Aquatic Toxicology 161 (2015) 102–107
Contents lists available at ScienceDirect
Aquatic Toxicology journal homepage: www.elsevier.com/locate/aquatox
MOAtox: A comprehensive mode of action and acute aquatic toxicity database for predictive model development M.G. Barron a,∗ , C.R. Lilavois a , T.M. Martin b a b
U.S. Environmental Protection Agency, Office of Research Development, Gulf Ecology Division, Gulf Breeze, FL, USA U.S. Environmental Protection Agency, Office of Research Development, Sustainable Technology Division, Cincinnati, OH, USA
a r t i c l e
i n f o
Article history: Received 19 November 2014 Received in revised form 31 January 2015 Accepted 2 February 2015 Available online 7 February 2015 Keywords: QSAR Toxicity Model Mode of action Fish Invertebrate
a b s t r a c t The mode of toxic action (MOA) has been recognized as a key determinant of chemical toxicity and as an alternative to chemical class-based predictive toxicity modeling. However, the development of quantitative structure activity relationship (QSAR) and other models has been limited by the availability of comprehensive high quality MOA and toxicity databases. The current study developed a dataset of MOA assignments for 1213 chemicals that included a diversity of metals, pesticides, and other organic compounds that encompassed six broad and 31 specific MOAs. MOA assignments were made using a combination of high confidence approaches that included international consensus classifications, QSAR predictions, and weight of evidence professional judgment based on an assessment of structure and literature information. A toxicity database of 674 acute values linked to chemical MOA was developed for fish and invertebrates. Additionally, species-specific measured or high confidence estimated acute values were developed for the four aquatic species with the most reported toxicity values: rainbow trout (Oncorhynchus mykiss), fathead minnow (Pimephales promelas), bluegill (Lepomis macrochirus), and the cladoceran (Daphnia magna). Measured acute toxicity values met strict standardization and quality assurance requirements. Toxicity values for chemicals with missing speciesspecific data were estimated using established interspecies correlation models and procedures (Web-ICE; http://epa.gov/ceampubl/fchain/webice/), with the highest confidence values selected. The resulting dataset of MOA assignments and paired toxicity values are provided in spreadsheet format as a comprehensive standardized dataset available for predictive aquatic toxicology model development. Published by Elsevier B.V.
1. Introduction In aquatic toxicology, mode of action (MOA) has been defined phenomenologically based on whole organism responses, identified from effects at the cellular or organ level, or based on the primary mechanism of toxicity initiated at the receptor level (Barron et al., 2002; Suter, 2007). MOA has been recognized as a key determinant of chemical toxicity and as an alternative to chemical class-based predictive toxicity modeling. Early work on determining chemical mode of action in aquatic organisms focused on industrial compounds that were largely of moderate polarity, low toxicity and acted through narcosis, a non-specific reversible baseline mode of whole organism toxicity (Konemann, 1981; Call et al., 1985). Based on a suite of behavioral/morphological
∗ Corresponding author at: U.S. Environmental Protection Agency, Office of Research and Development, Gulf Ecology Division, 1 Sabine Island Drive, Gulf Breeze, FL 32561, USA. Tel.: +1 850 934 9223; fax: +1 850 934 2402. E-mail address: [email protected] (M.G. Barron). http://dx.doi.org/10.1016/j.aquatox.2015.02.001 0166-445X/Published by Elsevier B.V.
