PROCEEDING OF 3 RD INTERNATIONAL CONFERENCE ON RESEARCH, IMPLEMENTATION AND EDUCATION OF MATHEMATICS AND SCIENCE YOGYAKARTA, 16 17 MAY 2016 QSAR STUDY OF ANTIMALARIA OF XANTHONE DERIVATIVES USING MULTIPLE LINEAR REGRESSION METHODS Dhina Fitriastuti a, *, Jumina b, Priatmoko b and Iqmal Tahir b a Department of Chemistry, Faculty of Mathematics and Natural Sciences, Universitas Islam Indonesia b Department of Chemistry, Faculty of Mathematics and Natural Sciences, Universitas Gadjah Mada email: dhinafitriastuti@yahoo.co.id C 04 Abstract Research on Quantitative Structure Activity Relationship (QSAR) analysis of xanthone derivatives had been conducted to obtain new antimalaria active compounds. The QSAR analysis was carried out using the semi empirical method of AM1 with the descriptors of atomic net charge, log P, dipole moment, polarizability, mass and volume. Statistical analysis was performed using the multilinear regression analysis. Then, the xanthone was designed using the validated best QSAR equation. The QSAR equation describing the relationship between antimalarial activity and parameter of lipophilicity, electronic and steric properties of xanthone was: Log (1/IC 50 ) = (2.726)qC 1 + (1.783)qC 10 + (0.2387)log P + 5.282 (n = 17, r 2 = 0.683, F calc /F table = 4.309, PRESS = 0.652). Keywords: QSAR, xanthone, antimalaria, multiple linear regression I. INTRODUCTION Malaria continues to be main health problem and deadly parasitic disease in the world. During 100 years, the world has not given clear contribution to the curing of the disease. In addition, World Health Organization (WHO) [1] reported that 41% of world population (2.3 billion people) was threatened, 300-500 million were infected, 1.5-2.7 million were died by malaria. In recent years, progress has been made in malaria control as a result of insecticide-treated bed nets and effective treatment. But the development of resistance to insecticides and medicines as well as the poor quality of the health systems in many affected countries pose threats to these achievements [1]. Malaria could be treated by oral medication. However, it constantly changes, especially through the development of parasite (such as P.falciparum), which is resistance to standard anti malaria drugs, such as chloroquin. Therefore, the discovery and development of new effective anti malaria drugs are urgently required to solve the problems. In this connection, Quantitative StructureActivity Relationship (QSAR) analysis plays an important role to minimize trial and error in designing new antimalarial drugs. In a QSAR analysis, the central task is to find a regression function that predicts the activity of the molecule in high accuracy. Hence, the present study is aimed at to establish the QSAR between experimental antiplasmodial activity and structure electronic descriptors which may focus on the molecular structures of the compounds. In last decades, QSAR have been applied in many areas enabling to prevent time consuming and cost during the analysis of biological activities of interest [2]. Experimental data used in the computational chemistry study were anti plasmodium activity of xanthone derivatives (IC 50 ) which were obtained from [3]. The lipophilicity, electronic and steric parameters of xanthone derivatives could be determined and then used as descriptor in Quantitative Structure Activity Relationship (QSAR). The QSAR study was conducted based on linear relationship between log IC 50 and atomic net charge of atoms on xanthone skeleton. Data of atomic net charge obtained from computational calculation using semi empirical method of AM1 was analyzed using linear regression analysis using statistic software. According to the results of QSAR analysis, the most active site of molecule, i.e. the one which is responsible to the antimalarial activity, could be predicted. Then, the design of xanthone molecule could be done by modifying the position of substituents. The position of substituents would be better to conduct C-23
