QSAR and Pharmacophore Modeling of Natural ... - Ingenta Connect

Send Orders for Reprints to [email protected] 350

Current Computer-Aided Drug Design, 2013, 9, 350-359

QSAR and Pharmacophore Antimalarial Prodiginines

Modeling

of

Natural

and

Synthetic

Baljinder Singh1,2, Ram A. Vishwakarma1,2,3 and Sandip B. Bharate*,2,3 1

Natural Products Chemistry Division, Indian Institute of Integrative Medicine (Council of Scientific and Industrial Research), Canal Road, Jammu - 180001, India 2

Academy of Scientific & Innovative Research (AcSIR), Anasandhan Bhawan, 2 Rafi Marg, New Delhi-110001, India

3

Medicinal Chemistry Division, Indian Institute of Integrative Medicine (Council of Scientific and Industrial Research), Canal Road, Jammu - 180001, India Abstract: Prodiginines are a family of linear and cyclic oligopyrrole red-pigmented compounds possessing antibacterial, anticancer and immunosuppressive activities and are produced by actinomycetes and other eubacteria. Recently, prodiginines have been reported to possess potent in vitro as well as in vivo antimalarial activity against chloroquine sensitive D6 and multi-drug resistant Dd2 strains of Plasmodium falciparum. In the present paper, a QSAR and pharmacophore modeling for a series of natural and synthetic prodiginines was performed to find out structural features which are crucial for antimalarial activity against these D6 and Dd2 Plasmodium strains. The study indicated that inertia moment 2 length, Kier Chi6 (path) index, kappa 3 index and Wiener topological index plays important role in antimalarial activity against D6 strain whereas descriptors inertia moment 2 length, ADME H-bond donors, VAMP polarization XX component and VAMP quadpole XZ component play important role in antimalarial activity against Dd2 strain. Furthermore, a five-point pharmacophore (ADHRR) model with one H-bond acceptor (A), one H-bond donor (D), one hydrophobic group (H) and two aromatic rings (R) as pharmacophore features was developed for D6 strain by PHASE module of Schrodinger suite. Similarly a six-point pharmacophore AADDRR was developed for Dd2 strain activity. All developed QSAR models showed good correlation coefficient (r2 > 0.7), higher F value (F >20) and excellent predictive power (Q2 > 0.6). Developed models will be highly useful for predicting antimalarial activity of new compounds and could help in designing better molecules with enhanced antimalarial activity. Furthermore, calculated ADME properties indicated drug-likeness of prodiginines.

Keywords: Antimalarial, malaria, pharmacophore model, prodiginines, QSAR. 1. INTRODUCTION Malaria is a life-threatening disease affecting at least 40% of the world’s population, and it remains a serious and complex health problem today [1]. Approximately 300-400 million people worldwide suffer from this infectious disease, around 3 million dying every year, mostly children younger than 5 years. This situation has become even more complex over the last decades with the increase in drug-resistance by parasite to available antimalarial drugs [2]. It is caused by protozoan parasites of the genus Plasmodium, and in humans four species namely P. falciparum, vivax, malariae and ovale are responsible for the spread of the disease. The most serious infections among these species are caused by Plasmodium falciparum. The emergence and spread of drugresistant malaria parasites further exacerbate this serious situation and has compounded the need for the development of new, cost effective antimalarials [3-5]. Since ancient times, natural products have provided great contribution in antimalarial drug discovery, the most notable examples being Cinchona alkaloids and artemisinin [6].

Prodiginines are a family of linear and cyclic oligopyrrole red-pigmented compounds with antibacterial [7], anticancer [8] and immunosuppressive activity [9], produced by actinomycetes and other eubacteria. Some prodiginines induce apoptotic effects, breaking genomic deoxyribonucleic acid (DNA) strands [10]. Prodiginines are known to exhibit potent in vitro activity against Plasmodium species, at much lower concentrations than seen with mammalian cells [11-15]. Recently Papireddy et al. discovered a series of 56 synthetic prodiginines along with 4 natural prodiginines possessing potent in vitro as well as in vivo antimalarial activity against chloroquine sensitive D6 and multi-drug resistant Dd2 strains of Plasmodium falciparum [16]. Natural prodiginines prodigiosin (1), undecylprodiginine (2), metacycloprodiginine (3) and streptorubin (4) showed potent antimalarial activity against D6 strain with IC50 values of 8, 7.7, 1.7 and 7.8 nM respectively (Fig. 1). In the present paper, we have performed quantitative structure-activity relationship (QSAR) and pharmacophore modeling on the data set consisting of 60 natural/synthetic prodiginines (Table 1) in order to identify physicochemical and structural features of these compounds responsible for antimalarial activity.

