Environ. Sci. Technol. 2003, 37, 4955-4961
Applicability and Limitation of QSARs for the Toxicity of Electrophilic Chemicals A N G E L A H A R D E R , †,‡,§ B E A T E I . E S C H E R , * ,† A N D R E N EÄ P . S C H W A R Z E N B A C H † , ‡ Swiss Federal Institute for Environmental Science and Technology (EAWAG), P.O. Box 611, CH-8600 Duebendorf, Switzerland, and Swiss Federal Institute of Technology (ETH), P.O. Box 611, CH-8600 Duebendorf, Switzerland
The appropriate selection and application of quantitative structure-activity relationships (QSARs) for the prediction of toxicity is based on the prior assignment of a chemical to its mode of toxic action. This classification is often derived from structural characteristics with the underlying assumption that chemically similar compounds have similar mechanisms of action, which is often but not necessarily the case. Instead of using structural characteristics for classification toward a mode of toxic action, we used Escherichia coli based bioanalytical assays to classify electrophilic chemicals. Analyzing a series of reactive organochlorines, epoxides, and compounds with an activated double bond, three subclasses of reactive toxicity were distinguished: “glutathione depletion-related toxicity”, “DNA damage”, and “unspecific reactivity”. For both subsets of specifically reacting compounds a direct correlation between effects and chemical reactivity was found. Reaction rate constants with either glutathione or 2′-deoxyguanosine, which was used as a model for complex DNA, served well to set up preliminary QSARs for either glutathione depletion-related toxicity or toxicity based on DNA damage in the model organism E. coli. The applicability of QSARs for electrophilic chemicals based on mechanistically relevant reaction rate constants is a priori limited to a small subset of compounds with strictly identical mechanism of toxic action and similar metabolic rates. In contrast, the proposed bioanalytical assays not only allowed the experimental identification of molecular mechanisms underlying the observable toxicity but also their toxicity values are applicable to quantitatively predict toxic effects in higher organisms by linear correlation models, independent of the assigned mode of toxic action.
Introduction Quantitative structure-activity relationships (QSARs) are widely used to predict toxicity from chemical structure and corresponding physicochemical properties. The development * Corresponding author phone: +41-1-823-5068; fax: +41-1-8235471; e-mail:
[email protected]. † EAWAG. ‡ ETH. § Present address: Swiss Federal Institute of Technology (ETH), Institute for Chemical and Bioengineering, ETH-Hoenggerberg, CH-8093 Zurich, Switzerland. 10.1021/es0341992 CCC: $25.00 Published on Web 09/25/2003
2003 American Chemical Society
and application of QSARs started with the prediction of toxicity caused by baseline toxicants (for a historical review, see ref 1). Species-specific toxicity of baseline toxicants was sufficiently described by a hydrophobicity factor (e.g., the octanol-water partition coefficient). Recent work showed that species differences of baseline toxicity can be explained by differences of membrane lipid content, and lethal membrane concentrations for algae, daphnia, and fish are nearly identical (2). QSARs using membrane-water partition coefficients are thus mechanistically meaningful and are based on a descriptor describing the crucial event leading to observed toxicity, the interaction of the toxicant with its target site, the biological membrane. A prerequisite for a correct predictive assessment of the toxicity of a chemical by using QSARs is the accurate assignment of the mode of toxic of action. Structural characteristics (3) and physicochemical properties of a chemical are helpful information to identify baseline toxicants and to discriminate them from potentially reactive chemicals (4). In contrast to baseline toxicants, the toxicity of reactive chemicals is determined by the intrinsic reactivity of the toxicant and the target occupation (5, 6). Important targets for electrophilic chemicals are nucleophilic sites (e.g., -SH, -NH2, -OH) in peptides and proteins as well as DNA, which may either have approximately constant cellular concentration (DNA) or are subject to a regular turnover (peptides and proteins). Toxicity of reactive chemicals is therefore time-dependent, and effects are irreversible unless defeated by an active cellular defense system, whereas baseline toxicity is reversible and occurs whenever a critical membrane concentration is exceeded, independent of the time of exposure (7). Electrophiles may react with biological nucleophiles through various mechanisms including nucleophilic substitution, Schiff’s base formation, or Michael addition (8). Thus a shared mechanistic basis, which is a prerequisite for a common QSAR, cannot be fulfilled a priori, which is reflected by numerous QSARs that are restricted to narrow groups of structurally related or congeneric chemicals that are likely to react according to the same reaction mechanism (1). However, structurally related nucleophiles