molecular modeling of pyridine derivatives for cox-2 inhibitors: quantitative structure–activity...
TRANSCRIPT
ORIGINAL RESEARCH
Molecular modeling of pyridine derivatives for COX-2 inhibitors:quantitative structure–activity relationship study
Amrita Dwivedi • A. K. Srivastava •
Ajeet Singh
Received: 4 March 2013 / Accepted: 6 August 2013 / Published online: 24 September 2013
� Springer Science+Business Media New York 2013
Abstract Quantitative structure–activity relationship
studies were performed on a set of compounds of pyridine
acyl sulfonamide derivative to understand the structural
features influencing the affinity toward the COX-2 enzyme
inhibition. In the present study, the density functional
theory-based descriptors were calculated at B3LYP/6-
31?G* level in gas phase. However, empirical molecular
descriptors were calculated with the help of different
softwares. Number of physicochemical, topological, and
electronic descriptors were computed, and the calculated
results revealed that the descriptors based on softness (S),
hardness (g), chemical potential (l), and the lowest unoc-
cupied molecular orbital energy seem to be responsible
factor for in vitro inhibition activity of COX-2 enzyme.
The possible descriptors which we have calculated for the
present series of compounds were selected for multiple
linear regression analysis. Various regression models have
been tested and the regression analysis data indicate that
some of the descriptors provide valuable information
toward designing of new COX-2 enzyme inhibitors with
high efficacy. The predictive ability of the models was
cross validated by observation of the low residual activity
values and appreciable cross validated R2 values (R2CV)
obtained by leave one out technique.
Keywords COX inhibition � DFT-QSAR �Regression analysis � Molecular descriptor �Frontier electron density
Introduction
Quantitative structure–activity relationship (QSAR) mod-
els are regression models used in the chemical and bio-
logical sciences and engineering. Like other regression
models, QSAR models relate measurements on a set of
‘‘predictor’’ variables to the behavior of the response var-
iable. QSAR represents an attempt to correlate activities or
properties with structural descriptors of compounds. To
correlate and predict physical, chemical and biological
activity from molecular structure is a very important and an
unsolved problem in theoretical and computational chem-
istry as well (Hu et al., 2003). QSAR analysis has been
widely used to modify lead compounds and to minimize
the toxic effects (Plummer, 1995). It provides better sense
to understand the interaction mechanisms between chemi-
cal compounds and biological targets (Gao and Hansch,
1996; Lien and Gao, 1995; Hansch et al., 1996). The
fundamental hypothesis of QSAR is that biological activity
is a function of molecular structure. Thus, molecules with
similar structures exert similar biological activities, and
changes in structure are thought to modulate biological
activities.
Cyclooxygenase (COX) is one of the key enzymes that
are responsible for the production of proinflammatory
prostaglandins and other isotypes of this compound, which
in turn play a pivotal role in gastrointestinal protection. It is
now established that two distinct isoforms of this enzyme
exist. The first one is the constitutive form (COX-1)
expressed virtually in all tissues, which is involved in the
Electronic supplementary material The online version of thisarticle (doi:10.1007/s00044-013-0691-4) contains supplementarymaterial, which is available to authorized users.
A. Dwivedi � A. K. Srivastava (&) � A. Singh (&)
QSAR Research Laboratory, Department of Chemistry,
University of Allahabad, Allahabad 211002, UP, India
e-mail: [email protected]
A. Singh
e-mail: [email protected]
123
Med Chem Res (2014) 23:1865–1877
DOI 10.1007/s00044-013-0691-4
MEDICINALCHEMISTRYRESEARCH
production of prostaglandins, important in gastrointestinal
protection (Vane, 1971; Smith and Willis, 1971). The
second isoform (COX-2) that is largely restricted to the
brain and kidney. During normal physiology COX-2 levels
are undetectable in most tissues. The level of COX-2 is
induced under the conditions of immunological insult (such
as inflammation) (Needleman and Isakson, 1997; Masferrer
et al., 1992). Hence, selective and potent inhibitors of
COX-2 might be anti-inflammatory without the side effects
associated with traditional NSAIDs (Non-Steroidal Anti-
inflammatory Drugs) such as aspirin, which inhibit both
COX-1 and COX-2 in a nonselective manner.
3D structure of COX-1 and COX-2 shows very much
similarities such as each consisting of a long, narrow
hydrophobic channel however, the difference of single
amino acid between two isoform of COX enzyme arises
difficulties for the selectivity of many COX inhibitory
drugs. The most critical structural feature conferring sen-
sitivity to inhibition by COX-2 is exchange of valine in
COX-2 at positions 434 and 523 in place of isoleucine in
COX-1 (Hawkey, 1999). This difference increases the
volume of the COX-2 side pocket, which is thought to be
the binding site of many selective drugs. Another factor
which increases sensitivity of drug toward COX-2 is
exchange of histamine at position 513 in COX-1 in place of
arginine in COX-2. Overall larger size of the central
channel of the COX-2 binding pocket increases selectivity
of drugs toward COX-2 enzyme.
Since a long time, attempts have been made by different
research groups for designing of various COX-2 inhibitors
experimentally; however, the QSAR studies on COX-2
inhibitor is scarce in the literature (Singh and Bhardwaj,
2010; Zarghi et al., 2009, 2010; Garg et al., 2003; Sarkar
and Mostafa, 2009). As part of a continuing effort to
develop COX-2 inhibitor, Zhu et al. (2011) recently
reported the synthesis of various pyridine acyl sulfonamide
derivatives as novel COX-2 inhibitors. Sulfonamide is a
promising skeleton that has shown not only the COX-2-
selective inhibition but also the minimization of the risk
potential of unwanted side effects. The two combined
substructures, pyridine ring and sulfonamide, might exhibit
synergistic effect and inflammatory activities. The objec-
tive of present study is to perform a QSAR study on the
present series to correlate the physicochemical and struc-
tural requirements of these compounds to exhibit optimal
inhibitory potency toward COX-2 enzyme which will in
turn help in the modeling of COX-2 inhibitors.
