1 2 Applicability of a keratinocyte gene signature to predict skin sensitizing potential 3 Jochem W. van der Veen a,b,⇑ Q1 , Tessa E. Pronk a,b , Henk van Loveren a,b , Janine Ezendam b 4 a Department of Toxicogenomics, Maastricht University, P.O. Box 616, NL-6200 MD Maastricht, The Netherlands 5 b Laboratory for Health Protection Research (GBO), National Institute for Public Health and the Environment (RIVM), P.O. Box 1, NL-3720BA Bilthoven, The Netherlands 6 7 9 article info 10 Article history: 11 Received 21 May 2012 12 Accepted 17 August 2012 13 Available online xxxx 14 Keywords: 15 Skin sensitization 16 Alternative methods 17 In vitro 18 Keratinocytes 19 Toxicogenomics 20 21 abstract 22 There is a need to replace animal tests for the identification of skin sensitizers and currently many alter- 23 native assays are being developed that have very promising results. In this study a gene signature capable 24 of very accurate identification of sensitizers was established in the HaCaT human keratinocyte cell line. 25 This signature was evaluated in a separate study using six chemicals that are either local lymph node 26 (LLNA) false-positive or false-negative chemicals in addition to nine sensitizers and four non-sensitizers. 27 Similar studies do not apply these more difficult to classify chemicals, which show the true potential for 28 human predictions of an assay. Although the gene signature has improved prediction accuracy compared 29 to the LLNA, the misclassified compounds were comparable between the two assays. Gene profiling also 30 showed a sensitizer specific response of the Nrf2-keap1 and Toll-like receptor signaling pathways. After 31 exposure to non-sensitizing chemicals that induce either of the pathways the signature misclassified all 32 Nrf2-inducers, while the Toll-like receptor ligands were correctly classified. In conclusion, we confirm 33 that keratinocyte based prediction assays may provide essential information on the properties of com- 34 pounds. Furthermore, chemical selection is critical for assessment of the performance of in vitro alterna- 35 tive assays. 36 Ó 2012 Elsevier Ltd. All rights reserved. 37 38 39 1. Introduction 40 Allergic contact dermatitis is a delayed-type IV hypersensitivity 41 reaction that can be induced after skin contact with chemical hap- 42 tens. It is a common occupational and consumer health problem 43 which develops through a series of immunological events caused 44 by repeated contact with compounds that have skin sensitizing po- 45 tential (Kimber et al., 2002). The current methods for assessing the 46 sensitizing potential of chemicals are the Local Lymph Node Assay 47 (LLNA) or guinea pig tests (Guinea Pig Maximization Test (GPMT) 48 or Buehler test) (Gerberick et al., 2007; Kimber et al., 1994). There 49 is great demand for validated non-animal alternatives to replace 50 these animal tests, due to the ban on animal testing described in 51 the 7th amendment to the European Union Cosmetics Directive. 52 In addition, the REACH (Registration, Evaluation, and Authorization 53 of Chemicals) regulation requires that the safety of a large amount 54 of chemicals has to be assessed and it stimulates the use of alter- 55 native test methods. These legislative changes, combined with eth- 56 ical issues and societal acceptance towards animal use in toxicity 57 testing, drive further development of alternative test methods. 58 Although much progress has been made for assessing skin sensitiz- 59 ing potential, no alternative test methods have been validated yet. 60 In recent years, it has become clear that a combination of methods 61 in a testing strategy will be required for correct identification of 62 sensitizers, rather than a single test (Vandebriel and van Loveren, 63 2010). 64 Many of current cell based alternative assays for skin sensitiza- 65 tion use either keratinocytes or dendritic cells. Read-outs are either 66 changes in gene regulation in these cells (Arkusz et al., 2010; 67 Johansson et al., 2011; Natsch, 2009; Vandebriel et al., 2010), pro- 68 duction of cytokines, such as IL-18 in keratinocytes (Corsini et al., 69 2009) or upregulation of cell surface markers, including CD86 70 and CD54 on dendritic cells (Aeby et al., 2004; Ashikaga et al., 71 2006; Sakaguchi et al., 2006; Schreiner et al., 2008) exposed to sen- 72 sitizers. The prediction accuracy of these assays range between 71% 73 and 99% (Bauch et al., 2011). 74 In the present study the focus is on the predictive power and 75 the driving pathways involved in the initial response of keratino- 76 cytes (KCs). KCs are abundantly present in the skin and play an 77 important role in the initial stages of skin sensitization as they 78 are the first cells to come into contact with chemicals. In addition, 79 KCs are able to secrete several pro-inflammatory mediators and 80 metabolize prohaptens into protein-reactive haptens (Jowsey 81 et al., 2006; Martin et al., 2011; Vandebriel and van Loveren, 82 2010). More recently, it has been proposed that KCs generate ‘dan- 83 ger’ signals in response to skin sensitizers that trigger the innate 84 immune system through TLR activation (Martin et al., 2011; 0887-2333/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tiv.2012.08.023 ⇑ Corresponding author at: Q2 Laboratory for Health Protection Research (GBO), National Institute for Public Health and the Environment (RIVM), P.O. Box 1, NL- 3720BA Bilthoven, The Netherlands. E-mail addresses: [email protected](J.W. van der Veen), Tessa [email protected](T.E. Pronk), [email protected](H. van Loveren), Janine [email protected](J. Ezendam). Toxicology in Vitro xxx (2012) xxx–xxx Contents lists available at SciVerse ScienceDirect Toxicology in Vitro journal homepage: www.elsevier.com/locate/toxinvit TIV 2934 No. of Pages 9, Model 5G 31 August 2012 Please cite this article in press as: van der Veen, J.W., et al. Applicability of a keratinocyte gene signature to predict skin sensitizing potential. Toxicol. in Vitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
Transcript
1
2
3 Q1
45
67
9
10111213
14151617181920
2 1
3839
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
Q2
Toxicology in Vitro xxx (2012) xxx–xxx
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
Contents lists available at SciVerse ScienceDirect
Toxicology in Vitro
journal homepage: www.elsevier .com/locate / toxinvi t
Applicability of a keratinocyte gene signature to predict skin sensitizing potential
Jochem W. van der Veen a,b,⇑, Tessa E. Pronk a,b, Henk van Loveren a,b, Janine Ezendam b
a Department of Toxicogenomics, Maastricht University, P.O. Box 616, NL-6200 MD Maastricht, The Netherlandsb Laboratory for Health Protection Research (GBO), National Institute for Public Health and the Environment (RIVM), P.O. Box 1, NL-3720BA Bilthoven, The Netherlands
2223242526272829303132
a r t i c l e i n f o
Article history:Received 21 May 2012Accepted 17 August 2012Available online xxxx
Please cite this article in press as: van der VeenVitro (2012), http://dx.doi.org/10.1016/j.tiv.201
a b s t r a c t
There is a need to replace animal tests for the identification of skin sensitizers and currently many alter-native assays are being developed that have very promising results. In this study a gene signature capableof very accurate identification of sensitizers was established in the HaCaT human keratinocyte cell line.This signature was evaluated in a separate study using six chemicals that are either local lymph node(LLNA) false-positive or false-negative chemicals in addition to nine sensitizers and four non-sensitizers.Similar studies do not apply these more difficult to classify chemicals, which show the true potential forhuman predictions of an assay. Although the gene signature has improved prediction accuracy comparedto the LLNA, the misclassified compounds were comparable between the two assays. Gene profiling alsoshowed a sensitizer specific response of the Nrf2-keap1 and Toll-like receptor signaling pathways. Afterexposure to non-sensitizing chemicals that induce either of the pathways the signature misclassified allNrf2-inducers, while the Toll-like receptor ligands were correctly classified. In conclusion, we confirmthat keratinocyte based prediction assays may provide essential information on the properties of com-pounds. Furthermore, chemical selection is critical for assessment of the performance of in vitro alterna-tive assays.