responses in rainbow trout (Onchorhynchus mykiss), McKim et al. (1987) established several modes of action including narcosis, oxidative phosphorylation uncoupler, acetylcholinesterase (AChE) inhibitor, or stimulant-like. Verhaar et al. (1992) categorized chemicals into one of four MOA classes based on structural rules: (1) inert compounds causing narcosis, (2) less inert more toxic compounds causing polar narcosis, (3) reactive compounds with enhanced toxicity, and (4) compounds with specific or receptor mediated toxicity. Russom et al. (1997) utilized an extensive fathead minnow (Pimephales promelas) acute toxicity and behavioral database to extend MOA categorization to three narcosis MOAs and five specific MOAs (AChE inhibition, uncoupling, reactivity, seizure mechanisms, and respiratory inhibition). Other approaches to MOA assignment and categorization have included assessments of in vitro activity and cross species receptor homology for pharmaceutical compounds (Escher et al., 2005; Christen et al., 2010). In contrast to empirical and rule-based categorization approaches, Martin et al. (2013) utilized two dimensional descriptor assignments and machine-based learning algorithms to develop a computational chemistry approach to MOA assignment
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
for six broad and seven specific MOAs. This computational chemistry-based approach calculated up to 700 structural and property descriptors for each chemical and developed prediction models based on multi-linear regression models and decision tree algorithms. As part of that work, a dataset of 924 acute aquatic toxicity MOA assignments were developed using a combination of high confidence approaches that included international consensus classifications, quantitative structure activity relationship (QSAR) predictions, and weight of evidence professional judgment based on an assessment of structure and literature information. The objective of the current study was to develop a comprehensive standardized dataset of aquatic toxicity values and MOA assignments. The current research expands the Martin et al. (2013) dataset to over 1200 chemicals including metals and organometallics, increases the number of specific MOA categories, and refines some of the pesticide neurotoxicity MOA categories to be less general and more mechanism-based. The categorization scheme was comprised of six broad and 31 specific MOA categories that encompassed the major acute toxicity MOAs reported in aquatic invertebrates and fish. Chemicals in the MOA dataset were then linked to a dataset of high confidence acute aquatic toxicity values developed for this study to create a database of 674 chemicals with species-specific acute toxicity values. The MOA-aquatic toxicity database (MOAtox) development focused on the four most commonly reported aquatic test species with toxicity endpoint data in the ECOTOX database (USEPA, 2014): rainbow trout, the cladoceran (Daphnia magna), fathead minnow, and bluegill (Lepomis macrochirus). These species represented 22% of the 300,000 ECOTOX records with toxicity endpoint data (Fairbrother et al., 2014). Additionally, generic fish and aquatic invertebrate acute toxicity values were developed from the geometric mean of all measured toxicity values that met strict quality assurance and standardization requirements. Toxicity values for chemicals with missing species-specific data were estimated using established interspecies correlation models and procedures (WebICE; http://epa.gov/ceampubl/fchain/webice/), with the highest confidence values selected (Raimondo et al., 2010, 2013). The development of this comprehensive standardized dataset is intended to facilitate the development of predictive aquatic toxicity and MOA models. The larger MOA dataset could support MOA-based assessments of chemical mixtures, such as applying a toxic units approach to chemicals with the same specific MOA.
2. Methods 2.1. MOA category and dataset development An aquatic toxicity MOA categorization scheme was developed for a diversity of metals, organometallics, pesticides, and other organics that appear in the ECOTOX database (USEPA, 2014), based on earlier work determining chemical modes of acute toxic action in fish (Konemann 1981; Call et al., 1985; McKim et al., 1987; Russom et al., 1997; USEPA, 2012; Martin et al., 2013). The categorization scheme was comprised of six broad and 31 specific MOA categories that encompassed the major MOAs reported in aquatic invertebrates and fish. Broad MOA categories included narcosis, AChE inhibition, neurotoxicity, iono/osmoregulatory/circulatory impairment, reactivity, and electron transport inhibition (Supplementary information; Table S1). Specific MOAs were developed as subcategories of the broad MOAs based on either chemical structure or known mechanism of action. A total of 1213 chemicals were assigned to a broad and specific MOA category following the general procedures of Martin et al. (2013). This involved using a combination of high confidence assignments including, international consensus classifications (e.g., IRAC, 2014), QSAR predictions
103
(USEPA, 2012), and weight of evidence professional judgment by considering a combination of assessment of structure (e.g., analogous structure; moiety/functional group presence) and literature information on MOA, mechanism of action and toxicity pathways (e.g., Russom et al., 1997; Nendza and Müller 2000; Schüürmann et al., 2003; Spycher et al., 2004). Chemicals with an uncertain MOA assignment were excluded from the database.
