TECHNICAL BASIS FOR NARCOTIC CHEMICALS AND POLYCYCLIC AROMATIC HYDROCARBON CRITERIA. I. WATER AND TISSUE

Transcription

1 Environmental Toxicology and Chemistry, Vol. 19, No. 8, pp , SETAC Printed in the USA /00 $ TECHNICAL BASIS FOR NARCOTIC CHEMICALS AND POLYCYCLIC AROMATIC HYDROCARBON CRITERIA. I. WATER AND TISSUE DOMINIC M. DI TORO,* JOY A. MCGRATH, and DAVID J. HANSEN HydroQual, Inc., 1 Lethbridge Plaza, Mahwah, New Jersey 07430, USA Environmental Engineering Department, Manhattan College, Riverdale, New York 10471, USA (Received 2 February 1999; Accepted 13 December 1999) Abstract A method is presented for developing water quality criteria (WQC) for type I narcotic chemicals in general and PAHs in particular. The criteria can be applied to any individual or mixture of narcotic chemicals using only the chemical s octanol-water partition coefficient K OW. It is derived from a database of LC50s comprising 156 chemicals and 33 species, including fish, amphibians, arthropods, mollusks, polychaetes, coelenterates, and protozoans. A target lipid model is proposed that accounts for variations in toxicity due to differing species sensitivities and chemical differences. The model is based on the idea that a target lipid is the site of action in the organism. Further, it is assumed that target lipid has the same lipid-octanol linear free energy relationship for all species. This implies that the slope of the log(lc50) log(k OW ) relationship is the same for all species. However, individual species may have varying target lipid body burdens that cause toxicity. The target lipid LC50 body burdens derived from concentration data in the water only are compared to measured total lipid LC50 body burdens for five species. They are essentially equal, indicating that the target lipid concentration is equal to the total extracted lipid concentration. The precise relationship between partitioning in target lipid and octanol is established. The species-specific body burdens are used to determine the WQC final acute value, i.e., the 95-percentile level of protection. An acute-to-chronic ratio is used to compute the body burden corresponding to the WQC final chronic value, which is the procedure used to derive the U.S. Environmental Protection Agency water quality criteria. The criteria are expressed either as dissolved concentrations in the water column or as tissue concentrations. Keywords Critical body burden Target lipid model Equilibrium partitioning INTRODUCTION Our motivation for developing a refined model of narcotic toxicity for aquatic organisms is to apply the results to mixtures of Polycyclic aromatic hydrocarbons (PAHs) in sediments. Type I narcotic chemicals are nonionic organic chemicals with a similar mode of toxic action, i.e., narcosis [1]. Since PAHs are expected to be type I narcotic chemicals [2,3], the toxicity of mixtures of PAHs should be additive. Thus, a model that can describe the toxicity for all type I narcotics, including PAHs, and that considers the species sensitivity explicitly can be used to develop water quality criteria following the U.S. Environmental Protection Agency (U.S. EPA) technical guidelines [4] and sediment quality guidelines using equilibrium partitioning (EqP) [5]. A comprehensive model for type I narcotic chemicals that considers multiple species has been presented by Van Leeuwen et al. [6]. Quantative structure activity relationships (QSARs) for individual species are developed and a species sensitivity analysis is performed. The analysis and model presented below is similar. The key differences are the use of a single universal slope for the log(lc50) versus log(k OW ) QSAR for all the species, the inclusion of corrections for chemical classes that are slightly more potent than baseline narcotics, and the interpretation of the y-intercepts as the species-specific critical body burdens for narcosis mortality. BODY BURDEN MODEL The initial quantitative structure activity models for narcotic toxicity relied on correlations of log(lc50) and log(k OW ) * To whom correspondence may be addressed (dditoro@hydroqual.com). [7,8]. An interesting and important interpretation of this inverse relationship that relates the toxicity to chemical body burden has been presented by McCarty et al. [9] and proceeds as follows. The relationship between the LC50 (mmol/l) and K OW for fish is approximately log(lc50) log(k ) 1.7 OW (1) For each LC50, a fish body burden C Org ( mol/g wet wt mmol/kg wet wt) corresponding to narcosis mortality can be computed using a bioconcentration factor (BCF) (L/kg), which is defined as the ratio of the chemical concentration in the organism C Org to the chemical concentration in the water C W (mmol/l): C Org BCF (2) C W Using the BCF, the organism concentration corresponding to the LC50, which is referred to as the critical body burden and is denoted by C Org, can be computed using C* BCF LC50 Org (3) The superscript * indicates that it is a critical body burden corresponding to the LC50. The BCF also varies with K OW. For fish, the relationship is log(bcf) log(k ) 1.3 OW (4) Therefore, the critical body burden corresponding to the LC50 for fish narcosis can be computed using the narcosis LC50 (Eqn. 1) and the BCF (Eqn. 4) as 1951

