QSAR study of the toxicity of nitrobenzenes to Tetrahymena pyriformis using quantum chemical descriptors

1 Introduction

The rapid development of new compounds by the chemical industry in general and in particular by the agrochemical, petrochemical and pharmaceutical industries is accompanied by an increasing toxic burden in the environment. Because of this, the development of tools able to assess hazardous effects on living species should receive high attention (Smiesko and Benfenati, 2004).

Quantitative Structure–Property/Activity Relationship (QSPR/QSAR) methods are among the most practical tools in computational physical chemistry. These methods are based on the axiom that the variance in the physicochemical properties and activities of chemical compounds is determined by the variance in their molecular structures. Thus, if experimental data are available for only some chemicals in a group, one can predict the missing from molecular descriptors calculated for the whole group and suitable mathematical model (Karcher and Devillers, 1990; Schultz et al., 2003; Katritzky et al., 2001). The global prediction of toxicity using QSARs has been the goal of many workers who utilized a variety of approaches. This goal is alluring, but has yet to be achieved satisfactorily. There are a number of reasons for the absence of success (Arulmozhiraja and Morita, 2004). The deficiency of available toxicity data has clearly held back progress. This lack of success has been compounded in many studies by a poor appreciation of the insufficient heterogeneity, or chemical diversity, in the dataset. Further, while some molecular properties (such as hydrophobicity) are well described, others, including electrophilic reactivity, ionization, and hydrogen bonding, are poorly parameterized. Last, mechanisms of toxic action are not fully understood or misinterpreted, or their relevance in the modeling of toxicity is ignored (Cronin et al., 2001).

Nitroaromatics are hazardous chemicals that display several manifestations of toxicity, including skin sensitization, immunotoxicity, germ cell degeneration, inhibition of liver enzymes and also a conjectured carcinogenicity. The modeling of toxicity of nitroaromatic compounds was complicated by the paucity of experimental data. Nitrobenzenes (NBs) are widely used as industrial chemicals, and consequently have high potential for environmental pollution, they have been reported (Zoeteman et al., 1980) to be present in surface waters. They are reactive chemicals, being reported to be uncouplers of oxidative phosphorylation (Purdy, 1988) and may be regarded as pro-electrophiles (Roberts, 1987) yielding the corresponding potentially highly toxic C-nitroso compounds. Because of their widespread use, the toxicity of NBs has been quite extensively examined, and they have been the subject of a number of QSAR studies. The NBs are representatives of electrophilic toxicants in that, depending on substitution pattern, they may undergo a number of different electrophilic reactions. NBs toxicity has been extensively studied by several groups of workers with the use of different methodologies. Attempts to model the acute toxicity of NBs have been reviewed by Dearden et al. (1995). Due to the reactive electrophilic nature of the NBs, it is not surprising that previous modeling efforts focused on the use of electronic molecular descriptor (Deneer et al., 1987, 1989; Roberts, 1987; Veith and Mekenyan, 1993; Lang et al., 1996; Yuan et al., 1997; Zhao et al., 1997; Debnath et al., 1991; Mekenyan et al., 1997).

Since the expression of chemical toxicity is a combination of penetration into, or through, biological membranes and the interaction of the toxicant with the site of action, this principle is represented mathematically as the following generic QSAR (Schultz, 1999):

Our aim of this study is to develop reliable and predictive QSAR models for identifying the primary molecular factors explaining the interaction effect (electronic/penetration) and governing the toxicity of these NB derivatives and to examine what factors other than log P are controlling the toxicity of compounds.

2 Materials and methods

2.1

2.1 Data set

A total of 50 NBs each containing either a halogenated or/and a methyl, nitro and other substituents is created from the data set described by Cronin et al. (2001). Scheme 1 shows the template structure of nitrobenzene.

Close

Scheme 1

The structural template of nitrobenzene (NB).

Export to PPT

The experimental toxicity data log(1/IGC₅₀) of the 50 congeners are listed in Table 1.

