Mode of Action (MOA) Assignment Classifications ... - ACS Publications

Subscriber access provided by University of Florida | Smathers Libraries

Article

Mode of Action (MOA) Assignment Classifications for Ecotoxicology: An Evaluation of Approaches Aude Kienzler, Mace G. Barron, Scott E Belanger, Amy Beasley, and Michelle R Embry Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b02337 • Publication Date (Web): 31 Jul 2017 Downloaded from http://pubs.acs.org on August 6, 2017

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 30

Environmental Science & Technology

1

TITLE: Mode of Action (MOA) Assignment Classifications for Ecotoxicology: An Evaluation of

2

Approaches

AUTHORS: A. Kienzlera, M. G. Barronb, S. E. Belangerc, A. Beasleyd, M.R. Embrye*

a

Scientific officer, Joint Research Centre, Directorate F – Health, Consumers and Reference

Materials, F.3 Chemicals Safety & Alternative Methods, TP 126, Via E. Fermi, 2749, I-21027 Ispra, Italy b

Research Toxicologist, United States Environmental Protection Agency, 1 Sabine Island Dr,

Gulf Breeze, FL 32561, USA c

Research Fellow, The Procter & Gamble Company, Mason Business Center, 8700 S Mason-

Montgomery Road, Mason, OH 45040, USA d

Product Sustainability Consultant, TERC Toxicology and Environmental Research and

Consulting, The Dow Chemical Company. 1803 Building, Midland, MI 48674, USA e*

Associate Director, Environmental Science, International Life Sciences Institute Health and Environmental Sciences Institute (HESI). 1156 15th Street, NW, Suite 200, Washington, DC, 20005, USA

*Corresponding author: HESI, 1156 15th Street, NW, Washington, DC 20005, USA; +1 202 659 3306 x183 (P); +1 202 659 3617 (F); [email protected]

1 ACS Paragon Plus Environment


Page 2 of 30

3

ABSTRACT

4

The mode of toxic action (MOA) is recognized as a key determinant of chemical toxicity and as

5

an alternative to chemical class-based predictive toxicity modeling. However, MOA classification

6

has never been standardized in ecotoxicology, and a comprehensive comparison of

7

classification tools and approaches has never been reported. Here we critically evaluate three

8

MOA classification methodologies using an aquatic toxicity dataset of 3488 chemicals, compare

9

the approaches, and assess utility and limitations in screening and early tier assessments. The

10

comparisons focused on three commonly used tools: Verhaar prediction of toxicity MOA, the

11

U.S. Environmental Protection Agency (EPA) ASsessment Tool for Evaluating Risk (ASTER)

12

QSAR (quantitative structure activity relationship) application, and the EPA Mode of Action and

13

Toxicity (MOATox) database. Of the 3448 MOAs predicted using the Verhaar scheme, 1165

14

were classified by ASTER and 802 were available in MOAtox. Of the subset of 432 chemicals

15

with MOA assignments for each of the three schemes, 42% had complete concordance in MOA

16

classification, and there was no agreement for 7% of the chemicals. The research shows

17

the potential for large differences in MOA classification between the five broad groups of the

18

Verhaar scheme and the more mechanism-based assignments of ASTER and MOAtox.

19

Harmonization of classification schemes is needed to use MOA classification in chemical hazard

20

and risk assessment more broadly.


Page 3 of 30

21


TOC ART

22



Page 4 of 30

23

INTRODUCTION

24

There has been a shift in the risk assessment paradigm towards a more mechanistically based

25

understanding of the mechanism and mode of action of chemicals.1 Improved understanding of

26

toxicological mechanisms would reduce reliance on traditional animal testing, allow chemical

27

grouping to increase assessment efficiency, and improve toxicity extrapolations.2,3 These

28

outcomes would increase the potential use of animal alternative methods, which is a long-term

29

policy goal of most chemical management programs.4,5.