2 1952 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. log(c* ) log(bcf) log(lc50) Org log(k OW) 1.3 log(k OW) or (5) C* 2.5 mol/g wet weight (6) Org Thus, McCarty et al. [9] rationalize the relationship between LC50s and K OW by suggesting that it is the result of a constant body burden of narcotic chemical that causes mortality. If the fraction lipid in the fish is assumed to be 5% (f Lipid 0.05), then the critical body burden in the lipid fraction of the fish is C* Org C* 50 mol/g lipid (7) L f Lipid which is the estimate of the chemical concentration in the lipid of these fish that causes 50% mortality. The reason it is a constant concentration for all the narcotic chemicals represented by Equation 1 is a consequence of the unity coefficient multiplying log(k OW ) in Equations 1 and 4, which are the slopes of the straight line relationships. The model presented below is an extension of the body burden model. TARGET LIPID MODEL The body burden model relates the narcosis concentration to a whole body concentration using a BCF. If different species are tested, then individual BCF expressions would be required to convert the LC50 concentration to a body burden for each species. A more direct approach is to relate narcotic lethality to the concentration of the chemical in the target tissue of the organism rather than to the concentration in the whole organism. If the partitioning into the target tissue is independent of species, then the need for species-specific BCFs is obviated. The identity of the target tissue is still being debated [10,11], but we assume that the target is a lipid fraction of the organism. Hence, the name target lipid. The reason for this choice will become clear subsequently. The target lipid model is based on the assumption that mortality occurs when the chemical concentration in the target lipid reaches a threshold concentration. This threshold is assumed to be species specific rather than a universal constant that is applicable to all organisms (e.g., 50 mol/g lipid, Eqn. 7). The formulation follows the body burden model [12]. The target lipid-water partition coefficient K LW (L/kg lipid) is defined as the ratio of chemical concentration in target lipid C L ( mol/g lipid mmol/kg lipid) to the aqueous concentration C W (mmol/l) C L C W K (8) LW This equation can be used to compute the chemical concentration in the target lipid phase producing narcotic mortality, the critical body burden C*, L when the chemical concentration in the water phase is equal to the LC50 as C* L KLW LC50 (9) Assuming the narcosis hypothesis is true, i.e., that 50% mortality occurs if any narcotic chemical reaches the concentration C * L, the LC50 for any chemical can be calculated using the same critical target lipid concentration C* L and the chemicalspecific target lipid-water partition coefficient as C* L LC50 or (10) K LW log(lc50) log(c*) L log(k LW) (11) It should be noted that, as a practical matter, narcotic chemicals, i.e., chemicals that cause death by the narcosis mechanism and not by some other, more potent, mechanism, are restricted to those that are sufficiently soluble in water. This will become evident when the data are analyzed below. The problem is to determine the target lipid-water partition coefficient K LW for narcotic chemicals. It is commonly observed for classes of organic molecules that the logarithms of the partition coefficient between two liquids are related by a straight line [13]. For target lipid and octanol, the relationship would be log(k LW) a0 a1log(k OW ) (12) Such a relationship is called a linear free energy relationship (LFER)[14]. Combining Equations 11 and 12 yields the following linear relationship between log LC50 and log K OW : log(lc50) log(c*) L a0 a1log(k OW ) (13) where log(c*) L a0 is the y-intercept and a 1 is the slope of the line. This derivation produces the linear relationship between log(lc50) and log(k OW ), which is found experimentally (cf., Table 4 in Hermens et al. [15]), as log(lc50) m log(k OW ) b (14) where m and b are the slope and intercept of the regression. In addition, it identifies the meanings of the parameters of the regression line. The slope of the line m is the negative of the slope of the LFER between target lipid and octanol a 1 (Eqn. 12). The intercept of the regression b log(c*) L a0 is com- posed of two parameters. C* L is the target lipid concentration at narcosis mortality (Eqn. 11) and a 0 is the constant in Equation 12. The difference between the target lipid model and Mc- Carty s body burden model is that, for the latter model, the coefficients a 0 and a 1 for fish are assumed to be known, a and a (see Eqns. 4 and 12). It is interesting to examine the consequences of a similar assumption applied to the target lipid model. If it is assumed that the partitioning of narcotic chemicals in lipid and octanol are equal, i.e., lipid is octanol, a common first approximation, then a 1 1 and a 0 0 and the y-intercept becomes b log(c*) L (15) which is the target lipid concentration producing 50% narcosis mortality. This result can be understood by examining Figure 1. The y-intercept b is the LC50 for a chemical with a log(k OW ) 0 or K OW 1. The octanol/water partition coefficient is the ratio of the chemical s concentration in octanol to its concentration in water. Hence, for this hypothetical chemical (an example would be 2-chloroethanol, for which log(k OW ) ), the chemical s concentration in water is equal to its concentration in octanol. But if the K LW equals the K OW, i.e., lipid is octanol, then its concentration in water must be equal to its concentration in the target lipid of the organism. Therefore, the y-intercept is the target lipid phase concentration at which 50% mortality is observed, i.e., LC50 K 1 b C* Octanol C* OW L (16)

3 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Fig. 1. Schematic diagram of the log(lc50) versus log(k OW ) relationship. At log(k OW ) 0, K OW 1, and the concentration in water concentration in octanol. Note that this interpretation is true only if a 0 0 (see Eqn. 13). Thus, the target lipid narcosis model differentiates between the chemical and biological parameters of the log(lc50) log(k OW ) regression coefficients (Eqn. 13) in the following way: slope chemical m a 1 (17) intercept chemical biological b a log(c*) 0 L The chemical parameters a 0 and a 1 are associated with the linear free energy relationship between octanol and the target lipid (Eqn. 12). The biological parameter is the critical target lipid concentration C*. L This result is important because it suggests that the slope m a 1 of the log(lc50) log(k OW ) relationship should be the same regardless of the species tested since it is a chemical property of the target lipid the slope of the LFER. Of course this assumes that the target lipids of all species have the same LFER relative to octanol. This seems to be a reasonable expectation since the mechanism of narcosis is presumed to involve the phospholipids in the cell membrane and it appears to be a ubiquitous mode of action. However, the biological component of the intercept C* L (Eqns. 13 and 17) should vary with species sensitivity to narcosis since it is commonly found that different species have varying sensitivity to the effects of exposure to the same chemical. The expectations that follow from the target lipid model that the slope should be constant among species and that the intercepts should vary among species is the basis for the data analysis presented below. DATA COMPILATION An acute lethality (LC50) database for type I narcotics was compiled from available literature sources. The principal criterion was that a number of chemicals were tested using the same species so that the slope and intercept of the log(lc50) log(k OW ) relationship could be estimated. The data were restricted to acute exposures and a mortality endpoint to limit the sources of variability. A total of 33 species including amphibians, fishes, arthropods (insect and crustacean), mollusks, polychaetes, coelenterates, and protozoans are represented. Fig. 2. (A) Comparison of K OW predicted by SPARC versus measured K OW using slow-stir method. Line is 1:1 (B) Comparison of reported LC50 versus aqueous solubility estimated by SPARC. Line is 1:1. Seventy-four individual data sets, listed in Appendix 1, were selected for inclusion in the database, which comprises 796 individual data points. The individual chemicals are listed in Appendix 2. There are 156 different chemicals, including halogenated and nonhalogenated aliphatic and aromatic hydrocarbons, PAHs, alcohols, ethers, furans, and ketones. Note that the highest log(k OW ) in the dataset is log(k OW ) 5.32, which is a reflection of the fact that aqueous solubility decreases more quickly than log(k OW ) increases. We will return to this point below. The octanol-water partition coefficients and aqueous solubilities of these chemicals were determined using SPARC (SPARC Performs Automated Reasoning in Chemistry) [16], which utilizes the chemical s structure to estimate various properties. The reliability of SPARC was tested using K OW s measured using the slow-stir flask technique [17]. Fifty-three compounds such as phenols, anilines, chlorinated monobenzenes, PAHs, PCBs, and pesticides were employed. A comparison to the SPARC estimates, presented in Figure 2A, demonstrates that SPARC can be used to reliably estimate K OW over nearly a seven order of magnitude range. Note that this comparison tests both SPARC and the slow-stir measurements since SPARC is not parameterized using octanol-water partition coefficients [18].