*Address correspondence to this author at the Medicinal Chemistry Division, CSIR-Indian Institute of Integrative Medicine, Canal Road, Jammu-180001, India; Tel. +91-191-2569000 (Extn. 345); Fax: +91-191-2569333; E-mail: [email protected] 1875-6697/13 $58.00+.00

© 2013 Bentham Science Publishers

QSAR and Pharmacophore Modeling

Current Computer-Aided Drug Design, 2013, Vol. 9, No. 3

OCH3

351

OCH3 4

N

N

HN

NH

HN

NH

Prodigiosin (1)

OCH3

N

10

Undecylprodiginine (2)

OCH3

N

HN

NH

Metacycloprodiginine (3)

HN

NH Streptorubin (4)

Fig. (1). Structures of natural prodiginines.

2. MATERIALS AND METHODS 2.1. Descriptor Based QSAR model Dataset: Dataset used in the present study is given in Table 1. For antimalarial activity against D6 strain, the data set of 60 prodiginines was separated into training and a test set of 45 and 15, respectively without any redundancy. The test set included compounds 3, 14, 17, 19, 21, 23, 24, 25, 29, 30, 33, 35, 41, 52 and 53. All remaining compounds were part of the training set. For multidrug resistant Dd2 strain, IC50 values for compounds 1, 2, 3, 4, 13, 18, 19 and 30 were not available. Thus, for Dd2 strain, remaining 52 compounds were divided into training and test set of 39 and 13 respectively in the way as described above for D6 strain. The test set included compounds 6, 8, 12, 15, 16, 22, 23, 35, 41, 50, 51, 53 and 58. All remaining compounds were part of the training set. The reported IC50 values were translated into pIC50 (pIC50 = -log IC50) values for QSAR purpose. Generation of QSAR models: The chemical structures were sketched using Chem-Draw Ultra software and were exported to TSAR 3.3 worksheet. In the TSAR worksheet, 3D-structures of all molecules were generated, partial charges were derived and their geometries were optimized [17]. For these optimized structures, then the molecular descriptors were calculated [18]. The descriptors with similar values were removed and the pair-wise correlation of remaining descriptors with antimalarial activity was carried out. The step-wise multiple regression analysis (MRA) was used to develop QSAR models. The F-to-enter and F-toleave values were kept as 4. For developed models, the noncollinearity of descriptors was checked using variance inflation factor (VIF = 1/1-r2) and tolerance (1/VIF) values [19, 20]. The regression was derived using descriptors which have high correlation with the antimalarial activity. Validation of QSAR models: The developed models were validated to check their predictive ability. Statistical quality of the generated QSAR equation was judged based

on the parameters like validated correlation coefficient (Q2) and predicted correlation coefficient (r2pred). For internal validation, the “leave-one-out” (Q2LOO) method was used [19]. In general, the value of Q2 > 0.5 is considered as reasonable predictive capability of the model; and the value of Q2 > 0.7 is indicative of the stable and predictive potential of the model [21]. The external validation was carried out by determining r2pred value [19, 22]. The QSAR model with number of descriptors < 6, r2CV and r2pred value > 0.6, F > 20, and r > 0.8 were chosen. 2.2. Pharmacophore Modeling Dataset: For D6 strain, out of 60 compounds, training and test set were chosen by randomization (50: 50) from Maestro project table. Test set included 30 compounds: 1, 4, 6, 8, 9, 12, 14, 15, 17, 18, 21, 22, 24, 26, 28, 29, 30, 32, 34, 35, 37, 38, 40, 42, 47, 48, 50, 57, 58 and 60. For multidrug resistant Dd2 strain, out of 52 compounds, training and test set were chosen by randomization (60: 40) from Maestro project table. Test set included 21 compounds: 6, 8, 12, 15, 16, 17, 20, 23, 26, 27, 28, 31, 32, 40, 46, 48, 50, 54, 57, 59 and 60. This distribution was done without keeping any redundancy with respect to structural, chemical, and activity distributions. The mean pIC50 of the training and test sets was found to be 7.26/7.39 and 7.53/7.20 for d2 and Dd2 strains respectively. These values clearly indicates the similar distribution of activity in both sets. Reported IC50 values were expressed as pIC50 values for pharmacophore modeling purpose. Pharmacophore-based QSAR model: A pharmacophore model was developed using PHASE module of Schrodinger molecular modeling package. The chemical structures of compounds were drawn in Chemdraw Ultra/Chem 3D and were ported to Maestro project table. These structures were then chosen for phase pharmacophore modeling. Structures were cleaned and conformers were generated using OPLS_2005 force field. The threshold of

352 Current Computer-Aided Drug Design, 2013, Vol. 9, No. 3

Table 1.

