A Review of Molecular Modelling Studies of Dihydrofolate Reductase

0 downloads 0 Views 2MB Size Report
modelling used for design of DHFR inhibitors against opportunistic microorganisms are reviewed, ... typical DHFR inhibitor, the 2-amino-4-oxo system in folate is.
712

Current Pharmaceutical Design, 2011, 17, 712-751

A Review of Molecular Modelling Studies of Dihydrofolate Reductase Inhibitors Against Opportunistic Microorganisms and Comprehensive Evaluation of New Models Nilesh R. Tawari, Seema Bag and Mariam S. Degani* Department of Pharmaceutical Sciences and Technology, Institute of Chemical Technology, Matunga (E), Mumbai 400019, India Abstract: Dihydrofolate reductase (DHFR) has been used as a target for antimicrobial drug discovery against a variety of pathogenic microorganisms, including opportunistic microorganisms; Pneumocystis carinii (pc), Toxoplasma gondii (tg) and Mycobacterium avium complex (ma). In this regard, several DHFR inhibitors are reported against pc and tg and ma. However, selectivity issue of these inhibitors over human DHFR often preclude their development and clinical use. In the first part of this work, various computational approaches including available crystallographic structures, binding affinity prediction, pharmacophore mapping, QSAR, homology modelling used for design of DHFR inhibitors against opportunistic microorganisms are reviewed, to understand specific interactions required for inhibition of microbial DHFR. Secondly, comprehensive molecular modelling techniques were used, to establish structure-chemical-feature-based pharmacophore models for pcDHFR, tgDHFR and mammalian DHFR. The results show that, the information encoded by ligand based approaches like pharmacophore mapping and 3D-QSAR methods are in well agreement with the information coded in the receptor structure. A combination of ligand and structure based approaches provides understanding of ligand-receptor interactions. The study indicated that the value of small alkyl moieties at position 5 of the bicyclic nitrogen containing nucleus along with a bulky group attached at the C-6 via suitable linker could optimize activity, with regard to both potency and selectivity.

Keywords: Dihydrofolate Reductase, QSAR, Pharmacophore mapping, CoMFA, CoMSIA, PHASE. 1. INTRODUCTION Dihydrofolate reductase (DHFR, EC 1.5.1.3), which is crucial for cell growth and division, catalyzes the reduction of 7,8dihydrofolate (DHF) by stereospecific hydride transfer from the dihydro-nicotinamide adenine dinucleotide phosphate (NADPH) cofactor, to the C6 atom of the pterin ring with concomitant protonation at N5 of DHF to form 5,6,7,8-tetrahydrofolate (THF). THF is a cofactor involved in one carbon donation for purine and pyrimidine de-novo synthesis, both in mammals as well as microorganisms [1]. The mechanism is depicted in Fig. 1. Inhibitors of DHFR have been in clinical use for over fifty years, for anticancer, antibacterial and antiprotozoal treatments. In a typical DHFR inhibitor, the 2-amino-4-oxo system in folate is changed to 2,4-diamino system, which increases the basicity of the nitrogen of the 1,3-diazine (Fig. 1). This modification results in aberrant protonation at the 1-position rather than the 8-position and as a result, this intermediate cannot accept a hydride ion and collapse to product. Consequently, the end product is an intermediate bound through a salt bridge to the conserved acidic residue (Glu or Asp) of the enzyme. This is the basis for majority of reported DHFR inhibitors. Thus the pursuit for species specificity is complex, particularly in case of highly conserved enzyme with high functional and structural enzyme. Opportunistic organisms such as Pneumocystis carinii (pc), Toxoplasma gondii (tg), and Mycobacterium avium complex (ma) often cause life threatening infections in immunocompromised hosts such as HIV patients. Currently, different DHFR inhibitors (Fig. 2) including trimethoprim (TMP), trimetrexate (TMQ) and piritrexim (PTX) are used for treatment of infections caused by pc and tg [2]. TMP, a well known antibacterial antifolate developed by Hitchings and Roth [3] is a selective but weakly potent agent, henceoften used in combination with sulfonamides e.g. *Address correspondence to this author at the Department of Pharmaceutical Sciences and Technology, Institute of Chemical Technology, Matunga (E), Mumbai 400019, India; Tel: +9122-33612213; Fax: +9122-24145414; E-mail; [email protected] 1381-6128/11 $58.00+.00

sulphamethoxazole, to enhance potency. TMQ and PTX are potent and non-selective DHFR inhibitors and are co-administered with leucovorin for host rescue. However, severe side effects attributed to the sensitivity of some patients to sulfa drugs often results in discontinuation of treatment. Furthermore, the DHFR inhibitor/leucovorin combination therapy also has several drawbacks including, the high cost of leucovorin and the inconsistent effect of leucovorin under all clinical conditions. Thus, there is a considerable interest to incorporate selectivity of TMP and potency of TMX/PTX into a single agent that can be used alone to treat these infections. DHFR inhibitors are yet to be clinically applied for the infections caused by ma [1]. In this light, several researchers including Gangjee et al., [4-18] Anderson et al.,[19] Rosowsky et al.[20-40] continue to work in search of selective, potent inhibitors of DHFR from opportunistic microorganisms. In many cases the issue of potency has been overcome, however selectivity over human DHFR remains a major stumbling block in these efforts. The structural requirements for potential DHFR inhibitors are summarized in recent review articles [41-44]. The availability of high resolution crystal structures of pcDHFR and human DHFR has provided impetus in the use of rational drug design techniques for the development of highly potent and specific DHFR inhibitors [45,46]. However, crystal structures of tgDHFR and maDHFR are not solved yet. In these cases, comparative modeling techniques have been used to enable use of rational drug design [47,48]. Also, availability of a plethora of information about the active and inactive ligands has provided a solid platform for ligand based drug design in this area [49,51]. Recently, we have described a combined method based on structure based and ligand based approaches to understand maDHFR inhibitory potential [52]. In this review, various rational drug design approaches for the development of specific anti-opportunistic DHFR inhibitors are surveyed and presented, to understand specific interactions required for inhibition of microbial DHFR. Secondly, in an attempt to understand the critical features required for specific DHFR inhibitors, we have employed combined structure based and ligand based approaches on a dataset of > 200 © 2011 Bentham Science Publishers Ltd.

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

713

Fig. (1). Mechanism of reaction catalyzed by DHFR.

NH2 N H2N

NH2 OCH3

N

OCH3 OCH3

Trimethoprim (TMP)

O

NH2

N H2N

N N

OCH3 OCH3

H2N

N H

OCH3

N

O Piritrexim (PTX)

Trimetrexate (TMQ)

Fig. (2). Non-classical inhibitors of DHFR used for treatment of opportunistic microorganisms.

molecules reported as DHFR inhibitors for opportunistic microorganisms to develop common structure-chemical-feature based pharmacophore models. In addition, structural requirements for mammalian DHFR binding were also studied in similar fashion to address the issue of selectivity. This study highlights several key parameters responsible for potency and selectivity of DHFR inhibitors, which could be explored for design of novel inhibitors, by incorporating structural features in the designed compounds in accordance with this study. 2. REVIEW OF MOLECULAR MODELING APPROACHES EMPLOYED FOR DESIGNING SELECTIVE DHFR INHIBITORS 2.1. Available Crystal Structure Information Co-crystallization and elucidation of the structure of the enzyme-inhibitor complex is often the first step in structure based drug design [53-55] X-ray crystallography [56-58] or NMR analysis [59-60] methods are used to experimentally determine the structure of target protein. Experimentally determined crystal structures

of pcDHFR, with or without inhibitor, are available in protein data bank (Table 1, Fig. 3) [61]. Furthermore, availability of crystal structure of human DHFR helped in addressing the selectivity issue. This section of the review discusses available crystal structures of pcDHFR. Edman et al.[62] have described cloning of pcDHFR, which contains 206 amino acids and shares 34% identity with human DHFR. Delves et al. have described high level of expression of pcDHFR gene using the promoter T7 system in Escherichia coli [63]. Further, Stammers et al. described crystallization conditions together with X-ray characterization of the grown crystals in ternary and binary complex from [64]. Champness et al. described, 1.86 Å resolution crystal structure of ternary complex of pc DHFR with TMP and NADP, similarly with PTX and NADP along with a binary complex holoenzyme at 2.5 Å resolution. These structures allowed direct comparison of interaction of TMP (less potent but selective) and PTX (potent but non-selective) inhibitors with the active site of fungal enzyme, in which the volume of the active site appeared to determine the strength of inhibitor binding [65].

714 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 1.

Degani et al.

List of Available Crystal Structures of pcDHFR in Protein Data Bank

Sr. No.

PDB ID

Resolution

Ligand

Reference

1.

1DYR

1.86

TMP, NDP

44

2.

1E26

2.00

GPB, NAP

46

3.

1DAJ

2.30

MOT, NDP

65

4.

1CD2

2.20

FOL, NAP

66

5.

2CD2

1.90

FOL, NAP

66

6.

3CD2

2.50

MTX, NAP

66

7.

4CD2

2.00

FA

66

8.

1VJ3

2.10

TAB, NDP

67

9.

1KLK

2.30

PMD (PT6 53), NDP

68

10.

1S3Y

2.25

TQT, NAP

70

11.

1LY3

1.90

COG, NAP

69

12.

1LY4

2.10

COQ, NAP

69

13.

2FZH

2.1

DH1, NAP

71

14.

2FZI

1.60

DH3,NAP

71

15.

2FZJ

2.00

DH3,NAP

71

O

O

OH

OH

N H

H2N

GBP

O

O

O

OH O

N

N

N H

N

FOL

NH2 N N

N

O

N

MTX

N

N

NH2 N HN

H 2N

O N

HN

Cl

NH2 N

H2N

N

N

H 2N

MOT

OH O

OH

N H NH2

O

OH

N

N N

O N H

O

N

NH2

O

N N

H2N

OH

N H

O

N H2N

O

OH

N H

NH2

O

N

H 2N

TAB

N

N N H

PMD

TTQ

O O

NH2 N

N H2N

N

O

COG

N

N H2N

N

O

NH2

NH2 N

O H 2N

N COQ

NH2

OH

O N

n N

O

H 2N DH1

O O

N O

DH3

Fig. (3). Structures of co-crystallized ligands with pcDHFR.

Cody et al. [66] described crystal structure determination of ternary complexes of a classical antitumor, furopyrimidine N-(4{N-[(2,4-diaminofuro-[2,3-d]pyrimidin-5-yl)methyl] methylamino}benzoyl)-Lglutamate (MOT) (pcDHFR IC50: 6.5 nM; human DHFR IC50: 2.7 nM), inhibitor with pcDHFR and recombinant wild-type human DHFR. These studies provided the first direct

comparison of the binding interactions of the same antifolate inhibitor in the active site for pcDHFR and human DHFR. Cody et al. [67] provided the first evidence for ligand induced conformational changes in pcDHFR. Comparison of the data for the folic acid binary complex of pcDHFR with those for the ternary structures reveals significant differences, with a >7 Å movement of

Comprehensive Evaluation of New Models

the loop region near residue 23 that results in a new “flap-open” position for the binary complex, and a “closed” position in the ternary complexes, similar to that reported for Escherichia coli (ec) DHFR complexes in monoclinic lattice with data to 2.5 Å resolution. In the orthorhombic lattice for the binary folic acid pcDHFR complex, an unwinding of a short helical region near residue 47 was also observed, which places hydrophobic residues Phe-46 and Phe-49 toward the outer surface, a conformation that is stabilized by intermolecular packing contacts. Cody et al. [68] also reported ternary complex of NADPH, a potent antifolate [2,4-diamino-5-{3-[3-(2-acetyloxyethyl)-3benzyltriazen-1-yl]-4-chlorophenyl}-6-ethylpyrimidine] (TAB) and pcDHFR refined to 2.1 Å resolution and molecular modeling studies including molecular dynamics (MD) studies in order to determine the binding orientation of the inhibitor. Another report by Cody et al. [69] describes structural data for N-(2,4-diaminopteridin-6-yl)methyldibenz[b,f]azepine (PT653) with pcDHFR. These studies revealed that, the favorable pcDHFR selectivity of PT653 could be as a result of ligand-induced fit of the large hydrophobic dibenzazepine ring, which occupies regions of the enzyme active site not probed by other antifolates, and which take advantage of sequence and conformational differences between the structures of human and pcDHFR. High resolution crystal structures of two ternary complexes of pcDHFR with the cofactor NADPH and potent antifolates; the N910 reversed-bridge inhibitor 2,4-diamino-6-[N-(20,50-dimethoxybenzyl)-N-methylamino] quinazoline and its 30,50-dimethoxypyrido[2,3- d]pyrimidine analog, were reported by Cody et al. [70] These studies revealed the first observation of an unusual conformation for the reversed-bridge geometry (C5-C6-N9-C10 torsion angle) in this antifolate. Structures of tetrahydroquinazoline antifolate (6R,6S)-2,4diamino-6-(1-indolinomethyl)-5,6,7,8-tetrahydroquinazoline and its trimethoxy analogue (6R,6S)-2,4-diamino-6-(30,40,50-trimethoxybenzyl)-5,6,7,8-tetrahydroquinazoline as inhibitor complexes with human DHFR and pcDHFR sources and correlations between enzyme selectivity and stereochemistry were decipher by Cody et al. [71] Structural analysis of these potent and selective DHFR inhibitor complexes revealed preferential binding of the 6S-equatorial isomer in each structure. Recently, Cody et al.[72] reported structural data for two highly potent antifolates, 2,4-diamino-5-[30,40-dimethoxy-50-(5-carboxy1-pentynyl)]benzylpyrimidine (PY1011), with 5000-fold selectivity for pcDHFR, relative to rat liver (rl) DHFR refined to 2.0 Å resolution and 2,4-diamino-5-[2-methoxy-5-(4-carboxybutyloxy)benzyl] pyrimidine (PY957), that has 80-fold selectivity for pcDHFR was refined to 2.2 Å resolution. Structure of PY1011 with mouse DHFR (mDHFR), refined to 2.2 Å resolution is also reported. From these structures it was observed that carboxylate of the -carboxyalkyloxy side chain of these inhibitors form ionic interactions with the conserved Arg in the substrate binding pocket of DHFR. The structural data further revealed reasons for differences in potency of two inhibitors. Thus, the available crystal structures allow direct comparison of potent, selective inhibitors with human DHFR. Furthermore, some of the structures indicate ligand induced specific conformational changes, which could be exploited for design of better inhibitors. 2.2. Homology Modeling The crystal structures of maDHFR and tgDHFR are not solved till date. Hence, in an attempt to invoke structure based drug design against these enzymes homology modeling technique was used to deduce the 3D structure of target protein. Kharkar et al. reported the homology model of maDHFR, using crystal structure of M. tuberculosis (Mtb) DHFR (PDB ID: 1DF7 [73]) as template [47]. The

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

715

generated homology model was used from molecular docking analysis of 5-deazapteridines reported as maDHFR inhibitors. Table 2.

Summary of Comparative Modeling Reports for maDHFR and tgDHFR

Species

Template

Software

Reference

maDHFR

1DF7

MOE

46

maDHFR

1DF7

Prime

72

tgDHFR

1U70

JIGSAW server

78

tgDHFR

1U70

Prime

Current work

Very recently, Bag et al. [52] reported various molecular modeling studies on maDHFR, in which the homology model of maDHFR was developed using Prime software based on MtbDHFR template. Cody et al. reported a model of the tgDHFR active site based on the sequence alignment of tgDHFR [74] and the crystal structures of ecDHFR, MTX complexes [75] and MtbDHFR, TMP ternary complex. Popov et al. recently reported an automatically generated homology model of tgDHFR using JIGSAW [76] server using first 300 residues of tg DHFR-TS sequence. The resulting model with 217 amino acids was further refined manually using Sybyl 7.0 (Tripos). Based on the superimposition with crystal structure of Plasmodium falciparum DHFR (PDB ID: 1J3I [77]), cofactor, NADPH and ligand, WR99210, were added to the model. This model was further used for correlating the docking scores from ensemble of poses for 11 docked inhibitors with their binding affinity. A correlation of 50.2% between docking score and activity was obtained in these studies [78]. Thus, in absence of experimentally determined crystal structure, comparative modeling techniques provide an understanding of DHFR inhibition. 2.3. Structure-based Approaches - molecular Docking and Binding Affinity Prediction Availability of a large number of high resolution crystal structures of both pcDHFR and human DHFR provide a solid platform for structure-based design studies for potent and selective DHFR inhibitors. As a thumb rule, the inhibitors designed against pcDHFR were screened against maDHFR and tgDHFR. Furthermore, in some cases comparative models of maDHFR and tgDHFR were used to understand the observed potency and selectivity. Molecular docking involves two steps; first placing the inhibitor correctly in the active site, also called as pose prediction, using various algorithms such as Monte Carlo, genetic algorithm, simulated annealing, and second is scoring the predicted poses. In general, majority of docking algorithms are able to predict the pose correctly, with accuracy of ~ 2 Å root-mean-square deviation (RMSD) to that of observed crystal structure. The challenge is in scoring these poses; an ideal scoring function should be able to reproduce binding energy and should be able to rank the ligands according to their binding affinity. However, the majority of scoring functions, bundled with docking packages, often perform poorly in reproduction of binding affinity; hence, use of these scoring functions is limited to screening of databases of large number of ligand. In order to predict binding affinity of small molecule inhibitors, a variety of postdocking methods have been established. These methods range from simple consensus scoring to free energy perturbation (FEP) [79-82]. This section of the review is limited to reports where attempts were made for predicting the binding affinity of opportunistic and human DHFR inhibitors.

716 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

O NH2 X

N H 2N

O

N

O

X

Degani et al.

