Alternatives to animal experiments for eye irritation.

Contents

Preface

Summary

1. Introduction

2. Reconstructed tissue models
2.1 Reconstruction of tissues

3. The COLIPA study
3.1 Design of the study
3.2 Alternative methods
3.3 Prediction models
3.4 Test substances
3.5 Distribution of test substances and collection of in vitro results
3.6 Statistical analyses
3.7 Further analyses of the results obtained in the COLIPA study.
3.8 Discussion

4. Test for recovery from ocular irritation
4.1 Method
4.2 Results
4.3 Discussion

5. Validation of alternatives to ocular irritancy tests

6. Evaluation of alternatives to ocular irritancy tests

7. Abbreviations

8. References

Preface

The Danish Environmental Protection Agency has sponsored the project "Validation of a three-dimensional model for eye and skin irritation". The study has been conducted as a joint venture between the Danish Environmental Protection Agency and the Danish Veterinary and Food Administration, Institute of Food Safety and Toxicology. The studies with the tissue model SKIN² ZK1200 have been published in 4 articles that are the basis of the present report:

Brantom, P. G. et al.: A summary report of the COLIPA international validation study on alternatives to the Draize rabbit eye irritation test. Toxicology in Vitro, 1997, 11, 141-179.

Espersen, R. J., Olsen, P., Nicolaisen, G., Jensen, B. L. and Rasmussen, E. S.: Assessment of recovery from ocular irritancy using a human tissue equivalent model. Toxicology in Vitro, 1997, 11, 81-88.

Rasmussen, E. S.: Use of reconstructed human tissue models in toxicological and pharmacological studies. Dansk Veterinærtidsskrift, 1/11, 1996, 943-948.

Southee, J. A., McPherson, J. P., Osborne, R., Carr, G. J. and Rasmussen, E. S.: The performance of the tissue equivalent assay using the SKIN^2TM ZK1200 model in the COLIPA international validation study on alternatives to the Draize eye irritation test. Toxicology in Vitro, 1999, 13, 355-373.

The project has been followed by a steering group with the following members:

The Danish Environmental Protection Agency:
Lone Mikkelsen (former chairman)
Lars Nørgaard (present chairman)
Anne Marie Linderstrøm

The Danish Veterinary and Food Administration:
John Christian Larsen
Eva Selzer Rasmussen (project leader)

The project has been accomplished by a project group at the Institute of Food Safety and Toxicology with the following members:

René Espersen
Preben Olsen
Bo Lund Jensen
Gert Maarløw Nicolaisen
Martin Bach
Eva Selzer Rasmussen

The members of the steering group and the project gruop are acknowledged for their excellent collaboration.

Summary

Models of human tissues can be reconstructed from single cells. Test substances can be applied directly to the tissues, and the potency for local irritancy can be evaluated by measurements of the viability of the cells. A corneal model, the SKIN² ZK1200 tissue equivalent, was evaluated as a possible replacement of tests for ocular irritation using rabbits. The tissue equivalent was used to evaluate 55 cosmetic ingredients and finished products in a large validation study arranged by the European cosmetics industry (COLIPA). The SKIN² ZK1200 model was shown to be very good in predicting ocular irritation (Draize test MMAS values). The tissue equivalent assay gave a good prediction of individual tissue reactions in the eye both for cosmetic ingredients and formulations, and all types of materials could be tested. In addition, a new testing procedure developed with the tissue equivalent showed promise for prediction of recovery from ocular irritation in a study on nine test substances. The results obtained indicate that the SKIN² ZK1200 tissue equivalent showed promise as a replacement of animal experiments for ocular irritancy testing. After the study was completed, the production of the SKIN² ZK1200 model has ceased. The results obtained with the model are, however, considered to be relevant for the assessment of the prospects for replacement of animal experiments for ocular irritancy with alternative methods.

The results of the COLIPA study also confirmed the results of other large validation studies on alternatives to animal experiments for eye irritancy testing. None of the other alternative methods in the COLIPA study were suitable to predict ocular irritation caused by the mixed group of ingredients and formulations. Several alternative methods were, however, very good in predicting ocular irritation caused by surfactant based ingredients. Tests using hens eggs had a poor interlaboratory reproducibility, whereas the reproducibility of the other alternative methods was good.

1. Introduction

World-wide there is a strongly increased interest in the use of alternatives to animal experiments, in particular concerning safety assessments of cosmetics. Most resources have been used on the development and validation of alternative methods for evaluation of ocular irritation, where the existing animal experiments have been subject to intensive critizism based on scientific and ethical arguments.

A lot of in vitro tests have been introduced as possible alternatives to in vivo eye irritation tests. These methods comprise simple physico-chemical tests with plant proteins, methods with cultures of cells from humans and other mammals, systems with isolated eyes or ocular tissues and tests with the chorioallantoic membrane of hens eggs. A new type of in vitro tests with human tissues that have been reconstructed from cell cultures has also been introduced. Good correlations have been found between results obtained with reconstructed tissue models and results from animal experiments and human data in several toxicological areas. The present report is initiated with a section on reconstruction of tissue models with human cells, especially concerning tests for local irritancy.

The report deals also with experiments with a corneal tissue model, SKIN² ZK1200, which has been evaluated as a possible replacement of eye irritation tests using rabbits. 55 cosmetic ingredients and finished products were tested in a validation study organized by the European cosmetics industry (COLIPA), and 10 alternative methods were evaluated. The Institute of Food Safety and Toxicology participated in the validation study sponsored by the Danish Environmental Protection Agency as one of the two laboratories using the SKIN² ZK1200 model.

COLIPA is the European trade association of cosmetics industry. The organisation was established in 1962, and it represents 95% of the cosmetics industry in the EU member states with approximately 2000 companies. Additionally, COLIPA comprises 6 groups of companies from non-EU states and 21 large international companies.

The cosmetics industry is one of the major promoters of development and validation of alternatives to animal experiments, and COLIPA has been arranging several programs on the validation of alternative tests. In 1992, COLIPA established a Steering Commitee on Alternatives to Animal Testing (SCAAT), which coordinates the efforts of the cosmetics industry in the development of alternative methods. The members of SCAAT come from the companies Biersdorf, L'Oreal, Procter & Gamble, SmithKline Beecham, Unilever and Wella, but all of the international companies in COLIPA provide support and funding to SCAAT. Currently, there are four SCAAT task forces focusing on: eye irritation, photoirritation, human skin compatibility and percutaneous absorption.

The provisions for a potential ban on animal testing of cosmetics and their ingredients implied in the EU Cosmetics Directive (EU, 1993) have added additional impetus to the progress of the alternative methods. COLIPA is one of the major interest groups involved in the discussions of the progress of alternative methods arranged by the European Commission.

In the COLIPA eye irritation validation study, very good correlations were demonstrated between prediction of eye irritancy from data obtained with the tissue model SKIN² ZK1200 and acute ocular irritation observed with the Draize test. The method used with the tissue model was, however, not suited for prediction of the persistency of ocular irritation. In an independent preliminary study with 9 test substances, a method for the evaluation of recovery from acute ocular irritation using the SKIN² ZK1200 model was developed. In the present report, the new model for recovery for ocular irritation is described after the presentation of the COLIPA eye irritation validation study.

A discussion of the future prospects for replacement of animal experiments for ocular irritation is also included in the present report. Most of the alternative methods have a better reproducibibility than the Draize test. For this reason, an evaluation of the ability of the alternatives to predict the in vivo response should be given priority to evaluation of the intra- and interlaboratory reproducibility of the methods. The SKIN² ZK1200 model is the first in vitro method, which has shown a convincing potential as a model which may totally replace the Draize test. The production and sales of the tissue model has ceased, but it may be possible to use other commercial tissue models or to develop new, non-commercial tissue models to be used in ocular irritancy testing. Other types of alternative methods have not been shown to be suited for a general prediction of ocular irritancy, but several of the existing alternative tests may be suited for test of water soluble compounds, for example, surfactant based products.

2. Reconstructed tissue models

2.1 Reconstruction of tissues

Several in vitro models with reconstructed tissues with cells from human skin have been introduced. These systems include ocular models, which are composed of a combined epithelium and stroma or just serve as models of the corneal epithelium. Several skin models are constructed in a similar way, but they also include a stratum corneum. The similarities between the reconstructed tissue models and the normal human tissues makes many practical applications in toxicology and pharmacology possible (Rasmussen, 1996). Experiments with reconstructed tissue models have so far given good concordances to results from animal experiments on many types of individual chemicals and finished products for eye and skin irritation, corrosivity and photorirritation. The reconstructed tissue models can also be used to evaluate efficacy issues, e.g. sun protection, wound healing and anti-aging.