ISBN 978-602-74529-0-9 on the active site. The molecule with low IC 50 has high activity and could be proposed to synthesize in laboratory. Therefore, the design of xanthone derivative with higher activity could reduce the trial and error experiments in the laboratory as well as the cost and time. II. EXPERIMENTAL SECTION A. Materials Xanthone derivatives compounds were taken from [3]. The structure of xanthone compound is illustrated in Figure 1, while the structures of xanthone derivatives as well as their antimalarial activities are presented in Table 1. B. Instrumentation For this study, a Laptop equipped with AMD E-450 APU with Radeon Processor 1.65 GHz and RAM 2.00 GB. All the compounds (Table 1) were calculated using package HyperChem Program Version 8.0 for Windows (license: Iqmal Tahir Universiti Malaysia Perlis) and complete geometry optimization with the semi-empirical AM1 method, statistical program IBM SPSS version 22 for Windows. C. Procedure QSAR descriptors were calculated for each of the compounds using the QSAR module of semiempirical AM1 method. The QSAR models are evaluated using sets of xanthone derivatives compounds whose molecular structure and antiplasmodial activity are known. Antiplasmodial activity of these compound were taken as the activity against chloroquine resistant P. falciparum (FcB1/Colombia strain) and is presented as the value of log 1/IC 50 where IC 50 is an effective concentration inhibiting 50% growth of the parasite. All the compounds were calculated using HyperChem and complete geometry optimization with the semi-empirical AM1 method was performed. The geometry was optimized to an RMS gradient of 0.001 kcal/(å mol) in vacuo (Polak-Ribière method). Quantum chemical descriptors were calculated, as for example: atomic net charges, dipole moment, log P, polarizability, volume and molar mass. From all the descriptors above mentioned, it can be considered that some of them give valuable information about the influence of electronic, steric and coefficient partition features upon the biological activity of drug molecules. In this work, the molecular descriptors were selected so that they represent the features necessary to quantify the activity. The QSAR model was generated by Multiple Linear Regression (MLR) Backward method by using SPSS package. It relates the dependent variable ŷ (biological activity) to a number of independent variables x i (molecular and electronic descriptors) by using linear equations. This method of regression estimates the values of the regression coefficients by applying least square curve fitting method. The model was chosen based on some statistical parameters such as r 2, F calc /F table, and PRESS. The developed QSAR models were validated using the following statistical measures: r 2 (coefficient of determination) and q 2 (cross-validated r 2 by LMO). III. RESULTS AND DISCUSSION Analysis of quantitative relationship between structure and activity conducted in this research included : determination of xanthone derivatives with high antimalarial activity which were obtained from the previous research [3]; optimization of xanthone structure; determination of descriptors; calculation of descriptor through the structure optimization; statistical analysis to give QSAR equations; determination of the best QSAR equation and design of xanthone derivative based on the best QSAR equation. In this research AM1 semi-empirical methods were used to calculate a number of descriptors, because the samples in this research i.e. xanthone derivatives have large number molecular compounds and bulky structures. The AM1 semi empirical method is easier to calculate a number of descriptors than using PM3 semi-empirical method. In current practice, AM1 semi-empirical methods serve as efficient C-24
