Acute Toxicity Prediction to Threatened and ... - ACS Publications


Acute Toxicity Prediction to Threatened and...

0 downloads 120 Views 464KB Size

Subscriber access provided by UNIV OF CALIFORNIA SAN DIEGO LIBRARIES

Article

Acute toxicity prediction to threatened and endangered species using Interspecies Correlation Estimation (ICE) models Morgan M Willming, Crystal R Lilavois, Mace G. Barron, and Sandy Raimondo Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.6b03009 • Publication Date (Web): 01 Sep 2016 Downloaded from http://pubs.acs.org on September 12, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 29

Environmental Science & Technology

1

Acute toxicity prediction to threatened and endangered species using Interspecies Correlation Estimation (ICE) models

2 3 4

Morgan M. Willming*†, Crystal R. Lilavois‡, Mace G. Barron‡, Sandy Raimondo‡

5 6 †

7 8 9 10



Oak Ridge Institute for Science and Education, U.S. Environmental Protection Agency, Gulf Ecology Division, 1 Sabine Island Drive, Gulf Breeze, FL USA 32561

U.S. Environmental Protection Agency, National Health and Environmental Effects Laboratory, Gulf Ecology Division, 1 Sabine Island Drive, Gulf Breeze, FL USA 32561

11 12

*Corresponding author: [email protected]; Phone: 850-934-9297; Fax: 851-934-2406

1 ACS Paragon Plus Environment

Environmental Science & Technology

13 14

Page 2 of 29

Abstract Evaluating contaminant sensitivity of threatened and endangered (listed) species and

15

protectiveness of chemical regulations often depends on toxicity data for commonly tested

16

surrogate species. The U.S. EPA’s internet application Web-ICE is a suite of Interspecies

17

Correlation Estimation (ICE) models that can extrapolate species sensitivity to listed taxa using

18

least squares regressions of the sensitivity of a surrogate species and a predicted taxon (species,

19

genus, or family). Web-ICE was expanded with new models that can predict toxicity to over 250

20

listed species. A case study was used to assess protectiveness of genus and family model

21

estimates derived from either geometric mean or minimum taxa toxicity values for listed species.

22

Models developed from the most sensitive value for each chemical were generally protective of

23

the most sensitive species within predicted taxa, including listed species, and were more

24

protective than geometric means models. ICE model estimates were compared to HC5 values

25

derived from Species Sensitivity Distributions for the case study chemicals to assess

26

protectiveness of the two approaches. ICE models provide robust toxicity predictions and can

27

generate protective toxicity estimates for assessing contaminant risk to listed species.

28

2 ACS Paragon Plus Environment

Page 3 of 29

29 30

Environmental Science & Technology

Introduction The U.S. Endangered Species Act (ESA) requires EPA to determine risks of pesticides

31

and other chemicals to federally endangered and threatened (listed) species to ensure that

32

chemical registration and water quality criteria are protective of listed taxa. A significant

33

challenge in this process is determining the sensitivity of the diversity of listed species to

34

chemicals that have only been evaluated using common test species or have limited data

35

available. Historically, toxicity data for the majority of listed species or closely related

36

representatives has been unavailable because of a lack of standardized culture and test methods

37

and limited organism availability. However, previous research suggests that listed species are not

38

consistently more sensitive than more commonly tested species1, 2. In the absence of species-

39

specific toxicity data, conservative approaches such as applying generic safety factors to toxicity

40

values of surrogate species have been used to develop hazard levels assumed to be protective of

41

listed species. Recently, the U.S. National Research Council3 recommended the use of

42

Interspecies Correlation Estimation (ICE) models in pesticide risk assessments as an alternative

43

to generic safety factors.

44

ICE models estimate acute toxicity to aquatic or terrestrial organisms using the known

45

toxicity data of a chemical for a surrogate species4, 5. The models are log-linear least squares

46

regressions of the sensitivity of a predicted taxon (species, genus, or family) and a surrogate

47

species. They are constructed from existing acute toxicity values determined in each pair of taxa

48

across a number of chemicals. ICE models for aquatic species contain acute toxicity values

49

(median lethal or median effect concentrations; LC50/EC50) for a minimum of any three

50

different chemicals and have been demonstrated to be robust, accurate estimators of toxicity

51

when the surrogate and predicted taxa are within the same taxonomic Order4. Species, genus, and

3 ACS Paragon Plus Environment

Environmental Science & Technology

52

family level ICE models are publically available on the U.S. EPA internet application Web-ICE

53

(www3.epa.gov/webice).