2.1 Reconstruction of tissues

Skin equivalents were first developed for treatment of large burns, where engineered grafts are needed. Both chemical systems with foam pads obtained by co-polymerizing dermal macromolecules, and biological systems with viable skin cells have been used. Culturing of keratinocytes directly on the burn wound bed sometimes results in wound contraction, and preliminary use of a dermal equivalent has been shown to improve the wound healing. Epidermal cell cultured sheets are used to replace the missing epidermis. Hormones and other growth factors are supplied from the patients body, and the keratinocytes will reconstitute to a fully differentiated epithelium with a stratum corneum.

Reconstructed skin models are composed either of pure keratinocyte cultures or of combinations of fibroblasts and keratinocytes. For this reason, they present a very simplified system compared to the in vivo situation. The large challenge in the construction of the tissue models has been to produce a multilayered epidermis with a stratum corneum without the assistance of a live organism. Keratinocytes are seeded to filters or other substrates, and they first form a coherent layer of basal cells, and then several epithelial cell layers. If the process is stopped at this stage, an ocular model has been produced. When the epidermal tissues afterwards are exposed to air, the multilayered epidermis may be covered by a coherent layer of corneocytes, a thin stratum corneum. The artificial epidermis is composed of layers of keratinocytes similar to the natural epidermis, and it has been shown to mimic the ultra structure well by light and electron microscopy.

Reconstructed tissue models can be used to study the differentiation of keratinocytes, including changes in proteins and lipids. Biochemically, reconstructed skin models have been shown to mimic normal skin well, but the models also differ significantly structually and biochemically from the in vivo situationen. Large differences in the relative composition of various lipids and their structural organization have been demonstrated. In addition, high rates of diffusion through the not fully cornified cells of the stratum corneum in vitro have been found.

Several tissue models are solely composed of an epidermal tissue cultured on non-viable dermal replacements, such as a deepidermized dead dermis, microporous filters or a collagen matrix. The presence of a viable dermis may, however, have a significant impact on the functions of the tissue. During wound healing in vitro, the growth of the keratinocytes is significantly increased by the presence of epidermal growth factors and other substances that are produced by the basal membrane and the dermal fibroblasts. Additionally, the presence of an epithelium, a basal membrane and a stroma in the in vitro models may be used in evaluations of recovery from lesions to the tissue. When the outer parts of an epithelium solely have been damaged, the consequences in general are small, because the tissue is able to regenerate. If the basal membrane and the stroma also are damaged, the risk that permanent tissue lesions are introduced is dramatically increased.

A live dermis can be constructed in several ways. Fibroblasts may, for example, be embedded in a collagen gel. The cells quickly adhere to the collagen fibers, the gel contracts, and the tissue culture medium is expelled. After 2-3 days in culture, a plate of white, condensed tissue, which is a suitable basis of culturing epidermal cells, has been created. A viable dermal tissue can also be constructed by culturing fibroblasts on a nylon net. The fibroblasts adheres to the net, secretes various dermal macromolecules, and the tissue gradually covers the meshes. A three-dimensional stroma, which can serve as a basis for culturing other cell types, has then been constructed.

A series of commercial tissue models with fibroblast tissue grown on nylon net as a basis have been available. The SKIN² ZK 1100 model was solely composed of fibroblast tissue. In the SKIN² ZK1200 ocular model, the fibroblast stroma was supplemented with 3-4 layers of keratinocytes. SKIN² ZK1200 was a model of the cornea, which has a layer of epidermal cells of corresponding thickness. SKIN² ZK 1300 systems with a stratum corneum have been used as skin models. A model with human oral epithelial cells has been introduced for test of dental materials. Additionally, artificial skin to be used in transplantations has also been constructed using nets as a basis.

The tissue models are metabolically active, and they can also be used to study the interaction between different cell types. Experiments with melanocytes cultures in multilayered keratinocyte tissues have shown that several characteristics from the in vivo situation can be reestablished. The melanocytes establish in the basal layer of keratinocytes, and start to produce the pigment melanin. The pigmentation of the tissue increases after exposure to UV light. This model has been evaluated to be well suited for test of chemicals and finished products for phototoxic or photoprotective effects.

Exposure of tissues to local irritants or UV light may change the enzyme activity and the growth pattern of the cells, and increase the levels of irritation markers such as prostaglandines, leukotrienes and cytokines. Measurements of such parameters in monolayer cell cultures have been evaluated as possible alternatives to animal experiments for test for local irritancy. In general, results obtained from this type of systems have been in good agreement with data on in vivo eye irritation for water soluble substances. In several large blind studies of mixed chemicals and products, such methods have, however, been shown to be inadequate in predicting the in vivo response. Experiments with ocular models composed of reconstructed tissue are now among the most promising alternatives to animal experiments. Several protocols with tissue models have been developed to test for corrosivity, skin irritation, and phototoxicity, and such methods are subject to further validation.

3.1 Design of the study
3.2 Alternative methods
3.3 Prediction models
3.4 Test substances
3.5 Distribution of test substances and collection of in vitro results
3.6 Statistical analyses
3.7 Further analyses of the results obtained in the COLIPA study.
3.8 Discussion

In 1993, COLIPA initiated a large program on validation of alternative methods to the Draize eye irritation test. The study was concentrated on test of cosmetics ingredients and formulations.The COLIPA study was designed to build on the lessons learned from a former validation study in this field arranged by the EU Commission and British Home Office (the EU/HO study) (Balls et al., 1995), where 40 % of the participating laboratories were from the cosmetics industry. 20 test substances were common to both studies, and several alternative methods were also used in both studies.

Figure 3.1   Look here
Management chart of the COLIPA eye irritation study. The tissue model SKIN² ZK1200 is used in the tissue equivalent assay.

The specific goal of the study was to determine whether a set of alternative methods would be valid for the prediction of the eye irritation potential of cosmetics ingredients and formulations. As a consequence, it was one of the specific goals of the study to evaluate if the alternative tests could be used as replacements for animal experiments on ocular irritancy (de Silva, 1996). Specifically, the program was designed to determine whether data from alternative methods could provide: 1. an acceptable agreement with the Draize test modified maximum average score (MMAS), 2. an acceptable agreement with the Draize test individual tissue scores and time to recovery, and/or 3. a proper prediction of the eye irritation potential in the rabbit eye according to a prediction model.

3.1 Design of the study

The study was sponsored and organized by COLIPA, but some independent research groups participated in the study. The Danish Veterinary and Food Administration, Institut of Food Safety and Toxicology, participated in the study on a grant from the Danish Environmental Protection Agency in a co-operative project. COLIPA formed a Task Force Committee, which was responsible for the overall design of the study and for establishment of the policies to be followed during its conduct, and a Management Team was appointed to oversee the day-to-day conduct of the study (see figure 3.1). Each group of laboratories had a lead laboratory, which was responsible for the contact between COLIPA and the participants, established the test protocol and prediction model for the test, and monitored the progress in other laboratories.

3.2 Alternative methods

The COLIPA study included 10 alternative methods, which already were being used in the cosmetics industry as screening tests.

SKIN² ZK1200

In this assay, a reconstructed tissue with keratinocytes seeded on a stroma of epidermal fibroblasts from human foreskin is used. The fibroblasts were grown on a nylon mesh matrix, and a coherent stroma was produced within a month. The keratinocytes were seeded to this stroma, and a 3-4 cell layer epithelium was generated. 1 x 1 cm tissues were produced as a commercial test kit, SKIN² ZK1200, by Advanced Tissue Sciences in California. Kits with 24 tissues were packed in trays, embedded in a nutrient agar, and transported to the participating laboratories by plane.

The test substances were applied undiluted to the epithelial surface of the tissues in fixed concentrations of 25 m l or 25 mg. After exposure to the test substances for up to 60 minutes, the tissues were washed and the viability of the cells was measured with the MTT test. MTT is a yellow tetrazolium compound, which is reduced to a purple formazan salt by cellular redox processes. The purple product was extracted with isopropanol, and the optical density was measured at 540 nm. For each test substance, a t₅₀ value was determined. This is the exposure time, which leads to a 50% reduction of the MTT reduction. The test had a mathematical prediction model that was established on the basis of historical data for 132 ingredients and products. All ingredients and products could be accommodated with the method. The production of the model has ceased after the COLIPA study was completed.

