Comparison of Species Sensitivity Distributions ... - ACS Publications

Comparison of Species Sensitivity Distributions Derived from Interspecies Correlation Models to Distributions used to Derive Water Quality Criteria...
0 downloads 0 Views 2MB Size
Download PDF
Environ. Sci. Technol. 2008, 42, 3076–3083

Comparison of Species Sensitivity Distributions Derived from Interspecies Correlation Models to Distributions used to Derive Water Quality Criteria S C O T T D . D Y E R , * ,† DONALD J. VERSTEEG,† SCOTT E. BELANGER,† JOEL G. CHANEY,† SANDY RAIMONDO,‡ AND MACE G. BARRON‡ The Procter and Gamble Company, 11810 East Miami River Road, Cincinnati, Ohio 45253-8707, and Gulf Ecology Division, National Health and Environmental Effects Laboratory, United States Environmental Protection Agency, 1 Sabine Island Dr., Gulf Breeze, Florida 32561

Received September 13, 2007. Revised manuscript received December 19, 2007. Accepted January 14, 2008.

Species sensitivity distributions (SSD) require a large number of measured toxicity values to define a hazard level protective of multiple species. This investigation comprehensively evaluated the accuracy of SSDs generated from toxicity values predicted from interspecies correlation estimation (ICE) models. ICE models are log–log correlations of multiple chemical toxicity values for a pair of species that allow the toxicity of multiple species to be predicted from a single measured acute toxicity value for a surrogate species. ICE SSDs were generated using four surrogate species (fathead minnow, Pimephales promelas; rainbow trout, Oncorhynchus mykiss; sheepshead minnow, Cyprinodon varigatus; and water flea, Daphnia magna). ICEbased hazard concentrations (HC5s) from the 5th percentile of the log–logistic distribution of toxicity values were compared to HC5s determined from the acute toxicity of 55 chemicals from the United States Environmental Protection Agency Ambient Water Quality Criteria (AWQC). Measured fish and invertebrate acute toxicity data and HC5s from the AWQC data sets were compared to ICE-based HC5s. Surrogate species choice was found to be an important consideration in developing predictive HC5s. These results illustrated that fish predict fish better than invertebrates and D. magna predicted invertebrates better than most fish. For example, a mixed model of predicted fish and invertebrates from fathead minnow and D. magna as surrogate species provided predictive relationships with an average factor of 3.0 ((6.7) over 7 orders of toxic magnitude and several chemical classes (HC5predicted/HC5measured). The application of ICE models is recommended as a valid approach for generating SSDs and hazard concentrations for chemicals with limited toxicity data.

* Corresponding author phone: 513-627-1163; fax: 513-627-1208; e-mail: [email protected]. † The Procter and Gamble Company. ‡ United States Environmental Protection Agency. 3076

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 8, 2008

Introduction Ecological risk assessments typically require characterizing the effects of multiple chemicals on a diversity of ecological receptors using toxicity data for only a limited number of species. Regulatory activities such as REACH (Registration, Evaluation and Authorization of Chemicals ( (1, 2); http:// europa.eu.int/comm/environment/chemicals/reach.htm), ICCA (International Council of Chemical Associations) High Production Volume (HPV) Chemicals Challenge (http:// www.icca-chem.org/), and Canada’s Domestic Substance List (3) will also create new demands for toxicity data. Note that the ICCA HPV Chemicals challenge is in cooperation with OECD and its member countries which then also include other HPV challenge programs such as those of the United States Environmental Protection Agency (USEPA) and Japan. Traditionally, aquatic toxicity data have been either empirically derived from laboratory tests or estimated from (quantitative) structure–activity relationships ((Q)SAR). Interspecies correlation estimation (ICE) statistical models have been developed as an additional approach for estimating the toxicity of chemicals to both aquatic organisms and terrestrial wildlife (4, 5). Each ICE model is a log–log correlation of multiple chemical toxicity values for a pair of species that allows the toxicity of multiple species to be predicted from a single measured acute toxicity value from a surrogate species (4, 5). If ICE can be shown to have strong predictive power, a reduction in empirical toxicity testing demands may ensue as is envisioned under the seventh Amendment to the Cosmetics Directive (6) and REACH (2). Such a tool would simultaneously reduce the numbers of fish used for experimental purposes as well as provide meaningful and equivalent hazard data for a broad range of species needed in ecological risk assessment. A current and particularly useful attribute of interspecies correlations is their use to minimize the testing of threatened and endangered species, based on test results from toxicity tests with a surrogate species (7). Surrogate species is defined here as the taxon which is used to develop predicted toxicity of new chemicals to additional species. Recently, Dyer et al. (8) introduced the use of ICE models to generate species sensitivity distributions (SSDs) and a hazard concentration (HC5) estimated from the fifth percentile of the distribution of toxicity values (9). SSDs have been used to develop water quality criteria and other protective environmental concentrations, but they currently require large data sets (e.g., USEPA ambient water quality criteria utilize at least eight acute toxicity values from several taxa spanning three trophic level, fish, invertebrates, and plants) of measured toxicity values (10). An initial assessment of five chemicals showed that ICE-based HC5s determined from a single surrogate toxicity value were within a factor of 10 of HC5s derived from SSDs of measured acute toxicity values (8). Dyer et al. (8) recommended additional assessment and validation of ICE-based HC5s, including an investigation of the effect of ICE surrogate species on the accuracy of ICEbased HC5s. The objective of the current work was to comprehensively evaluate the accuracy of ICE-generated SSDs and to validate a procedure for generating ICE-based HC5s. SSD accuracy was assessed by comparing ICE-based HC5s to HC5s derived from measured acute aquatic toxicity values for 55 chemicals of diverse structure and ecotoxicological modes of action. The effect of ICE surrogate species selection on HC5 accuracy was investigated by generating ICE SSDs using either one 10.1021/es702302e CCC: $40.75

