Toxicology and Applied Pharmacology 254 (2011) 198–205

Contents lists available at ScienceDirect

Toxicology and Applied Pharmacology

j ourna l homepage: www.e lsev ie r.com/ locate /ytaap

Assessing the toxic effects of ethylene glycol ethers using Quantitative StructureToxicity Relationship models☆

Patricia Ruiz a,⁎, Moiz Mumtaz a, Vijay Gombar b

a Computational Toxicology Methods Development Laboratory, Division of Toxicology and Environmental Medicine, Agency for Toxic Substances and Disease Registry, Atlanta, GA, USAb Lilly Research Laboratories, Drug Disposition and Toxicology, Lilly Corporate Center, Indianapolis, IN, USA

☆ The findings and conclusions in this paper are thosnecessarily represent the official position of their Agentrade names is not an endorsement of any commercial⁎ Corresponding author. Division of Toxicology a

Computational Toxicology and Methods DevelopmSubstances and Disease Registry. 1600 Clifton Road, MSFax: +1 770 488 3470.

E-mail address: pruiz@cdc.gov (P. Ruiz).

0041-008X/$ – see front matter. Published by Elsevierdoi:10.1016/j.taap.2010.10.024

a b s t r a c t

a r t i c l e i n f o

Article history:Received 6 November 2009Revised 12 May 2010Accepted 24 October 2010Available online 27 October 2010

Keywords:Quantitative structure toxicity relationshipEthylene glycol mono-n-alkyl ethersDevelopmental toxicityQSARQSTR

Experimental determination of toxicity profiles consumes a great deal of time, money, and other resources.Consequently, businesses, societies, and regulators strive for reliable alternatives such as QuantitativeStructure Toxicity Relationship (QSTR) models to fill gaps in toxicity profiles of compounds of concern tohuman health. The use of glycol ethers and their health effects have recently attracted the attention ofinternational organizations such as the World Health Organization (WHO). The board members of ConciseInternational Chemical Assessment Documents (CICAD) recently identified inadequate testing as well as gapsin toxicity profiles of ethylene glycol mono-n-alkyl ethers (EGEs). The CICAD board requested the ATSDRComputational Toxicology and Methods Development Laboratory to conduct QSTR assessments of certainspecific toxicity endpoints for these chemicals. In order to evaluate the potential health effects of EGEs, CICADproposed a critical QSTR analysis of the mutagenicity, carcinogenicity, and developmental effects of EGEs andother selected chemicals.We report here results of the application of QSTRs to assess rodent carcinogenicity, mutagenicity, anddevelopmental toxicity of four EGEs: 2-methoxyethanol, 2-ethoxyethanol, 2-propoxyethanol, and 2-butoxyethanol and their metabolites. Neither mutagenicity nor carcinogenicity is indicated for the parentcompounds, but these compounds are predicted to be developmental toxicants. The predicted toxicity effectswere subjected to reverse QSTR (rQSTR) analysis to identify structural attributes that may be the main driversof the developmental toxicity potential of these compounds.

e of the author(s) and do notcy or organization. Mention ofproduct.nd Environmental Medicine,ent Unit, Agency for Toxic-F62, Atlanta, GA 30333, USA.

Inc.

Published by Elsevier Inc.

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199Ames mutagenicity model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201Carcinogenicity models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201Developmental toxicity models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201Mutagenicity of alkoxyethanols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201Carcinogenicity of alkoxyethanols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202Developmental toxicity potential of alkoxyethanols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204Conflict of interest statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

Introduction

Ethylene glycol mono-n-alkyl ethers (EGEs) or alkoxyethanols area family of ethylene glycol ethers. The family includes 2-methox-yethanol (EGME), 2-ethoxyethanol (EGEE), 2-propoxyethanol (EGPE),and 2-butoxyethanol (EGBE) (Fig. 1). These chemicals are widely used

OOH

OOH

OOH

OOH

2-butoxyethanol

2-propoxyethanol

2-ethoxyethanol

2-methoxyethanol

Fig. 1. Chemical structures of the parent alkoxyethanols.

199P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

as organic solvents and thinners in resins, paints, and dyes (Boatmanand Knaak, 2001), but they also cause a variety of cancer andnoncancer health effects in humans and laboratory animals (Browningand Curry, 1994; Nagano et al., 1984; Starek and Szabla, 2008;Wang etal., 2004). They are readily absorbed following dermal, inhalation, ororal exposures, and are rapidly distributed throughout the body. Theyare known tometabolize principally via a pathway involving oxidationby alcohol dehydrogenase enzymes to an intermediate alkoxyacetal-dehyde, followed by rapid conversion by aldehyde dehydrogenases tothe corresponding alkoxyacetic acid (Fig. 2). The alkoxyacetic acidsmay be further conjugated with glycine or O-dealkylated andfurther metabolized to yield carbon dioxide. A secondary pathway formetabolism of the alkoxyethanols involves microsomal P-450 mixed-function oxidases with dealkylation to produce ethylene glycol (Ma etal., 1993). Direct conjugation with sulphate or glucuronic acid mayalso occur. The major hematological, reproductive, and developmen-tal toxic effects are caused by the metabolites—mainly the alkox-yacetic acids. The principal route of elimination for the alkoxyethanolsis via urine as the alkoxyacetic acid metabolite (Scott et al., 1989).

Although this class of chemicals has been well studied, severalgaps remain in their toxicological profiles, due primarily to theextensive amount of time and resources required for experimentaltoxicity measurements. To fill information gaps in toxicity databasesfor chemicals like EGEs with significant data gaps, academia andpharmaceutical, agrochemical, food, and other industries have turnedto computer-assisted prediction tools. Such tools play a complemen-tary role in assessing toxicity of chemicals for which no data areavailable or for which toxicological testing is impractical. In additionto minimizing animal testing and saving time and resources,computer-assisted prediction models also support screening andsubsequent prioritization of compounds for further toxicologicaltesting. Various advisory and regulatory government agencies have

also used such models as decision support tools (Demchuk et al.,2008; El-Masri et al., 2002; Mumtaz, et al., 1995).