Neutral red uptake

The neutral red uptake test was performed on 3T3 mouse fibroblasts. Neutral red is selectively retained by the lysosomes in living cells because of the differential pH of the lysosome and the cytoplasm. The amount of neutral red taken up by the cells is directly proportional to the number of viable cells present. The test substance concentration giving a 50% reduction in the neutral red uptake was determined. The test had a mathematical prediction model that could predict MMAS values up to 60. The model was established on the basis of historical data for 30 substances of which 29 were surfactants. A limitation of the method is that inorganic acids and bases can not be tested, and that only water soluble compounds can be accommodated.

Damage to the cell membrane is assessed by measuring the leakage of haemoglobin from red blood cells incubated with test substances. Protein denaturation is also measured by determining the reduction of oxyhaemoglobin. The test substance concentration causing 50% haemolysis relatively to a totally lysed sample was determined. Both the lowest concentration of the test substance causing denaturation and the maximum percentage denaturation were determined. The test had a mathematical prediction model that was established on the basis of historical data for 25 finished products, and a classification model that was established on the basis of historical data for 100 substances. A limitation of the method is that only water soluble or water miscible compounds can be accommodated.

In this assay the cell viability is assessed after a short exposure time (< 1 minute) to the test substances by measuring the neutral red release from pre-loaded SIRC cells (rabbit corneal cells). The test substance concentration giving a 50% reduction in the neutral red uptake is determined. The test had a mathematical prediction model that was established on the basis of historical data for 47 finished products. All ingredients and products could be accommodated.

Cultures of murine fibroblasts (L929 cells) were grown on Transwell membranes, and transferred to a sensor chamber on a Cytosensor^TM silicone microphysiometer. The metabolic rate was measured continuously as decreased extracellular pH. The test substance concentration causing a 50% reduction in the metabolic rate was determined. The test had a mathematical prediction model that was established on the basis of historical data for 133 surfactant and surfactant based products. Only water soluble substances can be accommodated.

In this assay, changes in the blood vessels (VAscular changes) of the chorioallantoic membrane (CAM) of fertilized hens eggs are determined. The eggs are incubated for 3 days and on day 4, a small hole is drilled into the egg and approximately 2.5 ml albumin removed. A 2 x 2 cm window is cut through the shell, and on day 10 a teflon ring is placed on the CAM and 40 m l of the test substance is pipetted into the ring. The window is sealed, and the egg is reincubated. The CAM response was evaluated after 30 minutes, and vascular haemorrhage, capillary injection and the presence of ghost vessels were considered to be a positive response. The calculated concentration theoretically producing a reaction in 50% of the eggs is determined. The CAMVA test had two different mathematical prediction models. One model for alcohols that was established on the basis of historical data for 4 substances, and one for other compounds that was established on the basis of historical data for 19 substances. The last model excludes results obtained with polyethylenglycol-fatty acids and the related fatty amide ethanolamides.

The EYTEX test was a commercial test kit with a plant protein that may be opacified after contact with a test substance. The turbidity was determined with a colorimeter. A set of calibrators provided a direct comparison with a Draize scale to determine an EYTEX Draize equvivalent. The EYTEX test had a mathematical prediction model that was established on the basis of historical data for a wide range of test substances. MMAS values up to 99 could be predicted. The model could not be used to predict the irritancy of products containing manganese violet, >5% urea, >3% aluminium chlorohydrate, >5% zink oxide or surfactants at >40% active ingredient. The production of the model has ceased after the COLIPA study was completed.

The assay is based on photometric quantification of pollen tube growth. Pollen grain from Tobacco plants are cultured for 18 hours in the presence of the test substance. The mass of pollen tubes produced during the incubation period was determined photometrically using the dye Alcian blue. The test substance concentration causing a 50% reduction in the production of pollen tube mass was determined. The test had a mathematical prediction model that was established on the basis of historical data for 43 finished products. All ingredients and products could be accommodated with the method.

In this assay, changes in the blood vessels in the chorioallantoic membrane of fertilized hens eggs are determined. The test substances were applied to the CAM on day 9, and the CAM response was evaluated after 5 minutes for transparent test substances and after 30 seconds for other substances. Vascular haemorrhage, lysis and coagulation were considered to be a positive response. The result was converted to an irritation index using a computer program. 5% Texapon SVF (an anionic surfactant) was used as a positive control. The HET-CAM test had a classification based prediction model that was established on the basis of historical data for 97 substances. A limitation of the method is that substances that stick to the membrane and highly coloured compounds cannot be accommodated.

In this assay, confluent Madin-Darby Canine kidney cells are exposed for 15 minutes to five fixed concentrations of the test substance. The amount of damage to the cellular monolayer was determined by determination of the amount of fluorescein leaking through the cell layer. The test had a classification based prediction model that was established on the basis of historical data for 43 surfactants and formulations. Only water soluble compounds could be accommodated.

3.3 Prediction models

Prediction models were established for each alternative test in order to assess the reliability of the methods. Prediction models are algoritms that converts the results from the assays into a prediction of the in vivo toxicity. It is difficult to assess the performance of an assay which does not have a prediction model since the relationship between in vitro and in vivo data has not been defined (Bruner et al., 1996).

The prediction models used in the COLIPA study were developed on the basis of historical data for the tests. The prediction models defined four elements needed to predict in vivo toxicity from in vitro results: 1. a description of the types of test substances for which the prediction model may be used, 2. a description of the types of data provided by the assay, including which data the prediction model can accommodate, 3. an algoritm defining how to convert the results from the assay into a prediction of the in vivo toxicity, and 4. an indication of the accuracy of the predictions.

The reliability of the prediction models was evaluated by determining whether the assay results were reproducible between the laboratories, and whether the data obtained fitted within the prediction intervals of the prediction models. After the COLIPA study was completed, the relevance of the alternative methods in predicting the eye irritation potential was considered to be a separate issue (Brantom et al., 1997).

Two different types of prediction models were used in the study. One group of models were mathematical functions converting the in vitro data to a broad spectrum of Draize MMAS values. Another group was non-continuous classification models that related the in vitro data to various irritancy classes. Different statistical methods had to be used to analyse the results obtained with the in vitro methods, depending on the type of prediction model used. Additionally, the basis for the comparison of the in vitro and in vivo results was not fully standardized. The data obtained with the HET-CAM test were compared to Draize test individual tissue scores, whereas Draize test MMAS values were used for all the other in vitro systems.

3.4 Test substances

The COLIPA study included 55 test materials covering a broad spectrum of cosmetics ingredients and finished products. The test substances were not only water soluble surfactant based products, but solids, powders, cremes, coloured products and alcohol based products were also included. The 55 test substances covered the whole spectrum of the Draize test MMAS 110 point scoring scale.

The first 23 test substances were ingredients (see table 3.1). The in vivo eye irritation data for these substances were supplied from the ECETOC data bank (ECETOC, 1998). All substances in this data bank have been tested for ocular irritancy in the Draize test according to OECD guideline 405. The ECETOC eye irritancy data bank includes both MMAS values and individual tissue scores for all the individual rabbits used in the experiments. 20 of these substances were common to the COLIPA study and the EU/Home Office study.

Table 3.1
Ingredients used as test substances in the COLIPA study

The other 32 test substances were finished cosmetic products (see table 3.2). The products were based on formulations that had been used in a large validation study arranged by the US cosmetics industry (CTFA), and the products had formerly been tested in vivo for eye irritancy. The results from these animal experiments could, however, not be used in the COLIPA study since a different protocol using local anaesthesia had been used. For this reason, new Draize tests according to OECD guideline 405 were performed on the 32 products by Agencé du Medicament. The in vivo experiments were performed with one to six rabbits per test substance. The experiments were only repeated for two test substances, and in these cases the average scoring values from the two experiments were used. Due to the lack of adequate data, the variation on the in vivo experiments could not be evaluated.

The COLIPA study included two phases: the first 10 test substances were tested in 1994, and the remaining 45 materials were tested in 1995. All substances were tested blind and they were coded individually in order to avoid that the participating laboratories could compare their results before the study was completed. Since hazardous compounds were tested, the participants were supplied with the telephone number of a Poison Information Centre in case of accidents.

3.5 Distribution of test substances and collection of in vitro results

BIBRA International was taking care of the distribution of test substances to the laboratories, the collection of the in vitro results, and the statistical analyses of the data. All these functions were conducted due to GLP.

The participating laboratories submitted the in vitro results on standardized data sheets directly to BIBRA. The data included information on the codes of the samples, a description of the test materials, in vitro raw data and predicted in vivo data. The quality of the submitted data was controlled by BIBRA.