Singh et al.

Antimalarial Activity of Prodiginines Against D6 and Dd2 Strains of P. falciparum OCH3 R3 N R1

HN R2

IC50 (nM) Entry

R1

R2

R3

CH3

IC50 (nM) Entry

D6

Dd2

CH3

4250

4950

24

n-C11H23

H

4060

6150

25

7

n-C11H23

H

10470

15750

26

8

CH3

CH3

19410

13640

27

9

n-C11H23

H

2920

3810

28

CH3

CH3

n-C11H23

H

CH3

CH3

n-C3H7

H

2300

N.D.

32

n-C4H9

H

1780

1590

33

n-C6H13

H

375

450

34

n-C8H17

H

80

130

35

n-C16H33

H

300

400

36

n-C11H22NH2

H

1700

N.D.

37

H

(CH2)3 COOCH3

4500

N.D.

38

H

CH2CH(CH 3)2

460

230

39

H

n-C4H9

80

18

40

5

N H

6

N H

O

10 S

11 S

12 O

13

14

15

16

17

18

19

20

21

H N

H N

H N

H N

H N

H N

H N

H N

H N

>25000 >25000

5940

7770

>25000 >25000

29

30

31

R1

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

R2

R3

H

n-C10H21

H

n-C16H33

H

C6H5CH 2

83

86

H

4-OCH3C6H 4CH2

170

156

H

4-ClC6H4 CH2

65

81

H

4-BrC6H4CH2

90

108

H

2-NaphthylCH2

56

N.D.

CH3

CH3

8900

8130

n-C6H13

n-C3H7

4.5

4.0

n-C8H17

n-C3H7

2.9

2.7

1.3

1.3

n-C3H7

D6

Dd2

8.0

10

>25000 >25000

n-C6H13

n-C6H13

1.7

1.1

n-C7H15

n-C6H13

2.1

1.2

n-C6H13

n-C8H17

4.9

2.0

n-C7H15

n-C8H17

6.2

2.9

n-C8H17

n-C8H17

92

129

5.3

3.5



353

(Table 1) contd…..

IC50 (nM) Entry

22

23

43

44

45

46

47

48

49

50

51

R1 H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

R2

R3

H

IC50 (nM) Entry

D6

Dd2

n-C6H13

28

7

41

H

n-C8H17

4.6

1.8

42

n-C6H13

4-ClC6H4 CH2

2.0

1.8

52

n-C7H15

4-ClC6H4 CH2

2.8

2.2

53

n-C8H17

4-ClC6H4 CH2

16.0

12.0

54

3.9

2.9

55

4-ClC6H4 CH2

n-C6H13

4-FC6H4 CH2

0.9

0.9

56

n-C8H17

4-FC6H4 CH2

1.3

1.2

57

n-C6H13

4-BrC6H4CH2

2.9

2.8

58

n-C8H17

4-BrC6H4CH2

4.0

2.9

59

4-ClC6H4 CH2

4-ClC6H4 CH2

6.1

4.8

60

R1 H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

H N

R2

R3

C2H5

D6

Dd2

4-ClC6H4 CH2

6.3

6.2

n-C3H7

4-ClC6H4 CH2

3.0

2.6

4-FC6H4 CH2

4-FC6H4 CH2

5.6

5.7

4-BrC6H4CH2

4-BrC6H4CH2

14.0

11.0

4-FC6H4 CH2

4-ClC6H4 CH2

6.1

6.1

4-BrC6H4CH2

4-ClC6H4 CH2

8.3

7.7

4-BrC6H4CH2

4-FC6H4 CH2

5.7

5.1

2,4-Cl2C6H 3CH 2

2,4-Cl2C6H 3CH 2

12.6

11.0

2,6-F2C6H 3CH 2

2,6-F2C6H 3CH 2

14.7

18.3

3-FC6H4 CH2

3-FC6H4 CH2

5.1

6.7

2-ClC6H4 CH2

2-ClC6H4 CH2

3.6

4.9

N.D.: not determined.