NH2

COOH N H

H2N

n E1-4; X= C, N; n = 0, 1

Fig. (4). General structure of classical ester soft drugs studied by GraffnerNordberg et al. using FEP.

Graffner-Nordberg et al. [83] have employed relative binding affinities, calculated from FEP simulations, for the selection of four esters soft drugs with quinazoline and pteridine nucleus (Fig. 4), as the compounds likely to inhibit human DHFR most efficiently. To account for the differences in protonation states of quinazoline and pteridine nucleus, simulations were carried out using the protonation state of the bound ligands with the free energy for protonation in water added as a correction. The results of the study demonstrated that the estimation of relative binding free energies by FEP simulations could be useful for the selection of target compounds to be synthesized for biological evaluation against DHFR; it further revealed the importance and viability of ester linkage in the MTX scaffold as potential DHFR inhibitors. Another report by Graffner-Nordberg et al. [84] discussed synthesis and the subsequent biological evaluation of a series of ester soft drugs (Fig. 5). Furthermore, MD simulations of three ligands in complex with pcDHFR and the human DHFR enzyme were conducted to understand the molecular basis for the observed selectivity. The LIE method was employed to predict the absolute binding free energies of molecules against pcDHFR and human DHFR. Interestingly, the predicted binding affinity and the selectivity ratio were well correlated with the experimental observations.

NH2 N H2N

O X

N

N

COOH

O

R

Y

Fig. (5). General structure of ester soft drugs structurally related to TMX and PTX studied by Graffner-Nordberg et al. using LIE.

A series of compounds in which the methylenamino-bridge of non-classical inhibitors was replaced with an ester function to provide potential soft drugs intended for inhalation against Pneumocystis carinii pneumonia (PCP) was reported by Graffner-Nordberg et al. (Fig. 6). In this study, the selection of the target compounds for synthesis was partly guided by an automated docking and scoring procedure using AutoDock 3.0 [85] as well as MD simulations. Even though, the AutoDock scores overestimated the magnitude of the binding free energies (-16 to -11 kcal/mol), the relative comparison was possible. In agreement with the predictions of AutoDock scores, compounds were not selective against human DHFR. Five of the docked compounds were also selected for studies using the more time consuming LIE method. Compounds were again predicted to be non selective by the LIE method. Thus, this study provided compounds with only slight preference for the fungal enzyme; furthermore, modest selectivity of the synthesized inhibitors was reasonably well predicted using the employed computational methods, although a correct ranking of the relative affinities was not successful in all cases [86].

O O

Ar

N

Fig. (6). General structure of non-classical easter soft drugs studied by Graffner-Nordberg et al. using AutoDock and LIE.

Gorse et al. [87] reported MD simulations on human DHFR in order to find out the putative stable binding conformers of the deazapterin analogs. Method based on standard thermodynamics cycles and linear approximation of polar and non-polar free energy contributions from MD averages was used to correlate the binding affinities of the different ligands in each binding site with experimental dissociation constants. The study has provided insights into structure-activity-relationships (SAR) for use in the design of modified inhibitors of DHFR. Pitts et al. [88] reported interaction energy calculations for various pcDHFR inhibitors including PTX, TMX, TMP and epiroprim using explicit solvent model. Each inhibitor was divided into different substructural regions and the minimized complexes were then used to calculate interaction energies for each intact antifolate and its corresponding substructural regions with the pcDHFR binding site residues. Substructural regions containing pteridyl, pyridopyrimidinyl and diaminopyrimidinyl subregions contributed most to the stability of antifolate interactions, while interaction energies for the hydrocarbon aromatic rings, methoxy and ethoxy groups were much less stable. Very recently, Bag et al. [89] described design, synthesis and biological evaluation of fourteen structurally diverse compounds. The top five docked poses using Glide-XP [90] score were minimized using the local optimization feature in Prime [91] and the energies were calculated using the OPLSAA force field [92] and the GBSA continuum model [93]. The docking scores from GlideXP method and MM-GBSA predicted Gbind were able to distinguish between the active and low active compounds. Furthermore, good correlation coefficient of 0.797 was obtained between the IC50 values and MM-GBSA predicted Gbind. Thus, a variety of methods ranging from simple docking scores, to computationally expensive and accurate methods like FEP, have been employed to rank order DHFR inhibitors according to their binding energy, with varying success. However, accurate prediction of binding affinity for a larger dataset of DHFR inhibitors still remains a challenge. 2.4. Ligand Based Approaches - QSAR and Pharmacophore Modeling Availability of large number of active and inactive ligands has provided a solid platform for rational ligand based drug design. In this light, several attempts were made to correlate the activity with structural properties. These methods vary from simple 2Ddescriptor based methods to more complex and advanced receptor based 3D-QSAR methods. Furthermore, the DHFR inhibitor dataset is often used to validate newly proposed ligand based approaches. A survey of various ligand based models developed for inhibitors of pcDHFR, tgDHFR and maDHFR is presented in this section. Agrawal et al. [94] reported development of QSAR models using a series of nineteen 2,6-substituted 2,4-diaminopyrido[3,2d]pyrimidine derivatives (Fig. 7) against pcDHFR using topological indexes. Mattioni et al. [95] used a data set of 345 diverse DHFR inhibitors to build QSAR models using artificial neural networks to corre-

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

R5

NH2 N

N

Z

NH2 R1

Fig. (7). 2,4-diaminopyrido[3,2-d]pyrimidine studied by Agrawal et al using QSAR.

late chemical structure and inhibition potency for pcDHFR, tgDHFR and rlDHFR. Classification models were also built using linear discriminant analysis (LDA) to predict the selectivity for pcDHFR and tgDHFR. A set of new nitrogen and oxygen-specific descriptors were developed to better encode structural features. The developed neural network models were able to accurately predict log IC50 values for the three types of DHFR to within ±0.65 log units. The best LDA models were able to correctly predict DHFR selectivity for approximately 70% of the external prediction set compounds. Debnath et al. [96] reported development of nonparabolic Hansch QSAR models using the software, Statistica, to find physicochemical and structural features of 2,4-diamino-5-methyl-6[(substituted anilino)methyl]pyrrido [2,3-d]pyrimidines (Fig. 8) required for pcDHFR, tgDHFR and rlDHFR. R5 R4

NH2 N N

N H

N H 2N

H 2N

R6

R4

R N

H 2N

717

N

N R1

R3 R2

Fig. (8). 2,4-diamino-5-methyl-6-[(substituted anilino)methyl]pyrrido [2,3d]pyrimidines studied by Debnath et al.

For pcDHFR the best developed model after removing four outliers was, pC1 = - 1.767(±0.064) + 0.800(±0.311) R_R2 - 0.578(±0.188) R3 + 0.433 (±0.159) I1 n=18, r = 0.813, %EV = 66.13, F = 9.11, p = 0.001, SEE = 0.205 For rlDHFR the best developed model after removing four outliers was, pC2 = - 2.039(±0.140) + 1.527(±0.306) R_R2 + 0.284(±0.106) MR + 0.683(±0.143) I2 n=18, r = 0.943, %EV = 88.90, F = 37.21, p = 0.000, SEE = 0.202 For tgDHFR the best developed model after removing four outliers was, pC3 = - 0.981(±0.198) - 0.908 (±0.199) R_R3 + 0.110(±0.052) log P + 0.712 (±0.128) I1 n=18, r = 0.918, %EV = 84.28, F = 25.01, p = 0.000, SEE = 0.169 This QSAR studies highlighted the importance of resonance effects, molar refractivity of substitutions on distal phenyl ring, lipophilicity of molecules and -OCH3, -CH3 substitutions on distal phenyl ring and bridge nitrogen for potent DHFR inhibitory activity. Debnath [97] reported development of pharmacophore hypotheses for a series of 2,4-diamino-5-deazapteridine inhibitors (Fig. 9) of maDHFR and human DHFR using the HypoGen module of Catalyst software.

N

N

R3 R2

Fig. (9). 2,4-diamino-5-deazapteridine used by Debnath to develop pharmacophore models for maDHFR and human DHFR.

The developed pharmacophore hypothesis for maDHFR consisted of two hydrogen bond acceptors, one hydrophobic feature, and one ring aromatic feature. This pharmacophore yielded an RMS deviation of 0.730 and a correlation coefficient of 0.967 with a cost difference (null cost minus total cost) of approximately 52. The validation of pharmacophore was carried out using a test set of compounds based on classification scheme, in which the success rate in predicting active compounds was greater than 92% while about 7% of the inactive compounds were predicted to be active. Further validation of pharmacophore was carried out using mapping of two most potent inhibitors on pharmacophore. These mapping studies provided additional confidence in developed pharmacophore. However, the developed pharmacophore lacked two donor features corresponding to 2,4-diamino group, which are critical for potent DHFR inhibition. Sutherland et al. [98] employed a dataset of 406 structurally diverse compounds to derive 3D-QSAR models using CoMSIA. Furthermore, classification models predicting selectivity for pcDHFR over rlDHFR were developed using SIMCA, with a selectivity ratio of 2 (IC50 rlDHFR/IC50 pcDHFR) used for delimiting classes. Alignment rules based on three different X-ray crystal structures were used to derive CoMSIA models. A 6-component model, using leave-10%-out cross-validation (n=240, q2=0.65) was found to be best for pcDHFR, while a 4-component model was selected for rlDHFR (n=237, q2=0.63); both include steric, electrostatic and hydrophobic contributions. The pcDHFR model showed good predictive r2=0.60 and mean absolute error (MAE) = 0.57 for the test set after removing 4 outliers, and the rlDHFR model showed r2=0.60 and MAE = 0.69 after removing 4 test set outliers. For the prediction of selectivity, a 5-component model, including steric and electrostatic contributions, showed cross-validated and test set classification rates of 0.67 and 0.68 for selective inhibitors, and 0.85 and 0.72 for selective inhibitors. The developed contour plots along with the SIMCA models provide clues for the design of potent and selective pcDHFR inhibitors. Gangjee et al. [99] reported 3D-QSAR model development using three methods, conventional CoMFA, all orientation search (AOS) CoMFA, and CoMSIA, using a dataset of 179 structurally diverse compounds from their previous publications. Low energy conformation of 5-((naphthalen-2-ylthio)methyl)furo[2,3d]pyrimidine-2,4-diamine (Fig. 10), one of the most potent and selective inhibitor was used as template for the flexible alignment using Molecular Operating Environment (MOE) suite. The models were derived against pcDHFR, tgDHFR and rlDHFR. AOS CoMFA models gave the best internal predictions (q2 = 0.604, 0.600, and 0.634 for pcDHFR, tgDHFR, and rl DHFR respectively). CoMSIA models gave the best external predictions (predictive r2 = 0.544, 0.648, and 0.488 for pcDHFR, tgDHFR, and rlDHFR respectively). Both AOS CoMFA and CoMSIA analyses were used to construct stdev*coeff. contour maps which provide an insight into SAR. Jain et al. [100] described QSAR analysis using 2D and 3D descriptors on a series of DHFR analogs of 2,4-diaminopyrido[2,3d]pyrimidines and 2,4-diaminopyrrolo[2,3-d]pyrimidines (Fig. 11).

718 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

S

NH2 N H2N

N

O

Fig. (10). 5-((naphthalen-2-ylthio)methyl)furo[2,3-d]pyrimidine-2,4diamine used as template for alignment of molecules for development of 3D-QASR models. NH2

NH2

N H 2N

N

R N

X

H2N

n N

R

N H

Fig. (11). 2,4-diaminopyrido[2,3-d]pyrimidines and 2,4-diaminopyrrolo[2,3d]pyrimidines used by Jain et al. to develop QSAR models using 2D and 3D descriptors.

Sequential multiple liner regression analysis method using VALSTAT program was employed to derive QSAR equations. Different equations were developed for ma DHFR, pcDHFR, tgDHFR and rlDHFR. For maDHFR the best developed model using 3D descriptors was, pIC50 = 2.244 (±0.786) LUMO - 3.049 (±0.386) SBE - 0.339 (± 0.082) DPL3 + 3.031 n = 31, r = 0.907, r2 = 0.822, SE = 0.453, F = 26.216, ICWP < 0.62 For pcDHFR the best developed model using 3D descriptors was, pIC50 = 2.367 (±0.296) SBE - 0.268 (± 0.071) DPL3 + 1.637 (± 0.603) LUMO + 1.741 n = 21, r = 0.910, r2 = 0.829, SE = 0.348, F = 27.431, ICWP < 0.62 For tgDHFR the best developed model using 3D descriptors was, pIC50 = 2.374 (±0.687) LUMO + 2.555 (± 0.338) SBE - 0.258 (± 0.081) DPL3 + 3.009 n = 21, r = 0.893, r2 = 0.797, SE = 0.395, F = 22.293, ICWP < 0.62 For rlDHFR the best developed model using 3D descriptors was, pIC50 = 1.879 (±0.474) HOMO + 2.687 (± 0.245) SBE - 3.400 (± 0.071) DPL3 -17.522 n = 21, r = 0.94, r2 = 0.884, SE = 0.336, F = 43.251, ICWP < 0.44 Thus, the developed models showed that, the electronic properties, energy of LUMO, Z-component of dipole moment (DPL3) and SBE of the molecules was positively correlated with activity. Richmond et al. [101] described a new algorithm GHALAND for pharmacophore identification by hypermolecular alignment of ligands in 3D. A set of six ligands belonging to folate and deazafolate classes was used to address the issue of alignment in pharmacophore generation. Pair-wise rigid ligand alignment based on linear assignment (the LAMDA algorithm) was developed as potential method for generation of alignments for pharmacophore identification. Evans et al. [102] described direct comparison of 3D-QSAR methods; Phase and Catalyst, using the dataset used by Sutherland et al. including DHFR inhibitors. Results revealed that the newly developed 3D-QSAR package Phase performed better or equal to Catalyst.

Manchester et al. [103] used a dataset of DHFR inhibitors from the work of Sutherland et al. to compare Simple Atom Mapping Following Alignment (SAMFA), a newly proposed 3D-QSAR method with CoMFA. Three regression approaches (PLS, SVM, RandomForest), as implemented in R, and Monte Carlo crossvalidation (MCCV) numerical experiments were used to derive models. The results indicated that SAMFA descriptors, despite their simplicity, perform well when compared to the much more refined CoMFA descriptors. Furthermore, SAMFA descriptors were found to be readily interpretable and applicable to the difficult problem of inverse QSAR. Santos-Filho et al. [104] developed a new drug design approach which includes docking, molecular fingerprints based cluster analysis, and ‘induced’ descriptors based receptor-dependent 3D-QSAR. The approach was validated using eight data sets sampled from the literature and from public databases including the pcDHFR dataset previously studied by Sutherland et al. Initially, the dataset of 756 pcDHFR studied by Sutherland et al. was used to build 3D-QSAR model. However, due to large structural variations, this entire dataset could not yield significant QSAR model. Hence, based on cluster analysis, a subset of 70 molecules (60 training set and 10 test set) was used to derive a statistically significant model. The best developed model for pcDHFR was, pIC50 (M) = 3.43 - 3.85 Rs_ARG_75_NH2 - 6.91 Rs_SER_64_OG + 10.14 Rs_THR_61_N (13) + 4.84 Sigma_GLU_32_OE1 - 3.10 Sigma_ILE_33_CD1 + 5.61 Sigma_ILE_65_CG2 -17.87 Sigma_LEU_72_N + 17.10 Sigma_LYS_37_N + 6.29 Sigma_PRO_66_CD -16.57Sigma_TRP_27_CA 5.03 Sigma_TRP_27_NE1 + 4.33 Sigma_TYR_129_OH N = 60, r2 = 0.82, q2 = 0.72 where, Rs corresponds to steric influence of the group and  corresponds to inductive effect of group. The QSAR model indicated importance of the hydrogen bond between the hydroxyl group of Tyr-129 and the cofactor NADPH. The negative regression coefficient of Sigma_LEU_72_N indicates that a polar group close to Leu-72 would disturb the favorable hydrophobic interactions with the ligand. Sigma_LYS_37_N shows importance of stable interaction of Lys-37, with the polar group on the ligand. The high regression coefficient of Rs_THR_61_N, indicate that a bulky group close to this small residue would result in unfavorable steric hindrance. Bag et al. [52] reported structure-based pharmacophore development; initially a homology model was developed for maDHFR based on the crystal structure of mtbDHFR. The developed homology model was used to identify putative bioactive conformers for various ligands. The conformers from molecular docking studies were used for predictive pharmacophore generation, which highlighted critical features required for binding. The pharmacophore was developed based on nineteen diverse, most active compounds using PHASE, the model was validated using atom based QSAR assessment employing a dataset of ninety 2,4 diaminodeazapteridine analogs. The alignment of bioactive conformers generated using PHASE was used to develop CoMFA and CoMSIA models. The major difference between the earlier pharmacophore hypothesis developed using Catalyst and the one proposed in this study was that the hydrogen bond acceptor feature corresponding to 3-nitrogen of 2,4-diaminodeazapteridin ring was missing in this pharmacophore. Also, one hydrogen bond acceptor feature corresponding to 8-nitrogen was present in Catalyst pharmacophore while in the one proposed in this study, hydrogen bond acceptor feature corresponding to nitrogen at 1-position in 2,4-diaminodeazapteridin ring was present. Moreover, the proposed new pharmacophore had two new donor features mapped to N-H bonds of two amino groups of 2,4-diaminodeazapteridin ring system. The proposed pharmacophore was found to refine the earlier pharmacophore and was in agreement with mechanism of the reaction cata-