3.6 Statistical analyses

BIBRA performed the statistical analyses. Different statistical methods were used to analyse the results obtained with the in vitro methods, depending on the type of prediction model for the method, e.g. mathematical function or non-continuous classification model.

For in vitro assays with mathematical prediction models, the relevance of the methods was evaluated by analyses of linear correlations between Draize MMAS values that were predicted on the basis of in vitro data and observed Draize MMAS values. In addition, it was evaluated to which extent the predicted Draize MMAS values fitted with the relevant prediction model, and an assessment of the fit of the predictions within the 95% and 99% prediction intervals was performed. BIBRA also performed a goodness-of-fit test based on the sum of squared differences between predicted and observed Draize MMAS values. In addition, the ability of the in vitro methods to predict the in vivo response was evaluated on the basis of Altman/Bland plots of the difference between predicted and observed Draize MMAS values from each laboratory.

In order to evaluate the reproducibility of the alternative methods between different laboratories means, standard deviations and coefficients of variation were calculated for all the assays. Both non-transformed and log-transformed data were used. In addition, Altman/Bland plots were made on paired differences between predicted Draize MMAS values from the participating laboratories.

Both the relevance and the reproducibility of in vitro methods with classification models were evaluated using kappa statistics. The measure of agreement in the classification, kappa (k ), has a maximum on 1 when the agreement is perfect. A kappa value of zero means that the agreement is not better than change, and negative values shows a higher disagreement. The

kappa statistics can be used with equal or different weighting of the results. By equal weighting all disagreements will be treated in a similar way. Kappa statistics with linear weighting puts more weight to effects of disagreements of more than two classifications, and quadratic weighting gives an even higher weight to effects of disagreements of more than two classifications.

One of the conclusions of the COLIPA study was that a further, more detailed analysis of the data generated in this study and the EU/Home Office study should be carried out, and that the knowledge gained should be used to design a future definitive validation study (Brantom et al., 1997). For the SKIN²ZK1200 model, additional data was available on 20 of the test substances from a third laboratory (Procter & Gamble, Cincinnati, USA). The SKIN²ZK1200 model was not planned to be included EU/Home Office study, but the assay was used by one of the participating laboratories on coded test substances, and the results had been submitted blind to BIBRA. In the further analysis, relationships between data obtained with the SKIN²ZK1200 model and individual Draize tissue scores were studied and an additional analysis of the assay reproducibility was performed. In addition, the data obtained in the COLIPA study with the SKIN²ZK1200 model was compared to the data obtained at Procter & Gamble (Southee et al., 1999).

3.8 Results

The SKIN²ZK1200 model

The SKIN²ZK1200 model performed well in predicting the Draize test response for all test materials. The test was performed in 2 laboratories: Microbiological Associates, Scotland (laboratory 21) and Institute for Food Safety and Toxicology, Denmark (laboratory 23). The first 10 test substances were also tested in a third laboratory, Laboratory Simon, Belgium (laboratory 22), and additional data on 20 of the compounds were available from Procter & Gamble, USA.

Figure 3.2

Relationships between Draize test MMAS values predicted with SKIN² ZK1200 in laboratory 21 (a) and 23 (b) and observed MMAS values. From Brantom et al., 1997.

Very good correlations were obtained both in laboratory 21 and 23 between predicted and observed MMAS values (Brantom et al., 1997), see figure 3.2. The good correlations were maintained, when a separate analysis was performed for formulations and ingredients, see table 3.3. Good correlations between the SKIN²ZK1200 model 'core data' (t₅₀ values) and individual Draize test tissue scores (r>0.8) were also obtained, but a relatively poor correlation (r = 0.66) was obtained to average days to clear the response in the Draize tests (Southee et al., 1999).

Table 3.3
Correlations between predicted and observed Draize test MMAS values in the COLIPA laboratories using the SKIN² ZK1200 model

The SKIN²ZK1200 model performed also well in reproducing the prediction model. No predictions fell outside the 95% and 99% prediction intervals for one laboratory (21) in the COLIPA study. For the other laboratory (23) only 5.3% of the data points did not fit the 95% prediction interval. These datapoints represented 7 test substances being slightly to moderately over predicted.

Figure 3.3

Relationship between t₅₀ values obtained with the SKIN² ZK1200 model in laboratory 21 (a) and 23 (b) observed Draize test MMAS values. The unbroken line shows the prediction model and the broken lines represents the 95% confidence intervals of the model. From Brantom et al., 1997.

Table 3.4
Correlations between predicted and observed Draize test MMAS values in the COLIPA study for all test substances

Interlaboratory reproducibility in the COLIPA study

As a supplementary test, a positive control (2% sodium lauryl sulphate, SDS) was included in all assays (see table 3.5).The mean % viability for the positive control was lower in laboratory 23 (29.4%) compared to laboratory 21 (49%) supporting the suggestion that there was a tendency to overpredictions in laboratory 23. Additionally, the mean OD₅₄₀ of the untreated controls was higher in laboratory 21 than in laboratory 23, suggesting that the viability of the tissues may have been higher before the assay started in laboratory 21 than in laboratory 23. The CV's of the untreated controls were 15%.

Table 3.5
Positive and negative controls in the SKIN² ZK1200 model

CV = coefficient of variation

The % CV around the positive control was lower in laboratory 21 (22.9%) than in laboratory 23 (48.3%) suggesting that the intralaboratory variation of the assay was lower in laboratory 21. However, considerably more assays were completed in laboratory 21 than in laboratory 23, and this suggests that it is better to limit the number of materials tested at one time in the assay.

In the prediction of the Draize test response, log-transformed t₅₀ values are used and this decreases the data variation considerably. The Altman/Bland plots of the difference between the observed and predicted MMAS values showed standard deviations of <15% i both laboratories using the SKIN²ZK1200 model. A scatterplot of the t50 values obtained with the model in laboratory 21 and 23 is presented in figure 3.3. There was a 100% interlaboratory agreement on 23 substances with t50 values of >60 minutes or <0.1 minute, and a linear correlation of r=0.92 (p<0.001) for 32 substances with t₅₀ values falling between the cut-off points. There was a marginally significant difference (p=0.06) in the paired differences for each laboratory's core data for all the test substances. This was probably due to the tendency to over predict the in vivo response for the most reactive test substances (e.g. substances with t₅₀< 0.1 minute) in laboratory 23. Therefore, the results obtained in the COLIPA study indicated that a good consistency in the data is obtainable with the assay (Southee et al., 1999).

Interlaboratory reproducibility in 3 laboratories

In general, the results with the SKIN²ZK1200 model obtained at Procter & Gamble under predicted the irritancy of the 20 substances tested compared to those generated in the COLIPA study (p<0.001)(Southee et al., 1999). The systematic under prediction at Procter & Gamble (USA) compared with the tendency to over predict at laboratory 23 (Denmark) relative to laboratory 21 (Scotland) suggests that the sensitivity of the SKIN²ZK1200 model may be strongly influenced by differences in the time of transportation of the tissues. During transportation the tissues were subject to unfavourable conditions as they were embedded in nutrient agar.

Another possible reason for the differences in results obtained is the use of a different dosing regimes. At Procter & Gamble the test substances were first applied to a coverslip and then applied to the tissues, whereas these substances primarily were applied directly to the tissues in the COLIPA study. Some test material may have been lost using the indirect application approach, and the toxicity of volatile substances may have been underestimated. In addition, the greatest discrepancy in results between the two laboratories participating in the COLIPA study was found for highly reactive and often solvent based substances. Subtle differences in dosing techniques with such substances may be expected to have a great influence on the results being obtained (Southee et al., 1999).

The consistency of data generated in different laboratories with the SKIN²ZK1200 model was apparent. The discrepancies seen between the laboratories suggest a high sensitivity of the SKIN²ZK1200 method to differences in application techniques, handling and time of transportation of the tissues from producer to customer. Production and sale of the SKIN²ZK1200 model have stopped, but other tissue equivalent assays that allow for topical application of test substances would be worthy of investigation for their ability to predict eye irritation (Southee et al., 1999).