Table 1 NBs and their toxicity against Tetrahymena pyriformis (Cronin et al., 2001).

No.	Compound	Observed toxicity $\text{log}({\text{IGC}}_{50}^{-1})$
1	2,6-Dimethylnitrobenzene	0.30
2	2,3-Dimethylnitrobezene	0.56
3	2-Methyl-3-chloronitrobenzene	0.68
4	2-Methylnitrobenzene	0.05
5	2-Chloronitrobenzene	0.68
6	2-Methyl-5-chloronitrobenzene	0.82
7	2,4,5-Trichloronitrobenzene	1.53
8	2,5-Dichloronitrobenzene	1.13
9	6-Chloro-1,3-dinitrobenzene	1.98
10	Nitrobenzene	0.14
11	3-Methylnitrobenzene	0.05
12	1,3-Dinitrobenzene	0.89
13	3,4-Dichloronitrobenzene	1.16
14	4-Methylnitrobenzene	0.17
15	1,4-Dinitrobenzene	1.30
16	4-Chlonitrobenzene	0.43
17	2,3,5,6-Tetrachloronitrobenzene	1.82
18	6-Methyl-1,3-dinitrobenzene	0.87
19	3-Chloronitrobenzene	0.73
20	1,2-Dinitrobenzene	1.25
21	2-Bromonitrobenzene	0.75
22	6-Bromo-1,3-dinitrobenzene	2.31
23	3-Bromonitrobenzene	1.03
24	4-Bromonitrobenzene	0.38
25	2,4,6-Trimethylnitrobenzene	0.86
26	5-Methyl-1,2-dinitrobenzene	1.52
27	2,4-Dichlorobenzene	0.99
28	3,5-Dichlorobenzene	1.13
29	6-Iodo-1,3-dinitrobenzene	2.12
30	2,3,4,5-Tetrachloronitrobenzene	1.78
31	2,3-Dichlorobenzene	1.07
32	2,5-Dibromobenzene	1.37
33	1,2-Dichloro-4,5-dinitrobenzene	2.21
34	3-Methyl-4-bromonitrobenzene	1.16
35	2,3,4-Trichloronitrobenzene	1.51
36	2,4,6-Trichlonitrobenzene	1.43
37	4,6-dichloro-1,2-Dinitrobenzene	2.42
38	3,5-Dinitrobenzyl alcohol	0.53
39	3,4-Dinitrobenzyl alcohol	1.09
40	2,4,6-Trichloro-1,3-dinitrobenzene	2.19
41	2,3,5,6-Tetrachloro-1,4-dinitrobenzene	2.74
42	2,4,5-Trichloro-1,3-dinitrobenzene	2.59
43	4-Fluoronitrobenzene	0.25
44	4-Fluoro-2-nitrotoluene	0.25
45	1-Fluoro-2-nitrobenzene	0.23
46	1-Fluoro-3-nitrobenzene	0.20
47	4-Nitrobenzaldehyde	0.20
48	2-Nitrobenzaldehyde	0.17
49	3-Nitrobenzaldehyde	0.14
50	3-Nitroacetophenone	0.32

3 Results

The calculated quantum chemical descriptors, namely, ɛ_lumo, the electrophilicity power ω, and the estimated partition coefficient log P are given in Table 2.

Several linear QSAR models involving one, two, and three descriptors are established and strongest multivariable correlations are identified by the “Best Subsets” regression analysis of the MINTAB program.

3.1

3.1 One-parameter QSAR models

Model #1

(5)

$$\text{log}(1/{\text{IGC}}_{50})=0.04+0.39\text{log}PN=50\text{,}{R}^2=0.16\text{,}{R}_{\text{adj}}^2=0.14\text{,}\text{SD}=0.68$$ Model #2

(6)

$$\text{log}(1/{\text{IGC}}_{50})=-0.89-1.23{\epsilon }_{\text{lumo}}N=50\text{,}{R}^2=0.66\text{,}{R}_{\text{adj}}^2=0.65\text{,}\text{SD}=0.43$$ Model #3

(7)

$$\text{log}(1/{\text{IGC}}_{50})=-2.89+0.94\omega N=50\text{,}{R}^2=0.68\text{,}{R}_{\text{adj}}^2=0.67\text{,}\text{SD}=0.42$$ The plots of log(1/IGC₅₀) versus the three descriptors, log P, ω, and ɛ_lumo are given in Fig. 1a–c, respectively.