Comprehensive Evaluation of New Models

lyzed by the enzyme. The alignment from pharmacophore model was used to develop CoMFA and CoMSIA models from 68 compounds taken from a dataset of 90 compounds. Each model was further validated using a test set of 22 compounds not included in the training set. Out of the various models evaluated CoMSIA model with a combination of steric, electrostatic, hydrophobic, and H-bond fields produced a statistically significant model with good correlation and predictive power. Furthermore, analysis of various contours provided details about the SAR. Very recently, Gangjee et al. [105] described CoMFA analysis of tgDHFR and rlDHFR based on 80 antifolates with 6-5 fused ring system using the all-orientation search (AOS) routine and a modified cross-validated r2-guided region selection (q2-GRS) routine. In this study, two modifications of q2-GRS routine were suggested to improve the predictability of models. In the first modification, the lowest corner of each modified subregion was the lowest grid point of the conventional CoMFA grid enclosed by the original q2-GRS subregion, and the highest corner of each modified subregion was the highest grid point of the conventional CoMFA grid enclosed within the subregion of the original q2-GRS routine. In the second modification, the region box was divided into 125 equal-sized subregions and the distance between the adjacent subregions was 1 Å, whereas in the original routine, the adjacent subregions touch each other. Among the various generated models, the q2-GRS model using the second modification had shown the best external predictive r2 (0.499) along with a satisfactory internal cross-validated q2 at 0.647 (ONC = 3). On the basis of the steric contour maps of the models, four new compounds were designed belonging to 2,4-diamino-5methyl-6-phenylsulfanyl-substituted pyrrolo[2,3-d]pyrimidine series. It was observed that, as predicted, the new compounds were potent and selective inhibitors of tgDHFR. One of them, 2,4diamino-5-methyl-6-(2’,6’-dimethylphenylthio)pyrrolo[2,3d]pyrimidine, had shown nanomolar tgDHFR inhibitory activity. Thus, the QSAR models derived from the homologues series (e.g. 2,4-diaminopyrido[3,2-d]pyrimidine, 2,4-diamino-5-methyl-6[(substituted anilino)methyl]pyrrido [2,3-d]pyrimidines, 2,4diaminopyrido[2,3-d]pyrimidines and 2,4-diaminopyrrolo[2,3d]pyrimidines), for DHFR inhibition, have applicability domains restricted to specific chemotypes. Furthermore, these models and even some of the models derived from larger datasets, provide no information about the binding site. On the other hand, the 3DQSAR models derived using methods such as, CoMFA and CoMSIA, have distinct advantage of stdev*coeff. contour maps, which provides an insight into SAR and generation of SAR in context of active site. However, these models lack inherent ability to mine 3D databases in search of potent/selective DHFR inhibitors. The models developed using receptor residue interaction energy, quantify the binding contribution of active site residues. Pharmacophore models are reported for maDHFR, however, there are no reports for development of pharmacophore models for pcDHFR and tgDHFR. Towards this direction, herein, we have developed structurechemical-feature based pharmacophore models which take advantage of both, the available structural and ligand information, to decipher SAR in a more comprehensive manner. 3. COMPREHENSIVE EVALUATION OF NEW MODELS Herein, we report development of structure-chemical-featurebased pharmacophore models and exploration of SAR for pcDHFR, tgDHFR and rlDHFR inhibitors. 3.1. Material and Methods 3.1.1. Software and Hardware The structures of all compounds from the reported literature were constructed using MarvinSketch version 5.2 from ChemAxon Ltd [106]. A database of structures was created and maintained

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

719

using JChem for Excel version 1.1. The MDL SDF file of the structures was imported in Maestro, version 8.5, Schrödinger, LLC, New York, NY, 2008, for further studies [107]. The homology modeling studies were carried out using Prime module version 3.0. Protein structure initial minimization was carried out using MacroModel module version 9.6 [108], Molecular dynamics studies were carried out using Impact module version 5.0 [109], Ligand three dimension structure generation was carried out using Ligprep module version 2.2 [110], Molecular docking studies were carried out using Glide module version 5.0. The pharmacophore hypothesis and 3D-QSAR model building based on it was carried out using PHASE module version 3.0 [111]. All softwares were installed on Fujitsu workstations M470 with Quadra core Xeon processor and 16 GB of RAM. 3.1.2. Dataset All ligand data used in this study was collected from literature reported by Rosowsky et al. [20-40] The compounds belong to diverse structural classes including, 2,4-diaminothieno[2,3d]pyrimidine [20,27], 2,4-diaminoquinazoline [21,22,24,32], 2,4diaminopteridine [22,30], 1,3-diamino-7,8,9,10-tetrahydropyrimido[4,5-c]isoquinoline [22], 1-aryl-4,6-diamino-1,2-dihydro-striazine [22], 2,4-diamino-5,6,7,8-tetrahydropyrido[4,3-d]pyrimidine [23], 2,4-diaminopyrido[3,2-d]pyrimidine [25,32,35], brominated TMX [26], 2,4-diamino-6,7-dihydro-5H-cyclopenta[d]pyrimidine [28], 2,4-diamino-6-(arylmethyl)-5,6,7,8-tetrahydroquinazoline [29], bridge alicyclic ketone [31], 2,4-diamino-7Hpyrrolo[2,3-d]pyrimidine [33], 2,4-diamino-5-[2-methoxy-5-(öcarboxyalkyloxy)benzyl]pyrimidine, 2,4-diamino-5-(2’,5’-disubstituted benzyl)pyrimidine [34], 2,4-diamino-5-(2’,5’-disubstituted benzyl)pyrimidines [35], 2,4-diamino-6-[2’-O-(ö-carboxyalkyl) oxydibenz[b,f]azepin-5-yl]-methylpteridine [36], 2,4-diaminopyrimidine [37-40]. The biological data used in this study were collected from pcDHFR, tgDHFR, and rl DHFR using a continuous spectrophotometric assay measuring oxidation at 340 nM of NADPH at 37ºC under conditions of saturating substrate and cofactor in the laboratory of S. F. Queneer and the same procedure for determining the IC50 value was followed [112,113]. The compounds with exactly determined biological activity were only considered for respective studies. The negative logarithm of the measured IC50 value (pIC50) was used in the 3D-QSAR study. The biological activity (pIC50) varies from ~ 4.5 to ~ 11. Thus, biological activity of the dataset spans ~5-6 log units. Considering the structural and biological activity diversity, this dataset provides a solid platform for understanding DHFR inhibitory potential. The structures of compounds with their biological activity are depicted in Table 3. 3.1.3. Protein Structure Ternary complex of pcDHFR crystal structure (PDB code: 1KLK [68]) with NADPH and [N-(2,4-diaminopteridin-6-yl)methyl]-dibenz[b,f]azepine refined to 2.2 Å resolution was used for the molecular docking of pcDHFR inhibitors. This structure was selected for docking studies, due to the observed ligand induced flexibility, owing to presence of large hydrophobic dibenzazepine ring substitution. As human DHFR and rlDHFR share 90% and their active site residues are almost identical, human DHFR (PDB code: 1KMS [46]) complexed with NADPH and 6-([5-quinolylamino]methyl)2,4-diamino-5-methylpyrido[2,3-d]pyrimidine (SRI-9439), a lipophilic antifolate refined to 1.09 Å resolution was used for docking instead of rlDHFR. Since, experimental structure for tgDHFR is not solved till date, a homology model was generated based on closely related Mus musculus DHFR [114]. The homology model was further refined using 1 nanosecond MD simulations. The refined model was then used for subsequent molecular modeling studies on tgDHFR.

720 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Generation of Homology Model for tgDHFR First 220 amino acids of bifunctional dihydrofolate reductasethymidylate synthase from tg were used to develop the homology model. The sequence of amino acids of the tgDHFR was retrieved from the Swiss-Prot Database (entry Q07422). Identification of homologues of tgDHFR was carried out by performing sequence database searches with BLAST using BLOSUM62 similarity matrix. Coordinates of the crystal structure of Mus musculus DHFR (PDB ID: 1U70, 2.5 Å resolution) with 32% sequence identity were used as template to build the initial tgDHFR structure. Alignment of the query sequence was then carried out on the template based on secondary structure using secondary structure prediction (SSP) program SSpro. The 3D-structure for tgDHFR was then built using the template. The cofactor (NADPH) and ligand were added to the model in the model building step. The resultant structure of the tgDHFR was then subjected to the side chain prediction and loop refinement. The final protein model was validated by Ramachandran’s plot. Molecular Dynamics (MD) Simulations The homology model generated in our laboratory was refinec using MD simulations. Initially, the protein complex was minimized using 1,000 cycles of steepest descent followed by truncated Newton conjugate gradient method, until the gradient reached a value smaller than 0.05, using OPLS_2005 force field in Macromodel. The minimized protein structure was immersed in a cubic box (555555 Å) SPC water with density of 1, using program Soak in Impact; to perform the MD in an explicit solvent model. Appropriate number of counterions were added to the solvent bulk of the protein/water complex to maintain neutrality of the system. Periodic boundary conditions were applied with Ewald long-range correction. The SHAKE algorithm was employed in order to constrain all bonds involving hydrogen atoms. The integration was carried out using Verlet algorithm. Initially, water molecules were equilibrated for 50 picoseconds (ps) keeping the non-water components fixed. Thereafter, a MD simulation was carried out for 1 ns using NVT ensemble at a constant temperature of 298.15 K. After every 1 ps of MD simulation actual frame was stored for analysis. The RMSD values were computed for all the frames relative to the starting structure, using the Protein structure alignment option in Maestro, and energy values were extracted using a Perl script from the generated trajectories. Similarly, 1ns MD was carried out on protein structure without inhibitor (MTX) and cofactor (NADPH). 3.1.4. Molecular Docking Studies With the help of Ligprep facility, appropriate hydrogens were added and a single, low-energy, 3D conformation was generated for each ligand in the dataset by energy minimization using OPLS_2005 force field with a dielectric constant of 1.0. The reported stereochemistry was preserved throughout the process and all the structures were considered in their neutral form. The protein structure of pcDHFR (PDB code: 1KLK) was used for the validation of the molecular docking algorithm. Hydrogens were added to the protein structure, NADPH and inhibitor N-(2,4diaminopteridin-6-yl)methyl]-dibenz[b,f]azepine (PMD); bond orders and proper charges were assigned. All the water molecules were deleted. Terminal rotameric states for Asn, Gln, and His as well as tautomeric and protonation states of His to optimize the hydrogen-bonding network in the complex were set automatically using the protein assignment program in Maestro. In addition, hydroxyl and thiol torsions in Cys, Ser, and Tyr were optimized. Finally, the protein-ligand complexes were subjected to a restrained minimization using the ‘‘impref’’ tool available in Protein Preparation Wizard in Maestro using RMSD cutoff value of 0.3 Å. A grid which is the representation of shape and properties of receptor using several different sets of fields was generated. The centroid

Degani et al.

of the bound crystallographic ligand was used to define the center of each grid box. The structure of PMD was randomized using Ligprep and used for the cognate docking studies. The docking study was carried out using extra precision (XP) mode with ‘must match’ of at least one hydrogen bond criteria; with Glu32, Ile10 and Ile123. Once the Glide algorithm was validated, molecular docking was carried out for the complete dataset of structures in the active site of pcDHFR, tgDHFR and human DHFR. The protein structure for human DHFR was prepared using protein preparation wizard in Maestro. Grids were generated for tgDHFR and human DHFR as described above. The molecular docking was carried out using XP mode with ‘must match’ of at least one hydrogen bond criteria; with Glu32, Ile10 and Ile123 in pcDHFR; Asp31, Val8 and Val151 in tgDHFR and with Glu30, Ile7 and Val115 in human DHFR, with the criteria for duplicate pose removal set to zero. A maximum of 50 poses were stored for each ligand. 3.1.5. Pharmacophore Mapping The multiple docked poses obtained from the molecule docking studies were imported in the “Develop Pharmacophore Hypothesis” panel of PHASE and were used for pharmacophore mapping studies as input conformers. The chemical features of ligand that may facilitate noncovalent bonding between the ligand and its target receptor were defined using six built-in pharmacophoric features: hydrogen bond acceptor (A), hydrogen bond donor (D), hydrophobic group (H), negatively charged group (N), positively charged group (P), aromatic ring (R), by a set of chemical structure patterns as SMARTS queries and assigned one of three possible geometries, which define physical characteristics of the site: • Point- the site is located on a single atom in the SMARTS query. • Vector- the site is located on a single atom in the SMARTS query, and assigned directionality according to one or more vectors originating from the atom. • Group- the site is located at the centroid of a group of atoms in the SMARTS query. For aromatic rings, the site is assigned directionality, defined by a vector that is normal to the plane of the ring. The aromatic rings were considered as both, the default aromatic ring (R) feature and hydrophobic feature with group geometry. Active analog approach was used to identify common pharmacophore hypotheses (CPHs), in which common pharmacophores were culled from the conformations of the set of highly active ligands (pIC50 > 9.0). This activity threshold resulted in 15 molecules for pcDHFR, 28 molecules for tgDHFR and 18 compounds for rlDHFR. Common pharmacophoric features were then identified from a set of variants, a set of feature types that define a possible pharmacophore, using a tree-based partitioning technique which groups together similar pharmacophores according to their intersite distances employing a tree depth of five with initial box size of 25.6 Å and final box size of 0.8 Å. The hypotheses obtained from this process were assigned a score comprising of geometric and heuristic factors with respect to actives, using overall maximum root-mean-square deviation (RMSD) value of 1.2 Å with default options for distance tolerance. Ligand activity, expressed as pIC50, was incorporated into the score with a weight of 1.0 and relative conformational energy, EConfref (kJ/mol) was included with a weight of 0.01, considering the fact that the bioactive structure will correspond to a reasonably low point on the energy surface of the free ligand. The total active score for a given hypothesis was defined as:

Comprehensive Evaluation of New Models

Active Score = Reference Score + wvolume Volume Score + wselectivity Selectivity Score + wconf EConfref + wM-1match (1) Where, w indicates weight and M is the number of actives that match the hypothesis. In the hypothesis generation, default values were used. Hypotheses emerging from this process were subsequently scored with respect to inactives (pIC50 < 5.5), using a weight of 1.0. This activity threshold resulted in 21 molecules for pcDHFR, 11 molecules for tgDHFR and 26 compounds for rlDHFR. 3D-QSAR Model Building The dataset was divided randomly into a training set and test set, incorporating biological and chemical diversity. A rectangular grid with spacing of 1.0  was defined to encompass the space occupied by the aligned training set molecules. Based on the occupation of a cube by a ligand atom of training set molecules, total number of volume bits were assigned to a given cube. Thus, molecules were represented by a string of zeros and ones, according to the cubes occupied and the different types of atoms/sites that reside in those cubes [115]. QSAR models were generated by applying partial least squares (PLS) regression to this pool of binary-valued independent variables and using pIC50 as independent variable. Models were generated for all developed pharmacophore based alignments using 1 to 5 PLS factors. Each of these models was validated using a test set molecules which were not considered during model generation. The training set comprised of 143, 130 and 144 compounds, for pcDHFR, tgDHFR and rlDHFR respectively. The test set comprised of 61, 56 and 61 compounds for pcDHFR, tgDHFR and rlDHFR respectively. 3.2. Results and Discussion Protocol for structure-chemical-feature-based pharmacophore model development is depicted in Fig. 12. Concisely, the input for study is 3D-structure of target receptor and database of ligands with structural and biological activity diversity. In the present study, 3D-strucure for pcDHFR and mammalian DHFR were taken from X-ray crystallographic results, while for tgDHFR, comparative model was developed. In the next step, flexible ligand docking was performed using Glide-XP. The conformers obtained from docking studies were considered as input for ligand based approaches such as pharmacophore and subsequent QSAR

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

721

analysis of pcDHFR, tgDHFR and rlDHFR inhibitors. Our results highlight certain critical features required for DHFR binding. 3.2.1. Construction of Homology Model for tgDHFR Since, no experimentally determined crystal structure of tgDHFR is present in public domain, to initiate development of comprehensive molecular modeling studies in an attempt to highlight the critical structural features for tgDHFR inhibition, a homology model was developed using Mus musculus DHFR template (PDB ID: 1U70) having a resolution of 2.5 Å with 32% sequence identity. Using the generated homology model, interactions of the reported inhibitors at the active site were explored. The sequence alignment between the template protein (Mus musculus DHFR; PDB ID: 1U70) and the target protein, tgDHFR by Align program of Prime module in Fig. 13. There are two major insertions in the tgDHFR sequence compared to Mus musculus DHFR; from residues 42-67 and 127-136. These regions are far away from the active site and hence do not interfere with the quality of active site modeled. The final homology modeling calculation achieved by PRIME has generated a reliable 3D structure of tgDHFR. The Ramachandran plot analysis revealed that most of the residues of the 3D structure lay in most favored regions and only few residues lay in additional allowed region thus validating the generated model. The template used along with the model developed, showed the overall fold conservation. The RMSD between the template and the target structures is 1.29 Å with alignment score of 0.067. The active site was defined by considering residues within 6 Å of the ligand MTX. The comparison between the template and homology model at the active site region revealed that overall interactions were conserved in both species. For instance, the Glu30, Ile7 and Val115 residues form the hydrogen bond with 2,4-diamino groups; a privileged feature for antifolate activity in the template protein while Asp31, Val8 and Val151 form hydrogen bond in tgDHFR respectively. One more residue, Phe34 of Mus musculus DHFR which may be involved in - stacking interaction with substrate is also conserved in tgDHFR (Phe35). However, some of the residues are not conserved at the binding pocket. As an example, hydrophilic residue, Asn64, just outside the distal binding pocket in Mus musculus DHFR is replaced with hydrophobic, Phe91 in tgDHFR.

Fig. (12). Protocol employed for development of structure-chemical-feature-based pharmacophore model development.