Table 3.6
Correlations between predicted and observed Draize test MMAS values in the COLIPA study for 23 formulations

Table 3.7
Correlations between predicted and observed Draize test MMAS values in the COLIPA study for 32 ingredients

Other methods with mathematical prediction models

Relatively poor average correlations to Draize MMAS values were obtained with the in vitro methods with mathematical prediction models for the whole set of substances tested (see table 3.4). The full set of test substances was only tested in two of the assays (the pollen tube growth test and the Predisafe test). Therefore, the correlation coefficients obtained for the other assays are most likely overestimating their performance in predicting the Draize test response for mixed groups of chemicals and products. In general, the ability to reproduce the prediction models of the assays was modest, and all the assays had considerably more than 5% of the data points falling outside the 95% confidence interval. When a separate analysis was performed for formulations and ingredients, considerably better correlations between predicted and observed Draize MMAS values were obtained for several of the tests (see table 3.6 and 3.7).

Interlaboratory reproducibility parameters for the methods with mathematical prediction models are shown in table 3.8. The most and least reproducible tests appear to be the EYTEX system and the red blood cell test, respectively, evaluated on the basis of non-transformed CV’s. Using log-transformed CV’s, the CAMVA test appeared to be very unreproducible, while the reproducibility of the red blood cell was in the middle range of the alternative methods. Using the Altman/Bland standard deviations, the pollen tube growth test, the neutral red uptake assay and the red blood cell test appeared to have the best interlaboratory agreement.

Table 3.8
Interlaboratory reproducibility parameters for assays with mathematical prediction models.

* :Coefficient of variation

The CAMVA test and the silicon microphysiometer both had a relatively strong tendency to over predict the in vivo response, whereas the red blood cell test had a marked, but less pronounced tendency to produce false positives. A pronounced tendency to underpredict the Draize MMAS values using the EYTEX system was found in one laboratory. The other alternative tests did not markedly tend to produce either false positives or false negatives compared to the above mentioned assays (Brantom et al., 1997).

Methods with classification based prediction models

A summary of the kappa statistics on the predictive capacity of in vitro methods with classification models is shown in table 3.9.

Table 3.9
Prediction of Draize test irritation classes by in vitro methods with classification models.

*: Average of kappa values obtained in the participating laboratories.

The predictive capacity of the HET-CAM test was evaluated to be poor. Each participating laboratory misclassified at least 7 of the 55 test substances by 2 categories. The HET-CAM test had its greatest success in identifying severe irritants, but there was an appreciable number of under predictions. The fluorescein leakage test performed better in predictive capacity, but only 40 of the 55 substances could be tested. However, only 4 test substances were classified to be moderate irritants, and this gives not enough data to evaluate the model. Additionally, the classification model does not discriminate between substances with MMAS values between 30 and 110. The interlaboratory reproducibility of the HET-CAM test was evaluated to be moderately good at the low and high end of the irritancy scale, but poorer in the middle of the scale. The fluorescein leakage test was used in 2 laboratories only, but there was a good interlaboratory agreement on the data generated (se table 3.10) (Brantom et al., 1997).

Table 3.10
Interlaboratory reproducibility of in vitro methods with classification models.

*: Average of kappa values obtained in the participating laboratories.

3.9 Discussion

After the COLIPA study was completed, it was concluded that none of the in vitro methods could meet both the criteria of reproducibility and relevance of the study. For this reason, none of the alternative methods were considered to be valid alternatives to the Draize test. Three of the methods used, the fluorescein leakage test, the red blood cell test and the SKIN²ZK1200 system, were evaluated to be either reproducible or relevant. Further analysis of the data obtained was recommended in order to establish new prediction models that could be tested in a future validation study (Brantom et al., 1997).

Several factors may be the basis for these conclusions. The COLIPA study was planned to be conducted with at least 3 participating laboratories per test. Several of the alternative tests were, however, only conducted in 2 laboratories. After the study was completed, the participants were informed that the reproducibility of the methods would only be evaluated, if the test had been performed in 4 laboratories. 70% of the alternative tests could not meet this demand. The results obtained concerning the relevance of the tests, e.g. their ability to predict the in vivo response, were not subject to a thorough discussion, where the testing had been completed in less than 4 laboratories. In addition, no overall comparsion of the performance of the alternative tests was carried out.

A further analysis of the results of the COLIPA study using additional data, and with more weight on the relevance of the alternative methods than the reproducibility, was the basis for a more detailed evaluation. The SKIN² ZK1200 model was demonstrated to be very good at predicting a broad spectrum of Draize test MMAS values, and the method could also predict individual tissue reaction caused by formulations and ingredients. All the 55 could be accommodated, and formerly promising results have been obtained by test of 132 materials with a very broad spectrum of irritancy potentials. The SKIN² ZK1200 model was also the only test in the COLIPA study that fully was able to reproduce the prediction model used. Preliminary results suggest that the SKIN²ZK1200 model may also be suited to studies of recovery from ocular irritancy. The interlaboratory reproducibility of the SKIN² ZK1200 method was relatively good, and modest reproducibility data are to be expected using methods, where tissues are exposed directly to crude test substances compared to e.g. methods using cellular monolayers being exposed to test substances dissolved in the medium. Like other in vitro systems, the SKIN² ZK1200 method is far more reproducible than the Draize test (see table 3.11).

Table 3.11
Historical data on the intralaboratory reproducibility of in vitro methods and the Draize test (from Bruner et al., 1996)

None of the other alternative methods in the COLIPA studiet were suited to predict a broad spectrum of Draize MMAS values for mixed ingredients and products. Several methods did, however, give good predictions of the in vivo response of ingredients, in particular of water soluble substances as surfactants. Both tests with chorioallantoic membranes of hens eggs had a very poor interlaboratory reproducibility. The reproducility of the other alternative tests was good. The neutral red uptake test and the red blood cell test were conducted by 4 or more laboratories, and these tests were considered to have a reasonable reproducibility (Brantom et al., 1997).

4. Test for recovery from ocular irritation

4.1 Method
4.2 Results
4.3 Discussion

Ocular irritancy is traditionally evaluated by use of the Draize test with albino rabbits receiving 0.1 mg or 0.1 ml of the test substance into the conjunctival sac of one of the eyes. The clinical responses of the cornea, conjunctiva and iris are observed until the initial lesions are cleared or up to 21 days. OECD guideline 405 on acute eye irritation/corrosion states that the eyes should be examined at 1, 24, 48 and 72 hours. If there is no evidence of irritation at 72 hours, the study may be ended. Extended observation may be needed if there is persistent corneal involvement or other ocular irritation in order to determine the progress of the lesions and their reversibility or irreversibility (OECD, 1987). The classification of ocular irritants by US-EPA/FDA is based on the use of maximum scores of corneal opacity, iritis and the conjunctival response, and the persistency of the lesions for more than 7 or 21 days (Gupta et al., 1993). The European Community is using the mean scores of the 24, 48 and 72 hour observation time for classification of R36 and R41 substances causing significant or severe ocular lesions, respectively. The persistency of the response is also evaluated, as the presence of ocular lesions at the 72 hours observation time or irreversible colouration of the eye are parameters releasing the R41 label (European Commission, 1996).

A wide range of in vitro assays for ocular irritation are being extensively used for screening purposes, but almost all of these tests have been designed to predict the severity of acute ocular effects of chemicals, and it has not been possible to demonstrate a correlation between the effects of chemicals over time in vitro and times to clear ocular damage in vivo. The Institute of Food Safety and Toxicology has as a part of the present project introduced a new testing procedure with SKIN² ZK1200 tissues to study the recovery from ocular irritation. Nine substances from the ECETOC eye irritation reference chemicals data bank were tested (see table 4.1).

4.1 Method

In the present exploratory study, the use of the three-dimensional ocular tissue model, SKIN^2TM ZK1200, was introduced for the study of tissue recovery after exposure to irritants. The study was designed to mimic the Draize test procedure with a topical exposure of the tissue specimens, and observation of the tissues at time points corresponding to observation times used in vivo. It was studied, whether the cellular viability of control tissue specimens could be maintained for periods of time comparable to the in vivo test procedure, and whether exposed tissues regained their viability in vitro in a manner comparable to days-to-clear the response in the Draize test.

The recovery model was developed before the identity of the COLIPA test substances was revealed to the participants. For this reason, the exposure times used in the recovery study were defined by a preliminary time range finding study, where exposure times reducing the MTT test response to 35-65% of the control level were established. The tissue specimens were exposed to the test substances according to the SKIN² ZK1200 dosing regime and the MTT test protocol used in the COLIPA study. The cellular viability of the tissue model was measured using the MTT assay immediately after chemical exposure and after incubation periods corresponding to observation times used in the Draize tests. The days-to-clear in vitro was defined as the number of days needed to obtain MTT activities in the exposed tissue specimens that were not significantly different from the matching control values.