Theoretical background
In the present study, steric parameters such as molecular
weight (Mw), molecular volume (Mv), molar refractivity
(Mr), parachor (Pc), index of refraction (IOR), surface
tension (st), density (D), molecular connectivity (v);
topological parameters such as Balaban indices (J), Wiener
index (W), mean Wiener index (WA), Balaban centric index
(BAC), information theoretic index (ID); electronic
parameters such as equalized electronegativity (Xeq),
polarizability (Pz); hydrophobic parameter such as parti-
tion coefficient (log P); and other parameters based on
density functional theory (DFT) calculations such as total
energy, dipole moment, highest occupied molecular orbital
energy (HOMO), the lowest unoccupied molecular orbital
energy (LUMO), hardness (g), chemical potential (l),
electrophilicity index (w), and softness (S) have been cal-
culated for all the derivatives of pyridine presented in the
series. Out of above descriptors, Mw, Mv, Mr, Pc, st, D,
IOR, and Pz were calculated by means of ACD Lab Chem.
Sketch version 12.0 Software (2009), whereas v, J, W, WA,
and BAC were evaluated by means of E-Dragon Software
(Tetko et al., 2005). DFT descriptors such as total energy,
dipole moment, HOMO, LUMO, g, l, w. and S were per-
formed with the help of the Turbomole program package
(Turbomole v6.0, 2009) using Becke’s three-parameter
exchange functional with correlation functional (Becke,
1993) of Lee, Yang, and Parr (B3LYP) (Lee et al., 1988;
Puzyn et al., 2008). All the species were fully optimized in
gas phase with 6-31?G* basis set, and harmonic vibra-
tional frequency calculations were used to confirm that the
optimized structures were minima, as characterized by
positive vibrational frequencies (Muller et al., 2001). The
statistical significance of the models was determined by
examining the regression coefficient, the standard devia-
tion, the number of variables, the cross validation leave-
one-out statistics, and the proportion between the cases and
variables in the equation (Coats et al., 1985). We have used
Hansch analysis for developing the models (Hansch and
Leo, 1995). The multiple regression analysis was used to
derive the correlation using SPSS 7.5 version program. The
detailed descriptions of the parameters that are used in the
present study are given in the supporting information.
Results and discussion
QSAR studies were performed on a set of compounds of
pyridine acyl sulfonamide derivative (Zhu et al., 2011).
The biological activity data of these compounds were
correlated with different steric parameters such as st, D, ID,
and molecular connectivity (3v and 5v); topological
parameters such as W, WA, Balaban connectivity distance
index (J); and DFT parameters such as S, g, l, and LUMO,
which have been found to be useful in QSAR-based drug
modeling (Srivastava et al., 2008a, b, c; Agarwal et al.,
2003, 2005, 2006; Khadikar et al., 2002, 2003a, b, 2005a,
1866 Med Chem Res (2014) 23:1865–1877
123
b). Structural details of the compounds with their experi-
mental activity (log IC50) and indicator parameter values
are given in Table 1. Indicator parameters are not QSAR
parameters; however, they will be used as dummy
parameters that indicate the significance of any particular
group or species at a particular substitution site in a given
series of drugs. In this study, we have taken the indicator
parameters as I1, I2, I3, I4, I5, and I6 and corresponding
indicator parameters are given in Scheme 1. In Table 1, the
value of indicator parameter is 1 if that group is present at a
particular position. For example, I4 indicator parameter is
considered as 1 ifN
(pyridine) is present at R2
position (11–14).
Various physicochemical, topological, and 3D parame-
ters which are used for making various significant models
are given in Table 2. Statistically significant models were
obtained when some of the physicochemical and topolog-
ical parameters such as st, D, ID, molecular connectivity
(3v and 5v), W, WA, Balaban connectivity distance index
(J) and DFT parameters such as S, g, l and LUMO are
combined with the indicator parameter.
The first step in obtaining a statistically significant
model is to investigate whether any collinearity exists
between the parameters used. This is achieved by obtaining
correlation matrix; such a matrix obtained in the present
case for the physicochemical, topological, and indicator
parameters along with respective activities is shown in
Table 3, and the one obtained for DFT parameter and
indicator parameter along with activities is shown in
Table 4. In practice, every term in correlation matrix [0.5
can be taken as suspicious because of collinearity.
The stepwise development of model along with changes
in statistical qualities on gradual addition of descriptors
pIC50 ¼ þ1:551 �0:588ð ÞS� 2:862
n ¼ 24; R ¼ 0:760; R2 ¼ 0:577; R2A ¼ 0:588;
SE ¼ 0:218; Fð1�22Þ ¼ 30:004; Q ¼ 3:486:
ð1Þ
Equation 1 explains only 57.7 % variance in the COX-2
inhibitory activity. It shows that descriptor S contributes
positively. Based on variance percentage, it is not a very
good significant equation; addition of indicator parameter
I6 , however, slightly improves the statics of the equation.
pIC50 ¼þ 1:557 �0:482ð ÞS
� 0:308 �0:186ð ÞI6 � 2:828
n ¼ 24; R ¼ 0:854; R2 ¼ 0:729; R2A ¼ 0:704;
SE ¼ 0:178; Fð2�21Þ ¼ 28:294; Q ¼ 4:798:
ð2Þ
Equation 2 explains only 72.9 % variance in the COX-2
inhibitory activity. It shows that indicator parameter I6 has
negative coefficient suggesting that the group presented by
I6 contributes negatively toward the activity of drugs.
Based on variance percentage, it is not a very good
significant equation; addition of another indicator
parameter I4 , however, slightly improves the statics of
the equation.
pIC50 ¼þ 1:488 �0:382ð ÞS
� 0:296 �0:147ð ÞI6
� 0:288 �0:161ð ÞI4 � 2:421
n ¼ 24; R ¼ 0:917; R2 ¼ 0:840; R2A ¼ 0:817;
SE ¼ 0:140; Fð3�20Þ ¼ 35:128; Q ¼ 6:550:
ð3Þ
Equation 3 explains only 84.0 % variance in the COX-2
inhibitory activity. It shows that indicator parameter I4 has
negative coefficient suggesting that the group presented by
I4 contributes negatively toward the activity of drugs.
