rate. Inhibition of lactate efflux by
potent blockade of lactate transport results in an accumu-
lation of lactate within the cell and feedback inhibition of
glycolysis. This suppression of cellular metabolism, with-
out being cytotoxic, results in an inability of T-lympho-
cytes to sustain the rapid rate of cell division, that takes
place during the early immune response to antigen
appearance. Thus, blockade of MCT1 is a new mechanism
of immunosuppression distinct from current therapies
(Murray et al., 2005).
Keywords QSAR Á Thieno[2,3-d]pyrimidine-2,4-dione Á
Monocarboxylate transporter 1 inhibitor Á
N-vinyl derivatives Á In silico screening Á Directed synthesis
O. T. Devinyak (&) Á V. M. Vais
Department of Pharmaceutical Disciplines, Uzhgorod National
University, Narodna sq., 1, Uzhgorod 88000, Ukraine
e-mail: o.devinyak@gmail.com
Mikh. V. Slivka Á Mar. V. Slivka Á V. G. Lendel
Department of Organic Chemistry, Uzhgorod National
University, O. Fedince Str., 53/1, Uzhgorod 88000, Ukraine
In addition, it has been reported that the blocking of
lactate transport to tumor cells is one of the reliable
methods to fight tumor growth. Experimental studies
O. T. Devinyak
Transcarpathian State University, Zankovecka Str., 89a,
Uzhgorod 88000, Ukraine
123

2264
Med Chem Res (2012) 21:2263–2272
promise MCT1 inhibition to be a new, efficient anticancer
treatment both by itself and combined with radiotherapy
(Sonveaux et al., 2008).
apprehensiveness to get too small training set. In an ideal
case, all the compounds have to be uniformly split into two
samples both in the descriptors’ space and by their activity.
In the present study, it was decided to split data set using
pKi values. For this purpose, compounds were sorted by the
activity in the ascending order, and every fourth compound
(starting from the first) was classified as test one. The
uniformity of test and training sets in the descriptors’ space
was checked using principal component analysis (PCA),
particularly by the projection of data into first two principal
components plane. The result (Fig. 1) shows satisfactory
compounds distribution by their descriptor values.
Thieno[2,3-d]pyrimidine-2,4-dione derivatives are
characterized by several biological activities. Alpha-1
adrenoreceptor antagonists (Russell et al., 1988), luteiniz-
ing hormone-releasing hormone receptor antagonists
(Sasaki et al., 2003), gonadotropin-releasing hormone
receptor antagonists (Betz et al., 2008), melanocortin
receptor agonists or antagonists (Sharma and Shi, 2005),
and also MCT1 inhibitors (Guile et al., 2006) have been
discovered in this class of compounds.
The purpose of current study is to reveal the dependence
of MCT1 inhibiting rate on molecular structure and to
execute directed synthesis of new thieno[2,3-d]pyrimidine-
2,4-dione derivatives with high MCT1 inhibiting activity.
A stepwise regression technique was used in order to
construct the model of MCT1 binding affinity. Stepwise
regression is a systematic method for adding and removing
terms from a multilinear model based on their statistical
significance in a regression (Draper and Smith, 1998).
Using default regression parameters (maximum p value for
a term to be added = 0.05, minimum p value for a term to
be removed = 0.10), only three descriptors were selected:
Results and discussion
Data sets
–
–
–
PEOE_RPC- relative negative partial charge;
E_sol solvation energy;
rgyr radius of gyration.
The comparativeness and homogeneity of experimental
data is an important precondition for perfect QSAR model
development because the values of the same activity for
same compounds vary due to the experimental conditions
and methods. Thus, MCT1 inhibition coefficients (Ki) used
for QSAR modeling were obtained during careful data
analysis from literature (Guile et al., 2006). The analysis
shows that out of 37 compounds with obtained Ki there
were only 22 compounds evaluated in the same assay (filter
binding assay). So only these compounds and their exper-
imental data were used for model development and vali-
dation. The complete set of compounds was divided into
corresponding training and test sets based on MCT1
binding affinity. Since the Ki varied by orders of magni-
tude, to guarantee the linear distribution of the dependent
variable, the Ki values were transformed to logarithmic
values [pKi = -log(1/Ki)], listed in Table 1. Before
descriptors calculations, molecular geometries were pre-
optimized by molecular mechanics calculations using the
MMFF94x force field as implemented in MOE, 2007 .09
(Chemical Computing Group Inc).