Table 4.1
Days needed for exposed SKIN² ZK1200 tissues to regain the control MTT activity (days-to-clear in vitro) together with average and median values of days-to-clear in vivo and Draize test MMAS values.

*: Average of two Draize tests.

Control tissues and tissue specimens exposed to 1% benzalkonium chloride, 3% sodium lauryl sulphate, and 5% Triton X-100 were examined histologically. The tissues were fixed in 10% neutral buffered formalin, parafin embedded, sectioned, and stained with haematoxylin and eosin. The tissues were then subject to histomorphological examination of numbers of epithelial cell layers present and dermal/epidermal cell degeneration.

A one-way analysis of variance was used to analyse the results on MTT activities. Analyses of linear correlation were used for comparison of exposure times used in vitro and Draize test MMAS values and of in vitro and in vivo recovery data.

4.2 Results

The average MTT values for the unexposed control tissues remained fairly stable for up to 14 days of incubation, but after 21 days a considerable decrease in the MTT activity was observed. The mild ocular irritants (100% glycerol and 0.1% cetylpyrimidinium bromide) did not induce a significant depression of the cellular viability of the tissue specimens after 60 minutes of exposure, but significant decreases in MTT activities were observed after 2 to 3 days of culture. The MTT activity of the ZK1200 tissue specimens exposed to 3% sodium lauryl sulphate and 5% Triton X-100 returned to control levels within 1 day, and tissues exposed to 1% sodium hydroxide, 100% isopropanol and 100% methyl acetate returned to control MTT levels within 7 days. Tissue specimens exposed to various concentrations of benzalkonium chloride did not return to the control level of MTT activity in a stable manner. In tissue specimens exposed to 1% benzalkonium chloride for 5 seconds, the MTT activity was not affected after 1 day, but significantly lower MTT values than the matching controls were found on day 3, 7, 10, and 14. Tissues exposed to 1% benzalkonium chloride for 10 seconds did not return to control MTT levels during the 7 day observation period. Tissue specimens exposed to 10% benzalkonium chloride for 1 second did not regain control MTT values within 21 days.

Histological analysis of cellular degeneration and necrosis in SKIN² ZK1200 tissues.

Histomorphological analysis of control and exposed SKIN² ZK1200 tissues. Dermal/epidermal degeneration/necrosis: (+) traces, + mild, ++ moderate, +++ severe, ++++ very severe. n.d.: not determined.

A moderately good linear correlation (r = 0.73, p<0.05) was obtained between the exposure times used in vitro (log transformed) and Draize MMAS values. The number of days needed to regain tissue viability in vitro and Draize MMAS values is shown in table 4.1 together with the average and median days-to-clear the ocular responses in the Draize tests. Good linear correlations were found both between days-to-clear in vitro and days-to-clear in vivo using average (r = 0.92, p<0.001) and median (r = 0.91, p<0.001) values.

By histological examination, the number of epithelial cell layers were observed to vary between 1 and 3 in all tissue specimens analyzed (see table 4.2). No consistent cellular changes were seen on day 0, 1 or 3. On day 7 moderate degree of pyknosis and karyorrhexis of epithelial cells was observed in tissue specimens exposed to 3% sodium lauryl sulphate and 5% Triton X-100, while severe cellular changes including confluent areas of stromal cellular necrosis were present in tissue specimen exposed to 1% benzalkonium chloride. Only mild cellular changes were seen in the control tissue. No cellular regeneration was observed (Espersen et al., 1997).

4.3 Discussion

By histological examination no consistent cellular changes in the tissues were observed before day 7 after treatment, indicating that measurements of the MTT activity of the tissues appear to be a more sensitive marker of changes of the cellular viability. This is in line with results obtained by histopathological examinations and MTT conversion tests in skin organ cultures exposed to chemical irritants (van de Sandt & Rutten, 1995). The SKIN² ZK1200 model appeared to be adequate for determination of delayed cytotoxicity, as the viability of the system was stable for up to two weeks. The use of the model for more prolonged periods of time would be questionable due to the impaired tissue viability. A good overall correlation was found between the times for regain of a full MTT activity in the exposed tissue specimens and the in vivo recovery data. The results obtained suggest that the SKIN² ZK1200 model have shown promise for use in assessing chemicals and products for their ability to induce repairable or persistent cellular damage to the eyes.

5. Validation of alternatives to ocular irritancy tests

World-wide a lot of resources have been invested in the development and validation of reliable alternatives to eye irritation tests. Formerly, many small validation studies have been performed, and recently several large validation programs have been completed. The last studies were performed by EU/Home Office and COLIPA, and their goal was to provide the basis for a definitive conclusion regarding the replacement of animal experiments for ocular irritancy testing with alternative in vitro methods. This proces has, however, been considerably more complicated than expected.

The EU/Home Office study

In a recent study arranged by the EU Commission and the British Home Office, it was investigated whether 9 alternative methods could replace the Draize eye irritation test (see table 5.1). No significant correlations between results from the in vitro methods and Draize test MMAS values were found for 59 test materials. In addition, no reliable predictions of Draize test results were obtained, when the test substances were grouped in water soluble chemicals (n = 30) and non-water soluble substances (n = 18). Further, the alternative methods could not be used to identify severe irritants. The precision of the predictions of the Draize test results with the alternative methods was so low that the practical utility of the prediction was considered to be questionable. The only positive result was that relatively good correlations to in vivo eye irritation data were obtained with several in vitro methods for 12 surfactants (Balls et al., 1995).

Table 5.1

The BGA study

The outcome of the EU/Home Office study was not unexpected. The conclusion of a large German validation study on 136 mixed test substances was, for instance, that no significant correlation was found between in vivo eye irritation data and results from a cytotoxicity test with neutral red uptake. In contrast, the HET-CAM test was able to identify about 25% of the severe irritants (Spielmann et al., 1993). A database on 200 test substances has recently been established. Results for 9 substances were excluded from further analysis due to an unacceptable quality of their in vitro data. In addition, results for 48 chemicals were excluded because original Draize eye test data were not available. Very precise individual rabbit eye irritancy data from Draize tests performed according to OECD guideline 405 for up to 21 days have, however, been publiced on more than 25% of the excluded chemicals (ECETOC, 1998), and it would be interesting to reinclude these substances in the BGA database. By analysis of the database with the remaining chemicals, it was apparent that neither the cytotoxicity test nor the hens egg test could identify R-41 irritants with more than about 50% sensitivity, and an acceptable specificity of more than 80% could only be obtained with the HET-CAM test. For 129 out of the 143 remaining chemicals data from both types of in vitro tests were available, and a linear discriminant analysis was performed on combined endpoints from the two alternative methods. Using this procedure, a false-negative rate for R-41 substances of 29%, and a false-positive rate for other chemicals of 22% were obtained. The classification of R41 substances was, however, improved by including considerations to the solubility of the chemicals in water and oil (Spielmann et al., 1996) .

The CTFA study

The US Cosmetics, Toiletry, and Fragrance Association (CTFA) conducted from 1990 to 1996 a three phase validation program on approximately 25 alternatives to rabbit eye irritation tests. Low-volume eye test (LVET) MAS values were applied in the two first phases of the program, and Draize MAS values were used in the last phase. In both tests albino rabbits are used, but in the LVET test, 10 µ l of the test substance is instilled on the cornea, whereas 100 µ l is instilled in the everted lower lid in the Draize test. Regression modelling was carried out on results from tests showing the best in vitro/in vivo concordances in an initial analysis. The best fitting data transformations were in general two or three parameter logistic models.The HET-CAM test, the neutral red release test, and the EYTEX^TM test gave relatively good predictions of 10 hydroalcoholic formulations tested in phase I of the program (Gettings et al., 1996a), but none of the alternative tests were able to reliable predict the in vivo response of 18 oil-water emulsions in phase II (Gettings et al., 1998). Several tests did, however, give good predictions of Draize MAS values of 25 surfactant based products in phase III (Gettings et al., 1996b). The variability of the HET-CAM test and the neutral red release test appeared to consistently exceed the variability of the in vivo tests. The other alternative tests were considerably more reproducible than the eye irritancy tests.

The IRAG evaluation

The results of the EU/Home Office study were in line with the outcome of an evaluation that was carried out by a group of experts from various regulatory authorities (IRAG) in 1993. IRAG performed an overall evaluation of existing data from a large number of in vitro assays for ocular irritation. Fourty-one laboratories world-wide submitted 55 data set obtained with 23 in vitro methods on 9 to 133 test substances to IRAG.