Most computational toxicity prediction systems are either Struc-ture Activity Relationship (SAR) or Quantitative Structure ActivityRelationship (QSAR) models (Gombar et al., 1995a,b; Greene, 2002;Richard et al., 2008). An SAR model, or an expert system, establishesqualitative association between a chemical's substructure, called alert,and its toxicity potential. In a well-tuned expert system, toxicitypredictions about new chemical entities are based on whether suchidentified structural alerts are present in the newmolecule of interest.Because expert systems are based on qualitative SARs, they usually donot make a quantitative prediction of toxic effect; predictions areexpressed as toxic/non-toxic. In many cases, such expert systems canalso providemechanistic information—as long as it was collected fromthe published literature during knowledge generation (Richard,1998). An experienced user of such systems may then assess therelevance of a toxicological alert.

A Quantitative Structure Toxicity Relationship (QSTR), on the otherhand, is a mathematical relationship between the chemical's quantita-tive molecular descriptors and its toxicological endpoint (Gombar,1997). The backbone of a predictive QSTR are molecular descriptorsderived from atomic ormolecular properties that encode physicochem-ical (e.g., octanol–water partition coefficient), topological (e.g., electro-topological states), and surface properties (e.g., polarity) of molecules(Hall et al., 1991; Kier, 1986). These descriptors are then correlatedwitha toxicological response of interest through a suitable statisticalapproach such as linear multiple regression, discriminant analysis,recursive partitioning, artificial neural networks, etc. (Gombar andEnslein, 1990; Gombar and Jain, 1987). Unlike an expert system, a QSTRprovides discrete quantitative predicted values given just the values ofthe descriptors finally selected in the QSTR model.

Many commercial and open source software packages such asTOPKAT (Accelrys, 2004), CASE and MultiCASE, (Klopman, 1984;Klopman and Rosenkranz, 1991), DEREK(LHASA, Ltd.; Sanderson andEarnshaw, 1991), OncoLogic™ (Woo and Lai, 2005), Lazar (Helma,2006), ToxTree (Benigni et al., 2008), etc. have embodied expertsystems or QSAR models to predict human health effects and relatedtoxicities (Enslein et al., 1994; Enslein, 1997; Klopman, 1984;Sanderson and Earnshaw, 1991). Since these packages allow predic-tion of toxicity solely from chemical structure, chemical andpharmaceutical industries and regulatory agencies, including theDanish EPA, U.S. EPA, U.S. FDA, and the U.K. Health and SafetyExecutive, have employed these tools to rapidly assess potentialtoxicity given just the 2D drawing of a chemical structure.

For the present work we decided to use TOPKAT not only becauseit offers modules to predict mutagenicity, carcinogenicity anddevelopmental toxicity but also because of its following uniquefeatures: use of information-rich E-states as descriptors, cleaned andharmonized model training sets, several prediction diagnostics toassess reliability of a prediction, and ability to conduct reverse QSAR(rQSAR) for quantifying the contribution of a desired molecularmoiety to the predicted value (Enslein, 1993; Gombar, 1997; Mumtazet al., 1995; Zeiger et al., 1996). A detailed description of the featuresand functionality leading to selection of TOPKAT for the present workand its toxicity modules employed in this work are given in thefollowing section.

Methods

TOPKAT is a commercially available software package with QSTRmodels of a variety of toxicity endpoints, including rodent carcino-genicity, Ames mutagenicity, and developmental effects. Each ofTOPKAT's QSTR model can be represented as:

T = ∑n

i=1βiXi + C

OH2 H2C C OH

H3CH3C

O CH2

CH2

O C CH3

O

H3C O CH2

CH

O

H3C O CH2

C OH

H3C O CH2

C NH

O

CH2

OCOOH

HO CH2

CH2

OHCH2

O

H3C O CH2

CH2

O Gluc

Cytochrome P-450

Alcohol dehydrogenase

Aldehyde dehydrogenase

Acyl transferase

DealkylaseCarboligase

Glucuronyltransferase

Methoxyethyl acetateMethoxyethanol

Ethylene glycolFormaldehyde

Methoxyacetaldehyde

Methoxyacetic acid

Methoxyethanolglucuronide

Methoxyacetic acidGlycine conjugate

H3C O CH2

CH2

O SO3H

MethoxyethanolSulfate

Sulfotransferase

CO2

Fig. 2. Metabolism pathway of 2-methoxyethanol.

200 P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

where T is the predicted toxicity score, C is a constant, Χi is the ith

structure descriptor value, and βi is the coefficient associated withthe descriptor. The product βi Χi represents the contribution of the ith

descriptor to the predicted toxicity value, which is the sum ofdescriptor contributions and the constant C. For a classificationmodel making a dichotomous prediction, this sum is transformedinto a probability value between 0 and 1 for the positive class (e.g.,mutagen, carcinogen, etc.). Compounds predicted with probabilityvalues between 0 and 0.3 are considered negative or of lowprobability of being positive, while those between 0.3 and 0.7 areconsidered indeterminate (i.e., no unequivocal prediction). Apositive prediction is indicated when the probability is greaterthan 0.7. The details of the assessment process using the TOPKATsoftware have been previously published (Moudgal et al., 2000; Ruizet al., 2008).