4 1954 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. Aqueous solubility The toxicity data were screened by comparing the LC50 to the aqueous solubility S of the chemical (Fig. 2B). Individual LC50s were eliminated from the database if LC50 S, which indicated the presence of a separate chemical phase in the experiment. For these cases, mortality must have occurred for reasons other than narcosis, e.g., the effect of the pure liquid or solid on respiratory surfaces, since the target lipid concentration cannot increase above that achieved at the water solubility concentration. A total of 55 data points were eliminated, decreasing the number to 736 and the number of chemicals to 145 (Appendix 2). Exposure duration The duration of exposure varies in the data set from 24 to 96 h (Appendix 1). Before the data can be combined for analysis, the individual data sets need to be adjusted to account for this difference. The required equilibration time may vary with both organism and chemical. An increase in either organism body size or chemical hydrophobicity and measures of molecular size may increase the time to reach equilibrium. To determine if acute lethality for narcotic chemicals varied with exposure time, data were selected where toxicity was reported at multiple exposure times for the same organism and the same chemical. For seven species, all fish (see Fig. 3 caption), data were available for 96-h and either 24- or 48-h or both 24- and 48-h exposures. Arithmetic ratios of the LC50 for 48- and 96-h and for the 24- and 96-h exposures are compared to log K OW in Figure 3A. The 48- to 96-h ratio is one for essentially all the data. The 24- to 96-h ratio is larger, approaching two for the higher K OW chemicals. A linear regression is used to fit the relationship (Fig. 3B) as LC50 (24)/LC50(96) log(k OW) (18) where LC50 (24) and LC50 (96) are the LC50s for 24- and 96-h exposures, respectively. Since the majority of the data points, approximately 46%, represent narcosis mortality after exposure to a chemical for 96 h the 24-h fish toxicity data are converted to a 96-h LC50 using Equation 18 for the range in log(k OW ) over which the ratio is 1. No correction factor is applied to 24-h toxicity data for species other than fish, such as Artemina, D. magna, and Tetrahymena. These three species are small in size, and equilibrium should be reached within a 24-h period. DATA ANALYSIS The analysis of the toxicity data is based on the target lipid model assumption that the slope of the log(lc50) log(k OW ) relationship is the same for all species. A linear regression model is formulated below that can be used to estimate the species-specific body burdens and the universal narcosis slope. Regression model Consider a species k and a chemical j. The LC50 k,j for that species chemical pair is (Eqn. 13) log(lc50 k,j) log(c*(k)) L a0 a1log(k OW(j)) (19) where bk a1log(k OW (j)) (20) b log(c*(k)) a k L 0 (21) Fig. 3. (A) Ratio of 48-h LC50 to 96-h LC50. (B) Ratio of 24-h LC50 to 96-h LC50. Line is the regression used to correct the 24-h LC50 to 96-h LC50. is the y-intercept. The problem to be solved is how to include all the b k, k 1,...,N S corresponding to the N S 33 species and a single slope a 1 in one multiple linear regression model equation. The solution is to use a set of indicator variables ki that are either zero or one depending on the species associated with the observation being considered. The definition is 1 if k i ki (22) 0 if k i which is the Kronecker delta [19]. The regression equation can be formulated using ki as follows: N s log(lc50 ) a log(k (j)) b (23) i,j 1 OW k ki k 1 Equation 23 is now a linear equation with N S 1 independent variables, log(k OW (j)) and ki, k 1,...,N S. There are N S 1 coefficients to be fit, a 1 and b k, k 1,..., N S. For each LC50 ij corresponding to species i and chemical j, one of the b k corresponding to the appropriate species k i has a unity coefficient ii 1, while the others are zero. The way to visualize this situation is to realize that each row of data con-

5 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Fig. 4. Log(LC50) versus log(k OW ) for the indicated species. Line has constant slope 0.970; y-intercepts vary for each species. Outliers are denoted by (see text for criterion). sists of the LC50 and these N S 1 independent variables. For example, for j 1 and i 3, log(lc50 ij ) log(k OW (j)) 1i 2i 3i i NS (24) which is actually the first of the 736 records in the database. The result is that b 3 is entered into the regression equation as the intercept term associated with species i 3 because that ki is one for that record. By contrast, the slope term a 1 log(k OW (j)) is always included in the regression because there is always an entry in the log(k OW (j)) column (Eqn. 24). Hence, the multiple linear regression estimates the common slope a 1 and the species-specific intercepts b k, k 1,...,N S. A graphical comparison of the results of fitting Equation 23 to the full data set are shown in Figures 4 and 5 for each of the 33 species. The regression coefficients are tabulated and discussed subsequently after a further refinement is made to the model. The lines appear to be representative of the data as a whole. There appear to be no significant deviations from the common slope. A few outliers, which are plotted as, were not included in the regression analysis. An outlier is identified if the difference between predicted and observed LC50 is greater than one log unit when they are included in the regression. This decreases the total number of data points from 736 to 722. Testing the model assumptions The adequacy of the regression model is tested in three ways: (1) Are the data consistent with the assumption that the slope is the same for each species tested? (2) Does the volume fraction hypothesis [10] provide a better fit? (3) Are there systematic variations for particular chemical classes? Equal slopes. The first assumption, that the slope estimated for a particular species is statistically indistinguishable from