2.2. Toxicity dataset development Toxicity dataset development focused on determining acute toxicity values for three fish species (rainbow trout, fathead minnow, and bluegill) and Daphnia for chemicals in the MOA dataset. These four taxa were selected, because they have been the most commonly reported aquatic test species with acute toxicity endpoint data in the ECOTOX database (USEPA, 2014). First, measured aquatic toxicity values were obtained from the US EPA’s ICE Aquatic Toxicity Database (Raimondo et al., 2013a). This database is comprised of 5500 acute EC/LC50 values for 180 species and 1260 chemicals that have been obtained from a diversity of sources (primarily ECOTOX) then subjected to a rigorous quality assurance review and standardization procedure (Raimondo et al., 2010). Each toxicity record included in the dataset met specific requirements for: (1) test chemical identity and purity, (2) test procedures, conditions, reporting, and quality control, and (3) test organism life stage. Additionally, toxicity values for selected compounds were normalized based on water quality parameters, and outliers and central tendency were determined in cases of multiple reported values as documented in Raimondo et al. (2013a). Mean fish and mean aquatic invertebrate toxicity values were also computed from the geometric mean of all standardized values available for each chemical. Similarly, a mean Daphnia genus value was computed from all standardized values for each chemical with available data. Missing species-specific toxicity data were then estimated using established interspecies correlation estimation (ICE) models available in the USEPA Web-ICE tool (http://epa.gov/ceampubl/fchain/webice/; Raimondo et al., 2010, 2013b). ICE models are log-linear least square regressions of chemical sensitivity relationships between a surrogate and predicted taxon. Toxicity values in the Aquatic Toxicity Database for all available surrogate species were used to estimate acute toxicity for those chemicals with missing measured data for rainbow trout, D. magna, fathead minnow, or bluegill. This procedure resulted in multiple estimated values for many chemicals because ICE models were available for multiple surrogate species.
2.3. Determination of best estimated value A systematic process was used to determine a best toxicity estimate for any of the four species with multiple predicted values based on model uncertainty and performance criteria. First, each surrogate species ICE model was evaluated to determine if it met the following statistical criteria: mean square error of less than or equal to 0.94, coefficient of determination greater than 0.6, and the upper 95th percentile confidence limit was within five fold of the predicted toxicity value. If multiple models met all three statistical criteria, these models were evaluated to determine if they met the following additional criteria: taxonomic distance between the surrogate and predicted species was less than or equal to four (same Class), model cross validation success was greater than 85%, and degrees of freedom were greater than 8. If multiple models met all criteria, the best estimated value was computed from the geometric mean of the model predicted values.
104
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
2.4. Ranking and assessment of estimated values Each estimated value in the MOAtox database received a qualitative ranking based on the number of statistical criteria met (Supplementary information; Table S2): level P1 (highest confidence estimate; all three model criteria met), P2 (two met), P3 (one met), and P4 (lowest confidence; no statistical criteria met). Estimated values were not included in the MOAtox database for those chemicals with species-specific measured values. However, toxicity values were also estimated and ranked for those chemicals with measured values to allow an assessment of accuracy. Accuracy was determined from the proportion of species-specific estimated values within two, five, and ten fold of measured values. Correlations between species-specific estimated and measured values were determined from simple log–log linear regressions.
3. Results 3.1. MOA assignments The MOA dataset was comprised of 1213 chemicals assigned to one of six broad acute toxicity MOA categories and one of 31 specific MOA subcategories (Tables 1, S1, and S2 ). The predominant MOAs and chemical classes in the database (>40 chemicals) were AChE inhibition (carbamates and organophosphates), narcosis (nonpolar, polar, and ester compounds), neurotoxicity (pyrethroids and alicyclics), and uncouplers of oxidative phosphorylation (e.g., polychlorinated and nitrophenols and amines) (Tables 1 and S1). Fig. 2. Prediction accuracy of estimated toxicity values by uncertainty level (upper panel), and correlation between measured and predicted toxicity values (lower panel) in fathead minnows.
3.2. Toxicity values Of the 1213 chemicals in the MOA dataset, 614 had measured values in fish and 424 had measured values in invertebrates determined as the geomeans of standardized acute toxicity data (Table 2). There were 671 chemicals in the MOA dataset that had either a measured or estimated taxa-specific acute toxicity value for Daphnia, bluegill, fathead minnow, or rainbow trout (Table 2). Estimated values for the three fish species were primarily comprised of level 1 (highest confidence; >85%) values, whereas, in D. magna the majority of estimated values were level 3 (62%) (Table 2 and Figs. 1–4). Over 60% of the estimated values in the three fish species were within two fold of standardized measured values, and 87% were within five fold (Table 2 and Figs. 1–4). Prediction accuracy was lower for D. magna, with only 61% of estimated values within five fold of measured values (Table 2 and Fig. 4). In the three fish species, estimated values were highly correlated with measured values (R2 > 0.87; Figs. 1–4), whereas, there was greater variance in the relationship for D. magna (R2 = 0.69; Fig. 4). 4. Discussion
Fig. 1. Prediction accuracy of estimated toxicity values by uncertainty level (upper panel), and correlation between measured and predicted toxicity values (lower panel) in bluegill.