Besides providing these probability values as a guide to confidencein a prediction, TOPKAT offers a number of other unique predictiondiagnostics to assess reliability of a prediction. First, the querystructure is checked for any new chemistry that has not beenrepresented in the training set of the (sub)model. After coverage ofthe query structure is confirmed, TOPKAT evaluates if the values of

model descriptors for the query structure are in the ranges ofdescriptor values defined by the training set compounds. Thisunivariate check confirms whether the query structure is inside thetraining data domain (TDD). After having performed the coverage andunivariate analyses, TOPKAT examines whether the query structure isin themodel's Optimum Prediction Space (OPS), which is smaller thanits TDD, especially when model descriptors are not completelyorthogonal (Gombar, 1997; Gombar and Enslein, 1996). It is insideand in the vicinity of the OPS where a model is applicable. Predictionsoutside the OPS are not supported by the model and generally exhibitlarge deviation from experimental values.

The coverage, univariate, and multivariate analyses combineddefine TOPKAT's validation criterion. For a prediction to be reliable thevalidation criterionmust be satisfied.When a query structure fails anyof the three tests, it is declared to fail the validation criterionindicating that the prediction may not be reliable. TOPKAT offersanother feature based on “QSTR similarity” analysis to enhanceconfidence in a prediction (Gombar, 1997; Moudgal et al., 2000;Venkatapathy et al., 2009). Unlike other similarity measures, QSTRsimilarity, expressed as the Euclidian distance computed from thevalues of model descriptors, is property sensitive because it reflects

Table 2Leave-one-out cross-validation performance of the TOPKAT NTP carcinogenicitymodule.

Species, sex Sensitivity (%) Specificity (%) Indeterminate (%)

Rat, male 82 82 11Rat, female 91 93 1Mouse, male 90 94 1Mouse, female 88 87 5

201P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

the similarity of descriptor values between two molecules withrespect to a specific property or endpoint (Accelrys, 2004). In thisimplementation, two molecules A and B can have different QSTRsimilarity distance depending on the property in question. One cansearch the training set of a model to look for the most similarcompounds and assign confidence in a prediction based on thesimilarity distance and concordance between the experimental andpredicted values of these compounds.

Unlike any other computational toxicity package, TOPKAT iscapable of conducting reverse QSTR (rQSTR), thus allowing compu-tation of contribution from any atom or group of atoms to thepredicted value. Termed moiety effects, these quantitative contribu-tions provide guidance in modulation of query structure toward adesired predicted toxicity value.

Given the aforementioned unique features of TOPKAT and giventhe topic of identifying potential drivers of mutagenicity, carcinoge-nicity, and developmental toxicity, if any, of alkoxyethanols, weselected TOPKAT for the present work. The quality of the mutagenic-ity, carcinogenicity, and developmental toxicity models of TOPKAT issummarized below.

Ames mutagenicity model

The Ames mutagenicity module of the TOPKAT package iscomposed of 10 QSAR sub-models developed from 1866 uniformstudies selected after a critical review of open-literature histidinereversion assays using Salmonella typhimurium strains. Each sub-model is a linear discriminant function derived from a specific class ofchemicals and predicts the probability of a submitted chemicalstructure being a mutagen in the histidine reversion assay (Accelrys,2004). The performance of the mutagenicity module in the leave-one-out cross-validation study is shown in Table 1. Since leave-one-out isthe least stringent of the cross-validation techniques, one may getsignificantly lower sensitivity, specificity, and forecast accuracy whenthis model is applied to new test sets.

Carcinogenicity models

TOPKAT also offers several modules for predicting rodentcarcinogenicity. The National Toxicology Program (NTP) RodentCarcinogenicity Module of the TOPKAT package has four QSTRmodels.These linear discriminant classification models are derived fromuniform studies selected after a critical review of technical reports on366 rodent carcinogenicity tests conducted by the National CancerInstitute (NCI) and the NTP using inbred rats and hybrid mice. EachQSTR model relates to a specific sex/species combination: male rat,female rat, male mouse, and female mouse. The models compute theprobability of a submitted chemical structure being a carcinogen. Theperformance of this carcinogenicity module in the leave-one-outcross-validation study is shown in Table 2. Since leave-one-out is theleast stringent of the cross-validation techniques, one may getsignificantly lower sensitivity, specificity, and forecast accuracywhen this model is applied to new test sets.

Table 1Leave-one-out cross-validation accuracy of the TOPKAT mutagenicity.

Expt call Predicted call Statistics

Mutagen Non-mutagen

Indeterminate Total 8 Noprediction

Mutagen 648 14 7 669 96.9% SpecificityNon-mutagen 5 408 1 414 98.6% SensitivityTotal 653 422 1075 98.2% AccuracyForecastaccuracy

99.2% 96.7% 99.3% Applicability

Developmental toxicity models

The Developmental Toxicity Potential module of TOPKAT has threeQSAR models; each developed from a specific class of chemicals(Gombar et al., 1995a,b). These discriminant models were derivedfrom 374 uniform experimental studies selected after a critical reviewof approximately 3000 open literature citations. They compute theprobability that a query structure will exhibit developmental toxicityin the rat when tested up to a dose inducing maternal toxicity. Theperformance of the developmental toxicity module in the leave-one-out cross-validation study is shown in Table 3. Since leave-one-out isthe least stringent of the cross-validation techniques, one may getsignificantly lower sensitivity, specificity, and forecast accuracy whenthis model is applied to new test sets.

Results and discussion

Mutagenicity, carcinogenicity, and developmental toxicity of thealkoxyethanol molecules (see structures in Fig. 1) and theirmetabolites, as predicted by TOPKAT, are summarized, respectively,in Tables 4–6. For each predicted toxicity class, these tables providethe following information: the probability of prediction, confidenceprediction level, whether the query structure is part of the TOPKATtraining set, any external experimental toxicity data. The findings formutagenicity, carcinogenicity, and developmental toxicity are sepa-rately discussed below.