30

Various structure-based classification schemes have been developed to categorize chemicals

31

based on the mode of toxic action (MOA). Thousands of chemicals are in commerce and most

32

have little available information other than structure. Understanding how the available MOA

33

schemes were devised, what information they are based on, and the limitations of each

34

approach are critical. The various MOA classification schemes provide information on a key

35

determinant of chemical toxicity and as an alternative to chemical class-based predictive toxicity

36

modeling. Direct regulatory application of these MOA classifications has been seen most

37

directly in chemical classification and in the use of Quantitative Structure-Activity Relationship

38

(QSAR) models in risk assessment. 6 MOA also plays a role in mixture risk assessment, where

39

both the grouping of chemicals and the choice of the concentration addition or independent

40

action hypothesis can be based on knowledge of the MOA. 7

41

MOA is an operational term that has been loosely defined in both human health and

42

ecotoxicology as a functional change at the cellular level, in contrast to the mechanism of action

43

or molecular initiating event. The definition of MOA is critical in the context of Adverse Outcome

44

Pathways (AOPs). Generally, MOA describes key events at various levels of biological

45

organization, starting with cellular interaction and leading to functional and/or anatomical

46

changes.8 MOA is important in classifying chemicals because it represents an intermediate level


Page 5 of 30


47

of complexity between molecular mechanisms and physiological or organismal outcomes and

48

provides an organizing scheme for chemical classification.9 MOA has also been used to classify

49

pesticides by the interaction with the receptor of the target species10 and to describe the

50

toxicology of human cancer and non-cancer agents by key cytological and biochemical

51

events.11,12 The definitions of MOA used are unique to each of the classification schemes

52

investigated: Verhaar used five broad categories based on general toxicological responses,13

53

whereas the U.S. Environmental Protection Agency (EPA) ASsessment Tool for Evaluating Risk

54

(ASTER) QSAR (quantitative structure activity relationship) application14 and the EPA Mode of

55

Action and Toxicity (MOAtox) database15 have a high degree of specificity based on fish

56

behavioral responses or weight of evidence classification, respectively. It is also important to

57

note that both within and between classification schemes, different levels of biological

58

organization may be represented (e.g., at the molecular, cellular, and organism-level).

59

Research to develop screening and early tier methods to more easily classify or assess

60

chemicals using approaches such as the ecological threshold of toxicological concern (eco-

61

TTC)16 and chemical activity17 is ongoing. Evaluation of these approaches and determination of

62

their domain of applicability is partly dependent on the MOA classification used to group

63

chemicals.

64

The objectives of this study were to critically evaluate available MOA classification

65

methodologies using a set of unique chemicals from a large aquatic toxicity dataset, compare

66

the various approaches, and evaluate their utility and limitations in screening early tier

67

assessments.

68

OVERVIEW OF MOA SCHEMES

69

The comparison of classification approaches focused on three currently available schemes: the

70

Verhaar scheme for prediction of toxicity MOA,13 the EPA ASTER QSAR application,18 and the 5 ACS Paragon Plus Environment


Page 6 of 30

71

EPA MOAtox database15 (Table S1). The Verhaar scheme classifies organic compounds into

72

one of four categories: inert chemicals (Class 1), less inert chemicals (Class 2), reactive

73

chemicals (Class 3), and chemicals acting by a specific mechanism (Class 4). Chemicals in

74

Class 1 exhibit non-polar narcosis or baseline toxicity and can only be predicted if they have log

75

octanol:water partition coefficient (Kow) values between 0 and 6 (e.g., benzenes).

76

Chemicals in Class 2 are more toxic and cause polar narcosis, and typically possess hydrogen

77

bond donor acidity (e.g. phenols & anilines).13 Chemicals in Class 3 demonstrate enhanced

78

toxicity as compared to baseline toxicity that react non-specifically with biomolecules (e.g.