 2008 American Chemical Society

Published on Web 02/28/2008

TABLE 1. List of Chemicals and Range of Acute Toxicity Values from Ambient Water Quality Criteria (AWQC) Documentsa class metals metals metals metals metals metals metals metals metals metals metals metals metals metals organometal organometal nonorganics nonorganics nonorganics nonorganics solvent-like nonaromatic solvent-like nonaromatic solvent-like nonaromatic solvent-like nonaromatic nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar aromatics nonpolar Aromatics nonpolar aromatics polar aromatic polar aromatic polar aromatic polar aromatic polar aromatic polar aromatic polar aromatic organophosphates organophosphates organophosphates organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines organochlorines

chemical Name aluminum beryllium cadmium chromium(III) chromium(VI) copper inorganic arsenic(III) inorganic arsenic(V) lead mercury II nickel selenate selenite zinc monosodium methanearsonate tributyltin ammonia chlorine cyanide Sodium chloride 1,1,2,2-tetrachloroethane 1,1-dichloro-ethylene hexachloroethane methylene chloride butylbenzyl phthalate di-n-butyl phthlate 1,2 or 3 or 4-dichlorobenzene 1,2,4-trichlorobenzene atrazine benzene chlorobenzene ethylbenzene naphthalene toluene 2,4,5 or 6-trichlorophenol 2,4-dinitrophenol 2-chlorophenol 4-nitrophenol nonylphenol pentachlorophenol phenol chlorpyrifos diazinon parathion aldrin chlordane DDT dieldrin endosulfan endrin Heptachlor hexachlorobutadiene hexachlorocyclohexane hexachlorocyclopentadiene toxaphene

total no. no. fish min. acute mean acute max. acute ref species species value (µg/L) value (µg/L) SD (µg/L) value (µg/L)

HC5 (µg/L)

11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

15 5 122 23 53 125 25 7 24 62 44 24 52 74 8

7 4 37 12 21 60 11 3 11 16 17 12 29 31 6

1900 3250 1.61 2221 23 2.54 232 850 143 2.22 152 593 255 51 1921

30 572 4742 5737 17 495 75 170 1502 10 421 14 369 22 753 296 37 581 129 855 14 960 11 974 295 506

21 582 1851 16 042 16 405 254 841 9274 10 145 17 368 51 967 508 68 103 326 805 29 009 38 825 492 486

79 900 7900 13 5000 71 060 187 0000 101 999 41 760 49 000 235 900 2000 320 000 1 515 661 203 000 320 400 140 3000

4120 2590 12.9 3770 279 4.61 514 766 90.3 4.58 887 612 564 145 324

26 27 38 29 30 31 32 33 34 35 36 37

46 48 56 26 13 5 6 7 5 6 6 8

13 29 32 13 5 3 4 4 3 4 4 5

0.61 11230 26 4.89 1470 9020 30 300 940 193 000 2944 940 1580

551 86 407 197 742 5277 14 448 155 867 2371 245 600 12 197 2953 6818

3625 213 843 240 1992 2898 5949 96 906 2568 52 681 13 335 2048 4327

24 600 0.503 1 466 000 9590 1418 29.5 10 000 16.5 11 940 1800 21 300 6680 250 000 34 300 8070 507 331 000 174 000 36 538 1230 6470 29.8 12 000 1340

38 39 40 41 42 43 (44) 45 46 47 48 49 50 (51) 52 53 54 55 56 57 58 59 60 (61) 62 63 64 65

6 27 14 7 10 7 10 5 6 5 5 28 59 21 33 31 38 36 22 59 39 22 48 36 9 40 10 50

4 11 8 5 7 4 6 3 4 4 3 15 23 10 17 12 14 23 13 35 21 10 27 20 6 26 6 27

450 2324 5300 10 500 430 2300 3700 450 620 4380 7170 17 4.36 5800 0.04 0.20 0.04 1.5 0.4 0.14 0.70 00.04 00.04 00.06 32 0.17 7 0.53

13 297 17 035 195 314 37 393 17 7897 53 231 140 050 3528 10 243 11 506 23 322 190 4681 52 624 169 2622 744 691 40 45 76 58 12 47 233 175 83 9259

19 694 15 019 266 389 26 275 308 890 84 072 322 186 3324 10 836 5879 22 243 163 17 826 55 086 410 3864 1303 3242 45 167 175 163 52 70 179 586 108 65 046

50 200 60 000 924 000 86 000 1 030 000 199 000 1 050 000 9040 29 400 20 170 60 500 774 116 000 248 000 1991 11 640 5230 19 000 190 1230 740 730 352 320 557 3680 371 460 000

245 2730 3860 8930 2680 442 2260 464 805 4410 3820 39.7 8.46 7510 0.0168 0.283 0.0847 1.42 2.41 0.265 0.865 0.0234 0.05 0.497 33.3 2.61 7.02 0.402

a HC5 corresponds to the hazard concentration at the 5th percentile based on a log–logistic regression of fish and invertebrate species within the AWQC data set.

fish species, one invertebrate species, or combinations of multiple surrogate species.