Mutagenicity of alkoxyethanols

Mutagenicity is thedamage caused in a cell that is transmissible fromcell to cell or generation to generation. Specifically, it is the ability of achemical to cause changes in the genetic material in the nucleus of thecells. Because mutagenicity could be the first step to carcinogenicity ofcertain chemicals, it is an important inherent characteristic ofenvironmental and pharmaceutical chemicals. Several alkoxyethanolswere tested (Hoflack et al., 1995; Ma et al., 1993) and consistentlyyielded negative results formutagenicity. Data from 2-methoxyethanol,2-ethoxyethanol, and 2-butoxyethanol show lack of significant geno-toxic activity. Mutagenicity tests with 2-methoxyethanol, however,reveal genetic effects in induced male germ cells. No evidence ofmutations in vitro was seen for the two lower molecular weightalkoxyethanols—2-methoxyethanol and 2-ethoxyethanol. For 2-butox-yethanol results of mutagenicity assays in S. typhimurium wereinconsistent, although assay results in most of the standard strainstested showed no mutagenic activity.

Table 3Leave-one-out cross-validation performance of the TOPKAT developmental toxicitypotential (DTP) module.

Expt call Predicted call Statistics

DTP No DTP Indeterminate Total 9 No prediction

DTP 116 17 5 138 87.2% SensitivityNo DTP 12 107 4 123 89.9% SpecificityTotal 128 124 9 261 88.5% AccuracyForecast accuracy 90.6% 86.3% 3.45% 96.6% Applicability

Table 4Predicted mutagenicity of alkoxyethanols and their metabolites.

Alkoxyethanol Experimentalmutagenicitya

TOPKAT predictedmutagenicity

2-Methoxyethyl acetate na −(M)2-Methoxyethanol − −(H)(DB)2-Methoxyacetaldehyde +/−b −(M)2-Methoxyacetic acid − −(L)2-Ethoxyethyl acetate − −(H)2-Ethoxyethanol − −(H) (DB)2-Ethoxyacetaldehyde − −(L)2-Ethoxyacetic acid − −(L)2-Propoxyethyl acetate na −(H)2-Propoxyethanol na −(H)2-Propoxyacetaldehyde na −(L)2-Propoxyacetic acid na −(L)2-Butoxyethyl acetate na −(H)2-Butoxyethanol − −(H) (DB)2-Butoxyacetaldehyde na −(L)2-Butoxyacetic acid − −(L)

na=not available; + = mutagen; −=non-mutagen.DB=compound in TOPKAT training dataset.H=High confidence prediction; M=Moderate confidence prediction; L=Lowconfidence prediction.Assessment Documents [CICADs]; WHO/IPCS, 2009a).

a Concise International Chemical Assessment Documents (CICADs) (WHO/IPCS,2005, 2009a,b).

b Positive with S. typhimurium strain TA97A, negative with TA98, TA100, TA102(Concise International Chemical).

Table 6Predicted developmental toxicity of alkoxyethanols and their metabolites.

Alkoxyethanols Experimentaldevelopmental toxicity(oral route)a

TOPKAT predictionsdevelopmental toxicity(oral route)

2-Methoxyethyl acetate na +(H)2-Methoxyethanol + +(H) (DB)2-Methoxyacetaldehyde na −(M)2-Methoxyacetic acid na +(M)2-Ethoxyethyl acetate na −(H)(DB)2-Ethoxyethanol + +(H) (DB)2-Ethoxyacetaldehyde na −(M)2-Ethoxyacetic acid na +(H)2-Propoxyethyl acetate na −(H)(DB)2-Propoxyethanol nab +(H)2-Propoxyacetaldehyde na −(M)2-Propoxyacetic acid na +(M)2-Butoxyethyl acetate na −(H)2-Butoxyethanol – −(H) (DB)2-Butoxyacetaldehyde na −(L)2-Butoxyacetic acid na −(H)

+ = potential developmental toxicant; −=negative developmental toxicantprediction; na=not available; DB=compounds in TOPKAT training dataset.H=High confidence prediction; M=Moderate confidence prediction; L=Lowconfidence prediction.

a Concise International Chemical Assessment Documents (CICADs) (WHO/IPCS2005,2009a,b).

b Experimental inhalation data, no developmental toxicity was observed on thisinhalation study (WHO/IPCS, 2009b). Oral study was not available.

202 P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

Mutagenicity of alkoxyethanols and their metabolites as predictedby the TOPKAT program is compared with the experimental data inTable 4. Of these 16 compounds, experimental mutagenicity data areavailable for nine, based on information reported in the ConciseInternational Chemical Assessment Documents (CICADs) (WHO/IPCS,2005, 2009a,b). It can be seen that not only the high confidencepredictions but also the low confidence predictions show highconcordance with the available experimental data. It is acknowledgedthat three compounds are found in the model's training set.

For 2-methoxyacetaldehyde, a metabolite of 2-methoxyethanol,the mutagenic potential predicted by TOPKAT lacks concordancewith the published experimental results. To further investigate thisdisagreement, we looked at the training set data of the TOPKATmutagenicity model. TOPKAT follows the U.S. Environmental Protec-tion Agency (U.S. EPA) Gene-Tox protocol to label a compound“mutagenic”, i.e., it must test positive in one or more of the fivestrains of Salmonella, namely TA100, TA1535, TA1537, TA1538, and

Table 5Predicted carcinogenicity of alkoxyethanols and their metabolites.