In the IRAG study, results from tests with chorioallantoic membranes of hens eggs were in poor to moderate agreement with Draize test results for up to 93 mixed test substances. Using the HET-CAM test, good correlations to in vivo data were, however, obtained for surfactant based substances and products, and the CAMVA test gave the best prediction of alcohol based products (Spielmann et al., 1997). An assay with isolated rabbit eyes and a test with isolated bovine corneas (the BCOP-test) were both considered to have potentials for the identification of severe irritants, but no general prediction of Draize test results was obtained. Results from a test with isolated chicken eyes and an assay with isolated bovine lenses were both in good agreement with in vivo eye irritancy data, but the data sets of the two tests were too limited to allow a general evaluation (Chamberlain et al., 1997). IRAG considered various cytotoxicity tests to have a potential for the prediction of Draize test results for water soluble substances at normal pH values (Botham et al., 1997; Harbell et al., 1997). The EYTEX test showed a poor correlation to in vivo data for 454 cosmetic ingredients and formulations, but the predictions were good for individual groups of materials, e.g. petrochemicals and solvents. The bacterial Microtox assay was considered to be potentially useful for test of surfactant based products. The SKIN² ZK1200 tissue model was evaluated to be very useful in the screening of cosmetics and household products, when measurements of the viability of the tissues with the MTT test were used. The use of prostaglandin E₂ as an irritation marker was evaluated to be problematic. Further development of the SKIN² ZK1200 model was recommended, whereas the ZK1100 model with fibroblasts was not found to be suited for irritancy testing (Curren et al., 1997).

The JMHW/JCIA study

The Japanese Ministry of Health and Welfare (JMHW) initiated in 1991 a validation study together with the Japanese Cosmetics Industry Association (JCIA), national research institutes, universities and kit suppliers. A three-step validation study was initiated in 1993 and completed in 1996. Twelve alternative methods were evaluated using 38 cosmetic ingredients, and Draize tests were performed on the same lot of test substances.

Two SKIN² models were used, both with a MTT test protocol. The ZK1100 fibroblast model was used in 6 to 8 laboratories, and the ZK1200 co-culture model (TEA) was used in 2 laboratories for the full set of test substances, and in 6 to 7 laboratories for 13 test substances. Using the fibroblast model, the tissues were submerged in the culture medium and exposed to test substances dissolved in the medium. Using the TEA model, the reference substances were applied to the surface of the epithelium, but the protocol was not similar to the dosing regime used in the COLIPA study, since the test substances were dissolved or suspended in the culture medium at 10%. T₅₀ values were derived from MTT time-response graphs. High interlaboratory variabilities were observed with both tissue models, with an average CV of 44.5% (n=30) on the data obtained with the ZK1100 model and with an average CV of 61.9 % (n=9) on exact t₅₀ values obtained with the ZK1200 model in 6-7 laboratories. There was, however, total agreement between the 2 laboratories testing the full set of substances on establishing cut-off values for 17 compounds, and a very good correlation (r=0.84) was obtained by reanalysis of log-transformed t50 values on 16 of the remaining test compounds with exact t50 values obtained in the 2 laboratories. Relatively poor linear correlations were obtained to Draize test MAS values both with in vitro data from the ZK1100 model (r=0.71) and the ZK1200 model (r=0.63) (Kurishita et al., 1999). A reanalysis of the data obtained with the ZK1200 model in the two laboratories testing the full set of test substances was performed in the present study based on the prediction model used in the COLIPA study. This improved the prediction of the in vivo results, and relatively good correlations (r>0.78) were obtained between observed and predicted Draize test MMAS values. Exclusion of test substances incompatible with the culture media (acids, alkalies and alcohols of low molecular weight), however, considerably improved the in vitro/in vivo correlation obtained with the ZK1200 model (r=0.84) (Ohno et al., 1999).

A MATREX tissue model, consisting of human fibroblasts grown in a collagen lattice, was also used in the study. Test substances were applied neat to the surface of the tissue for 24 hours, and the concentrations causing a 50% decrease in the MTT response were determined (EC50 values). Alternatively, a MATREX score indicating the lowest concentration reducing the viability by 20-80% was used. A total of 12 laboratories used the model, but the full set of reference substances was only tested in 3 laboratories. The interlaboratory reproducibility was considerably better using MATREX scores (CV=9.6%, n=39) than using EC50 values (CV=34.6%, n=33)(Ohno et al., 1999). Similar correlations (r=0.67) were obtained between EC50 values and MATREX scores and Draize MAS values (Ohuchi et al., 1999). It appear to be possible to improve the predictive ability of the assay considerably, if a non-linear logistic prediction model is developed.

A poor correlation (r=0.31) to Draize MAS values was obtained with the EYTEX test (Matsukawa et al., 1999). In addition, moderate in vitro/in vivo correlations with a test based on measurements of denaturation of isolated bovine haemoglobin, and the interlaboratory CV’s exceeded 240%. An excellent correlation to Draize MAS values (r=0.91) was reported using 50% denaturation as an endpoint (Hatao et al., 1999). The correlation was, however, based on tests of 8 substances only, and it appeared to be due to clustering of the data. Large interlaboratory variations (CV’s>50%) were also observed with various hens eggs tests, and moderate correlations to Draize MAS values were obtained with the CAM tryphan blue absorption test (r=0.69, n=52) and the HET-CAM test (r=0.72, n=55) (Hagino et al., 1999). The predictive ability of the HET-CAM test may be considerably improved if a non-linear prediction model is applied. The results from various cytotoxicity tests were moderately to well correlated to Draize MAS values: normal rabbit corneal cells (r=0.53, n=28), the red blood cell haemolysis test (r=0.63, n=17), mammalian cell lines (r>0.71, n=29), and rabbit corneal SIRC cells (r>0.81, n=29-30). For these tests, the interlaboratory reproducibility was acceptable with CV’s ranging from 24% to 37% (Ohno et al., 1999).

The BCOP workshop, 1997

The predictive ability of the BCOP test has recently been evaluated by a working group of researchers from laboratories with a large in-house experience on the assay. A database of in vitro results on more than 200 test substances has shown concordances to Draize irritancy classes of 80-85%, and the assay has a good reproducibility (Sina and Gautheron, 1998). Tests of a large number of positive controls have shown an excellent intralaboratory reproducibility with CV’s of total BCOP scores ranging from 12% to 16% (Harbell and Curren, 1998). The fluorescein leakage test with MDCK cells has demonstrated a very poor ability to discriminate between Draize test irritancy classes, and measurements of fluorescein leakage through the corneas have been shown to be relatively non-predictive of ocular irritancy (Sina and Gautheron, 1998).

The COLIPA study

The results of the COLIPA validation study on alternatives to eye irritation tests confirmed the results of the previous studies. Most of the in vitro methods used in the COLIPA study were not suited for the prediction of acute eye irritation caused by mixed ingredients and products (Brantom et al., 1997). This is in line with the conclusion of the EC/Home Office study, where none of the alternative methods used were found to be promising candidates for replacement of the Draize test (Balls et al., 1995). In the COLIPA study, most in vitro methods appeared to be suited for the prediction of ocular irritation caused by cosmetic ingredients, in particular of surfactant based substances. This was also the case in the EU/Home Office study, and in several former validation studies (Rasmussen, 1993). In the COLIPA study, tests with chorioallantoic membranes of hens eggs had a very poor reprodicibility, whereas the other in vitro assays had relative good interlaboratory reproducibilities.

The most important result of the COLIPA study was that an in vitro model, SKIN² ZK1200, was demonstrated to give a very good prediction of a broad spectrum of eye irritancy data. The SKIN² ZK1200 model was the only alternative method that was able to fully meet the criteria on the reproduction of the prediction model used in the COLIPA study. In addition, the interlaboratory reproducibility of the SKIN² ZK1200 method was evaluated as satisfactory based on results obtained in 3 laboratories (Southee et al., 1999), and the method is like other alternative tests considerably more reproducible than the Draize test. The main outcome of the COLIPA study was that an alternative test for the first time showed a good potential for replacing animal experiments for acute eye irritation in a large blind validation study with mixed chemicals and products The SKIN² ZK1200 method has also been shown to be useful in the prediction of recovery from ocular irritation in a preliminary study (Espersen et al., 1997). For this reason, the method can be evaluated to show promise as a full replacement of the Draize test. The manufacture and sale of the SKIN² models ceased shortly after the COLIPA study was completed. This, however, does not make the findings with the systems insignificant, since general knowledge on the prospects for use of tissue equivalent models has been gained.