The calculated values for selected descriptors are listed
in Table 2.
In order to reveal intercorrelations of independent
variables and to discover the role of each descriptor in the
resulting model, a pair-wise correlation analysis was per-
formed, and correlation matrix of selected descriptors and
pKi was calculated (Table 3). Multi-collinearity between
the above mentioned three descriptors was tested by cal-
culating their variance inflation factors (VIF) (Studenmund,
2006). If VIF equals to 1.0—no intercorrelation exists for
each variable, and if it is larger than 5.0—that means that
the related model is unstable and a recheck is necessary.
The corresponding VIF values of selected descriptors are
shown in Table 4. As can be seen from this table, all the
variables have VIF values less than 2.0. Both the results
show the absence of significant intercorrelations between
selected descriptors, indicating that obtained model’s
coefficients have obvious statistic significance (Allen,
1997).
Statistical parameters of QSAR model are the following:
QSAR model building
N = 16; R2 = 0.878; p \ 10-5; F = 28.71; Q2LOO
=
=
0.8105; Q2ext = 0.7946, RMSE = 0.2236; RMSEPLOO
About 250 descriptors used as independent variables in
QSAR modeling were calculated using MOE software.
Then, all data processing was performed in MATLAB 7.7
(The MathWorks, Inc). The whole data was divided into a
training (16 compounds) and a test sets (6 compounds).
This proportion was selected as the compromise between
1/3 of data as recommended (Gramatica, 2007) and the
0.2789; RMSEPext = 0.3342, where N is the number of
compounds used, R2 is the squared correlation coefficient,
p is the significance level of the model, F is the Fisher
ratio, Q2LOO and Qe2xt are the squared correlation coefficients
for Leave-One-Out and external validations, respectively,
RMSE is the Root Mean Squared Error of the model (also
known as standard deviation of regression, S), RMSEPLOO
123

Med Chem Res (2012) 21:2263–2272
2265
Table 1 Molecular structures with the corresponding observed and predicted MCT1 inhibition activity
Molecular
Structure
pKi
pKi
Molecular
Structure
pKi
No
No
observed calculated
pKi observed calculated
1
9.48
10.00
9.55
9.63
9.95
12*
9.38
9.17
8.19
9.16
9.21
8.28
2
13
14
3*
10.08
4
5
9.46
8.22
9.59
8.18
15
16
7.92
8.96
8.30
8.56
6
7
8.66
7.89
8.37
8.19
17
18
8.31
8.04
8.55
7.82
123

2266
Med Chem Res (2012) 21:2263–2272
Table 1 continued
8
9
8.49
8.22
8.57
8.54
8.29
8.94
19*
20*
21*
8.26
8.06
7.43
8.00
8.03
7.80
10*
11
9.54
9.15
22
8.46
8.39
Compounds labeled with ‘‘*’’ are the test set; and other compounds are the training set
and RMSEPext are Root Mean Squared Errors of Prediction
in Leave-One-Out and external validation procedures,
respectively.
cross-validation using genetic algorithm (GA-MCVS)
with iterations number N = 100 000, validation set size
n_v = 7, and number of variables n = 3 (Konovalov et al.,
2008). The data filtering was executed; and constant,
repeated, and intercorrelated (with |r| [ 0.9) descriptors
were discarded preliminarily. The results have shown that
the minimum root mean squared error of prediction
(RMSEP = 0.3607) is represented by the previously
obtained descriptor set. It means that the obtained model
has a great probability to be globally optimal.
The obtained statistical parameters and model validation
results indicate its high quality and predictive power. A
plot of predicted pKi versus experimental pKi values for
both the training and test sets is shown in Fig. 2.
The agreement observed between the predicted and
experimental values confirmed the efficiency of this QSAR
model. In order to find outliers, the plot of the residuals
(predicted pKi - experimental pKi) versus experimental
pKi (Fig. 3) was studied. Considering the 3 standard
deviation limit line (3S) for spotting outliers, all the data
were retained in the model.