79

epoxides) or are metabolized into more toxic species (e.g., nitriles).13 Chemicals in Class 4

80

cause toxicity through a specific mechanism such as acetylcholinesterase (AChE) inhibition by

81

carbamate insecticides. The assignment of a chemical to a class is based on a decision tree

82

that utilizes the presence or absence of certain chemical structures and moieties. The Verhaar

83

classification scheme has been further modified by changing the order of some classification

84

rules in the decision tree and by adding a fifth class to which chemicals are assigned if the

85

chemical could not be placed in Class 1–4.19

86

Verhaar and modified Verhaar schemes have been implemented into online tools in the public

87

domain such as the Organization for Economic Cooperation and Development (OECD) QSAR

88

Toolbox (https://www.qsartoolbox.org/home) and Toxtree (https://eurl-

89

ecvam.jrc.ec.europa.eu/laboratories-research/predictive_toxicology/qsar_tools/toxtree).20 The

90

OECD QSAR Toolbox is a software application used in the grouping of chemicals and in filling

91

(eco)toxicity data gaps for assessing the hazards of chemicals. Chemical profiling is based on

92

the Chemical Abstracts Service numbers (CAS) and/or SMILES (Simplified Molecular Line Entry

93

System). Toxtree, an open-source application, groups chemicals into categories using several

94

schemes based on SMILES notation and decision trees. A post-processing filter encoded as a

95

KNIME (Konstanz Information Miner) workflow adds amendments to the classification rules21 to 6 ACS Paragon Plus Environment

Page 7 of 30


96

improve the Toxtree classification. The post-filter addresses three types of misclassification:

97

Class 5 of some aliphatic alcohols, some halogenated compounds that should both be assigned

98

to Class 1 (rule 1.5.2 and rule 1.7.1, respectively, of the Verhaar classification scheme), and

99

non- or weakly acidic phenols (rule 2.1). Fewer compounds are misclassified as outside of the

100

model domain, further improving the prediction accuracy of the scheme.

101

ASTER is a rule-based expert system that screens a chemical structure for structural fragments

102

to determine MOA.18 The tool uses acute toxicity of over 600 chemicals in fathead minnows.

103

Non-polar narcosis is the default MOA if no MOA specific structural fragments are identified. If

104

fragments satisfy requirements for more than one MOA, then the QSAR for the most potent

105

toxicity prediction is selected.18 ASTER is based on the MOA categories in Russom et al.22 and

106

contains information on over 50,000 molecular structures. ASTER was primarily developed with

107

neutral industrial organic chemicals acting through narcosis and has a more limited

108

representation of pesticides and other chemicals with more specific modes of action. The tool is

109

not publicly available.

110

MOAtox is comprised of a database of MOA assignments for 1208 chemicals including metals,

111

organometallics, pesticides, and other organic compounds.15 The categorization scheme was

112

based on earlier work determining chemical modes of acute toxic action in fish and

113

encompasses six broad and 31 specific MOAs.23 The resulting dataset used a combination of

114

high-confidence MOA assignments, including biological responses in fish acute toxicity

115

assays,22 pesticide classifications schemes (e.g., Insecticide Resistance Action Committee

116

[IRAC]), QSAR predictions (e.g., ASTER), and weight of evidence professional judgement

117

incorporating an assessment of chemical structure (e.g., analogous structure; moiety/functional

118

group presence) and available information on MOA, mechanism of action, and toxicity

119

pathways. Chemicals with an uncertain MOA assignment and MOAs specific to invertebrates



120

were excluded. Specific MOAs were developed as subcategories of the broad MOAs based on

121

either chemical structure or known mechanism of action.

Page 8 of 30

122 123

MATERIALS AND METHODS

124

Figure S1 provides an overview of the materials and methods used in this analysis.

125

Database Construction and Curation

126

A master dataset was constructed using pre-determined inclusion criteria (Figure S2) following

127

the Stepwise Information-Filtering Tool (SIFT) of Beasley et al.24 An initial dataset of

128

approximately 200,000 studies and chemical property information was compiled, including

129

MOA classification. Approximately 110,000 toxicity data records meeting inclusion criteria was

130

prepared to inform the MOA assignment objectives.