6 1956 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. Fig. 5. Log(LC50) versus log(k OW ) for the indicated species. Line has constant slope y-intercepts vary for each species. Outliers are denoted by (see text for criterion). the universal slope a , can be tested using conventional statistical tests for linear regression analysis [20]. The method is to fit the data for each species individually to determine a species-specific slope. Then that slope is tested against the universal slope a to determine what the probability is that this difference could have occurred by chance alone. The probability and the number of data points for each species are shown in Figure 6A. The slope deviations are shown in Figure 6B. Some are quite large. However, only three species equal or exceed the conventional significance level of 5% for rejecting the equal slope hypothesis. Testing at the 5% level of significance is misleading, however. Since there are 33 species being tested simultaneously, there is more than an even chance of rejecting one species falsely. The reason is that the expected number of rejections for a 5% level of significance would be , i.e., more than one species on average would be rejected due to statistical fluctuations even though all the slopes are actually equal. In fact, only 20 tests at 5% would, on average, yield one slope that would be incorrectly judged as different. The correct level of significance is (1/33)(1/20) 0.152% so that the expected number of rejections is or 5% [20]. This level of significance is displayed together with 5% in Figure 6A. As can be seen, there is no statistical evidence for rejecting the claim of equal slopes for the tested species taken as a whole and only two species if considered individually. Volume fraction hypothesis. The volume fraction hypothesis asserts that narcotic mortality occurs at a constant volume fraction of chemical at the target site of the organism [10]. Basically this involves expressing the LC50 as a volume fraction of chemical rather than a molar concentration. This is done using the molar volume of the chemicals (MV in Appendix 2). The LC50 on a molar volume basis is 3 3 LC50 (cm /L) LC50 (mmol/l) MV (cm /mmol) (25)

7 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Fig. 6. (A) Statistical comparison of slopes fitted to the individual species to the universal slope The probability that the difference occurred by chance (filled bar) and number of data points in the comparison (hatched bar) for each species in the database. Abbreviations are based on first letter of the genus and either the first or second letters of the species names given in Appendix 1, e.g., Aae Aedes aegypti and Am Ambystoma mexicanum. (B) The deviations of the individual estimates from the universal slope. The question is whether using molar volume as the concentration unit improves the regression analysis. The results are shown in Table 1. There appears to be no significant advantage to using the volumetric units. The fits do not improve, i.e., the R 2 s are essentially the same, and the slope is still not unity. Chemical classes. The analysis presented above assumes that all the 145 chemicals listed in Appendix 2 are narcotic chemicals and that their only distinguishing chemical property that affects toxicity is K OW. A criteria has been suggested that can be used to determine whether a chemical is a narcotic, namely that it demonstrates additive toxicity with a reference narcotic [1]. But it is not practical to test each possible chemical. The more practical test is whether the toxicity can be predicted solely from the log(lc50) log(k OW ) regression. In fact, this is used in methods that attempt to discriminate baseline narcotics from other classes [3]. Using this approach, differences in toxicity among chemical classes would be difficult to detect if differing species were aggregated or different slopes were allowed in the regression analysis. However, with the large data set employed above, these differences can be seen by analyzing the residuals grouped by chemical class. The criteria for choosing the relevant classes are not obvious without a detailed understanding of the mechanism of Table 1. Narcosis slope estimate using molar concentrations or molar volume units for LC50 Molar concentrations (mmol/l) Molar volumes (cm 3 /L) Slope R narcotic toxicity. Hence, the conventional organic chemical classes based on structural similarities, e.g., ethers, alcohols, ketones, etc., are used. The results are shown in Figure 7A. The means 2 standard error of the means are shown for each class. Although not a rigorous test, the 2 standard error range does not encompass zero for certain classes. Thus, it is likely that there are statistically significant chemical class effects. A rigorous test is conducted by including correction constants for each of the chemical classes in a manner that is analogous to Equation 23. The model equation is formulated using N C 1 corrections, c, corresponding to the 1,...,N C 1 chemical classes. These are interpreted as corrections relative to the baseline class that is chosen to be aliphatic non halogenated hydrocarbons. The regression equation is formulated as before with a variable j that is one if chemical j is in chemical class and zero otherwise, i.e., 1 if chemical j is in class j (26) 0 otherwise The regression equation that results is NS NC 1 i,j 1 OW k ki j log(lc50 ) a log(k (j)) b c k 1 1 (27) Each data record now contains the dependent variables log(lc50 i,j ), the independent variables log(k OW (j)) and the ki, k 1,...,N S, and j, 1,...,N C 1 indicator variables which are 0 or 1 depending on which species and which chemical class is represented by the LC50 i,j. Only N C 1 chemical class corrections are required because including N C class corrections underdetermines the equation set with one too many unknowns. The reason is that every