Materials and Methods AWQC (Ambient Water Quality Criteria) Data Set. Fish and invertebrate acute toxicity data were collected from data sets that supported USEPA’s Ambient Water Quality Criteria (AWQC) for 55 metals, organics, and organometals (11–65). These compounds represent a wide range of chemical classes and ecotoxic modes of toxic action. Compounds were not prescreened and included all compounds in the AWQC database that contained acute toxicity values (primarily 24–48

h for invertebrates and 96 h for fish) for a minimum of five species, leading to a total of 55 chemicals (Table 1). Overall, the total number of fish and invertebrate acute toxicity data records for all compounds was 1578. Species mean acute values as reported in the AWQC documents were used when multiple toxicity values were available per species and chemical. The range of the total number of species per chemical was 5-125, with the greatest numbers occurring for metals such as copper (125 species), cadmium (122 species), and zinc (74 species). There were slightly more fish species in the data set with an average of 54% of all toxicity values (range of 26–80.0% for individual chemicals). Toxicity VOL. 42, NO. 8, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3077

FIGURE 1. A and B. Robust regressions (>10 chemical pairs per species pair and highly significant, p e 0.05) from ICE for fathead minnow and Daphnia magna. Bold regression lines correspond to counter intuitive relationships where predicted species were relatively tolerant for chemicals with high toxicity (e.g., 1 µg/L), yet amongst the most sensitive species for the least toxic chemicals. values were wide ranging, from a minimum value (maximum toxicity) of 0.04 µg/L (chlorpyrifos, parathion, endrin, and endosulfan) to a maximum value (minimum toxicity) of 1 870 000 µg/L for Neophasganophora capitata exposed to hexavalent chromium. ICE Data Set. Fish and invertebrate toxicity data that were used in the development of ICE models were compiled by Mayer (66), Mayer and Ellerseick (67), AQUIRE (68), and USEPA’s Office of Pesticide Programs aquatic database. Standardized acute toxicity test conditions and quality criteria were described by Buckler et al. (69). Test results in the database had the following characteristics: 96 h LC50 for fish, weighing 0.2–0.5 g, less than 3 months old or less than 50 mm in length; 48 h EC/LC50 data for most invertebrates (e.g., Daphnia sp.); test chemicals were identified as technical grade or >90% active ingredient; water hardness between 30 and 60 mg CaCO3/L; pH 6.8–7.8; and temperatures appropriate per species (e.g., rainbow trout 10–15 °C; Daphnia magna ∼20 °C). The entire data set consisted of 4772 test results comprising 247 aquatic species and 660 chemicals, including amphibians, bryozoans, fish invertebrates, and algae. Due to the limited availability of significant correlations between algae and animal taxa, we focused our work on invertebrate and fish species. Independency of ICE and AWQC data sets were ensured by not admitting identical values in the AQWC data set already found in the ICE data set. ICE Regressions. Interspecies correlations were determined via Model II least-squares regressions where both 3078

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 8, 2008

independent and dependent variables are random (70). Slopes (β0) and intercepts (β1) were derived from the eq 1 log X2 ) β0 + β1 × (log X1)

(1)

where X1 is the acute toxicity of the surrogate species (e.g., Pimephales promelas, Oncorhynchus mykiss, Cyprinodon variegatus, Americamysis bahia, and Daphnia magna) and X2 is the predicted species toxicity value. These surrogate species were selected as they are among the most commonly tested species in the ICE database (4). A “robust” subset of ICE models was created since 67% of the ICE models were not statistically significant. This robust model subset was developed by including only significant correlations (p < 0.05) with at least 10 observations per species pair (8). Figure 1(A and B) illustrates robust regressions for selected surrogate species: fathead minnow and D. magna. Some robust regressions illustrated a shift in relative species sensitivity across the range of toxicity values. For instance, regressions for fathead minnow vs Pteronarcys californica, Pteronarcella badia, and Cypridopsis vidua indicated that as toxicity decreased, the relative sensitivity of the predicted species increased (Figure 1). Similar trends were also observed for other surrogate vs predicted species relationships. A plot of correlations coefficients compared to slope (Figure 2) clearly illustrated that slopes approaching unity provided better-fit ICE models. Further investigation showed that fish surrogates provided better models for fish than invertebrates, and vice versa. That is, Daphnia magna

FIGURE 2. Scatter-plot of correlation coefficients vs slopes from robust regressions for five surrogates species: fathead minnow, rainbow trout, sheepshead minnow, Daphnia magna, and Americamysis bahia. relationships with other invertebrates had better correlation coefficients and steeper slopes than when D. magna was used to predict fish toxicity values. This was also shown through cross-validation of ICE models developed for wildlife species, in which model robustness was greatest for surrogate and predicted species that were most closely related (5). Since the goal of this exercise was to maximize the number of predicted species using ICE, yet use the most reliable regressions, we included an additional criterion to our robust data set: a slope of greater or equal to 0.65. This criterion resulted in predictions for 19, 23, 7, 3, and 7 species from surrogate species Pimephales promelas, Oncorhynchus mykiss, Cyprinodon variegatus, Americamysis bahia, and Daphnia magna, respectively. Since Americamysis bahia only had three models which meet this criterion, it was eliminated from further evaluation. All seven predicted species Daphnia ICE were arthropods. Development of ICE- and Measured-Based SSDs and HC5s. No transformation always provides the best fit, however, for consistency across substances, species toxicity values were plotted as cumulative percent versus log LC50 values with the best-fitting log-logistic distribution (71). Species sensitivity distributions (SSD) were described by the following two-parameter log-logistic model fit via maximum likelihood estimation (eq 2): Y ) 1 ⁄ (1 + exp((A - X) ⁄ B))

(2)