131

Records in the final dataset were subjected to additional curation steps:

132

•

Harmonization of data (e.g., taxonomic name, duration, test statistic units)

133

•

Spot verification of source information, durations, test organism

134

•

Determination of acute/chronic test type and toxicity endpoint

135

CAS numbers were extracted from the working dataset to inform the MOA assignment

136

investigation. A total of 5638 unique CAS were subjected to further reconciliation, including

137

identification of duplicates and mismatches.

138

The CAS list was processed with OECD QSAR Toolbox 3.4 to obtain the related SMILES code,

139

the EPA Ecological Structure Activity Relationships (ECOSAR)25 chemical classification, and the

140

aquatic MOA according to the modified Verhaar classification. The SMILES are essential as the


Page 9 of 30


141

two-dimensional representation of the molecule processed by the MOA assignment tools. The

142

CAS list was also employed to obtain the ASTER and the MOATox classifications.

143

257 CAS were not recognized (no match, no name, and no SMILES were retrieved in the

144

various databases) and therefore could not be processed. 151 CAS were linked to a chemical

145

name without a corresponding SMILES and therefore resulted as “blank” or “non-applicable”

146

regarding the MOA assignment. For some CAS, several SMILES matched, which was either

147

linked to the presence of a salt form of the chemical or to an error in the SMILES. In some rare

148

cases, the same SMILES linked to two different CAS, which was mainly related to different

149

enantiomeric forms of the chemical.

150

Once the CAS list was scrubbed to remove incorrect or unrecognized CAS, 3448 of the original

151

5638 chemicals remained (61%) for which there were CAS matched to a name and a correct

152

SMILES notation.

153

MOA Assignment

154

Two different tools (OECD QSAR Toolbox 3.4 and Toxtree 2.6.620) were used to compare the

155

MOA assignment results for the modified Verhaar scheme, as Bhatia et al.26 showed that

156

discrepancies might exist between the tools. The CAS was used as an input in the OECD

157

QSAR Toolbox, which also retrieves the name and SMILES for each chemical. The SMILES

158

were then processed with Toxtree 2.6.6, to obtain the Verhaar classification. The results

159

obtained with the Toxtree software were then processed through a KNIME post-filter workflow,21

160

which aims at lowering the number of compounds being classified as outside of the model

161

domain and further improving the prediction accuracy of the scheme.

162

To obtain ASTER classifications, the SMILES corresponding to the listed CAS were first

163

processed through the proprietary ASTER tool. If SMILES was not present or returned an error,



Page 10 of 30

164

the corresponding CAS was submitted to maximize the number of CAS given an ASTER

165

assignment. When the SMILES represented a structure described as “impossible,” ASTER

166

returned an error message describing the structure discrepancies and no MOA was assigned.

167

When the SMILES represented a mixture, no MOA was assigned, as mixtures are outside the

168

ASTER applicability domain. When no match was found from the SMILES or CAS, no MOA

169

assignment was returned.

170

MOATox assignments were obtained from previously published tables,15 rather than from model

171

predictions. The individual CAS numbers of all chemicals in the database were cross-referenced

172

to the curated CAS-linked MOA classifications in Table S2.15

173

Along with the MOA, the ECOSAR classes of the chemicals also were extracted to characterize

174

the dataset and to identify the chemicals and classes present in the dataset. The ECOSAR

175

classification distinguishes up to 111 classes of organic chemicals, including neutral organics,

176

organic chemicals with excess toxicity, and chemicals that are determined to be surfactants.

177

Analysis

178

The Verhaar MOAs obtained with the OECD QSAR Toolbox and the Toxtree software for each

179

chemical were compared to identify potential discrepancies (n = 3448). The result of the use of

180

the KNIME post-filter on the classification given by Toxtree was also evaluated for

181

discrepancies.