In aquatic toxicology, MOA-based toxicity prediction has primarily focused on organic chemicals acting through polar and nonpolar narcosis. Although compounds causing narcosis are structurally diverse and represent the largest class of organic chemicals produced globally, the narcosis MOA does not encompass more toxic compounds, such as pesticides, metals, and other chemicals acting through specific mechanisms and toxicity pathways. The current study builds upon the earlier work of Konemann (1981), Russom et al. (1997), Verhaar et al. (2000), and others to expand the chemical coverage and MOA categories to include
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
105
Table 1 Description of mode of action (MOA) categories used in the MOAtox database and number of chemicals within each MOA.a Broad MOA (n)b
AChE inhibition (285)
Iono/osmoregulatory/ circulatory impairment (55)
Narcosis (465)c
Neurotoxicity (200)
Reactivity (111)
Electron transport inhibition (97)
a b c
Specific MOA
Chemical class
Carbamate
Carbamate insecticides
Organophosphate
Organophosphorus insecticides Heavy metals, transition metals
Structural features
Example chemicals
#Chemicals in Databasea MOA
MOAtox
74
19
211
58
18
15
Oxime and methyl carbamate Organophosphates, organothiophosphates Divalent metals
Aldicarb, carbaryl, methiocarb Chlorpyrifos, malathion, terbufos Copper, zinc
Inorganics Chemical specific, nitrites
Chemical specific Chemical specific
6 6
4 1
Anticoagulation Ester
Coumarin rodenticides Mono or diesters,
Coumarin moiety Diverse structures, non-reactive
25 48
7 41
Non-polar
Neutral organics
Moderate polarity
347
283
Polar
Polar organics
Single nitro or
Contents lists available at ScienceDirect
Aquatic Toxicology journal homepage: www.elsevier.com/locate/aquatox
MOAtox: A comprehensive mode of action and acute aquatic toxicity database for predictive model development M.G. Barron a,∗ , C.R. Lilavois a , T.M. Martin b a b
U.S. Environmental Protection Agency, Office of Research Development, Gulf Ecology Division, Gulf Breeze, FL, USA U.S. Environmental Protection Agency, Office of Research Development, Sustainable Technology Division, Cincinnati, OH, USA
a r t i c l e
i n f o
Article history: Received 19 November 2014 Received in revised form 31 January 2015 Accepted 2 February 2015 Available online 7 February 2015 Keywords: QSAR Toxicity Model Mode of action Fish Invertebrate
a b s t r a c t The mode of toxic action (MOA) has been recognized as a key determinant of chemical toxicity and as an alternative to chemical class-based predictive toxicity modeling. However, the development of quantitative structure activity relationship (QSAR) and other models has been limited by the availability of comprehensive high quality MOA and toxicity databases. The current study developed a dataset of MOA assignments for 1213 chemicals that included a diversity of metals, pesticides, and other organic compounds that encompassed six broad and 31 specific MOAs. MOA assignments were made using a combination of high confidence approaches that included international consensus classifications, QSAR predictions, and weight of evidence professional judgment based on an assessment of structure and literature information. A toxicity database of 674 acute values linked to chemical MOA was developed for fish and invertebrates. Additionally, species-specific measured or high confidence estimated acute values were developed for the four aquatic species with the most reported toxicity values: rainbow trout (Oncorhynchus mykiss), fathead minnow (Pimephales promelas), bluegill (Lepomis macrochirus), and the cladoceran (Daphnia magna). Measured acute toxicity values met strict standardization and quality assurance requirements. Toxicity values for chemicals with missing speciesspecific data were estimated using established interspecies correlation models and procedures (Web-ICE; http://epa.gov/ceampubl/fchain/webice/), with the highest confidence values selected. The resulting dataset of MOA assignments and paired toxicity values are provided in spreadsheet format as a comprehensive standardized dataset available for predictive aquatic toxicology model development. Published by Elsevier B.V.