Where: X ) independent variable (i.e., log L(E)C50), Y ) cumulative probability (i.e., percentage of affected species), A ) parameter representing location (i.e., the intercept parameter), B ) parameter representing the slope of the curve (i.e., the scale parameter). Residuals from the models were generally normally distributed. For each of the 55 chemicals, four types of SSDs were created using different values of acute toxicity: (1) measured values for all fish and invertebrate species from each chemical’s AWQC data set. (2) ICE- generated values using fish surrogate species (fathead minnow, rainbow trout, or sheepshead minnow) toxicity value to predict all species toxicity values. (3) ICE-generated values using fish surrogate species toxicity value to predict toxicity values for fish. (4) ICE-generated values using the D. magna toxicity value from each AWQC chemical to predict invertebrate toxicity values. The predictions generated in approaches 3 and 4 were then combined to create a fifth SSD: ICE-based fish + invertebrate SSD (Figure 3). The acute hazardous concentration protective of 95% of species (HC5) for all SSDs was determined by the methods of Aldenberg and Jaworska (72). The HC5 was selected as this parameter is being widely used in environmental risk assessment (9) thus providing a metric to assess whether these techniques might lead to over- or

FIGURE 3. Process by which HC5 values were obtained from measured values from AWQC data and from ICE. The figure illustrates a species sensitivity distribution (solid line SSD) for a toxicant with the HC5 designated as an “X”. Measured rainbow trout data was input into robust regressions to derive a SSD for fish species only ( · · · · · · · · ), with the HC5 designated as a square 9. Rainbow trout-based ICE SDD for all species and HC5 for the same toxicant are illustrated as the dashed line (- -) diamond (. An SSD constructed from rainbow trout-ICE predictions for fish only and from Daphnia magna-ICE predictions for invertebrates is provided by the small dashed line (-----) and the HC5 indicated by the triangle 2. Predicted HC5s from ICE were plotted against measured HC5 for each AWQC chemical. underprotection of fish and invertebrate communities. The relationships of measured AWQC HC5s to ICE-based HC5s were determined via least-squares linear regression.

Results AWQC HC5s. HC5 values based on measured acute toxicity values for both fish and invertebrates ranged over 7 orders of magnitude, from a low of 0.017 µg/L (chlorpyrifos) to a high of 174 000 µg/L (methylene chloride) (Table 1). The most toxic classes of chemicals were the organophosphates (e.g., 0.017 µg chlorpyrifos/L, 0.28 µg diazinon/L) and organochlorines (ranging from 0.023 µg/L endosulfan to 33.3 µg/L hexachlorobutadiene). Ranges in the HC5 values of all other classes (metals, nonorganics, nonpolar organics, organometals, polar aromatics, and solvent-like nonaromatics) overlapped. AWQC HC5 vs ICE HC5. Figure 4 illustrates four scatterplots and associated regressions of measured (AWQC) HC5 values compared to ICE-predicted HC5 values based on the surrogate species fathead minnow, rainbow trout, sheephead minnow, and D. magna. The number of compounds in these regressions differ due to the availability of a measured surrogate value. For example, a D. magna acute toxicity value is only available for 49 compounds, thus the D. magna-based regressions have 49 compounds. All compounds had a measured fathead minnow toxicity value, thus regressions based on fathead minnow had 55 compounds. When fathead minnow and D. magna predictions were combined, HC5 values on only 49 compounds could be generated. The regression of AWQC HC5 values compared to ICEbased HC5 values using the fathead minnow as the surrogate provided a highly significant relationship with a slope of 0.56 and R2 of 0.70 (p < 0.001) (Figure 4). Points below the oneto-one line (dotted gray line) indicated that ICE predictions were more protective than measured HC5s, i.e., ICE-predicted HC5s were less than the measured HC5s. For materials with HC5s >1.0 µg/L, fathead minnow-based HC5s were generally protective compared to AWQC data. HC5s based on fishonly predictions yielded a greater slope and better over all fit (slope 0.76, R2 ) 0.71), yet a clearly under-protective relationship compared to measured HC5s. However, when VOL. 42, NO. 8, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3079

FIGURE 4. Relationships of predicted HC5s based on robust ICE regressions for four surrogate species (fathead minnow, rainbow trout, sheepshead minnow, Daphnia magna) to measured HC5s using the process illustrated in Figure 3. For fish surrogates, diamond symbols and blue regression line show ICE-based HC5 predictions for all species. Measured HC5 compared to ICE-based fish-only HC5s are illustrated as squares and the red line. Measured HC5s compared to HC5s derived via addition of fish-only predictions plus D. magna predictions for invertebrates are symbolized as triangles and the green line. Daphnia magna predictions for all species are provided in red and symbolized in green for invertebrates-only. Perfect unity predictions are represented as a gray line in each graphic. fathead minnow-predicted fish values are combined with D. magna-predicted invertebrate values to create a SSD and HC5, the result was a regression (slope 0.88, R2 ) 0.89) that more closely resembled the one-to-one line. On average, predicted HC5 values were within a factor of 3 ((6.7 SD) of measured HC5s. Stated another way, 90% of the predicted HC5 values were within a factor of 10 and 45% within a factor 2 of the measured HC5s across 7 orders of magnitude. Similar relationships were found with rainbow trout-based HC5s and measured HC5s. Rainbow trout HC5s for all fish and invertebrate species provided a slope and coefficient of determination (R2) of 0.52 and 0.66, respectively. As with the fathead minnow, rainbow trout predicting fish-only HC5s compared to the measured HC5 yielded a steeper slope (0.76) and greater R2 (0.71), but an under-protective relationship. However, when rainbow trout to fish-only toxicity predictions were combined with those with D. Magna-predicting invertebrates, the result was a steeper slope (0.80) and greater coefficient of determination (0.88). On average, predicted HC5 values were within a factor of 1.8 ((3.0 SD) of measured HC5s. As with the fathead minnow, the vast majority of ICEbased predictions (94%) were within an order of magnitude of the AWQC-based HC5s. Sheepshead minnow-based predicted HC5s compared to measured HC5s had a slope of 0.78 and an R2 of 0.73, with predictions below approximately 1000 µg/L being under protective. A decreased slope (0.64) and similar coefficient of determination (0.73) was observed for fish-only predic3080

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 8, 2008

tions. The addition of D. magna-based invertebrate predictions to the fish-only predictions did not lead to a slope approaching the one-to-one line, although the R2 improved to 0.81. Fifty-three percent of the entire HC5 range was protective using ICE-based predictions with sheepshead minnow and D. magna. D. magna predicted HC5s for all species provided a slope of 0.64 and an R2 of 0.70. Similarly, D. magna predictions for invertebrates-only had a slope and R2 of 0.55 and 0.70, respectively. Forty three percent of the predicted HC5s via D. magna were underprotective compared to the AWQCbased HC5s.