QSAR model interpretation
As it is indicated by model, normalized coefficients shown
in Table 4 and correlation coefficients between descriptors
and pKi (Table 3), the highest significance for MCT1
inhibition activity belongs to relative negative partial
charge (PEOE_RPC-). It is calculated as the smallest
negative partial charge qi divided by the sum of the negative
partial charges qi. And atomic partial charges in a molecule
are obtained using Partial Equalization of Orbital Electro-
negativities (PEOE) method (Gasteiger and Marsili, 1980).
Stepwise regression method gives the ability to find a
minimum number of variables necessary to build an ade-
quate model, but there is no guarantee that the obtained
model is globally optimal. Thus, the genetic algorithm as
more powerful global search heuristics was used as
implemented in QSAR_BENCH software (Konovalov
et al., 2007). In particular, the variable selection procedure
was carried out by RMSEP minimization in Monte-Carlo
123

Med Chem Res (2012) 21:2263–2272
2267
Table 3 Correlation matrix of experimental pKi values and molec-
ular descriptors selected for MCT1 binding affinity model
development
pKi observed
PEOE_RPC-
E_sol
rgyr
pKi observed
PEOE_RPC-
E_sol
1.000
0.782
1.000
0.057
0.537
0.172
0.079
1.000
-0.373
1.000
rgyr
It means an important contribution of electrostatic interac-
tion to MCT1 low molecular weight ligand binding. The
highest PEOE_RPC- values are represented by the com-
pounds that contain an atom with considerable negative
partial charge on it. This opinion is perfectly harmonized by
the fact that natural MCT1 ligands are monocarboxylates–
anions of short-chain organic acids (such as lactate, pyru-
vate, acetate, acetoacetate, b-hydroxybutyrate) (Halestrap
and Price, 1999). Monocarboxylates contain negative
charge delocalized over two carboxylate oxygen atoms.
As a result, MCT1 blockers have to be found among
organic acids, and other compounds that can form anions.
Nevertheless, it is known that ions with a localized charge
exist in the solvated state in polar solvents (such as water
and biological fluids). Therefore, solvent molecules coor-
dinated to ligand become a barrier for ligand binding. Thus,
the solvation energy barrier overcoming is a necessary
precondition for the interaction. Just that very case is
considered by the second descriptor—solvation energy
E_sol. Its positive coefficient value means that the pKi
increases when solvent coordination is more difficult, and
solvation sphere removing is easier. Since, there is not even
a minimal correlation between E_sol and pKi (Table 3),
this descriptor does not define MCT1 affinity itself, but it is
an important restrictive factor for PEOE_RPC-. This is
confirmed by the negative value of correlation coefficient
between PEOE_RPC- and E_sol. Thereby, the results
obtained in this study QSAR model once more confirms the
impossibility to develop perfect models using approaches
that discard so called insignificant (from the standpoint of a
single-parameter correlation) descriptors (e.g., heuristic
descriptor pre-selection procedure implemented in CO-
DESSA PRO - University of Florida, 2005).
Fig. 1 Projection of the compounds into first two principal compo-
nents plane (PC 1 and PC 2 describe 34.2 and 19.7% of total variance,
respectively)
Table 2 Calculated descriptor values for chemical structures
Compound
PEOE_RPC-
E_sol
rgyr
1
0.1563
0.1835
0.1696
0.1508
0.1336
0.1312
0.1297
0.1268
0.1241
0.1446
0.1467
0.1467
0.1296
0.1316
0.1364
0.1282
0.1266
0.1251
0.1229
0.1206
0.1279
0.1244
3.3014
4.3210
4.2096
4.4744
4.3448
3.8918
4.0009
3.9850
4.1868
4.3740
4.3396
4.4109
4.4112
4.3766
4.1793
4.1285
4.2967
4.3732
3.9626
4.2640
4.3337
3.8486
4.2520
2
-16.4083
-6.2025
8.1924
3*
4
5
1.8911
6
6.9452
7
2.6764
8
9.3244
9
-7.7311
-8.9567
-7.5846
-7.2084
21.6715
-7.5933
-10.9105
1.8018
10*
11
12*
13
14
15
16
17
18
19*
20*
21*
22
The third descriptor included in the model is the radius
of gyration rgyr. Its positive coefficient value means that
spatially branched molecules will show higher affinity to
MCT1. According to the authors’ opinion, it can be the
result of additional weak hydrogen and van der Waals
bonds formation. This formation plays a key role in ligand
emplacement near the active site of protein molecule in the
moment of proton accepting by anion because of equilib-
rium proton exchange between anionic ligand and the
positively charged amino acid.