Based on variance percentage, it is not a very good
significant equation; addition of another indicator
parameter I5 , however, slightly improves the statics of
the equation.
pIC50 ¼þ 1:378 �0:311ð ÞS
� 0:239 �0:122ð ÞI6
� 0:307 �0:129ð ÞI4
þ 0:212 �0:124ð ÞI5 � 1:902
n ¼ 24; R ¼ 0:951; R2 ¼ 0:905; R2A ¼ 0:884;
SE ¼ 0:111; Fð4�19Þ ¼ 45:011; Q ¼ 8:568:
ð4Þ
Equation 4 explains 90.5 % variance in the COX-2
inhibitory activity. It shows that indicator parameter I5
has positive coefficient suggesting that the group presented
by I5 contributes positively toward the activity of drugs.
Introducing g parameter in place of S parameter also gives
significant model. Based on variance percentage, it is a
good significant equation and model.
pIC50 ¼� 76:188 �17:23ð Þg
þ 0:228 �0:124ð ÞI5
� 0:240 �0:122ð ÞI6
� 0:301 �0:129ð ÞI4 þ 12:590
n ¼ 24; R ¼ 0:951; R2 ¼ 0:904; R2A ¼ 0:884;
SE ¼ 0:111; Fð4�19Þ ¼ 44:946; Q ¼ 8:568:
ð5Þ
Equation 5 explains 90.4 % variance in the COX-2 inhib-
itory activity. It shows that descriptor g contributes nega-
tively. Thus, Eqs. 4 and 5 represent best model among all
the obtained models.
Med Chem Res (2014) 23:1865–1877 1867
123
Table 1 Biological activity and indicator parameter of pyridine acyl sulfonamide derivatives
S
NH O
OO
R1
R2
Compound no. R2 R1 I1 I2 I3 I4 I5 I6 pIC50
1
N
H 0 0 1 0 0 0 4.801
2
N
F 0 0 0 0 0 0 4.932
3
N
Cl 0 0 0 0 0 0 5.180
4
N
Br 0 0 0 0 1 0 5.310
5
N
CH3 0 0 0 0 0 1 4.740
6 S
N
NHO
O O
H
H 1 0 1 0 0 0 5.102
7 S
N
NHO
O O
F
F 0 0 0 0 0 0 5.252
8 S
N
NHO
O O
Cl
Cl 0 0 0 0 0 0 5.468
1868 Med Chem Res (2014) 23:1865–1877
123
Table 1 continued
Compound no. R2 R1 I1 I2 I3 I4 I5 I6 pIC50
9
S
N
NHO
O O
Br
Br 0 0 0 0 1 0 5.553
10 S
N
NHO
O O
Br
CH3 0 0 0 0 0 1 5.070
11
N
F 0 0 0 1 0 0 4.917
12
N
Cl 0 0 0 1 0 0 4.900
13
N
Br 0 0 0 1 1 0 5.108
14
N
CH3 0 0 0 1 0 1 4.712
15
N
H 0 0 1 0 0 0 5.056
16
N
F 0 0 0 0 0 0 5.201
17
N
Cl 0 0 0 0 0 0 5.222
18
N
Br 0 0 0 0 1 0 5.443
19
N
CH3 0 0 0 0 0 1 5.097
Med Chem Res (2014) 23:1865–1877 1869
123
Table 1 continued
Compound no. R2 R1 I1 I2 I3 I4 I5 I6 pIC50
20
S N
NHO
O O
H
H 0 0 1 0 0 0 5.432
21 S N
NHO
O O
F
F 0 0 0 0 0 0 5.585
22 S N
NHO
O O
Cl
Cl 0 0 0 0 0 0 5.721
23 S N
NHO
O O
Br
Br 0 1 0 0 1 0 6.097
24 S N
NHO
O O
H3C
CH3 0 0 0 0 0 1 5.252
I1 = 1 if R2 is
S
N
NHO
O O
H
I2 = 1 if R2 isS N
NHO
O O
Br
I3 = 1 if R1 is H
I4 = 1 if R2 is
N
I5 = 1 if R1 is Br
I6 = 1 if R1 is CH3
1870 Med Chem Res (2014) 23:1865–1877
123
The models which are less significant compared with 4
and 5
pIC50 ¼0:056 �0:02ð Þst
þ 0:503 �0:4ð ÞI2 þ 1:933
n ¼ 24; R ¼ 0:861; R2 ¼ 0:741; R2A ¼ 0:717;
SE ¼ 0:174; Fð2�21Þ ¼ 30:095; Q ¼ 4:949:
ð6Þ
Equation 6 explains only 74.1 % variance in the COX-2
inhibitory activity. It shows that descriptor st and indicator
parameter I2 contribute positively toward the activity of
drugs. Based on variance percentage, it is not a very good
significant equation; therefore new model is required for
good explained variance.
pIC50 ¼þ 0:337 �0:188ð ÞI5
� 0:051 �0:199ð ÞI6
� 0:355 �0:198ð ÞI4
� 29:753 �12:528ð Þl� 0:259
n ¼ 24; R ¼ 0:878; R2 ¼ 0:771; R2A ¼ 0:723;
SE ¼ 0:172; Fð4�19Þ ¼ 16:003; Q ¼ 5:105:
ð7Þ
Equation 7 explains only 77.1 % variance in the COX-2
inhibitory activity. Based on variance percentage, it is not a
very good significant equation; therefore, new model is
required for good explained variance.
pIC50 ¼0:358 �0:407ð ÞI2
� 0:172 �0:364ð ÞI1
þ 1:241 �0:669ð ÞD
þ 0:00009 �0:00009ð ÞW þ 3:185
n ¼ 24; R ¼ 0:887; R2 ¼ 0:787; R2A ¼ 0:742;
SE ¼ 0:166; Fð4�19Þ ¼ 17:524; Q ¼ 5:343:
ð8Þ
Equation 8 explains only 78.7 % variance in the COX-2
inhibitory activity. It shows that descriptor D and
W contribute positively, and indicator parameter I1
contributes negatively toward the activity of drugs. In
this model, descriptor D and W show slight correlation
(Table 3) and it is not a very good significant equation,
therefore new model required for good explained variance.
