Modeling the toxicity of polar and nonpolar narcotic compounds to luminescent bacterium Shk1

INTRODUCTION

Bioassays have been widely used for toxicity testing in various compartments of the environment, such as water/ wastewater, sediment, and soil. An ideal bioassay for toxicity testing should be inexpensive, easy to perform, accurate, reproducible, and both relevant and appropriate for on-line or in situ toxicity monitoring [1]. Luminescent bacteria-based bioassays have an additional advantage in that they usually have a short response time. The commercial basis for a luminescent bacteria-based assay was first described by Bulich [2]. The assay employed the marine luminescent bacterium Vibrio fischeri (also known as Photobacterium phosphoreum). This assay was later marketed by Azur Environmental (Carlsbad, CA, USA) and is commonly applied as the Microtoxr̀ assay, which has become a standardized method for assessing possible ecotoxicological effects of water samples [3].

In recent years, interest has been increasing in regard to testing and estimating the toxicity of influent to activated sludge wastewater treatment plants to prevent upsets in plant operations caused by influent toxicity. Whereas V. fischeri has been used for the toxicity monitoring of effluent discharges [4, 5], studies in the literature have frequently concluded that V. fischeri is not well suited for predicting the possible toxic effects of influent wastewater on the activated sludge [6-9], because V. fischeri is not an indigenous activated sludge microorganism.

The continuous Shk1 assay was developed at the Center for Environmental Biotechnology at the University of Tennessee for the purpose of influent wastewater toxicity monitoring and has been described previously [10]. The bacterium used in this assay, Shk1, is a genetically modified luminescent bacterium whose host strain was isolated from the activated sludge in an industrial wastewater treatment plant [11]. When applied to influent wastewater toxicity assessment, the continuous Shk1 assay has the advantage of being relevant to the activated sludge compared to other assays that use luminescent organisms not native to activated sludge. In our previous study [10], we demonstrated that compared to the Microtox assay, the results of the continuous Shk1 assay for phenols are more similar to those of activated sludge respiration or growth-inhibition assays.

The number of chemical compounds in industrial use is huge; currently, more than 100,000 chemical compounds are used worldwide [12]. A significant number of environmental pollutants are eventually discharged into wastewater. However, due to cost and other constraints, testing the toxicity of all compounds that can eventually exist in wastewater is difficult. A quantitative structure-activity relationship (QSAR) can be developed and used to predict toxicity when little or no experimental data are available. We have developed QSAR models based on the logarithm of the octanol-water partition coefficient (log K_ow) for phenols in our previous study [10]. The continuous Shk1 assay can be applied to compounds other than phenols, but we limited the scope of that previous study to phenols because of their distinct mechanism of toxicity.

Several mechanisms of aquatic toxicity have been identified, such as polar and nonpolar narcosis, respiratory uncoupling, etc. [13]. These mechanisms of toxicity can be broadly categorized as general or specific [14]. General toxicity refers to polar and nonpolar narcosis, which act by nonspecific disruption of the functioning of the cell membrane, and specific toxicity refers to reactive toxicity, which occurs through disruption of the function of a defined receptor site in the cell [14]. Most industrial organic chemicals have a narcotic mode of toxic action [15]. The correlation of the observed toxicity of nonpolar narcotic compounds with log K_ow has been used to define a baseline toxicity [13]. Polar narcotic compounds, which include phenols and anilines, exhibit toxicity slightly greater than that predicted by the baseline toxicity, whereas reactive compounds can exhibit toxicity orders of magnitude higher than that predicted by baseline toxicity [16].

Because of the deviations from the baseline toxicity of polar narcotic compounds, properly modeling the toxicity of polar and nonpolar narcotic compounds together using log K_ow usually is difficult. For example, when Tang et al. [17] modeled the toxicity of both nonpolar and polar narcotic (namely, phenols) compounds with log K_ow, a low r² (coefficient of determination) of 0.11 was obtained. As those authors noted, two groups of data points were present; consequently, the authors developed separate QSAR models. After separating the two classes of compounds, the r² values increased to 0.88 and 0.79 for nonpolar narcotic compounds and polar narcotic phenols, respectively.