182

Mapping the three classification schemes shows how they correspond and where they overlap

183

(Table 1). The subset of chemicals for which at least three classifications were obtained using

184

Verhaar, OECD QSAR Toolbox, ASTER, and the MOATox classifications (n = 432) was further

185

analyzed to compare the MOA assigned according to the various schemes, specifically where

186

those classification assignments overlap. The complete dataset is given in Table S2.

187 10 188

ACS Paragon Plus Environment

Page 11 of 30

189


Table 1. Comparison of the Three MOA Classification Schemes Verhaar (Modified) Class 1 (narcosis or baseline toxicity) Class 2 (less inert compounds)

Class 3 (unspecific reactivity)

ASTER MOA

MOATox Broad

Non-polar narcosis

Non-polar Narcosis

Polar narcosis Ester narcosis Diester toxicity Reactive Chloro-diester-based reactivity Carbonyl (C=O)-based reactivity Carbonyl reactivity Alkylation/arylation-based reactivity Acylation-based reactivity Sulfhydryl (-S-H)-based reactivity Reactive dinitroaromatic group Nitroso-based reactivity Quinoline reactivity Acetamidophenol reactivity Reactive diketones Acrylate toxicity N-halogenated acetophenone inhibition Hydrazine-based reactivity

Isocyanate (-N=C=O)-based reactivity Pyridinium compounds

Polar Ester

Reactivity Reactivity

Carbonyl Carbonyl

Reactivity

Alkylation

Reactivity

Di/trinitroaromatic

Reactivity

Acrylate

Reactivity Reactivity Reactivity Reactivity

Hydrazine Chromate Phosphide Other

Reactivity

Neurotoxicant: DDT-type Neurotoxicant: pyrethroid Neurotoxicant: cyclodiene-type Neurotoxicity Neurotoxicant: strychnine Neurotoxicant: nicotine Class 4 (compounds and groups Organophosphate-mediated AChE of compounds acting by a inhibition AChE inhibition Carbamate-mediated AChE specific mechanism) inhibition Uncoupler of oxidative phosphorylation Electron transport inhibition Respiratory blocker: azides and cyanides Iono/osmoregulatory/ circulatory impairment Class 5

MOATox Specific

Unknown mode of action

Unknown

Cyanate/nitrile (OCN-; CΞN) Sodium channel blocking Diphenyl sodium channel modulation Pyrethroid sodium channel modulation Alicyclic GABA antagonism Pyrazole GABA antagonism GABA agonism Strychnine Nicotinic acetylcholine receptor agonism Other Organophosphate Carbamate Uncoupling oxidative phosphorylation Arsenical respiratory inhibition Oxidative phosphorylation inhibition Metallic iono/osmoregulatory impairment Methemoglobinemia Anticoagulation Other osmoregulatory Unknown



Page 12 of 30

190

AChE, acetylcholinesterase; ASTER, Assessment Tool for Evaluating Risk; MOA, mode of toxic action; MOATox,

191

Mode of Action and Toxicity database.

192

RESULTS

193

Chemical Classes in Dataset

194

The dataset of 3448 chemicals was evaluated using ECOSAR to provide an overview of the

195

chemical class coverage. For chemicals where multiple classifications were generated by

196

ECOSAR, the first reported was used. These 89 categories were further collapsed into 46 more

197

general categories. Collapsed ECOSAR classifications that each represent >4% of the total

198

dataset are shown in Figure 1. Figure 1 includes the seven major categories in which the most

199

chemicals were classified (e.g., those individual categories in which ≥4% of the total number of

200

chemicals were classified). 24% of chemicals in the dataset were classified as neutral organics,

201

with the second most common category being esters (13%). Importantly, approximately 6% of

202

chemicals were classified as “should not be profiled” and therefore were not assigned a

203

chemical category. The “Other” category (24%) represents other non-major chemical categories

204

that are each represented by

Mode of Action (MOA) Assignment Classifications ... - ACS Publications

Recommend Documents