Close

Figure 1

(a–c) Plot showing the relationship between each parameter and the observed toxicity.

Export to PPT

It turns out that the best one-parameter QSAR model is obtained with Parr’s electrophilicity index ω (model #3, R² = 0.68). In order to improve the predictivity of the QSAR models, it is necessary to investigate multilinear QSAR models involving two and three parameters.

3.3

3.3 Three-parameters QSAR model

Model #7

(11)

$$\text{log}(1/{\text{IGC}}_{50})=-11.70+4.7\omega +4.9{\epsilon }_{\text{lumo}}+0.42\text{log}PN=50R^2=0.87\text{,}{R}_{\text{adj}}^2=0.86\text{,}\text{SD}=0.27\text{,}F=106.11$$ A significant improvement of the quality of the QSAR model is obtained with a combination of the three parameters, namely, Parr’s electrophilicity index ω, the partition coefficient log P, and the LUMO energy ɛ_lumo. Fig. 2 shows the linear correlation between the observed and the predicted toxicity values obtained using Eq. (11).

Close

Figure 2

Predicted vs. observed toxicity using Eq. (11).

Export to PPT

In Table 3 are given the coefficients, the coefficient errors and t-test values of the three molecular parameters corresponding to the best model #7.

Table 3 Coefficients, coefficient errors and t-test values of the descriptors of the best QSAR model (Eq. (11)).

Number	X	DX	t-Test	Descriptor
0	−11.717	1.942	−6.03	Intercept
1	0.4204	0.0532	7.90	log P
2	4.7055	0.9157	5.14	ω
3	4.983	1.208	4.13	ɛlumo

4 Cross-validation

In order to check the reliability and the stability of the best elaborated QSAR model (Eq. (11)), we have used the leave-1/3-of-set-out validation in the following manner: the parent data points were divided according to the experimental values into three subsets (the 1st, 4th, 7th, etc. entries go into the first subset A, the 2nd, 5th, 8th, etc. into the second subset B, and the 3rd, 6th, 9th, etc. into the third subset C). In each of three combinations, two of the subsets were combined into one and the correlation equation was derived with the same descriptors. The obtained equation was used to predict data for the remaining subset. It turns out that the predicted R² values using subsets (A + B), (B + C), (A + C) are close to that corresponding to the full training set (A + B + C) and the average values of R² (Fit) and R² (Predicted) (see Table 4) are also close. Note that the $R_{\text{adj}}^2$ value of the models corresponding to the subsets A + B, A + C, and B + C is much larger than 0.80, indicating that our model is stable and can be efficiently used for estimating the toxicity of other nitrobenzenes for which no experimental data are available.