54

Page 4 of 29

ICE model predictions are intended to supplement toxicity databases where species of

55

concern or a required diversity of species have not been or cannot be adequately tested6. Previous

56

studies evaluating the robustness and application of ICE model predictions have focused

57

primarily on species level models and their use in either direct toxicity estimation or in

58

developing species sensitivity distributions (SSDs)4-8. Less work has explored the protectiveness

59

of ICE predictions in risk assessments for listed species or the use of genus and family models.

60

Few species-specific ICE models can be developed for listed species due to limited existing

61

toxicity data, therefore genus or family models may be required to predict toxicity to the higher

62

taxonomic level. Genus and family models have historically been developed using geometric

63

means of toxicity values from multiple species, which are useful in generating toxicity estimates

64

in cases such as water quality criteria development. However, the protectiveness of these

65

predictions across the range of species sensitivity within the predicted taxon has not been

66

investigated, especially for listed or sensitive species. Demonstrating protectiveness of model

67

estimates is a critical step for their application in risk assessment of listed species.

68

The present study evaluates the use of ICE models for listed species and compares the

69

protectiveness of genus and family level models developed from either geometric means or

70

minimum toxicity values. Models used in this evaluation were developed from an acute toxicity

71

database that was substantially expanded from the previous version of Web-ICE (v3.2, release

72

April 2013) and contains 314 species and 1501 chemicals, including new toxicity records for

73

listed species. Because there are 87 federally listed species of freshwater unionid mussels, which

74

are often the most sensitive taxa to some contaminants9, development of models for this taxon

4 ACS Paragon Plus Environment

Page 5 of 29

Environmental Science & Technology

75

was a focus for the present study. We compare the development of genus and family models

76

using the most sensitive toxicity value for each chemical within the predicted taxon (minimum

77

models) to models built with geometric means in order to evaluate an approach for generating

78

protective estimates of toxicity. Models developed with minimum toxicity values were expected

79

to result in toxicity predictions that are protective of the most sensitive species within a taxon.

80

We also revised model selection guidelines to reduce reliance on professional judgement and

81

increase reproducibility of their application. Lastly, we demonstrate the application of models in

82

a case study of 11 federally listed species exposed to a diversity of contaminants with a range of

83

aquatic toxicity modes of action (MOA) and compared ICE predictions to hazard concentrations

84

derived from species sensitivity distributions (SSDs).

85 86

Methods

87

Database development

88

Standardization of toxicity data for inclusion in ICE models followed the approach and

89

selection criteria described in Raimondo et al.4 This included confirming chemical and species

90

identity; compiling acute toxicity values as 48 hour EC50/LC50 for specific invertebrate taxa

91

(e.g., daphnids, fairy shrimp) or 96 hour EC50/LC50 for fish, amphibians and other aquatic

92

invertebrates (e.g., insects); and determining whether each data record met standardization

93

criteria for life stage, test conditions, and water quality parameters (e.g., temperature, dissolved

94

oxygen, salinity)4. The database included records from the previous Web-ICE database appended

95

with new records from ECOTOX (www.epa.gov/ecotox/; downloaded September 2014), and

96

recently collected primary data on freshwater mussels9 and fairy shrimp (unpublished data).

97

Acute toxicity data, which can represent intrinsic species sensitivity, were used because of the

98

large diversity of chemicals and taxa available. Data were standardized for life stage by using 5 ACS Paragon Plus Environment

Environmental Science & Technology

Page 6 of 29

99

only juveniles for fish and decapods, immature aquatic life stages for amphibians and insects,

100

and juveniles and spat for molluscs. All life stages were included for other taxa groups except

101

egg or embryo stages. Specific aspects of the mussel toxicity dataset are detailed in Raimondo et

102

al.9. All chemicals in the ICE database were curated using the distributed structure-searchable

103 104

toxicity database (DSSTox; www.epa.gov/ncct/dsstox/). A single name and the confirmed

105

chemical abstract services registry number from the source material were checked against

106

DSSTox to validate their consistency. Names that were not contained within the DSSTox list of

107

synonyms for a particular chemical were manually checked to validate the agreement between

108

the chemical identifiers and confirm the chemical-data linkage. The ICE database toxicity value

109

was specified as the compound tested, except for some metal salts (recorded as the element or

110

hardness normalized element), pentachlorophenol (pH normalized) and ammonia (pH and

111

temperature normalized). Only records for chemicals with an active ingredient purity of ≥ 90%

112

were accepted. Open ended (e.g., LC50 > 100 mg/L) or unconfirmed toxicity values were not

113

used.