Alkoxyethanols Carcinogenicity experimental dataa

Rat, male Rat, female Mouse, male Mou

2-Methoxyethyl acetate na na na na2-Methoxyethanol na na na na2-Methoxyacetaldehyde na na na na2-Methoxyacetic acid na na na na2-Ethoxyethyl acetate na na na na2-Ethoxyethanol na na na na2-Ethoxyacetaldehyde na na na na2-Ethoxyacetic acid na na na na2-Propoxyethyl acetate na na na na2-Propoxyethanol na na na na2-Propoxyacetaldehyde na na na na2-Propoxyacetic acid na na na na2-Butoxyethyl acetate na na na na2-Butoxyethanol IND IND IND IND2-Butoxyacetaldehyde na na na na2-Butoxyacetic acid na na na na

+ = Carcinogen; −=non-carcinogen; na=not available; IND=indeterminate; DB=comH=High confidence prediction; M=Moderate confidence prediction; L=Low confidence

a Concise International Chemical Assessment Documents (CICADs) (WHO/IPCS, 2005, 20

TA98. However, the positive mutagenicity result for 2-methoxyace-taldehyde published by Hoflack et al. (1995) is based on experi-mental testing only in the TA97a strain, which is not part of theGene-Tox protocol. Therefore, the apparent lack of concordancebetween experimental and predicted mutagenic potential for 2-methoxyacetaldehyde is perhaps due to difference of strains ofSalmonella used rather than misprediction by TOPKAT. A model onlypredicts the outcome in the test whose results are used to constructthe predictive model. When examining the model performance, onemust always make sure that the training data and the test data wereproduced in a uniform, consistent manner from assays usingidentical protocol.

Carcinogenicity of alkoxyethanols

Through chronic bioassays conducted by the National ToxicologyProgram (NTP), the carcinogenicity of a large number of chemicals has

Carcinogenicity TOPKAT predictions

se, female Rat, male Rat, female Mouse, male Mouse, female

+(M) +(M) −(H) +(H)+(M) +(M) −(H) IND(H)+(L) +(M) −(H) +(L)+(H) +(L) −(H) IND(H)+(M) −(L) −(H) −(L)−(M) IND(L) −(M) −(H)IND(L) −(L) −(H) −(M)−(H) +(L) −(H) −(M)+(L) −(L) −(H) −(L)−(L) −(L) −(M) −(H)IND(L) −(L) −(H) IND(M)−(M) +(L) −(H) −(M)+(L) −(L) −(H) −(L)−(L) −(L) −(M) −(H)+(L) −(L) −(H) IND(M)−(L) +(L) −(H) −(M)

pounds in TOPKAT training dataset.prediction.09a,b).

Table 7Contributions of E-states and count atoms fragment to the final measure of thedevelopmental toxicity of alkoxyethanols and its derivatives.

Chemicals Count of[CH2CHO]

Shapeindexorder 7

E-state of[aliphatic O]

Predicteddevelopmentaltoxicity

CH3OCH2CH2OH 0 0 5.821 +CH3CH2OCH2CH2OH 0 0 6.020 +CH3CH2CH2OCH2CH2OH 0 0 6.135 +CH3 CH2CH2 CH2OCH2CH2OH 0 −6.481 6.212 −CH3OCH2CHO 12.724 0 6.395 −CH3CH2OCH2CHO 12.724 0 6.607 −CH3CH2CH2OCH2CHO 12.724 0 6.731 −CH3CH2CH2CH2OCH2CHO 12.724 −6.481 6.816 −CH3OCH2COOH 0 0 10.092 +CH3CH2OCH2COOH 0 0 10.365 +CH3CH2CH2OCH2COOH 0 0 10.535 +CH3CH2CH2CH2OCH2COOH 0 −6.481 10.653 −CH3OCH2CH2OCOCH3 0 0 9.014 +CH3CH2OCH2CH2OCOCH3 0 −6.481 9.222 −CH3CH2CH2OCH2CH2OCOCH3 0 −8.641 9.345 −CH3CH2CH2CH2OCH2CH2OCOCH3 0 −9.721 9.430 −

[CH2CHO]=acetaldehyde fragment; [aliphatic O] = aliphatic oxygens.

203P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

been extensively studied. However, most of the alkoxyethanols andtheir metabolites have not been experimentally tested for carcinoge-nicity. Only 2-butoxyethanol has been tested adequately for carcino-genic potential in long-term studies in rats and mice. The bioassay of2-butoxyethanol by NTP (NTP, 2000) provides some evidence ofcarcinogenicity in mice (haemangiosarcomas of the liver andsquamous cell papilloma or carcinomas of the forestomach) andequivocal evidence of carcinogenicity in female rats (marginalincrease in pheochromocytomas of the adrenal gland) (WHO/IPCS,2005).

According to the battery of TOPKAT carcinogenicity models, allparent molecules, except for 2-methoxyethanol, are predicted ashaving no carcinogenic potential (Table 5). The male and femalerat sub-models predicted potential probability of carcinogenicity for2-methoxyethanol and its metabolites. The male rat sub-modelpredicted 2-butoxyacetaldehyde as a potential carcinogen; 2-ethox-yacetaldehyde and 2-propoxyacetaldehyde were predicted as inde-terminate. Similarly, the male rat sub-model predicted thealkoxyethyl acetates as potential carcinogens, while the alkoxyaceticacids were predicted as carcinogenic in the female rat sub-model.

Developmental toxicity potential of alkoxyethanols

Developmental toxicity is any adverse effect on the developingorganism from implantation through prenatal development orpostnatally to the time of sexual maturation. Manifestations ofdevelopmental toxicology include structural malformations, growthretardation, functional impairment, and/or death of the organism.

2-Methoxyethanol has been shown to consistently induce develop-mental effects in studies involving several species of laboratory animalsfollowing exposure by multiple routes, including oral, inhalation, anddermal (WHO/IPCS, 2009a). For example, decreased fetal body weightswere noted in rats exposed to 2-methoxyethanol at doses of 16 mg/kgbody weight per day, with malformations at 31 mg/kg body weight perday or more, while maternal toxicity was only evident at 140 mg/kgbody weight per day. In inhalation studies in rats, developmentaltoxicity occurred following repeated maternal exposure to 2-methox-yethanol concentrations of≥50 ppm (156 mg/m3). The cardiovascularsystem, kidney, skeletal system, and heart were the principal targets of2-methoxyethanol developmental effects.