5 Discussion

The elaborated QSAR models (Eqs. (5)–(11)) reveal that the toxicity of the nitrobenzenes could be explained by a number of electronic and transport factors. Electrophilicity, as defined by ɛ_lumo and ω is important in describing the electronic interaction and the reactivity of these toxins; whereas hydrophobicity, as expressed by log P is important to describe the transport to the site of action. To put in evidence the contribution of each parameter (ω, ɛ_lumo, log P) to the toxicity, we studied the relationship between these parameters and the toxicity $\text{log}({\text{IGC}}_{50}^{-1})$ . Although the electrophilicity indexes ω and ɛ_lumo are found to be more significant in the one-parameter QSAR models (Eqs. (2) and (3)) in comparison with log P (Eq. (5)), the inclusion of the latter in the two- and three-parameter QSAR models is of great importance. Indeed, the combination of log P with either ω and/or ɛ_lumo provides more reliable models. The two-parameter model involving ω and ɛ_lumo explains only 71% of the variance of the toxicity; whereas the two-parameter models involving the combinations of ω with log P and ɛ_lumo with log P explain around 80–82% of the variance of the toxicity. This indicates the importance of the hydrophobicity index in the multilinear regression models. Indeed, it turns out that the combination of the three parameters increases remarkably the predictive power of the QSAR model given by Eq. (11) (R² = 0.87, $R_{\text{adj}}^2=0.86$ , SD = 0.27, F = 106.11). As can be seen from the statistical parameters of the above equation, a considerable improvement is achieved by combining the three descriptors. (Eq. (11)) can explain about 87% of the experimental variance of the dependent variable $\text{log}({\text{IGC}}_{50}^{-1})$ besides it presents a high F of Fischer (F = 106.11) and a low standard deviation (SD = 0.27) which confirms that the model #7 predicts the toxicity (dependent variable) in a statistically satisfactory significant manner. According to the t-test values (|t|), the importance of the descriptors involved in the model decreases in the following order: log P > ω > ɛ_lumo. The most significant descriptor according to the t-test (see Table 4) is the partition coefficient log P. The second significant descriptor is Parr’s electrophilicity index ω and the last one is the LUMO energy ɛ_lumo. The ɛ_lumo is related directly to the electron affinity of a molecule and as such characterizes the susceptibility of the molecule to be attacked by nucleophiles; whereas, Parr’s electrophilicity index, ω, defined in terms of the electronic chemical potential and the chemical hardness, expresses the stabilization energy when the system acquires an additional electronic charge from the environment. Our QSAR models reveal that Parr’s electrophilicity index ω constitutes the main among the three descriptors in explaining the toxicity of the nitrobenzene derivatives. Indeed, ω alone explains about 68% of the variance of the toxicity (see Eq. (7)).

To confirm the electrophilic behavior of these nitrobenzenes, we have performed a comparison of the electrophilicity power of the 50 toxins with electrophilicity power of some nucleic acids (NA) bases. The values of electronic chemical potential, chemical hardness and electrophilicity indexes for adenine A, guanine G, cytosine C, uracil U and thymine T are given in Table 5.

It turns out that the electrophilicity indexes of the 50 toxins (see Table 2) are all greater than those of the NA bases (see Table 5). Consequently, the toxin will act as an electrophile (electron acceptor); whereas the NA will act as a nucleophile (electron donor) during the toxin–NA interaction. The positive coefficient obtained for Parr’s electrophilicity index ω as a quantum chemistry descriptor in all QSAR models #1–7, supports the earlier concept that the toxicity of nitrobenzene derivatives increases with the increase in their electron accepting capability, indicating, that the electron flow takes place from the organism to the toxins. The same statements were obtained for the toxicity of polychlorinated dibenzofurans (Sarkar et al., 2006) and aromatic compounds (Cronin et al., 2001). Consequently, the most toxic nitrobenzenes are predicted to be characterized by high electrophilicity power (strong electron acceptors) and high hydrophobicity (lipophilicity) values.

6 Conclusion

The present study shows that quantum chemistry descriptors expressing the electrophilicity power, namely, the LUMO energy and Parr’s electrophilicity index ω in combination with the hydrophobicity index, log P, expressing the transport factor, are useful for the prediction of the toxicity of a nitrobenzenes to Tetrahymena pyriformis ( $\text{log}({\text{IGC}}_{50}^{-1})$ . The best QSAR model (Eq. (11)) is able to describe about 87% of the variance in the experimental toxicity and could be efficiently used for estimating the toxicity of nitrobenzene derivatives for which the experimental data are unavailable. Our study shows that Parr’s electrophilicity index constitutes the main descriptor in explaining the toxicity of these toxins although the contributions of the log P and ɛ_lumo descriptors are also important. Indeed, the elaborated QSAR model reveals that the most toxic nitrobenzenes are characterized by large hydrophobicities and high electrophilicity powers.