8 1958 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. RESULTS The results of the final regression analysis are listed in Table 2. Both the logarithmic b i and arithmetic 10 b i values of the intercepts are included together with their standard errors. Chemical classes that demonstrate higher potency than the reference class are ketones and PAHs. Halogenation increases the potency as well. After accounting for different potencies in the chemical classes, the mean residuals are statistically indistinguishable from zero (Fig. 7B). Standard errors and residuals The standard errors of the body burdens SE(b i ) found from the regression (Eqn. 27) are in an almost one-to-one correspondence with the number of data points for that species. Thus, the b i for Pimephales with 182 data points has a 10% coefficient of variation CV(b i ) SE(b i )/b i, while the b i for Neanthes with four data points has a 50% error (Table 1). Figure 8A presents CV(b i ) versus N. This determines the number of data points that are required to produce accurate estimates of critical body burdens. The residuals are lognormally distributed (Fig. 8B), which confirms the assumption underlying the use of regression analysis, and exhibit no trend with respect to K OW (Fig. 8C). The reason they are restricted to approximately 1 order of magnitude is that 14 data points outside that range were excluded as outliers. Fig. 7. Chemical class comparison. (A) Residuals from the regression grouped by class. Mean 2 standard errors. (B) Residuals grouped by class with chemical class corrections included in the regression. equation would have one b i and one c for species i and chemical j in chemical class. Since this condition would occur in every equation, there is no unique solution for the b k s and the c s. One of these constants could be adjusted by an arbitrary amount and the rest could then be adjusted to compensate while still achieving the same fit of the data. Thus, a reference chemical class, nonhalogenated aliphatic hydrocarbons for which c 0, is chosen. The remaining regression constants c, 1,..., N C 1, are then the differential toxicity of chemical class relative to the reference class. This is the reason for the c notation. Whether a chemical class correction is required is decided using a statistical test that compares the c s that result from the regression to the hypothesis c 0. For the classes that are not statistically different, they are included in the baseline class and the parameters are re-estimated. This is continued until all the remaining c s are statistically different from zero. After a number of trials, it was found that treating halogen substitutions as a separate additive correction gave the least number of statistically significant class corrections. Thus, chemical class corrections are applied to the base structure if necessary and an additional correction is made if any substituent is a halogen. Thus, for halogenated chemicals, it is possible that two j 1 in Equation 27. The chemical classes are listed in Appendix 2. Chemical class corrections The corrections due to chemical classes reduce the critical body burden by a factor of approximately one half for ketones (0.569) and PAHs (0.546). Halogenation reduces it further by (Table 2). Thus, a chlorinated PAH would exhibit a critical body burden of approximately one third of a baseline narcotic ( ). The coefficients of variation for these corrections are approximately 10%. The chemical class differences among the type I narcotics affect the LC50 K OW relationship. The model no longer predicts a single straight line for the log(lc50) log(k OW ) relationship for all narcotic chemicals. What is happening is that the y-intercepts are changing due to the changing c s. The model (Eqn. 27), when applied to a single species k, is N 1 C log(lc50 ) a log(k (j)) b c (28) k,j 1 OW k j 1 This is a straight line if only baseline narcotics are considered, c 0, or if only one chemical class correction is involved, e.g., all halogenated baseline narcotics. Otherwise, more than one c enter into Equation 28 and the line is jagged. Figure 9 presents three examples. The deviations from the baseline narcosis straight line are caused by the different chemical class potencies. Comparison to observed body burdens The target lipid model predicts the concentration in octanol (the y-intercept) that causes 50% mortality in 96 h. The question is how these compare to measured critical body burdens. The species-specific y-intercepts b i are related to the target lipid concentration by the relationship (Eqn. 21) y-intercept bi log(c*(i)) L a0 (29) or, with chemical class corrections, y-intercept bi c log(c*(i)) L a0 (30)

9 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Table 2. Regression results: y-intercepts and chemical class corrections a 10 bi SE( 10 bi ) b Species i N b i SE(b i ) ( mol/ g octanol) Mysidopsis bahia Portunus pelagicus Leptocheirus plumulosus Palaemonetes pugio Oncorhynchus mykiss Jordanella floridae Ictalurus punctatus Pimephales promelas Lepomis macrochirus Daphnia magna Cyprinodon variegatus Oryzias latipes Carassius auratus Rana catesbian Tanytarsus dissimilis Orconectes immunis Alburnus alburnus Nitocra spinipes Gambusia affinis Leucisus idus melanotus Neanthes arenaceodentata Artemia salina nauplii Lymnaea stagnalis Xenopus laevis Hydra oligactis Culex pipiens Poecilia reticulata Menidia beryllina Daphnia pulex Ambystoma mexicanum Daphnia cucullata Aedes aegypti Tetrahymena elliotti Chemical class N c SE( c ) 10 c SE( 10 c ) Aliphatics Ethers Alcohols Aromatics Halogenated chemicals Ketones Polycyclic aromatic hydrocarbons Slope a a See Equation 27. N number of data points; b i y-intercept; SE(b i ) standard error of b i ; c chemical class correction to the y-intercept; SE( c ) standard error of c. b Standard errors of 10bi and 10 c are based on the assumption that the estimation errors for b i and c are gaussian. The formulas follow from the standard error of a log normally distributed random variable [26]. For x b i or c, 2.303x, SE(x), and SE(10 x ) SE(e 2.303x ) e e e. for species i and chemical class, where a 0 is the parameter in the linear free energy relationship between octanol and the target lipid (Eqn. 12). The relationship between the predicted concentration in octanol b i c to the concentration measured in extracted lipid log(c*) L is examined in Table 3, which lists observed LC50 body burdens ( mol/g lipid) and predicted critical body burdens ( mol/g octanol) for organisms in the database for which measured lipid normalized critical body burdens were available. Three fish species, Gambusia, (mosquito fish), Poecilia (guppy), and Pimephales (fathead minnow), and two crustaceans Leptocheirus (amphipod) and Portunus (crab), are compared in Figure 10. With the exception of the fathead minnow data, the predicted and measured body burdens appear to be equal. There are substantial differences between the species, which the model reproduces. The fish are predicted and observed to have higher critical body burdens than the crustaceans. The apparent equality between the estimated and measured critical body burdens that come from two independent sets of data strongly suggest that, in fact, so that a 0 0 (31) log(c*(i)) b c y-intercept L i (32) This relationship implies that the target lipid is the lipid measured by the extraction technique used in the body burden data sets. This is an important practical result since it suggests that