1. Introduction In aquatic toxicology, mode of action (MOA) has been defined phenomenologically based on whole organism responses, identified from effects at the cellular or organ level, or based on the primary mechanism of toxicity initiated at the receptor level (Barron et al., 2002; Suter, 2007). MOA has been recognized as a key determinant of chemical toxicity and as an alternative to chemical class-based predictive toxicity modeling. Early work on determining chemical mode of action in aquatic organisms focused on industrial compounds that were largely of moderate polarity, low toxicity and acted through narcosis, a non-specific reversible baseline mode of whole organism toxicity (Konemann, 1981; Call et al., 1985). Based on a suite of behavioral/morphological
∗ Corresponding author at: U.S. Environmental Protection Agency, Office of Research and Development, Gulf Ecology Division, 1 Sabine Island Drive, Gulf Breeze, FL 32561, USA. Tel.: +1 850 934 9223; fax: +1 850 934 2402. E-mail address: [email protected] (M.G. Barron). http://dx.doi.org/10.1016/j.aquatox.2015.02.001 0166-445X/Published by Elsevier B.V.
responses in rainbow trout (Onchorhynchus mykiss), McKim et al. (1987) established several modes of action including narcosis, oxidative phosphorylation uncoupler, acetylcholinesterase (AChE) inhibitor, or stimulant-like. Verhaar et al. (1992) categorized chemicals into one of four MOA classes based on structural rules: (1) inert compounds causing narcosis, (2) less inert more toxic compounds causing polar narcosis, (3) reactive compounds with enhanced toxicity, and (4) compounds with specific or receptor mediated toxicity. Russom et al. (1997) utilized an extensive fathead minnow (Pimephales promelas) acute toxicity and behavioral database to extend MOA categorization to three narcosis MOAs and five specific MOAs (AChE inhibition, uncoupling, reactivity, seizure mechanisms, and respiratory inhibition). Other approaches to MOA assignment and categorization have included assessments of in vitro activity and cross species receptor homology for pharmaceutical compounds (Escher et al., 2005; Christen et al., 2010). In contrast to empirical and rule-based categorization approaches, Martin et al. (2013) utilized two dimensional descriptor assignments and machine-based learning algorithms to develop a computational chemistry approach to MOA assignment
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
for six broad and seven specific MOAs. This computational chemistry-based approach calculated up to 700 structural and property descriptors for each chemical and developed prediction models based on multi-linear regression models and decision tree algorithms. As part of that work, a dataset of 924 acute aquatic toxicity MOA assignments were developed using a combination of high confidence approaches that included international consensus classifications, quantitative structure activity relationship (QSAR) predictions, and weight of evidence professional judgment based on an assessment of structure and literature information. The objective of the current study was to develop a comprehensive standardized dataset of aquatic toxicity values and MOA assignments. The current research expands the Martin et al. (2013) dataset to over 1200 chemicals including metals and organometallics, increases the number of specific MOA categories, and refines some of the pesticide neurotoxicity MOA categories to be less general and more mechanism-based. The categorization scheme was comprised of six broad and 31 specific MOA categories that encompassed the major acute toxicity MOAs reported in aquatic invertebrates and fish. Chemicals in the MOA dataset were then linked to a dataset of high confidence acute aquatic toxicity values developed for this study to create a database of 674 chemicals with species-specific acute toxicity values. The MOA-aquatic toxicity database (MOAtox) development focused on the four most commonly reported aquatic test species with toxicity endpoint data in the ECOTOX database (USEPA, 2014): rainbow trout, the cladoceran (Daphnia magna), fathead minnow, and bluegill (Lepomis macrochirus). These species represented 22% of the 300,000 ECOTOX records with toxicity endpoint data (Fairbrother et al., 2014). Additionally, generic fish and aquatic invertebrate acute toxicity values were developed from the geometric mean of all measured toxicity values that met strict quality assurance and standardization requirements. Toxicity values for chemicals with missing species-specific data were estimated using established interspecies correlation models and procedures (WebICE; http://epa.gov/ceampubl/fchain/webice/), with the highest confidence values selected (Raimondo et al., 2010, 2013). The development of this comprehensive standardized dataset is intended to facilitate the development of predictive aquatic toxicity and MOA models. The larger MOA dataset could support MOA-based assessments of chemical mixtures, such as applying a toxic units approach to chemicals with the same specific MOA.