� 2012 Elsevier Ltd. All rights reserved.
37
58
1. Introduction Although much progress has been made for assessing skin sensitiz- 59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
Allergic contact dermatitis is a delayed-type IV hypersensitivityreaction that can be induced after skin contact with chemical hap-tens. It is a common occupational and consumer health problemwhich develops through a series of immunological events causedby repeated contact with compounds that have skin sensitizing po-tential (Kimber et al., 2002). The current methods for assessing thesensitizing potential of chemicals are the Local Lymph Node Assay(LLNA) or guinea pig tests (Guinea Pig Maximization Test (GPMT)or Buehler test) (Gerberick et al., 2007; Kimber et al., 1994). Thereis great demand for validated non-animal alternatives to replacethese animal tests, due to the ban on animal testing described inthe 7th amendment to the European Union Cosmetics Directive.In addition, the REACH (Registration, Evaluation, and Authorizationof Chemicals) regulation requires that the safety of a large amountof chemicals has to be assessed and it stimulates the use of alter-native test methods. These legislative changes, combined with eth-ical issues and societal acceptance towards animal use in toxicitytesting, drive further development of alternative test methods.
78
79
80
81
82
83
84
ll rights reserved.
Protection Research (GBO),ent (RIVM), P.O. Box 1, NL-
(J.W. van der Veen), Tessa.nl (H. van Loveren), Janine
, J.W., et al. Applicability of a k2.08.023
ing potential, no alternative test methods have been validated yet.In recent years, it has become clear that a combination of methodsin a testing strategy will be required for correct identification ofsensitizers, rather than a single test (Vandebriel and van Loveren,2010).
Many of current cell based alternative assays for skin sensitiza-tion use either keratinocytes or dendritic cells. Read-outs are eitherchanges in gene regulation in these cells (Arkusz et al., 2010;Johansson et al., 2011; Natsch, 2009; Vandebriel et al., 2010), pro-duction of cytokines, such as IL-18 in keratinocytes (Corsini et al.,2009) or upregulation of cell surface markers, including CD86and CD54 on dendritic cells (Aeby et al., 2004; Ashikaga et al.,2006; Sakaguchi et al., 2006; Schreiner et al., 2008) exposed to sen-sitizers. The prediction accuracy of these assays range between 71%and 99% (Bauch et al., 2011).
In the present study the focus is on the predictive power andthe driving pathways involved in the initial response of keratino-cytes (KCs). KCs are abundantly present in the skin and play animportant role in the initial stages of skin sensitization as theyare the first cells to come into contact with chemicals. In addition,KCs are able to secrete several pro-inflammatory mediators andmetabolize prohaptens into protein-reactive haptens (Jowseyet al., 2006; Martin et al., 2011; Vandebriel and van Loveren,2010). More recently, it has been proposed that KCs generate ‘dan-ger’ signals in response to skin sensitizers that trigger the innateimmune system through TLR activation (Martin et al., 2011;
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
2 J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
Natsch, 2009). The signaling cascade following TLR activation leadsto MAPK signaling and NF-jb activation, which induces the releaseof pro-inflammatory mediators (Kumar et al., 2009). Gene profilingstudies in KC indicate a role for the cytoprotective Nrf2-Keap1pathway in the response to skin sensitizers (Natsch, 2009; Vande-briel et al., 2010). This pathway is involved in antioxidant responsesignaling and antioxidant response genes such as hemeoxygenase1 (HMOX1) and NADPH quinone oxidoreductase 1 (NQO1) havebeen shown to be under Nrf2 control (Martin et al., 2011; Vande-briel et al., 2010; Vandebriel and van Loveren, 2010).
In an earlier array study, we found biologically relevant path-ways regulated by skin sensitizers in KCs. In addition, sensitizerscould be distinguished from non-sensitizers based on gene regula-tion patterns with 70% accuracy (Vandebriel et al., 2010). To con-firm these findings we have performed a new gene profilingstudy with more statistical power through an increased numberof chemicals. In addition, the accuracy of a gene signature obtainedfrom this study was thoroughly tested in novel approach thatincluded a relatively high number of either false-negative orfalse-positive chemicals from the LLNA. In addition, as a proof ofconcept, the performance of the gene signature was challengedby including chemicals that activate the Nrf2-Keap1 or TLR path-ways yet are not sensitizers.
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
2. Materials and methods
2.1. Cell culture
The human keratinocyte cell line HaCaT (Boukamp et al., 1988)was purchased from Cell Lines Service (Eppelheim, Germany). Cellswere grown in culture flasks to 80% confluence in Dulbecco’s mod-ified Eagle’s medium, supplemented with 1% nonessential aminoacids, 100 U/ml penicillin, 100 lg/ml streptomycin (all from Gibco,Breda, the Netherlands), and 10% heat-inactivated Fetal Calf Serum(Integro, Zaandam, the Netherlands) (complete medium), at 37 �Cin a humidified atmosphere of 5% CO2 in air. For passaging the cellswere washed twice with PBS and then trypsinized (0.05% Trypsinwith EDTA 4Na; Gibco). New culture flasks were seeded in com-plete medium with 1/3rd or 1/6th of the total number of cells ofthe previous passage.
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
2.2. Chemical exposure
Before exposure trypsinized cells were resuspended in freshcomplete medium to a concentration of 3 � 105 cells/ml. The cellsuspension was seeded into 12-well plates (1.5 ml per well; Grein-er, Alphen aan den Rijn, the Netherlands). The cells were allowed toadhere and form a monolayer during 24 h, after which the wellswere washed with PBS and exposed to the different chemicals incomplete medium.
The chemicals that were used in this study are shown in Tables1 and 2. All compounds were obtained from Sigma–Aldrich (Zwijn-drecht, the Netherlands), except for 2-mercaptobenzothiazole,which was obtained from Merck (Schiphol-Rijk, the Netherlands).Chemicals were dissolved in complete medium, absolute ethanol,or dimethylsulfoxide (DMSO). In addition to the compound, thevehicle (DMSO or ethanol) was added to a final concentration of1% to the cell cultures exposed to compounds dissolved in com-plete medium. Cells were exposed to ethanol and DMSO to obtainvehicle control samples. For each chemical the concentrationresulting in 80% viability (CV80) was determined using colorimet-ric measurement of WST-1 cleavage. To this end, HaCaT cells wereseeded in 12-well plates (4.5 � 105 cells/well) and incubated witha concentration range of the chemicals in duplicate or solvent con-trol in triplicate at 37 �C in a humidified atmosphere of 5% CO2 in
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
air. After 21 h of exposure, 1.2 mL medium was removed fromthe wells, leaving 300 lL exposure medium in the wells, and40 lL/well WST-1 (Roche, Woerden, the Netherlands) was added.After 4 h, 100 lL was transferred to a microtiter plate and WST-1cleavage was quantified at 450 nm using a microplate reader(Spectramax 190, Molecular devices, Wokingham, UK). After blankcorrection, WST-1 without cells, the mean optical densities of thereplicates were compared to the mean of the corresponding vehiclecontrols in order to calculate relative viability (data not shown). Ina single experiment the HaCaT cells were exposed to each chemicalin four replicates for 4 h, which was determined to be the optimalexposure time for classification as this exposure period had higherprediction accuracy compared to 8 h exposure (Vandebriel et al.,2010).
For the Toll-like receptor ligands it proved impossible to deter-mine a CV80 value. The applied concentration induced the same le-vel of IL-8 in the supernatant after 24 h of exposure as did thestrong sensitizer benzoquinone, IL-8 was measured using ELISAaccording to manufacturer’s instructions (eBioscience, Vienna,Austria). For the Nrf2 activators the concentration was based onthe CV80. Additionally, the ability to induce HMOX1 was assessedby ELISA (R&D systems, Abingdon, UK) for the selectedconcentration.
2.3. RNA isolation
At the end of the exposure period 400 lL RNAprotect cell re-agent (Qiagen, Westburg, the Netherlands) was added to each well.Cells were resuspended and were stored at �80 �C until furtheranalysis. For RNA isolation the cells were lysed after removing theRNAprotect using Qiazol and the lysates were homogenized usingQiashredder columns. RNA was isolated by using miRNeasy MiniKit in combination with RNeasy MinElute Cleanup Kit (all from Qia-gen) according to the manufacturer’s instructions. RNA quantitywas spectrophotometrically assessed (Nanodrop Technologies,Wilmington, DE), and integrity was determined by automated gelelectrophoresis (Bioanalyzer 2000; Agilent technologies, Amstelv-een, the Netherlands). For each compound three replicates were se-lected out of the four RNA samples for DNA microarray analysis,based on concentration and RNA Integrity Number. Control RNAsamples from ethanol and DMSO were included in analysis.
2.4. DNA microarray and data analysis
The samples were prepared, hybridized to Affymetrix HT HG-U133 + 2.0 PM arrays and measured by ServiceXS (Leiden, theNetherlands). The quality of the raw data was checked using RMA-Express (Bolstad et al., 2003), which was also used in combinationwith the BrainArray CustomCDF version 13 for the annotation of18.040 genes. The expression values were then log2 transformedand corrected for the corresponding vehicle control. The raw datais accessible at Array Express (http://www.ebi.ac.uk/arrayex-press/) under the accession number 943-MTAB-E.
2.5. Identification of significant genes
To detect if genes significantly changed between samples of thesensitizers and non-sensitizers, a t-test was done on the controlcorrected samples of those respective classes. p-Values were falsediscovery rate (FDR) corrected (Benjamini and Hochberg, 1995),an FDR below 0.05 was considered significant. In addition to thecomparison between all sensitizing compounds and all irritatingcompounds, this approach was applied to the potency subsets ofsensitizers, as defined by the Globally Harmonized System (GHS).Strong sensitizers (EC3 6 2) or other than strong sensitizers(EC3 > 2) were independently compared to all non-sensitizers.
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
Table 1Sensitizers and non-sensitizers used for biomarker identification. The HaCaT cell line was exposed for 4 h to the CV80 concentrations of these compounds. Chemicals in bolt wereused in both gene signature identification in the DNA microarray and the confirmation experiment using RT-PCR, while chemicals with an asterisk.
Name Abrreviation CAS-nr Class (LLNA) EC3 (%) CV80 (lg/mL)
5-Chloro-2-methyl-4-isothiazolin-3-one CMI 26172-55-4 Sensitizer 0.009 149.6Diphenylcyclopropenone DCP 886-38-4 Sensitizer 0.003 2.1Oxazolonea OXA 15646-46-5 Sensitizer 0.003 217.2p-Benzoquinone BQ 106-51-4 Sensitizer 0.0099 9.71-Chloro-2,4-dinitrobenzene DNCB 97-00-7 Sensitizer 0.04 4.1p-Phenylenediamine PPD 106-50-3 Sensitizer 0.11 162.2Cobalt(II) chloride CCl 7646-79-9 Sensitizer 0.6 39.0Isoeugenol IsoEug 97-54-1 Sensitizer 1.5 88.72-Mercaptobenzothiazole 2-MBT 149-30-4 Sensitizer 1.7 167.3Tetramethylthiuram disulfidea TMTD 137-26-8 Sensitizer 5.2 48.1Resorcinola RES 108-46-3 Sensitizer 5.5 55.1Diethyl maleate DM 141-05-9 Sensitizer 5.8 43.0Citral Ci 5392-40-5 Sensitizer 9.2 34.3a-Hexylcinnamaldehyde HCA 101-86-0 Sensitizer 9.7 30.3Eugenol EU 97-53-0 Sensitizer 10.1 164.2Cinnamyl alcohol CA 104-54-1 Sensitizer 21 228.1Benzocainea BENZ 94-09-7 Sensitizer 22 470.8Hydroxycitronellala HC 107-75-5 Sensitizer 23 689.0Imidazolidinyl urea IU 39236-46-9 Sensitizer 24 38.8Ethylene glycol dimethacrylate EGDM 97-90-5 Sensitizer 28 79.3Ethyl acrylate EA 140-88-5 Sensitizer 28 40.0Methyl methacrylate MMA 80-62-6 Sensitizer 90 400.5Tert-butylhydroquinonea tBHQ 1948-33-0 Sensitizer NA 33.2Nickel(II) chloride NiCl 7718-54-9 False-negative NA 155.5Triisobutylphosphatea TIBP 126-71-6 False-negative NA 186.4Chlorobenzene DCB 108-90-7 Non-sensitizer NA 225.12-Propanol Iso 67-63-0 Non-sensitizer NA 15.0Lactic acid LA 50-21-5 Non-sensitizer NA 3.6Methyl salicylate MS 119-36-8 Non-sensitizer NA 426.0Salicyclic acid SA 69-72-7 Non-sensitizer NA 345.3Hexane Hex 110-54-3 Non-sensitizer NA 3.9Dextrana DEX 9004-54-0 Non-sensitizer NA 2000.0Propylene glycola PG 57-55-6 Non-sensitizer NA 304.4Tween80a T80 9005-65-6 Non-sensitizer NA 1000.0Xylenes Xy 1330-20-7 False-positive 95.8 21.2Benzalkonium chloridea BK 63449-41-2 False-positive 0.07 6.0Sodium dodecyl sulfate SDS 151-21-3 False-positive 4.0 144.2Maleic acida MA 110-16-7 False-positive 7.0 464.3Hexaethylene glycol monotetradocyl ethera HEG 5168-89-8 False-positive NA 169.0
a Only used in the second experiment using RT-PCR.
Table 2Pathway specific compounds used in RT-PCR gene list validation. For evaluation of the effects of pathway specific compounds of the gene list, the HaCaT cell line was exposed for4 h to the CV80 concentrations of these non-sensitizing compounds. For the toll-like receptor ligands it proved impossible to determine a CV80 value, the exposure concentrationwas based on the levels of IL-8 in the supernatant after 24 h exposure (CIL8).
Name Abrreviation CAS-nr Class EC3 (%) CV80 (lg/mL) CIL8 (lg/mL)
Z-Leu-Leu-Leu-al MG132 133407-82-6 Nrf2 NA 4.8Hydrogen peroxide H2O2 7722-84-1 Nrf2 NA 34.0Sodium arsenite Ars 7784-46-5 Nrf2 NA 13.0
Lipopolysaccharide e. coli 055:B5 LPS ec NA TLR NA 5.0Lipopolysaccharide Salmonella enterica LPS se NA TLR NA 5.0Peptidoglycan S. aureus PGN NA TLR NA 12.5Poly I:C Poly IC 31852-29-6 TLR NA 20.0
J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx 3
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
2.6. Compound classification
Classification was performed using the environment for statisti-cal computing of R (R Development Core Team, 2010). Classifica-tion was done in a leave-one-compound-out cross-validationapproach. In this approach one compound is used as test com-pound, while the others function as training set (Vandebrielet al., 2010). The training set is used to classify the test compoundand this is repeated until each compound has been classified. Eachclassifier algorithm builds a classifier in a different way; thereforethree different approaches were applied to the data. The Random
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
forests (RF) (Breiman, 2001) is based on the creation of predictiontrees. The support vector machine (SVM) (Rifkin et al., 2003), usingthe radial kernel on scaled data, creates a separating hyper plane.Lastly, the Prediction analysis for Microarrays in R (PAM-R) usesshrunken centroids to classify samples (Tibshirani et al., 2001).
To reduce the algorithm dependent class prediction, the out-comes of the three classifier algorithms were combined. A predic-tion was generated for each sample using the algorithms,generating three predictions per sample. The classification of acompound was based on the prediction of the triplicate samples,with a total of nine predictions. Majority voting was used when
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
4 J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
the prediction for the replicate samples was not identical; the com-pound is classified according to the prediction of the majority ofthe samples.
2.7. Gene signature identification
Biomarkers for further validation were selected based on themost predictive genes in each of the three classifier algorithmsfor every leave-one-compound-out cross-validation. In this meth-od each algorithm generates 26 lists of predictive genes, one foreach compound. Although these gene lists differ for each com-pound left out, overlap in the genes exists. The overlap can be usedto identify the most predictive genes for each algorithm. Genesthat occur in multiple lists can be considered the most robust.Within the algorithms, a gene was considered to be robust whenit was present in at least 13 of the gene lists. Now that the mostcommon genes for an algorithm were selected, the three algorithmgene sets were compared. Genes present in two of the three listswere selected. The resulting gene signature comprises the mostcommon predictive genes across algorithms.
2.8. Real-time PCR analysis and classification
After 4 h exposure to the compounds listed in Tables 1 and 2400 lL RNAprotect cell reagent (Qiagen, Westburg, the Nether-lands) was added to each well. Cells were resuspended and storedat �80 �C until further use. For RNA isolation the cells were lysedafter removing the RNAprotect. RNA was isolated using the RNeasykit with DNase (Qiagen) according to the manufacturer’s instruc-tions. The samples were analyzed using RT-PCR assays (AppliedBiosystems) for the biomarker gene list and housekeeping genes(HPRT1, POLR2A and TBP) by ServiceXS (Leiden, the Netherlands).The gene expression data was log2 transformed and normalizedagainst the housekeeping genes.
The RF, SVM and PAM-R classifier algorithms were used forclass prediction of the compounds. The data obtained from themicroarray study was used for training the classifiers, while theRT-PCR data served as a test set. For final prediction the combina-tion of the three algorithms was applied as described for classifica-tion of the gene profiling results.
2.9. Pathway analysis and heatmaps
Pathway analysis was performed using ToxProfiler (Boorsmaet al., 2005) with all microarray data as input. The WikiPathwaysand BioCarta databases were used for reference. The selected path-ways are significantly affected (E-value < 0.05) by sensitizers, butnot by non-sensitizers. The Multiple Array Viewer program wasused to generate supervised heatmaps.
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
3. Results
3.1. Significantly regulated genes in the array
The differences in gene expression in HaCaT cells exposed tosensitizers and non-sensitizers were evaluated. The comparisonof sensitizers to non-sensitizers showed that 166 genes were dif-ferentially regulated. The most significantly upregulated geneswere HMOX1, ZNF805, ADM and FosL1, while DNMT3b was down-regulated. The number of significantly affected genes differed be-tween the strong sensitizers (EC3 values 6 2) and less potentsensitizers (EC3 value > 2). Strong sensitizers significantly affected661 genes, whereas less potent sensitizers affected only 76 genes.There was an overlap of 45 genes between these two potencyclasses.
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
3.2. Pathways significantly regulated by sensitizers
The array data were used to identify important pathways in-duced by skin sensitizers in keratinocytes using the ToxProfilerpathway analysis tool. This showed that the Nrf2-Keap1 pathway,involved in antioxidant response, was significantly regulated bysensitizers. Also the MAPKinase signaling pathway and pathwaysinvolved in innate immunity, such as Toll-like receptor (TLR) sig-naling and cytokine network were significantly affected by sensi-tizers (Table 3).
3.3. Prediction accuracy of the microarray
Three different algorithms were used to assess the predictionaccuracy of the microarray, resulting in high prediction accuracyfor all approaches. The highest prediction was achieved with therandom forest (RF) approach: 96.2%. The prediction accuracieswith PAM-R and support vector machine (SVM) were 92.3% and88.5% respectively. To reduce the data set dependent predictionof the classifier algorithms, the predictions of these three algo-rithms were combined. The combined classifier algorithms classi-fied sensitizers and non-sensitizers with 96.2% accuracy (Table4). In all approaches the non-sensitizer SDS was considered to bea false positive, similar to the LLNA. In contrast, the gene list cor-rectly identified xylene, which is also a false positive in the LLNA,as a non-sensitizer. In addition, nickel, a false-negative in the LLNA,was correctly identified as a sensitizer.
3.4. Gene signature
A predictive gene signature was generated by selecting genesthat have good predictivity in each algorithm. Table 5 shows the10 genes that were retrieved with this methodology; these wereall found to be significantly regulated by sensitizers. The gene sig-nature contains HMOX1, STC2, ADM and SRD1, which play a role inthe oxidative stress response, in which the Nrf2-Keap1 pathway isalso involved. In addition, the gene signature contains genes re-lated to the inflammatory response (cFOS and FosL1).
The expression of these genes is shown in Fig. 1. The sensitizersshow a gene expression pattern that differs from the non-sensitiz-ers. Whereas most genes show an upregulation due to sensitizerexposure, DNMT3b, RBMP5 and CDK12 were downregulated.HMOX1 and cFOS are the most affected genes with the highest foldchange; averaging 15.5 fold and 11.7 fold upregulation respectively(see Supplementary data I). The gene expression profile induced bySDS is similar to that of the skin sensitizers, which explains thefalse positive classification of SDS as a sensitizer by all classifieralgorithms.
3.5. Prediction accuracy of the gene signature
To put the prediction accuracy of the gene signature to the test,HaCaT cells were exposed to a new set of sensitizers and non-sen-sitizers, including several chemicals that overlap with the micro-array experiment and some that have proven difficult to classifyin the LLNA. The expression patterns are shown in Fig. 2. Overall,the expression of these 10 genes is affected by sensitizers; whereasnon-sensitizing chemicals show limited changes in expression (seeSupplementary data I). In this experiment the gene regulationdirectionality is very similar to that observed in the microarrayexperiment (Fig. 1). It also becomes apparent that false positivecompounds in the LLNA show regulation and similar directionalityof gene expression compared to that of the sensitizers and thatfalse negative compounds show similarities to non-sensitizers.
The predictivity of the gene signature was evaluated using thegene expression data of the RT-PCR as the test set in the prediction
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
Table 3Pathways significantly regulated by sensitizers. The ToxProfiler pathway analysis tool was used to identify pathways affected by sensitizers in the WikiPathways and BioCartadatabases. The threshold for significance was set at E < 0.05. Pathways with related functions are grouped.
J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx 5
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
algorithms, while the microarray data functioned as the trainingset. It was shown that the most accurate classification wasachieved using the SVM algorithm, reaching a prediction accuracyof 78.9% (Table 6). The prediction accuracy of RF and PAM-R were57.9% and 68.4% respectively, showing that performance of analgorithm can be variable, depending on the dataset. To reduce thisdependence on classification algorithm, the combination of theclassifiers was used. This approach yielded an accuracy of 73.7%,
Table 4Prediction by classifier algorithms using full data set. The class of compounds is predicted u(a–c) or a combination of the algorithms (d). The in vivo classification by the LLNA is show
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
which is slightly higher than the LLNA (68.4%) for the tested com-pounds (Table 6e). Nickel, benzalkonium chloride, SDS, and hex-amethylene glycol monotetradocyl ether were incorrectlyclassified by the gene signature. These chemicals have been provendifficult to classify in the LLNA as well. Maleic acid and triisobutyl-phosphate, respectively false-positive and false-negative in theLLNA, were correctly classified by the gene signature as non-sensi-tizer and sensitizer, respectively.
sing the three algorithms (random forests, SVM and Pam-R) based on majority votingn in (e).
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
Table 5Gene signature. The selected genes were most the most common within each algorithm and present in two or more of the leave-one-compound-out classification approaches,using the classifier algorithms of random forests, SVM and Pam-R.
Entrez ID Gene name Function
1789 DNA (cytosine-5-)-methyltransferase 3 beta DNMT3b DNA methylation Hervouet et al. (2009)3162 Hemeoxygenase 1 HMOX1 Oxidative stress Morse and Choi (2002)8614 Stanniocalcin 2 STC2 Oxidative stress Law and Wong (2010)133 Adrenomedullin ADM Oxidative stress Katsuki et al. (2003)140,809 Sulfiredoxin 1 homolog SRD1 Oxidative stress Singh et al. (2009)2353 FBJ murine osteosarcoma viral oncogene homolog cFos Development and inflammation Zenz et al. (2008)8061 FOS-like antigen 1 FosL1 Development and inflammation Zenz et al. (2008)10,181 RNA binding motif protein 5 RBM5 Alternative splicing Kotlajich and Hertel (2008)51,755 Cyclin-dependent kinase 12 CDK12 Alternative splicing Bartkowiak et al. (2010)353,322 Ankyrin repeat domain 37 ARD37 Unknown Benita et al. (2009)
Fig. 1. Regulation of the biomarkers by compounds in the microarray. The heatmap is based on the fold change difference between chemical exposed and the correspondingvehicle control.
Fig. 2. Regulation of the biomarker gene list by compounds in RT-PCR. The heatmap is based on the fold change difference between chemical exposed and the correspondingvehicle control.
6 J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a keratinocyte gene signature to predict skin sensitizing potential. Toxicol. inVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
Table 6Prediction by classifier algorithms using RT-PCR data set. The class of compounds is predicted using the three algorithms (random forests, SVM and Pam-R) based on majorityvoting (a–c) or a combination of the algorithms (d). The in vivo classification by the LLNA is shown in (e).
J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx 7
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
3.6. Classification of chemicals that activate the key signalingpathways
To assess if classification using the gene signature is specific forsensitizing and not for non-sensitizing chemicals that activate theNrf2-Keap1 or TLR signaling pathways, a selection of such com-pounds were included. It was shown that the three Nrf2-inducerswere all incorrectly classified as sensitizers, whereas the fourTLR-ligands were correctly classified as non-sensitizers. The genesignature appears to have a bias towards Nrf2-induction (Table 7).
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
4. Discussion
We have shown that keratinocyte based prediction assays canprovide essential information on the sensitizing potential of chem-icals. In addition, gene profiling of keratinocytes exposed to sensi-tizers confirmed the importance of the previously identified Nrf2-Keap1 and MAPKinase signaling pathways (Natsch, 2009; Vande-briel et al., 2010). Furthermore, additional pathways related tothe innate immune response were identified in this study, includ-ing the TLR signaling pathway and the Cytokine and inflammatoryresponse pathway. This is in line with recent studies that haveindicated an important role for the innate immune response inthe development of allergic contact dermatitis (Klekotka et al.,2010; Martin et al., 2008, 2011). In analysis of the significantly
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
regulated genes it became apparent that strong sensitizers triggersignificantly more genes than weak sensitizers. In addition, themajority of the genes regulated by weak sensitizers were also af-fected by strong sensitizers, indicating that strong sensitizers trig-ger additional processes or pathways. A similar effect of strongersensitizers was also observed in dendritic cells (Johansson et al.,2011). Understanding these differences may bring an in vitro po-tency ranking closer, but this has to be explored in further detail.The data set was also used for classification of sensitizers, resultingin a high prediction accuracy of 96.2%, which was higher than the88.5% of the LLNA for this particular compound set (Gerbericket al., 2005). The non-sensitizer SDS was the only compound thatwas incorrectly classified, which is a false-positive in the LLNA aswell. The prehaptens isoeugenol and p-Phenylenediamine and pro-hapten eugenol were correctly classified, indicating that the kerat-inocytes have sufficient metabolic activity. Our results are in linewith prediction accuracies obtained in other in vitro test systems(Arkusz et al., 2010; Bauch et al., 2011; Corsini et al., 2009; Johans-son et al., 2011; Natsch et al., 2011; Sakaguchi et al., 2010). A genesignature was generated to further test the applicability of in vitrogene expression profiling for the identification of sensitizers.
The gene signature comprises 10 genes that can be linked to thesignificantly regulated pathways, indicating that these genes aremechanistically relevant. Four genes, including HMOX1, encodeproteins that have a role related to oxidative stress, which is alsothe main function of the Nrf2-Keap1 pathway (Motohashi and
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
Table 7Class prediction of pathway specific inducers based on combination of classifieralgorithms. The class of compounds is predicted using a combination of threealgorithms; random forests, SVM and Pam-R.
Compound Stimulates Prediction
LPS E. coli TLR4 Non-sensitizerLPS S. aureus TLR4 Non-sensitizerPoly I:C TLR3 Non-sensitizerPeptidoglycan TLR2 Non-sensitizerH2O2 SensitizerSodium arsenite Nrf2 SensitizerMG132 Sensitizer
8 J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
Yamamoto, 2004; Natsch, 2009). The cFOS and FosL1 proteins arepart of the AP-1 transcription factor, which has been linked to gen-eral stress response and TLR signaling (Zenz et al., 2008). A downregulation of DNMT3b protein has been related to an upregulationof cFOS (Hervouet et al., 2009), which is in line with our findings.Remarkably, HMOX1 was the only gene that was also present inthe gene list of Vandebriel et al. (2010). Also, the prediction accu-racy is a major improvement over the previous array study, inwhich an accuracy of 71% was achieved. This limited overlap be-tween these studies and the increase in prediction accuracy canbe explained by a number of differences between these studies.First of which is the increase in the number of compounds, whichensures that the selected genes are more specific for the class ofsensitizers. The time point is a second influential factor, the genelist in the study of Vandebriel et al. (2010) was generated usingdata from cells exposed for 4 and 8 h. Due to the dynamic natureof gene expression other genes can become important at a latertime point. The third factor is the difference in biomarker selectionprocedure as it was optimized for the three specific classificationalgorithms, whereas previously only two algorithms were used.Changes to the inclusion criteria set during selection can lead tovariations in the eventual biomarker list. However, notwithstand-ing the differences in affected genes, it should be stressed thatthe identified pathways were highly similar between the Vandebri-el et al. (2010) and the present study.
The performance was more thoroughly tested in a secondexperiment using 19 chemicals, of which 6 were overlapping withthe microarray. The number of chemicals either false-positive orfalse-negative in the LLNA was increased compared to the micro-array. The prediction accuracy of the gene signature in this exper-iment was 73.7%, which is an improvement over the LLNA for thetested compounds (68.4%). It is currently unclear why some chem-icals are misclassified in both the LLNA and this in vitro test. Cor-rect classification of these chemicals turns out to be difficult,since the gene signature misclassified 4 of these 6 chemicals. Itwould be interesting to investigate their physical-chemical proper-ties and to evaluate these compounds in other in vitro test to obtainmore insight in their effects on immune cells.
The initial gene selection procedure was evaluated in two ways.First, the classifier algorithms were run in a leave-one-compound-out approach on the microarray dataset, using only the gene signa-ture genes. This resulted in a prediction accuracy of 100% of thecombined algorithms (see Supplementary data II). Secondly, theRT-PCR results were used to train the classification algorithms,while the microarray data functioned as a test set. This approachresulted in a prediction accuracy of 96.2%, with the weak sensitizermethyl methacrylate as the only misclassified chemical (see Sup-plementary data III). Surprisingly, the previously false-positiveSDS was correctly identified in this approach. However, it is possi-ble that the approach is biased due to the fact that several chemi-cals were included in both experiments. The algorithms mighthave been trained on these chemicals increasing the chance of a
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
correct prediction. This evaluation shows that the selected genesare very accurate in predicting sensitizing potential; however, theyhave only been tested on the microarray data set.
The majority of the published studies use a selection of sensitiz-ers and non-sensitizers based on predefined lists, such as the onescomposed by the ECVAM and ICCVAM (Casati et al., 2009; ICCVAM,2008), which contain only a very limited number of chemicals thatare either false-negative or false-positive in the LLNA. For a realis-tic assessment of the prediction accuracy for human sensitizationpotential of a test method, it is essential to also include chemicalsthat are either false-positive or false-negative in the LLNA. Due tothe inclusion of the more difficult to classify compounds the accu-racy of the gene signature seems relatively low compared to simi-lar studies. However, chemicals properly classified by the LLNA arealso correctly identified using the gene signature in both themicroarray and RT-PCR studies, with the exception of EGDM inthe RT-PCR study.
The performance of the gene signature was further evaluated byincluding a limited number of chemicals was included that activateeither the Nrf2-Keap1 or the TLR signaling pathway. These wereused to test the concept that chemicals that trigger sensitizationrelated pathways might be identified as sensitizers. Human kerat-inocytes do not express TLR4, however the TLR4 ligand LPS wasalso included because work by (Kollisch et al., 2005) has shownthat the HaCaT cell line does express TLR4 (Kollisch et al., 2005).Nevertheless, the HaCaT cells are only activated by a relatively highamount of LPS. Although the selected pathways seem to be an inte-gral part of the response to sensitizers, it is not unlikely that othernon-sensitizing compounds can activate these pathways as well.Interestingly, the gene signature misclassified the Nrf2-inducersand not the TLR-ligands. This demonstrates that the gene signaturemight be biased towards Nrf2 activation. Possibly, this issue mightbe relevant to other in vitro systems as well, since there is evidencethat TLR activation can generate false-positive results in dendriticcells, as LPS is known to upregulate CD86 and CD54 (Coutantet al., 1999; Francisco et al., 2009; Schreiner et al., 2008).
The proof of concept study underlines the importance of chem-ical selection in the prevalidation phase of promising in vitro tests.For improved understanding of an assay it is recommended to in-clude a significant number of chemicals that are difficult to classifyin the LLNA and chemicals that are related to pathways that areimportant in the specific test. More insight in which classes ofchemicals can be accurately identified by an in vitro test will im-prove the understanding of the capabilities of an assay. However,for the correct classification of LLNA false-positive and false-nega-tive chemicals combination with other alternative assays will berequired. These should exploit different stages in the induction ofsensitization and can include in silico (Q)SAR methods, in chemicopeptide reactivity of chemicals or in vitro dendritic cell based as-says (Ashikaga et al., 2006; Gerberick et al., 2004; Patlewiczet al., 2007; Python et al., 2007). For effective integration of severalalternative approaches insight into the limitations of each assay isof importance. In conclusion, we believe that keratinocyte basedprediction assays provide essential information on the sensitizingpotential of chemicals and are therefore crucial in a integrated test-ing strategy.
Acknowledgements
This work was supported by a grant of the Netherlands Genom-ics Initiative/Netherlands Organization for Scientific Research(NWO) No.: 050-060-510. In addition, the Dutch Ministry of Infra-structure and the Environment is acknowledged for funding thiswork.
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
J.W. van der Veen et al. / Toxicology in Vitro xxx (2012) xxx–xxx 9
TIV 2934 No. of Pages 9, Model 5G
31 August 2012
Appendix A. Supplementary material
Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.tiv.2012.08.023.
References
Aeby, P., Wyss, C., Beck, H., Griem, P., Scheffler, H., Goebel, C., 2004. Characterizationof the sensitizing potential of chemicals by in vitro analysis of dendritic cellactivation and skin penetration. J. Invest. Dermatol. 122, 1154–1164.
Arkusz, J., Stepnik, M., Sobala, W., Dastych, J., 2010. Prediction of the contactsensitizing potential of chemicals using analysis of gene expression changes inhuman THP-1 monocytes. Toxicol. Lett. 199, 51–59.
Ashikaga, T., Yoshida, Y., Hirota, M., Yoneyama, K., Itagaki, H., Sakaguchi, H.,Miyazawa, M., Ito, Y., Suzuki, H., Toyoda, H., 2006. Development of an in vitroskin sensitization test using human cell lines: the human Cell Line ActivationTest (h-CLAT). I. Optimization of the h-CLAT protocol. Toxicol. In Vitro 20, 767–773.
Bauch, C., Kolle, S.N., Fabian, E., Pachel, C., Ramirez, T., Wiench, B., Wruck, C.J.,Ravenzwaay, B.V., Landsiedel, R., 2011. Intralaboratory validation of fourin vitro assays for the prediction of the skin sensitizing potential ofchemicals. Toxicol. In Vitro 10, 567–734.
Benjamini, Y., Hochberg, Y., 1995. Controlling the false discovery rate: a practicaland powerful approach to multiple testing. J. Roy. Stat. Soc.: Ser. B (Methodol.)57, 289–300.
Bolstad, B.M., Irizarry, R.A., Astrand, M., Speed, T.P., 2003. A comparison ofnormalization methods for high density oligonucleotide array data based onvariance and bias. Bioinformatics 19, 185–193.
Boorsma, A., Foat, B.C., Vis, D., Klis, F., Bussemaker, H.J., 2005. T-profiler: scoring theactivity of predefined groups of genes using gene expression data. Nucleic AcidsRes. 33, W592–595.
Boukamp, P., Petrussevska, R.T., Breitkreutz, D., Hornung, J., Markham, A., Fusenig,N.E., 1988. Normal keratinization in a spontaneously immortalized aneuploidhuman keratinocyte cell line. J. Cell Biol. 106, 761–771.
Breiman, L., 2001. Random forests. Machine Learning 45, 5–32.Casati, S., Aeby, P., Kimber, I., Maxwell, G., Ovigne, J.M., Roggen, E., Rovida, C., Tosti,
L., Basketter, D., 2009. Selection of chemicals for the development andevaluation of in vitro methods for skin sensitisation testing. Altern. Lab.Anim. 37, 305–312.
Corsini, E., Mitjans, M., Galbiati, V., Lucchi, L., Galli, C.L., Marinovich, M., 2009. Use ofIL-18 production in a human keratinocyte cell line to discriminate contactsensitizers from irritants and low molecular weight respiratory allergens.Toxicol. In Vitro 23, 789–796.
Coutant, K.D., de Fraissinette, A.B., Cordier, A., Ulrich, P., 1999. Modulation of theactivity of human monocyte-derived dendritic cells by chemical haptens, ametal allergen, and a staphylococcal superantigen. Toxicol. Sci. 52, 189–198.
Francisco, V., Neves, B.M., Cruz, M.T., Goncalo, M., Figueiredo, A., Duarte, C.B., Lopes,M.C., 2009. Effect of lipopolysaccharide, skin sensitizers and irritants onthioredoxin-1 expression in dendritic cells: relevance of different signallingpathways. Arch. Dermatol. Res..
Gerberick, G.F., Ryan, C.A., Dearman, R.J., Kimber, I., 2007. Local lymph node assay(LLNA) for detection of sensitization capacity of chemicals. Methods 41, 54–60.
Gerberick, G.F., Ryan, C.A., Kern, P.S., Schlatter, H., Dearman, R.J., Kimber, I.,Patlewicz, G.Y., Basketter, D.A., 2005. Compilation of historical local lymph nodedata for evaluation of skin sensitization alternative methods. Dermatitis 16,157–202.
Gerberick, G.F., Vassallo, J.D., Bailey, R.E., Chaney, J.G., Morrall, S.W., Lepoittevin, J.P.,2004. Development of a peptide reactivity assay for screening contact allergens.Toxicol. Sci. 81, 332–343.
Hervouet, E., Vallette, F.M., Cartron, P.F., 2009. Dnmt3/transcription factorinteractions as crucial players in targeted DNA methylation. Epigenetics 4,487–499.
ICCVAM, 2008. Revised Draft ICCVAM LLNA Performance Standards.Johansson, H., Lindstedt, M., Albrekt, A.S., Borrebaeck, C.A., 2011. A genomic
biomarker signature can predict skin sensitizers using a cell-based in vitroalternative to animal tests. BMC Genomics 12, 399.
Jowsey, I.R., Basketter, D.A., Westmoreland, C., Kimber, I., 2006. A future approach tomeasuring relative skin sensitising potency: a proposal. J. Appl. Toxicol. 26,341–350.
656
Please cite this article in press as: van der Veen, J.W., et al. Applicability of a kVitro (2012), http://dx.doi.org/10.1016/j.tiv.2012.08.023
Katsuki, A., Sumida, Y., Urakawa, H., Gabazza, E.C., Maruyama, N., Morioka, K.,Kitagawa, N., Hori, Y., Nakatani, K., Yano, Y., Adachi, Y., 2003. Increasedoxidative stress is associated with elevated plasma levels of adrenomedullin inhypertensive patients with type 2 diabetes. Diabetes Care 26, 1642–1643.
Kimber, I., Dearman, R.J., Scholes, E.W., Basketter, D.A., 1994. The local lymph nodeassay: developments and applications. Toxicology 93, 13–31.
Klekotka, P.A., Yang, L., Yokoyama, W.M., 2010. Contrasting roles of the IL-1 and IL-18 receptors in MyD88-dependent contact hypersensitivity. J Invest Dermatol130, 184–191.
Kollisch, G., Kalali, B.N., Voelcker, V., Wallich, R., Behrendt, H., Ring, J., Bauer, S.,Jakob, T., Mempel, M., Ollert, M., 2005. Various members of the Toll-likereceptor family contribute to the innate immune response of human epidermalkeratinocytes. Immunology 114, 531–541.
Kotlajich, M.V., Hertel, K.J., 2008. Death by splicing: tumor suppressor RBM5 freezessplice-site pairing. Mol. Cell 32, 162–164.
Kumar, H., Kawai, T., Akira, S., 2009. Pathogen recognition in the innate immuneresponse. Biochem. J. 420, 1–16.
Martin, S.F., Dudda, J.C., Bachtanian, E., Lembo, A., Liller, S., Durr, C., Heimesaat,M.M., Bereswill, S., Fejer, G., Vassileva, R., Jakob, T., Freudenberg, N., Termeer,C.C., Johner, C., Galanos, C., Freudenberg, M.A., 2008. Toll-like receptor and IL-12signaling control susceptibility to contact hypersensitivity. J. Exp. Med. 205,2151–2162.
Morse, D., Choi, A.M., 2002. Heme oxygenase-1: the ‘‘emerging molecule’’ hasarrived. Am. J. Respir. Cell Mol. Biol. 27, 8–16.
Motohashi, H., Yamamoto, M., 2004. Nrf2-Keap1 defines a physiologicallyimportant stress response mechanism. Trends Mol. Med. 10, 549–557.
Natsch, A., 2009. The Nrf2-Keap1-ARE toxicity pathway as a cellular sensor for skinsensitizers-functional relevance and a hypothesis on innate reactions to skinsensitizers. Toxicol. Sci. 23, 23–56.
Natsch, A., Bauch, C., Foertsch, L.M., Gerberick, G.F., Norman, K., Hilberer, A., Inglis,H., Landsiedel, R., Onken, S., Reuter, H., Schepky, A., Emter, R., 2011. The intra-and inter-laboratory reproducibility and predictivity of the KeratinoSens assayto predict skin sensitizers in vitro: results of a ring-study in five laboratories.Toxicol. In Vitro 25, 733–744.
Patlewicz, G., Aptula, A.O., Uriarte, E., Roberts, D.W., Kern, P.S., Gerberick, G.F.,Kimber, I., Dearman, R.J., Ryan, C.A., Basketter, D.A., 2007. An evaluation ofselected global (Q)SARs/expert systems for the prediction of skin sensitisationpotential. SAR QSAR Environ. Res. 18, 515–541.
Python, F., Goebel, C., Aeby, P., 2007. Assessment of the U937 cell line for thedetection of contact allergens. Toxicol. Appl. Pharmacol. 220, 113–124.
R Development Core Team, 2010. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing.
Rifkin, R., Mukherjee, S., Tamayo, P., Ramaswamy, S., Yeang, C.H., Angelo, M., Reich,M., Poggio, T., Lander, E.S., Golub, T.R., Mesirov, J.P., 2003. An analytical methodfor multiclass molecular cancer classification. SIAM Rev. 45, 706–723.
Sakaguchi, H., Ashikaga, T., Miyazawa, M., Yoshida, Y., Ito, Y., Yoneyama, K., Hirota,M., Itagaki, H., Toyoda, H., Suzuki, H., 2006. Development of an in vitro skinsensitization test using human cell lines; human Cell Line Activation Test (h-CLAT). II. An inter-laboratory study of the h-CLAT. Toxicol. In Vitro 20, 774–784.
Sakaguchi, H., Ryan, C., Ovigne, J.M., Schroeder, K.R., Ashikaga, T., 2010. Predictingskin sensitization potential and inter-laboratory reproducibility of a human CellLine Activation Test (h-CLAT) in the European Cosmetics Association (COLIPA)ring trials. Toxicol. In Vitro 24, 1810–1820.
Schreiner, M., Peiser, M., Briechle, D., Stahlmann, R., Zuberbier, T., Wanner, R., 2008.A new dendritic cell type suitable as sentinel of contact allergens. Toxicology249, 146–152.
Tibshirani, R., Hastie, T., Narasimhan, N., Chu, G., 2001. Diagnosis of multiple cancertypes by shrunken centroids of gene expression. PNAS 99, 6567–6572.
Zenz, R., Eferl, R., Scheinecker, C., Redlich, K., Smolen, J., Schonthaler, H.B., Kenner, L.,Tschachler, E., Wagner, E.F., 2008. Activator protein 1 (Fos/Jun) functions ininflammatory bone and skin disease. Arthritis Res. Ther. 10, 201.
eratinocyte gene signature to predict skin sensitizing potential. Toxicol. in