10 1960 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. Fig. 8. (A) Coefficient of variation of the estimated species-specific body burdens versus number of data points for that species. (B) Log probability plot of the residuals. (C) Residuals versus K OW. Fig. 9. Log(LC50) versus log(k OW ) for the indicated species. Line connects the individual estimates of the LC50 including the chemical class correction. body burdens normalized to extracted lipid are expressed relative to the appropriate phase for narcotic toxicity. Since the intercepts appear to be the organism s lipid concentration, the y-intercepts, b i c, in the discussion presented below are referred to as body burden lipid concentrations although the units ( mol/g octanol) are retained since these are, in fact, the actual units of the intercepts. Species sensitivity The critical body burdens for baseline chemical b i range from 34 to 286 mol/g octanol (Table 2). The most sensitive species are crustacea and fish. Critical levels for the most sensitive crustacea range from 34 to 48 mol/g octanol. Fish range from 62 mol/g octanol for Oncorhynchus (rainbow trout) to 227 mol/g octanol for Poecilia (guppy). One consequence of the use of a universal narcosis slope is that the species sensitivity ranking derived from comparing either the LC50s or the critical body burdens of various species are the same. This occurs because the critical body burden is calculated from the LC50 and the universal slope (Eqns. 14, 15, and 31), log(c*) L log(lc50) log(k OW ) (33) If this were not the case, then the species sensitivity order could be reversed if LC50s or C* L were considered. Equation 33 is important because it can be used to compute the critical body burden of any type I narcotic chemical. Thus, it predicts what the critical body burden should be for a particular species at its LC50. This would be the concentration that would be compared to a directly measured critical body burden. It can be thought of as a normalization procedure that corrects type I narcotics for the varying K OW and places them on a common footing, namely, the critical body burden. Universal narcosis slope The universal narcosis slope, m , that results from the final analysis that includes chemical class corrections (Table 2) is smaller than that determined above without chemical class corrections ( ). It is close

11 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Table 3. Predicted and observed body burdens C* Org Organism Chemical log(k OW ) t (h) Observed Mean ( mol/ g lipid) Predicted ( mol/g octanol) Reference Gambusia 1,4-dibromobenzene [27] affinis 1,2,3-trichlorobenzene ,2,4-trichlorobenzene pentachlorobenzene Poecilia 1,4-difluorobenzene [28] reticulata 1,2-dichlorobenzene ,4-dichlorobenzene ,2-dibromobenzene ,4-dibromobenzene Pimephales 1,2-dichlorobenzene [28] promelas 1,4-dichlorobenzene ,2-dibromobenzene ,4-dibromobenzene ,2,4-trichlorobenzene [29] 1,1,2,2-tetrachloroethane dichlorobenzene dichlorobenzene a ,2-dichlorobenzene [30] 1,2-dichlorobenzene a ,4-dichlorobenzene ,4-dichlorobenzene a ,2 1,4 dichlorobenzene 107 1,2 1,4 dichlorobenzene a 110 1,2 1,4 dichlorobenzene a 138 1,2 1,4 dichlorobenzene a 150 naphthalene [31] 1,2,4-trichlorobenzene Leptocheirus fluoranthene [32] plumulosus fluoranthene Portunus 1,4-dichlorobenzene [33] pelagicus 1,2,3-trichlorobenzene ,2,3,4-tetrachlorobenzene pentachlorobenzene a Replicated exposure. to unity, a value commonly found [21], and is larger than the average of individual slopes reported by Van Leeuwen et al. [6] ( ) but comparable with a recent estimate for fathead minnows ( 0.94) [22]. The fact that the slope is not exactly one suggests that octanol is not quite lipid. However, it is also possible that, for the more hydrophobic chemicals in the database, the exposure time may not have been long enough for complete equilibration of water and lipid to have occurred. To test this hypothesis, the regression analysis is restricted to successively smaller upper limits of log(k OW ). The results are listed in Table 4. The variation is within the standard errors of estimation, indicating that there is no statistically significant difference if the higher log(k OW ) data are removed from the regression. This suggests that the universal narcosis slope is not 1 but is actually Table 4. Narcosis slope estimate for limited range in log (K OW ) Maximum log(k OW ) Slope Standard error CRITERIA DEVELOPMENT A principal motivation for developing the target lipid model for narcosis toxicity is that it can be used to derive water and sediment quality criteria. These criteria are based on the data set analyzed above and therefore are restricted to chemicals with log(k OW ) 5.3 (see Appendix 2). The application to these criteria to mixtures of more hydrophobic chemicals is discussed in the companion paper [23]. Species sensitivity The problem of how to reconcile varying species sensitivity with the need for a single number criteria that is protective has been addressed in the U.S. EPA water quality criteria guidelines [4]. The approach is to estimate the five percentile concentration from the ranked ordering of the genus mean acute values (GMAVs) for each genera in the acute toxicity database for that chemical. The concentration that results for this level of protection is referred to as the final acute value (FAV). For the target lipid model, the analogous species-dependent parameter is C*(i), L the critical lipid concentration for organism i. In this analysis, with the exception of Daphnia, for which there are three species in the database (Appendix 1), the species-specific critical target lipid body burdens are the GMAVs.