722 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (13). Sequence alignment result between the template protein (Mus musculus DHFR [PDB ID: 1U70]) and the target protein (tgDHFR).

Molecular Dynamics (MD) Simulations The tgDHFR structure was energy-minimized to remove bad contacts derived from the homology modeling and to achieve a good starting structure with which to perform the MD simulations. To assess the stability and refine the developed homology model, 1 ns MD simulations were carried out using explicit solvent model under periodic boundary conditions. Change in RMSD values, a commonly used parameter indicating the structural movement was used to assess the dynamic behaviour and structural changes of the receptor. Change in potential energy values was also monitored during simulations. The MD simulations were carried out for both protein-inhibitor-cofactor complex and for only protein. It can be seen from the Fig. 14, that the potential energy of the tgDHFR model without any ligand drops rapidly during first ~300 ps and then drops slowly to achieve a plateau. The RMSD of the structure without cofactor (NADPH) and inhibitor (MTX), increases rapidly, during first ~300 ps to ~ 2.0Å thereafter increases slowly except a rapid fluctuation to ~2.5Å during 550-600 ps. The RMSD remains at equilibrium value of ~ 2.2 Å during last 400 ps. This patterns of changes in potential energy is very similar for enzyme in presence of ligand and inhibitor, however, the RMSD increases slowly till 800 ps to ~ 2.5 Å and the remains at equilibrium during last 200 ps. These results are indicative of the stability of generated homology model of tgDHFR. 3.2.2. Molecular Docking Studies Validation of Molecular Docking Algorithm The accuracy of molecular docking simulation depends on the molecular docking algorithm and the parameters employed for docking. As discussed previously, molecular docking consists of two steps; fist, placing the inhibitors in active site i.e. pose prediction and second, scoring the obtained poses. Since, the aim of this study is to predict multiple poses in active site and use these poses as input for ligand based drug design method, only the effectiveness of docking algorithm to reproduce observed crystal structure was evaluated and the quality of scoring function was not considered. The experimentally determined structures of pcDHFR and human DHFR were used as test cases. The model preparation and docking studies were carried out using the protocol outlined in the experimental section. Since, DHFR has a highly conserved ligand binding orientation (Fig. 15), ‘must match’ of at least one hydrogen bond criteria; with Glu32, Ile10 and Ile123 in pcDHFR; Asp31, Val8 and Val151 in tgDHFR and with Glu30, Ile7 and Val115 in rlDHFR residues was constrained during docking simulations. Using flexible docking procedure in Glide-XP, the observed crystallographic structure was reproduced with RMSD value of 1.65 for pcDHFR and 1.85 for human DHFR, indicating that this docking method is

valid and can be used for predicting binding modes of DHFR inhibitors. Molecular Docking Studies for pcDHFR, tgDHFR and Human DHFR Once suitable parameters for docking were established, further studies using Glide, were carried out for DHFR inhibitors, to understand the forms of interaction of the inhibitors and predict their binding orientation. Docking of the inhibitors was carried out using Glide-XP protocol with the criteria for duplicate pose eliminations reduced to zero, in order to ensure a full sampling of conformational space irrespective of scoring function. Maximum of 50 poses were saved for each ligand. Only compounds for which at least one valid pose was returned using molecular docking studies were considered further. 3.2.3. Development of Common Pharmacophore Hypotheses (CPH) The results of ligand based approaches like pharmacophore mapping and 3D-QSAR analysis are mostly dependent on the quality of input conformers and the resulting alignment. To address this, herein we have employed an ensemble of conformers generated using molecular docking studies for pharmacophore mapping studies. The alignment generated based on pharmacophore models was then used for 3D-QSAR model development. CPH were generated by systematically varying the number of sites (nsites) and the number of matching active compounds (nact). The nact were selected by using a cutoff of pIC50 > 9.00 which corresponds to IC50 in nanomols, nsites was varied from 6 to 5 until at least one hypothesis was found and scored successfully; if this failed, nact was decreased by one and the nsites cycle was repeated. Considering the flexibility and the length of molecules under study, to optimize the number of generated pharmacophores, a terminal box size of 0.8 Å and tree depth of 5, which resulted in initial box size of 25.6 Å, was employed to search for CPH a in tree based partition algorithm. Based on the molecule occupancy of the pharmacophoric features and statistical performance, six featured CPHs were considered for pcDHFR, tgDHFR and rlDHFR. A total of 7,655 probable six featured CPHs belonging to five types, ADDDRR, AADDRR, AADRRR, ADDRRR and ADDDDR for pcDHFR; 578 probable six featured CPHs belonging to six types, ADDHPR, ADDHHR, DDHHRR, ADDPRR and DDHHPR for tgDHFR; 695 probable six featured CPHs belonging to fifteen types, ADDHPR, ADDHRR, ADDHHR, AADDHP, DDHPRR, DDHHRR, DDHHHR, ADDRRR, AADDPR, DDHHHP, AADDRR, AADDHR, ADDPRR, DDHHPR and ADDHHP for rlDHFR, were subjected to stringent scoring function analysis with respect to actives using default parameters for site, vector and volume. Reference relative conformational energy (kJ/mol) was included in the score with a

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

723

Fig. (14). Change in energy and RMSD of tgDHFR protein without ligand, during the simulation time of 1 nanosecond. (Pink colour points represent change in RMSD for protein with ligand).

Aliphatic residue

O O

Aliphatic residue

H

H

R N H N H

H

O

O

Acidic residue Fig. (15). Experimentally observed conserved geometry in the active site of DHFR for compounds with a 2,4-diaminopyrimidine nucleus.

weight of 0.01 and ligand activity, expressed as pIC50, was incorporated with a weight of 1.0. Hypotheses emerging from this process were subsequently scored with respect to the inactives, using a weight of 1.0. The inactives were defined by a cut off of pIC50 < 5.5 which corresponds to ligands with approximately >200 M activity. The hypotheses which survived the scoring process were used to build both, the pharmacophore based and atom-based 3D-QSAR model by correlating the observed and estimated activity for the internal set of training molecules using partial least-square analysis. All the models were validated using a test set of molecules randomly selected from the complete dataset with biological and structural diversity. Only these compounds, which successfully aligned using pharmacophoric points were considered for 3DQSAR studies. The reliability and robustness of QSAR models was judged based on overall performance of various pharmacophore models with respect to different statistical parameters, such as SD standard deviation of the regression, r2 value of r2 for the regression, F variance ratio, P significance level of variance ratio, RMSE root-meansquare error, Q2 value of Q2 for the predicted activities and Pear-

son-R Correlation between the predicted and observed activity for the test set. The atom based QSAR models were found to be more predictive and statistically significant than pharmacophore based models. Hence, thereafter atom based QSAR models were considered for further studies. Actual and predicted values of the training set and test set molecules are given in Table 3. The statistical parameters for some of the representative atom based QSAR models with survival scores of pharmacophores are reported in Tables 4-6. Pc-CPH1 model, associated with 3 PLS factors, was found to be the best for pcDHFR (Table 4). The model Pc-CPH1 showed excellent correlation for training set molecules (r2 = 0.80, SD = 0.58, F = 196, N = 143); furthermore, the model also showed good predictive power for test set (Q2 = 0.57, RMSE = 0.85, Pearson-R = 0.76, N = 61). Fig. 16 shows the linearity trend in the graphs of actual vs. predicted activity for training and test set molecules for pcDHFR. Hence, the hypothesis Pc-CPH1 (Fig. 17) with one hydrogen bond acceptor (A), two hydrogen bond donors (D), two hydrophobic groups (H) and one positive ionic group (P) as pharmacophoric features was retained for further studies. The pharmacophore

724 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 3.

Degani et al.

Structures of Compounds Used in the Present Study Along with Their Actual and Predicted Activities, - Indicates Activity not Considered for the Study

Table 3a NH2

NH2

N

N

(CH2)n

H 2N

S

N

H 2N

Z

Z S

N

1_007-1_008

1_001-1_006

Compd

n

(CH2)n

X

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

1_001

0

2,5-(MeO)2

6.32

7.89

7.43

6.25

6.52

7.33

1_002

1

2,5-(MeO)2

5.85

8.15

7.40

7.04

7.48

7.1

1_003

2

2,5-(MeO)2

6.92

6.48

6.23

6.85

7.02

6.25

1_004

0

3,4,5-(MeO)3

-

7.49

6.74

-

6.89

6.49

1_005

1

3,4,5-(MeO)3

-

7.20

5.29

-

7.23

5.64

1_006

2

3,4,5-(MeO)3

-

5.74

5.60

-

7.04

5.69

1_007

2

2,5-(MeO)2

5.55

6.24

6.51

5.45

6.69

6.85

1_008

3

2,5-(MeO)2

5.26

5.62

5.59

6.17

6.61

6.86

Table 3b:

NH2 N H 2N

NH2

R1

S

N

H2N

1_009-1_013

Compd

R1

R2

R1

N

R2

S

N

1_014-1_018

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

1_009

CH3

C6H5

5.59

6.41

5.24

6.4

7.03

6.07

1_010

CH3

3,4-Cl2 C6H3

5.80

-

6.34

6.07

-

-

1_011

CH3

CH2 C6H5

5.46

6.21

5.85

6.07

6.92

6.72

1_012

4-Cl C6H 5

CH3

-

-

-

-

-

-

1_013

CH2 C6H5

CH3

-

-

-

-

-

-

1_014

CH2

-

6.68

6.55

6.41

5.83

6.08

6.25

1_015

(CH2)2

-

6.85

7.07

6.89

5.92

6.42

6.26

1_016

(CH2)3

-

6.19

-

6.72

5.75

-

6.24

1_017

(CH2)7

-

-

-

6.03

-

-

-

1_018

N(CH2 C6H5)

-

6.28

7.07

7.24

5.55

6.64

6.89

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 3c: NH2 Cl Y

N H2N

MeO

NH2 N

N

Z N

H2N

Y

OMe

N

2_019-2_025

Compd

N H 2_026

Z

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

2_019

NHCH2

2,5-(MeO)2

8.28

8.77

7.51

8.13

7.67

8.73

2_020

NH(CH2)2

2,5-(MeO)2

5.92

7.00

7.45

7.8

7.08

7.64

2_021

NH(CH2)3

2,5-(MeO)2

5.59

7.00

6.74

7.45

6.97

6.73

2_022

NHCH2

3,4,5-(MeO)3

8.48

9.15

9.13

8.32

7.77

8.62

2_023

NH(CH2)2

3,4,5-(MeO)3

5.37

6.59

6.65

5.85

7.84

6.44

2_024

NH(CH2)3

3,4,5-(MeO)3

5.02

5.85

5.97

6.82

6.71

5.33

2_025

N(Me)CH2

3,4,5-(MeO)3

7.77

8.80

8.98

8.08

7.74

8.31

2_026

-

-

6.85

8.00

7.37

7.22

8.33

8.36

Table 3d: NH2 R1 R2

N H 2N Compd

R1

R2

N

R3

R3 Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

3_027

CH3

H

H

6.82

6.30

6.13

6.97

6.56

5.98

3_028

CF3

H

H

5.09

-

4.67

-

-

5.82

3_029

F

H

H

7.39

-

6.30

6.49

-

5.63

3_030

Cl

H

H

7.38

-

6.41

6.74

-

6.05

3_031

Br

H

H

7.55

-

6.64

6.86

-

6.13

3_032

I

H

H

7.28

-

6.62

6.86

-

6.32

3_033

OCH3

H

H

6.35

7.32

6.92

6.44

6.21

6.24

3_034

OC2H5

H

H

6.33

6.06

5.96

6.26

6.12

6.42

3_035

OC3H7

H

H

7.02

7.82

7.92

6.25

6.48

6.8

3_036

OCH2CF3

H

H

5.89

-

7.85

5.93

-

6.89

3_037

OCH2C2 F5

H

H

5.27

-

7.39

5.9

-

6.48

3_038

OCH2C3 F7

H

H

5.15

-

6.21

4.99

-

6.42

3_039

SCH3

H

H

6.96

-

7.27

6.37

-

6.29

3_040

H

CH3

H

5.70

-

5.26

6.37

-

5.69

3_041

H

CF3

H

6.57

6.51

6.39

6.25

6.73

6.2

725

726 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

(Table 3d) Contd…. Compd

R1

R2

R3

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

3_042

H

F

H

4.92

4.98

4.34

6.37

6.37

5.42

3_043

H

Cl

H

6.44

5.85

5.54

6.33

6.56

5.6

3_044

H

Br

H

6.80

5.85

6.37

3_045

H

I

H

6.80

6.80

6.23

6.33

6.65

5.84

3_046

H

OCH3

H

5.24

5.66

5.34

6.26

6.79

5.83

3_047

H

H

CH3

5.19

5.09

5.38

6.26

6.26

5.31

3_048

H

H

CF3

5.51

-

5.20

6.04

-

5.31

3_049

H

H

F

5.39

-

4.72

6.36

-

5.24

3_050

H

H

Cl

5.35

-

5.28

6.07

-

5.25

3_051

H

H

Br

5.92

-

5.72

6.08

-

5.23

3_052

Cl

Cl

H

6.01

-

6.46

7.03

-

6.42

3_053

Cl

Br

H

7.85

-

7.10

7.07

-

6.52

3_054

H

CH3

CH3

6.28

6.11

6.10

6.16

6.54

5.58

5.83

Table 3e:

NH2 N H 2N Compd

R

N

N

R

N

R

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

6.7

-

3_055

4-n-C4H9C6H 4CH2

6.41

5.85

5.36

3_056

3-CF3C6H4CH 2

5.59

5.41

5.19

6.37

6.08

-

3_057

4-ClC6H4 CH2

6.23

6.60

6.08

6.84

6.77

-

3_058

3,4-Cl2C6H 3CH 2

6.36

-

5.74

-

-

-

3_059

C6H4CH 2CH 2

6.31

6.32

6.35

6.16

6.81

5.76

3_060

C6H4CH 2CH 2CH2

6.96

7.00

6.57

6.87

7.33

7.28

3_061

C6H11CH2

6.59

7.34

5.57

6.87

7.28

6.17

3_062

2-C10H7 CH2

6.40

-

-

-

-

-

3_063

1,2-Naphtho

5.74

6.31

6.89

5.92

6.28

-

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 3f: R3 R4

N

R5

N H 2N

R1

R2 NH2

N

N

R1

N

R1

R2

R3

R3

N

H 2N

3_069-3_071

3_064-3_068

Compd

R2

NH2

R4

Actual activity

R5

Predicted activity

pc

tg

rl

pc

tg

rl

3_064

CH3

OCH3

H

H

OCH3

-

-

-

-

-

-

3_065

CH3

OCH3

OCH3

OCH3

H

5.69

6.92

6.21

6.09

-

-

3_066

CH3

OCH3

H

OCH3

H

-

-

-

-

-

-

3_067

CH3

H

OCH3

H

OCH3

5.89

5.79

5.94

5.93

-

-

3_068

CH3

H

OCH3

OCH3

OCH3

-

5.08

-

-

-

-

3_069

H

Cl

H

-

-

6.36

6.89

6.04

6.28

6.82

6.33

3_070

Cl

Cl

H

-

-

6.52

6.85

6.51

6.66

6.84

6.27

3_071

Cl

H

Cl

-

-

7.08

7.39

6.92

6.65

6.79

6.43

Table 3g: NH2 N N

H2N

Compd

R

Actual activity

R

Predicted activity

pc

tg

rl

pc

tg

rl

3_072

CH3

7.00

-

8.21

-

-

-

3_073

CH3CH2

-

-

7.74

-

-

6.86

3_074

CH3S

-

-

7.74

-

-

7.71

3_075

C6H5

6.17

-

7.35

-

-

-

3_076

H

-

-

6.55

-

-

6.55

Table 3h: NH2 N H2N

Compd

n

N

(CH2)nAr

N

Actual activity

Ar

Predicted activity

pc

tg

rl

pc

tg

rl

4_077

1

2’,5’-(MeO)2C6H3

6.48

7.52

6.85

6.89

8.2

7.98

4_078

1

3’,4’,5’-(MeO)3C6H3

6.16

7.70

6.66

6.99

7.92

7.28

4_079

1

2’-Br-,3’,4-,5’-(MeO) 3C6H 2

7.29

8.05

7.46

7.01

8.56

7.5

4_080

2

3’,4’,5’-(MeO)3C6H2

5.52

6.49

6.02

7.2

7.81

7.46

4_081

3

3’,4’,5’-(MeO)3C6H2

6.03

6.77

7.03

7.12

7.81

6.79

727

728 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Table 3i: NH2 X

NH2 X

H2N

H 2N

N

5_082-5_087 Compd

X

N

Y

N

N

5_088-5_091 Actual activity

Y

Predicted activity

pc

tg

rl

pc

tg

rl

5_082

CH2NH

2’,5’-(MeO)2

-

-

-

-

-

-

5_083

CH2NH

3’,4’,5’-(MeO)3

-

-

-

-

-

-

5_084

CH2N(Me)

2’,5’-(MeO)2

-

-

5.08

-

-

6.1

5_085

CH2N(Me)

3’,4’,5’-(MeO)3

-

-

5.28

-

-

5.61

5_086

CH2CH2

2’,5’-(MeO)2

5.72

6.23

6.39

5.95

6.41

6.32

5_087

CH2CH2

3’,4’,5’-(MeO)3

5.85

7.03

6.43

5.43

6.89

7.69

5_088

-CCC6H 5

-

-

-

-

-

-

-

5_089

-CHCHC6H 5

-

6.00

6.92

6.82

5.76

6.34

-

5_090

-CH2CH 2C6H 5

-

6.49

7.43

7.29

5.84

7.82

7.32

5_091

-CC C 6H2 (3’,4’,5’MeO)3

-

-

-

-

-

-

-

Table 3j: NH2 N H2N

NH2

Z

N N

H2N

6_092-6_096

Compd

R

R1

NH2 R1

N

N R

Z

N N

N 6_097

Z

H2N

N R

N

6_098-6_103

OMe Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

6_092

H

-

3’,4’,5’-(MeO)3

6.59

8.09

7.52

6.98

8.33

7.97

6_093

Me

-

3’,4’,5’-(MeO)3

7.89

9.33

8.59

7.41

8.09

8.88

6_094

Me

-

4’-Cl

8.21

8.82

8.66

7.17

8.05

8.63

6_095

Me

-

3’-Cl

6.68

8.70

8.17

6.55

8.71

8.72

6_096

Me

-

3’,4’-Cl

8.66

8.01

8.49

7.28

8.22

8.41

6_097

-

-

-

-

6.73

-

-

7.12

-

6_098

H

Me

3’,4’,5’-(MeO)3

8.07

9.13

9.68

8.16

8.35

8.91

6_099

Me

Me

3’,4’,5’-(MeO)3

8.89

10.07

9.12

9.51

8.25

9.36

6_100

Me

Me

3’,4’-Cl

8.00

8.57

8.38

7.92

8.25

8.54

6_101

H

Cl

2’,5’-(MeO)2

8.29

8.52

8.36

8.6

8.39

8.95

6_102

H

Cl

3’,4’,5’-(MeO)3

8.48

9.28

9.23

8.18

8.19

9.39

6_103

Me

Cl

3’,4’,5’-(MeO)3

8.92

9.19

8.92

8.96

9.04

8.81

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 3k: Br NH2

Br

H2N

NH2

OMe

N

N

S

N

OMe

OMe

H2N

OMe

NH2

N

7_104

OMe

H2N

OMe

7_105

7_106

H 2N

OMe OMe NH2

N

N H

OMe

N

N

H2N

Br

OMe

N H

OMe

N 7_108

7_107

Compd

OMe

N

OMe NH2 Cl

OMe

N

OMe S

Br

H N

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

7_104

-

7.68

7.03

-

7.24

6.97

7_105

5.31

6.60

6.55

4.96

7.35

6.6

7_106

4.70

5.60

5.37

5.59

6.45

5.37

7_107

8.55

9.11

9.01

8.14

8.71

8.94

7_108

7.26

8.41

7.48

7.33

8.38

7.76

Table 3l:

R N

NH2 N H2N Compd

X

R

Z N

S

X

Z

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

8_109

Br

H-

3’,4’,5’-(MeO)3

5.89

5.47

5.77

-

6.31

5.82

8_110

Br

H-

2’,5’-(MeO)2

-

-

5.48

-

-

-

8_111

Br

CH3-

3’,4’,5’-(MeO)3

5.51

4.90

5.55

5.3

-

5.85

8_112

Br

CH3-

2’,5’-(MeO)2

-

-

-

-

-

-

8_113

Br

H-

3’,5’-Cl2,4’pyrrol

6.12

5.59

6.00

5.17

5.81

6.07

8_114

H

H-

3’,4’,5’-(MeO)3

-

-

-

-

-

-

8_115

H

H-

2’,5’-(MeO)2

-

-

-

-

-

-

8_116

H

CH3-

3’,4’,5’-(MeO)3

-

-

-

-

-

-

8_117

H

CH3-

2’,5’-(MeO)2

-

-

-

-

-

-

8_118

H

H-

3’,5’-Cl2,4’pyrrol

-

-

-

-

-

-

729

730 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Table 3m: OMe NH2

OMe

n

N OMe H 2N

Compd

n

N

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

9_119

1

-

5.68

5.16

-

5.98

5.26

9_120

2

6.74

7.85

7.29

6.19

7.28

7.23

9_121

3

6.89

7.85

7.66

6.36

7.82

7.67

Table 3n: NH2

H2N

NH2

NH2

N

N

R N

H2N

10_122-10_134

R H2N

N

10_138 NH2

N

N S

N

H2N

O

N

10_139

Compd

S

N

10_135-10_137

NH2

H2N

N

10_140

Actual activity

R

Predicted activity

pc

tg

rl

pc

tg

rl

10_122

H

7.54

8.49

7.74

7.09

8.17

7.87

10_123

2-Me

7.60

8.64

7.96

6.93

8.65

7.43

10_124

3-Me

7.47

8.44

7.96

6.82

8.58

7.78

10_125

2-OMe

7.35

8.85

7.92

7.4

8.06

8.24

10_126

3-OMe

7.57

8.68

7.96

7.3

8.85

8.18

10_127

4-OMe

7.36

8.30

8.11

7.38

8.54

7.53

10_128

3-CF3

7.40

7.85

7.72

7.33

8.62

7.87

10_129

3-OCF3

8.00

8.85

8.10

7.75

8.87

8.01

10_130

4-OCF3

7.24

8.14

7.77

7.5

8.66

8.07

10_131

2’,5’-(MeO)2

8.24

8.68

8.47

7.64

9.48

8.38

10_132

3’,4’-(MeO)2

8.00

8.64

8.20

7.46

8.33

7.99

10_133

3’,4’,5’-(MeO)3

8.04

8.62

8.42

7.07

8.67

7.97

10_134

3,4-diCl

5.82

6.59

6.14

5.61

8.47

7.65

10_135

Et

6.16

6.96

6.52

6.33

7.06

6.58

10_136

t-Bu

7.74

8.74

8.19

6.46

6.91

6.85

10_137

Ph

6.66

6.89

6.72

6.83

6.4

6.26

10_138

-

7.68

8.21

8.10

6.93

8.33

7.96

10_139

-

6.77

8.31

7.89

6.67

8.2

7.92

10_140

-

7.06

8.40

7.57

6.18

7.73

7.77

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Table 3o:

NH2 N

N H 2N

Compd

X

N

N

X

N

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

11_141

CH2CH2

6.85

7.04

-

7.67

6.14

-

11_142

CH2

8.38

8.54

8.57

8.16

7.53

7.58

11_143

O

6.47

6.66

5.89

7.74

7.56

8.01

11_144

S

7.92

7.96

7.70

7.95

7.18

-

11_145

Direct

8.00

8.92

8.26

8.03

8.17

8.24

11_146

None

6.31

6.89

6.40

7.57

7.69

6.95

TMX

-

8.38

9.00

9.52

8.64

8.41

8.78

Table 3p: NH2

NH2

N

NH2

N

H2N

N

H2N

12_148 NH2

N H2N

N 12_149

O

H2N

N

H2N

12_152

Compd

H2N

N 12_151 NH2

NH2

N

N

N N

H2N

S H2N

N 12_154

12_153

N

N N 12_150

NH2

N

NH2

N 12_155

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

12_148

6.21

6.29

5.23

6.43

6.33

5.96

12_149

4.93

6.25

5.37

6.29

6.13

5.42

12_150

5.11

5.41

4.95

6.25

6.34

5.28

12_151

-

5.34

-

-

5.86

-

12_152

4.59

5.47

4.63

6.02

5.8

5.96

12_153

4.75

5.52

5.00

6.07

6.09

4.98

12_154

-

5.00

5.52

-

5.46

5.38

12_155

-

5.70

5.54

-

6.09

5.47

731

732 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Table 3q:

NH2

N

H2N

NH2 X

N

N

Y

NH2

H N

N N

H2N

H2N

13_156, X=CH, Y=N 13_157, X=Y=CH

N

13_158

13_158

Actual Activity

Compd

H N

N

Predicted Activity

pc

tg

rl

pc

tg

rl

13_156

8.37

8.40

7.72

8.09

-

-

13_157

8.43

8.47

8.28

7.48

-

-

13_158

5.89

6.24

6.82

5.91

-

-

13_159

7.29

7.89

8.72

6.64

-

-

Table 3r: NH2 n

N

X H2N

Compd

n

N

X

N H

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

14_160

0

H

-

5.92

6.04

-

5.78

5.44

14_161

0

4-Cl

-

4.95

5.21

-

5.56

5.3

14_162

0

3,4-Cl2

5.48

5.80

5.64

5.17

5.62

5.47

14_163

0

3,4,5-(OMe)3

6.08

5.85

5.57

5.46

5.67

5.29

14_164

1

H

-

-

4.57

-

-

5.92

Table 3s: NH2

OMe

N H 2N

N O(CH2)nR

Compd

n

15_165

Actual activity

R

TMP

Predicted activity

pc

tg

rl

pc

tg

rl

5.92

6.57

4.89

6.49

6.83

7.11

15_166

4

Me

5.72

6.19

5.36

7

7.18

5.95

15_167

5

Me

5.85

6.04

5.54

6.36

6.97

5.6

15_168

6

Me

6.25

6.30

5.96

6.64

7.01

6.17

15_169

7

Me

5.36

5.82

5.52

6.39

7.28

5.51

15_170

8

Me

5.29

5.49

5.07

6.52

6.25

5.28

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

733

(Table 3s) Contd…. Compd

n

R

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

15_171

3

-COOH

7.60

7.74

6.59

7.6

7.8

7.06

15_172

4

-COOH

8.31

7.96

6.41

7.1

7.89

6.5

15_173

5

-COOH

7.10

7.32

5.77

7.56

7.46

5.91

15_174

6

-COOH

6.59

7.74

5.92

6.73

7.98

6.11

15_175

7

-COOH

6.15

7.44

5.92

6.59

7.35

5.73

15_176

8

-COOH

6.32

7.37

5.92

6.25

7.41

6.11

Table 3t: NH2

NH2

N H2N

H2N

16_177-16_182

Compd

N

X N

N

NH2 n

N

N

X

N H

H2N

X N 16_186-16_198

16_183-16_185

Actual Activity

X

Predicted Activity

pc

tg

rl

pc

tg

rl

16_177

PTX

8.89

9.37

10.48

8.77

9.19

8.64

16_178

2,5-(OMe)2

7.55

8.47

8.27

8.05

8.99

7.99

16_179

3,4-(OMe)2

7.62

8.19

7.68

7.61

8.59

8.06

16_180

3,5-(OMe)2

7.68

8.44

7.43

8.19

8.79

8.16

16_181

2-Cl

7.29

8.77

8.14

8.01

8.53

8.13

16_182

4-F

7.28

8.40

7.89

7.13

8.73

7.92

16_183

3,4-Cl2, n=1

5.54

6.48

6.02

6.31

6.1

5.68

16_184

3,4,5-(OMe)3, n=1

5.14

5.85

5.28

5.82

5.98

5.27

16_185

3,4,5-(OMe)3, n=2

7.11

8.43

7.70

6.67

7.59

7.33

16_186

2-OMe

7.51

8.96

8.27

6.64

8.63

8.4

16_187

3-OMe

7.70

9.00

8.42

7.06

7.78

8.34

16_188

4-OMe

7.64

8.74

8.00

7.21

8.54

8.22

16_189

2,5-(OMe)2

7.96

9.00

9.03

7.5

8.82

9.01

16_190

2,4-(OMe)2

8.04

8.89

8.55

7.48

8.5

7.66

16_191

3,5-(OMe)2

7.96

9.00

8.52

7.19

8.78

8.51

16_192

3,4,5-(OMe)3

7.82

8.77

8.38

7.88

8.36

8.45

16_193

2-Cl

7.96

9.19

8.96

6.78

8.5

7.81

16_194

3-Cl

7.77

8.80

8.40

7.07

8.02

8.12

16_195

4-Cl

7.82

8.66

8.40

6.9

8.19

8.15

16_196

3,4-Cl2

8.30

8.25

8.55

7.89

7.87

8.07

16_197

4-F

7.17

8.59

8.17

7.1

8.53

7.88

16_198

a

8.06

8.66

8.82

7.53

8.25

8.3

a=2-napthylmethyl

734 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Table 3u: NH2

O

NH2

N H 2N

NH2

O N

N N

H 2N

O

N

H 2N

N

O

(CH2)xCO2H (CH2)xCO2H

X

17_199-17_201

Compd

17_204-17_205

17_202-17_203

X

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

17_199

2

7.89

8.01

6.40

7.77

7.65

6.63

17_200

3

8.55

8.49

6.66

8.1

7.88

6.77

17_201

4

7.06

8.14

5.60

7.24

7.72

6.29

17_202

4

7.82

8.24

6.49

7.37

8.02

6.84

17_203

5

7.82

9.08

6.39

7.07

7.67

6.48

17_204

3-COOH

7.05

7.22

5.72

6.86

7.13

5.27

17_205

4-COOH

6.92

6.70

5.68

7.73

6.9

5.71

Table 3v: NH2 NH2 N H 2N

N

OMe

N N

N

H 2N

OMe

N

N

X 19_213

18_206-18_212

Compd

X

COOH

Actual Activity

Predicted Activity

pc

tg

rl

pc

tg

rl

18_206

H

8.10

8.41

6.52

7.81

-

7.00

18_207

OH

8.51

8.59

8.39

7.84

7.69

-

18_208

OCH2CO2 H

8.68

8.77

7.55

8.33

-

-

18_209

O(CH2)2 CO2H

8.62

8.55

6.96

8.87

-

7.00

18_210

O(CH2)3 CO2H

9.96

9.00

6.82

8.9

-

-

18_211

O(CH2)4 CO2H

8.96

8.68

6.89

8.88

-

-

18_212

OCH2C6H 4 (4-CO 2H)

10.00

8.66

7.24

-

-

7.4

6.92

7.38

5.06

6.88

7.28

4.69

19_213

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

735

Table 3w:

NH2

NH2

O N

N H2N

N

H2N

O

N X

(CH2)6CO2H

Y

20_214

Compd

X

20_215-20_225

Y

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

20_214

-

-

7.28

8.52

6.34

6.91

7.59

6.41

20_215

OCH2

3-CO2H

7.05

7.22

5.72

7.02

7.44

5.19

20_216

OCH2

4-CO2H

6.92

6.70

5.68

6.92

6.98

5.7

20_217

CCCH2O

2-CO2H

6.64

7.30

6.21

7.14

7

6.3

20_218

CCCH2O

3-CO2H

6.25

6.80

5.85

6.73

7.11

5.9

20_219

CHCHCH2O

2-CO2H

6.07

6.60

6.19

7.39

7.5

6.56

20_220

(CH2)3O

2-CO2H

6.11

7.15

5.60

5.83

7.78

5.95

20_221

(CH2)3O

3-CO2H

6.96

7.68

6.40

7.14

7.56

6.78

20_222

CC

2-CO2H

6.39

6.72

5.60

6.71

7.37

5.76

20_223

CC

3-CO2H

8.64

9.26

7.19

6.85

8.22

6.75

20_224

CC

4-CO2H

6.89

7.47

6.09

6.76

8.03

5.79

20_225

CH2CH2

4-CO2H

6.15

6.82

5.89

6.43

6.95

5.55

contains one hydrogen bond acceptor (A) mapping on the lone pair of vectors of 1-nitrogen atom of 2,4-diamino substituted heterocycle, two hydrogen bond donor (D) groups, one corresponding to the polar hydrogen of 2,4-diamino groups. One positive ionic feature was mapped on 3-nitrogen atom of 2,4-diamino substituted heterocycle while two hydrophobic features map on the distal substitutions. Tg-CPH3 model associated with 2 PLS factors, was found to be the best for tgDHFR (Table 5). The model Tg-CPH3 showed good correlation for training set molecules (r2 = 0.74, SD = 0.65, F = 184.6, N = 130); furthermore, the model also showed good predictive power for test set (Q2 = 0.65, RMSE = 0.75, Pearson-R = 0.82, N = 56). Fig. 18 shows the linearity trend in the graphs of actual vs. predicted activity for training and test set molecules for tgDHFR. Hence, the hypothesis Tg-CPH3 (Fig. 19) with one hydrogen bond acceptor (A), two hydrogen bond donors (D), two aromatic rings (R) and one positive ionic group (P) as pharmacophoric features was retained for further studies. The pharmacophore contains one hydrogen bond acceptor (A) mapping on the lone pair of vectors of 1-nitrogen atom of 2,4-diamino substituted heterocycle, two hydrogen bond donor (D) groups one corresponding to the polar hydrogen of 2,4-diamino groups, One positive ionic feature was mapped on 3-nitrogen atom of 2,4-diamino substituted heterocycle; two aromatic ring features, one mapping on fused aromatic ring and other on the distal aromatic substitution. Rl-CPH5 model associated with 5 PLS factors, was found to be the best for rlDHFR (Table 6). The model Rl-CPH5 showed excellent correlation for training set molecules (r2 = 0.91, SD = 0.42, F = 263.1, N = 144); furthermore, the model also showed good predic-

tive power for test set (Q2 = 0.66, RMSE = 0.79, Pearson-R = 0.82, N = 61). Fig. 20 shows the linearity trend in the graphs of actual vs. predicted activity for training and test set molecules for rlDHFR. Hence, the hypothesis Rl-CPH5 (Fig. 21) with one hydrogen bond acceptor (A), two hydrogen bond donors (D), two aromatic rings (R) and one positive ionic group (P) as pharmacophoric features was retained for further studies. The pharmacophore contains one hydrogen bond acceptor (A) mapping on the lone pair of vectors of 1-nitrogen atom of 2,4-diamino substituted heterocycle, two hydrogen bond donor (D) groups one corresponding to the polar hydrogen of 2,4-diamino groups, one positive ionic feature was mapped on 3-nitrogen atom of 2,4-diamino substituted heterocycle; two aromatic ring features, one mapping on fused aromatic ring and other on the distal aromatic substitution. 3.2.4. Implication in Drug Design One of the major advantages of employing conformers generated in receptor active site using molecular docking algorithms, for pharmacophore mapping and subsequent 3D-QSAR model building is that, the generated pharmacophore could be mapped back to receptor active site to decipher and understand the pharmacophore sites in context of ligand-receptor interactions. Furthermore, the regression cubes generated using PLS analysis could be visualized in 3D space in the receptor active site, which could provide better visualization of active site and SAR.116-118 The PLS regression cubes generated using PHASE 3D-QSAR model can be visualized in two ways, first in the context of QSAR model and second in the context of the ligand included in the workspace. The PLS regression cubes generated in context of QSAR model are constant for all ligands and give the idea of which 3D space when occupied by

736 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Table 3x: NH2 N

O

H2N

N

(CH2)3CO2H

N

H2N

O

21_227-21_229

21_230 21_234-21_235

NH2

O

N N

H2N

NH2

OMe

N

N

N

N

21_231-21_233

NH2 N

N

H2N

N

N

X (CH2)nCO2H

21_226

N

(CH2)nCO2H X

NH2

O

N N

H2N

NH2

O

Br

H2N

N

N O(CH2)nCO2H

X(CH2)nCO2H

(CH2)2CO2H 21_239

21_236-21_238

NH2

O

CH2Ph

NH2

N N

H2N

N

H2N

N

n

N

N Y

O

21_243

X

X

N

O

Compd

NH2

O(CH2)3CO2Et

N N

H2N

21_240-21_242

21_244

Y

21_245-21_246

Actual activity

Predicted activity

pc

tg

rl

pc

tg

rl

21_226

-

-

-

10.00

8.47

6.30

8.33

8.2

7.09

21_227

OCH2

2

-

10.37

9.72

9.66

9.74

8.89

9.81

21_228

OCH2

3

-

10.29

9.64

9.66

10.53

9.95

9.72

21_229

OCH2

4

-

9.89

9.70

9.70

9.64

9.6

9.43

21_230

OCH2

2

-

10.77

9.64

9.04

10.09

9.79

8.85

21_231

CC

2

-

11.01

10.34

10.49

10.77

10.55

9.96

21_232

CC

3

-

9.85

9.37

9.24

9.86

7.06

9.86

21_233

CC

4

-

9.74

9.49

9.40

10.49

10.17

9.56

21_234

CC

2

-

10.19

9.39

8.26

10.34

9.83

9.06

21_235

CC

3

-

9.92

9.24

8.62

10.19

9.59

8.85

21_236

CC

1

-

9.03

8.00

7.13

8.91

7.21

6.42

21_237

CC

2

-

7.64

8.11

7.25

8.25

21_238

CC

3

-

8.92

8.68

6.89

8.32

9.05

6.24

21_239

-

-

-

8.15

8.82

7.20

8.57

8.48

6.97

21_240

-

3

-

9.19

9.47

9.38

8.81

8.73

9.41

21_241

-

4

-

8.55

7.95

8.85

8.89

9.05

8.61

21_242

-

5

-

8.46

8.89

8.68

8.54

8.42

8.72

21_243

-

-

-

8.02

8.80

7.77

8.42

8.61

8.38

21_244

-

-

-

8.08

9.00

8.03

9.21

9.28

7.99

21_245

OMe

-

I

9.24

9.39

9.30

8.92

8.72

9.41

21_246

I

-

OMe

9.42

9.24

9.40

9.04

8.84

8.87

6.55

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

737

Table 4: Summary of QSAR results for five best common pharmacophore hypotheses (CPHs) for pcDHFR inhibitory activity Pc-CPH1 ADDHHP

Pc-CPH2 AADDHH

Pc-CPH3 AADDHP

Pc-CPH4 DDHHRR

Pc-CPH5 DDHHPR

Survival score

14.38

13.69

13.71

13.77

13.71

SD

0.58

0.54

0.57

0.52

0.54

2

0.80

0.83

0.81

0.84

0.83

F

196

232.2

210.2

251

230.4

r

P

9.83e

RMSE Q

2

Pearson-R

-50

0.85

6.26e

-54

0.96

1.89e

-51

0.96

1.10e

-55

0.97

9.81e-54 0.75

0.57

0.44

0.44

0.43

0.51

0.76

0.67

0.66

0.64

0.72

Table 5: Summary of QSAR results for five best common pharmacophore hypotheses (CPHs) for tgDHFR inhibitory activity Tg-CPH1 ADDPRR

Tg-CPH2 ADDPRR

Tg-CPH3 ADDPRR

Tg-CPH4 ADDHPR

Tg-CPH5 ADDHPR

Survival score

13.07

13.06

13.04

13.6

12.75

SD

0.65

0.66

0.65

0.78

0.82

0.73

0.74

0.74

0.63

0.59

F

178.5

179.8

184.6

115

96.2

P

1.26e-37

9.05e-38

2.61e-38

1.77e-29

2.27e-26

RMSE

0.85

0.82

0.75

0.93

0.86

0.56

0.59

0.65

0.47

0.54

0.75

0.77

0.82

0.68

0.73

r

2

Q

2

Pearson-R

Table 6.

Summary of QSAR Results for Five Best Common Pharmacophore Hypotheses (CPHs) for rlDHFR Inhibitory Activity Rl-CPH1 ADDHPR

Rl-CPH2 ADDHRR

Rl-CPH3 DDHHPR

Rl-CPH4 ADDRRR

Rl-CPH5 ADDPRR

Survival score

13.94

12.64

12.6

12.63

13.97

SD

0.43

0.44

0.42

0.42

0.42

r2

0.90

0.89

0.90

0.90

0.91

F

254.2

232.5

266.5

266.2

263.1

P

4.49e-68

1.14e-65

2.32e-69

5.11e-69

1.06e-68

RMSE

0.80

0.88

0.90

0.90

0.79

Q2

0.64

0.57

0.56

0.52

0.66

Pearson-R

0.81

0.76

0.75

0.73

0.82

certain type of ligand atom can lead to increase or decrease in the activity, thus these cubes give an idea of receptor active site topology. On the other hand, the cubes generated in the context of ligand are specific for each ligand and suggest which groups of ligand are responsible for observed (high-low) activity, thus these cubes can help in understanding the SAR in more lucid way. The PLS regression cubes could be generated for different properties such as hydrophobic, hydrogen bond donor, hydrogen bond acceptor, electron

withdrawing, positive and negative ionic features, which define the non-covalent interactions with receptor. The majority of compounds used in this study are either monocyclic or bicyclic with 6-6 or 6-5 fused ring system. Few compounds have tricyclic or tetracyclic ring system. In general, the compounds could be represented by various domains depicted in Fig. 22. Thus, most of the DHFR inhibitors have a 2,4-diaminopyrimidine ring system (or 2,4-diaminotriazine ring system) “A”,

738 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (16). Scatter plots for the predicted and experimental pIC50 values for the pcDHFR 3 PLS factor 3D-QSAR model applied to training set and test set.

Fig. (17). Geometry of the pharmacophore for developed for pcDHFR; red sphere with vector shows acceptor feature, blue sphere with vector shows donor feature, orange torus shows the aromatic ring feature and green sphere shows hydrophobic feature.

Fig. (18). Scatter plots for the predicted and experimental pIC50 values for the tgDHFR 2 PLS factor 3D-QSAR model applied to training set and test set.

either monocyclic or fused with various 6 or 5 membered ring “B”. The bridge “C” joins the distal substitution “D” with “A-B” ring systems. Active Site Analysis The DHFR enzyme is highly conserved among various species, thus it shares high sequence similarity and overall fold conservation across the species. However, there are subtle differences between human DHFR and microbial DHFR, making selective inhibitor design possible. Comparison of amino acid residues in pcDHFR,

tgDHFR and human DHFR was made in order to understand the specific differences (Table 7). Thus, in spite of overall sequence similarity and fold conservation, the specific differences mentioned in Table 7, could be explored in quest of specific inhibitors. For instance, bulky Phe35 in human DHFR is replaced by less bulky Ile33 (Maroon colour); Lys35, Lys68, Ile60 of humanDHFR are replaced by Ser36, Val95, Met87 in tgDHFR respectively (black and red colour). Furthermore, it is interesting to note that the hydrophilic Asn64 in human DHFR

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

739

Fig. (19). Geometry of the pharmacophore for developed for tgDHFR; red sphere with vector shows acceptor feature, blue sphere with vector shows donor feature and orange torus shows the aromatic ring feature.

Fig. (20). Scatter plots for the predicted and experimental pIC50 values for the rlDHFR 5 PLS factor 3D-QSAR model applied to training set and test set.

Table 7.

Differences in Amino Acid Residues for Active Sites of pcDHFR, tgDHFR and Human DHFR

Position

Human DHFR

pc DHFR

tg DHFR

4

10:VAL

13:LEU

11:MET

16

31:PHE

33:ILE

32:PHE

18

33:TYR

35:TYR

34:HIS

20

35:GLN

37:LYS

36:SER

29

60:ILE

65:ILE

87:MET

31

64:ASN

69:PHE

91:PHE

33

68:LYS

73:LYS

95:VAL

53

134:PHE

142:MET

170:TYR

(dark blue colour), located just outside the binding site, is replaced by hydrophobic Phe69 in pcDHFR, Phe91 in tgDHFR. This can be explored for the design of selective compounds. Fig. 23 shows the 3D comparative differences in the active site along with one of the potent inhibitor 21_231. Interpretation of Developed Models Molecular docking studies revealed many conserved interactions of inhibitors across various species; nevertheless, there are

certain specific interactions which could be exploited for selective inhibitor design. The docked poses for most active compound, 21_231, along with developed pharmacophore for pcDHFR, tgDHFR and rlDHFR are shown in Fig. 23. The analysis of docked poses selected using predictive pharmacophore model revealed that, the 2,4-diaminosubstituted ring system (A-B) of all compounds positions itself at approximately the same position in all DHFR species, i.e. at the bottom of the active site and shows conserved hydrogen bonding interactions. For instance, in most active compound, 21_231, the 2,4diamino groups of pyrido[2,3-d]pyrimidine ring form hydrogen bonding interactions with Glu32, Ile10 and Ile123 in pcDHFR; with Asp31, Val8 and Val151 in tgDHFR and with Glu30, Ile7 and Val115 in humanDHFR (Fig. 24). The developed pharmacophores for pcDHFR, tgDHFR and humanDHFR also reveal the presence of two donor groups (blue spheres with vectors) corresponding to 2,4diamino groups (Fig. 24). All three pharmacophores contain a hydrogen bond acceptor group, depicted by a red sphere with vector, mapping on the lone pair of electrons on 1-nitrogen and a positive ionic feature, mapping on 3-nitrogen of the pyrimidine ring. In terms of receptor interactions, the 1-nitrogen is important for salt bridge formation with conserved acid residue for e.g. Glu32, Asp31 and Glu30 in pcDHFR, tgDHFR and human DHFR; whereas no direct interactions were observed between the 3-nitrogen of pyrimidine ring and receptor active site. However, this nitrogen is essential for the basicity of pyrimidine ring system [119]. The developed pharmacophores here are in well agreement with the mechanism of reaction of small molecules inhibition catalyzed by DHFR.

740 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (21). Geometry of the pharmacophore for developed for rlDHFR; red sphere with vector shows acceptor feature, blue sphere with vector shows donor feature and orange torus shows the aromatic ring feature.

NH2

N

A H2N

B

N

Fig. (22). General domains of a typical DHFR inhibitor.

Fig. (23). Key differences in the active site of pcDHFR, tgDHFR and human DHFR.

C

D

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

741

Fig. (24). Developed pharmacophore models for pcDHFR, tgDHFR and rlDHFR, mapped on the most active compound 21_231, in the active site of enzyme. Enzyme is shown in brown colored line model and ligand is depicted in cyan stick model. Hydrogen bonding of 2,4-diamino groups with active site residues depicted in doted black lines is noticeable.

The highly active compounds of the series map all the functional features of the best predictive hypotheses, whereas the less active compounds map only a few of the features (Fig. 25). For instance in case of pcDHFR the least active molecule like, 11_152, is unable to map on two distal hydrophobic features, while in case of tgDHFR the least active molecule, 14_161, is unable to map on distal aromatic ring feature and in case of rlDHFR compound, 3_042 is also unable to map on distal aromatic ring feature, thus indicating a reason for observed low potency of these compounds. These mapping studies provide confidence in the ability of pharmacophores to differentiate between active and inactive ligands. The generated PLS cubes for hydrogen bond donor group are depicted in Figs. 26-28. In these figures, light blue color represents the regions favored for activity, while light orange color represents the regions disfavored for activity. Fig. 26 represent the PLS cubes generated in context of QSAR model, which provide an idea of regions of active site favored/disfavored for hydrogen bond donor groups. Hydrogen bong donor favored light blue colored cubes were visible in the vicinity of 2,4-diamino groups substituted on the nitrogen ring system. Thus the PLS cubes also revealed the importance of 2,4-diamino groups for pcDHFR, tgDHFR and rlDHFR inhibitory potential. Inspection of the PLS cubes also reveal the presence of few light orange colored cubes along with the light blue colored cubes in the vicinity of 2,4-diamino groups, which means

not only the presence but also proper orientation of 2,4-diamino groups is important for the activity. PLS cubes in the context of QSAR model also revealed the presence of both hydrogen bond disfavored light orange cubes and hydrogen bond favored light blue cubes in the bridge part “C” of molecule in case of pcDHFR, tgDHFR and rlDHFR (Fig. 26). However, rlDHFR showed clear predominance of favorable light blue cubes in bridge part, this observation can be used judiciously to obtain selective ligands for microbial DHFR. Furthermore, hydrogen bond donor favored light blue cubes were seen near the 5´ and 6-position of distal aromatic substitution “D”, which can be used to get tgDHFR selective ligands. PLS cubes visualization in context of most active ligands like, 21_231, revealed predominance of light blue cubes in the vicinity of 2,4-diamino groups, indicating the positive contribution of these groups towards activity (Fig. 27). For the cubes generated in context of QSAR model (Fig. 28), distinct hydrogen bond disfavored light orange colored cubes were visible near the 8th position in 6-6 fused ring or 7th position in 6-5 fused ring system. Visualization of PLS regression cubes in context of one or more inactive ligand clearly support this observation, for instance, when the cubes generated for hydrogen bond donor groups were visualized in context of inactive ligand, 16_184 for pcDHFR, 14_161 for tgDHFR and 14_164 for rlDHFR; unfavourable orange cubes were seen near -NH- at 7th position in fused ring system, this could be one of the reasons for observed lower activity

742 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (25). Developed pharmacophore models for pcDHFR, tgDHFR and rlDHFR, mapped on the most inactive ligands, 11_152 for pcDHFR, 14_161 for tgDHFR and 3_042 for rlDHFR, in the active site of enzyme. Enzyme is shown in brown colored line model and ligand is depicted in cyan stick model. Inability of these compounds to map on distal pharmacophoric features is noticeable.

Fig. (26). Pictorial representation of the contours generated for hydrogen bond donor property in context of QSAR model, light blue cubes indicate favorable regions while orange cube indicates unfavorable region for the activity.

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

743

Fig. (27). Pictorial representation of the contours generated for hydrogen bond donor property for most active ligand 21_321, light blue cubes indicate favorable regions while orange cube indicates unfavorable region for the activity.

of these compounds. Thus, relatively lower activity of compounds, 14_160 to 14_164 and 16_183-16_184, could be attributed to presence of hydrogen bond donor group at 7th position in fused ring system. The ring “B” is involved in - stacking interaction with Phe36 (centroid-centroid distance 4.50Å) in pcDHFR; with Phe35 (centroid-centroid distance 4.58Å) in tgDHFR; with Phe35 (centroidcentroid distance 3.87Å) in human DHFR. The - stacking interaction is stronger in human DHFR. The developed pharmacophore also revealed presence of aromatic ring (orange torus, R15) in the “B” region for tg and rl DHFR. Thus bicyclic molecules are more likely to have good human DHFR activity thereby decreasing the selectivity. This observation is in well agreement with the fact that, the bicyclic molecules like MTX, TMX, and PTX are potent human DHFR inhibitors, whereas monocyclic molecule such as TMP is selective for microbial DHFR. Visualization of QSAR model cubes associated with hydrophobic property give approximate idea of receptor site topology. Fig. 29, show cubes generated for hydrophobic property using QSAR model, in this figure green cubes represent the sterically favored spatial regions for activity while magenta cubes represent the sterically disfavored regions; for better visualization the most active ligand 21_231 is included. Sterically favored green cubes were visible above the 5th position ring “B”. The molecular docking studies revealed that small hydrophobic substitution such as -CH3 (3_027), -F (3_029), -Cl (3_030), -Br (3_031) and -I (3_032) at 5th position “B” ring forms favorable hydrophobic interaction with Ile 123 in pcDHFR, Val151 in tgDHFR and Val115 in human DHFR. However, bigger substitutions such as, -OCH2CF3 (3_037) and OCH3CF7 (3_038), could lead to steric clash with Ile 123 in pcDHFR and there by leading to decrease in activity. The space

above the 5th position ring “B” in human DHFR appears to be somewhat greater than in pcDHFR as Val15 is present in human DHFR instead of Ile123; hence, even compounds with bulkier substitutions show good activity in human DHFR. Thus, compounds (5_089, 5_090, 7_105 to 7_108, 9_120, 9_121) with bulkier substitution at 5th position show greater activity against rlDHFR there by leading to decreased selectivity. These compounds also show good activity against tgDHFR with Val151. Thus, selection of optimal substitution at 5th position could lead to selective compounds. Ring “D” binds in the hydrophobic pocket formed by; Ser64, Ile65, Pro66, Leu72, Lys37, Phe36 and Ile33 in pcDHFR; in Leu23, Met87, Pro88, Phe91, Leu94, Ser36 and Phe32 tgDHFR and Leu22, Phe31, Ile60, Pro61and Leu67 in rlDHFR. The developed pharmacophores also revealed importance of hydrophobic feature corresponding to ring “D” in case of pcDHFR; while aromatic ring feature corresponding to ring “D” was found to be important for tgDHFR and rlDHFR inhibitory activity. In addition to a hydrophobic feature, corresponding to ring “D” in pcDHFR, second hydrophobic feature was found to be important in the distal part of molecule. Thus, active site of pcDHFR appears to be more tolerable to bulk at distal part. This observation has been explored for designing pcDHFR inhibitors by Rosowasky et al. and very recently in our laboratory by using bulky tricyclic moieties as distal substitution. The QSAR model generated cubes revealed presence of pronounced green cubes in the vicinity of distal aromatic ring “D”. Surfaces generated for hydrophobic property of the enzyme active site also reveal the hydrophobic nature (brown color) of the distal binding pocket and the hydrophilic nature (light blue color) of the binding pocket at bottom, where the 2,4-diamino groups bind (Fig. 30).

744 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (28). Pictorial representation of the contours generated for hydrogen bond donor property for some of the inactive ligands, 16_184 for pcDHFR, 14_161 for tgDHFR and 14_164 for rlDHFR, light blue cubes indicate favorable regions while light orange cube indicates unfavorable region for the activity.

Fig. (29). Pictorial representation of the activity cubes generated for hydrophobic property using QSAR model. In these figures, the green cubes indicate favorable regions while magenta cube indicates unfavorable region for the activity.

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

745

Fig. (30). Pictorial representation of the lipophilic, hydrophilic surfaces generated for pcDHFR, tgDHFR and rlDHFR. In this figure, brown color represents lipophilic regions, while light blue surfaces represent hydrophilic regions of enzyme active site.

Direct interactions were not observable with the bridge “C” in majority of the molecules with the active site; it mainly affects the orientation of ring “D” (Fig. 29). However, selection of an appropriate bridge is crucial factor in determining the selectivity and potency. Visualization of PLS generated cubes for hydrophobic property for one of the most active ligands, 21_231 (Fig. 31), revealed the presence of pronounced green cubes in the vicinity of distal aromatic substituted ring “D”. In of pcDHFR QSAR model cubes, (Fig. 29), distinct hydrophobic disfavored magenta cubes were visible near the 1-nitrogen, amino group at 2nd position, over bridge “C” part and near the 7th position substitution. While in case of tgDHFR QSAR model cubes, (Fig. 29), distinct hydrophobic disfavored violet cubes were visible near the 7-8th position substitution, and few in distal pocket. Visualization of PLS generated cubes for hydrophobic property in context of one or more inactive ligands in the series clearly supports this observation (Fig. 32). For e.g. in case of inactive molecule 7_106, in pcDHFR the bromo substitution on distal ring “D” orients in unfavourable magenta cubes, furthermore, the distal substitution at 5th position causes dislocation of the 2,4-diaminothieno[2,3-d]pyrimidine ring in binding pocket, thus leading to orientation of 1-nitrogen, amino group at 2nd position, in unfavourable region. Hence, compounds

with 6-5 fused ring system having bulky distal aromatic substitution on 5th position (1_007-1_008, 7_105, 7_106, 9_120, 9_121, 14_162 and 16_183-16_185), in general show less activity in pcDHFR compared to tgDHFR and rlDHFR. In case of less active molecule 12_154 in tgDHFR, the tricyclc moiety orients in unfavourable magenta region, (Fig. 32), thereby justifying the observed low activity. Thus, proper orientation of distal hydrophobic region in the hydrophobic binding pocket is important for good DHFR inhibitory activity and selectivity. Figures 33-34 show the cubes generated for negative and positive ionic groups, in these figures, red color represents the regions favored for negative ionic features while yellow color represents regions disfavored for positive ionic groups. Both, the cubes generated in context of QSAR model and for various ligands have similar interpretation. Hence, here cubes generated for various ligands are discussed. Red colored cubes were visible near the terminal carboxylic acid group in most active ligands like 21_226-21_246, implying the importance of negatively charged feature in distal moiety (Fig. 33). In terms of receptor-ligand interactions, it was observed that, the terminal carboxylic acid group in these molecules is involved in ionic interactions with basic residues, Arg75, Arg97 and Arg70, in pcDHFR, tgDHFR and human DHFR respectively. Positive charged unfavourable yellow regions where visible in the

746 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Degani et al.

Fig. (31). Pictorial representation of the contours generated for hydrophobic property for the most active ligand, 21_321, green cubes indicate favorable regions, while magenta cube indicate unfavorable region for the activity.

Fig. (32). Pictorial representation of the contours generated for hydrophobic property for inactive ligand, 7_106 for pcDHFR and 12_154 for tgDHFR, green cubes indicate favorable regions, while cube magenta indicate unfavorable region for the activity.

Comprehensive Evaluation of New Models

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

747

Fig. (33). Pictorial representation of the contours generated for negative ionic feature for the most active ligand, 21_321, red cubes indicate favorable regions.

Fig. (34). Pictorial representation of the contours generated for positive ionic feature for one of the low active ligand, 4_080, yellow cubes indicate unfavorable regions.

748 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

bridge region for the low active compounds like 4_080, indicating the structural feature disfavored for the activity (Fig. 34). 4. CONCLUSION A survey of molecular modeling reports, for rational drug design of DHFR inhibitors active against opportunistic microorganisms including pc, tg and ma, was carried out, which provided an understanding of specific interactions required for inhibition of microbial DHFR. Considering the availability of plethora of structural information and data of active and inactive ligands, a comprehensive evaluation of new models was undertaken. In absence of experimentally determined crystal structure of tgDHFR, the generated homology model was used. The structure thus generated, was further refined using 1 nanosecond molecular dynamics simulation. Thereafter, a combination of structure based and ligand based approaches was employed for pcDHFR, tgDHFR and rlDHFR to decipher structure-chemical-feature based pharmacophores. Thus, the putative bioactive conformers obtained from molecular docking studies, were used as starting conformational ensembles for pharmacophore based 3D-QSAR studies. The developed pharmacophore models are in well agreement with the reaction catalyzed by DHFR. Furthermore, the pharmacophores along with PLS regression cubes for various properties such as hydrogen bond donor, hydrophobic, positive and negative ionic features, help to decipher SAR in more lucid way. Taken together, these studies unearth many structural features, which could be considered for the design of potent and selective DHFR inhibitors. 5. ACKNOWLEDGMENTS Nilesh R. Tawari is thankful to Department of Biotechnology (DBT), India and Seema Bag is thankful to University Grant Commission (UGC), India for financial support. 6. REFERENCES [1]

[2] [3] [4]

[5]

[6]

[7]

[8]

[9]

[10]

Schnell JR, Dyson HJ, Wright PE. Structure, dynamics, and catalytic function of dihydrofolate reductase. Annu Rev Biophys Biomol Struct 2004; 33: 119-40. Klepser ME, Klepser, TB. Drug Treatment of HIV-Related Opportunistic Infections. Drugs 1997; 53: 40-73. Roth B, Flaco EA, Hitchings GH, Bushby SRM. 5-benzyl-2,4diaminopyrimidines as antibacterial agents. I. synthesis and antibacterial activity in vitro. J Med Chem 1962; 5: 1103-23. Gangjee A, Shi J, Queener SF, Barrows L, Kisliuk RL. Synthesis of 5-methyl-5-deaza nonclassical antifolates as inhibitors of dihydrofolate reductases and as potential antipneumocystis, antitoxoplasma, and antitumor agents. J Med Chem 1993; 36: 3437-43. Gangjee A, Mavandadi F, Queener SF, McGuire JJ. Novel 2,4diamino-5-substituted-pyrrolo[2,3-d]pyrimidines as classical and nonclassical antifolate inhibitors of dihydrofolate reductases. J Med Chem 1995; 38: 2158-65. Gangjee A, Zaveri N, Kothare M, Queener SF. nonclassical 2,4diamino-6-(aminomethyl)-5,6,7,8-tetrahydroquinazoline antifolates: synthesis and biological activities. J Med Chem 1995; 38: 3660-8. Gangjee A, Zhu Y, Queener SF, Francom P, Broom AD. Nonclassical 2,4-diamino-8-deazafolate analogues as inhibitors of dihydrofolate reductases from rat liver, pneumocystis carinii, and toxoplasma gondii. J Med Chem 1996; 39: 1836-45. Gangjee A, Vasudevan A, Queener SF, Kisliuk RL. 2,4-Diamino5-deaza-6-substituted pyrido[2,3-d]pyrimidine antifolates as potent and selective nonclassical inhibitors of dihydrofolate reductases. J Med Chem 1996; 39: 1438-46. Gangjee A, Devraj R, Queener SF. Synthesis and dihydrofolate reductase inhibitory activities of 2,4-diamino-5-deaza and 2,4diamino-5,10-dideaza lipophilic antifolates. J Med Chem 1997; 40: 470-8. Gangjee A, Vasudevan A, Queener SF. Synthesis and biological evaluation of nonclassical 2,4-diamino-5-methylpyrido[2,3d]pyrimidines with novel side chain substituents as potential inhibitors of dihydrofolate reductases. J Med Chem 1997; 40: 479-85.

Degani et al. [11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

Gangjee A, Mavandadi F, Queener SF. Effect of N9-methylation and bridge atom variation on the activity of 5-substituted 2,4diaminopyrrolo[2,3-d]pyrimidines against dihydrofolate reductases from Pneumocystis carinii and Toxoplasma gondii. J Med Chem 1997; 40: 1173-77 Gangjee A, Shi J, Queener SF. Synthesis and biological activities of conformationally restricted, tricyclic nonclassical antifolates as inhibitors of dihydrofolate reductases. J Med Chem 1997; 40: 1930-6. Gangjee A, Guo X, Queener SF, et al. Selective Pneumocystis carinii dihydrofolate reductase inhibitors: design, synthesis, and biological evaluation of new 2,4-diamino-5-substituted-furo[2,3d]pyrimidines. J Med Chem 1998; 41: 1263-71. Gangjee A, Vidwans AP, Vasudevan A, et al. Structure-based design and synthesis of lipophilic 2,4-diamino-6-substituted quinazolines and their evaluation as inhibitors of dihydrofolate reductases and potential antitumor agents. J Med Chem 1998; 41: 342634. Gangjee A, Zhu Y, Queener SF. 6-Substituted 2,4diaminopyrido[3,2-d]pyrimidine analogues of piritrexim as inhibitors of dihydrofolate reductase from rat liver, Pneumocystis carinii, and Toxoplasma gondii and as antitumor agents. J Med Chem 1998; 41: 4533-41. Gangjee A, Adair O, Queener SF. Pneumocystis carinii and Toxoplasma gondii Dihydrofolate reductase inhibitors and antitumor agents: synthesis and biological activities of 2,4-diamino-5methyl-6-[(monosubstituted anilino)methyl]pyrido-[2,3d]pyrimidines. J Med Chem 1999; 42: 2447-55. Gangjee A, Yang J, Queener SF. Novel non-classical C9-methyl-5substituted-2,4-diaminopyrrolo-[2,3-d]pyrimidines as potential inhibitors of dihydrofolate reductase and as anti-opportunistic agent. Bioorg Med Chem 2006; 14: 8341-51. Gangjee A, Adair OO, Pagley M, Queener SF. N9-Substituted 2,4diaminoquinazolines: synthesis and biological evaluation of lipophilic inhibitors of pneumocystis carinii and toxoplasma gondii dihydrofolate reductase. J Med Chem 2008; 5: 6195-200. Pelphrey PM, Popov VM, Joska TM, et al. Highly efficient ligands for dihydrofolate reductase from Cryptosporidium hominis and Toxoplasma gondii inspired by structural analysis. J Med Chem 2007; 50: 940-50. Rosowsky A, Mota CE, Wright JE, et al. 2,4-diaminothieno[2,3d]pyrimidine analogs of trimetrexate and piritrexim as potential inhibitors of Pneumocystis carinii and Toxoplasma gondii dihydrofolate-reductase. J Med Chem 1993; 36: 3103-12. Rosowsky A, Mota CE, Wright JE, Queener SF. 2,4-diamino-5chloroquinazoline analogs of trimetrexate and piritrexim - Synthesis and antifolate activity. J Med Chem 1994; 37: 4522-8. Rosowsky A, Hynes JB, Queener SF. Structure-activity and structure-selectivity studies on diaminoquinazolines and other inhibitors of Pneumocystis carinii and Toxoplasma gondii dihydrofolate reductase. Antimicrob Agents Chemother. 1995; 39: 79-86. Rosowsky A, Mota CE, Queener SF. Synthesis and antifolate activity of 2,4-diamino-5,6,7,8-tetrahydropyrido[4,3-d]pyrimidine analogs of trimetrexate and piritrexim. J Heterocycl Chem 1995; 32: 335-40. Rosowsky A, Mota CE, Queener SF, Waltham M, Ercikanabali E, Bertino JR. 2,4-diamino-5-substitutedquinazolines as inhibitors of a human dihydrofolate-reductase with a site-directed mutation at position-22 and of the dihydrofolate reductases from Pneumocystis carinii and Toxoplasma gondii. J Med Chem 1995; 38: 745-52. Rosowsky A, Forsch RA, Queener SF. 2,4-diaminopyrido[3,2d]pyrimidine inhibitors of dihydrofolate reductase from Pneumocystis carinii and Toxoplasma gondii. J Med Chem 1995; 38: 261520. Rosowsky A, Mota CE, Queener SF. Brominated trimetrexate analogues as inhibitors of Pneumocystis carinii and Toxoplasma gondii dihydrofolate reductase. J Heterocycl Chem 1996; 33: 195966. Rosowsky A, Papoulis AT, Queener SF. 2,4-diaminothieno[2,3d]pyrimidine lipophilic antifolates as inhibitors of Pneumocystis carinii and Toxoplasma gondii dihydrofolate reductase. J Med Chem 1997; 40: 3694-9. Rosowsky A, Papoulis AT, Queener SF. 2,4-diamino-6,7-dihydro5H-cyclopenta[d]pyrimidine analogues of trimethoprim as inhibitors of Pneumocystis carinii and Toxoplasma gondii dihydrofolate reductase. J Med Chem 1998; 41: 913-8.

Comprehensive Evaluation of New Models [29]

[30]

[31] [32]

[33]

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41] [42] [43]

[44]

[45]

[46]

Rosowsky A, Papoulis AT, Forsch RA, Queener SF. Synthesis and antiparasitic and antitumor activity of 2,4-diamino-6-(arylmethyl)5,6,7,8-tetrahydroquinazoline analogues of piritrexim. J Med Chem 1999; 42: 1007-17. Rosowsky A, Cody V, Galitsky N, Fu HN, Papoulis AT, Queener SF. Structure-based design of selective inhibitors of dihydrofolate reductase: Synthesis and antiparasitic activity of 2,4diaminopteridine analogues with a bridged diarylamine side chain. J Med Chem 1999; 42: 4853-60. Rosowsky A, Papoulis AT, Queener SF. One-step synthesis of novel 2,4-diaminopyrimidine antifolates from bridged alicyclic ketones and cyanoguanidine. J Heterocycl Chem 1999; 36: 723-8. Rosowsky A, Fu HN, Queener SF. Synthesis of 2,4diaminopyrido[2,3-d]pyrimidines and 2,4-diaminoquinazolines with bulky dibenz[b,f]azepine and dibenzo[a,d]-cycloheptene substituents at the 6-position as inhibitors of dihydrofolate reductases from Pneumocystis carinii, Toxoplasma gondii, and Mycobacterium avium. J Heterocycl Chem 2000; 37: 921-6. Rosowsky A, Fu HN, Queener SF. Synthesis of new 2,4-diamino7H-pyrrolo[2,3-d]pyrimidines via the Taylor ring transformation/ring annulation strategy. J Heterocycl Chem 2001; 38: 1197202. Rosowsky A, Forsch RA, Queener SF. Inhibition of Pneumocystis carinii, Toxoplasma gondii, and Mycobacterium avium dihydrofolate reductases by 2,4-diamino-5-[2-methoxy-5[omegacarboxyalkyloxy)benzyl]pyrimidines: Marked improvement in potency relative to trimethoprim and species selectivity relative to piritrexim. J Med Chem 2002; 45: 233-41. Rosowsky A, Chen H, Fua H, Queener SF. Synthesis of new 2,4diaminopyrido[2,3-d]pyrimidine and 2,4-diaminopyrrolo[2,3-] pyrimidine inhibitors of Pneumocystis carinii, Toxoplasma gondii, Mycobacterium avium dihydrofolate reductase Bioorg Med Chem 2003; 11; 59-67 Rosowsky A, Forsch RA, Queener SF. Further Studies on 2,4diamino-5-(2’,5’-disubstituted benzyl)pyrimidines as potent and selective inhibitors of dihydrofolate reductases from three major opportunistic pathogens of AIDS. J Med Chem 2003; 46: 1726-36 Rosowsky A, Fu H, Chan DCM, Queener SF. Synthesis of 2,4diamino-6-[2’-o-(-carboxyalkyl)oxydibenz[b,f]azepin-5-yl]- methylpteridines as potent and selective inhibitors of Pneumocystis carinii, Toxoplasma gondii, and Mycobacterium avium dihydrofolate reductase. J Med Chem 2004; 47: 2475-85 Forsch RA, Queener SF, Rosowsky A. Preliminary in vitro studies on two potent, water-soluble trimethoprim analogues with exceptional species selectivity against dihydrofolate reductase from Pneumocystis carinii and Mycobacterium avium. Bioorg Med Chem Lett 2004; 14: 1811-5. Rosowsky A, Forsch RA, Sibley CH, Inderlied CB. Queener SF. New 2,4-diamino-5-(2’5’-substituted benzyl)pyrimidines as potential drugs against opportunistic infections of AIDS and other immune disorders. Synthesis and species-dependent antifolate activity. J Med Chem 2004; 47: 1475-86. Chan DCM, Fu H, Forsch RA, Queener SF, Rosowsky A. Design, synthesis, and antifolate activity of new analogues of piritrexim and other diaminopyrimidine dihydrofolate reductase inhibitors with öcarboxyalkoxy or ö-carboxy-1-alkynyl substitution in the side chain. J Med Chem 2005; 48: 4420-31. Chan DCM, Anderson AC. Towards Species-specific antifolates. Curr Med Chem 2006; 13: 377-98. Gangjee A, Kurup S, Namjoshi O. Dihydrofolate reductase as a target for chemotherapy in parasites. Curr Pharm Des 2007; 13: 609-39. Gangjee A, Jain HD, Kurup S. Recent advances in classical and non-classical antifolates as antitumor and antiopportunistic infection agents: Part I Anti-cancer agents in medicinal chemistry. 2007; 7: 524-42 Gangjee A, Jain HD, Kurup S. Recent advances in classical and non-classical antifolates as antitumor and antiopportunistic infection agents: Part II. Anti-cancer agents in medicinal chemistry. 2008; 8: 205-231 Champness JN, Achari A, Ballantine SP, Bryant PK, Delves CJ. Stammers DK. The structure of Pneumocystis carinii dihydrofolate reductase to 1.9 Å resolution. Structure 1994; 2: 915-24 Klon AE, Heroux A, Ross LJ, et al. Atomic structures of human dihydrofolate reductase complexed with NADPH and two lipo-

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

[47]

[48] [49]

[50] [51]

[52]

[53] [54]

[55] [56]

[57] [58]

[59] [60]

[61] [62]

[63]

[64] [65]

[66]

[67]

[68]

[69]

749

philic antifolates at 1.09 Å and 1.05 Å resolution. J Mol Biol 2002; 320: 677-93. Kharkar PS, Kulkarni VM. A proposed model of Mycobacterium avium complex dihydrofolate reductase and its utility for drug design. Org Biomol Chem 2003; 1: 1315-22 Popov V, Yee WA, Anderson A. Towards in silico lead optimization: scores from ensembles of protein-ligand conformations reliably correlate with biological activity. Proteins 2007; 66: 375-87. Sutherland JJ, O’Brien, LA, Weaver DF. Spline-fitting with a genetic algorithm: a method for developing classification structureactivity relationships. J Chem Inf Comput Sci 2003; 43: 1906-15 Mattioni BE, Jurs PC. Prediction of dihydrofolate reductase inhibition and selectivity using computational neural networks and linear discriminant analysis. J Mol Graphics Model 2003; 21: 391-419. Gangjee A, Lin X, CoMFA and CoMSIA analyses of Pneumocystis carinii dihydrofolate reductase, Toxoplasma gondii dihydrofolate reductase, and rat liver dihydrofolate reductase. J Med Chem 2005; 48: 1448-69. Bag S, Tawari, NR, Degani MS. Insight into inhibitory activity of mycobacterial dihydrofolate reductase inhibitors by in-silico molecular modeling approaches. QSAR Comb Sci 2009; 28: 296-311. Kitchen DB, Decornez H, Furr JR, Bajorath J. Docking and scoring in virtual screening for drug discovery: methods and applications. Nat Rev Drug Discov 2004; 3: 935-49. Anderson AC, Wright DL. The design and docking of virtual compound libraries to structures of drug targets. Curr Comput Aided Drug Des 2005; 1: 103-27. Valler MJ, Green D. Rhodopsin crystal structure: provides information on GPCR-ligand binding in general? Drug Discov Today 2000; 5: 288-9. Natchiar SK, Jeyaprakash AA, Ramya TN, et al. Structural plasticity of peanut lectin: An X-ray analysis involving variation in pH, ligand binding and crystal structure. J Pep Res 2004; 60: 211-9. Miao J, Hodgson KO, Sayre D. An approach to three-dimensional structures of biomolecules by using single-molecule diffraction images. Proc Natl Acad Sci USA 2001; 98: 6641-5. Kapoor M, Mukhi PL, Surolia N, Suguna K, Surolia K. Kinetic and structural analysis of the increased affinity of enoyl-ACP (acylcarrier protein) reductase for triclosan in the presence of NAD+. Biochem J 2004; 381: 725-33. Freeny J. NMR studies of ligand binding to dihydrofolate reductase. Agnew Chem Int Ed Engl 2000; 39: 290-312. Schenell JR, Dyson HJ, Wright PE. Structure, dynamics, and catalytic function of dihydrofolate reductase. Ann Rev Biophys Biomol Struct 2004; 33: 119-40. http: //www.rcsb.org/pdb/home/home.do Edman U, Edman JC, Lundgren B, Santi DV. Isolation and expression of the Pneumocystis carinii thymidylate synthase gene PNAS 1989; 86: 6503-7. Delves CJ, Ballantine SP, Tansik RL, Baccanari DP, Stammers DK. Refolding of recombinant Pneumocystis carinii dihydrofolate reductase and characterization of the enzyme. Protein expression and purification 1993; 4: 16-23. Stammers DK, Delves C, Ballantine S, et al. Preliminary crystallographic data for Pneumocystis carinii dihydrofolate reductase. J Mol Bio 1993; 230: 679-80. Champness JN, Achari A, Ballantine SP, Bryant PK, Delves CJ, Stammers DK. The structure of Pneumocystis carinii dihydrofolate reductase to 1.9 Å resolution. Structure 1994; 2: 915-24. Cody V, Galitsky N, Lufr JR, et al. Blakley RL. Comparison of ternary complexes of Pneumocystis carinii and wild-type human dihydrofolate reductase with coenzyme NADPH and a novel classical antitumor furo[2,3-d]pyrimidine antifolate. Acta Cryst. 1997; D53: 638-49. Cody V, Galitsky N, Rak D, Luft JR, Pangborn W, Queener SF. Ligand-induced conformational changes in the crystal structures of Pneumocystis carinii dihydrofolate reductase complexes with folate and NADP. Biochemistry 1999; 38: 4303-12. Cody V, Chan D, Galitsky N, et al. Structural studies on bioactive compounds. 30. Crystal structure and molecular modeling studies on the Pneumocystis carinii dihydrofolate reductase cofactor complex with TAB, a highly selective antifolate. Biochemistry 2000; 39: 3556-64. Cody V, Galitsky N, Luft JR, Pangborn W, Rosowsky A, Queener SF. Structure-based enzyme inhibitor design: modeling studies and crystal structure analysis of Pneumocystis carinii dihydrofolate re-

750 Current Pharmaceutical Design, 2011, Vol. 17, No. 7

[70]

[71]

[72]

[73]

[74] [75]

[76]

[77]

[78]

[79] [80] [81]

[82] [83]

[84]

[85]

[86]

[87] [88]

[89]

ductase ternary complex with PT653 and NADPH relative to rat liver dihydrofolate reductase (rlDHFR). Acta Cryst 2002; D58: 946-54. Cody V, Galitsky N, Luft JR, Pangborn W, Queener SF, Gangjee A. Analysis of quinazoline and pyrido[2,3-d]-pyrimidine N9-C10 reversed-bridge antifolates in complex with NADP+ and Pneumocystis carinii dihydrofolate reductase. Acta Cryst 2002; D58: 13939 Cody V, Galitsky N, Luft JR, Pangborn W, Gangjee A, Queener SF. Structure determination of tetrahydroquinazoline antifolates in complex with human and Pneumocystis carinii dihydrofolate reductase: correlations between enzyme selectivity and stereochemistry. Acta Cryst. 2004; D60: 646-55. Cody V, Pace J, Chisum K, Rosowsky A. New insights into DHFR interactions: analysis of Pneumocystis carinii and mouse DHFR complexes with NADPH and two highly potent 5-(omegacarboxy(alkyloxy) trimethoprim derivatives reveals conformational correlations with activity and novel parallel ring stacking interactions. Proteins 2006; 65: 959-69. Li R, Sirawaraporn R, Chitnumsub P, et al. Three-dimensional structure of M. tuberculosis dihydrofolate reductase reveals opportunities for the design of novel tuberculosis drugs. J Mol Biol 2000; 295: 307-23. Roos DS. Primary structure of the dihydrofolate reductasethymidylate synthase gene from Toxoplasma gondii. J Biol Chem 1993; 268: 6269-80 Matthews DA, Bolin JT, Burridge JM, et al. Crystallographic investigation of the cooperative interaction between trimethoprim, reduced cofactor and dihydrofolate reductase. J Biol Chem 1985; 260: 381-91. Bates P, Kelley L, MacCallum R, Sternberg M. Enhancement of protein modeling by human intervention in applying the automatic programs 3D-JIGSAW and 3D-PSSM. Proteins 2001; 45(Suppl 5): 39-46. Yuvaniyama J, Chitnumsub P, Kamchonwongpaisan S, et al. Insights into antifolate resistance from malarial DHFR-TS structures. Nat Struct Biol 2003; 10: 357-65. Veljko M, Popov W, Yee A, Anderson AC. Towards in-silico lead optimization: scores from ensembles of protein/ligand conformations reliably correlate with biological activity. Proteins 2007; 66: 375-87. Kollman PA. Free energy calculations: applications to chemical and biochemical phenomena. Chem Rev 1993; 93: 2395-417. Carlson HA, Jorgensen WL. An extended linear response method for determining free energies of hydration. J Phys Chem 1995; 99: 10667-73. Kollman PA, Massova I, Reyes C, et al. E. III. Calculating structures and free energies of complex molecules: Combining molecular mechanics and continuum models. Acc Chem Res 2000; 33: 889-97. Teramoto R, Fukunishi H. Supervised consensus scoring for docking and virtual screening. J Chem Inf Model 2007; 47: 526-34. Graffner-Nordberg M, Marelius J, Ohlsson S, et al. Computational predictions of binding affinities to dihydrofolate reductase: synthesis and biological evaluation of methotrexate analogues. J Med Chem 2000; 43: 3852-61. Graffner-Nordberg M, Kolmodin K, Åqvist J, Queener SF, Hallberg A. Design, synthesis, computational prediction and biological evaluation of ester soft drugs as inhibitors of dihydrofolate reductase from Pneumocystis carinii. J Med Chem 2001; 44: 2391-402. Morris GM, Goodsell DS, Halliday RS, et al. Automated docking using a Lamarckian genetic algorithm and empirical binding free energy function. J Comput Chem 1998; 19: 1639-62. Graffner-Nordberg M, Kolmodin K, Åqvist J, Queener SF, Hallberg A. Design, synthesis, and computational affinity prediction of ester soft drugs as inhibitors of dihydrofolate reductase from Pneumocystis carinii. Eur J Pharm Sci 2004; 22: 43-54. Gorse AD, Gready JE. Molecular dynamics simulations of the docking of substituted N5-deazapterins to dihydrofolate reductase. Protein Eng 1997; 10: 23-30. Pitts C, Yin J, Bowen D, Maxwell CJ, Southerland WM. Interaction energy analysis of nonclassical antifolates with Pneumocystis carinii dihydrofolate reductase. Int J Mol Sci 2002; 3: 1188-202. Bag S, Tawari NR, Degani MS, Queener SF. Design, synthesis, biological evaluation and computational investigation of novel in-

Degani et al.

[90] [91] [92]

[93]

[94] [95]

[96] [97]

[98] [99]

[100]

[101]

[102]

[103] [104]

[105]

[106] [107] [108] [109] [110] [111] [112] [113]

[114]

hibitors of dihydrofolate reductase of opportunistic pathogens. Bioorg Med Chem 2010; 18: 3128-97. Glide, version 5.0, Schrödinger, LLC, New York, NY, 2008, User manual. Prime, version 3.0, Schrödinger, LLC, New York, NY, 2008, User manual. Jacobson MP, Pincus DL, Rapp CS, Day TJ, Honig B, Shaw DE, Friesner RA. A Hierarchical approach to all-atom protein loop prediction. Proteins 2004; 55: 351-367. Yu Z, Jacobson MP, Friesner RA. What role do surfaces play in GB models? A new generation of surface-generalized born model based on a novel Gaussian surface for biomolecules. J Comput Chem 2006; 27: 72-89. Agrawal VK, Sohgaura R, Khadikar PV. QSAR studies on biological activity of piritrexim analogues against pcDHFR. Bioorg Med Chem 2002; 10: 2919-26. Mattioni BE, Jurs PC. Prediction of dihydrofolate reductase inhibition and selectivity using computational neural networks and linear discriminant analysis. J Mol Graph Model 2003; 21: 391-419. Debnath B, Vishnoi SP, Sa B, Jha T. QSAR study on some dihydrofolate reductase inhibitors. Internet Electronic Journal of Molecular Design 2003; 2: 128-36. Debnath AK. Pharmacophore mapping of a series of 2,4-diamino5-deazapteridine inhibitors of Mycobacterium avium complex dihydrofolate reductase. J Med Chem 2002; 45: 41-53. Sutherland JJ, Weaver DF. Three-dimensional quantitative structure-activity and structure-selectivity relationships of dihydrofolate reductase inhibitors. J Comp-Aided Mol Design 2004; 18: 309-31. Gangjee A, Lin X. CoMFA and CoMSIA analyses of Pneumocystis carinii dihydrofolate reductase, Toxoplasma gondii dihydrofolate reductase, and rat liver dihydrofolate reductase. J Med Chem 2005; 48: 1448-69. Jain P, Soni LK, Gupta AK, Kashkedikar SG. QSAR analysis of 2,4-diamino pyrido[2,3-d]pyrimidines and 2,4-diaminopyrrolo[2,3d]pyrimidines as dihydrofolate reductase inhibitors. Indian J Biochem Biophysics 2005; 42; 315-20. Richmond NJ, Abrams CA, Wolohan PRN, Abrahamian E, Willett P, Clark RDK. GALAHAD: 1. Pharmacophore identification by hypermolecular alignment of ligands in 3D. J Comput-Aided Mol Design 2006; 20: 567-87. Evans DA, Doman TN, Thorner DA, Bodkin MJ. 3D-QSAR Methods: Phase and Catalyst Compared. J Chem Inf Model 2007; 47: 1248-57. Manchester J, Czerminski R. SAMFA simplifying molecular description for 3D-QSAR. J Chem Inf Model 2008; 48: 1167-73. Santos-Filho OA, Cherkasov A. Using molecular docking, 3DQSAR, and cluster analysis for screening structurally diverse data sets of pharmacological interest. J Chem Inf Model 2008; 8: 205465. Gangjee A, Lin X, Biondo LR, Queener SF. CoMFA analysis of tgDHFR and rlDHFR based on antifolates with 6-5 fused ring system using the all-orientation search (AOS) routine and a modified cross-validated r2-guided region selection (q2-GRS) routine and its initial application. Bioorg Med Chem 2010; 18: 1684-701 MarvinSketch version 5.2 ChemAxon Ltd 2008. Maestro, version 8.5, (2008) Schrödinger, LLC, New York, NY, User manual. MacroModel, version 9.6, Schrödinger, LLC, New York, NY, 2008, User manual. Impact, version 5.0, Schrödinger, LLC, New York, NY, 2008, User manual. LigPrep, version 3.0, Schrödinger, LLC, New York, NY, 2008, User manual. Phase, version 3.0, Schrödinger, LLC, New York, NY, 2008, User manual. Broughton M C, Queener SF. Pneumocystis carinii dihydrofolate reductase used to screen potential antipneumocystis drugs. Antimicrob Agents Chemother 1991; 35: 1348-55. Chio LC, Queener SF. Identification of highly potent and selective inhibitors of Toxoplasma gondii dihydrofolate reductase. Antimicrob Agents Chemother 1993; 37: 1914-23. Cody V, Luft JR, Pangborn W. Understanding the role of Leu22 variants in methotrexate resistance: comparison of wild-type and Leu22Arg variant mouse and human dihydrofolate reductase ternary crystal complexes with methotrexate and NADPH. Acta Crystallogr 2005; 61: 147-55.

Comprehensive Evaluation of New Models [115]

[116]

Current Pharmaceutical Design, 2011, Vol. 17, No. 7

Dixon SL, Smondyrev AM, Knoll EH, Rao SN, Shaw DE, Friesner RA. PHASE: a new engine for pharmacophore perception, 3D QSAR model development, and 3D database screening: 1. Methodology and preliminary results. J Computer-Aided Mol Des 2006; 20: 647-71. Tawari NR, Bag S, Degani MS. Pharmacophore mapping of a series of pyrrolopyrimidines, indolopyrimidines and their congeners as multidrug-resistance-associated protein (MRP1) modulators. J Mol Mod 2008; 14: 911-21.

Received: January 10, 2011

Accepted: January 28, 2011

[117]

[118]

[119]

751

Tawari NR, Degani MS. Pharmacophore mapping and electronic feature analysis for a series of nitroaromatic compounds with antitubercular activity. J Comput Chem 2010; 31: 739-51. Tawari NR, Degani MS. Predictive models for nucleoside bisubstrate analogs as inhibitors of siderophore biosynthesis in Mycobacterium tuberculosis: pharmacophore mapping and chemometric QSAR study. Mol Diver In press. DOI: 10.1007/s11030-010-92438 Barrow EW, Suling WJ, Seitz LE, Reynolds RC, Barrow WW. New antifolate inhibitors for Mycobacterium avium. Med Chem 2006; 2: 505-10.