PROCEEDING OF 3 RD INTERNATIONAL CONFERENCE ON RESEARCH, IMPLEMENTATION AND EDUCATION OF MATHEMATICS AND SCIENCE YOGYAKARTA, 16 17 MAY 2016 computational tools which can yield fast quantitative estimates for a number of descriptors. This may be particularly useful for correlating large sets of experimental and theoretical data, for establishing trends in classes of related molecules, and for scanning a computational problem before proceeding with higher level treatments [5]. The structure of xanthone was presented in Figure 1. Xanthone derivatives consisted of functional groups attached on carbon atom and their antimalarial activities reported by [3] were presented in Table 1. Figure 1 Structure of xanthone Table 1. Xanthone derivatives and their antimalarial activities by [3] Comp Structure IC 50 (μg/ml) 1 a 1.3 ± 0.5 Comp Structure IC 50 (μg/ml) 12 a 3.5 ± 0.3 2 b 0.9 ± 0.2 13 a 3.5 ± 0.3 3 a 1.0 ± 0.1 14 a 24.7 ± 2.1 4 a 2.7 ± 0.8 15 a 4.9 ± 1.5 5 a 4.4 ± 0.2 16 b 1.8 ± 0.7 C-25
ISBN 978-602-74529-0-9 6 a 1.9 ± 0.3 17 a 4.1 ± 0.5 7 a 0.8 ± 0.1 18 a 3.7 ± 0.4 8 b 4.2 ± 2.0 19 a 6.6 ± 0.8 9 a 1.6 ± 0.4 20 a 1.4 ± 0.4 10 b 3.2 ± 0.6 21 a 5.4 ± 1.8 11 a 2.3 ± 0.6 a Training set b Test set Descriptor is parameter of molecular properties employed as variable in the QSAR research. The lipophilicity, electronic and steric parameters of xanthone derivatives could be determined and then used as descriptor in Quantitative Structure Activity Relationship (QSAR). Descriptors used in this research were atomic net charge, dipole moment (µ), logarithmic partition coefficient (log P), polarizability (α), mass (M) and volume molar (V). The optimization process was conducted to obtain the properties of structures and followed with the data recording via single point menu. Method used to calculate the electronic properties and descriptors was AM1. Results of QSAR analysis A good fit was assessed based on the determination squared correlation coefficients (R 2 ), standard deviation (SD) and Fisher s statistic (F). Most of the QSAR modeling methods implement the C-26
PROCEEDING OF 3 RD INTERNATIONAL CONFERENCE ON RESEARCH, IMPLEMENTATION AND EDUCATION OF MATHEMATICS AND SCIENCE YOGYAKARTA, 16 17 MAY 2016 leave-one-out (LOO) or leave-many out (LMO) cross-validation procedure, which are internal validation techniques [6]. LOO cross-validation procedure consists of removing one data point from the training set and constructing the model only on the basis of the remaining training data and then testing on the removed point. LMO cross-validation procedure calculates the models leaving multiple observations out at a time, reducing the number of times it has to recalculate a model. The outcome from the crossvalidation procedure is cross-validated R 2 (LOO-q 2 or LMO-q 2 ), which is used as a criterion of both robustness and predictive ability of the model. In this research, the leave-4-out cross-validation method as the internal validation tool was performed. The total set of compounds was manually divided into a training set (17 compounds) for generating 2D QSAR models and a test set (4 compounds) for validating the quality of the models. Selection of the training set and test set molecules was done on the basis of structural diversity and a wide range of activity such that the test-set molecules represent a range of biological activity similar to that of the training set; thus, the test set is truly representative of the training set [7]. This approach resulted in selection of compounds 2, 8, 10 and 16 as the test set and the remaining 17 compounds as the training set. The uni column statistics of the training and test sets is reported in Table 2. The maximum and minimum value in training and test set were compared in a way considering that: (i) the maximum value of log1/ic 50 of test set should be less than or equal to maximum value of log1/ic 50 of training set and (ii) the minimum value of log1/ic 50 of test set should be higher than or equal to minimum value of log1/ic 50 of training set. Table 2 shows that average and standard deviation values of training and test set are not different significantly, indicating a similar data distribution in both. Table 2. Uni-column statistics of the training and test sets for QSAR models Set Number of Data Average Max Min St. Dev Sum Training 17 5.530 6.097 4.607 0.359 94.013 Test 4 5.666 6.046 