2-Ethoxyethanol also showed developmental toxicity in theabsence of significant maternal toxicity in rats, mice, and rabbits,with exposures involving oral, inhalation, and dermal routes (WHO/IPCS, 2009b). The cardiovascular and skeletal systems were againmajor targets of developmental toxicity. The available experimentaldevelopmental toxicity studies for 2-propoxyethanol (WHO/IPCS,2009b) and 2-butoxyethanol (WHO/IPCS, 2005) generally suggest alack of developmental toxicity.

TOPKAT analysis revealed that all parent compounds and theiralkoxyacetic acid metabolites, except 2-butoxy ethanol and 2-butoxyacetic acid, were predicted positive for potential to inducedevelopmental effects (Table 6). For all other alkoxyacetaldehydesand alkoxyethyl acetates the TOPKAT models predicted no potentialdevelopmental effects except methoxyethyl acetate. It can be seen(Table 6) that the experimental and predicted developmental toxicityvalues agree for three of the four alkoxyethanols. 2-Propoxyethanolwas predicted to be a potential developmental toxicant by the oralQSTR model. No oral experimental data were available. However, theresults of an inhalation study did not show a potential fordevelopmental toxicity (Krasavage and Katz, 1985). Since the TOPKATmodel was developed from oral developmental toxicity data in rats, acorrelation between inhalation and oral toxicity of this chemical is notexpected. When examining the model performance, one must alwaysmake sure that the training data and the test data were produced in auniform, consistent manner from assays using identical route andspecies.

Our oral QSTR model predictions indicate that alkoxyethanols andalkoxyacetic acids, with the exception of the 2-butoxyethanol and 2-butoxyacetic acid, are likely developmental toxicants. By contrast, thealkoxyacetaldehydes and alkoxyethylacetates metabolites are notexpected to produce developmental effects.

In order to identify descriptors that drive prediction of develop-mental toxicity of a given compound, we examined their standard-ized contributions. It was found that the count of [CH2CHO]fragment, E-state of [aliphatic O] and the molecular shape index #7were significant drivers of developmental toxicity (Table 7). Theresults indicate that the specific count of the [CH2CHO] fragment isassociated with the lack of developmental toxicity. Similarly, thelarge negative contribution of the molecular shape index #7, whichquantifies the shape of a molecule as a function of the number ofatoms and their bonding relationship, indicates its association withnegative developmental toxicity predictions. This implies that largemolecules with extensive branching are likely not to inducedevelopmental toxicity. In contrast, all the negative developmentalchemicals had a nonzero value for [CH2CHO] descriptor, implyingthat the presence of the [CH2CHO] fragment prevents the occurrenceof developmental toxicity, as shown in Table 7. In addition, thepresence of the CH2COOH fragment or the CH2CH2OH fragmentseemed to promote developmental toxicity. It was further observedthat the developmental toxicity of alkoxyethanols decreases as thesize of the alkyl ether group increases. One may speculate that thesmaller alkyl ether side chain (1 to 3 carbons long) could be astructural feature leading to developmental toxicity but any furtherincrease in the side chain length would diminish the possibility ofdevelopmental toxicity.

In order to systematically identify the substructures that lead topositive developmental toxicity prediction, we employed the reverseQSTR (rQSTR) functionality of TOPKAT. This allows calculation ofcontribution, termed moiety effect, of any atom, group of atoms, or auser-defined moiety to the predicted discriminant score. The chosenmoiety is assigned red color if its contribution is positive, or blue if itscontribution is negative; the darker the color, the larger the absolutevalue of the contribution (Fig. 3).

Before undertaking rQSTR, the confidence in each prediction wasestablished. In addition to evaluating the validation criteria, weutilized the “Similarity Search” functionality of TOPKAT andemployed the process shown in Fig. 4 to assign high, moderate, orlow confidence in a prediction. For example, ethoxyacetic acid ispredicted to be a developmental toxicant (positive) with aprobability of 0.987. The similarity search on the entire database

Fig. 3. Moiety effects for the developmental toxicity of 2-methoxyethanol and its metabolites. The red color represents the positive contribution of the moiety and the blue colorrepresents the negative contribution of the moiety. The darker color represents the larger absolute value of the contribution.

204 P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

revealed that four compounds–valproic acid, dimethadione, 1,3-butylene glycol, and propylene glycol monomethyl ether–have thehighest QSTR similarity to ethoxyacetic acid with similarity distanceb0.25. Since all four training set compounds are correctly predicteddevelopmental toxicants, we assigned high confidence to the positiveprediction for ethoxyacetic acid. Predictions of alkoxyethanol andacetate metabolites also have a high level of confidence; however, 2-butoxyacetaldehyde showed a lower level of confidence that is thesimilarity distance is >0.25.

For functional groups, the greatest contribution to the calculateddevelopmental toxicity was due to the fragment [aliphatic O].Therefore, this fragment may be the main driver of developmental

Cmpdswith Sim.

Dist. <0.25

0

ModerateConfidence

0

NoConfidence

1,2Compounds with correct prediction

1,2

LowConfidence

Search Database for

Similar Compounds

Fig. 4. The decision matrix for assigning

toxicity. It must be noted that since TOPKAT employs context-sensitive E-state values as descriptors, the contribution of a givensubstructural fragment may change depending on the entiremolecular context. This feature of TOPKAT distinguishes itself fromother group-contribution approaches in which the contribution of agiven group is constant irrespective of its environment and totalmolecular context.

Conclusions

Alkoxyethanols are a family of ethylene glycol ethers widely usedas organic solvents and thinners for resins, paints, and dyes. QSTR

3,4 Compounds with correct prediction

3,4High

Confidence

1,2

0

NoConfidence

ModerateConfidence

confidence to the model predictions.

205P. Ruiz et al. / Toxicology and Applied Pharmacology 254 (2011) 198–205

models as embedded in a commercial toxicity prediction system,TOPKAT, were applied to assess toxicity potential of these compounds.