The fact that the toxicity of polar and nonpolar narcotic compounds cannot be satisfactorily modeled together using log K_ow as the single model descriptor reflects the failure of octanol to adequately model the polar interactions associated with the partition of polar narcotic compounds into the cell membrane [14]. Separate QSAR models can be developed for polar and nonpolar narcotic compounds, as was done by Tang et al. [17], but a single model for both classes of compounds is desirable. The solvatochromic method developed by Kamlet et al. [18, 19] has been successfully used to model the aquatic toxicity to several organisms. In particular, the use of solvation parameters appears to provide a promising alternative for modeling the toxicity of polar and nonpolar narcotics together [14]. In the present study, we investigated the use of solvation parameters in modeling the toxicity of polar and nonpolar narcotic compounds to Shk1 simultaneously.

MATERIALS AND METHODS

A detailed description of the solvation parameter model, which is based on a cavity model of solvation, can be found in Gunatilleka and Poole [14]. The solvation parameter model describes the change in free energy for the transfer of a solute (toxicant) from the donor (aqueous phase) to the receptor (site of toxic action, generally considered to be the cell membrane). The transfer of the solute from the donor phase to the acceptor phase requires the formation of a cavity in the acceptor phase. As Gunatilleka and Poole described, the solvation parameter model for aquatic toxicity can be expressed as

(1)

The solvation parameters in Equation 1 are McGowan's characteristic volume (V_x), the excess molar refraction (R), the solute's dipolarity/dipolarizability (∑^H), the solute's effective or summation hydrogen-bond acidity (∑α^H), and the solute's effective or summation hydrogen-bond basicity (∑β^H). The constants c, m, r, s, a, and b can be obtained from regression. Observed toxicity is represented by log (1/EC50), where EC50 is the median effective concentration. The terms mVx, rR, sπ^H, a∑α^H, and β∑β^H reflect the relative contribution of each molecular interaction between the solute and the donor and acceptor phases to the observed toxicity, and the constants (m, r, s, a, and b) characterize the differences between the capacities of the donor phase and the acceptor phase to interact with the solute in specific manners [20]: Ease of forming a cavity (the m constant), interaction with solute n electrons or π electrons (the r constant), participation in dipole-dipole or dipoleinduced dipole plus some polarizability interactions (the s constant), relative hydrogen bond acidity (the a constant), and hydrogen-bond basicity (the b constant).

The Shk1 luminescence EC50 values of 40 organic compounds were obtained in our laboratory following an assay procedure described and reported previously [10, 21]. These compounds included both polar narcotic phenols and nonpolar narcotic compounds. The organic compounds, their log (1/EC50) values, and solvation parameters are shown in Table 1. Extrapolated EC50 values (50% Shk1 luminescence repression was not observed at saturation) are indicated in the table as well. The solvation parameters were taken from the study of Gunatilleka and Poole [14]. A multiple regression of log (1/ EC50) values on the solvation parameters was performed using JMP 4.0 software (SAS Institute, Cary, NC, USA).

RESULTS

An initial multiple regression showed that the variables R, π^H, and ∑α^H were not statistically significant (p = 0.49–0.97). Consequently, these variables were excluded from the subsequent multiple regression. Using the remaining two variables (V_x and ∑β^H), a multiple regression resulted in an r (multiple correlation coefficient) value of 0.86. Examining the Cook's distance revealed that octane was a statistical outlier. After removing this outlier, the final multiple regression yielded the following regression equation:

(2)

The t tests for the regression coefficients showed that all were statistically significant (p < 0.0001 for c and m; p = 0.0002 for b). The lack-of-fit test indicated no lack of fit (p = 0.3053). The predicting capacity of the model shown by Equation 2 was examined based on the values for error sum of squares (SSE), predicted error sum of squares (PRESS), and mean error sum of squares (MSE). Using the JMP 4.0 software, the values of SSE, PRESS, and MSE were calculated to be 4.95, 5.92, and 0.14, respectively.

As Neter et al. [22] pointed out, a fairly close agreement between PRESS and SSE would suggest that MSE may be a reasonably valid indicator of the selected model's predictive capability. Because the PRESS and SSE values of the model shown by Equation 2 were judged to be close, the MSE value of 0.14 was used to indicate the predicting capacity of the model. Taking the square root of 0.14, a value of 0.37 is obtained, which means that an error of 0.37 is expected when predicting log (1/EC50) using the model. As Blum [23] pointed out, the achievable accuracy of a QSAR for predicting toxicity to bacteria is approximately one order of magnitude. Therefore, the prediction accuracy of the model shown by Equation 2 is acceptable.