Discussion In this study, we comprehensively evaluated the use of robust ICE models to predict HC5 values by comparing them to HC5 values obtained from measured and noncensored acute toxicity data for 55 chemicals documented from USEPA Ambient Water Quality Criteria reports. That is, only predicted species from ICE using highly significant and interpretable relationships were used to estimate HC5s and compared to all available acute aquatic toxicity data from the AWQC data sets. The approach clearly illustrated that a combination of fish and invertebrate surrogate species could typically predict HC5 values within an order of magnitude of measured-based toxicity HC5s. For example, the use of the fathead minnow to predict fish toxicity values and D. magna to predict

invertebrate toxicity values, when combined, provided predicted HC5 values within a mean factor of 3 of the HC5 based on measured values from the AWQC data sets. Remarkably, these relationships were established with chemicals with diverse mechanisms of action and HC5 values which ranged over 7 orders of magnitude. Even so, the good agreement is the logical extension that surrogate and predicted species are based on species pairs exposed to the same chemicals. Hence, it is likely that the same mechanisms of action for both the surrogate and predicted species are at play in these robust relationships. Raimondo et al. (5) indicated that goodness of fit was best with least taxonomic distance. The approach was based on an initial assessment by Dyer et al. (8), that illustrated that robust ICE-modeled HC5s for chemicals with disparate modes of actions, including a neurotoxicant (fenvalerate), two narcotics (LAS, nonylphenol), a herbicide (atrazine), and a heavy metal (copper), predicted HC5 values within a factor of 10 of the measured HC5 value. Further, these predictions were protective compared to experimental ecosystem results (8). When all predicted species in ICE were used to determine SSDs and HC5s, the results were typically a decreased scale (i.e., slope) and decreased HC5, compared to SSDs and HC5s using the robust regressions. The decreased slope and decreased HC5 were a reflection of poorer predictability from the use of nonsignificant regressions and/or regressions with few species pairs. Our expanded, diverse and comprehensive data set of chemicals from the AWQC continue to illustrate the consistency of the relationship between ICE-based HC5 predictions and measured values. This study showed that surrogate species selection in ICE was important for predicting accurate HC5s. Selected interspecies correlations containing highly significant correlations (p < 0.05), a minimum of 10 pairwise observations, and have slopes greater 0.65 led to a majority of predictions within a factor of 10 of AWQC HC5s. For instance, 94% of predicted HC5s using the fathead minnow as the surrogate for fish and Daphnia magna as the surrogate species for invertebrates provided predictions within a factor of 10 of AWQC HC5s with nearly half of all predictions within a factor of 2. Similar results were observed with rainbow trout and Daphnia magna. Using a mechanistic methodology, Jager et al. (73) recently conducted an analysis benchmarking the vulnerabilities of species to toxicants with diverse modes of action (nonpolar narcotics, photosynthesis inhibitors, organophosphates, and carbamates). The ratio of toxicity results for various species to that of the fathead minnow and the algal taxa Scenedesmus demonstrated that estimated species toxicity values that incorporate both toxicant potency and vulnerabilities were generally within an order of magnitude. SSDs from a distribution of the estimated species toxicity values were also determined, though no estimations of an HC5 were provided (73). Even so, evidence is growing that indicates that appropriate species selection can provide highly predictive toxicity and protective criteria. Our study is the most exhaustive study to date investigating the use of interspecies correlations across diverse chemical class and mechanisms of action, spanning nearly 7 orders of magnitude in L(E)C50 values. A key need for environmental protection is to put this information into practice in a risk assessment context. At present, it is common practice to assume that protection of a single species will protect the ecosystem (74). To protect a wide variety of species based on a few acute toxicity tests (often one alga, a daphnid ,and one fish species each), assessment factors (AF) are applied to the L(E)C50 values to account for interlaboratory and interspecies variability and extrapolations from short (acute) to long (chronic) term exposure and laboratory to field differences. The EU (74) currently uses a default AF of 1000