Concordance between experimental and TOPKAT-predicted muta-genicity data on many of these compounds confirmed the accuracy ofTOPKAT model predictions. Althoughmutagenicity and carcinogenicitywere not indicated for the majority of these molecules, predictionsindicated their developmental toxicity potential. By applying uniquefeatures of TOPKAT, we were able to identify substructures andstructural features that may lead to increased developmental toxicitypotential. Some insightwas gained into the substructures thatmay alterpotential for developmental toxicity.

As indicated by a growing number of computational tools andmodels being developed for various toxicity endpoints, computationaltoxicology approaches are becoming an important tool for assessingthe health risk of environmental and pharmaceutical chemicals wheregaps in the toxicological data exist (El-Masri et al., 2002; Gombaret al., 2003; Richard et al., 2008; Ruiz, et al., 2008). These areinvaluable tools for assessing chemical hazards, especially when theonly characteristic known about a chemical is its structure. Athoughtful application and prudent use of QSTR models can helpestimate the toxicity of certain chemicals, prioritize those that lackdata, and through this process save cost and time.

Regulators can effectively use these methods, as part of a weightof evidence evaluation, to support health-based decision making formanaging environmental chemical exposures and thus leverage togreat advantage the public's safe use of chemicals (Pavan and Worth,2008; Worth et al., 2007). To more efficiently identify the potentialtoxicity of chemicals, the European Commission recommends usingboth SAR and QSAR methods for the registration, evaluation,authorization and restriction of chemical substances (REACH)(Cronin et al., 2003). But to use computational tools in a reasonablemanner, we need to continue to develop, evaluate, validate, andassess their utility.

Conflict of interest statement

The authors declare that they have no conflicts of interest.

References

Accelrys, 2004. TOPKAT User Guide Version 6.2. Accelrys, Inc, Burlington, MA.Benigni, R., Bossa, C., Jeliazkova, N., Netzeva, T., Worth, A., 2008. The Benigni / Bossa

rulebase for mutagenicity and carcinogenicity—a module of Toxtree. EuropeanCommission report EUR 23241 EN.

Boatman, R.J., Knaak, J.B., 2001. Ethers of ethylene glycol and derivatives, fifth ed. In:Bingham, E., Cohrssen, B., Powell, C.H. (Eds.), Patty's Toxicology, Vol 7. John Wiley,New York, pp. 73–270.

Browning, R.G., Curry, S.C., 1994. Clinical toxicology of ethylene glycol monoalkylethers. Hum. Exp. Toxicol. 3 (5), 325–335.

Cronin, M.T., Jaworska, J.S., Walker, J.D., Comber, M.H., Watts, C.D., Worth, A.P., 2003.Use of QSARs in international decision-making frameworks to predict health effectsof chemical substances. Environ. Health Perspect. 111, 1391–1401.

Demchuk, E., Ruiz, P., Wilson, J.D., Scinicariello, F., Pohl, H.R., Fay, M., Mumtaz, M.M.,Hansen, H., De Rosa, C.T., 2008. Computational toxicology methods in public healthpractice. Toxicol. Mech. Methods 18, 119–135.

El-Masri, H.A., Mumtaz, M.M., Choudhary, G., Cibulas, W., De Rosa, C.T., 2002.Applications of computational toxicology methods at the Agency for ToxicSubstances and Disease Registry. Intern. J. Hyg. Environ. Health. 205, 63–69.

Enslein, K., 1993. The future of toxicity predictionwithQSAR. In. Vitr. Toxicol. 6, 163–169.Enslein, K., 1997. QSTR applications in acute, chronic, and developmental toxicity, and

carcinogenicity. Adv. Mol. Toxicol. 141–164.Enslein, K., Gombar, V.K., Blake, B., 1994. Use of SAR in computer-assisted prediction of

carcinogenicity and mutagenicity of chemicals by the TOPKAT program. Mut Res.305, 47–61.

Gombar, V.K., 1997. Quantitative structure–activity relationships in toxicology: fromfundamentals to applications. Adv. Mol. Toxicol. 125–139.

Gombar, V.K., Enslein, K., 1990. Quantitative structure–activity relationship (QSAR)studies using electronic descriptors calculated from topological and molecularorbital (MO) methods. Quant. Struct. Act. Relat. 9, 321–325.

Gombar, V.K., Enslein, K., 1996. Assessment of n-octanol/water partition coefficient:when is the assessment reliable? J. Chem. Inf. Comput. Sci. 36, 1127–1134.

Gombar, V.K., Jain, D.V.S., 1987. Quantification of molecular shape and its correlationwith physico-chemical properties. Indian J. Chem. 26A, 554–555.

Gombar, V.K., Enslein, K., Blake, B.W., 1995a. Assessment of developmental toxicitypotential of chemicals by quantitative structure–toxicity relationship models.Chemosphere 31, 2499–2510.

Gombar, V.K., Enslein, K., Blake, B., 1995b. Assessment of developmental toxicitypotential of chemicals by quantitative structure–toxicity relationship models.Chemosphere 31 (1), 2499–2510.

Gombar, V.K., Silver, I.S., Zhao, Z., 2003. Role of ADME characteristics in drug discoveryand their in silico evaluation: in silico screening of chemicals for their metabolicstability. Curr. Top. Med. Chem. 3, 1205–1225.

Greene, N., 2002. Computer systems for the prediction of toxicity: an update. Adv. Drug.Del. Rev. 54, 417–431.

Hall, L.H.,Mohney, B., Kier, L.B., 1991. The electrotopological state: structure information atthe atomic level for molecular graphs. J. Chem. Inf. Comput. Sci. 31, 76–82.

Helma, C., 2006. Lazy structure-activity relationships (lazar) for the prediction ofrodent carcinogenicity and Salmonella mutagenicity. Mol. Divers. 10, 147–158.

Hoflack, J.C., Lambolez, L., Elias, Z., Vasseur, P., 1995. Mutagenicity of ethylene glycolethers and of their metabolites in Salmonella typhimurium his. Mutat. Res. 341,281–287.