DISCUSSION

As Gunatilleka and Poole [14] pointed out, the approach of the solvation parameter model provides an interpretation of nonspecific toxicity in terms of the characteristic membrane sorption properties. The results of the multiple regression showed that three variables (R, π^H, and ∑α^H) were not statistically significant. Equivalently, their regression coefficients (i.e., r, s, and a) were all taken to be zero in Equation 2. As mentioned in Materials and Methods, the coefficients in the product terms of Equation 1 characterize the differences in the aqueous phase and the site of toxic action. Those regression coefficients in Equation 2 that equal zero indicate the following: The capacity of the site of toxic action in Shk1 to interact with lone-pair electrons is the same as water (the r constant), the site of toxic action in Shk1 is as dipolar or polarizable as water (the s constant), and the site of toxic action in Shk1 is as hydrogen-bond basic as water (the a constant). Consequently, the existence of lone-pair electrons in the toxicants, the dipolarity, and the hydrogen-bond acidity of the toxicants do not contribute to the nonspecific toxicity of organic compounds to Shk1.

The nonzero regression coefficients for V_x and ∑β^H in Equation 2 indicate that forming a cavity at the site of toxic action in Shk1 is relatively easier than in water (the m constant) and that the site of toxic action in Shk1 is more hydrogen-bond acidic than water (the b constant). In Equation 2, the regression coefficient for V_x is positive and that for ∑β^H is negative, which indicates that toxicity increases as the McGowan's characteristic volume, or size (discussed later), of the toxicants increases and that toxicity decreases as the hydrogen-bond basicity decreases.

It is necessary to compare Equation 2 with an equivalent QSAR model for V. fischeri, because both Shk1 and V. fischeri are luminescent bacteria used in aquatic toxicity assays. To do this, V. fischeri toxicity data were collected from studies in the literature (Table 1). A multiple regression of V. fischeri log (1/EC50) values on the solvation parameters was performed using JMP 4.0 software, and the following regression equation was obtained:

(3)

The r and r_adj values for Equation 3 are relatively low compared to those for Equation 2. However, no outliers were identified, because all studentized residuals (not shown) had absolute values smaller than four. Examining the Cook's distance (not shown) also did not reveal any influential observations. As Cronin and Schultz [24] pointed out, an r₂ value of 0.6 to 0.7 is sometimes all that can realistically be expected in QSAR models, and improvement of statistics should not be the driving force for QSAR analysis. Consequently, no action was taken, such as removing certain observations, to improve the correlation in Equation 3.

Gunatilleka and Poole [14] obtained the following equation when they modeled the nonspecific toxicity to V. fischeri using a different set of organic compounds:

(4)

The regression coefficients in Equations 3 and 4 are not exactly the same, because different sets of compounds were used to develop these equations. Note that the difference in the intercepts in Equations 3 and 4 is caused by the different units associated with V. fischeri EC50 values in these equations. Also, note that most V. fischeri data used in our study are 5-min EC50 values, whereas 15-min EC50 values were used in the study of Gunatilleka and Poole [14] to develop Equation 4. However, the use of EC50 values obtained at different contact times should not have a significant impact on the regression coefficients, because Schultz et al. [13] demonstrated that for nonreactive compounds, the correlation between V. fischeri 5-min EC50 values (independent variable) and 15-min EC50 values (dependent variable) has a slope of 1.01 and an intercept of 0.01. This implies, from a regression point of view, that the 5-min EC50 values and 15-min EC50 values are practically equivalent.

The regression coefficients in Equations 3 and 4 are essentially the same (noting the difference in the units of the EC50 values mentioned previously) except for the coefficient multiplying R. Comparing Equation 2 and Equations 3 and 4 it can be seen that the site of toxic action in Shk1 interacts less easily with lone-pair electrons of the toxicant molecules compared to the site of toxic action in V. fischeri, as indicated by the absence of a term containing R in Equation 2 and the presence of such a term in Equations 3 and 4. The site of toxic action in Shk1 is approximately as hydrogen-bond acidic as that in V. fischeri, as indicated by similar regression coefficients for ∑β^H in Equations 2, 3, and 4. Finally, the site of toxic action in Shk1 is less cohesive than that of V. fischeri, and cavities can be more easily formed, both of which are indicated by a larger regression coefficient for V_x in Equation 2 than those in Equations 3 and 4.