114 115 116

ICE Model Development Models were developed as least squares log-linear regressions by pairing all possible

117

surrogate species toxicity values with all possible predicted taxa toxicity values (i.e., a species,

118

genus, or family) by common chemical. Each ICE model describes the relationship of sensitivity

119

of the two species as Log10(Predicted Toxicity) = a*Log10(Surrogate Toxicity)+b.

120

Here, Predicted Toxicity is the EC/LC50 value of the predicted taxa, Surrogate Toxicity is the

121

EC/LC50 value of the surrogate species, a is the slope of the regression line and b is the

122

intercept. A slope of 1 describes two species whose relative sensitivity is consistent across a wide 6 ACS Paragon Plus Environment

Page 7 of 29

Environmental Science & Technology

123

range of chemicals. A slope of 1 and an intercept of 0 describes two species with identical

124

sensitivity. Each ICE model is also described by an R2 and a Mean Square Error (MSE). The R2

125

is a measure of goodness of fit of the regression, where a value closer to one represents a model

126

where data are tightly fit around the line and variability in the data are well represented by the

127

model. Model MSE is an estimate of average error contained within the model. Small MSE

128

indicates that the model is robust, or contains high probability of good performance for data

129

drawn from a wide range of samples.

130

Each model required toxicity values from at least three different chemicals available in

131

the database for both the surrogate species and predicted taxon. Therefore, each model data point

132

represents a unique chemical. For species level models, when multiple database records occurred

133

for the same species and chemical, the geometric mean of toxicity values was used. For genus

134

and family level models, two sets of models were developed: geometric mean and minimum

135

toxicity models. First, all species available in the database within a genus or family were

136

identified by common chemical. For geometric mean models, the species mean acute value

137

(SMAV; geometric mean for each chemical) was calculated. For genus geometric mean models,

138

the genus mean acute value was then calculated for each chemical as the geometric mean of the

139

SMAVs. For family geometric mean models, the family mean acute value was calculated for

140

each chemical as the geometric mean of the SMAV. For minimum toxicity models predicting to

141

both genus and family, the minimum toxicity value of all species within each predicted taxa was

142

used for each chemical. This second approach using the most sensitive toxicity value was

143

intended to generate toxicity estimations protective of more species within that genus or family

144

compared to geometric mean models. For both sets of genus and family level models, the

7 ACS Paragon Plus Environment

Environmental Science & Technology

145

predicted taxon toxicity value was paired with the SMAV of the respective chemical for the

146

surrogate species.

147

Page 8 of 29

For ICE models with four or more chemicals, model prediction accuracy was determined

148

using leave-one-out cross-validation4. In this process, each data point within a model (e.g.,

149

chemical pair of acute values for surrogate and predicted taxon) was systematically removed

150

from the original model and a new model was rebuilt with the remaining data. The new model

151

was used to estimate the toxicity value of the removed predicted taxa from the removed

152

surrogate species toxicity value. For each removed data point, the N-fold difference was

153

calculated as the greater value of the estimated/actual or actual/estimated for non-transformed

154

values. For each model, the number of removed data points predicted within 5-fold of the actual

155

value was determined and a cross-validation (prediction) success rate was calculated as the

156

percentage of removed values predicted within 5-fold of the actual value 4. The 5-fold range

157

represents standard inter-laboratory variation of a toxicity value tested for the same species and

158

chemical as demonstrated in previous studies4, 5, 9-11.