Kier, L.B., 1986. Shape indices of orders one and three from molecular graphs. Quant.Struct. Act. Relat. 5, 1–7.

Klopman, G., 1984. Artificial intelligence approach to structure–activity studies:computer automated structure evaluation of biological activity of organicmolecules. J. Am. Chem. Soc. 106, 7315–7321.

Klopman,G., Rosenkranz, H., 1991. Structure-activity relations:maximizing the usefulnessof mutagenicity and carcinogenicity databases. Environ. Health Perspect. 96, 67–75.

Krasavage, W.J., Katz, G.V., 1985. Developmental toxicity of ethylene glycol monopropylether in the rat. Teratology 32, 93–102.

Ma, H., An, J., Hsie, A.W., Au, W.W., 1993. Mutagenicity and cytotoxicity of 2-methoxyethanol and its metabolites in Chinese hamster cells (the CHO/HPRT andAS52/GPT assays. Mut. Res. 298, 219–225.

Moudgal, C.J., Lipscomb, J.C., Bruce, R.M., 2000. Potential health effects of drinking waterdisinfection by-products using quantitative structure toxicity relationship. Toxicology147, 109–131.

Mumtaz, M.M., Knauf, L.A., Reisman, D.J., Peirano, W.B., DeRosa, C.T., Gombar, V.K.,Enslein, K., Carter, J.R., Blake, B.W.,Huque, K.I., Ramanujam, V.M.S., 1995. Assessmentof effect levels of chemicals fromquantitative structure-activity relationship (QSAR)models. I. chronic lowest-observed-adverse-effect level (LOAEL). Toxicol. Lett. 79,131–143.

Nagano, K., Nakayama, E., Oobayashi, H., Nishizawa, T., Okuda, H., Yamazaki, K., 1984.Experimental studies on toxicity of ethylene glycol alkyl ethers in Japan. Environ.Health Perspect. 57, 75–84.

NTP, 2000. Toxicology and carcinogenesis studies of 2-butoxyethanol (CAS no. 111-76-2)in F344/N rats and B6C3F1mice (inhalation studies). Research Triangle Park, NC,UnitedStatesDepartment ofHealth andHumanServices, National Institutes ofHealth,National Toxicology Program (NTP TR 484).

Pavan, M., Worth, A.P., 2008. Publicly-accessible QSAR software tools developed by theJoint Research Centre. SAR QSAR Environ. Res. 19 (7-8), 785–799.

Richard, A.M., 1998. Commercial toxicology prediction systems: a regulatoryperspective. Toxicol. Lett. 103, 611–616.

Richard, A., Yang, C., Judson, R., 2008. Toxicity data informatics: supporting a newparadigm for toxicity prediction. Toxicol. Mech. Methods 18, 103–118.

Ruiz, P., Faroon, O., Moudgal, C.J., Hansen, H., De Rosa, C.T., Mumtaz, M., 2008. Predictionof the health effects of polychlorinated biphenyls (PCBs) and theirmetabolites usingquantitative structure-activity relationship (QSAR). Toxicol. Lett. 181 (1), 53–65.

Sanderson, D.M., Earnshaw, C.G., 1991. Computer prediction of possible toxic actionfrom chemical structure; the DEREK system. Hum. Exp. Toxicol. 10, 261–273.

Scott, W.J., Fradkin, Wittfoht, W., Nau, H., 1989. Teratologic potential of 2-methoxyethanol and transplacental distribution of its metabolite, 2-methoxyaceticacid, in non-human primates. Teratology 39, 363–373.

Starek, A., Szabla, J., 2008. Ethylene glycol alkyl ethers–the substances noxious tohealth. Med. Pr. 59, 179–185.

Venkatapathy, R., Wang, C.Y., Bruce, R.M., Moudgal, C., 2009. Development of quantitativestructure-activity relationship (QSAR) models to predict the carcinogenic potency ofchemicals I. Alternative toxicity measures as an estimator of carcinogenic potency.Toxicol. Appl. Pharmacol. 234, 209–221.

Wang, R.S., Suda, M., Gao, X., Wang, B., Nakajima, T., Honma, T., 2004. Health effects ofexposure to ethylene glycol monoethyl ether in female workers. Ind. Health 42,447–451.

WHO/IPCS, 2005. Concise International Chemical Assessment Documents (CICAD): 2-Butoxyethanol (update)(No. 67, 2005). http://www.who.int/ipcs/publications/cicad/cicad_67_2_butoxyethanol.pdf. Accessed May, 2010.

WHO/IPCS, 2009a. Concise International Chemical Assessment Documents (CICAD): 2-Methoxyethanol (No. 67, 2009). http://www.who.int/ipcs/publications/cicad/methoxyethanol.pdf. Accessed May, 2010.

WHO/IPCS, 2009b. Concise International Chemical Assessment Documents (CICAD): 2-Ethoxyethanol and 2-Propoxyethanol (No. 67, 2009). http://www.who.int/ipcs/publications/cicad/ethoxy_propoxyethanol.pdf. Accessed May,2010.

Woo, Y.T., Lai, D.Y., 2005. OncoLogic: a mechanism-based expert system for predictingthe carcinogenic potential of chemicals. In: Helma, C. (Ed.), Predictive Toxicology.CRC Press, Boca Raton, pp. 385–414.

Worth, A.P., Bassan, A., De Bruijn, J., Gallegos-Saliner, A., Netzeva, T., Patlewicz, G.,Pavan, M., Tsakovska, I., Eisenreich, S., 2007. The role of the European ChemicalsBureau in promoting the regulatory use of (Q)SAR methods. SAR QSAR Environ.Res. 18 (1–2), 111–125.

Zeiger, E., Ashby, J., Bakale, G., Einslein, K., Klopman, G., Rosenkranz, H.S., 1996.Prediction of Salmonella mutagenicity. Mutagenesis 11 (5), 471–484.