-0.9526
-4.7219
-11.4607
-10.8194
-2.6092
3.1249
Compounds labeled with ‘‘*’’ are the test set; and other compounds
are the training set
123

2268
Med Chem Res (2012) 21:2263–2272
Table 4 Descriptors, coefficients, normalized coefficients, standard errors, t values, and variance inflation factors for the linear model
Physico-chemical meaning
Descriptors
Coefficient
Normalized coefficient
Standard error
t value
VIF
Intercept
Constant
PEOE_RPC-
E_sol
-2.1411
35.3745
0.0245
1.425
-3.3483
0.8479
0.3448
0.3637
1.6909
4.6509
0.0078
0.4066
-1.266
7.6059
3.1294
3.5047
Relative negative partial charge
Solvation energy
1.2193
1.1908
1.0563
Radius of gyration
rgyr
Fig. 2 Calculated versus experimental pKi values for both the
training and test sets. The diagonal in the plot is the y = x line
Fig. 3 Residuals versus experimental pKi values for both training and
test sets
QSAR model application
than the corresponding monomers (24a–g). But the anion
formation from compounds 25a–g, that may be essential
for MCT1 binding, is possible only after former hydrolysis.
So these compounds can be regarded only as prodrugs.
In modern drug discovery programs, compliance of
molecular structures with drug likeness criteria has a high
importance. Therefore, Lipinski’s drug-like test (Lipinski
et al., 1997) and Oprea’s lead-like test (Oprea, 2000) (both
implemented in MOE) were conducted for the studied
compounds. Results showed that molecular structures 23a–
g and 24a–g meet all the criteria. Molecular structures
25a–g failed both the drug-like and lead-like tests.
In order to reveal leading compounds among potent MCT1
blockers, in silico screening of N-vinyl derivatives of thi-
eno[2,3-d]pyrimidine-2,4-dione was carried out. The whole
set of evaluated compounds is presented in Table 5.
Deliberately reducing the molecular weight and com-
plexity (compounds 23a–g and 24a–g in comparison with
the compounds 1–22 used for model building and valida-
tion) an increase of PEOE_RPC- values was achieved, but
rgyr and E_sol values decreased (Table 2 vs. Table 5). As
a result of the model application, high pKi values for a
number of derivatives containing alcoholic hydroxyl
(compounds 23a–g, Table 5) were predicted. The predicted
pKi values for thiol derivatives (compounds 24a–g) were
lower. And for both the series of molecules, larger pKi
were observed for compounds with condensed cyclopen-
tene or cyclohexene rings (23f,g and 24f,g). Several dimers
with considerably decreased PEOE_RPC- simultaneously
with the increased rgyr and E_sol values (compounds 25a–
g) were also included to the set of screening compounds.
This dimers have shown higher predicted MCT1 affinities
Since the compounds 23f and 23g, and 24f and 24g have
the very close predicted pKi values, in order to select
leading compounds to synthesize, the following analysis
was made. From the drug structure database DrugBank
(http://www.drugbank.ca) the two samples of molecules
containing, respectively, cyclopentene or cyclohexene ring
as a substructure were received. So 115 structures with
cyclohexene ring, and only 10 structures with cyclopentene
ring were found. Since there is more than a tenfold
123

Med Chem Res (2012) 21:2263–2272
2269
Table 5 Selected descriptors and predicted pKi values for analyzed compounds
Compound
R1
R2
PEOE_RPC-
E_sol
rgyr
pKi (predicted)
23a
23b
23c
23d
23e
23f
CH3
CH3
CH3
CH3
CH3
H
0.2005
0.1979
0.1931
0.1882
0.1890
0.1965
0.1913
0.1595
0.1572
0.1529
0.1485
0.1493
0.1559
0.1513
0.0839
0.0826
0.0803
0.0779
0.0782
0.0819
0.0794
-14.8547
-13.2489
-13.7838
-9.7605
-12.4710
-12.7736
-12.5710
-12.8694
-12.1241
-11.8150
-11.4918
-12.5344
-12.7663
-12.6884
-15.4518
0.0983
3.4041
3.4415
3.5130
3.5775
3.5961
3.5247
3.6372
3.4455
3.4873
3.5528
3.6061
3.6415
3.5669
3.6773
5.2819
5.2233
5.8252