7.5 (for D6 activity) 7.0 (for Dd2 activity) was set for discrimination between inactives and actives. Pharmacophore sites were created using SMARTS patterns - point, vector and group. By choosing appropriate number of maximum and minimum sites and number of actives should match hypothesis, common pharmacophore hypotheses (CPH) containing different combinations of Hbond acceptor (A), H-bond donor (D), hydrophobic (H), negatively charged group (N), positively charged group (P) and aromatic (R) were generated. All the steps used for pharmacophore model development were similar to our earlier work [22]. The atom-based QSAR model was generated for common pharmacophore hypotheses with one to seven PLS factors. Developed models were validated by three external test set predictions, namely correlation coefficient Q2, Pearson r and RMSE [23, 24]. 2.3. Prediction of ADME Properties ADME properties of the selected potent compounds were determined using QikProp [25] module of Schrodinger software. Calculated properties and their acceptable limits are as follows: predicted octanol/water partition co-efficient QPlogPo/ w (acceptable range: 2.0 to 6.5), aqueous solubility QPlogS (acceptable range: 6.5 to 0.5), HERG K+ channel blockage

QPlogHERG (concern below -5), apparent Caco-2 cells permeability QPPCaco (500 great), apparent MDCK permeability QPPMDCK (500 great), % human oral absorption ( 0.6) comprises molecular volume, molecular mass, inertia moment 2 size,


Singh et al.

inertia moment 2 length, dipole moment Z component, lipole Z component, Kier Chi6 (path) index, ADME H-bond donors, ADME rotatable bonds, VAMP Quadpole YY, VAMP octupole YYY and VAMP octupole XYZ. Using these collinear descriptors, the QSAR model was generated in TSAR software. A QSAR model with r2 = 0.652 (F = 25.62, s = 0.846) involving 4 descriptors viz. inertia moment 2 length, dipole moment Z component, Kier Chi6 (path) index, and Kappa3 index was generated. The removal of one descriptor “dipole moment Z component” led to improved correlation coefficient (r2 = 0.743), F value (F = 28.91) and s value (s = 0.736). Further, inclusion of Wiener topological index improved the correlation coefficient (r2 = 0.778), F value (F = 35.11) and s value (s = 0.684). The final QSAR model developed for antimalarial activity against D6 strain using MRA method is mentioned below. Y = (0.343 X1) + (2.373 X2) + (1.014 X3) - (0.003 X4) - 4.438

(1)

where X1 is the inertia moment 2 length; X2 is Kier Chi6 (path) index; X3 is the kappa3 index and X4 is the Wiener topological index. Statistical parameters for equation 1 are: N = 45; s = 0.684; r = 0.882; r2 = 0.778; F = 35.11; r2cv = 0.616; F probability = 3.6 10-13; residual sum of squares = 18.709; predictive sum of squares, PRESS = 32.419; predictive r2 (r2pred) = 0.608.

The descriptors included in the final model are listed in Table S1 (Supporting information). The Pearson correlation matrix of included descriptors in Equation 1 is provided in Table S3. The experimental and predicted activities of training set using MRA method are listed in Table S2. The predictive ability of the developed model was analyzed by test set (N = 15). Two test set compounds 23 and 24 were identified as outliers and the pIC50 (predicted) values of rest of 13 compounds were predicted using equation 1. Compounds 23 and 24 were removed in order to improve the statistical parameters (r2CV) of the test set. The plot of experimental and predicted activity for training and test set compounds is shown in Fig. (2a). As depicted in Fig. (2c), each descriptor contributes to the different extent to the variation in pIC50 values. Three descriptors viz. inertia moment 2 length, Kier Chi6 (path) index and Kappa 3 index component are contributing positively to the activity whereas Wiener topological index YYX are contributing negatively to the activity. Amongst all descriptors, the Kier Chi6 (path) index contributes more significantly. QSAR model for Dd2 strain: The highly collinear (with Dd2 strain activity) descriptors (r > 0.6) include molecular mass, inertia moment 1 size, inertia moment 2 length, lipole X component, Kier ChiV0 (atoms) index, several Kier Chi and ChiV bond/cluster/path indexes, 5-membered aliphatic rings, 5-membered rings, ADME weight, ADME H-bond