to extrapolate from three acute toxicity values to the predicted no effect concentration, PNEC, in the ecosystem. The AF of 1000 accounts for the extrapolation of acute to chronic toxicity, interspecies variability, interlaboratory variability, and the extrapolation of laboratory to field. Since SSDs reflect variability in chemical sensitivity between species (and strain) as well as variability in laboratory testing, there is the potential to use these distributions and reduce the magnitude of AFs used on ecological risk assessments. Thus, we compared the measured HC5 values taken from the AWQC with measured acute toxicity values divided by 10 to account for interlaboratory and -species variability. When the measured fathead minnow, rainbow trout, or D. magna acute toxicity values were divided by 10, the resulting value was more than 10fold below the measured HC5 value in 2.5, 10.8, and 1.8% of the cases, respectively. However, the acute toxicity value divided by 10 was more than 10 times higher than the measured HC5 value for 12.7, 12.5, and 10.2% of the chemicals. For many risk assessors it may be acceptable for the lowest acute toxicity value divided by 10 to be overprotective relative to the measured HC5 values; however, it is likely unacceptable for the lowest acute divided by 10 to be underprotective. While increasing the assessment factor is one approach, we recommend use of the HC5 values based on measured fish and invertebrate toxicity values and ICE predictions. This data-based approach to incorporate interspecies variability is preferred over the use of generic, deterministic assessment factors. While the use of ICE to provide accurate estimates of toxicity and protective criterion is encouraging, there are some aspects that need further research before the approach should be accepted as a potential replacement for the current factors of 10-based assessment factors. First, aquatic plant species are absent in this analysis. We are presently developing ICE models for algal species and plants to test its inclusion into fish + Daphnia, based SSDs and HC5s. Second, USEPA is developing additional robust ICE models for aquatic and wildlife species that are available from the Internet (WebICE: http://www.epa.gov/ceampubl/fchain/webice/index. htm). Lastly, we intend to explore methods using ICE where effects assessments can be based less on North American species. For example, Maltby et al. (75) have suggested that geographic distribution of species is not very important when establishing chemical hazard. This will be investigated in two ways including non-North American species into ICE and trait-based relationships including morphology, life history, physiology, and feeding ecology. Species traits may have potential for explaining species vulnerability (76), and species sensitivity relationships may be improved by incorporating them. If so, ICE can be re-evaluated based on these traits and provide SSDs and HC5s that involve species with diverse distributions. Finally, it is important to consider the use of animals in toxicity testing. Some species cannot be tested due to their status as threatened and endangered species. While Sappington et al. (77) and Hansen et al. (78) have indicated that some of these species are not especially sensitive to toxicants, ICE provides a mechanism to obtain toxicity estimates to augment their protection. In conclusion, we strongly encourage the use of ICE for the generation of SSDs and HC5s for effects evaluations and environmental risk assessments for chemicals with limited data. With further work, it is our hope that use of ICE will supplant the need for extensive laboratory testing, thereby reducing the use of animals.

Acknowledgments We thank Marion Marchetto, Deborah Vivian, Anthony DiGirolamo, Shaina Duffy, Bret Blackwell, Brandon Jarvis, Christel Chancy, Nathan Lemoine, Nicole Allard, and Laura Dobbins for AWQC data entry and quality assurance. VOL. 42, NO. 8, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3081

Literature Cited (1) Bodar, C.; de Bruijn, J.; Vermeire, T.; van der Zant, P. Trends in risk assessment of chemicals in the European Union. Human Ecol. Risk Assess. 2002, 8, 1825–1843. (2) EC. Regulation (EC) no. 1907/2006 of the European Parliament and of the Council of 18 December 2006 concerning the registration, evaluation, authorisation and restriction of chemicals (REACH), establishing a European Chemicals Agency, amending Directive 1999/45/EC and repealing Council Regulation (EEC) no. 793/93 and Commission Regulation (EC) no. 1488/94 as well as Council Directive 76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/105/EC, and 2000/21/EC. Off. J. Eur. Union 2007L 136, pg 3–280. (3) Environment Canada. Draft new substances notification regulations (chemicals and polymers) Can. Gaz., Part 1 2004, 3032– 3082. (4) Asfaw, A.; Ellersieck, M. R.; Mayer, F. L. Interspecies Correlation Estimations (ICE) for Acute Toxicity to Aquatic Organisms and Wildlife. II User Manual and Software, EPA/600/R-03/106; United States Environmental Protection Agency: Washington DC, 2003. (5) Raimondo, S.; Mineau, P.; Barron, M. G. Estimation of chemical toxicity to wildlife species using interspecies correlation models. Environ. Sci. Technol. 2007, 41, 5888–5894. (6) EC. Directive 2003/15/EC of the European Parliament and of the Council of 27 February 2003 amending Council Directive 76/768/EEC on the approximation of the laws of the member states relating to cosmetic products Off. J. Eur. Union 2003, L 66, pp. 26–35. (7) Dwyer, F. J.; Mayer, F. L.; Sappington, L. C.; Buckler, D. R.; Bridges, Bridges, C.M.; Greer, I. E.; Hardesty, D. K.; Henke, C. E.; Ingersoll, C. G.; Kunz, J. L.; White, D. W.; Augsberger, T.; Mount, D. R.; Hattala, K.; Neuderfer, G. N. Assessing contaminant sensitivity of endangered and threatened aquatic species: Part I. Acute toxicity of five chemicals. Arch. Environ. Contam. Toxicol. 2005, 48, 143–154. (8) Dyer, S. D.; Versteeg, D. J.; Belanger, S. E.; Chaney, J. G.; Mayer, F. L. Interspecies correlation estimates predict protective environmental concentrations. Environ. Sci. Technol. 2006, 40, 3102–3111. (9) Species Sensitivity Distributions in Ecotoxicology; Posthuma, L., Suter, G. W., Traas, T. P., Eds.; Lewis Publishers: Boca Raton, FL, 2002. (10) Barron, M. G.; Wharton, S. R. 2005 survey of methodologies for developing media screening values for ecological risk assessment. Integr. Environ. Assess. Manage. 2005, 1, 320–332. (11) USEPA. Ambient Water Quality Criteria for Aluminum, 440/588-008; U.S. Environmental Protection Agency: Washington, DC, 1988. (12) USEPA. Ambient Water Quality Criteria for Beryllium, 440/580-024; U.S. Environmental Protection Agency: Washington DC, 1980. (13) USEPA. 2001 Update of Ambient Water Quality Criteria for Cadmium, 822-R-01-001; U.S. Environmental Protection Agency: Washington, DC, 2001. (14) USEPA. Ambient Water Quality Criteria for Chromium, 440/ 5-84-029; U.S. Environmental Protection Agency: Washington, DC, 1984. (15) USEPA. Ambient Water Quality Criteria for Chromium, 440/ 5-84-029; U.S. Environmental Protection Agency: Washington, DC, 1984. (16) USEPA. 2003 Draft Update of Ambient Water Quality Criteria for Copper, 822-R-03-026;. U.S. Environmental Protection Agency: Washington, DC, 2003. (17) USEPA Ambient Water Quality Criteria for Lead, 440/5-84-027; U.S. Environmental Protection Agency: Washington, DC, 1984. (18) USEPA. Ambient Water Quality Criteria for Arsenic, 440/5-84033; U.S. Environmental Protection Agency: Washington, DC, 1984. (19) USEPA. Ambient Water Quality Criteria for Mercury, 440/5-84026; U.S. Environmental Protection Agency: Washington, DC, 1984. (20) USEPA. Ambient Water Quality Criteria for Arsenic, 440/5-84033; U.S. Environmental Protection Agency: Washington, DC, 1984. (21) USEPA. Ambient Water Quality Criteria for Nickel, 440/5-86004; U.S. Environmental Protection Agency: Washington, DC, 1986. (22) USEPA. Ambient Water Quality Criteria for Selenium, 822-D04-001; U.S. Environmental Protection Agency: Washington DC, 2004. 3082