5.377 0.296 22.662 Multiple linear regression method is the standard method for multivariate data analysis. It estimates the values of the regression coefficients by applying least squares curve fitting method. For getting reliable results, dataset having typically 5 times as many data points (molecules) as independent variables (descriptors) is required. In first step of model development, all of the descriptors (atomic net charge, dipole moment (µ), logarithmic partition coefficient (log P), polarizability (α), molecular mass (M) and molar volume (V)) were included. These parameters were used to represent properties such as lipophilicity, shape and electron distribution, which were believed to have a major influence on the drug s activity [8]. Based on statistical backward analysis, the non-significant descriptors were excluded from the model. QSAR investigations of the xanthone series resulted in several QSAR equations, considering the term selection criterion as R 2, F and PRESS. Some statistically significant 2D QSAR models were chosen for discussion (Table 3 and 4). Table 3. Statistical parameters of 5 QSAR models of xanthone derivatives Model Descriptor R R 2 F calc F table F calc /F table PRESS 1 qc10, qc1, log P, qc3, qc9 0.853 0.727 5.855 2.167 2.702 0.562 2 qc10, qc1, log P, qc9 0.837 0.701 7.029 2.167 3.244 0.615 3 qc10, qc1, log P 0.826 0.683 9.336 2.167 4.309 0.652 4 qc10, log P 0.806 0.649 12.957 2.167 5.980 0.721 Table 4. Coefficient of selected independent variables for QSAR models as obtained from multilinear regression analysis Model Atomic net charges (coulomb) qc 1 qc 3 qc 9 qc 10 Log P Constant C-27
ISBN 978-602-74529-0-9 1 4.5514-0.7321 4.3214 3.1194 0.1680 5.0792 2 3.4938 2.6007 2.4519 0.2022 5.1260 3 2.726 1.783.2387 5.282 4 1.773.307 4.897 According to the statistical calculation, it was obtained the strong correlation between the electronic and lipophilic parameters to the antimalarial activity of the substituted xanthone. All of the QSAR models have shown good correlation between their corresponding descriptors and biological activity. The higher the F value is, the more significant the data would be. Data would be significant if F calc /F table > 1. According to the F calc /F table value, it indicated that the models fits in all cases was not a chance occurrence and all models were statistically significant. Based on the R 2 criteria (R 2 > 0.6), all of the model were passed to validation step. To validate the selected prediction function, a cross-validation and an external test were carried out. Cross-validation is a practical and reliable method for testing the significance. The developed QSAR models were validated using the following statistical measures: r 2 (coefficient of determination) and q 2 (cross-validated r 2 by LMO). A QSAR model is considered to be predictive, if the following conditions are satisfied: r 2 > 0.6 and q 2 > 0.6. The r 2 and q 2 values were used as deciding factors in selecting the optimal models [9]. The predicted (LMO) activities of the xanthone derivatives by the above models are shown in Table 5. The result of evaluation antimalarial activity [predicted log IC 50 ] and correlation with antimalarial activity [experiment log IC 50 ] for the model 3 by using semi-empiric AM1 method of test set and training set can be seen at Figure 2 and Figure 3. The statistically best significant model obtained by MLR method (Model 3) with r 2 = 0.919 was considered, as the model showed good internal predictive power (q 2 = 0.68) of 68% and predictivity for the external test set (pred_r 2 = 0.50) of about 50%. Consequently, QSAR model 3 can be considered as the most suitable model for antimalarial activity against FcB1/Colombia strain of P.falciparum with both high statistical significant and excellent predictive ability. The best QSAR equation was as follows: Log (1/IC 50 ) = (2.726)qC 1 + (1.783)qC 10 + (0.2387)log P + 5.282 (n = 17, r 2 = 0.683, F calc /F table = 4.309, PRESS = 0.652). Compound Table 5. Comparison of observed and predicted values of log 1/IC 50 log1/ic50 observed Predicted log1/ic50 Model 1 Model 2 Model 3 Model 4 2 6.046 6.414 6.341 6.348 6.331 16 5.745 5.887 5.902 5.933 6.102 10 5.495 5.692 5.769 5.702 5.647 8 5.377 5.252 5.377 5.385 5.173 PRESS 0.210 0.187 0.170 0.273 r 2 0.831 0.894 0.919 0.775 q 2 0.73 0.70 0.68 0.65 pred_r 2 0.38 0.44 0.50 0.19 C-28