(a)

(b)

11

Pred dictedpIC50(dd2strain)

10

11

y=0.7783x+1.6213 R²=0.7783

Training(N=45)

Predicte ed pIC50 (d6strain)

Test(N=13) 9

Linear(Training(N=45))

8 7 6 5

Test(N=11) 9

Linear(Training(N=39))

8 7 6 5 4

4 4

5

6

7

8

9

10

4

5

ExperimentalpIC50(dd2strain)

15 10 5 0 5

InertiaMoment KierChi6(path) Kappa3index 2Length index (wholemol)

10

6

7

8

9

10

ExperimentalpIC50(dd2strain)

Wiener topological index

Coefficient*A AverageofDescriptorValue

Coefficient*A AverageofDescriptorValue

y=0.7314x+1.9985 R²=0.7314

Training(N=39)

10

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 0.5

InertiaMomentKierChi6(path) Kappa3index 2Length index (wholemol)

Wiener topological index

15

(c)

(d)

Fig. (2). Correlation graphs of experimental versus predicted activity and descriptor contribution charts are shown. (a). Correlation graph of pIC50 (experimental) versus pIC50 (predicted) for D6 strain using equation 1; (b). Correlation graph of pIC50 (experimental) versus pIC50 (predicted) for Dd2 strain using equation 2; (c). Descriptor contribution chart for QSAR equation 1; (d). Descriptor contribution chart for QSAR equation 2.


Table 2.


355

Summary of Phase 3D-QSAR Statistical Results D6 strain

Pharmacophore

Dd2 Strain

Pharm_1

Pharm_2

Pharm_3

ADHRR.8

AADHR.35

AADDRR.8

r2

0.9928

0.9944

0.9924

SD

0.1379

0.1220

0.1442

530

678

523

Hypothesis a

PLS Statistics of QSAR Model

F

-23

3 x 10-24

P

2 x 10

Stability

0.3067

0.3204

0.6162

6

6

6

0.6547

0.4512

0.8564

Number of PLS factors

1 x 10

-24

b

External Test Set Prediction Q2 (r2 pred ) rp (Pearson r)

0.8115

0.6738

0.9263

RMSE

0.7355

0.9273

0.5119

a 2

r , a coefficient of determination; SD, the standard deviation of regression; F, the ratio of the model variance to the observed activity variance; P, significance level of F when treated as a ratio of Chi-squared distributions; stability, stability of the model predictions to changes in the training set composition Q2, directly analogous to r2 but based on the test set predictions; rp, Pearson r value for the correlation between the predicted and observed activity for the test set; RMSE, the RMS error in the test set predictions.

b

acceptors, ADME H-bond donors, VAMP nuclear energy, VAMP polarization XX component and VAMP quadpole XZ component. The QSAR model was then generated as described above. A QSAR model with r2 = 0.814 (F = 23.29, s = 0.669, r2cv = 0.600) involving 6 descriptors viz. Inertia moment 2 length, ADME H-bond donors, VAMP polarization XX, VAMP dipole X component, VAMP dipole Z component, VAMP Quadpole XZ was obtained. Removal of descriptors with similar values viz. VAMP dipole X component and VAMP dipole Z component resulted in improvement of r2cv (0.631). The final QSAR model developed for antimalarial activity against Dd2 strain using MRA method is mentioned below. Y = (0.380 X1) + (1.384 X2) + (0.047 X3) - (0.086 X4) - 1.846

internal validation method and predicted correlation coefficient (r2pred) by external validation method. The internal validation of data set yielded r2cv = 0.616 (for equation 1), 0.631 (for equation 2) indicating the stable and predictive potential of the model (r2cv > 0.6). The external validation correlation coefficient r2pred > 0.6 indicates excellent predictive ability of both models (equation 1 and 2). The VIF values are not higher than 5 (or tolerance values more than 0.20) (Tables S7 and S8) which indicates that there do not exists multi-collinearity among descriptors. Thus the present equation could be used to design newer prodiginines as potent antimalarial agents against chloroquine sensitive D6 and multi-drug resistant Dd2 strains of Plasmodium falciparum.