2. Methods 2.1. MOA category and dataset development An aquatic toxicity MOA categorization scheme was developed for a diversity of metals, organometallics, pesticides, and other organics that appear in the ECOTOX database (USEPA, 2014), based on earlier work determining chemical modes of acute toxic action in fish (Konemann 1981; Call et al., 1985; McKim et al., 1987; Russom et al., 1997; USEPA, 2012; Martin et al., 2013). The categorization scheme was comprised of six broad and 31 specific MOA categories that encompassed the major MOAs reported in aquatic invertebrates and fish. Broad MOA categories included narcosis, AChE inhibition, neurotoxicity, iono/osmoregulatory/circulatory impairment, reactivity, and electron transport inhibition (Supplementary information; Table S1). Specific MOAs were developed as subcategories of the broad MOAs based on either chemical structure or known mechanism of action. A total of 1213 chemicals were assigned to a broad and specific MOA category following the general procedures of Martin et al. (2013). This involved using a combination of high confidence assignments including, international consensus classifications (e.g., IRAC, 2014), QSAR predictions
103
(USEPA, 2012), and weight of evidence professional judgment by considering a combination of assessment of structure (e.g., analogous structure; moiety/functional group presence) and literature information on MOA, mechanism of action and toxicity pathways (e.g., Russom et al., 1997; Nendza and Müller 2000; Schüürmann et al., 2003; Spycher et al., 2004). Chemicals with an uncertain MOA assignment were excluded from the database.
2.2. Toxicity dataset development Toxicity dataset development focused on determining acute toxicity values for three fish species (rainbow trout, fathead minnow, and bluegill) and Daphnia for chemicals in the MOA dataset. These four taxa were selected, because they have been the most commonly reported aquatic test species with acute toxicity endpoint data in the ECOTOX database (USEPA, 2014). First, measured aquatic toxicity values were obtained from the US EPA’s ICE Aquatic Toxicity Database (Raimondo et al., 2013a). This database is comprised of 5500 acute EC/LC50 values for 180 species and 1260 chemicals that have been obtained from a diversity of sources (primarily ECOTOX) then subjected to a rigorous quality assurance review and standardization procedure (Raimondo et al., 2010). Each toxicity record included in the dataset met specific requirements for: (1) test chemical identity and purity, (2) test procedures, conditions, reporting, and quality control, and (3) test organism life stage. Additionally, toxicity values for selected compounds were normalized based on water quality parameters, and outliers and central tendency were determined in cases of multiple reported values as documented in Raimondo et al. (2013a). Mean fish and mean aquatic invertebrate toxicity values were also computed from the geometric mean of all standardized values available for each chemical. Similarly, a mean Daphnia genus value was computed from all standardized values for each chemical with available data. Missing species-specific toxicity data were then estimated using established interspecies correlation estimation (ICE) models available in the USEPA Web-ICE tool (http://epa.gov/ceampubl/fchain/webice/; Raimondo et al., 2010, 2013b). ICE models are log-linear least square regressions of chemical sensitivity relationships between a surrogate and predicted taxon. Toxicity values in the Aquatic Toxicity Database for all available surrogate species were used to estimate acute toxicity for those chemicals with missing measured data for rainbow trout, D. magna, fathead minnow, or bluegill. This procedure resulted in multiple estimated values for many chemicals because ICE models were available for multiple surrogate species.
2.3. Determination of best estimated value A systematic process was used to determine a best toxicity estimate for any of the four species with multiple predicted values based on model uncertainty and performance criteria. First, each surrogate species ICE model was evaluated to determine if it met the following statistical criteria: mean square error of less than or equal to 0.94, coefficient of determination greater than 0.6, and the upper 95th percentile confidence limit was within five fold of the predicted toxicity value. If multiple models met all three statistical criteria, these models were evaluated to determine if they met the following additional criteria: taxonomic distance between the surrogate and predicted species was less than or equal to four (same Class), model cross validation success was greater than 85%, and degrees of freedom were greater than 8. If multiple models met all criteria, the best estimated value was computed from the geometric mean of the model predicted values.