PROCEEDING OF 3 RD INTERNATIONAL CONFERENCE ON RESEARCH, IMPLEMENTATION AND EDUCATION OF MATHEMATICS AND SCIENCE YOGYAKARTA, 16 17 MAY 2016 Predicted log 1/IC 50 6.5 6.0 5.5 5.0 y = 1.0318x r² = 0.9197 5.0 5.5 6.0 6.5 Observed log 1/IC 50 (a) (b) Figure 2. Plot of observed versus predicted log 1/IC 50 values of xanthone derivatives of (a) Test set and (b) Training set by Model 3 Based on the coefficient of descriptor parameters involved in the QSAR model 3 in which seen on the atomic net charge descriptor, thus the active region of xanthone compounds can be predicted. The atoms which have an influence on antimalarial activity were C 1 and C 10. The active region can be illustrated in Figure 3. The bold line in this figure shows the active central area allegedly providing antimalarial effect. This can happen, because they contribute quite active to the influence of the quantity of biological activity. The changes in the values of the atomic charges in active areas would result in changes in the value of antimalarial activity. Predicted log 1/IC 50 6.5 6.0 5.5 5.0 4.5 y = 0.683x + 1.7532 r² = 0.683 4.5 5.0 5.5 6.0 6.5 Observed log 1/IC 50 Figure 3. The hypothetical active region of xanthone structure Design of new xanthone derivatives The best QSAR equation was model 3, in which there were variables of qc1, qc10 and log P that strongly affect the antimalarial activity. Therefore, to design xanthone derivatives with high antimalarial activity, all of those variables should be considered. Substitution was conducted by varying the functional groups on the active site and also considering the ease of synthesis process as well as the availability of chemical in the laboratory. The new xanthone derivatives and the calculation of its log 1/IC 50 were presented on Table 6. Table 6. The new xanthone derivatives and their IC 50 prediction No Compound qc1 qc10 logp log1/ic 50 IC 50 (μg/ml) 1 0.2196 0.0615 1.80 6.42 0.38 2 0.0351 0.0224 3.95 6.36 0.44 C-29
ISBN 978-602-74529-0-9 3 0.2184 0.0252 1.83 6.36 0.44 4-0.0131 0.0573 3.16 6.10 0.79 IV. CONCLUSION The QSAR equation describing the relationship between antimalarial activity and parameter of lipophilicity, electronic and steric properties of xanthone was: Log (1/IC 50 ) = (2.726)qC 1 + (1.783)qC 10 + (0.2387)log P + 5.282 (n = 17, r 2 = 0.683, F calc /F table = 4.309, PRESS = 0.652). REFERENCES [1] WHO, 1997, Roll Back Malaria, A Global Partnership, Geneva. [2] Mustofa, Yappi, A.D., Valentin, A., and Tahir, I.,2003, J. Med. Sci., 35 (2), 67 74. [3] Hay, A., E., He lesbeux, J., J., Duval, O., LabaRed, M., Grellier, P. and Richomme, P., 2004, Antimalarial Xanthones from Calophyllum caledonicum and Garcinia vieillardii, J. Lif. Sci., 75, 3077-3085. [4] Basilico, N., Pagani, E., Monti, D., Olliaro, P., dan Taramelli, D., 1998, A microtitrebased method for measuring the haem polymerization inhibitory activity (HPIA) of antimalarial drugs, J. Antimicrob. Chemother., 42, 55-60. [5] Hadanu, R., Idris, S., and Sutapa, I.W., 2015, QSAR Analysis of Benzothiazole Derivatives of Antimalarial Compounds Based on AM1 Semi-Empirical Method, Indo. J. Chem., 15 (1), 86-92. [6] Baumann, K. and N. Stiefl, 2004, Validation tools for variable subset regression, J.of Comp.-Aided Mol.Design, 18 (7-9), 549-562. [7] Golbraikh A. and Tropsha A., 2003, J. Chem. Inf. Comp. Sci. (43), 144. [8] Purcell, W.P., 1974, Quantitative Structure-Activity Relationship, in Bergmann, E. D., Pullman B., Molecular and Quantum Pharmacology, Reidel Publishing Company, Dordrecht. [9] Sahu, N.K., M. Sharma, V. Mourya, and D.V. Kohli, 2012, QSAR Study of Some Substituted 4-Quinolinyl and 9-Acridinyl Hydrazones as Antimalarial Agents, Acta Poloniae Pharmaceutica-Drug Research, 69 (6), 1153-1165. C-30