12 1962 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. Fig. 10. Predicted and measured body burdens for five species. For the three Daphnia species, a geometric mean is used. Applying the interpolation methods suggested in Stephan et al. [4] for deriving water quality criteria, the FAV is estimated to be C* L (5%, baseline) 35.3 mol/g octanol (34) for baseline chemicals (Fig. 11A). For other chemical classes, Equation 32 is used with the c from Table 2, and C* L (5%, class ) C* L (5%, baseline) c (35) Thus, for PAHs, the FAV is 35.3 mol/g octanol mol/g octanol. These are the tissue lipid concentrations corresponding to the FAV water quality criteria. Acute-to-chronic ratio The final step in developing a criteria is to estimate the acute-to-chronic ratio (ACR) [4]. Individual acute and chronic toxicity data pairs are available for 6 species and 20 chemicals (Appendix 3). The data span over four orders of magnitude in concentration, and the species represented are evenly distributed across the species sensitivity range of the acute data (Fig. 11B). Individual ACRs range from 1.2 to 23, with one exception. Also, there is no apparent trend between ACR and log K OW. This can be deduced from Figure 11B since the acute LC50 varies inversely with K OW. Thus, low LC50s correspond to high K OW chemicals. This observation is consistent with the observation that the 48- and 96-h LC50s are the same (Fig. 3A). These analyses indicate that the ACR is independent of chemical and species and can therefore be applied to all species and chemicals in this analysis. The final ACR is computed as the geometric mean of the paired values in Appendix 3. Water quality criteria The final chronic value is computed using the FAV and ACR as FAV FCV (36) ACR Fig. 11. (A) Probability distribution of the estimated critical body burdens C* L (Table 2). The fifth-percentile concentration is indicated. (B) Acute toxicity versus chronic toxicity concentrations for the indicated species (Appendix 3). Line represents an acute-to-chronic ratio ACR Using the results from the above analysis (Eqns. 33 and 35) yields log(fcv) log[c*(5%, L baseline) c /ACR] log(k ) OW log[6.94] c log(k OW) (37) where FCV is the final chronic value (mmol/l), C* L (5%,baseline) 35.3 mol/g octanol, c is given in Table 2, and ACR The K OW s used in this analysis are listed in Appendix 2. The resulting FCVs are listed in Table 5 as a function of K OW and in Table 6 for a selection of PAHs. If the toxicity is limited by solubility, no value is included, which indicates that the chemical acting alone would not cause chronic toxicity. The procedure for applying these results to mixtures and to sediments is discussed in the accompanying paper [23]. It should be pointed out that, for the PAHs with log(k OW ) 5, these criteria are an extrapolation beyond the limits of

13 Technical basis for narcotic chemicals and PAH criteria Environ. Toxicol. Chem. 19, Table 5. Final chronic values for narcotic chemicals Water column concentrations ( mol/l) log(k OW ) Baseline Halogenated baseline Ketones Halogenated Halogenated ketones PAHs a PAHs 0.0 6,940 3,960 3,950 2,250 3,790 2, ,340 1,360 1, , Tissue concentrations ( mol/g lipid) a PAHs polycyclic aromatic hydrocarbons. the data set used to validate the target lipid model. Their validity as water quality criteria can only be inferred from their validity as sediment criteria. In fact, when these criteria are applied to sediments, they are predictive, as discussed in the accompanying paper [23]. Although we have used most of the methodology prescribed by the U.S. EPA Technical Guidelines for developing water quality criteria [4], we have not adhered to them all. For example, we used 24-h data with adjustment, combined freshwater and saltwater data, included Artemia data and non-north American species, etc. We judged that, with the larger dataset, the estimate of the universal narcosis slope would be more robust and that the WQC would be more representative. Further, the exclusions can be made easily using Table 2 and Appendix 3 recompute the criteria if desired. Tissue criteria The tissue concentrations corresponding to the FCVs are also listed in Table 5. Since we have demonstrated that the critical body burdens computed from the target lipid model agree with measured concentrations (Fig. 10), these concentrations can be used for criteria as well. It has been shown that narcotic chemicals exhibit strictly additive toxicity (see [23] for a discussion). Therefore, the tissue criteria are applied to the molar sum of all the narcotic chemicals in extracted lipid. It should be pointed out that, unlike the acute tissue LC50s (Fig. 10), no experimental data have been presented that supports the validity of the chronic tissue concentrations. Thus, they are simply predictions at this point and await experimental confirmation. Table 6. Final chronic values for polycyclic aromatic hydrocarbons a Chemical name CAS MW (g/mol) log(k OW ) log(s) b (mol/l) log(fcv) (mol/l) Acenaphthylene Naphthalene Methylnaphthalene Methylnaphthalene Acenaphthene Fluorene ,6-Dimethylnaphthalene Anthracene Phenanthrene ,3,5-Trimethylnaphthalene c 7.01 Pyrene Fluoranthene Benzo[a]anthracene Chrysene Benzo[a]pyrene Perylene d Benzo[e]pyrene Benzo[b]fluoranthene Benzo[k]fluoranthene d Benzo[ghi]perylene d Dibenz[a,h]anthracene a CAS chemical abstract number; MW molecular weight; FCV final chronic value. b [34]. c Computed using SPARC [16]. d Limited by aqueous solubility.

