Cytotoxicity and 2D-QSAR study of some heterocyclic compounds

1 Introduction

Hydantoin and thiohydantoin are five-member heterocyclic system containing very reactive cyclic urea and thiourea cores (López and Trigo, 1985; Meusel and Gütschow, 2004). In recent years, heterocyclic compound and its derivatives are very familiar to their anti-inflammatory (Ghate et al., 2003), antimicrobial (Khan et al., 2005), antitubercular (Gupta and Prabhu, 2004), antipyretic (Shastri et al., 2004), analgesic (Ghate et al., 2005), antioxidant (Torres and Faini, 2006) and cytotoxic (Kostava and Momekov, 2008) activities. They are now widely used as drugs, medicines, dyes and raw materials in manufacturing industries etc. Much attention has been paid to the synthesis of nitrogen (N)-, oxygen (O)-, sulfur (S)- containing heterocyclic compounds and their derivatives mainly due to their broad spectrum biological and pharmaceutical activities. In our laboratory, several substituted heterocyclic compounds such as methyl and bromine groups in the benzene ring of isatin, Δ²-1,3,5-thiadiaozolines were synthesized and found a good cytotoxic activity (Islam et al., 2001a, 2001b; Lingcon et al., 2001). In light of this biological activity, we thought of synthesizing different types of hydantoin and thiohydantoin related heterocyclic compounds and their derivatives which were tested for cytotoxic activity. Though development of drugs is lengthy, laborious and expensive, computer aided drug design (CADD) can help us to increase the pool of interesting structures that can be evaluated. The most important step is to find the possible structural feature of compounds with desired biological activity. Over the years of development many methods, algorithms and techniques have been discovered and applied in QSAR studies (He and Jurs, 2005; Eldred et al., 1999). Nowadays QSARs are being applied in many disciplines with much emphasis in drug design. As a well accepted technique, two-dimensional quantitative structure–activity relationship (2D-QSAR) was carried out to study the biological activity. It is a mathematical model that was used to evaluate the toxicity of a compound from its physiochemical properties of molecular structures.

2 Experimental

2.2

2.2 Synthesis

Twenty two (22) compounds were prepared according to the literature procedure of Muccioli et al. (2003) by direct heating instead of micro-wave assistance. All the synthesized compounds illustrated in Table 1 were confirmed with melting point, IR and ¹H NMR spectral analysis. The synthetic portion was not included in this manuscript due to the main focus on 2D-QSAR study.

Table 1 Synthesized compounds which are employed for 2D-QSAR study.

Compound ID	R1	R2	X	Y
1	H	H	H	O
2	H	C6H5 (phenyl)	H	O
3	C6H5 (phenyl)	C6H5 (phenyl)	H	O
4	H	C6H11 (cyclohexyl)	H	O
5	C6H11 (cyclohexyl)	C6H11 (cyclohexyl)	H	O
6	H	H	Br	O
7	H	CH3	Br	O
8	H	CH3–CH2	Br	O
9	H	CH3–CH2–CH2–CH2	Br	O
10	H	CH3–CH2–CH2–CH2–CH2	Br	O
11	H	H	Cl	O
12	H	CH3CH2	Cl	O
13	H	CH3–CH2–CH2–CH2	Cl	O
14	H	CH3–CH2–CH2–CH2–CH2–CH2–CH2	Cl	O
15	H	H	H	S
16	H	CH3	H	S
17	CH3	CH3	H	S
18	H	CH3CH2	H	S
19	CH3CH2	CH3CH2	H	S
20	H	CH3CH2CH2CH2	H	S
21	CH3CH2CH2CH2	CH3CH2CH2CH2	H	S
22	H	C6H5 (phenyl)	H	S

3 Calculating descriptors

All the calculations were performed on Intel core-i5 Fujitsu Laptop Computer with 6–8 GB of memory and 100 GB of scratch disk space. The 3D-structure of 22 molecules was sketched by the Gauss view03 software. The Molecular Mechanics (MM+) force field with the aid of Chemoffice Ultra-5 software package was applied for preliminary geometry optimization. The final geometry optimization of the MM+ force field optimized structures was done by Gaussian98 (Revision A.9) (Frisch et al., 1998) software applying B3LYP/6-31G(d,p) level of theory. Semi-empirical PM3 (Stewart, 1989a, 1989b) method by Gaussian98 software had been used in order to obtain an accurate charge distribution and quantum-chemical descriptors (Net atomic charge, Q_A; mean absolute atomic charge, Q_M; Highest occupied molecular orbital, HOMO; Lowest unoccupied molecular orbital, LUMO; Energy gap, E_GAP; Absolute hardness, η; Activation hardness, Δη; σ electron densities, Q_σ; π electron densities, Q_π; Molecular polarizability, α; Ionization energy, I_P; Total energy, E_T; Heat of formation, ΔH_F; Steric energy, E_S; Molecular surface area, MSA; Molecular volume, MV; Molecular dipole moment, MDM etc.) for each compound in the series and some other descriptors (Gibbs’s free energy change of solvation, $\delta G_{\text{solv}}^{°}$ ; electrostatic Gibb’s free energy change in solution, $\delta G_{\text{elec}}^{°}$ etc.) were also calculated by Gaussian98 at B3LYP/6-31G(d,p)-CPCM solvation (DMSO) method. Initially, a total of twenty five types of semi-empirical and thermodynamic descriptors were calculated and these descriptors were finally reduced into four (Table 4) to study the 2D-QSAR models. The MLR, PLS and ANN methods were applied to perform the QSAR models and cross validation method (LOO) was also used to validate the predictive ability of the model obtained by MLR method.

