MODELING OF BAKER TRIAZINE DERIVATIVES AS DHFR INHIBITORS USING QUANTUM CHEMICAL DESCRIPTORS
HTML Full TextMODELING OF BAKER TRIAZINE DERIVATIVES AS DHFR INHIBITORS USING QUANTUM CHEMICAL DESCRIPTORS
 Chouksey^{*1}, S. Malik ^{2}, A. Thakur ^{3 }and N. Upadhyay ^{1}
Department of chemistry ^{1}, Chameli Devi Group of Institutions, Indore452020, M.P., India
Department of Chemistry ^{2}, Sadhu Vaswani College, Bhopal462030, M.P., India
Department of Applied Sciences ^{3}, NITTTR, Bhopal462030; M.P., India
ABSTRACT: DHFR being involved in many important cell processes, its inhibition has long been an attractive goal for the development of chemotherapeutic agents. In present work, efforts have been made to model the DHFR inhibitory activity of a series of 4,6diamino2,2dimethyl1,2dihydro1,3,5triazine derivatives to identify the structural requirements for the binding affinity between the receptor and Triazine derivatives. Quantum chemical properties like electron density on specific atoms, net charge on specific atom, binding energy, HOMO, LUMO etc. were used for various structural activity relationship investigations for a series of derivatives of Baker’s triazine for proposals of new compounds which might be useful for the development of effective drugs. The parameters were calculated by optimizing the molecule using MM+ force field. To develop the model with significant statistics and predicting ability, quantum chemical computations were made by using stepwise regression analysis and validated by various crossvalidation parameters. The results were discussed on the basis of maximum R^{2} value which indicates that pentaparametric model is the most significant model for the activity
Keywords:

DHFR inhibitors,
Triazines, Regression analysis, QuantumChemical descriptors, Electron density
INTRODUCTION: Dihydrofolate reductase (DHFR) catalyzes the reduction of folate or 7, 8−dihydrofolate to tetrahydrofolate and intimately couples with thymidylate synthase (TS). TS is a crucial enzyme that catalyzes the reductive methylation of deoxyuridine monophosphate (dUMP) to deoxythymidine monophosphate (dTMP) utilizing N_{5}, N_{10}methylenetetrahydrofolate as a cofactor, which functions as the source of the methyl group as well as the reductant.^{1} This is the exclusive de novo sources of dTMP, hence inhibition of DHFR or TS activity in the absence of salvage, leads to “thymineless death”.^{23}
Thus, DHFR inhibition has long been an attractive goal for the development of chemotherapeutic agents against bacterial and parasitic infections as well as cancer.^{46}
One class of compounds that has been identified as a potential DHFR inhibitor is 4,6Diamino2,2dimethyl1,2dihydro1,3,5triazine, well known as Baker’s triazine.^{7} The triazines are among the oldest known organic nitrogencontaining heterocycles noted for their therapeutic value as anticancer, antimalarial, antibacterial and antiprotozoal agents.^{8, 9} Baker triazines are becoming increasingly important as pharmaceuticals. The extension of studies have explored that 4, 6 diamino1,3,5triazines would also possess other forms of pharmacological activities like inhibitory action against various microorganisms, cancer cells and neuronal sodium channels in addition to the antifolate activity.^{10} Recently, a series of 4,6diamino1,2dihydro1aryl2(1adamantyl) 1,3,5triazines was patented as potent Toxoplasma gondii dihydrofolate Reductase inhibitors with IC_{50}= 2×10^{4}M to 9.7×10^{8} M .^{11}
Many structure based techniques of drug discovery and development have evolved in the past 20 years during the search of therapeutically useful agents. QSAR is one of the drug design method and leads to discovery of potent drugs.
In the present work, a quantitative structure activity relationship has been performed to develop mathematical relationship between quantum molecular descriptors and antimalarial activity to obtain more information about the structural requirement underlined the inhibition of Plasmodial vivax DHFR enzyme. Quantum molecular properties were used for various structure activity relationship investigations for a series of derivatives of Baker’s triazine for proposals of new compounds which might be useful for development of effective antimalarial drugs against resistant Plasmodial vivax. For QSAR modeling we have used maximum R^{2} method and followed stepwise regression analysis. To achieve responsible molecular features for DHFR inhibition activity, we used the molecular modeling technique. To model the most potent triazine derivative we optimized the molecules using molecular mechanics method applying MM+ force field.