104
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
2.4. Ranking and assessment of estimated values Each estimated value in the MOAtox database received a qualitative ranking based on the number of statistical criteria met (Supplementary information; Table S2): level P1 (highest confidence estimate; all three model criteria met), P2 (two met), P3 (one met), and P4 (lowest confidence; no statistical criteria met). Estimated values were not included in the MOAtox database for those chemicals with species-specific measured values. However, toxicity values were also estimated and ranked for those chemicals with measured values to allow an assessment of accuracy. Accuracy was determined from the proportion of species-specific estimated values within two, five, and ten fold of measured values. Correlations between species-specific estimated and measured values were determined from simple log–log linear regressions.
3. Results 3.1. MOA assignments The MOA dataset was comprised of 1213 chemicals assigned to one of six broad acute toxicity MOA categories and one of 31 specific MOA subcategories (Tables 1, S1, and S2 ). The predominant MOAs and chemical classes in the database (>40 chemicals) were AChE inhibition (carbamates and organophosphates), narcosis (nonpolar, polar, and ester compounds), neurotoxicity (pyrethroids and alicyclics), and uncouplers of oxidative phosphorylation (e.g., polychlorinated and nitrophenols and amines) (Tables 1 and S1). Fig. 2. Prediction accuracy of estimated toxicity values by uncertainty level (upper panel), and correlation between measured and predicted toxicity values (lower panel) in fathead minnows.
3.2. Toxicity values Of the 1213 chemicals in the MOA dataset, 614 had measured values in fish and 424 had measured values in invertebrates determined as the geomeans of standardized acute toxicity data (Table 2). There were 671 chemicals in the MOA dataset that had either a measured or estimated taxa-specific acute toxicity value for Daphnia, bluegill, fathead minnow, or rainbow trout (Table 2). Estimated values for the three fish species were primarily comprised of level 1 (highest confidence; >85%) values, whereas, in D. magna the majority of estimated values were level 3 (62%) (Table 2 and Figs. 1–4). Over 60% of the estimated values in the three fish species were within two fold of standardized measured values, and 87% were within five fold (Table 2 and Figs. 1–4). Prediction accuracy was lower for D. magna, with only 61% of estimated values within five fold of measured values (Table 2 and Fig. 4). In the three fish species, estimated values were highly correlated with measured values (R2 > 0.87; Figs. 1–4), whereas, there was greater variance in the relationship for D. magna (R2 = 0.69; Fig. 4). 4. Discussion
Fig. 1. Prediction accuracy of estimated toxicity values by uncertainty level (upper panel), and correlation between measured and predicted toxicity values (lower panel) in bluegill.
In aquatic toxicology, MOA-based toxicity prediction has primarily focused on organic chemicals acting through polar and nonpolar narcosis. Although compounds causing narcosis are structurally diverse and represent the largest class of organic chemicals produced globally, the narcosis MOA does not encompass more toxic compounds, such as pesticides, metals, and other chemicals acting through specific mechanisms and toxicity pathways. The current study builds upon the earlier work of Konemann (1981), Russom et al. (1997), Verhaar et al. (2000), and others to expand the chemical coverage and MOA categories to include
M.G. Barron et al. / Aquatic Toxicology 161 (2015) 102–107
105
Table 1 Description of mode of action (MOA) categories used in the MOAtox database and number of chemicals within each MOA.a Broad MOA (n)b
AChE inhibition (285)
Iono/osmoregulatory/ circulatory impairment (55)
Narcosis (465)c
Neurotoxicity (200)
Reactivity (111)
Electron transport inhibition (97)
a b c
Specific MOA
Chemical class
Carbamate
Carbamate insecticides
Organophosphate
Organophosphorus insecticides Heavy metals, transition metals
Structural features
Example chemicals
#Chemicals in Databasea MOA
MOAtox
74
19
211
58
18
15
Oxime and methyl carbamate Organophosphates, organothiophosphates Divalent metals
Aldicarb, carbaryl, methiocarb Chlorpyrifos, malathion, terbufos Copper, zinc
Inorganics Chemical specific, nitrites
Chemical specific Chemical specific
6 6
4 1
Anticoagulation Ester
Coumarin rodenticides Mono or diesters,
Coumarin moiety Diverse structures, non-reactive
25 48
7 41
Non-polar
Neutral organics
Moderate polarity
347
283
Polar
Polar organics
Single nitro or