do not necessarily have the same reactive mode of toxic action, nor do they necessarily react according to the same reaction mechanism with different nucleophiles (9, 10). These are additionally complicating factors that may hamper setting up reliable QSARs with a sound mechanistic basis. QSARs for prediction of toxicity of reactive chemicals typically account for the intrinsic reactivity by measured or calculated reaction rate constants. However, no parameter explicitly accounts for the target site occupation that is influenced by toxicokinetic processes such as uptake and metabolism. Thus, the improvement of the quality of QSARs for certain reactive chemicals, for example, for epoxides (11), achieved by adding a hydrophobicity parameter was in fact explained by the influence of uptake. However, it remains unclear why reactive chemicals with other reactive moieties (e.g., organochlorines) but the same range of hydrophobicity are satisfactorily modeled without this additional parameter (12, 13). Of course, it is always possible that within a given test set certain properties related to uptake and metabolisms are constant, for example, uptake is not rate-limiting or catalytic enhancement of reactivity due to the glutathione S-transferases is proportional for all compounds. We have chosen bacteria as model systems to study the direct effects of electrophilic chemicals on nucleophilic biological targets because reactions with target sites are presumably less influenced by toxicokinetic processes. This VOL. 37, NO. 21, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4955
TABLE 1. Toxicity Values for E. coli, Algae, Daphnia, and Fish; Mode of Action Classification; Octanol-Water Partition Coefficients; and Reaction Rate Constants with Glutathione (GSH) and 2′-Deoxyguanosine (GUA) electrophilea
log EC50 E. coli (mM)b
ACR IBA HEA EA ACN ACA EPOX SOX EOX NBCl DClB BCl DClP NOX EPI MVIN
-1.78 0.16 0.24 0.25 0.73 1.89 0.45 0.56 1.67 -0.93 -0.42 -0.33 -0.29 -0.26 0.53 1.51
log EC50 algae (mM)c
0.99 0.88 2.13
log EC50 daphnia (mM)d
log 96 h LC50 fish (mM)e
-2.99
-0.73 0.35
-3.60 -1.79 -1.38 -1.60 -0.80 0.19
-1.01
-1.42
0.03 0.05 1.14
-0.64
log 14 d LC50 fish (mM)f
-2.16 -0.31 -1.23 -0.34
-0.94
-3.16 -2.51 -1.99
-1.43
-2.15
mode of actiong
log KOWh
log kGSH (M-1 min-1) i
log kGUA (M-1 min-1) j
GSH GSH GSH GSH GSH GSH DNA DNA DNA ur url ur ur url url ur
-0.01 2.22 -0.21 1.32 0.25 -0.067 2.00 (1.97-2.02) 1.65 (1.64-1.67) 0.64 (0.62-0.68) 2.39 (2.38-2.40) 2.18 (2.14-2.21) 2.57 (2.51-2.64) 2.15 (2.14-2.15) 1.68 (1.67-1.69) 0.42 (0.42-0.43) 1.03 (1.01-1.05)
3.92 1.47 1.71 1.60 0.87 -0.33, -0.23k -0.42 0.11 -0.82 0.43 0.36 SN1m -0.70 -0.12 0.15 -0.55
-2.09 -2.15 -3.02 not detectable -1.38 SN 1 m S N 1m -1.89 -2.30 not detectable
a Abbreviation for compounds with an activated double bond: ACR, acrolein; IBA, isobutyl acrylate; HEA, 2-hydroxyethyl acrylate; EA, ethyl acrylate; ACN, acrylonitrile; ACA, acryl amide. Abbreviations for other electrophiles are given in Methods section. b EC50 values for growth inhibition of the microorganism Escherichia coli strain CC102 (10). c 2 h EC50 values for photosystem II inhibition of the algae Scenedesmus vacuolatus (16). d 48 h LC 50 values for the crustacean Daphnia magna from the EPA database ECOTOX (17). Data were derived in different systems: ACA, flowthrough experiments; ACN and SOX, semi-static; ACR and EPI, static. e 96 h LC50 values for the fish fathead minnow (Pimephales promelas) from the EPA database ECOTOX (17). Selected were lowest values reported with preference for data derived in flow-through systems (ACA, IBA, HEA, EA, ACN, ACA, and SOX). Data for BCl and EPI are from static experiments. f 14 d LC50 toxicity values for the fish guppy (Poecelia reticulata) from semi-static experiments. Data for EA and HEA from ref 18; data for SOX, EOX, and EPI from ref 11; data for DClB, BCl, and DClP from ref 12.g Mode of toxic action of electrophilic chemicals. Classification: GSH, GSH depletion-related toxicity; DNA, DNA damage; ur, unspecific reactivity (see ref 10). h log KOW for ACR, EA, ACN, and ACA recommended values from ref 19; values for HEA and IBA from ref 20. Log KOW for the other compounds are experimental results (see Methods) with 95% confidence intervals in parentheses. i Values for reaction with glutathione (GSH) for ACR to ACA measured at 20 °C, pH 8.8 (13). Values for all other compounds were determined at 30 °C, pH 7.65 (9). j Values for reaction with 2′-deoxyguanosine (GUA) measured at 30 °C at pH 7.65 (9). k Measured at 30 °C, pH 7.65, and comparable to the reaction rate constant measured at 20 °C and pH 8.8 (13). This indicates that the higher temperature compensates the higher proportion of the glutathione anion at higher pH, which allowed a combined use of kGSH values from both series. l Correlation to reaction rate constants indicate the higher importance of DNA damage as compared to GSH depletion-related toxicity (eq 2 and Figure 2). m Reaction of electrophile follows first-order nucleophilic substitution (SN1) and is therefore independent of the concentration of nucleophile the GSH or GUA, respectively.
assumption was made because catalysis by glutathione S-transferase is less efficient in bacteria than in higher organisms (14, 15) and DNA is not compartmentalized and less shielded by proteins than in higher organisms. The set of bacterial bioanalytical assays presented in ref 10 allowed a clear classification of a test set of epoxides, reactive organochlorines, and compounds with an activated double bond into three subclasses of reactive toxicity: “GSH depletion-related toxicity”, “DNA damage”, and “unspecific reactivity”, which is governed by both of the aforementioned mechanisms. The results from these bioanalytical assays can not only be exploited for classification according to mode of toxic action but may also be useful for setting up QSARs. We therefore explored the possibility to predict the toxicity data of Escherichia coli from physicochemical descriptors as well as the potential to use these easily available data to set up predictive models for higher organisms. We examined the correlation between the toxicity of each class of reactive chemical in E. coli, the octanol-water partitioning coefficient, and the appropriate reaction rate constant (i.e., the reaction rate constant toward GSH or 2′-deoxyguanosine as a model for DNA, respectively, as well as correlations between E. coli and algae, daphnia, and fish). The results were discussed in view of limitations and potential for use in predictive (eco)toxicological risk assessment.
Methods Data Sets. To examine the relationship of reaction rate constants, octanol-water partition coefficients (KOW), and toxicity of electrophiles with respect to their reactive mode of toxic action, toxicity values for four aquatic species groups were selected and complemented with reaction rate constants 4956
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 21, 2003
and KOW values as far as available (Table 1). Examined electrophilic classes comprise compounds with an activated double bond, epoxides and reactive organochlorine compounds. Toxicity values include concentrations resulting in 50% growth inhibition (EC50) of E. coli, EC50 values of the inhibition of the photosystem II quantum yield measured by chlorophyll fluorescence of the unicellular algae Scenedesmus vacuolatus after 2 h of exposure, concentrations causing 50% lethality (LC50) after 48 h for the crustacean Daphnia magna, 96 h LC50 for the fish fathead minnow (Pimephales promelas), and 14 d LC50 values of the guppy (Poecilia reticulata). Physicochemical descriptors include the octanol-water partition coefficient and second-order reaction rate constants with the nucleophiles glutathione (GSH) and 2′-deoxyguanosine (GUA). KOW values were measured in this study (details given below). The sources of the other data are given in Table 1. Additionally, the classification of the examined chemicals toward the reactive mode of toxic action is given (for classification, refer to ref 10). The classification is based on growth differences observed for a set of bacterial strains either lacking the nucleophile GSH or lacking DNA repair systems. Comparison of toxicity observed in those strains to the toxicity of their unaltered parent strains allowed the distinction of three modes of action for electrophilic chemicals: DNA damage, glutathione depletion-related toxicity, and unspecific reactivity. Determination of Octanol-Water Partition Coefficients. Octanol-water partition coefficients were measured for the following reactive organochlorine compounds and epoxides: benzyl chloride (BCl, CAS Registry No. 100-44-7), 4-nitrobenzyl chloride (NBCl, CAS Registry No. 100-14-1), 2,3-dichloro-1-propene (DClP, CAS Registry No. 78-88-6), trans-1,4-dichloro-2-butene (DClB, CAS Registry No. 110-
57-6), styrene oxide (SOX, CAS Registry No. 96-09-3), 2-(4nitro-phenyl)oxirane (NOX, CAS Registry No. 6388-74-5), (2,3epoxypropyl)benzene (EPOX, CAS Registry No. 4436-24-2), 1,2-epoxybutane (EOX, CAS Registry No. 106-88-7), epichlorohydrin (EPI, CAS Registry No. 106-89-8), and 2-methyl2-vinyloxirane (MVIN, CAS Registry No. 1838-94-4). BCl, NBCl, EPI, and EOX were purchased from Fluka Chemie AG, Buchs, Switzerland. DClP and SOX were obtained from Sigma-Aldrich Chemie AG, Steinheim, Germany. DClB, NOX, EPOX, MVIN, and 1-octanol (CAS Registry No. 111-87-5) were obtained from Aldrich Chemical Company Inc., Milwaukee, WI. All chemicals were of highest purity available (g95%) and were used as received. 1-Octanol of ACS reagent grade and deionized water filtered by a Millipore filter system were mutually saturated for 12 h on a shaking incubator. Octanol-saturated water and water-saturated octanol phases separated within 4 d. KOW values were determined in duplicate in 1:5.3, 1:2.6, and 1:1.3 water:octanol mixtures using serum vials with crimped viton rubber stoppers. Liquid electrophiles were directly added to the mixture. Different masses of solid electrophiles (i.e., NBCl and NOX) were weighed prior to addition of octanol and water. As some of the examined electrophiles hydrolyze quite quickly (9), the whole procedure of determination of partition coefficients had to be very fast but sufficiently long for the compounds to attain equilibrium. The following procedure satisfied both requirements: Mixtures with liquid electrophiles were vigorously mixed for six times 20 s on a vortex; 30 min of ultrasonic treatment lead to dissolution of the slowly hydrolyzing solid electrophiles. Thereafter, similar to the liquid electrophiles, they were additionally mixed on a vortex. Subsequently, the mixtures were centrifuged for 2 min at 2500g, leading to a fast separation of phases. Samples of the octanol and water phases were taken using syringes with very fine cannulas. To avoid traces of octanol in the water phase, air was gently expelled while passing the octanol layer. Samples were analyzed immediately afterward. Aliphatic electrophiles (i.e., DClP, DClB, EOX, EPI, and MVIN) were analyzed by GC-FID detection (GC 8000, Fision Instruments, Milano, Italy) using a Stabilwax column (30 m × 0.32 mm, 1 µm; BGB Analytik, Anwil, Switzerland) with direct on-column injection. The aromatic electrophiles BCl, NBCl, SOX, NOX, and EPOX were analyzed by HPLC with UV detection (pump M480, Gina 160 autosampler, Gynkotek, Germering, Germany; 875-UV detector, Jasco, Gross-Umstadt, Germany) using C-8 (LiChrosphere, 125 × 4 mm, 5 µm spheres; Merck, Darmstadt, Germany) and C-18 (Nucleosil, 250 × 4 mm, 5 µm spheres; Macherey-Nagel, Dueren, Germany) reversed-phase columns with differing methanolwater mixtures as mobile phase. Concentrations in the water phase could be measured directly. For HPLC analysis, the octanol phase had to be diluted with methanol. GC measurements of octanol phases could be done directly using a very slow on-column injection. KOW values were calculated as the ratio of the concentration in the octanol phase to the concentration in the water phase. Reported values are average values from six octanol-water mixtures.
Results and Discussion Correlation of Bacterial Toxicity to Reaction Rate Constants with Biological Nucleophiles. Based on the assignment of a chemical toward a reactive mode of toxic action (Table 1), the relationship of bacterial toxicity toward the corresponding reaction rate constants was examined. Toxicity values of compounds with glutathione depletion-related toxicity log linearly correlated with the chemical reaction rate constant with glutathione kGSH (eq 1 and Figure 1). Toxicity values showed no dependence on the hydrophobicity term log KOW as a single descriptor (n ) 6, r 2 ) 0.11, F ) 0.5). A combination of kGSH and log KOW did not improve the quality of the
FIGURE 1. Plot of log EC50 values of E. coli of specifically reacting compounds vs logarithmic reaction rate constants with glutathione (kGSH) (b) and with 2′-deoxyguanosine (kGUA) (9).
FIGURE 2. Plot of log EC50 values of E. coli of unspecifically reacting compounds vs logarithmic reaction rate constants with glutathione (kGSH) (epoxides 0, organochlorines 4) and with 2′-deoxyguanosine (kGUA) (epoxides 9, organochlorines 2). correlation (n ) 6, r 2 ) 1.00, F ) 322). The log linear correlation of toxicity to kGSH suggests that the chemical reaction of examined compounds with glutathione is the limiting step in a series of events leading to observed toxicity:
log
(
)
EC50 E. coli ) -0.87(( 0.04) × mM kGSH log + 1.60(( 0.08) (1) -1 M min-1
(
)
n ) 6, r 2 ) 0.99, F ) 440 Correspondingly, the toxicity of DNA damaging compounds (i.e., EPOX, SOX, and EOX) log linearly correlated with the reaction rate constants with 2′-deoxyguanosine (kGUA). However, as the data include two compounds of comparable reactivity and toxicity, the correlation is governed by two points (Figure 1). Therefore, it is shown only to illustrate the case and not to actually derive a valid QSAR. These results are consistent with expectation that the dominant mode of toxic action is clearly assigned, and the process triggering the toxicity can be described by physicochemical parameters. The main prerequisites for setting up QSAR models are thus fulfilled. Unspecifically reactive compounds exhibit multiple toxic mechanisms. It is not a priori clear which, if any, toxic mechanism is dominant. However, when plotting the data pairs EC50 - log kGUA and EC50 - log kGSH in the correlations derived for the clearly assigned data sets, it becomes evident that there is no correlation with kGSH but that the EC50 values of DClB, EPI, and NOX (Figure 2) reacting with GUA lay well on the extension of the line combining reactivity and toxicity of DNA-damaging compounds (i.e., EOX, SOX, and EPOX) VOL. 37, NO. 21, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4957
TABLE 2. Enhancement of Aliphatic and Aromatic Nucleophilic Substitution and Michael Addition by Glutathione S-Transferases from Rat Livera substance
reaction mechanism with GSH
enzymatical rate/chemical rate
4-nitrobenzyl chloride 1-chloro-2,4-dinitrobenzene 1,2-dichloro-4-nitrobenzene 1-(p-nitrophenoxy) propane-2,3-oxide t-4-phenyl-3-buten-2-one
nucleophilic substitution of chlorine nucleophilic aromatic substitution of chlorine nucleophilic aromatic substitution of chlorine nucleophilic substitution of epoxide Michael addition on activated double bond
1.3 × 104 8.6 × 103 3.6 × 103 1.8 × 103 2.0 × 103
a
Adapted from figure and text (26).
(Figure 1). These three compoundssthe organochlorine DClB and the epoxides EPI and NOXshappen to be the only three compounds with a statistically significant difference of growth between the DNA repair-sufficient and -deficient strain (10) and a detectable reaction rate with GUA (9). The good agreement of toxicity and reaction rate constant with GUA of these compounds suggests that toxicity is more influenced by the reaction with DNA bases than with GSH. In cases of unspecific reactivity, the comparison of toxicity and reaction rate constants thus might be a method to detect which reaction determines toxicity. Equation 2 shows the correlation of E. coli EC50 values to the reaction rate constant with GUA for the combined data set of EOX, SOX, and EPOX with EPI, NOX, and DClB. The reaction rate constant with GUA was sufficient to describe the toxicity, and inclusion of a hydrophobicity term did not improve the model (n ) 6, r 2 ) 0.94, F ) 24 for correlation with kGUA and log KOW).
log
(
)
EC50 E. coli ) - 1.34(( 0.18) × mM kGUA log -1 - 2.43(( 0.40) (2) M min-1
(
)
n ) 6, r 2 ) 0.93, F ) 52 The unspecifically reacting compound MVIN could not be included in the comparison because due to the fast hydrolysis, the reaction rate with GUA could not be determined. Thus, it can be concluded that the toxicity of all examined epoxides (EPOX, SOX, EOX, NOX, and EPI) is primarily based on the reaction with DNA and that the toxicity of all examined compounds with an activated double bond (ACR, IBA, HEA, EA, ACN, and ACA) is triggered by reaction with glutathione. These results are in line with the hard and soft acid and base concept (21), confirming the hypothesis that hard acids (electrophiles) (e.g., epoxides) tend to react with hard bases (nucleophiles) (e.g., DNA) and that soft acids (electrophiles) (e.g., compounds with an activated double bond) tend to react with soft bases (nucleophiles) (e.g., glutathione) (22). For DClB, the direct correlation between the reaction rate constant kGUA and toxicity implies that DClB is a directly alkylating chemical, which does not need prior activation, for example, by epoxidation of the allyl bond (23) or by formation of a reactive glutathione conjugate (24), as was observed for other bifunctional organochlorines. Note that the difference in reactivity between the nucleophiles GSH and GUA is not constant for different electrophiles. This seems trivial, as it can be readily deduced from the Swain-Scott relationship (25), which describes the dependence of second-order reaction rate constants on a nucleophilicity constant (n) and a substrate (electrophile) constant (s). Since the difference of nucleophilicity between GSH and GUA for the examined compounds can be assumed to be constant, the difference of reaction rate constants is higher for electrophiles with a large substrate constant than for compounds with a small substrate constant. Thus, not only the intrinsic reactivity of electrophiles but also the selected nucleophiles influence QSARs set up to describe 4958
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 21, 2003
toxic effects of electrophiles. Reaction rate constants with different nucleophiles are consequently not interchangeable. Nevertheless, this simple rule is often overlooked when rate constants for reactions with other nucleophiles are used as surrogates for target molecules in cells. Therefore, it is mandatory to select the “right” nucleophile for the description of toxic effects. To get a closer link to observed toxicity of electrophiles, reactions rates with target nucleophiles such as GSH or GUA thus seem to be most suitable descriptors. For a correct application of a QSAR for reactive chemicals, the knowledge about the reactive mode of toxic action and measurement of the appropriate chemical reaction rate constant might not be sufficient. It is possible that the toxicity of reactive chemicals is additionally influenced, for example, by catalysis of glutathione S-transferases (GSTs) or hydrolases. We have only limited knowledge about the efficiency of such detoxification processes. For GSTs, one example indicates that the influence of GST-catalyzed reaction on the correct application of a QSAR is small given a high range of chemical reaction rate constants. Comparing a chemical reaction with GSH with an enzymatically catalyzed reaction by GSTs of very dissimilar reaction mechanisms (26), the differences of the enhancement factor are less than a factor of 10 (Table 2). Thus, model systems describing the biological reaction with GSH with chemical reaction rate constants are applicable as long as the impact of differences in GST catalysis efficiency for different substrates is negligible, because the chemical reaction rate constants itself span more than 1 order of magnitude. This is the case for the series of the reaction of compounds with an activated double bond with glutathione. Their reaction rate constants span a range of 104. However, when the range of chemical reaction rate constants is small, catalysis by GSTs may change the relative ranking of reactivity. For the epoxides EOX, EPI, and MVIN that react according to second-order nucleophilic substitution (9), chemical reaction rate constants increase in the order EOX < MVIN < EPI and differ by a factor of 12. In contrast, enzymatic catalysis of a GST from the microorganism Rhodococcus sp. for EOX and EPI was three times lower than that for MVIN (27). In case of small ranges of reaction rate constants, QSARs can only then successfully be applied when the examined compounds are proportionally catalyzed to their chemical reactivity. Naturally, GSTs could have an even higher impact on the correct application of QSAR for the toxicity in a given organism if the GSTs of that organism have substrate specificities (see discussion below) and do not catalyze particular electrophiles. Other detoxification processes, for example, the detoxification of epoxides by epoxides hydrolases (28, 29), directly influence the target site concentration of the electrophiles. This example demonstrates clearly that processes not directly linked to the reactive mode of toxic action may hamper direct correlations between chemical reactivity and toxicity. QSARs for reactive chemicals that are intended to describe toxicity by chemical reaction rate constants can therefore be applied for only a limited range of compounds for which the mode of toxic action and a rough perception of metabolic
FIGURE 3. Linear correlation with 95% confidence interval between EC50 growth inhibition of E. coli and EC50 of photosystem II inhibition of Scenedesmus vacuolatus. The broken line marks the 1:1 correlation. influences are known. Because the correct application of such QSARs already requires important toxicological knowledge, the claim that QSARs provide a predictive tool must, at least for reactive chemicals, be challenged. Alternative Approaches for Ecotoxicological Risk Assessment of Electrophilic Chemicals. As the development of QSARs for toxicity of electrophiles based on reaction rate constants is a questionable endeavor, we examined if bacterial toxicity that is even simpler to measure than reaction rate constants is correlated to toxicity in higher organisms and, thus, can be used as a risk assessment tool. We therefore correlated available consistent toxicity data sets for algae, daphnia, and fish with toxicity data for E. coli. Most reliable data for easy hydrolyzing and partly volatile reactive chemicals are short-term tests (e.g., as the newly developed algae test (16) and toxicity tests derived in flow-through systems). For six compounds for which data were available, a comparison between bacterial toxicity and algae toxicity was made. As can be seen from the intercept of the linear correlation between algal and bacterial toxicity (eq 3 and Figure 3), the bacterial toxicity end point was slightly more sensitive. Indicated by the unit slope of the regression, the relative toxicity of the examined compounds however is identical in both organisms. This observed sensitivity difference does not allow the conclusion to be drawn that bacteria are more sensitive than algae as the two end points are arbitrarily chosen. Noteworthy is that bacterial tests are performed within a time frame of up to 3 generation times of the E. coli strain used (3 h of exposure, specific growth rate µ ) 0.57 h-1, experiments not shown), whereas the duration of the algal tests span only 10% of the generation time (2 h of exposure, specific growth rate µ ) 0.039 h-1 (16)). Thus, the sensitivity difference might reflect the relative differences of exposure times:
(
)
EC50 S. vacuolatus log ) 1.02(( 0.07) × mM EC50 E. coli log + 0.45(( 0.06) (3) mM
(
)
n ) 6, r 2 ) 0.98, F ) 225
FIGURE 4. Linear correlation with 95% confidence interval between EC50 growth inhibition of E. coli and 48 h LC50 of Daphnia magna. The broken line marks the 1:1 correlation. and Figure 4), indicating the constant relative toxicity of the examined compounds:
log
(
)
EC50 Daphnia magna ) 0.91(( 0.06) × mM EC50 E. coli log - 1.36(( 0.09) (4) mM
(
)
n ) 5, r 2 ) 0.99, F ) 215 Comparison of toxicity in fish with toxicity in E. coli was made with 96 h LC50 values of P. promelas and 14 d LC50 values of Po. reticulata. Because toxicity values of EPI and BCl were determined in static experiments and EPI and especially BCl hydrolyze quite rapidly (9), the data points were omitted from the correlation to 96 h toxicity of the fish P. promelas but are still shown in Figure 5a. Whereas for data from flow-through tests a sound correlation could be set up (eq 5), the correlation with toxicity data from 14 d semistatic tests was poor (eq 6 and Figure 5b), probably reflecting depletion of fast hydrolyzing substances, like BCl. Toxicity of SOX in the 14 d semi-static tests was inexplicably lower than in the 96 h experiments. Fish toxicity end points were more sensitive as could be deduced from the intercept of the linear correlation between fish and bacterial toxicity. Indicated by the slope of the regression, the relative toxicity of examined compounds however is identical for both fish species and bacteria:
(
log
)
LC50 P. promelas ) 1.04(( 0.07) × mM EC50 E. coli log - 1.79(( 0.07) (5) mM
(
)
n ) 7, r 2 ) 0.96, F ) 242
(
log
)
LC50 Po. reticulata ) 1.06(( 0.17) × mM EC50 E. coli log - 2.25(( 0.17) (6) mM
(
)
n ) 8, r 2 ) 0.86, F ) 38 Comparison of toxicity in Daphnia magna with toxicity in E. coli is based on 48 h LC50 values for the compounds ACR, ACN, ACA, SOX, and EPI. Despite the mixed quality of data for daphnia derived in static, semi-static and flowthrough tests, a linear 1:1 correlation could be set up (eq 4
The compounds in the linear correlation for algae (eq 3) include DNA-damaging compounds (EPOX, SOX, EOX, DClB, and EPI) and an unspecific reactive compound (BCl). Daphnia and fish toxicity data also comprise compounds from different VOL. 37, NO. 21, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4959
FIGURE 5. Linear correlation with 95% confidence interval between EC50 growth inhibition of E. coli and (a) 96 h LC50 of the fish Pimephales promelas (b flow through tests, O static tests) and (b) 14 d LC50 of the fish Poecilia reticulata (9 semi-static tests). The broken line marks the 1:1 correlation. reactive modes of toxic action. The log linear regressions of bacterial toxicity to toxicity in higher organisms show unit slope and thus indicate the same relative sensitivity in all here investigated organisms. This constant relative sensitivity is independent of the electrophile’s mode of toxic action, which suggests that the mode of toxic action is the same in all organisms. Taken together, the correlations given in eqs 3-6 provide a first indication that the E. coli toxicity test may indeed be used as a predictive tool in ecotoxicological risk assessment. Further studies are clearly required to confirm the toxicity correlation between different organisms and to examine the reactive modes of toxic action in higher organisms on a molecular basis. A study by Freidig et al (13) supports the assumption of consistent mode of toxic action. He observed that compounds with an activated double bond, which are assigned to GSH depletion-related toxicity, cause a critical depletion rate constant of glutathione resulting in 50% lethality of fish, which suggests that the mode of toxic action in bacteria and fish remains the same. In a toxicity study of EA in rat no DNA adducts but protein adducts were found (30). Van Welie et al. (22) found that exposure to ethylene oxide and 1,2dibromoethane in rat resulted in ratios of 1:10 and 1:107 of DNA to GSH adducts, respectively. This underlines that DNA interactions of epoxides are retained in higher organisms and indicates that haloalkanes might, as haloalkenes, react unspecifically with both important biomolecules. However, a comparison of cytotoxicity and glutathione depletion in algae and E. coli indicates a potential influence of metabolism on the occurrence of reactive modes of toxic action. Comparing concentrations of cytotoxicity (Figure 3) for algae and bacteria, algae were found to be less sensitive than bacteria. Glutathione depletion in algae (16), however, occurred at much lower toxicant concentrations than in bacteria (10). Whereas in bacteria no GSH depletion was observed for DClB and SOX, both compounds depleted GSH in algae. Testing GSH depletion in E. coli was however hampered in the case of DClB due to poor water solubility. A GSH-deficient strain was much more sensitive toward DClB than a GSH-competent E. coli strain (10). One explanation for the discrepancy of the GSH-depleting potency in algae and E. coli might be the influence of GSTs. Bacteria are known to possess only small concentrations of GSTs (14), catalyzing the reactions of GSH with electrophiles. But not only the concentrations are low, additionally bacterial GSTs were found to have lower activities than, for example, mammalian GST (15), which enhance chemical reaction rate constants up to 104-105 times (31). Thus, it can be assumed that the higher susceptibility of the GSH homeostasis in algae as compared to bacteria is determined by the activity of GSTs. 4960
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 21, 2003
An additional point is the difference of substrate specificity. For example, whereas mammalian GSTs effectively catalyze the substitution of SOX, GSTs of E. coli were found to have no activity with SOX (15). Reaction of SOX with glutathione in higher organisms was found to be an important detoxification mechanism (32), which reduces but not inhibits the extent of toxic effects on DNA (33). Hazard Assessment of Electrophilic Chemicals. Toxicity of electrophilic chemicals is determined by their reaction with susceptible biological nucleophiles (i.e., peptides, proteins, and DNA). As was shown with a set of E. coli strains (ref 10 and this paper), epoxides tend to react with DNA while compounds with an activated double bond prefer the tripeptide GSH as a target. For other unspecific reactive compounds, reactions with both targets determine their toxicity. This clear differentiation of reactive modes of toxic action diminishes in higher organisms, because they may have a broader metabolic competence, as was shown for GSTs catalyzing the reaction of GSH with SOX. However, as was shown by multiple studies (34-37), the hazard of chemicals recognized as DNA-damaging chemicals to impair DNA remains. As was shown by correlations of toxicity of electrophilic chemicals between different organisms, the relative toxicity of chemicals of different mode of toxic action remains the same. This is an indication that the pattern of glutathione depletion-related toxicity and toxicity based on DNA damage is constant for microorganisms, algae, daphnids, and fish despite certain differences in metabolic transformation reactions. However, clear evidence is missing, and research on toxicity of electrophiles in higher organisms should concentrate on finding toxicity indicators that clearly link observable toxicity with mechanisms of toxic action. Despite lacking a mechanistic proof, the observed linearity (eqs 3-6) of toxicity in algae, daphnia, and fish provides a practical tool to predict toxicity with a simple bacterial test. QSARs based on reaction rate constants as descriptors are suitable as an exploratory tool but lack suitability as a predictive tool. Such QSARs are only valid for very small sets of chemicals that act not only according to the same mechanism in a given biological organism but also have identical chemical reaction mechanism. Consequently, correct assignment of a compound to the appropriate QSAR requires such detailed information on the mode of action and toxicokinetic processes that experimental data on a more integrative level (e.g., the E. coli based bioanalytical assays) are better suited as descriptors in predictive models.
Acknowledgments We are grateful to Joop L. M. Hermens (Utrecht University, The Netherlands) for helpful discussions and Jakov Bolotin
(EAWAG, Duebendorf, Switzerland) for laboratory assistance. Additional thanks to Kathrin Fenner, Zachariah Schreiber, and Rik I. L. Eggen (all EAWAG, Duebendorf, Switzerland) for reviewing the manuscript.
Literature Cited (1) Lipnick, R. L. In Aquatic Toxicology; Rand, G. M., Ed.; Taylor & Francis: London, Great Britain, 1995; pp 609-655. (2) Escher, B. I.; Schwarzenbach, R. P. Aquat. Sci. 2002, 64, 20-35. (3) Verhaar, H. J. M.; van Leeuwen, C. J.; Hermens, J. L. M. Chemosphere 1992, 25, 471-491. (4) EC. Technical Guidance Document in Support of Commission Directive 93/67/EEC on Risk Assessment for New Notified Substances and Commission Regulation (EC) No. 1488/94 on Risk Assessment for Existing Substances; Office for Official Publications of the European Communities: Luxembourg, 1996. (5) Legierse, K. C. H. M.; Verhaar, H. J. M.; Vaes, W. H. J.; De Bruijn, J. H. M.; Hermens, J. L. M. Environ. Sci. Technol. 1999, 33, 917925. (6) Verhaar, H. J. M.; De Wolf, W.; Dyer, S.; Legierse, K. C. H. M.; Seinen, W.; Hermens, J. L. M. Environ. Sci. Technol. 1999, 33, 758-763. (7) Escher, B. I.; Hermens, J. L. M. Environ. Sci. Technol. 2002, 36, 4201-4217. (8) Hermens, J. L. M. Environ. Health Perspect. 1990, 87, 219-225. (9) Harder, A. Assessment of the Risk Potential of Reactive Chemicals with Multiple Modes of Toxic Action. Ph.D. Thesis, Swiss Federal Institute of Technology (ETH), Zurich, Switzerland, 2002. Accessible at http://e-collection.ethbib.ethz.ch./show? type)diss&nr)14966. (10) Harder, A.; Escher, B. I.; Landini, P.; Tobler, N.; Schwarzenbach, R. P. Environ. Sci. Technol. 2003, 37, 0000-0000. (11) Deneer, J. W.; Sinnige, T. L.; Seinen, W.; Hermens, J. L. M. Aquat. Toxicol. 1988, 13, 195-204. (12) Hermens, J. L. M.; Busser, F.; Leeuwach, P.; Musch, A. Toxicol. Environ. Chem. 1985, 9, 219-236. (13) Freidig, A.; Verhaar, H. J. M.; Hermens, J. L. M. Environ. Sci. Technol. 1999, 33, 3038-3043. (14) Penninckx, M. J.; Elskens, M. T. Adv. Microb. Physiol. 1993, 34, 239-301. (15) Kerklaan, P. M.; Zoetemelk, C. E. M.; Mohn, G. R. Biochem. Pharmacol. 1985, 34, 2151-2156. (16) Niederer, C.; Behra, R.; Harder, A.; Schwarzenbach, R. P.; Escher, B. I. Environ. Toxicol. Chem. (in press).
(17) U.S. EPA. http://www.epa.gov/ecotox (accessed on November 17, 2002). (18) Hermens, J. M. L.; Leeuwangh, P. Ecotoxicol. Environ. Saf. 1982, 6, 302-310. (19) Chemfate. Chemfate Database, http://esc.syrres/efdb/Chemfate.htm (accessed on November 21, 2002). (20) Tanii, H.; Hashimoto, K. Toxicol. Lett. 1982, 11, 125-129. (21) Edwards, J. O.; Pearson, R. G. J. Am. Chem. Soc. 1962, 84, 16-24. (22) van Welie, R. T.; van Dijck, R. G.; Vermeulen, N. P.; van Sittert, N. J. Crit. Rev. Toxicol. 1992, 22, 271-306. (23) Thier, R.; Mu ¨ ller, M.; Taylor, J. B.; Pemble, S. E.; Ketterer, B.; Guengrich, F. P. Chem. Res. Toxicol. 1995, 8, 465-472. (24) Beaten, A.; Tafazoli, M.; Kirsch-Volders, M.; Geerlings, P. Int. J. Quantum Chem. 1999, 74, 315-355. (25) Swain, C. G.; Scott, C. B. J. Am. Chem. Soc. 1953, 75, 141-147. (26) Ketterer, B. Drug Metab. Rev. 1982, 13, 161-187. (27) van Hylckama Vieg, J. E. T.; Kingma, J.; Kruizinga, W.; Janssen, D. B. J. Bacteriol. 1999, 181, 2094-2101. (28) Fretland, A. J.; Omiecinski, C. J. Chem.-Biol. Interact. 2000, 129, 41-59. (29) Swaving, J.; de Bont, J. A. M. Enzyme. Microb. Technol. 1998, 22, 19-26. (30) Ghanayem, B. I.; Burka, L. T.; Metthews, H. B. Fundam. Appl. Toxicol. 1987, 9, 389-397. (31) Douglas, K. In Glutathione Conjugation: Mechanisms and Biological Significance; Sies, H., Ketterer, B., Eds.; Academic Press: London, Great Britain, 1988; pp 1-41. (32) Pacifici, G. M.; Warholm, M.; Guthenberg, C.; Mannervik, B.; Rane, A. Human Toxicol. 1987, 6, 483-489. (33) Shield, A. J.; Sanderson, B. Environ. Mol. Mutagen. 2001, 37, 285-289. (34) Giri, A. K. Mutat. Res. Rev. 1997, 386, 25-38. (35) Ehrenberg, L.; Hussain, S. Mutat. Res. 1981, 86, 1-113. (36) Uziel, M.; Munro, N. B.; Katz, D. S.; Vo Dinh, T.; Zeighami, E. A.; Waters, M. D.; Griffith, J. D. Mutat. Res. 1992, 277, 35-90. (37) Griem, H. In Biological Reactive Intermediates; Dansette, P. M., Snyder, R., Delaforge, M., Gibson, G. G., Griem, H., Jollow, D. J., Monks, T. J., Sipes, I. G., Eds.; Kluwer Academic/Plenum Publishers: New York, 2001; pp 715-722.
Received for review March 6, 2003. Revised manuscript received July 24, 2003. Accepted August 11, 2003. ES0341992
VOL. 37, NO. 21, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4961