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 42, NO. 8, 2008

(23) USEPA. Ambient Water Quality Criteria for Selenium, 822-D04-001; U.S. Environmental Protection Agency: Washington DC, 2004. (24) USEPA Ambient Water Quality Criteria for Zinc, 440/5-87-003; U.S. Environmental Protection Agency: Washington, DC, 1987. (25) USEPA Ambient Water Quality Criteria for Arsenic, 440/5-84033; U.S. Environmental Protection Agency: Washington, DC, 1984. (26) USEPA. Ambient Water Quality Criteria for Tributyltin, 822-R03-031; U.S. Environmental Protection Agency: Washington, DC, 2003. (27) USEPA. 1999 Update of Ambient Water Quality Criteria for Ammonia, 822-R-99-014; U.S. Environmental Protection Agency: Washington, DC, 1999. (28) USEPA. Ambient Water Quality Criteria for Chlorine, 440/584-030; U.S. Environmental Protection Agency: Washington DC, 1984. (29) USEPA. Ambient Water Quality Criteria for Cyanide, 440/5-84028; U.S. Environmental Protection Agency: Washington, DC, 1984. (30) USEPA. Ambient Water Quality Criteria for Chloride, 440/588-001; U.S. Environmental Protection Agency: Washington, DC, 1988. (31) USEPA. Ambient Water Quality Criteria for Chlorinated Ethanes, 440/5-80-029; U.S. Environmental Protection Agency: Washington. DC, 1980. (32) USEPA. Ambient Water Quality Criteria for Dichloroethylenes, 440/5-80-041; U.S. Environmental Protection Agency: Washington, DC, 1980. (33) USEPA. Ambient Water Quality Criteria for Chlorinated Ethanes, 440/5-80-029; U.S. Environmental Protection Agency: Washington, DC, 1980. (34) USEPA. Ambient Water Quality Criteria for Halomethanes, 440/ 5-80-051;. U. Environmental Protection Agency: Washington, DC, 1980. (35) USEPA. Ambient Water Quality Criteria for Phthalate Esters, 440/5-80-067; U.S. Environmental Protection Agency: Washington, DC, 1980. (36) USEPA. Ambient Water Quality Criteria for Phthalate Esters, 440/5-80-067; U.S. Environmental Protection Agency: Washington, DC, 1980. (37) USEPA. Ambient Water Quality Criteria for Dichlorobenzenes, 440/5-80-039; U.S. Environmental Protection Agency: Washington, DC, 1980. (38) USEPA. Ambient Water Quality Criteria for Chlorinated Benzenes, 440/5-80-028; U.S. Environmental Protection Agency: Washington, DC, 1980. (39) USEPA. Ambient Water Quality Criteria for Atrazine, 822-R03-023; U.S. Environmental Protection Agency: Washington, DC, 2003. (40) USEPA. Ambient Water Quality Criteria for Benzene, 440/5-80018; U.S. Environmental Protection Agency: Washington, DC, 1980. (41) USEPA Ambient Water Quality Criteria for Chlorinated Benzenes, 440/5-80-028; U.S. Environmental Protection Agency: Washington, DC, 1980. (42) USEPA Ambient Water Quality Criteria for Ethylbenzene, 440/ 5-80-048; U.S. Environmental Protection Agency: Washington, DC, 1980. (43) USEPA Ambient Water Quality Criteria for Naphthalene, 440/ 5-80-059; U.S. Environmental Protection Agency: Washington, DC, 1980. (44) USEPA Ambient Water Quality Criteria for Toluene, 440/5-80075; U.S. Environmental Protection Agency: Washington, DC, 1980. (45) USEPA Ambient Water Quality Criteria for Chlorinated Phenols, 440/5-80-032; U.S. Environmental Protection Agency: Washington, DC, 1980. (46) USEPA Ambient Water Quality Criteria for Nitrophenols, 440/ 5-80-063; U.S. Environmental Protection Agency: Washington, DC, 1980. (47) USEPA Ambient Water Quality Criteria for 2-Chlorophenol, 440/ 5-80-034; U.S. Environmental Protection Agency: Washington, DC, 1980. (48) USEPA Ambient Water Quality Criteria for Nitrophenols, 440/ 5-80-063; U.S. Environmental Protection Agency: Washington, DC, 1980. (49) USEPA Ambient Water Quality Criteria for Nonylphenol, 822R-05-005; U.S. Environmental Protection Agency: Washington, DC, 2005.