(2)

where X1 is the inertia moment 2 length; X2 is ADME Hbond donors; X3 is the VAMP polarization XX and X4 is the VAMP Quadpole XZ. Statistical parameters for equation 2 are: N = 39; s = 0.779; r = 0.855; r2 = 0.731; F = 23.14; r2cv = 0.631; F probability = 9.5 10-10; residual sum of squares = 20.641; predictive sum of squares, PRESS = 28.380; predictive r2 (r2pred) = 0.633. Descriptors included in the final model are listed in Table S4. The Pearson correlation matrix of included descriptors in equation 2 is provided in Table S6. The experimental and predicted activity of training set compounds is listed in Table S5. The predictive ability of the model was checked by test set (N = 15). The plot of experimental and predicted activity for both sets is shown in Fig. (2b). As depicted in Fig. (2d), the descriptor ADME H-bond donor contributes more significantly to pIC50 variation. Validation of QSAR models: The statistical quality of the generated QSAR equation was judged based on the parameters like validated correlation coefficient (r2cv) by

3.2. Pharmacophore Modeling The pharmacophore model for a series of natural and synthetic prodiginines possessing antimalarial activity against both strains D6 and Dd2 was developed. The common pharmacophore hypothesis (CPH) was idenfied using training set compounds. The 3D-QSAR models were then generated using these CPH by Partial Least Square (PLS) method. The statistical results for best QSAR models are summarized in Table 2. In case of antimalarial activity against D6 strain, the best-fitted model Pharm_1 (ADHRR: r2 = 0.9928, F = 530, Q2 = 0.6547) and for Dd2 strain, model Pharm_3 (AADDRR: r2 = 0.9924, F = 530, Q2 = 0.8564) were used for activity prediction of training and test sets. For D6 antimalarial activity, total 15 pharmacophores were obtained, amongst which best two are listed in Table 2. The best pharmacophore model (Pharm_1) was selected based on their statistical parameters, primarily Q2 and RMSE value. Best pharmacophore hypotheses for D6 strain (Pharm_1: ADHRR) and Dd2 strain (Pharm_3: AADDRR) are shown in Fig. (3) with their inter-feature constraints in Å.


Singh et al.

Fig. (3). The best pharmacophores: (a). ADHRR (for D6 strain) and (b). AADDRR (for Dd2 strain) are shown with their inter-feature distance constraints in angstrom.

Fig. (4a-c) present the D6 strain Pharm_1 CPH aligned with one of the most active compound (36, IC50 = 2.1 nM), a moderately active compound (55, IC50 = 8.3 nM) and one of the least active compounds (10, IC50 = >25000 nM) in the training set, respectively. The vectors showed hydrogen bond acceptor site (A) and hydrogen bond donor site (D). Ring demonstrates the aromatic ring (R) and hydrophobic group (H) pharmacophore. The empty gray cubes represents the regions characteristics of the ligand structure that have positive effect on the calculated activity; whereas the gray cubes with black circle inside represent unfavourable regions. Similarly for Dd2 activity, amongst 10 CPH obtained, best hypothesis Pharm_3 selected based on Q2 and RMSE is listed in Table 2. One of the most active compound (35, IC50 = 1.1 nM), a moderately active compound (45, IC50 = 12 nM) and one of the least active compounds (10, IC50 = >25000 nM) from training set have been aligned to developed pharmacophore Pharm_3 (Fig. 4d-f). The PHASE QSAR models do not use internal crossvalidation techniques but rather use distinct training and test sets. The predictive ability of each analysis was determined from a test of 30 (in case of D6) and 21 (in case of Dd2) ligands that were not included in the training set. The test set molecules were aligned and their activities were predicted by each PLS analysis. The developed D6 model (Pharm_1) showed r2 = 0. 9928, Q2 = 0.6547 which was significant in whole series of ligands. Similarly, in case of Dd2 model (Pharm_3), excellent correlation coefficients (r2 = 0. 9924, Q2 = 0.8564) were obtained. The value of squared correlation coefficient Q2 > 0.6 indicates good predictability of the model. Moreover, the predicted and experimental activities have good correlation (rp > 0.8) [23, 24]. A plot of experimental versus predicted activity using Pharm_1 and Pharm_3 models is shown in Fig. (5a, b), which shows excellent correlation coefficient. The experimental and

PHASE predicted pIC50 values for both training and test set are provided in Tables S9 and S10 of supporting information. 3.3. ADME Properties Lipinski’s rule of 5 is a rule of thumb to evaluate drug likeness, or determine if a chemical compound with a certain pharmacological or biological activity has properties that would make it a likely orally active drug in humans. The rule describes molecular properties important for a drug’s pharmacokinetics in the human body, including its ADME. Top 8 compounds were evaluated for their drug-like behaviour through analysis of pharmacokinetic parameters required for absorption, distribution, metabolism and excretion (ADME) by use of QikProp. The partition coefficient (QPlogPo/w) and water solubility (QPlogS) critical for estimation of absorption and distribution of drugs within the body ranged between 3.34 to 4.85 and 3.227 to 0.315. Cell permeability (QPPCaco), a key factor governing drug metabolism and its access to biological membranes, ranged from 4830 to 6344, QPPMDCK ranges from 3024 to 7106. Overall, the percentage human oral absorption for all 8 compounds was predicted to be 100% and none of them are violating Lipinki Rule of 5. These ADME parameters are within the acceptable range defined for human use (see Table 3 footnote), indicating their drug-likeness. 4. CONCLUSION In summary, in this study we have derived a validated QSAR model to predict antimalarial activity of prodiginines based on their physicochemical parameters. Additionally, the statistically-significant pharmacophore based 3D-QSAR model has also been established. These models will be



357

Fig. (4). Pharmacophore overlay of highly active, medium active and least active compounds. (a-c). CPH Pharm_1 (ADHRR, for prodiginines against D6 strain) aligned to one of the most active 36, medium active 55 and least active compound 10 respectively; (d-f). CPH Pharm_3 (AADDRR, for prodiginines against Dd2 strain) aligned to one of the most active 35, medium active 45 and least active compound 10 respectively. Vectors are shown for hydrogen-bond acceptors (A) and hydrogen-bond donors (d). Rings are shown for aromatic groups (R), and balls for hydrophobic functions (H). In the background, the gray cubes indicates favorable regions and gray cubes with small black circle indicates unfavorable regions.

highly useful for both virtual screening and designing of new potent antimalarial agents.

CONFLICT OF INTEREST The authors confirm that this article content has no conflict of interest.

(b)

(a)

9

Testset(N=30)

8

Linear(Trainingset(N=30))

y=0.9928x+0.0523 R²=0.9929

Trainingset(N=31) Testset(N=21)

9 PredicctedpIC50 (dd2strain)

Pred dictedpIC50 (d6strain)

10

Trainingset(N=30)

10

7 6

y=0.9924x+0.058 R²=0.9924

Linear(Trainingset(N=31))

8 7 6 5

5

4

4 4

5

6

7

8

ExperimentalpIC50 (d6strain)

9

10

4

5

6 7 8 ExperimentalpIC50 (dd2strain)

9

Fig. (5). Correlation graphs of experimental antimalarial activity versus PHASE predicted activity. (a) D6 strain; (b) Dd2 strain.

10


Table 3.

Singh et al.

ADME Properties of Selected Potent Moleculesa Antimalarial Activity (pIC50)

QPlogP o/wb

QPlogSc

QPlog HERGd

QPP Cacoe

QPP MDCKf

% Human Oral Absorptiong

Lipinski Violationsh

8.57

3.336

-0.083

0.226

5337.747

3023.694

100

0

8.77

8.96

3.624

0.227

1.172

6344.150

3644.355

100

0

36

8.68

8.92

3.814

0.315

1.365

6274.841

3601.340

100

0

42

8.52

8.59

3.981

-2.998

-3.772

5383.657

7106.211

100

0

43

8.70

8.75

4.421

-2.529

-2.938

5013.342

6557.471

100

0

44

8.55

8.66

4.858

-3.227

-3.199

4960.860

6497.087

100

0

47

9.05

9.05

4.218

-2.245

-2.891

4970.574

4732.185

100

0

49

8.54

8.55

4.477

-2.678

-3.087

4829.690

6772.098

100

0

Entry

D6

Dd2

33

8.54

35

a

ADME properties were determined using QikProp module of Schrodinger 9.1 software. Predicted octanol/water partition co-efficient log p (acceptable range: 2.0 to 6.5). Predicted aqueous solubility; S in mol/L (acceptable range: 6.5 to 0.5). d HERG K+ Channel Blockage: log IC50 (concern below -5). e Apparent Caco-2 Permeability (nm/sec) (500 great). f Apparent MDCK Permeability (nm/sec) (500 great). g Percentage human oral absorption in GI (+-20%) (