14 1964 Environ. Toxicol. Chem. 19, 2000 D.M. Di Toro et al. There may be a problem with a direct application of these concentrations to observed body burdens in organisms that can metabolize specific narcotic chemicals such as PAHs. In analyses of the ratio of nonionic chemicals in organism lipid and sediment organic carbon, the PAHs are found to have significantly lower ratios than other, presumably non metabolized, chemicals [24,25]. Clearly, if metabolism is important, then measuring the body burden of only the parent compounds would underestimate the toxicity of these chemicals. This topic is discussed further in the accompanying paper [23]. CONCLUSIONS Water quality criteria for type I narcotic chemicals have been derived using critical body burdens. In particular, criteria for PAHs have been established. The body burdens deduced from the target lipid model are comparable to measured concentrations in extracted lipid, suggesting that the target lipid is indeed organism lipid. The criteria are based on the final acute and final chronic value methodology used by the U.S. EPA. These criteria will be used to derive sediment quality guidelines in the accompanying paper. Acknowledgement This work was performed for the U.S. Environmental Protection Agency, Office of Water, with additional support from a Cooperative Agreement with Manhattan College. We wish to thank our project officer Mary Reiley and Heidi Bell for their support, and our colleagues Walter Berry, Rob Burgess, Dave Mount, Bob Ozretich, Bob Spehar, and Rick Swartz for their help and advice. REFERENCES 1. Bradbury S, Carlson R, Henry T Polar narcosis in aquatic organisms. Aquat Toxicol Hazard Assess 12: Hermens J Quantitative structure activity relationships of environmental pollutants. In Hutzinger O, ed, Handbook of Environmental Chemistry, Vol 2E, Reactions and Processes. Springer-Verlag, Berlin, Germany, pp Verhaar H, van Leeuwen C, Hermens J Classifying environmental pollutants. 1. Structure activity relationships for prediction of aquatic toxicity. Chemosphere 25: Stephan C, Mount D, Hansen DJ, Gentile J, Chapman G, Brungs W Guidelines for deriving numerical national water quality criteria for the protection of aquatic organisms and their uses. PB National Technical Information Service, Springfield, VA, USA. 5. Di Toro DM, et al Technical basis for the equilibrium partitioning method for establishing sediment quality criteria. Environ Toxicol Chem 11: Van Leeuwen CJ, Zandt PVD, Aldenberg T, Verhaar H, Hermens J Application of QSARs, extrapolation and equilibrium partitioning in aquatic effects assessment. I. Narcotic industrial pollutants. Environ Toxicol Chem 11: Konemann H Quantitative structure toxicity relationships in fish toxicity studies. Part I: Relationship for 50 industrial pollutants. Toxicology 19: Veith G, Call DJ, Brooke L Structure toxicity relationships for the fathead minnow, Pimephales promelas: Narcotic industrial chemicals. Can J Fish Aquat Sci 40: McCarty L, Mackay D, Smith A, Ozburn G, Dixon D Interpreting aquatic toxicity QSARs: The significance of toxicant body residues at the pharmacologic endpoint. In Nrigu JO, ed, The Science of the Total Environment. Elsevier, Amsterdam, The Netherlands, pp Abernnethy S, Mackay D, McCarty L Volume fraction correlation for narcosis in aquatic organisms: The key role of partitioning. Environ Toxicol Chem 7: Franks N, Lieb W Mechanisms of general anesthesia. Environ Health Perspect 87: McCarty L Toxicant body residues: Implications for aquatic bioassays with some organic chemicals. In Mayes MA, Barron MG, eds, Aquatic Toxicology and Risk Assessment, Vol 14. American Society for Testing and Materials, Philadelphia, PA, pp Leo AJ Relationships between partitioning solvent systems. In Biological Correlations The Hansch Approach Advances in Chemistry, Series 114. American Chemical Society, Washington, DC, pp Brezonik P Chemical Kinetics and Process Dynamics in Aquatic Systems. CRC, Boca Raton, FL, USA. 15. Hermens J, Leeuwangh P, Musch A Quantitative structure activity relationships and mixture toxicity studies of chloro- and alkylanilines at an acute lethal toxicity level to the guppy (Poecilia reticulata). Ecotoxicol Environ Saf 8: Karickhoff SW, McDaniel VK, Melton C, Vellino AN, Nute DE, Carreira LA Predicting chemical reactivity by computer. Environ Toxicol Chem 10: de Bruijn J, Busser F, Seinen W, Hermens J Determination of octanol/water partition coefficients for hydrophobic organic chemicals with the slow-stirring method. Environ Toxicol Chem 8: Hilal S, Carreira L, Karickhoff S Estimation of chemical reactivity and parameters and physical properties of organic molecules using SPARC. In Politzer P, Murray J, eds, Quantitative Treatments of Solute/Solvent Interactions, Vol 1. Elsevier, Amsterdam, The Netherlands, pp Kreyszig E Advanced Engineering Mathematics. John Wiley & Sons, New York, NY, USA. 20. Wilkinson L SYSTAT: The System for Statistics, SYSTAT, Inc., Evanston, IL, USA. 21. Hansch C, Leo A Exploring QSAR. Fundamentals and Applications in Chemistry and Biology. American Chemical Society, Washington, DC. 22. Russom C, Bradbury S, Broderius S, Hammermeister D, Drummond R Predicting modes of toxic action from chemical structure: Acute toxicity in the fathead minnow (Pimephales promelas). Environ Toxicol Chem 16: Di Toro DM, McGrath JA Technical basis for narcotic chemicals and polycyclic aromatic hydrocarbon criteria. II. Mixtures and sediments. Environ Toxicol Chem 19: Parkerton T, Connolly J, Thomann R, Uchrin C Do aquatic effects or human health end points govern the development of sediment-quality criteria for nonionic organic chemicals? Environ Toxicol Chem 12: Tracey G, Hansen D Use of biota-sediment accumulation factors to assess similarity of nonionic organic chemical exposure to benthically-coupled organisms of differing trophic mode. Arch Environ Contam Toxicol 30: Aitchison J, Brown JAC The Lognormal Distribution. Cambridge University Press, Cambridge, UK. 27. Chaisuksant Y, Yu Q, Connell D Internal lethal concentrations of halobenzenes with fish (Gambusia affinis). Ecotoxicol Environ Saf 37: Sijm D, Schipper M, Opperhuizen A Toxicokinetics of halogenated benzenes in fish: Lethal body burden as a toxicological end point. Environ Sci Technol 12: van Wezel A, de Vries D, Kostense S, Sijm D, Opperhuizen A Intraspecies variation in lethal body burdens of narcotic compounds. Aquat Toxicol 33: van Wezel A, de Vries D, Sijm D, Opperhuizen A Use of the lethal body burden in the evaluation of mixture toxicity. Ecotoxol Environ Saf 35: de Maagd GJM Polycyclic aromatic hydrocarbons: Fate and effects in the aquatic environment. PhD thesis. University of Utrecht, Utrecht, The Netherlands, pp Kane Driscoll S, Schaffner L A comparison of equilibrium partitioning an critical body residue approaches for predicting toxicity of fluoranthene to amphipods. Abstracts, 18th Annual Meeting, Society of Environmental Toxicology and Chemistry, San Francisco, CA, USA, p Mortimer M, Connell D Critical internal and aqueous lethal concentrations of chlorobenzenes with the crab Portunus pelagicus (L). Ecotoxicol Environ Saf 28: Mackay D, Shiu W, Ma K Illustrated Handbook of Physical-Chemical Properties and Environmental Fate of Organic Chemicals, Vol II Polynuclear Aromatic HydroCarbons, Polychlorinated Dioxins and Dibenzofurans. Lewis, Chelsea, MI, USA. 35. Slooff W, Canton J, Hermens J Comparison of the susceptibility of 22 freshwater species to 15 chemical compounds. I. (Sub)acute toxicity tests. Aquat Toxicol 4: