Screening of 397 Chemicals and Development of a ... - ACS Publications

We have screened 397 chemicals for human androgen receptor (AR) antagonism by a sensitive reporter gene assay to generate data for the development of ...
0 downloads 0 Views 2MB Size
Chem. Res. Toxicol. 2008, 21, 813–823

813

Screening of 397 Chemicals and Development of a Quantitative Structure-Activity Relationship Model for Androgen Receptor Antagonism Anne Marie Vinggaard,* Jay Niemelä,* Eva Bay Wedebye, and Gunde Egeskov Jensen National Food Institute, Department of Toxicology and Risk Assessment, Technical UniVersity of Denmark, Mørkhøj Bygade 19, DK-2860 Søborg, Denmark ReceiVed July 3, 2007

We have screened 397 chemicals for human androgen receptor (AR) antagonism by a sensitive reporter gene assay to generate data for the development of a quantitative structure-activity relationship (QSAR) model. A total of 523 chemicals comprising data on 292 chemicals from our laboratory and data on 231 chemicals from the literature constituted the training set for the model. The chemicals were selected with the purpose of representing a wide range of chemical structures (e.g., organochlorines and polycyclic aromatic hydrocarbons) and various functions (e.g., natural hormones, pesticides, plastizicers, plastic additives, brominated flame retardants, and roast mutagens). In addition, the intention was to obtain an equal number of positive and negative chemicals. Among our own data for the training set, 45.7% exhibited inhibitory activity against the transcriptional activity induced by the synthetic androgen R1881. The MultiCASE expert system was used to construct a QSAR model for AR antagonizing potential. A “5 Times, 2-Fold 50% Cross Validation” of the model showed a sensitivity of 64%, a specificity of 84%, and a concordance of 76%. Data for 102 chemicals were generated for an external validation of the model resulting in a sensitivity of 57%, a specificity of 98%, and a concordance of 92% of the model. The model was run on a set of 176103 chemicals, and 47% were within the domain of the model. Approximately 8% of chemicals was predicted active for AR antagonism. We conclude that the predictability of the global QSAR model for this end point is good. This most comprehensive QSAR model may become a valuable tool for screening large numbers of chemicals for AR antagonism. Introduction There is increasing evidence that a variety of chemicals have the potential to disrupt the endocrine system by mimicking or inhibiting endogenous hormones such as estrogens and androgens. Many endocrine-disrupting chemicals (EDCs) have a potential to adversely affect development and/or reproductive function in wildlife, experimental animals, and humans. Among the many biological mechanisms that can result in endocrine disruption, one important is the expression of an antiandrogenic response. Chemicals with antiandrogenic activity counteract the effect of the male sex steroid hormones either by affecting their synthesis or metabolism or by blocking the effects of androgens. Androgens such as testosterone and dihydrotestosterone play a crucial role at several stages of male development and in the maintenance of the male phenotype. The development of the male phenotype during gestation is totally dependent on the action of androgens (1), and interference with AR action at this point of development is hypothesized as being linked to the increased frequency of male reproductive disorders such as testicular dysgenesis syndrome (2). The blocking of androgen action may be exerted by antagonism of the androgen receptor (AR), that is, by direct interaction of the chemical with AR. The AR is a member of the nuclear receptor superfamily, a class of receptors that function through ligand-dependent transcription of specific genes (3). Receptor binding is the primary intracel* To whom correspondence should be addressed. (A.M.V.) E-mail: [email protected]. (J.N.) E-mail: [email protected].

lular step, and transactivation of the receptor is the critical step for androgen-dependent gene expression in vitro and in vivo (4, 5). The structural diversity of chemicals, which can bind to and affect transactivation of AR, is very broad. Chemicals from several different categories including steroids, synthetic hormones, polycyclic aromatic hydrocarbons (PAHs) (6), polychlorinated biphenyls (PCBs) (7), plastic additives (8), diphenyl derivatives, pesticides (9-19), pharmaceuticals, and a number of other miscellaneous chemicals interact with AR. It is conceivable that many other chemicals with antiandrogenic activity still have to be identified. In response to scientific and public concerns on EDCs, strategies are needed for screening and testing of a large number of chemicals that are present in our food, environment, consumer products, etc. In vivo assays for the detection of antiandrogenic action are time-consuming, costly, and labor intensive, which makes them impractical for routine screening and testing of a large number of chemicals. Although in vitro data for AR antagonism alone are not sufficient to characterize a compound as an EDC, information on the ability of a chemical to antagonize AR provides an important piece of information for priority setting of chemicals for more elaborate in vivo assays. Because of the high cost and time associated with screening and testing of EDCs, it is crucial that priorities be set to ensure that chemicals with the highest predicted probabilities or measured activities are given first priority for entry into the screening procedure. The challenge of evaluating a very large number of chemicals for their antiandrogenic activity is offset by the ability to construct quantitative structure-activity

10.1021/tx7002382 CCC: $40.75  2008 American Chemical Society Published on Web 03/07/2008

814

Chem. Res. Toxicol., Vol. 21, No. 4, 2008

relationship (QSAR) models for the rapid prediction of activity. Such models may have a great potential for use in the identification of large numbers of potential AR antagonists. At the very least, QSAR models could be employed to establish a prioritization procedure for subsequent biological testing. The use and regulatory acceptance of QSAR models for predicting biological effects have been increasing for the last 15 years. QSAR models for a wide range of end points, including end points for endocrine disruption, such as binding to the AR, have been developed (20-24). These models may be local models covering one chemical category, for example, unsaturated aldehydes (25), or global ligand-based models covering more chemical categories, for example, polyaromatic hydrocarbons, benzophenones, diphenyl ethers, phthalate esters, etc (26). Our actual model is a global fragment-based model covering many chemical categories. For this paper, we used an AR transactivation or reporter gene assay that we have developed (27) and further modified (28). This assay has proved to be a powerful tool for detecting AR antagonists among chemicals as shown for instance in an interlaboratory study, where four different AR assays were compared (28). A total of 397 chemicals were tested in our laboratory. Some of these chemicals are environmentally persistent and/or commercially important, and for many of them, the AR activity has not previously been reported. A total of 292 chemicals served as the training set together with additional data for 231 chemicals obtained from the literature. We used the MultiCASE methodology to analyze the structure-activity relationship of this large data set of diverse molecules. MultiCASE is a QSAR expert system capable of training automatically from data and organizing that knowledge into an expert system (29). The model was cross-validated by the leave-groupsout method, and a further 102 chemicals were tested as part of an external validation of the QSAR model.

Materials and Methods In Vitro AR Assay. We tested AR transactivation in a sensitive luciferase reporter assay (27, 18, 28). Chinese hamster ovary cells (CHO K1 from ATCC, Maryland) were maintained in DMEM/F12 (Gibco, Paisley, United Kingdom) supplemented with 100 U/mL penicillin, 100 µg/mL streptomycin, and 0.25 µg/mL amphotericin B (PAA Laboratories GmbH, Pasching, Austria) and 10% fetal bovine serum (BioWhittaker Inc., Walkersville, MD). The cells were seeded in white 96 well plates (PerkinElmer Life Sciences, Packard) at a density of 7000 cells/well in DMEM/F12 containing 10% charcoal-treated fetal bovine serum (Biological Industries Ltd., Kibbutz Beit Haemek, Israel) and incubated at 37 °C in a humidified atmosphere of 5% CO2/air. After 24 h, cells were transfected for 5 h with a total of 75 ng cDNA/well consisting of the expression vector pSVAR0 (human AR) and the MMTV-LUC reporter plasmid (gifts from Dr. Albert Brinkmann, Erasmus University, Rotterdam) in a ratio of 1:100 using 300 nL of the transfection reagent FuGene (Roche Diagnostics A/S, Hvidovre, Denmark). The ratio of DNA (µg) to Fugene (µL) was kept at 0.25. Test solutions were prepared from 10 mM stock solutions in ethanol, DMSO, orsif insoluble in these solventssin dimethylformamide. None of the solvents had an effect in the assay at relevant concentrations. Test compounds were tested in the presence of 0.1 nM R1881 typically at concentrations of 1, 3, 10, and 30 µM in triplicates (except where stated in Table 1), and within each assay, all data were related to 0.1 nM R1881 (NEN, Boston, MA), which was set to 100%. After incubation for 20 h, cells were lysed by adding 20 µL/well of a lysis buffer containing

Vinggaard et al.

25 mM trisphosphate, pH 7.8, 15% glycerol, 1% Triton X-100, 1 mM DTT, and 8 mM MgCl2, followed by shaking at room temperature for 10 min. The luciferase activity was measured directly using a BioOrbit Galaxy luminometer by automatic injection of 40 µL of substrate containing 1 mM luciferin (Amersham Int., Buckinghamshire, United Kingdom) and 1 mM ATP (Boehringer Mannheim, Germany) in lysis buffer, and the chemiluminescence generated from each well was measured over a 1 s interval after an incubation time of 2 s. We determined cytotoxicity in parallel by transfecting cells with a plasmid (pSVAR13) encoding for a constitutively active AR, which lacks the ligand-binding domain (a gift from Dr. Albert Brinkmann, Erasmus University, Rotterdam). These experiments were designed exactly as was the AR reporter gene assay except that the ratio between pSVAR13 and MMTV-LUC was 2:100. Most pipetting procedures were performed using a Biomek2000 laboratory robot (Beckman Coulter, Fullerton, CA). Study Design and Data Analysis. Chemicals were tested at concentrations of 1, 3, 10, and 30 µM in triplicates, and within each assay, all data were related to the response of 0.1 nM R1881, which was set to 100%. In the case of equivocal results (for antagonism and/or cytotoxicity) or if the concentration range of test compound had to be adjusted, testing was repeated. For some more potent compounds, test concentrations were 10 times lower, that is, 0.1, 0.3, 1, and 3 µM, except for hydroxyflutamide, which was diluted 100-fold. For TCDD, PeCDD, PeCDF, and HxCDD, test concentrations were 8, 25, 80, and 250 nM due to technical reasons. The dilution factors for these compoundsaregiveninparenthesesinTable1.Concentration-response analyses were performed, and the IC25, that is, the concentration of test compound showing a 25% inhibition of the activity induced by 0.1 nM R1881, was calculated for each compound. The criteria for determination of “a positive” was that a 25% inhibition of the 0.1 nM R1881-induced response should be reached at noncytotoxic concentrations e10 µM. Modeling Methodology. Algorithm. The Multiple Computer Automated Structure Evaluation (MultiCASE, version 2006) system was used (29). MultiCASE is a fragment-based statistical model system. The program aims to discover substructures that appear in active molecules and therefore may be responsible for the observed activity. It starts by identifying the statistically most significant substructure present in the training set. This fragment, labeled the top biophore, is considered responsible for the activity of the largest possible number of active molecules. The molecules containing this biophore are then removed from the database, and the remaining ones are submitted to a new analysis leading to the identification of the next biophore. This procedure is repeated, either until the activity of all of the molecules in the training set has been accounted for or until no additional statistically significant substructure can be found. The chemicals containing the same biophore are grouped together. For each set of molecules containing a specific biophore, MultiCASE identifies additional parameters, called modulators. These modulators may be structural fragments or chemical properties (e.g., log P, molecular orbital energies) that can either enhance or inhibit the activity of the chemicals containing the biophore. The relevant modulators are then used to derive a QSAR model, restricted to the chemicals containing the biophore. The objective of the MultiCASE analysis is to identify the most representative chemical fragments believed to be responsible for the effect, in this case the AR antagonism of molecules within the training set. When challenged with a new chemical, the program will search for the presence of one of the biophores previously discovered and apply the appropriate

QSAR Model for Antiandrogenic Effect

Chem. Res. Toxicol., Vol. 21, No. 4, 2008 815

Table 1. Data Generated and Used as a Training Set for QSAR Modelinga CAS no. estriol estrone corticosterone progesterone (10×) acetoxyprogesterone betulin bicalutamide (10×) busulfane cimetidine cyproteronacetate (10×) cortisonacetate dexamethasone dihydroxyprogesterone (17R,20β) diethylstilbestrol epoxyprogesterone finasteride flutamide hydroxyflutamide (100×) ICI 182 780 alachlor anilazin atrazin benzenamine, 2,3,4,5,6-pentachlorobifenox carbamic acid, (3-chlorophenyl)-, 4-chloro-2-butynyl ester chlormequatchloride chlordecone (kepone) chlorobenzilate chlozolinate cypermethrin cyproconazole cyromazin deltamethrin dichlone dichlorvos dicofol dieldrin dimetridazole diuron β-endosulfan epoxiconazole etridiazole fenarimol fenchlorfos fenitrothion fenthion (10×) fenvalerate imazalil iprodion isoproturone ketoconazol linuron M1 PCB PCB PCB PCB PCB PCB PCB PCB PCB PCB PCB

3 ∼ 4-chlorobiphenyl 4 ∼ 2,2′-dichlorobiphenyl 5 ∼ 2,3-dichlorobiphenyl 8 ∼ 2,4′-dichlorobiphenyl 10 ∼ 2,6-dichlorobiphenyl 11 ∼ 3,3′-dichlorobiphenyl 14 ∼ 3,5-dichlorobiphenyl 19 ∼ 2,2′,6-trichlorobiphenyl 22 ∼ 2,3,4′-trichlorobiphenyl 24 ∼ 2,3,6-trichlorobiphenyl 28 ∼ 2,4,4′-trichlorobiphenyl

50-27-1 53-16-7 50-22-6 57-83-0

classification

CAS no.

Natural hormones 4 4-androsten-3,17-dione neg 17R-estradiol (10×) 3 17β-estradiol (10×) 2

Synthetic hormones and drugs neg medroxyprogesterone neg medroxyprogesterone, 17-acetate 1 metyrapone 4 mifepristone (10×) neg nilutamide (10×) 4 spironolactone (10×) neg spirocort 4 testosterone propionate 3 1-butanone, 4-[4-(4-chlorophenyl)4-hydroxy-1-piperidinyl]-1(4-fluorophenyl)56-53-1 4 2H-1,2,4-benzothiadiazine-7sulfonamide, 6-chloro-3,4-dihydro-, 1,1-dioxide 1097-51-4 3 2,4-pyrimidinediamine, 5-(4-chlorophenyl)-6-ethyl98319-26-7 4 17R-ethynylestradiol 13311-84-7 3 17R-methyltestosterone 52806-53-8 1 17β-estradiolbenzoate 129453-61-8 tox 17β-trenbolone (10×) 302-23-8 473-98-3 90357-06-5 55-98-1 51481-61-9 427-51-0 50-04-4 50-02-2 1662-06-2

15972-60-8 101-05-3 1912-24-9 527-20-8 42576-02-3 101-27-9

5 neg neg 3 4 neg

999-81-5 143-50-0 510-15-6 84332-86-5 52315-07-8 113096-99-4 66215-27-8 52918-63-5 117-80-6 62-73-7 115-32-2 60-57-1 551-92-8 330-54-1 33213-65-9 135319-73-2 2593-15-9 60168-88-9 299-84-3 122-14-5 55-38-9 51630-58-1 35554-44-0 36734-19-7 34123-59-6 65277-42-1 330-55-2 119209-27-7

neg neg 3 5 5 neg neg tox 5 neg 4 4 neg 3 5 5 neg 5 4 2 5 neg 5 neg neg 5 4 2

2051-62-9 13029-08-8 16605-91-7 34883-43-7 33146-45-1 2050-67-1 34883-41-5 38444-73-4 38444-85-8 55702-45-9 7012-37-5

neg 4 4 5 5 4 5 3 3 4 5

Pesticides M2 mercaptodimethur methoxychlor methylparathion metribuzin nitrofen o,p-DDT (2,4-DDT) p,p-DDT (4,4-DDT) p,p′-DDE (4,4-DDE) o,p-DDE pentane, 1,5-diiodopermethrin permethrin-cis permethrin-trans prochloraz procymidone (10×) propiconazole propyzamide pyrethrins pyrimidine, 4,6-dichloro-2-phenylresmethrin simazin sulfamethizole sulfamethoxazole sulfathiazole tebuconazole terbuthylazin toxaphen triadimefon triadimenol vinclozolin (10×) 1-chloro-2,3-propanediol (MCPD) 1,4-benzenediol, 2,3,5,6-tetrachloro4-hexylresorcinol PCBs PCB PCB PCB PCB PCB PCB PCB PCB PCB PCB PCB

30 34 39 40 46 49 52 58 64 65 66

∼ ∼ ∼ ∼ ∼ ∼ ∼ ∼ ∼ ∼ ∼

2,4,6- trichlorobiphenyl 2′,3,5-trichlorobiphenyl 3,4′,5-trichlorobiphenyl 2,2′,3,3′-tetrachlorobiphenyl 2,2′,3,6′-tetrachlorobiphenyl 2,2′,4,5′-tetrachlorobiphenyl 2,2′,5,5′-tetrachlorobiphenyl 2,3,3′,5′-tetrachlorobiphenyl 2,3,4′,6-tetrachlorobiphenyl 2,3,5,6-tetrachlorobiphenyl 2,3′,4,4′-tetrachlorobiphenyl

classification

63-05-8 57-91-0 50-28-2

neg 2 4

520-85-4 71-58-9 54-36-4 84371-65-3 63612-50-0 52-01-7 51333-22-3 57-85-2 52-86-8

neg neg neg 1 2 1 3 neg neg

58-93-5

neg

58-14-0

neg

57-63-6 58-18-4 50-50-0 10161-33-8

3 neg neg 3

83792-61-4 2032-65-7 72-43-5 298-00-0 21087-64-9 1836-75-5

1 5 3 3 neg 3

789-02-6 50-29-3 72-55-9 3424-82-6 628-77-3 52645-53-1 61949-76-6 61949-77-7 67747-09-5 32809-16-8 60207-90-1 23950-58-5 121-29-9 3740-92-9 10453-86-6 122-34-9 144-82-1 723-46-6 72-14-0 107534-96-3 5915-41-3 8001-35-2 43121-43-3 55219-65-3 50471-44-8 96-24-2 87-87-6 136-77-6

3 4 3 4 neg neg neg neg 5 3 5 neg neg tox neg neg neg neg neg 4 neg 4 neg 5 2 neg 4 5

35693-92-6 37680-68-5 38444-88-1 38444-93-8 41464-47-5 41464-40-8 35693-99-3 41464-49-7 52663-58-8 33284-54-7 32598-10-0

4 3 3 neg 3 5 neg neg 4 4 3

816

Chem. Res. Toxicol., Vol. 21, No. 4, 2008

Vinggaard et al. Table 1. Continued

CAS no.

classification

74 ∼ 2,4,4′,5-tetrachlorobiphenyl 75 ∼ 2,4,4′,6-tetrachlorobiphenyl 101 ∼ 2,2′,4,5,5′-pentachlorobiphenyl 105 ∼ 2,3,3′,4,4′-pentachlorobiphenyl 114 ∼ 2,3,4,4′,5-pentachlorobiphenyl 118 ∼ 2,3′,4,4′,5-pentachlorobiphenyl

32690-93-0 32598-12-2 37680-73-2 32598-14-4 74472-37-0 31508-00-6

4 3 neg 3 5 4

PCB 126 ∼ 3,3′,4,4′,5-pentachlorobiphenyl

57465-28-8

4

PCB 138 ∼ 2,2′,3,4,4′,5′-hexachlorobiphenyl

35065-28-2

5

PCB 153 ∼ 2,2′,4,4′,5,5′-hexachlorobiphenyl

35065-27-1

neg

anthracene, 1-chloro-9,10-bis(phenylethynyl)anthracene, 9,10-dimethylanthracene, 9,10-dichloroanthracene, 9,10-diethoxyanthracene, 9,10-diphenylanthracene, 9,10-dibromoanthracene benzo[a]pyrene benzo[b]fluoranthene benzo(ghi)perylene benzo[j]fluoranthene benzo[k]fluoranthene chrysene cyclopentane(def)phenanthene dibenzo[a,h]anthracene fluoranthene fluorene naphthalene

41105-35-5 781-43-1 605-48-1 68818-86-0 1499-10-1 523-27-3 120-12-7 50-32-8 205-99-2 191-24-2 205-82-3 207-08-9 218-01-9 203-64-5 53-70-3 206-44-0 86-73-7 91-20-3

neg 5 5 neg tox neg 5 5 neg neg 4 3 5 4 neg 4 neg neg

PCB PCB PCB PCB PCB PCB

BDE 19 ∼2,2′,6-tribromophenylether BDE 49 ∼2,2,′4,5′-tetrabromophenylether BDE 100 ∼2,2′,4,4′,6pentabromophenylether AC DiMeIQx Glu-P1j Glu-P-2 IQ benzyl-butylphthalate (BBP) bisphenol A bisphenol-A-dimethacrylate bisphenol F bisphenol A diglycidylether dibutylphthalate (DBP) diethylhexyladipate (DEHA)

PCBs PCB 155 ∼ 2,2′,4,4′,6,6′-hexachlorobiphenyl PCB 156 ∼ 2,3,3′,4,4′,5-hexachlorobiphenyl PCB 157 ∼ 2,3,3′,4,4′,5′-hexachlorobiphenyl PCB 167 ∼ 2,3′,4,4′,5,5′-hexachlorobiphenyl PCB 169 ∼ 3,3′,4,4′,5,5′-hexachlorobiphenyl PCB 180 ∼ 2,2′,3,4,4′,5,5′heptachlorobiphenyl PCB 202 ∼ 2,2′,3,3′,5,5′,6,6′octachlorobiphenyl PCB 209 ∼ 2,2′,3,3′,4,4′,5,5′,6,6′decachlorobiphenyl PAHs pentacene phenanthrene pyrene pyrene, 1-nitroretene triphenylene 1-acetylpyrene 1-aminopyrene 1-hydroxypyrene 1-pyrenebutyric acid 1,2-benzanthrazene 2-aminoanthracene 5,6,11,12-tetraphenyl-naphthacene 9,10-anthracenedicarboxaldehyd 9,10-bis(phenylethynyl)-anthracene 9,10-dimethyl-1,2-benzanthracene 9-methylanthracene

Brominated flame retardants 3 BDE 181∼2,2′,3′,4,4′,5,6heptabromodiphenylether 40088-47-9 4 BDE 190 ∼2,3,3′,4,4′,5,6heptabromodiphenylether 32534-81-9 3 tetrabromobisphenol A 49690-94-0

Roast mutagens (heterocyclic amines) 26148-68-5 neg MeAC 95896-78-9 tox MeIQ 67730-11-4 neg MeIQx 67730-10-3 neg PHIP 76180-96-6 neg Plasticizers and plast additives neg diethylhexylphthalate (DEHP) 4 diethylphthalate (DEP) 4 diisononyl phthalate (DINP) 4 mono-2-ethylhexylphthalate (MEHP) 1675-54-3 neg nonylphenol 84-74-2 neg 4-tert-octylphenol 103-23-1 neg 85-68-7 80-05-7 3253-39-2 620-92-8

CAS no.

classification

33979-03-2 38380-08-4 69782-90-7 52668-72-6 32774-16-6 35065-29-3

neg 5 4 5 neg neg

2136-99-4

4

2051-24-3

neg

135-48-8 85-01-8 129-00-0 5522-43-0 483-65-8 217-59-4 3264-21-9 1606-67-3 5315-79-7 3443-45-6 56-55-3 613-13-8 517-51-1 7044-91-9 10075-85-1 57-97-6 779-02-2

neg neg 5 3 3 neg 5 4 5 neg 4 4 tox tox tox 5 4

189084-67-1

5

189084-68-2

neg

79-94-7

4

68006-83-7 77094-11-2 77500-04-0 105650-23-5

neg neg neg neg

117-81-7 84-66-2 28553-12-0 4376-20-9

neg neg neg neg

104-40-5 27193-28-8

neg 3

Food additives and cosmetic additives 458-37-7 neg triclosan 94-26-8 neg 2-ethylhexylen 4-methoxycinnamate 120-47-8 neg 2-hydroxy-4-methoxybenzophenone (benzophenon-3) 118-56-9 3 3-benzylidene camphor (3-BC) ∼ unisol

3380-34-5 83834-59-7 131-57-7

5 tox neg

15087-24-8

neg

93-15-2

neg

36861-47-9

neg

methyl-4-hydroxybenzoate (methylparaben) octadecanoic acid (lauric acid) propyl-p-hydroxybenzoate (propylparaben)

99-76-3 143-07-7 94-13-3

neg neg neg

4674-50-4 463-40-1 506-26-3

4 neg 4

butylated hydroxyanisole (BHA) ethoxyquin laurylgallate nordihydroguaiaretic acid tert-buytlhydroquinon (TBHQ)

25013-16-5 91-53-2 1166-52-5 500-38-9 1948-33-0

Antioxidants neg 1,4-benzenediamine, N,N′-diphenylneg 2,4,5-trihydroxy-butyrophenon tox 2,5-di-tert-butylhydroquinone tox 2,6-di-tert-butyl-hydroxymethylphenol 5 3,5-di-tert-butyl-4-hydroxytoluen (BHT)

74-31-7 1421-63-2 88-58-4 88-26-6 128-37-0

tox neg tox 4 neg

curcumin butyl-p-hydroxybenzoate (butylparaben) ethyl-4-hydroxybenzoate (ethylparaben) homosalate (3,3,5-trimethylcyclohexyl salicylate) methyl eugenol

3-(4-methylbenzylidene)camphor (4-MBC) (+)-nootkatone R-linolenic acid λ-linolenic acid

QSAR Model for Antiandrogenic Effect

Chem. Res. Toxicol., Vol. 21, No. 4, 2008 817 Table 1. Continued CAS no.

classification

CAS no.

classification

biochanin A coumestrol (2×) daidzein

491-80-5 479-13-0 486-66-8

Plant compounds neg equol neg genistein neg phloretin

531-95-3 446-72-0 60-82-2

neg neg neg

acetone acrylamide benzene, (chloromethyl)pentamethylbenzene, 2,4-dimethoxy-1-nitrobenzene, 1,1′,1′′-(chloromethylidyne)trisbenzenemethanol, R,R-diphenylbenzeneacetic acid, R-hydroxy-R-phenylbenzenethiol, 4-methylbenzenethiol, 4-chlorobenzenesulfonic acid benzoic acid, 3,5-dichloro-4-hydroxy-, ethyl ester benzonitrile, 2-chloro-5-(trifluoromethyl)benzonitrile, 2-(trifluoromethyl)-

67-64-1 79-06-1 484-65-1 4920-84-7 76-83-5 76-84-6 76-93-7 106-45-6 106-54-7 98-11-3 17302-82-8

Miscellaneous neg 1,2,3,6,7,8-HxCDD (120×) neg 1,2,3,7,8-PeCDD (120×) neg 2-chloroaniline 5 2,3,4,7,8-PeCDF (120×) 5 2,3,7,8 TCDD (120x×) neg iso-octan neg mesitylene-2-sulfonic acid dihydrate 4 quinazoline, 4-chloro-2-phenyl tox 4-aminobenzoic acid (PABA) neg perfluoroctanoic acid (n-isomers) neg perfluoroctanoate, K-salt

57653-85-7 40321-76-4 95-51-2 57117-31-4 1746-01-6 26635-64-3 3453-83-6 6484-25-9 150-13-0 335-67-1 2795-39-3

1 neg neg 1 neg neg neg neg neg neg neg

328-87-0 447-60-9

neg neg

59-50-7 39225-17-7

5 tox

benzonitrile, 2-methylcarbostyril cholesterol cholic acid cyclohexylamine, N,N-diethyldimethylsulphoxide

529-19-1 59-31-4 57-88-5 81-25-4 91-65-6 67-68-5

neg neg neg neg 4 neg

66-22-8 17924-92-4 10373-78-1 54-88-6 81-83-4 1843-03-4

neg 5 neg neg neg tox

dimethylformamide dodecylamine ethanol ethanol, 2(dibutylamino)ethanone,2-hydroxy-1,2-bis(4-methoxyphenyl)-

68-12-2 124-22-1 64-17-5 102-81-8 119-52-8

neg neg neg 5 neg

91-15-6 120-82-1 117-08-8 88-06-2 81-30-1

neg neg neg neg neg

glycerol hexachlorbenzene (1,1′-biphenyl)-4-ol, 3,5-dichloro [1,1′-biphenyl]-4,4′-diamine, 3,3′-dichloro-

56-81-5 118-74-1 1137-59-3 91-94-1

neg neg 5 3

937-00-8 119-90-4 99-96-7 85-19-8

tox 5 neg 4

phenol, 4-chloro-3-methylphosphonic acid, [(4-chlorophenyl)methyl]-, diethyl ester uracil zearalenon D,L-camphorquinone N,N-dimethyl-4-phenylazo-m-toluidine 1H-benz[de]isoquinoline-1,3(2H)-dione 1,1,3-tris-(2-methyl-4-hydroxy-5t-butylphenyl)butan 1,2-benzenedicarbonitrile 1,2,4-trichlorbenzen 1,3-isobenzofurandione, 4,5,6,7-tetrachloro2,4,6-trichlorphenol [2]benzopyrano[6,5,4-def][2]benzopyran1,3,6,8-tetrone 3-(triflouromethyl)thiophenol 3,3-dianisidine 4-hydroxybenzoic acid 5-chloro-2-hydroxybenzophenone

a The name and CAS number of each chemical as well as the classification of the chemical as positive, negative, or toxic are shown. Chemicals were tested at 1, 3, 10, and 30 µM except for the more potent ones, which were diluted by the factors given in parentheses after the chemical name. Potency codes for the positive chemicals were as follows: 1, the IC25 e 0.1 µM; 2, 0.1 µM < IC25 e 0.3 µM; 3, 0.3 µM < IC25 e 1 µM; 4, 1 µM < IC25 e 3 µM; 5, 3 µM < IC25 e 10 µM; neg, negative chemicals; tox, cytotoxic chemicals. Noncytotoxic chemicals belonging to groups 1-5 were classified as positive for AR antagonism.

QSAR to project the AR antagonistic effect of that molecule. As well as the program aims to discover substructures (biophores) that appear in active molecules, it also discovers substructures (biophobes) that appear in inactive molecules. Units of Measurement. The experimental data express the ability of each chemical to inhibit the luminescence response induced by the synthetic androgen, R1881 (IC25). These responses were transformed into “CASE activity units” used for input to the MultiCASE program. If the chemical reached an IC25 at a test concentration e10 µM, the chemical was defined as “active” and assigned a CASE activity of 45. If an IC25 was not reached or if the chemical gave rise to cytotoxicity at concentrations g3 µM, the chemical was defined as “inactive” and assigned a CASE activity of 10. Applicability Domain of the Model. During the prediction process for a given substance, MultiCASE may provide warnings due to the presence of fragments not present in the training set and not covered by the model or the presence of inactivating fragments associated with an active prediction (or the opposite). In this study, any MultiCASE warning was considered being an indication that the molecule was outside the model domain. The applicability domain of the QSAR model is estimated by use of the Danish National Food Institute QSAR database, a local database with structural information on 176103 chemicals. The database includes chemicals from a number of inventories, for example, EINECS (European Inventory of

Existing Commercial Chemical Substances), ENCS (Japanese Existing and New Chemical Substances), TSCA (U.S. Toxic Substances Control Act), and NCI US (National Cancer Institute database). Training Set. The training set consisted of experimental data on 523 chemicals: 292 chemicals tested in our own laboratory and an additional data set for 231 chemicals obtained from the literature. Selection of chemicals to be tested for the training set was based on the following criteria: • Molecular weights were between 100 and 1000. • Chemicals represented various structures, although the majority contained an aromatic ring. • Known antiandrogens were as far as possible included. • An equal amount of positive and negative chemicals was intended and was obtained. • Half-way through testing, a preliminary QSAR was developed, based on data generated so far, and applied for selecting some of the remaining chemicals. Chemicals included in the training set encompass a broad range of structural classes including pesticides, PAH, PCBs, dioxins/furans, natural hormones, synthetic hormones, plasticizers, antioxidants, brominated flame retardants, parabens, mutagens of roast chemicals, bisphenol A derivatives, alkyl phenols, and plant constituents. Selected data from the following papers

818

Chem. Res. Toxicol., Vol. 21, No. 4, 2008

Vinggaard et al.

Figure 1. Illustrated is the distribution into various chemical classes of the 295 chemicals that were tested in our laboratory and included in the training set. The number of chemicals in each chemical class and the distribution within each class according to AR antagonizing potency of the chemicals are shown. Potency codes were as follows: neg, negative chemicals; 1, IC25 e 0.1 µM; 2, 0.1 µM < IC25 e 0.3 µM; 3, 0.3 µM < IC25 e 1 µM; 4, 1 µM < IC25 e 3 µM; and 5, 3 µM < IC25 e 10 µM. Noncytotoxic chemicals belonging to groups 1-5 were classified as positive for AR antagonism.

Table 2. Cross-Validation of the QSAR Model (5 Times, 2-Fold, 50% Cross-Validation) sensitivity (%) 64.4

specificity (%)

concordance (%)

Domain onlya 84.2

76.1

b

67.9 a

All predictions 75.5

Chemicals with warnings excluded. included.

72.1 b

Chemicals with warnings

were included in the training set to generate the QSAR model (13, 26, 30-44). Among the chemicals tested in our laboratory, cis- and transpermethrin were included in the QSAR model as a single chemical because cis/trans isomerism cannot be distinguished by the chemical codes used in the MultiCASE system (2D SMILES codes; Simplified Molecular Input Line Entry System). Benzophenone-3, which in our laboratory was tested negative at a concentration of 10 µM (the cutoff point) but positive at 30 µM and also was found positive by others (39), was included in the QSAR model as a positive. 17R- and 17β-Estradiol were included in the QSAR model as a single substance because of identical 2D SMILES codes. Fourteen chemicals (including cholesterol-synthetizing/metabolic compounds and glucose) with no experimental data were included in the QSAR model as negatives due to biological/physiological conditions. The training

set is available at the Internet address http://qsar.food.dtu.dk/ AntiAndrogenQSAR.zip, which contains information on CAS numbers, machine-readable structure notations, activities assigned in this model, and references to original literature. Data Set for External Validation. The internal crossvalidated QSAR model was “closed” and used to select 102 chemicals within the domain for external validation. The selection of chemicals was done according to the following criteria: • Only EINECS chemicals (European INventory of Existing Commercial chemical Substances), approximately 47000 chemicals, were considered. • Two lists of chemicals within the model domain representing positive and negative predictions, respectively, were generated. • The chemicals in each list were randomized, and chemicals in the top were selected for testing. If a chemical was not commercially available, the next chemical on the list was taken. The distribution of selected chemicals for external validation was approximately 10% predicted positive and 90% predicted negative. This approach was taken to reflect the prevalence of chemicals with positive and negative activity as predicted by the QSAR model (Table 5). Chemicals were blinded until test and data treatments were completed.

QSAR Model for Antiandrogenic Effect

Chem. Res. Toxicol., Vol. 21, No. 4, 2008 819

Table 3. Data for External Validation of the QSAR Modela CAS no. androsta-4,6-dien-3-one, 17-hydroxy-, (17β)benzene, 1,1′-(chlorophenylmethylene)bis [4-methoxybenzenamine, 4,4′,4′′-methylidynetris [N,N-dimethylbicyclo[2.2.1]heptan-2-ol, 1,3,3-trimethyl-, acetate 1,1′-biphenyl, 2-fluoro-

2484-30-2 40615-36-9

[1,1′-biphenyl]-2,5-diol pentacosanethiourea, N,N′-bis(4-methylphenyl)acetic acid, (2,4,5-trichlorophenoxy)dodecanedioic acid, diethyl ester uridine, 4-thioeicosanoic-acidacetic acid, pentyl ester 1,3-hexanediol, 2-ethylbenzene, 1-fluoro-4-(1-methylethenyl)benzene, 1-chloro-3-fluoroacetic acid, phenyl ester ethanone, 2,2,2-trifluoro-1-(3-fluorophenyl)2-propanamine, N-(1-methylethyl)[1,1′-biphenyl]-4-carboxylic acid pyridine, 3-bromophenol, 2,6-dichloro-4-nitro2-butanone, 4-(4-methoxyphenyl)benzonitrile, 4-aminoD-glucose, cyclic 1,2-ethanediyl mercaptal benzene, 1,1′-(chloroethenylidene)bis [4-chloro1-pentene, 3-methylbenzene, 1,3-dichlorocyclopentane, chlorophenol, 2,4,5-trichlorobenzoic acid, 2,6-dichlorocyclopentanecarboxylic acid, 1-(4-chlorophenyl)benzene, 1,2,4,5-tetrachlorobenzeneethanamine, N-methylpyridine, 5-ethyl-2-methyl2,4-pyrimidinediamine, 6-chloro2-pyridinecarboxaldehyde, 6-methyldecanoic acid, ethenyl ester

classification

CAS no.

chemicals predicted positive neg [1,1′-biphenyl]-2,2′-diol 5 1,1′-biphenyl, 2-chloro-

classification

1806-29-7 2051-60-7

4 5

603-48-5

tox

olean-12-en-28-oic acid, 3-hydroxy-, (3β)-

508-02-1

tox

13851-11-1

neg

1173-26-8

1

321-60-8

5

1247-42-3

3

1079-21-6

5

pregn-4-ene-3,20-dione, 21-(acetyloxy)-11-hydroxy-, (11β)pregna-1,4-diene-3,11,20-trione, 17,21-dihydroxy-16-methyl-, (16β)2-aminobiphenyl

90-41-5

4

141-03-7 504-63-2 2568-30-1 3404-61-3 315-14-0 1539-42-0

neg neg neg 5 neg neg

434-13-9 4411-80-7 56-12-2 589-16-2 464-72-2 941-98-0 5736-89-0 41114-00-5 1196-57-2 2116-65-6 84-65-1 366-18-7 3709-18-0 106-52-5

neg tox neg neg neg 5 neg neg neg 4 neg neg neg neg

chemicals predicted negative 629-99-2 neg butanedioic acid, dibutyl ester 621-01-2 neg 1,3-propanediol 93-76-5 neg 1,3-dioxolane, 2-(chloromethyl)10471-28-0 neg 1-hexene, 3-methyl13957-31-8 tox benzene, 1,3,5-trifluoro-2-nitro506-30-9 neg 2-pyridinemethanamine, N-(2-pyridinylmethyl)628-63-7 neg cholan-24-oic acid, 3-hydroxy-, (3R,5β)94-96-2 neg 2,2′-bipyridine, 6,6′-dimethyl350-40-3 4 butanoic acid, 4-amino625-98-9 neg benzenamine, 4-ethyl122-79-2 neg 1,2-ethanediol, 1,1,2,2-tetraphenyl708-64-5 neg ethanone, 1-(1-naphthalenyl)108-18-9 neg ethanone, 1-(4-butoxyphenyl)92-92-2 neg pentadecanoic acid, ethyl ester 626-55-1 neg 2(1H)-quinoxalinone 618-80-4 neg pyridine, 4-(phenylmethyl)104-20-1 neg 9,10-anthracenedione 873-74-5 neg 2,2′-bipyridine 3650-65-5 neg 1,3-dioxane-4,6-dione, 2,2,5-trimethyl1022-22-6 neg 4-piperidinol, 1-methyl760-20-3 541-73-1 930-28-9 95-95-4 50-30-6 80789-69-1

neg neg neg neg neg neg

benzenesulphonamide, 4-nitro1,3,5-triazine-2,4-diamine, 6-methylbenzenemethanol, 3-amino-R-methylbenzene, (2-methylpropyl)D-talose 2-propenoic acid, hexyl ester

6325-93-5 542-02-9 2454-37-7 538-93-2 2595-98-4 2499-95-8

neg neg neg neg neg neg

95-94-3 589-08-2 104-90-5 156-83-2 1122-72-1 4704-31-8

neg neg neg neg neg neg

589-91-3 3663-82-9 328-70-1 509-09-1 2350-89-2 2550-73-4

neg 5 neg tox neg neg

methanone, phenyl-2-pyridinyl-, oxime cyclohexanecarboxylic acid, 2-oxo-, ethyl ester 1-butanone, 4-chloro-1-(4-methoxyphenyl)-

1826-28-4 1655-07-8

neg neg

cyclohexanol, 4-methyl1,4-benzodioxin-2-methanol, 2,3-dihydrobenzene, 1-bromo-3,5-bis(trifluoromethyl)propanoic acid, pentafluoro-, silver(1+) salt 1,1′-biphenyl, 4-ethenylbenzo[1,2-c:4,5-c′]dipyrrole-1,3,5,7(2H,6H)tetrone benzene, 1-methoxy-4-(2-propenyl)5-quinolinamine

140-67-0 611-34-7

neg neg

40877-19-8

neg

132-18-3

neg

stigmast-5-en-3-ol, (3β)benzenebutanoic acid, 4-nitrodecanedioic acid, bis(2-ethylhexyl) ester 2-propenamide, 3-phenylbenzene, 1-fluoro-2-nitrodecanoic acid, ethyl ester propanoic acid, 2-hydroxy-, ethyl ester 1,4-cyclohexanedione

83-46-5 5600-62-4 122-62-3 621-79-4 1493-27-2 110-38-3 97-64-3 637-88-7

neg neg neg neg neg neg neg neg

4706-81-4 623-05-2 88-88-0 1125-88-8 321-28-8 137-32-6 575-38-2 632-51-9

neg neg neg 3 neg neg neg neg

ethanone, 1-(2,4,6-trihydroxyphenyl)1,4-hexadiene, (Z)-

480-66-0 7318-67-4

neg neg

piperidine, 4-(diphenylmethoxy)-1-methyl-, hydrochloride 2-tetradecanol benzenemethanol, 4-hydroxybenzene, 2-chloro-1,3,5-trinitrobenzene, (dimethoxymethyl)benzene, 1-fluoro-2-methoxy1-butanol, 2-methyl1,7-naphthalenediol benzene, 1,1′,1′′,1′′′-(1,2-ethenediylidene) tetrakis1,3-pentanediol, 2,2,4-trimethylphenol, 3-(diethylamino)-

144-19-4 91-68-9

neg tox

a The name and CAS number of each chemical as well as the classification of the chemical as positive, negative, or toxic is shown. Chemicals were tested at 1, 3, 10, and 30 µM except for the more potent ones that were diluted by the factors given in parentheses after the chemical name. Potency codes for the positive chemicals were as follows: 1, the IC25 e 0.1 µM; 2, 0.1 µM < IC25 e 0.3 µM; 3, 0.3 µM < IC25 e 1 µM; 4, 1 µM < IC25 e 3 µM; 5, 3 µM < IC25 e 10 µM; neg, negative chemicals; and tox, cytotoxic chemicals. Noncytotoxic chemicals belonging to groups 1-5 were classified as positive for AR antagonism.

Statistical Analyses. The internal validation was performed using leave-groups-out (LGO) cross-validation (45). In LGO, part (e.g., 10, 20, or 50%) of the substances in the training set was removed at random and a new model was developed for

the remaining substances. This new model did not in any way use the information from the removed substances. This process was repeated a predefined number of times. In this study, 50% cross-validation was used, as a sample of 261 chemicals was

820

Chem. Res. Toxicol., Vol. 21, No. 4, 2008

randomly taken from the training set, and a new model was constructed using only this subset as training data. The balance between positives and negatives was kept similar to the one in the full training set. The new model was then used to predict the remaining 262 chemicals from the initial training set. After this, the procedure was reversed, that is, using the former training set as the test set. This procedure was repeated five times, and all results were pooled and used as the basis for calculating overall sensitivity, specificity, and concordance.

Results and Discussion Data on AR Activity. To our knowledge, there are three reports on the screening of AR activity of a large number of chemicals. Fang et al. (34) used an AR binding assay to investigate 202 natural, synthetic, and environmental chemicals, and structure-activity relationships were analyzed. A total of 146 chemicals out of 202 were found to be AR binders, and out of them, 14 chemicals were strong binders, all of which were steroids. It should be noted that no distinction between agonism and antagonism could be made in this study. Kojima et al. (13) investigated 200 pesticides in an AR reporter gene assay that is comparable to the one used in our laboratory. The pesticides were tested for agonism as well as antagonism. No chemicals were found agonizing AR, whereas 66 out of 200 pesticides exhibited AR antagonism. Finally, Araki et al. (46) screened 253 industrial chemicals for AR agonism as well as antagonism, and they identified two agonists and nine antagonists. These chemicals were essentially randomly selected, although they were chosen in part to generate a training set for development of a QSAR model. The present study is the first to present AR transactivation data for 397 chemicals of various structures and for varying uses. The chemicals were selected and tested in two separate phases: First of all, 295 chemicals were tested to generate data for the training set, and second, 102 chemicals were tested as part of an external validation. The AR antagonism data for the 295 chemicals that were included in the training set are shown in Table 1. It should be noted that the selection of chemicals for the training set was biased, as most AR antagonizing chemicalsswe were aware ofswere included in the test, if possible. Furthermore, halfway during testing, a preliminary QSAR model was developed and applied to select some of the next chemicals to be tested with the purpose of strengthening the model or broadening its domain. The chemicals were divided into the following chemical classes: natural hormones (seven chemicals, 71% positive), synthetic hormones and drugs (30 chemicals, 53% positive), pesticides (68 chemicals, 56% positive), PCBs (39 chemicals, 74% positive), PAHs (35 chemicals, 54% positive), brominated flame retardants (six chemicals, 83% positive), roast mutagens (nine chemicals, no positives), plastizicers and plast additives (13 chemicals, 31% positive), food additives and cosmetics (16 chemicals, 25% positive), antioxidants (10 chemicals, 20% positive), plant compounds (six chemicals, no positives), and a miscellaneous group of chemicals (56 chemicals, 23% positive). The numbers in parentheses refer to the total number of tested chemicals in each class and the percentage of positive compounds. An overview of the total number of chemicals tested in each group together with the distribution of active chemicals is given in Figure 1. In total, 45.7% of the chemicals were positive AR antagonists according to our criteria, 48.5% of the chemicals were negative, and 5.8% of all chemicals were cytotoxic. A total of 4.4% of all tested chemicals were relatively potent with an IC25 e 0.3 µM (belonging to potency groups 1 and 2).

Vinggaard et al. Table 4. External Validation of the QSAR Modela sensitivity (%)

specificity (%)

concordance (%)

57.1 (8/14)

97.6 (80/82)

91.7 (88/96)

a

A total of 102 chemicals were selected and tested for AR antagonism in vitro for an external validation of the QSAR model.

Table 5. QSAR Model Run on Selected Lists of Chemicals

chemical list total set of chemicalsa b

EINECS chemicals

n % n %

chemical no.

chemicals within the domain

chemicals within the domain with AR antagonistic effects

176103 100 47000 100

82308 47 26245 56

6816 8.3 2561 9.8

a Chemicals in the National Food Institute QSAR database. See the Materials and Methods for further information. b Only organic chemicals with unequivocal chemical names and CAS numbers were included.

They were either natural hormones (e.g., progesterone and estradiol), synthetic hormones and drugs (bicalutamide, hydroxyflutamide, mifepristone, spironolactone, and nilutamide), pesticides (vinclozolin and its metabolites M1 or M2, and fenitrothion), or organochlorines (1,2,3,6,7,8-HxCDD and 2,3,4,7,8-PeCDF). The positive PAHs, brominated flame retardants, plasticizers and plastic additives, and food additives and cosmetics all had IC25 > 0.3 µM. Among the most potent PAHs were benzo[k]fluoranthene, 1-nitro-pyrene, and retene with an IC25 between 0.3 and 1 µM. Among the positive plastizicers and plast additives were bisphenol A, bisphenol A dimetacrylate, bisphenol F, and 4-tertoctylphenol, which had IC25 values between 0.3 and 3 µM. No plant compounds or roast mutagens were detected positive. Development and Validation of the QSAR Model. The MultiCASE expert system was used to construct a QSAR model for screening of chemicals with AR antagonizing potential. The ability of the MultiCASE expert system to handle large databases with diverse chemical structures and to rapidly identify structural features responsible for activity makes it suitable for QSAR analysis of our data. Our own AR reporter gene data for 292 chemicals (Table 1), and an additional data set for 231 chemicals from the literature, were used as inputs to train the expert system. The training data set covered inactive, weak, as well as very powerful AR antagonists and represented a variety of chemical compounds. A “5 times, 2-fold, 50% crossvalidation” of the model showed a sensitivity of 64%, a specificity of 84%, and a concordance of 76% (Table 2). Experimental data for the 102 chemicals were generated in our laboratory for an external validation of the QSAR model (Table 3). The ratio between positives and negatives was intended to be 10:90 to approximately reflect the “true” ratio of positives/negatives predicted by the QSAR model (see Table 5). Among the 12 chemicals predicted positive by the QSAR model, eight chemicals were positive, two were toxic, and two were negative. Among the 90 chemicals predicted negative by the QSAR model, 80 chemicals were negative, four were toxic, and six were positive. The validation resulted in a sensitivity of 57% [8/(8 + 6) ) 0.57], a specificity of 98% [80/(80 + 2) ) 0.98], and a concordance of 92% [88/(82 + 14) ) 0.92] (Table 4). According to Devillers et al. (47), the majority of the existing endocrine disruptor QSAR models has not been properly validated, as external validations are usually not performed; rather, cross-validations have been commonly performed. However, cross-validation may in some cases be sufficient. In

QSAR Model for Antiandrogenic Effect

Chem. Res. Toxicol., Vol. 21, No. 4, 2008 821 Table 6. Significant Biophores in the Training Set

biophore no.

fragmenta

compounds total

compounds inactive

compounds active

1 2 7 12 13 15 18 20 21 26 42 44 45 47 51 52 53 54 61

c )cH -cH )c -c ) c )cH -c