pIC50 ¼þ 0:360 �0:406ð ÞI2
� 0:185 �0:364ð ÞI1
þ 1:234 �0:668ð ÞD
þ 0:108 �0:075ð ÞWA þ 2:802
n ¼ 24; R ¼ 0:888; R2 ¼ 0:788; R2A ¼ 0:744;
SE ¼ 0:166; Fð4�19Þ ¼ 17:693; Q ¼ 5:349:
ð9Þ
Equation 9 explains only 78.8 % variance in the COX-2
inhibitory activity. It shows that descriptor WA contributes
positively. In this model, descriptor D and WA show slight
correlation (Table 3)and it isnot averygoodsignificantequation,
therefore new model required for good explained variance.
pIC50 ¼þ 0:360 �0:405ð ÞI2
� 0:187 �0:364ð ÞI1
þ 1:235 �0:667ð ÞD
þ 0:051 �0:034ð Þ3vþ 2:941
n ¼ 24; R ¼ 0:888; R2 ¼ 0:789; R2A ¼ 0:744;
SE ¼ 0:166; Fð4�19Þ ¼ 17:718; Q ¼ 5:349:
ð10Þ
Equation 10 explains only 78.9 % variance in the COX-2
inhibitory activity. It shows that descriptor molecular
connectivity (3v) contributes positively. In this model,
descriptor D and 3v show slight correlation (Table 3) and it
is not a very good significant equation, therefore new
model required for good explained variance.
pIC50 ¼þ 0:359 �0:404ð ÞI2
� 0:201 �0:365ð ÞI1
þ 1:239 �0:663ð ÞD
þ 0:084 �0:056ð Þ5vþ 2:959
n ¼ 24; R ¼ 0:889; R2 ¼ 0:790; R2A ¼ 0:746;
SE ¼ 0:165; Fð4�19Þ ¼ 17:890; Q ¼ 5:388:
ð11Þ
For I1:
S
N
NHO
O O
H group at R2 position For I4: N group at R2 position
For I2:
S N
NHO
O O
Br group at R2 position For I5: -Br group at R1 position
For I3: H at R1 position For I6: -CH3 group at R1 position
Scheme 1 Different indicator
(I) parameters
Med Chem Res (2014) 23:1865–1877 1871
123
Equation 11 explains only 79.0 % variance in the COX-2
inhibitory activity. It shows that descriptor molecular
connectivity (5v) contributes positively. In this model,
descriptor D and 5v show slight correlation (Table 3) and It
is not a very good significant equation.
pIC50 ¼þ 0:359 �0:401ð ÞI2 � 0:220 �0:366ð ÞI1
þ 1:247 �0:657ð ÞD� 0:740 �0:494ð ÞJ þ 0:461
n ¼ 24; R ¼ 0:890; R2 ¼ 0:792; R2A ¼ 0:748;
SE ¼ 0:164; Fð4�19Þ ¼ 18:100; Q ¼ 5:427: ð12Þ
Table 2 Values of the calculated empirical and DFT-based parameters for 1–24
Compound no. st D W WA3v 5v J LUMO g S l ID
1 55.6 1.361 626 4.092 6.013 3.419 1.848 -0.07651 0.100 4.996 -0.177 4.146
2 53.9 1.423 736 4.304 6.424 3.581 1.845 -0.07936 0.100 5.005 -0.179 4.207
3 57.4 1.450 736 4.304 6.424 3.581 1.845 -0.08015 0.100 5.013 -0.180 4.207
4 58.3 1.633 736 4.304 6.424 3.581 1.845 -0.08024 0.100 5.025 -0.180 4.207
5 53.2 1.322 736 4.304 6.424 3.581 1.845 -0.07468 0.098 5.092 -0.175 4.207
6 62.6 1.471 2689 6.182 10.482 6.225 1.517 -0.09193 0.096 5.210 -0.188 4.881
7 60.0 1.547 3265 6.583 11.303 6.549 1.509 -0.09657 0.096 5.224 -0.193 4.759
8 64.5 1.575 3265 6.583 11.303 6.549 1.509 -0.09681 0.096 5.224 -0.193 4.759
9 65.6 1.800 3265 6.583 11.303 6.549 1.509 -0.09662 0.094 5.296 -0.191 4.759
10 58.7 1.412 3265 6.583 11.303 6.549 1.509 -0.28083 0.096 5.220 -0.185 4.759
11 53.9 1.423 736 4.304 6.424 3.581 1.845 -0.08925 0.096 5.220 -0.185 4.207
12 57.4 1.450 736 4.304 6.424 3.581 1.845 -0.08729 0.097 5.167 -0.184 4.207
13 58.3 1.633 736 4.304 6.424 3.581 1.845 -0.08757 0.097 5.167 -0.184 4.207
14 53.2 1.322 736 4.304 6.424 3.581 1.845 -0.08214 0.097 5.130 -0.180 4.207
15 55.6 1.361 626 4.092 6.013 3.419 1.845 -0.07885 0.098 5.125 -0.176 4.146
16 53.9 1.423 736 4.304 6.424 3.581 1.848 -0.08189 0.097 5.131 -0.180 4.207
17 57.4 1.450 736 4.304 6.424 3.581 1.845 -0.08231 0.097 5.138 -0.180 4.207
18 58.3 1.633 736 4.304 6.424 3.581 1.845 -0.08215 0.097 5.170 -0.179 4.207
19 53.2 1.322 736 4.304 6.424 3.581 1.845 -0.07730 0.097 5.154 -0.174 4.207
20 62.6 1.471 2689 6.182 10.482 6.225 1.517 -0.09651 0.095 5.285 -0.191 4.881
21 60.0 1.547 3265 6.583 11.303 6.549 1.509 -0.10113 0.094 5.341 -0.195 4.759
22 64.5 1.575 3265 6.583 11.303 6.549 1.509 -0.10136 0.091 5.508 -0.192 4.759
23 65.6 1.800 3265 6.583 11.303 6.549 1.509 -0.10190 0.088 5.665 -0.190 4.759
24 58.7 1.412 3265 6.583 11.303 6.549 1.509 -0.09447 0.092 5.447 -0.186 4.759
Table 3 Correlation matrix between empirical parameters and indicator parameter
pIC50 st D I1 I2 I3 W WA3v 5v J ID
pIC50 1.000
st 0.798 1.000
D 0.798 0.773 1.000
I1 -0.073 0.219 -0.033 1.000
I2 0.574 0.376 0.483 0.043 1.000
I3 -0.163 0.075 -0.257 0.466 -0.093 1.000
W 0.650 0.808 0.465 0.165 0.265 -0.028 1.000
WA 0.649 0.811 0.465 0.185 0.260 -0.027 0.999 1.0003v 0.648 0.812 0.463 0.188 0.260 -0.020 0.999 1.000 1.0005v 0.642 0.815 0.454 0.209 0.256 0.014 0.998 0.999 0.999 1.000
J -0.632 -0.817 -0.441 -0.239 -0.249 -0.063 -0.994 -0.996 -0.996 -0.999 1.000
ID 0.611 0.815 0.421 0.315 0.227 0.110 0.973 0.980 0.980 0.986 0.991 1.000
1872 Med Chem Res (2014) 23:1865–1877
123
Equation 12 explains only 79.2 % variance in the COX-2
inhibitory activity. It shows that descriptor Balaban
connectivity distance index (J) contributes negatively. It
is not a very good significant equation.