159 160 161

Model Selection Guidelines Model selection guidelines were developed to reduce the amount of professional

162

judgement needed to select the best available model when multiple ICE models (i.e., multiple

163

surrogate species) are available for predicting to one taxon. Details of guidance development are

164

provided in Supplementary Information (SI 001). Briefly, selection guidance was determined

165

from the combination of regression model attributes (MSE, R2, and slope) that corresponded to

166

the highest percent of cross-validated data points predicted within 5-fold of the measured value

167

using an iterative approach. The highest MSE, lowest R2 and lowest slope that corresponded to

168

the highest percent accuracy were then applied as the model selection guidelines. 8 ACS Paragon Plus Environment

Page 9 of 29

Environmental Science & Technology

169 170 171

Evaluation of Web-ICE for Listed Species A case study was used to assess the protectiveness of genus and family models applied to

172

listed species. The 11 selected species represented a diverse taxonomic range including fish,

173

amphibians, and fairy shrimp from the Sacramento California Valley12, and freshwater unionid

174

mussels from the Ohio River watershed and southeastern United States (Table 1). The 16

175

chemicals chosen for the case study represented priority chemicals encompassing a broad range

176

of MOAs (primarily pesticides from the Sacramento Valley) and contaminants of concern for

177

unionid mussels (Table 1). First, all minimum and geometric mean ICE models available for

178

predicting to the genera and families of the case study species were identified. For each of these

179

models, toxicity values in the Web-ICE database were extracted for each surrogate species and

180

case study chemical. When multiple toxicity records were available for the same surrogate

181

species and chemical, the SMAV was used as the model input. Toxicity estimates and 95%

182

confidence intervals (CI) were calculated from available surrogate species toxicity and models

183

for each chemical and taxa (genus, family). Where the taxa of a listed species could be predicted

184

from multiple surrogates, we applied the model selection guidance determined in the methods

185

outlined above to identify the best model (Fig. 1). According to the guidance, model selection

186

was first based on the taxonomic distance of the surrogate and predicted taxon, because previous

187

analyses indicated higher prediction accuracy for models from closely related taxa. Models were

188

then chosen with a low MSE (≤ 0.95), and a narrow 95% CI. If models had similar MSE, the

189

model with the greater N was selected.

190

Protectiveness of the toxicity estimates and their lower 95% CI predicted from the

191

selected model for the genus and/or family of each listed species was determined by comparing

192

these values to the measured minimum toxicity value for that taxon available in the ICE 9 ACS Paragon Plus Environment

Environmental Science & Technology

Page 10 of 29

193

database. For example, the model prediction to the family Salmonidae for malathion was

194

compared to the minimum Salmonidae LC50 (i.e., the most sensitive measured species value for

195

all salmonids) for malathion available in the ICE database. If the model prediction was greater

196

than the minimum toxicity value in the database, then the lower 95% CI of the prediction was

197

compared to the minimum taxa toxicity value. For each listed taxa and chemical, it was

198

determined if the prediction was protective, the lower 95% CI was protective, or models were

199

not protective at all. This was performed for genus and family models developed using both

200

minimum toxicity values and geometric means.

201

Species sensitivity distributions (SSDs) were developed for each chemical to compare the

202

5th percentile hazard concentrations (HC5) values to ICE generated predictions and sensitive taxa

203

values for the case study species. SSDs were developed from freshwater species data available in

204

the ICE database for each case study chemical2, 6. Toxicity data selection followed the same

205

standardization criteria as ICE database development, using the SMAV for each chemical to

206

build the SSD. SSDs were fit using maximum likelihood estimation in R statistical software13 to

207

a log-logistic cumulative distribution so that Y = 1/{1 + exp[(α- X)/β]}

208 209

where Y is the cumulative proportion, X is the log-transformed toxicity, α is the intercept, and β

210

is the scale parameter. The HC5 values were derived from this distribution for each chemical and

211

were compared to the most sensitive taxa values and ICE minimum model predictions to assess

212

protectiveness of each approach.

213 214

Results

215

Database and model development

10 ACS Paragon Plus Environment

Page 11 of 29

216

Environmental Science & Technology

The Web-ICE database (v. 3.3) has an expanded number of toxicity records, species, and

217

chemicals, and yielded considerably more models, including those predicting to listed species,

218

compared to the previous version (Table 2). This suite of ICE models can predict toxicity values

219

to 25 listed species and 20 genera and 20 families containing U.S. federally listed species (Table

220

3). Overall, species model prediction accuracy and cross-validation success were consistent with

221

previous results for the previous dataset 4. Models developed with predicted and surrogate

222

species within the same family estimated toxicity within 5- and 10-fold of the actual value for 92

223

and 98% of data points, respectively (Table S1). All models are available on the EPA Web-ICE

224

application webpage (www3.epa.gov/webice).