5.9173
5.6233
5.6502
5.8241
9.44
9.44
9.36
9.37
9.36
9.51
9.50
8.09
8.09
8.04
7.97
8.02
8.14
8.14
7.97
8.23
8.83
8.62
8.27
8.96
8.57
CH3
C2H5
n-C3H7
iso-C3H7
(CH2)3
(CH2)4
23g
24a
24b
24c
24d
24e
24f
CH3
CH3
CH3
CH3
CH3
H
CH3
C2H5
n-C3H7
iso-C3H7
(CH2)3
(CH2)4
24g
25a
25b
25c
25d
25e
25f
CH3
CH3
CH3
CH3
CH3
H
CH3
C2H5
n-C3H7
iso-C3H7
-6.9993
-17.4794
-15.2369
6.0621
(CH2)3
(CH2)4
25g
-16.3663
difference in the sizes of the received samples, it was
decided to synthesize cyclohexene derivatives, specifically:
splitting oxazolino(thiazolino-)thienopyrimidinium bromides
26a–c were developed by varying synthesis conditions and
the nature of nucleophilic reagent (Scheme 1).
•
•
Compound 23g—as a leading compound with the
highest predicted pKi value;
Nucleophilic cleavage was carried out by the action of
sodium carbonate. In the first phase, the attack of hydroxyl
group on the ‘‘nodal’’ carbon of pyrimidine cycle is imple-
mented with the next addition and pseudo-base (A) forma-
tion. Then the destruction of oxazolidine (thiazolidine) ring
takes place with the intermediate (B) formation, and under
the influence of strong nucleophile the elimination of
hydrogen bromide follows. This leads to the formation of
N-vinyl derivatives of thieno[2,3-d]pyrimidine-2,4-dione
23g and 24g. In the case of formation of N-thiovinyl deriv-
atives from corresponding tribromides compound 25g is
received. This can be explained by the presence of free
bromine molecules in the reaction atmosphere, acting as
Compounds 24g and 25g—as model representatives of
the series 24a–g and 25a–g, respectively.
Chemistry
A convenient way to get vinyl functional derivatives of
heterocycles by splitting oxazoline or thiazoline rings was
chosen (Slivka et al., 2008). Oxazolino(thiazolino-)thieno-
pyrimidine salts 26a–c were chosen as starting compounds.
An active electrophilic center located at the ‘‘nodal’’ Carbone
was used as the reaction center. The preparative methods of
123

2270
Med Chem Res (2012) 21:2263–2272
Scheme 1 Synthesis of
selected compounds
oxidants of previously formed aliphatic mercapto groups.
The overall reaction is shown in Scheme 1.
coefficient (R2); the Root Mean Squared Error of the model
(RMSE, also known as the standard deviation of regression
S); the significance of the model (p); and the Fisher ratio
value (F).
Known methods of synthesis from the corresponding
allyl(thio-)ethers were used for the preparation of starting
oxazolino(thiazolino-)thienopyrimidinium bromides 26a–c
(Khripak et al., 2004).
The predictive stability and robustness of the model
depends not only on model building approaches, but also
on the quality of initial data. Therefore, in order to detect
multicollinearity between the selected descriptors the var-
iance inflation factors (VIF) were calculated as follows:
VIF = 1/(1 - R2), where R2 is the correlation coefficient
of multiple regression between one variable and the others
in the model.
Conclusions
The study of quantitative relationship between molecular
structure and monocarboxylate transporter 1 inhibition is
carried out and QSAR model that can predict pKi is
obtained in present study. It is shown that the inhibition
constant increases with increasing relative negative partial
charge, solvation energy, and radius of gyration of molecules.
Using obtained QSAR model, the activities of N-vinyl
derivatives of thieno[2,3-d]pyrimidin-2,4-dione are pre-
dicted, and the leading compound is identified. The synthesis
of the leading compound and two model representatives of
the studied molecules is executed, and preparative methods of
current synthesis are developed. Further investigations of the
synthesized compounds in vitro and in vivo are necessary
for QSAR model optimization and for optimization of the
leading compound structure.