pIC50 ¼þ 0:365 �0:398ð ÞI2
� 0:274 �0:374ð ÞI1
þ 1:244 �0:649ð ÞD
þ 0:436 �0:283ð ÞIDþ 1:418
n ¼ 24; R ¼ 0:892; R2 ¼ 0:796; R2A ¼ 0:753;
SE ¼ 0:163; Fð4�19Þ ¼ 18:533; Q ¼ 5:472:
ð13Þ
Equation 13 explains only 79.6 % variance in the COX-2
inhibitory activity. It shows that descriptor ID contributes
positively. However, variance is low so that this model is
not a very significant equation.
pIC50 ¼ 0:061 �0:019ð Þst
þ 0:437 �0:377ð ÞI2
� 0:278 �0:4ð ÞI1
� 0:098 �0:21ð ÞI3 þ 1:671
n ¼ 24; R ¼ 0:894; R2 ¼ 0:799; R2A ¼ 0:757;
SE ¼ 0:161; Fð4�19Þ ¼ 18:883; Q ¼ 5:553:
ð14Þ
Equation 14 explains only 79.9 % variance in the COX-2
inhibitory activity. It is not a very good significant
equation.
pIC50 ¼þ 0:297 �0:151ð ÞI5
� 0:084 �0:155ð ÞI6
� 0:343 �0:159ð ÞI4
� 24:002 �7:173ð ÞLUMOþ 3:121
n ¼ 24; R ¼ 0:924; R2 ¼ 0:853; R2A ¼ 0:822;
SE ¼ 0:138; Fð4�19Þ ¼ 27:559; Q ¼ 6:696:
ð15Þ
Equation 15 explains only 85.3 % variance in the COX-2
inhibitory activity. This model equation shows that the
LUMO descriptors contribute negatively toward activity. It
is not a very good significant equation. Where n is the
number of compounds in the dataset, R is the correlation
coefficient, R2 is the coefficient of determination, R2A is the
adjusted coefficient of determination, SE is the standard
error of estimate, F is the variance ratio (Bikash et al.,
2003; Diudea, 2001), and Q is the quality of fit (Pogliani,
1994, 1996).
The above equations show that the coefficients of st, W,
WA, ID, D and molecular connectivity of third and fifth
order are positive this indicate that more bulkier, denser
group having more branching is favorable for the activity.
The negative coefficient of J in the above model is probably
due to its high collinearity with other parameters. The
negative sign of coefficients of indicator I1 and I3 is showing
that the group presented by I1 and I3 at R2 and R1 positions
(see footnote of Table 1), respectively, has a negative
influence on activity and should not be used for the future
drug modeling at R2 and R1 positions, respectively. The
positive sign of coefficients of indicator parameter I2 is
showing that the group presented by I2 at R2 group (see
footnote of Table 1) has a positive influence on activity and
should be used at R2 position for future drug modeling.
DFT-based calculated models reveal that the parameters
g, l, LUMO have negative coefficient, while one of them is
positive (S). Indicator parameter (I4, I6) have negative
coefficient, while I5 have positive. Thus, the decrease in
values of these parameters g, l, and LUMO and the
absence of the group indicated by I4, and I6 at R2 and R1
positions, respectively (see footnote of Table 1), are
favorable for the inhibitory activity. Similarly, the increase
in the value of S parameter and the presence of group
indicated by I5 at R1 position (see footnote of Table 1) are
favorable for the inhibitory activity. In all the above
equations, R, R2, and R2A are substantially high and SE is
fairly low, indicating that these models are statistically
significant in all the statistical terms.
Out of several QSAR models, models (4) and (5) show
best results. In order to confirm that the model with
excellent statistics has excellent predictive power too, we
Table 4 Correlation matrix between DFT-based parameters and indicator parameter
pIC50 l g el S I4 I5 I6
pIC50 1.000
l -0.647 1.000
g -0.752 0.695 1.000
el -0.734 0.969 0.846 1.000
S 0.760 -0.687 -0.998 -0.841 1.000
I4 -0.426 0.051 0.112 0.063 -0.101 1.000
I5 0.460 -0.062 -0.159 -0.117 0.190 0.046 1.000
I6 -0.385 0.327 -0.018 0.252 0.007 0.046 -0.263 1.000
Med Chem Res (2014) 23:1865–1877 1873
123
have evaluated quality factor Q. The predictive power as
determined by the Pogliani Q parameter for the model
expressed by Eqs. 4 and 5 (Q = 8.568) confirms that this
model has excellent statistics as well as excellent predictive
power.