4 Statistical methods

For the QSAR study, model selection was performed by Build QSAR (version 2.1.0.0) (De Oliveira and Gaudio, 2001) software program. Statistical calculations allowed for the selection of the models with the following characteristics: high squared correlation coefficient (R²), high Fischer’s value (F-test), low standard error of estimate (SEE), correlation matrix (Table 5) among the parameters and the least number of descriptors involved. Next, the IBM SPSS Statistics20 software packages were applied for detailed statistical analysis of the QSAR models.

5 Results and discussions

The correlation between various descriptors (Kier and Hall, 1986, 1999) with biological activity is the most important means of structure–activity relationship (SAR) study. Thus the equation should use the minimum number of descriptors to obtain the best fit. To achieve this, a popular procedure is used to find out the saturation point, a point beyond which there is no considerable improvement in regression coefficient (R²) values has been observed. By interpreting the resulting descriptors, it is possible to gain some insight into factors that are likely to govern the cytotoxic activity. The best QSAR model built using multiple linear regression (MLR) method is represented by the following equation:

The high correlation coefficient R (0.91) indicates the susceptibility of descriptors (HOMO, ΔH_F, E_T and E_S) to form the above model (1). Squared correlation coefficient (R²) of 0.83 explains 83% variance in biological activity of the tested compounds. It also indicates the statistical significance >99.9% with F-values (20.70). Cross-validated square correlation coefficient (Q²) by LOO technique was 0.74 which showed a good internal predictive ability of the model. The model was also validated by applying the Y-randomization test. Several random shuffles of the Y vector were performed and the obtained results are in good agreement with the suggested limits (Eriksson et al., 2003). The low $R_{\text{train}}^2$ and $Q_{\text{LOO}}^2$ values shown in Table 6 indicate that there is no chance of correlation or structural dependency in the proposed model. Consequently Eq. (1) can be considered as a perfect model with both high statistical significant and excellent predictive ability. The predictive ability of the model was further confirmed by leave-5-out cross validation with $R_{\text{cv(ext)}}^2$ . The $R_{\text{cv(ext)}}^2$ values are shown in Table 7. It is observed that HOMO, ΔH_F, E_T and E_S are the best descriptors in the establishment of the QSAR model for heterocyclic derivatives such as hydantoin and thiohydantoin related compounds. The correlation of the experimental activities with the MLR calculated ones is illustrated in Fig. 1. Partial Least Square (PLS) was also applied to generate a model (Geladi and Kowalski, 1986; Cramer et al., 1988) for quantitative structure–activity relationship (QSAR) between a set of molecular descriptors used in the MLR method and experimental activity. The correlation of the experimental and calculated activities with the PLS method is shown in Fig. 2. The correlation coefficient R (0.90), squared correlation coefficient R² (0.81), standard error of estimate SEE (0.32), and Fischer Statistics F (18.55) obtained with the PLS method indicate that the model proposed to predict activity is significant and pertinent to that of MLR method. In addition the architecture 5-5-1 of three layer artificial neural networks (ANN) shown in Fig. 5 is used to calculate the biological activities with the help of a set of molecular descriptors and experimental activity. The correlation between experimental and calculated activities with the ANN method is shown in Fig. 3. The correlation coefficient R (0.95), squared correlation coefficient R² (0.91), standard error of estimate SEE (0.23), and Fischer Statistics F (41.94) obtained with the neural network show that model proposed to predict activity by ANN was good and relevant to that of MLR method. The correlation of the experimental values with the calculated values in LOO procedure shown in Fig. 4 was reliable due to the high values of the correlation coefficient R_CV (0.86), squared correlation coefficient $R_{\text{cv}}^2$ (0.74), standard error of estimate SEE (0.40) and Fischer Statistics F (14.65).

Close

Figure 1

Predicted cytotoxic activities by MLR in comparison with experimental.

Export to PPT

Close

Figure 2

Predicted cytotoxic activities by PLS in comparison with experimental.

Export to PPT

Close

Figure 3

Predicted cytotoxic activities by ANN in comparison with experimental.

Export to PPT

Close

Figure 4

Predicted cytotoxic activities by cross validation (LOO) in comparison with experimental.

Export to PPT

Over all, the biological activities predicted by MLR, PLS, ANN and LOO procedure with respect to their experimental values are shown in Table 8. The predicted activities in Fig. 6 show the approximately same approaches to the experimental values.

Close

Figure 5

Schematic representation of architecture (5-5-1) of the three layer neural network (ANN).

Export to PPT

Close

Figure 6

Graphical representation of biological activities predicted by MLR, PLS, ANN, LOO as well as experimental activities.

Export to PPT

6 Conclusion

The brine shrimp lethality bioassay is considered as a useful tool for rapid and preliminary assessment of toxicity of the compounds. Further studies are required to calculate the more accurate bioactivity and, to find the mode of pharmacological activities. Significant regression equations were obtained by MLR, PLS and ANN methods with respect to their experimental cytotoxic activities. The best regression equation was obtained on the following descriptors: Highest occupied molecular orbital (HOMO), heat of formation (ΔH_F), total energy (E_T) and steric energy (E_S). These variables allowed physical explanation of electronic molecular properties contributing to the cytotoxic activity as the electronic character relates directly to the electron distribution of interacting molecules. The predicted biological activities by MLR, PLS and ANN showed a good agreement with experimental values but the activities obtained from ANN were relatively better among them. The LOO, Leave-5-out cross validation and the Y-randomization techniques indicate that the model is significant, robust and has a good predictive ability.