Validation procedure is the most important step in the
development of reliable QSAR models. Obtained QSAR
model was first verified by internal cross-validation by
calculating the following parameters: Q2LOO (the squared
correlation coefficient for Leave-One-Out cross-validation)
and RMSEPLOO (the Root Mean Squared Error of Predic-
tion in Leave-One-Out validation). Using test set, the
model was further checked by external validation by cal-
culating parameters: Q2ext (the squared correlation coeffi-
cient for external validation) and RMSEPext (the Root
Mean Squared Error of Prediction in external validation
procedure).
Chemistry
Experimental section
Melting points were determined in open capillaries on a
Mel-Temp apparatus and are uncorrected. The purity of the
compounds was confirmed by the thin-layer chromatogra-
phy (TLC) performed on Sorbfil plates (Russia) at 27ꢁC
(silicagel, ethanol/diethyl ether/hexane 1:3:1). Spots were
Statistical methods and model validation
The goodness of fit of the model was evaluated using the
following statistical parameters: the squared correlation
123

Med Chem Res (2012) 21:2263–2272
2271
detected by their absorption under UV light and by visu-
alization using 0.05 mol I2 in 10% HCl. IR spectra were
recorded on an UR-20 spectrometer, as KBr pellets and
d6) d: 1.76 (m, 4H, 2CH2), 2.72 (m, 2H, CH2), 2.78 (m, 2H,
CH2), 4.12 (s, 2H, CH2), 5.66 (s, 1H, =CH2), 5.92 (s, 1H,
=CH2), 7.20–7.50 (m, 5H, C6H5) ppm; IR (KBr), m: 2480
(S–H), 1640 (C=O) cm-1. Anal. Calcd. for C19H18N2O2S2
(370.488), %: C, 61.62; H, 4.86; N, 7.57; S, 17.30. Found,
%: C, 61.51; H, 4.81; N, 7.62; S, 17.39.
wave numbers were given in cm-1
.
1H NMR spectra were obtained using a Varian VXR-
300 spectrometer (300 MHz) in DMSO-d6. Chemical
shifts are reported in d values (ppm) relative to TMS d = 0
(1H), as internal standard. The microanalyses were per-
formed on Perkine-Elmer 240C elemental analyzer.
The starting oxazolino(thiazolino-)thienopyrimidinium
bromides 26a–c were prepared according to the literature
procedure (Khripak et al., 2004).
1,1’-(disulfanediyldiprop-1-ene-3,2-diyl)bis(3-phenyl-
5,6,7,8-tetrahydrobenzo[b]thieno[2,3-d]pyrimidine-
2,4(1H, 3H)-dione) (25g)
Tribromide 26c (0.68 g, 1 mmol) was dissolved in 20 ml
of heated DMSO. To the stirring and cooling resulting
mixture, a solution of sodium carbonate (0.53 g, 5 mmol)
in 5 ml of water was added dropwise. The reaction mixture
was stirred at room temperature for 10 min, and then the
hydrolysis product was precipitated by adding 100 ml of
cooled water. Precipitate was filtered and washed by 50 ml
of warm water acidified by 1 ml of acetate acid. The
reaction product was crystallized from dioxane. Yield:
72%. Light yellow crystals; mp: 227–228ꢁC; TLC:
Rf = 0.64; 1H NMR (300 MHz, DMSO-d6) d: 1.74 (m, 4H,
2CH2), 2.74 (m, 2H, CH2), 2.79 (m, 2H, CH2), 4.06 (s, 2H,
CH2), 5.64 (s, 1H, =CH2), 5.90 (s, 1H, =CH2), 7.22–7.48
(m, 5H, C6H5) ppm; IR (KBr), m: 1640 (C=O) cm-1. Anal.
Calcd. for C38H34O4N4S4 (738.961), %: C, 61.79; H, 4.61;
N, 7.59; S, 17.34. Found, %: C, 61.62; H, 4.51; N, 7.62; S,
17.42.