According to Eqs. 4 and 5, the compounds having the
highest value of softness parameter, the lowest value of
hardness parameter, and the presence of bromine (–Br) at
R1 position show the highest inhibitory effect toward COX-
2 enzyme. Focusing on Tables 1 and 2, we find that 23 has
the maximum value of softness parameter, minimum value
of hardness parameter, and besides, Br is also present at R1
position, which all indicate that 23 inhibits COX-2 enzyme
more potentially, which is in good agreement with exper-
imental finding and SAR studies (Zhu et al., 2011).
Predicted and residual activity values for models (4) and
(5) are given in Table 5. Predicted values are the calculated
activities obtained from models (4) and (5), and the
residual values are the differences between the observed
biological activities and calculated activities The plot of
observed pIC50 verses predicted pIC50 for Eqs. 4 and 5 is
shown in graphical format (Figs. 1, 2), and the predicted R2
values were found to be fairly larger.
Further, we have calculated the molecular electrostatic
potential (MESP) and frontier orbitals for quantitative ana-
lysis of the activity of the present series using DFT methods by
employing 6-31?G* basis set in gas phase. The calculated
MESP at iso-value of 0.06 revealed that 22 and 23 have quite
less negative potential on sulfonamides group than that of the
rest of the compounds (Supporting information Fig. S1), and
also the frontier orbitals (HOMO and LUMO) of 22 and 23
were away from sulfonamides group and mostly concentrated
on benzene ring. However, in rest of the compounds, the
frontier orbitals are dispersed (Supporting information Fig.
S2). The calculated results of MESP and frontier orbitals
suggest that the activity of compound depends upon the less
negative potential on sulfonamide group and that the frontier
orbitals lie on one side of the compounds. The predicted
results also support the experimental observation.
Cross validation
The cross validation analysis was performed using leave-
one-out method (Cramer et al., 1988), in which oneTable 5 Comparison between observed and predicted activities and
their residual values
Observed
activity
Equation 4 Equation 5
Predicted
activity
Residual Predicted
activity
Residual
1 4.801 4.982 -0.181 4.971 -0.170
2 4.932 4.994 -0.062 4.979 -0.047
3 5.180 5.005 0.175 4.994 0.186
4 5.310 5.233 0.077 5.238 0.072
5 4.740 4.875 -0.135 4.868 -0.128
6 5.102 5.276 -0.174 5.284 -0.182
7 5.252 5.296 -0.044 5.299 -0.047
8 5.468 5.296 0.172 5.299 0.169
9 5.553 5.607 -0.054 5.626 -0.073
10 5.070 5.051 0.019 5.051 0.019
11 4.917 4.983 -0.066 4.990 -0.073
12 4.900 4.911 -0.011 4.922 -0.022
13 5.108 5.123 -0.015 5.097 0.011
14 4.712 4.621 0.091 4.628 0.084
15 5.056 5.159 -0.103 5.162 -0.106
16 5.201 5.168 0.033 5.170 0.031
17 5.222 5.177 0.045 5.177 0.045
18 5.443 5.433 0.010 5.451 -0.008
19 5.097 4.960 0.137 4.960 0.137
20 5.432 5.380 0.052 5.383 0.049
21 5.585 5.457 0.128 5.459 0.126
22 5.721 5.687 0.034 5.680 0.041
23 6.097 6.115 -0.018 6.099 -0.002
24 5.252 5.364 -0.112 5.364 -0.112
Fig. 1 A plot showing comparison between observed activity and
predicted activity based on model no. 4
Fig. 2 A plot showing comparison between observed activity and
predicted activity based on model no. 5
1874 Med Chem Res (2014) 23:1865–1877
123
compound is removed from the dataset and the activity is
correlated using the rest of the dataset. The cross validated
R2 in each case was found to be very close to the value of
R2 for the entire dataset, and hence, these models can be
termed as statistically significant. Cross validation provides
the values of PRESS, SSY, predictive square error (PSE),
R2CV and R2
A from which we can test the predictive power of
the proposed model. These statistical parameters can be
calculated from following equations:
PRESS ¼X
Xobs � Xcalð Þ2 ðiÞ
SSY ¼X
Xobs � Xmeanð Þ2 ðiiÞ
PSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPRESSp
=n ðiiiÞ
R2CV ¼ 1� PRESS
SSYðivÞ
R2A ¼ 1� r2
� � n� 1
n� p� 1
� �: ðvÞ
It is argued that PRESS is a good estimate of the real
predictive error of the model, and if it is smaller than SSY,
then the model predicts better than chance and can be
considered as statistically significant. Furthermore, the
ratio of PRESS/SSY can be used to calculate the approx-
imate confidence intervals of prediction of a new com-
pound. To be a reasonably perfect QSAR model, PRESS/
SSY should preferably be smaller than 0.4. Also, if PRESS
value is transformed in a dimensionless term by relating it
to the initial sum of squares, we obtain R2CV i.e., the
complement to the traces on of unexplained variance over
the total variance. The PRESS and R2CV have good prop-
erties. However, for practical purposes of the end users, the
use of square root of PRESS/n, which is called PSE, is
more directly related to the uncertainty of the predictions.
The PSE values also support our results. The calculated
cross validated parameters confirm the validity of the
models. All the requirements for an ideal model have been
fulfilled by model nos. 4 and 5; that is why, we have
considered these two models as the best models. R2A takes
into account the adjustment of R2. R2A is a measure of the
percentage of explained variation in the dependent vari-
able, which takes into account the relationship between the
number of cases and the number of independent variables
in the regression model, whereas R2 will always increase
when an independent variable is added. R2A will decrease if
the added variable does not reduce the unexplained vari-
able enough to offset the loss of decrease of freedom.