The results of our study indicate that V_x is a significant factor that determines the nonspecific toxicity of organic compounds to Shk1. This conforms, to some extent, with the critical volume hypothesis that was proposed first by Overton [25] and later by Mullins [26]. The critical volume hypothesis states that when the volume fraction of a toxicant in the membrane reaches the critical level, narcosis occurs. A model that is based on the critical volume hypothesis describes the narcotic effect as proportional to the critical volume of a toxicant at the site of toxic action. Zhao et al. [27] used this model, which is shown by Equation 5, to model nonpolar narcotic toxicity to several aquatic species:

(5)

where E represents the narcotic effect, K_C is a proportionality constant, V_C is the critical volume of a toxicant at the site of toxic action, V_M is the molar volume of the toxicant, [A] is the concentration of the toxicant at the site of toxic action, and V_T is the total volume of the site of toxic action. As stated previously, our results conform, to some degree, with the critical volume hypothesis in that V_X, the model descriptor used in the present study, has been shown to correlate highly with V_M [28]. Therefore, V_X can be used in Equation 5 instead of V_M without altering the validity of the equation. However, our results deviate from the critical volume hypothesis, because our results show that the hydrogen-bond basicity is a significant factor, in addition to volume-related factor(s), in determining the nonspecific toxicity of organic compounds to Shk1.

Some researchers have used both hydrophobicity (log K_ow) and hydrogen-bonding descriptors to model the toxicity of polar and nonpolar narcotic compounds [20, 29, 30]. We did not include log K_ow in our model, because a preliminary multiple regression with backward variable elimination showed that log K_ow was not statistically significant and, therefore, could be removed from the regression (not shown). This is not surprising, because both V_X (used in our model) and log K_ow can be thought to characterize the ability of a toxicant to move from the aqueous phase to the cell membrane.

As mentioned, the present study was motivated by the difficulty of using a simple log K_ow-based model to properly describe the toxicity of toxicants acting by mechanisms of polar and nonpolar narcosis. To demonstrate the superiority of the solvation parameter model developed in this study, we also attempted to model the toxicity of the compounds shown in Table 1 using log K_ow as the sole model descriptor. A linear regression resulted in an r value of 0.78, much lower than the r_adj value (0.91) associated with Equation 2, which takes into account having more than one model predictor in Equation 2. In addition to a lower r value, regressing log (1/EC50) on log K_ow also yielded an MSE value of 0.32, more than twice that associated with Equation 2. Consequently, compared to a log K_ow-based model, the solvation parameter model yields better modeling capacity when polar and nonpolar narcosis toxicity to Shk1 are modeled together and offers greater prediction accuracy.

Values of solute descriptors used in the model developed in the present study are available for a number of compounds. Regarding compounds for which values of the descriptors are not available, convenient methods to calculate them have been published [28, 31]. Therefore, the nonspecific toxicity of organic compounds to Shk1 can be predicted using the model we developed, which provides a useful tool in predictive toxicology. The capacity to predict toxicity to Shk1 is valuable because of the similarity between the response of Shk1 to toxicants and that of the activated sludge. The QSAR model based on solvation parameters that we developed in the present study provides a convenient way for wastewater treatment plant operators and engineers to estimate the nonspecific toxicity of organic compounds without having to distinguish between polar and nonpolar narcosis.

CONCLUSIONS

We developed a solvation parameter model (Equation 2) to describe and predict the nonspecific toxicity (polar and nonpolar narcosis) of organic compounds to the luminescent bacterium Shk1. The statistically significant model descriptors were the McGowan's characteristic volume (V_x) and the hydrogen-bond basicity (∑β^H). The predictive capacity of the model was indicated by a small MSE value of 0.14 associated with the multiple regression that was performed to obtain the model. The model (Equation 2) is similar to the solvation parameter models developed for V. fischeri, but it does not include a term containing excess molar refraction (R). Because the values of the model descriptors are either reported in the literature or can be calculated, this model can be used as a tool to predict the nonspecific toxicity of organic compounds to Shk1.