1-(3-hydroxyprop-1-en-2-yl)-3-phenyl-5,6,7,8-
tetrahydrobenzo[b]thieno[2,3-d]pyrimidine-2,4(1H, 3H)-
dione (23g)
Monobromide 26a (0.50 g, 1 mmol) was dissolved, when
heated in 20 ml DMSO. To the stirred resulting mixture, a
solution of sodium carbonate (0.53 g, 5 mmol) in 5 ml of
water was added. The reaction mixture was stirred under
heating at 70–80ꢁC for 4 h, and then the hydrolysis product
was precipitated by adding 100 ml of cooled water. Pre-
cipitate was filtered and washed by 50 ml of warm water
acidified by 1 ml of acetate acid, and then was dissolved in
ethanol heated to boiling. The mixture was cooled and
formed during cooling precipitate was separated. The
reaction product was participated from the filtrate by add-
ing water and was crystallized from 50% ethanol–water
solution. Yield: 69%. Colorless crystals; mp: 148–149ꢁC;
Acknowledgments The authors thank Transcarpathian Regional
Charitable Foundation ‘‘Assistance to Students and Teachers of
Medical Faculty Renaissance’’ for supporting this study.
1
TLC: Rf = 0.78; H NMR (300 MHz, DMSO-d6) d: 1.75
(m, 4H, 2CH2), 2.69 (m, 2H, CH2), 2.76 (m, 2H, CH2),
4.81–4.83 (m, 2H, CH2), 5.78 (s, 1H, =CH2), 6.14 (s, 1H,
=CH2), 7.25–7.51 (m, 5H, C6H5) ppm; IR (KBr), m:
3400–3600 (O–H), 1680 (C=O) cm-1. Anal. Calcd. for
C19H18N2O3S (354.423), %: C, 64.41; H, 5.08; N, 7.91; S,
9.04. Found, %: C, 64.11; H, 5.11; N, 7.97; S, 9.09.
References
Allen MP (1997) Understanding regression analysis. Plenum Press,
New York, pp 176–180
Berchtold P, Seitz M (1996) Immunosuppression—a tightrope walk
between iatrogenic harm and therapy. Schweiz Med Wochenschr
38:1603–1609
Betz SF, Zhu Y, Chen C, Struthers RS (2008) Non-peptide
gonadotropin-releasing hormone receptor antagonists. J Med
Chem 51(12):3331–3348
1-(3-mercaptoprop-1-en-2-yl)-3-phenyl-5,6,7,8-
tetrahydrobenzo[b]thieno[2,3-d]pyrimidine-2,4(1H, 3H)-
dione (24g)
Draper NR, Smith H (1998) Applied regression analysis. Wiley-
Interscience, Hoboken, pp 307–312
Gasteiger J, Marsili M (1980) Iterative partial equalization of orbital
electronegativity—a rapid access to atomic charges. Tetrahedron
36:3219–3228
Monobromide 26b (1.03 g, 2 mmol) was dissolved in
20 ml of heated DMSO. To the stirring and cooling
resulting mixture, a solution of sodium carbonate (0.53 g,
5 mmol) in 5 ml of water was added dropwise. The reac-
tion mixture was stirred at room temperature for 10 min,
and then the hydrolysis product was precipitated by adding
100 ml of cooled water. Precipitate was filtered and
washed by 50 ml of warm water acidified by 1 ml of
acetate acid. The reaction product was crystallized from
methanol. Yield: 54%. Light yellow crystals; mp:
123–125ꢁC; TLC: Rf = 0.86; 1H NMR (300 MHz, DMSO-
Gramatica P (2007) Principles of QSAR models validation: internal
and external. QSAR Comb Sci 26:694–701
Guile SD, Bantick JR, Cheshire DR, Cooper ME, Davis AM, Donald
DK, Evans R, Eyssade C, Ferguson DD, Hill S, Hutchinson R,
Ingall AH, Kingston LP, Martin I, Martin BP, Mohammed RT,
Murray C, Perry MWD, Reynolds RH, Thorne PV, Wilkinson
DJ, Withnall J (2006) Potent blockers of the monocarboxylate
transporter MCT1: novel immunomodulatory compounds. Bio-
org Med Chem Lett 16:2260–2265
123

2272
Med Chem Res (2012) 21:2263–2272
Halestrap AP, Price NT (1999) The proton-linked monocarboxylate
transporter (MCT) family: structure, function and regulation.