Predictive error of coefficient of correlation (PE)
The PE (Chaterjee et al., 2000) is yet another parameter
used to decide the predictive power of the proposed mod-
els. We have calculated PE values of all the proposed
models, and they are reported in Table 6. It is argued that if
the values of R \ PE, then such correlation is not signifi-
cant; however, if values of R [ PE by several folds (at least
three times), then the values are correlated. However, if
values of R [ 6PE, then mathematically the correlation is
unquestionably good. For all the models developed, the
condition R [ 6PE is satisfied, and hence, they can be said
to have a good predictive power.
Conclusions
The current study was performed to examine the applica-
bility of the empirical and DFT-based descriptors in QSAR
analysis for studying the biological activities of a series of
pyridine acyl sulfonamide derivatives which are known as
potent nonsteroidal anti-inflammatory and analgesic
agents. The obtained QSAR results based on empirical
parameters and DFT-based descriptors demonstrate that the
Table 6 Cross validated parameters and PE for the proposed models
Model no. n Parameter used PRESS SSY PRESS/SSY R2CV PSE R 1 - R2 PE 6PE
4 24 S ? I4 ? I5 ? I6 0.235 2.229 0.105 0.895 0.099 0.951 0.095 0.013 0.078
5 24 g ? I4 ? I5 ? I6 0.236 2.228 0.106 0.894 0.099 0.951 0.096 0.013 0.078
6 24 st ? I2 0.637 1.827 0.349 0.651 0.169 0.861 0.259 0.035 0.210
7 24 l ? I4 ? I5 ? I6 0.564 1.900 0.297 0.703 0.153 0.878 0.229 0.031 0.186
8 24 D ? I1 ? I2 ? W 0.525 1.939 0.271 0.729 0.147 0.887 0.213 0.028 0.168
9 24 D ? I1 ? I2 ? WA 0.521 1.943 0.268 0.732 0.147 0.888 0.212 0.029 0.174
10 24 D ? I1 ? I2 ? 3v 0.521 1.943 0.268 0.732 0.147 0.888 0.211 0.029 0.174
11 24 D ? I1 ? I2 ? 5v 0.517 1.947 0.266 0.734 0.146 0.889 0.210 0.028 0.168
12 24 D ? I1 ? I2 ? J 0.512 1.952 0.262 0.738 0.146 0.890 0.208 0.028 0.168
13 24 D ? I1 ? I2 ? ID 0.502 1.962 0.256 0.744 0.145 0.892 0.204 0.028 0.1.67
14 24 st ? I1 ? I2 ? I3 0.495 1.969 0.251 0.749 0.143 0.894 0.201 0.027 0.162
15 24 LUMO ? I4 ? I5 ? I6 0.362 2.102 0.172 0.828 0.123 0.924 0.147 0.020 0.120
Med Chem Res (2014) 23:1865–1877 1875
123
DFT-based descriptors are in very good agreement with
experimental findings. However, the calculated empirical
parameters also shed light on factors which influence the
activity. Among the DFT-based parameters, such as S, g, l,
and LUMO, the descriptors S and g were found to be the
most prominent to affect the activity.
Further, the calculated QSAR model, MESP, and fron-
tier orbital results revealed that the electron-donating group
at R1 position will activate the ring (e.g., –CH3), and it will
reduce the activity; however, the deactivating nature at R1
position with electron-donating group (e.g., –Br) will
enhance the activity of the present series of drug. Fur-
thermore, our calculated results also demonstrate that the
group of pyridine derivatives (11–14) at R2 position will
reduce the activity.
Acknowledgments The author A. K. Srivastava gratefully
acknowledges the grant under the Grant No. 40-70/2011 (SR) from
the University Grants Commission (UGC), New Delhi, India. Author
A. Singh wishes to acknowledge the award of Dr. D. S. Kothari’s
postdoctoral fellowship by the UGC, New Delhi, India.
References
ACD Lab Chem. Sketch version 12.0 Software (2009) Acd-Lab
software for calculating the referred physicochemical parameters.
Advanced chemistry development inc./Chemsketch ver. 12.0.
www.acdlabs.com/acdlabs-rss-feed.xml. Accessed 30 Aug 2013
Agarwal VK, Bano S, Khadikar PV (2003) Topological approach to
quantifying molecular lipophilicity of heterogenous set of
organic compounds. Bioorg Med Chem 11:4039–4047
Agarwal VK, Gupta M, Singh J, Khadikar PV (2005) A novel method
of estimation of lipophilicity using distance based topological
indices: dominating role of equalized electronegativity. Bioorg
Med Chem 13:2109–2120
Agarwal VK, Singh J, Louis B, Khadikar PV (2006) The topology of
molecule and its lipophilicity. Curr Comput Aided Des
2:369–403
Becke AD (1993) Density-functional thermochemistry. III. The role
of exact exchange. J Chem Phys 98:5648–5652
Bikash D, Shovanlal G, Subrata B, Soma S, Tarun J (2003) QSAR
study on some pyridoacridine ascididiamine analogues as
antitumor agents. Bioorg Med Chem 11:5493–5499
Chaterjee S, Hadi AS, Price B (2000) Regression analysis by
example, 3rd edn. Wiley, New York
Coats EA, Genther CS, Selassie CD, Strong CD, Hansch C (1985)
Quantitative structure–activity relationship of antifolate inhibi-
tion of bacteria cell cultures resistant and sensitive to metho-
trexate. J Med Chem 28:1910–1916
Cramer RD III, Bunce JD, Patterson DE, Frank IE (1988) Cross
validation, bootstrapping, and partial least squares compared
with multiple regression in conventional QSAR studies. Quant
Struct Act Relat 7:18–25
Diudea MV (2001) QSPR/QSAR studies for molecular descriptors.