225

226 227

Model selection & prediction guidance Analysis of model parameters (MSE, R2, and slope) indicated that models with MSE <

228

0.95, R2 > 0.6, and slope > 0.6 will have the highest prediction accuracy (Table S2). For models

229

with a taxonomic distance of 6 (same Kingdom) that have N > 10, models with MSE < 0.55 will

230

have improved prediction accuracy because of the increased variation in toxicity data associated

231

with less closely related taxa (Table S2). Distributions of each model parameter by taxonomic

232

distance are available in the Supporting Information (Fig. S1).

233

Figure 1 summarizes the procedures for model selection based on the model guidelines

234

when models for multiple surrogates are available. Web-ICE users should first select models

235

with the closest taxonomic distance and then apply the model parameter guidelines to identify

236

the best model. The analysis also indicated selection of models with confidence intervals within

237

5-fold of the predicted value, although using 3-fold slightly improves prediction accuracy, but

238

not substantially. 11 ACS Paragon Plus Environment

Environmental Science & Technology

Page 12 of 29

239

240 241

Case study analysis of model protectiveness For the case study of 16 chemicals and 11 listed species, there were 787 minimum

242

toxicity models and 807 geometric mean models from which predictions could be made from

243

available surrogate toxicity values and ICE models for the predicted taxa. Applying model

244

selection guidance resulted in 50 ICE predictions for the listed species for each set of models

245

(Supporting Information SI 002; Table S3). For both minimum and geometric mean models, 25

246

of the 50 predicted values were from family models and 25 were genus models. The Supporting

247

Information (SI 002) provides all available minimum models predicting to these taxa with the

248

best selected model identified. When model predictions were available for both a family and a

249

genus within that same family (e.g., Oncorhynchus and Salmonidae), we selected the family

250

model if both models had similar parameters (Table S4). Because family models are comprised

251

of greater taxa diversity, estimates derived from these models can be considered to be protective

252

of more species. In some cases, genus models may generate more conservative estimates, but

253

model selection should ultimately be guided by the goal of the particular analysis; if the goal is

254

specific to a species and location, a genus level model will more accurately reflect the sensitivity

255

of the species, if the goal is comprehensive protection of listed species, the family model will be

256

more inclusive.

257

The predicted values for genus level minimum models were protective of the most

258

sensitive species within the genus in 84% of cases and for 100% of instances when using the

259

lower 95% CI (Fig. 2). For family level minimum models, predictions were protective in 72% of

260

cases and the lower 95% CI was protective in 88% of instances (Fig. 2). Of the family level

261

models where the toxicity of the most sensitive species was less than the lower 95% CI of the

12 ACS Paragon Plus Environment

Page 13 of 29

Environmental Science & Technology

262

prediction (12%, N = 3 models), two of these models were at a taxonomic distance of greater

263

than 5 because no models with a closer taxonomic distance were available. For models

264

developed with geometric means, model predictions were protective of the most sensitive species

265

in 28% of cases for both genus and family models (Fig. 2). The lower 95% CI was protective in

266

50% of genus models and 48% of family models built with geometric means.

267

SSDs were developed for 13 chemicals that had sufficient data. Fipronil, glyphosate, and

268

imidacloprid each only had 4 standardized toxicity values, so SSDs were not constructed. The

269

number of species in the SSDs ranged from 7 (thiobencarb) to 66 (copper), and HC5 values

270

ranged from 0.08 to 1,389,897 µg/L (Table 4). SSDs with large N contained diverse taxa while

271

those with small N ( 0.6 and slope > 0.6

Fig. 1. ICE model selection guidance steps based on model parameters used to identify the best model when predictions from multiple surrogates are available. TD = Taxonomic distance of surrogate and predicted taxon; MSE = Mean square error; N = Number of data points in the model (df + 2); CI = Confidence interval.

27 ACS Paragon Plus Environment

Environmental Science & Technology

Page 28 of 29

Figure 2

Fig. 2. Comparison of model protectiveness for geometric mean and minimum genus and family models used with priority chemicals and taxa containing listed species in the Sacramento Valley and listed mussels. There were 25 genus model and 25 family model predictions for each set of models (geometric mean or minimum models).

28 ACS Paragon Plus Environment

Page 29 of 29

Environmental Science & Technology

TOC/Abstract Art

29 ACS Paragon Plus Environment