(50) USEPA Ambient Water Quality Criteria for Pentachlorophenol, 440/5-86-009; U.S. Environmental Protection Agency: Washington, DC, 1986. (51) USEPA Ambient Water Quality Criteria for Phenol, 440/5-80066; U.S. Environmental Protection Agency: Washington, DC, 1980. (52) USEPA Ambient Water Quality Criteria for Chlorpyrifos, 440/ 5-86-005; U.S. Environmental Protection Agency: Washington, DC, 1986. (53) USEPA Ambient Water Quality Criteria for Diazinon, 822-R05-006; U.S. Environmental Protection Agency: Washington, DC, 2005. (54) USEPA Ambient Water Quality Criteria for Parathion, 440/586-007; U.S. Environmental Protection Agency: Washington, DC, 1986. (55) USEPA Ambient Water Quality Criteria for Aldrin/Dieldrin, 440/ 5-80-019; U.S. Environmental Protection Agency: Washington, DC, 1980. (56) USEPA Ambient Water Quality Criteria for Chlordane, 440/580-027; U.S. Environmental Protection Agency: Washington, DC 1980. (57) USEPA Ambient Water Quality Criteria for DDT, 440/5–80–038; U.S. Environmental Protection Agency: Washington, DC, 1980. (58) USEPA Ambient Water Quality Criteria for Aldrin/Dieldrin, 440/ 5-80-019; U.S. Environmental Protection Agency: Washington, DC, 1980. (59) USEPA Ambient Water Quality Criteria for Endosulfan, 440/580-046; U.S. Environmental Protection Agency: Washington, DC, 1980. (60) USEPA Ambient Water Quality Criteria for Endrin, 440/5-80047; U.S. Environmental Protection Agency: Washington, DC, 1980. (61) USEPA Ambient Water Quality Criteria for Heptachlor, 440/580-052; U.S. Environmental Protection Agency: Washington, DC 1980. (62) USEPA Ambient Water Quality Criteria for Hexachlorobutadiene, 440/5-80-053;U.S. Environmental Protection Agency: Washington, DC, 1980. (63) USEPA Ambient Water Quality Criteria for Hexachlorocyclohexane, 440/5-80-054;U.S. Environmental Protection Agency: Washington, DC, 1980. (64) USEPA Ambient Water Quality Criteria for Hexachlorocyclopentadience, 440/5-80-055; U.S. Environmental Protection Agency: Washington, DC, 1980. (65) USEPA Ambient Water Quality Criteria for Toxaphene, 440/586-006; U.S. Environmental Protection Agency: Washington, DC, 1986.

(66) Mayer, F. L. Acute Toxicity Handbook of Chemicals to Estuarine Organisms, report no. EPA/600/8–87/017; U.S. Environmental Protection Agency: Gulf Breeze, FL, 1987; p 274. (67) Mayer, F. L.; Ellersieck, M. R. 1986 Manual of Acute Toxicity: Interpretation and Data Base for 410 Chemicals and 66 Species of Freshwater Animals, Resource Publication 160; U.S. Fish and Wildlife Service: Washington, DC, 1986. (68) USEPA. ECOTOX. Release 2.0. Ecotoxicology Database; U.S. Environmental Protection Agency: Duluth, MN, 2002; www. epa.gov/ecotox. (69) Buckler, D. R.; Mayer F. L.; Ellersieck M. R., Asfaw, A. Evaluation of Minimum Data Requirements for Acute Toxicity Value Extrapolation, EPA/600/R-03/105; United States Environmental Protection Agency: Washington, DC, 2003. (70) Snedecor, G. W.; Cochran, W. G. Statistical Methods, 7th ed.; Iowa State University Press: Ames, IA, 1980. (71) Versteeg, D. J.; Belanger, S. E.; Carr, G. J. Understanding single species and model ecosystem sensitivity: A data based comparison. Environ. Toxicol. Chem. 1999, 18, 1329–1346. (72) Aldenberg, T.; Jaworska, J. Uncertainty of the hazardous concentration and fraction affected for normal species sensitivity distributions. Ecotoxicol. Environ. Saf. 2000, 46, 1–18. (73) Jager, T.; Posthuma, L.; de Zwart, D.; van den Meent, D. Novel view on predicting acute toxicity: decomposing toxicity data in species vulnerability and chemical potency. Ecotox. Environ. Safety 2007, 67, 311–322. (74) EU. Technical Guidance Document on Risk Assessment, EUR 20418 EN/2; European Chemicals Bureau: Ispra, Italy, 2003. (75) Maltby, L.; Blake, N.; Brock, T. C. M.; Van den Brink, P. J. Insecticide species sensitivity distributions: Importance of test species selection and relevance to aquatic ecosystems. Environ. Toxicol. Chem. 2005, 24, 379–388. (76) Baird, D. J.; Van den Brink, P. J. Using biological traits to predict species sensitivities to toxic substances. Ecotox. Environ. Safety 2007, 67, 296–301. (77) Sappington, L. C.; Mayer, F. L.; Dwyer, F. J.; Buckler, D. R.; Jones, J. R.; Ellersieck, M. R. Contaminant sensitivity of threatened and endangered fishes compared to standard surrogate species. Environ. Toxicol. Chem. 2001, 20, 2869–2876. (78) Hansen, J. A.; Lipton, J.; Welsh, P. G. Relative sensitivity of bull trout (Salvelinus confluentus) and rainbow trout (Oncorhynchus mykiss) to acute copper toxicity. Environ. Toxicol. Chem. 2002, 21, 633–639.

ES702302E

VOL. 42, NO. 8, 2008 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3083