Nova Science, Huntington
Gao H, Hansch C (1996) QSAR of p450 oxidation: on the value of
comparing kcat and Km with kcat/Km. Drug Metab Rev
28:513–526
Garg R, Kurup A, Mekapati SB, Hansch C (2003) Cyclooxygenase
(COX) inhibitors: a comparative QSAR study. Chem Rev
103:703–731
Hansch C, Leo A (1995) Exploring QSAR—fundamentally and
application in chemistry and biology. American Chemical
Society, Washington, DC
Hansch C, Hoekaman D, Gao H (1996) Comparative QSAR: toward a
deeper understanding of chemicobiological interaction. Chem
Rev 96:1045–1075
Hawkey C (1999) COX-2 inhibitors. J Lancet 353:307–314
Hu QN, Liang YZ, Fang KT (2003) The matrix expression,
topological index and atomic attribute of molecular topological
structure. J Data Sci 1:361–389
Khadikar PV, Singh S, Srivastava A (2002) Novel estimation of
lipophilic behavioural polychlorinated biphenys. Bioorg Med
Chem Lett 12:1125–1128
Khadikar PV, Singh S, Mandoloi D, Joshi S, Bajaj AV (2003a) QSAR
study on bioconcentration factor (BCF) of polyhaloginated
biphenlys using the PI index. Bioorg Med Chem 11:5045–5050
Khadikar PV, Mandoloi D, Bajaj AV, Joshi S (2003b) QSAR study on
solubility of alkanes in water and their partition coefficient in
different solvent system using PI index. Bioorg Med Chem Lett
13:419–422
Khadikar PV, Jaiswal M, Gupta M, Mandoloi D, Sisodia RS (2005a)
QSAR studies on 1,2 dithiol-3-thiones: modeling of lipophilicity,
quinone reductase specific activity and production of growth
hormone. Bioorg Med Chem Lett 15:1249–1255
Khadikar PV, Sharma V, Verma RG (2005b) Novel estimation of
lipophilicity using 13C NMR chemical shifts as molecular
descriptor. Bioorg Med Chem Lett 15:421–425
Lee CT, Wang WT, Parr RG (1988) Development of the Colle–
Salvetti correlation-energy formula into a functional of the
electron density. Phys Rev B 37:785–789
Lien EJ, Gao H (1995) QSAR analysis of skin permeability of various
drugs in man as compared in vivo and in vitro studies in rodents.
Pharm Res 12:583–587
Masferrer J, Zweifel B, Seibert K, Needleman P (1992) Endogenous
glucocorticoids regulate an inducible cyclooxygenase enzyme.
Proc Natl Acad USA 89:3917–3921
Muller K, Altmann R, Prinz H (2001) 2-Arylalkyl-substituted
anthracenones as inhibitors of 12-lipoxygenase enzymes. 1.
Structure–activity relationships of the terminal aryl ring. Eur J
Med Chem 36:569–575
Needleman P, Isakson PC (1997) The discovery and function of
COX-2. J Rheumatol 49:6–8
Plummer EL (1995) Successful application of the QSAR paradigm in
discovery programs. In: Hansch C, Fujita T (eds) Classical and
three-dimensional QSAR in agrochemistry. American Chemical
Society, Washington, DC
Pogliani L (1994) Structural property relationships of amine acids and
some peptides. Amino Acids 6:141–156
Pogliani L (1996) Modeling with special descriptors derived from a
medium size set of connectivity indices. J Phys Chem
100:18065–18077
Puzyn T, Suzuki N, Haranczyk M, Rak J (2008) Calculation of
quantum-mechanical descriptors for QSPR at the DFT level: Is it
necessary? J Chem Inf Model 48:1174–1180
Sarkar A, Mostafa G (2009) Towards the design of cyclooxygenase
(COX) inhibitors based on 40,5 di-substituted biphenyl acetic acid
molecules: a QSAR study with a new DFT based descriptor—
nucleus independent chemical shift. J Mol Model 15:1221–1228
Singh P, Bhardwaj A (2010) Mono-, di-, and triaryl substituted
tetrahydropyrans as cyclooxygenase-2 and tumor growth inhib-
itors. Synthesis and biological evaluation. J Med Chem
53:3707–3717
1876 Med Chem Res (2014) 23:1865–1877
123
Smith JB, Willis AL (1971) Aspirin selectively inhibits prostaglandin
production in human platelets. Nature (New Biol) 231:235–237
Srivastava AK, Archana, Jaiswal M (2008a) A series of p-arylthio
cinnamides as QSAR studies on antagonists of biochemical
ICAM-1/LFA-1 interaction in relation to anti inflammatory
activity. Oxid Commun 31:44–51
Srivastava AK, Srivastava A, Archana, Jaiswal M (2008b) QSAR of
substituted N-benzyl piperidines in the GBR series. J Ind Chem
Soc 85:842–848
Srivastava AK, Pathak, Archana VK, Jaiswal M (2008c) QSAR based
modeling of a new class of RNA polymerase inhibitors. J Ind
Chem Soc 85:627–631
Tetko IV, Gasteiger J, Todeschini R, Mauri A, Livingstone D, Ertl P,
Palyulin VA, Radchenko EV, Zefirov NS, Makarenko AS,
Tanchuk VY, Prokopenko VV (2005) Virtual computational
chemistry laboratory design and description. J Comput Aid Mol
Des 19:453–463
Turbomole v6.0 (2009) Turbomole GmbH, Karlsruhe. http://www.
turbomole.com. Accessed 1 Aug 2013
Vane JR (1971) Inhibition of prostaglandin synthesis as a mecha-
nism of action for aspirin-like drugs. Nature (New Biol)
231:232–235
Zarghi A, Zebardast T, Daraie B, Hedayati M (2009) Design and
synthesis of new 1,3-benzthiazinan-4-one derivatives as selec-
tive cyclooxygenase (COX-2) inhibitors. Bioorg Med Chem
17:5369–5373
Zarghi A, Ghodsi R, Daraie B, Hedayati M (2010) Design, synthesis
and biological evaluation of new 2,3-diarylquinoline derivatives
as selective cyclooxygenase-2 inhibitors. Bioorg Med Chem
18:1029–1033
Zhu HL, Lu X, Zhang H, Xi Li, Chen G, Li QS, Luo Y, Ruan BF,
Chen XW (2011) Design, synthesis and biological evaluation of
pyridine acyl sulfonamide derivatives as novel COX-2 inhibi-
tors. Bioorg Med Chem 19:6827–6832
Med Chem Res (2014) 23:1865–1877 1877
123