We noted that the maximum R^{2} method actually includes a combination of standard error, adjusted R^{2} value, standard error of estimation and Fratio value. The predictive ability of the model is discussed on the basis of crossvalidation method. The accuracy of a molecular mechanics or quantum mechanical method depends on the database used to parameterize the method.
MATERIALS AND METHOD:
Quantitative Structure Activity Relationships (QSAR) have been established for a set of 32 analogues of 4, 6diamino1,2dihydro2,2dimethyl1Rstriazine (Baker’s triazine), potent inhibitor of DHFR enzyme. The DHFR inhibition activity of these compounds analyzed as log 1/C was adopted from the literature and the various substituents selected are as shown in Table 1. The structures of different derivatives of triazine used in present study were drawn using ACD Lab software (Chem Sketch 5.0). Figure 1 shows the parent structure of triazine analogues used in the study.
We have used the quantum chemical approach to identify the structural requirements for the efficient binding between receptor and triazine derivatives used. For this molecular modeling parameters like binding energy, total energy, electron density at different atoms in molecules, net charge on different atoms, HOMO, LUMO, dipole moment, X, Y, Z Coordinates etc. have been selected. All the molecular modeling parameters were calculated from the optimized molecules in the minimum energy conformer with the help of computer software Hyperchem 7.0 (demoversion).^{12} Figure 2 shows optimized structures of compound 9 and 15. Since the calculation methods of these parameters with the software are well documented in the literature, it is not necessary to duplicate the same here.
The molecular parameters having maximum effect on the biological activity were selected by performing correlation studies in –mono, bi, tri, tetra and penta parametric combinations. The quantum chemical parameters selected for the study are as mentioned in Table 2.
Regression^{1315} analyses were made using maximum R^{2} method (Chaterjee et al, 2000) to propose statistically significant models and identify prediction models adopting stepwise regression. Several combinations of variables were examined to identify combinations of variables with good prediction capabilities. Regression results were further validated with the help of some crossvalidation parameters. All the computations were carried out on NCSS statistical software.^{16}
The generated QSAR models were selected on the basis of various statistical parameters such as squared correlation coefficient (R^{2}) which is relative measure of quality of fit; Fischer’s value which represents Fratio between the variance of calculated and observed activity; standard error of estimation (Se) representing absolute measure of quality of fit, crossvalidation parameters viz. PRESS, SSY, R^{2}a, Pogliani’s Q parameter^{1719} etc. to estimate the predictive potential of models.
RESULTS AND DISCUSSION:
As mentioned in the introduction in order to carry out quantum computations we have first carried out the molecular geometry optimization to find out the structural behavior of the compounds as a function of attached groups and their positions. To analyze the relationship between binding affinity and the structure of the molecule, various molecular properties of triazine derivatives mentioned in Table 1, we tested a pool of quantum chemical parameters, presented in Table 2.
As may be seen from Table 2, a very lowlevel degeneracy is present in the activity log 1/c. as a result of the occurrence of degeneracy in the activity; it becomes essential to examine the degeneracy in the molecular modeling descriptors also. A perusal of Table 2 which contains molecular modeling descriptors shows that high to low degeneracy is observed. So, these descriptors can be used successfully in developing statistically significant models.
FIG. 1: PARENT STRUCTURE OF TRIAZINE ANALOGUES USED IN THE STUDY
The univariate correlation among the selected parameters in the form of correlation matrix is presented in Table 3, shows that except for binding energy having correlation coefficient 0.6009, none of the other parameters tested in the study have significant correlation with the binding affinity of the triazine derivatives.
All those correlations resulting in the low value of R (< 0.50) are not considered because those were statistically insignificant. The parameters not participating in the developing QSAR models are not listed in Table 2 but discussed in the methodology section.
TABLE 1: SUBSTITUENT AND BIOLOGICAL ACTIVITY OF TRIAZINE DERIVATIVES USED IN THE PRESENT STUDY
Com. No.  R1  R2  R3  R4  BA (obs.) 
1  H  Cl  Cl  H  8.54 
2  H  Ph(CH_{2})_{2}  H  H  8.19 
3  H  H  Ph(CH_{2})  H  8.05 
4  H  Ph(CH_{2})  H  H  8 
5  H  H  Ph(CH_{2})_{2}  H  7.89 
6  H  CF_{3}  H  H  7.76 
7  H  Cl  H  H  7.76 
8  H  Cl  OCH_{2}Ph  H  7.52 
9  H  SO_{2}F  H  H  7.27 
10  H  Ph  OH  H  7.14 
11  H  NO_{2}  H  H  7.07 
12  H  H  CH_{2}CN  H  6.92 
13  H  H  H  H  6.92 
14  H  Ph  H  H  6.85 
15  H  COCH_{3}  H  H  6.79 
16  Cl  Cl  H  H  6.52 
17  H  H  COCH_{2}Cl  H  6.45 
18  H  COCH_{2}Cl  H  H  6.21 
19  H  OCH_{3}  H  H  6.17 
20  H  H  CN  H  5.14 
21  F  H  H  H  4.74 
22  H  H  Ph  H  4.7 
23  Cl  H  H  H  4.15 
24  OCH_{3}  H  H  H  3.68 
25  Cl  H  H  Cl  3.43 
26  CH_{3}  H  H  H  4 
27  H  SO_{2}NH_{2}  H  H  5.32 
28  H  CONH_{2}  H  H  5.7 
29  H  OH  H  H  6.38 
30  H  F  H  H  7.45 
31  H  C(CH_{3})_{3}  H  H  7.5 
32  H  CN  H  H  7.69 
For QSAR studies inhibitory activity of triazine derivatives was selected as dependent variable and calculated parameters as independent variables. In any thorough investigation of the effects of molecular properties, it is essential to prove that the results are both statistically valid and make chemical sense. It would be appropriate to obtain insight into the physical meaning of the correlation obtained as an output of the regression analysis. The magnitude of descriptors could be used as guideline to improve the selected activity of molecules.
The best uniparametric model was obtained with binding energy (BE) is as given below:
Model1
log 1/C= 0.0001 BE + 8.1654 Eq. 1
N= 32, R= 0.6009, R^{2}= 0.3610, Se= 1.1546, R^{2}a = 0.3398, Fratio= 16.953, Q= 0.4508
Equation 1 signifies that although the value of R is considerable and show dominant effect of BE but, the model has error of estimation much higher than multiple regression coefficient.
TABLE 2: QUANTUM CHEMICAL PARAMETERS SELECTED FOR THE STUDY
Comp. No.  BE  EDN_{5}  EDC_{15}  NCC_{16}  EDN_{1}  HOMO 
1  862.7441  5.715223  3.757367  0.188933  4.9855322  0.192234 
2  2964.054  5.714498  3.896849  7.05574  5.0270383  0.190726 
3  2964.054  5.716309  4.004531  2.081527  5.0303681  0.18387 
4  5209.875  5.720363  3.921552  0.12692  5.0289514  0.180842 
5  7567.46  5.717482  4.009658  2.268232  5.025991  0.183979 
6  8652.506  5.716333  3.982788  3.53549  5.0277707  0.190367 
7  9442.047  5.716795  3.806387  2.542702  5.0172471  0.191029 
8  10732.49  5.720583  3.871574  0.303094  5.0295766  0.176614 
9  11793.42  5.71638  3.83058  3.986662  5.0197869  0.190486 
10  12961.28  5.716624  4.006274  0.230771  5.0394514  0.177236 
11  13903.23  5.71748  3.847654  2.355836  5.0243674  0.190199 
12  14780.46  5.717487  4.001658  0.060965  5.0265014  0.187215 
13  15498.56  5.717487  4.001658  0.060965  5.0265014  0.190604 
14  16562.3  5.716979  3.976768  3.63443  5.0235963  0.187341 
15  17483.52  5.719423  3.927178  6.11597  5.0288385  0.190587 
16  18373.54  5.698127  3.782911  3.944025  4.9789254  0.185998 
17  19369.43  5.712848  3.917474  8.6766  4.995715  0.185212 
18  20365.8  5.717092  4.049561  3.77793  4.9953978  0.179795 
19  21242.56  5.71719  3.711039  0.1065  5.024374  0.190294 
20  22051.48  5.716869  3.943815  6.455801  5.0057395  0.190859 
21  22880.94  5.718125  4.05297  2.57519  5.0198782  0.184684 
22  23928.46  5.717319  4.001572  1.73883  5.0188382  0.184872 
23  24741.2  5.696098  3.938367  2.08235  4.9866837  0.18611 
24  25644.05  5.721142  4.070199  3.00492  5.0225462  0.178169 
25  26530.16  5.696745  3.929862  3.814452  4.9844428  0.187586 
26  27332.74  5.714763  4.002779  0.03795  5.0333191  0.149784 
27  28379.41  5.71826  3.854024  2.892349  5.0202191  0.191285 
28  30355.22  5.717676  3.94491  2.97627  5.0230081  0.190193 
29  31155.3  5.717098  3.710291  0.11986  5.0232801  0.189934 
30  31964  5.717813  3.653065  0.10689  5.0221244  0.190558 
31  32986.86  5.717181  3.90089  7.47083  5.0282344  0.189431 
32  33816.05  5.812435  3.854066  0.057763  5.0423157  0.17131 
TABLE 3: CORRELATION AMONG THE VARIABLES SELECTED FOR THE STUDY
BA  BE  EDC_{15}  EDN_{5}  NCC_{16}  HOMO  EDN_{1}  
BA  1  
BE  0.601  1  
EDC15  0.328  0.1147  1  
EDN5  0.2675  0.229  0.0562  1  
NCC16  0.086  0.0879  0.1903  0.0201  1  
HOMO  0.271  0.1725  0.35531  0.29822  0.07242  1  
EDN1  0.315  0.0112  0.15215  0.49489  0.05667  0.26992  1 
From various biparametric combinations tried a very few gave the statistically significant results and the best result was obtained by the combination of BE with electron density on nitrogen atom at 5th position (ED N_{5}). The model obtained was:
Model2:
log 1/C = 0.0001 BE + 33.2620 (±10.1299) ED N_{5} – 181.7655 Eq. 2
N=32, R= 0.739, R^{2}= 0.5342, Se= 1.002, R^{2}a= 0.5021, Fratio= 16.631, Q= 0.7289
Equation 3 signifies that although the statistical improvement is marginal but the combination of BE and ED N_{5} plays the important role and favors binding affinity of the molecules. The model was good in terms of statistics but not as much as required for describing the structure activity relationship in quantitative manner. So, we tried for triparametric model including BE, ED N_{5} and ED C_{15}. The model so obtained is
Model3
log 1/C= 0.0001 BE – 5.1773 (±1.4202) ED C_{15} + 32.3034 (±8.4936) ED N_{5} – 155.917 Eq. 3
N= 32, R= 0.8271, R^{2}= 0.6841, Se= 0.8403, R^{2}a= 0.6503, Fratio= 20.215, Q= 0.9843
Equation 3 suggests that positive correlation coefficient of ED N_{5} shows direct relationship with BA (log 1/C). This signifies that the increase in electron density on N atom at 5^{th} position has positive impact on biological activity quantitatively.
For the further improvement in predictive potential and modeling efficiency, we tried for tetra and pent parametric models. The models developed are:
Model4
log 1/C = 0.0001 BE 5.79251 (±1.3565) ED C_{15} + 32.1902 (±7.9442) ED N_{5} – 0.0894 ± (0.0400) NC C_{16} Eq. 4
N= 32, R= 0.8565, R^{2}= 0.7336, Se= 0.7859, R^{2}a= 0.6941, Fratio= 28.584, Q= 1.089
Model5
log 1/C = 194.386 – 0.00011×10^{5} (±1.49×10^{5}) BE + 23.1741 (±8.8672) EDN5 – 6.2681(±1.3142) EDC15 –0.08727 (±0.03805) NCC16 + 18.9148 (±9.7127) EDN1 Eq. 5
N= 32, R= 0.8761, R^{2}= 0.7675, R^{2}a= 0.7228, Se= 0.7482, F= 17.1636, Q= 1.1708
Equation 4 demonstrates that high net charge on carbon atom at 16th position has unfavorable effect for the inhibition of DHFR by triazine derivatives. The improvement in regression coefficient (R), R^{2}a and Q supports Model4 for the prediction of the activity.
To confirm our findings, we estimated the log 1/C values from the above models and compared them with the observed values of biological activity. The residual values showed that compound number 31 as outlier for the modeling. The model obtained after deletion of this compound from the series resulted unto much improved model as represented by Equation 6.
Model6
log 1/C =  194.595 – 0.00012×10^{5} (±1.34×10^{5}) BE + 27.4464 (±7.777) EDN5 – 5.9057 (±1.1407) EDC15 – 0.0468 (±0.0352) NCC16 + 13.8380 (±8.5417) EDN1 Eq.6
N= 31, R= 0.91129, R^{2}= 0.83046, R^{2}a= 0.79656, Se= 0.64613, F= 24.49243, Q= 1.41038
Equation 6 shows significant increase in R^{2} and R^{2}a values from 0.7675 to 0.8305 & 0.7228 to 0.7966 respectively. This increase in values simply indicates that the deleted compound has the unfair share in modeling of the activity and confirms the exceptional behavior from their parent series. The value of R^{2}a will decrease if the deletion of compounds does not reduce the unexplained variation in the model enough to offset the loss degree of freedom.
Predictive Power of Proposed Models:
It is not necessary that a model which gives excellent statistics will also have excellent predictive power. So, the predictive potential of proposed models is validated by using variety of crossvalidation parameters. Crossvalidation method evaluates the validity of a model by how well it predicts data rather than how well it fits data. For this, various crossvalidation parameters are used viz. PRESS (Predictive Residual Error Sum Squares), SSY (Sum of the Squares of response values), R^{2}a (overall predictive ability), S_{PRESS} (Standard Error of Predicted Residual Sum Squares), PSE (Predicted Square Error). The various crossvalidation parameters calculated are summarized in Table 4.
PRESS is good to estimate the real predictive error of the models to assess BA and the ratio PRESS/SSY is to estimate the confidence interval of the binding affinity. To have a reliable QSAR model the ratio PRESS/SSY should be less than 0.4. The significant lowering in the value of crossvalidation parameters PRESS and PRESS/SSY ratio from Model1 to 6 justify the models.
The indication of the performance of model is obtained from R^{2}cv. ^{ }In our case, the highest R^{2}cv is found for the Model6 indicating it has excellent predicting power with value 0.7959. Figure 3 shows the graph of correlation between observed and estimated biological activity according to Model6
TABLE 4: CROSS VALIDATION PARAMETERS FOR THE PROPOSED MODELS
MODEL  PARAMETERS  N  R^{2}a  R^{2}cv  PRESS/SSY  SPRESS  PSE 
1  BE  32  0.339765  0.7696  1.7696  1.1546  1.118 
2  BE, EDN5  32  0.502107  0.12815  0.87185  1.003  0.9546 
3  BE, EDN5, EDC15  32  0.650296  0.53831  0.46169  0.8404  0.7861 
4  BE, EDN5, EDC15, NCC16  32  0.694093  0.63679  0.36321  0.7859  0.7219 
5  BE, EDN5, EDC15,NCC16,EDN1  32  0.722765  0.69703  0.30297  0.7482  0.6744 
6  BE, EDN5, EDC15,NCC16,EDN1  31  0.796553  0.79586  0.20414  0.6461  0.5802 
FIG. 2: OPTIMIZED STRUCTURES OF COMPOUND 9 AND 15
FIG. 3: GRAPH BETWEEN OBSERVED AND ESTIMATED BA FOR MODEL6
CONCLUSION: From the results and discussions made above, we conclude that quantum chemical parameters can be used successfully for modeling the inhibition activities of DHFR by triazine derivatives. The results obtained express the
unfavorable effect of presence of high electron density at carbon atom on 15^{th} position and the presence of high net charge on carbon at 16^{th} position for the inhibition of DHFR by selected triazine derivatives. The comparison of all the models obtained exhibit the dominant and significant role of electron density at nitrogen at the 5^{th} position in the molecule over the carbon atoms at 15^{th} and 16^{th} positions. The results also indicate that molecular (3D) modeling can be used for the understanding of structural behavior and selecting the compounds with potential activity.
ACKNOWLEDGMENT: We greatly acknowledge support from National Institute for Technical Teachers Training and Research, Bhopal. We also acknowledge late Dr. P.V. Khadikar sir for sharing his knowledge and constant motivation for this research.
REFERENCES:
 Blaney J.M., Hansch C., Silipo C. and Vittorio A. Rev. 2000; 84: 333338.
 Hadjipavlou–Litina D. and Pontiki E. Internet Electron. J. Mol. Des. 2002; 1:134–141.
 García–Domenech R., Catalá–Gregori A., Calabuig C., Antón–Fos G., Castillo L. and Gálvez J. Internet Electron. J. Mol. Des. 2002; 1: 339–350.
 Goodey N.M., Herbert K.G., Hall S.M. and Bagley K.C. Prediction of residues involved in inhibitors specificity in the DHFR family. Biochimica et Biophysica Acta, 2011; 1814: 18701879.
 Goodey N.M., Alapa M.T., Hagmann D.F., Korunow S.G., Mauro A.K., Kwon K.S. and Hall S.M. Development of a fluorescently labeled thermostable DHFR for studying conformational changes associated with inhibitor binding. Biochemical and Biophysical Research Communications, 2011; 413: 442447.
 Legesse A. and Prasad V Bharatam. Computeraided molecular design od 1Himidazole2,4diamine derivatives as potential inhibitors of Plasmodium falciparum DHFR enzyme. J. Molecular Modeling, 2011; 17: 657667.
 Vijayraj R. 3DQSAR Studies on the Inhibitory Activity of Trimethoprim Analogues against Escherichia coli Dihydrofolate Reductase. J. chemical biology and drug design, 2012; 79: 935942.
 Kulkarni R. and Konda S. 3D QSAR Studies on nonclassical DHFR Inhibitors. RGUHS J Pharm Sci, 2013; 3: 6066.
 Debnath B., Srikanth K., Banarjee S. and Jha T. Internet Electron. J. Mol. Des. 2002; 1: 488–502.
 Hawser S., Lociuro S., and Islam K. Dihydrofolate reductase inhibitors as antibacterial agents. Biochem. Pharmacol. 2006; 71 (7): 941–8.
 Ekaterini A.V., Polyxeni T. and Christos R. PCT Int. Appl. 2002; 10: 784.
 Hyperchem 7 demo version from hyperchem.com.
 Borst P. and Quellette M. New mechanism of drug resistance in parasitic protozoa. Annu. Rev. Microbiol. 1995; 49: 427460.
 Kompis I.M., Islam K. and Then R.L. DNA and RNA synthesis: antifolates. Chem. Rev. 2005; 2: 593620.
 Jolivet J., Cowan K.H., Curt G.A., Clendeninn N.J. and Chabner B.A. Med. Chem. 2008; 309: 10941104.
 NCSS 9 demo version from www.ncss.com.
 Tanrikulu Y., Schneider G. Pseudo receptor model in drug design: bridging ligand and receptor based virtual screening. Nature Reviews Drug Discovery, 2008; 7: 667677.
 Bhagwat V.W., Khadikar P.V., Tiwari A., Solanki A., Choubey A., Neel Kamal, Manana P. and Lowekar V. QSAR evidences for topological indices to mimic lipophilicity of Thia and AzaCrown ethers. Asian J. Chem. 2009; 21 (7): 52125220.
 Vijey A. Muthukumar and Hemalatha C. Nagaraj. Virtual screening studies to design dual inhibitors of integrase and caspase: QSAR, pharmacophore modeling and docking Studies. International Journal of Pharmacy and Pharmaceutical Sciences, 2013; 5: 09751491
How to cite this article:
Chouksey, S. Malik, A. Thakur and N. Upadhyay: Modeling of Baker Triazine Derivatives as DHFR Inhibitors Using Quantum Chemical Descriptors. Int J Pharm Sci Res 2015; 6(5): 209096.doi: 10.13040/IJPSR.09758232.6(5).209096.
All © 2013 are reserved by International Journal of Pharmaceutical Sciences and Research. This Journal licensed under a Creative Commons AttributionNonCommercialShareAlike 3.0 Unported License.
Article Information
33
209096
492
767
English
Ijpsr
R. Chouksey*, S. Malik , A. Thakur and N. Upadhyay
Assistant Professor Chameli Devi Group of Institutions Near Toll Naka, Khandwa Road Indore (M.P.) 452020, India.
rchouksey26@gmail.com
22 September, 2014
12 November, 2014
11 January, 2015
10.13040/IJPSR.09758232.6(5).209096
01 May, 2015