Biochem J 343:281–299
Khripak SM, Plesha MV, Slivka MV, Yakubets VI, Krivovyaz AA
(2004) Syntesis and reactivity of 1-bromomethyl-5-oxo-4-
phenyl-1, 2, 4, 5, 6, 7, 8, 9-octahydrobenzo[4, 5]thieno[3,
2-e][1, 3]oxazolo[3, 2-a]-pyrimidin-11-ium bromides. Zhurnal
Organicheskoi Khimii 40:1705–1706
Konovalov DA, Coomans D, Deconinck E, Vander Heyden Y (2007)
Benchmarking of QSAR models for blood-brain barrier perme-
ation. J Chem Inf Model 47:1648–1656
Konovalov DA, Sim N, Deconinck E, Vander Heyden Y, Coomans D
(2008) Statistical Confidence for Variable Selection in QSAR
models via Monte Carlo Cross-Validation. J Chem Inf Model
48:370–383
Oprea TI (2000) Property distribution of drug-related chemical
databases. J Comp Aided Mol Des 14:251–264
Rifai K, Kirchner GI, Bahr MJ, Cantz T, Rosenau J, Nashan B,
Klempnauer JL, Manns MP, Strassburg CP (2006) A new side
effect of immunosuppression: High incidence of hearing impair-
ment after liver transplantation. Liver Transpl 12(3):411–415
Russell RK, Press JB, Rampulla RA, McNally JJ, Falotico R, Keiser
JA, Bright DA, Tobia A (1988) Thiophene systems. 9. Thieno-
pyrimidinedione derivatives as potential antihypertensive agents.
J Med Chem 31(9):1786–1793
Sasaki S, Cho N, Nara Y, Harada M, Endo S, Suzuki N, Furuya S,
Fujino M (2003) Discovery of a thieno[2, 3-d]pyrimidine-2,
4-dione bearing a p-methoxyureidophenyl moiety at the 6-posi-
tion: a highly potent and orally bioavailable non-peptide
antagonist for the human luteinizing hormone-releasing hormone
receptor. J Med Chem 46(1):113–124
Lipinski CA, Lombardo F, Dominy BW, Feeney PJ (1997) Exper-
imental and computational approaches to estimate solubility and
permeability in drug discovery and development settings. Adv
Drug Deliv Rev 23:3–25
Sharma SD, Shi Y (2005) Thieno[2,3-d]pyrimidine-2,4-dione mela-
nocortin-specific compounds. US Patent 2005/0124636, 9 June
2005
Chemical Computing Group Inc. MOE 2007.09 (Molecular Operating
Environment software). http://www.chemcomp.com
Murray CM, Hutchinson R, Bantick JR, Belfield GP, Benjamin AD,
Brazma D, Bundick RV, Cook ID, Craggs RI, Edwards S, Evans
LR, Harrison R, Holness E, Jackson AP, Jackson CG, Kingston
LP, Perry MW, Ross AR, Rugman PA, Sidhu SS, Sullivan M,
Taylor-Fishwick DA, Walker PC, Whitehead YM, Wilkinson
DJ, Wright A, Donald DK (2005) Monocarboxylate transporter
MCT1 is a target for immunosuppression. Nat Chem Biol
1:371–376
Slivka MV, Slivka MV, Usenko RM, Lendel VG (2008) Method of
obtaining N-vinyl functional derivatives of 5,6-membered nitro-
gen-containing heterocycles. UA Patent 37721, 10 Dec 2008
Sonveaux P, Vegran F, Schroeder T, Wergin MC, Verrax J, Rabbani
ZN, De Saedeleer CJ, Kennedy KM, Diepart C, Jordan BF,
Kelley MJ, Gallez B, Wahl ML, Feron O, Dewhirst MW (2008)
Targeting lactate-fueled respiration selectively kills hypoxic
tumor cells in mice. J Clin Invest 118(12):3930–3942
Studenmund AH (2006) Using econometrics: a practical guide, 5th
edn. Pearson International Edition, New Delhi, pp 258–259
The MathWorks, Inc. MATLAB (R2008b) (The Language of Tech-
nical Computing) Version 7.7.0.471. http://www.mathworks.com
University of Florida (2005) CODESSA PRO user’s manual. p 82.
http://www.codessa-pro.com
Niethammer D, Ku¨mmerle-Deschner J, Dannecker GE (1999) Side-
effects of long-term immunosuppression versus morbidity in
autologous stem cell rescue: striking the balance. Rheumatology
38:747–750
123