Alternative ocular tissue models

Other available eye models with tissues of keratinocytes grown on microporous membranes include an EpiOcular ^TM model and a REC model, which both have shown promise as tests that may replace the Draize ocular irritancy test. In both models, substances are applied neat to the surface of the tissues, and the endpoints used are based on time-response relations obtained with the MTT assay. Using the EpiOcular ^TM model, t50 values for 28 chemicals from the ECETOC databank were very well correlated (r=0.90) to Draize rabbit eye scores. Using a prediction model developed from this study, a good correlation (r=0.87) between predicted Draize scores and actual Draize scores of 41 finished products was obtained (Sheasgreen et al., 1996). In addition, relatively good concordance was found between t50 values obtained with 43 samples including liquids, powders and gels in the EpiOcular^TM model and Draize test classifications (Stern et al., 1998). A two parameter logistic prediction model for MTT data obtained with EpiOcular^TM tissues has recently been developed based on in vitro/in vivo data for 19 water-soluble chemicals and 41 finished products. A good correlation (r=0.90) was obtained between predicted and observed Draize MMAS values. Using the prediction model, a good prediction of Draize MMAS values (r=0.89) was later obtained for 11 finished products. The reproducibility of the MTT test with the EpiOcular^TM model appear to be good. Based on test of 132 samples, average CV’s of approximately 5% were obtained with negative controls, and average CV’s of approximately 25% were obtained with a positive control (0.3% Triton X-100) (Klausner et al., 1999).

The REC model resembles closely the EpiOcular^TM model, and a good correlation (r=0.89) has recently been found between MTT test data obtained with the REC model and Draize test MMAS values for 40 cosmetic formulations covering a broad spectrum of the scoring scale. In addition, acceptable CV% were found in a reproducibility study with 1% SLS (n=12, CV=18%) and a surfactant based product (n=15, CV=24%) (Doucet et al., 1998).

6. Evaluation of alternatives to ocular irritancy tests

Based on the results obtained in the EU/Home Office study, it was estimated that alternative methods replacing the Draize eye irritation test could only be established after 2005 (Purchase, 1997). In addition, the results of both the EU/Home Office study and the COLIPA study were characterized as disappointing in the EU Commission’s reports on alternatives as replacements of animal experiments for testing of cosmetics (EU Commission, 1995 and 1996b). On this basis, the ban against animal experiments on cosmetics implied in the EU cosmetics directive was postponed to January, 2000.

Recently, more optimistic view-points on the state-of-the-art of eye irritancy testing in vitro have been agreed upon on at workshops arranged by COLIPA (Bruner et al., 1998) and the EU Commissions center for validation of alternative methods (ECVAM)(Balls et al., 1999). The importance of understanding the mechanisms of eye irritation, particularly when attempting to improve in vitro prediction of in vivo eye irritancy, was emphasised at both workshops, and suggested research areas include:

	better characterisation of damage to the eye tissues, including the development of early markers for eye injury,
	increasing understanding of the effects of chemicals on the tear film and consequences of tear film disruption,
	characterisation of the acute and medium/chronic inflammatory response,
	validation of the area and depth of corneal injury as markers for eye injury,
	development of methods for assessing wound healing, pain, and the kinetics of the eye response, and
	development of methods for assessing persistence or reversibility of eye effects.

At the COLIPA workshop, there was a general agreement that the value of the Draize test for accurately assessing eye hazards in humans and for the development of alternative tests is limited. Therefore, the need for use of data on the human eye irritation response was stressed. It was recommended to identify a reference set of 10-20 model chemicals, selected on the basis of human eye irritation data when possible, for the development of mechanism based in vitro tests. It was recommended to assess the performance of the currently available in vitro tests with better in vivo data, and to develop and refine mechanism based in vitro methods for ocular irritancy testing.

The currently available in vitro methods, which are most likely to be relevant to eye irritation responses were identified as:

	cytotoxicity assays (including multilayered co-culture models),
	assays for measuring changes in epithelial function (e.g. the fluorescein leakage test),
	organotypic tests, as the BCOP test and tests with isolated chicken or rabbit eyes,
	CAM tests to model primary conjunctival effects, and
	long term models of the cornea (for example, cultured corneas).

An approach to the validation of in vitro tests for eye irritancy based on the use of reference standards was suggested at the ECVAM workshop, and ECVAM plans to validate further a test with isolated chicken eyes, the BCOP test, the combined used of the HET-CAM test and the neutral red uptake test, the red blood cell haemolysis test, and the EpiOcular^TM tissue model. In addition, it was recommended to compare the results from the EU/Home Office study and the COLIPA study with eye irritation classifications, and to evaluate the predictive abilities of testing strategies (combinations of in vitro tests.)

Prospects for progress in the replacement of eye irritancy studies

Today a very large body of data on the performance on various alternatives to eye irritancy tests has been generated. Various tissue models with topical application of the test substances have been demonstrated to be very good at predicting a broad spectrum of Draize test MMAS values, and the models appear to be able to accomodate all types of formulations and ingredients. Preliminary results also suggest that tissue models may be suited to studies of recovery from ocular irritancy. Tests with cellular monolayers being exposed to test substances dissolved in the culture medium appear to be limited to test of water soluble substances, and good concordances to eye irritation data have in general been obtained with surfactants. Results from various methods with isolated eyes or corneas have in several studies been shown to correlate well with in vivo data or to give good concordances to eye irritancy classes. The HET-CAM test appears to be able to identify some severe irritants, but in general it has been demonstrated to be less reproducible than the in vivo tests. In addition, very poor in vitro/in vivo concordances have in several studies been obtained with the EYTEX test.

Most of the alternative methods have, however, been documented to be far more reproducible than the in vivo eye irritancy tests. For this reason, much more weight should be given to an evaluation of the ability of the alternative methods to predict the in vivo response than to evaluations of their reproducibility. Test of a subset of chemicals may be a sufficient basis for an evaluation of the intra- and interlaboratory variability of the in vitro methods. Rigoristic demands to validation of in vitro methods may both lead to neglectance of valid alternatives and to acceptance of invalid tests..

The process of evaluating the information obtained appears to be much more complicated than expected, and a new, independent reevaluation of the existing data in line with the 1997 IRAG evaluation might very valuable. The establishment of prediction models for the alternative methods prior to their validation is crucial in order to obtain useable predictions of the in vivo response. For the majority of the alternative tests, it would be possible to develop more sophisticated prediction models based on advanced regression modelling on the relationships between historical in vitro and in vivo data.

The ban againts animal experiments on cosmetics implied in the EU cosmetics directive has been considered to be an impetus to the development and validation of alternative tests. However, the ban may be a major hinderance to the acceptance of alternatives, since a ban on products that have been subject to specific types of animal experiments in the European Union may be causing severe difficulties to the trade of cosmetics between the EU member states and other countries. The specific ban on animal tests on cosmetics and their ingredients appears also to be superfluous, since alternatives are to be used when available due to the EU directive on animal experimentation. An efficient move against improved possibilities to obtain a general acceptance of alternatives to the animal experiments required for safety assessements concerning cosmetics may be to delete the ban on animal experiments in the fortcoming revisions of the EU cosmetics directive.

7. Abbreviations

8. References

Balls, M., Botham, P. A., Bruner, L. H. and Spielmann, H.: The EC/HO international validation study on alternatives to the Draize eye irritation test. Toxicology in Vitro. 1995, 9, 871-929.

Balls, M., Berg, N., Bruner, L. H. et al.: Eye irritation testing: the way forward. Alternatives to Laboratory Animals, 1999, 27, 53-77.

Botham, P., Osborne, R., Atkinson, K., Carr, G., Cottin, M. and van Buskirk, R. G.: Cell function based assays. IRAG Working Group 3. Food and Chemical Toxicology, 1997, 35, 67-77.

Brantom, P. G., Bruner, L. H., Chamberlain, M. et al.: A summary report of the COLIPA international validation study on alternatives to the Draize rabbit eye irritation test. Toxicology in Vitro, 1997, 11, 141-179.

Bruner, L. H., Carr, G. J., Chamberlain, M. and Curren, R. D.: Validation of alternative methods for toxicity testing. Toxicology in Vitro, 1996, 10, 479-501.

Bruner, L. H., De Silva, O., Earl, L. K., Easty, D. L., Pape, W. J. W. and Spielmann, H.: Report on the COLIPA workshop on mechanisms of eye irritation. Alternatives to Laboratory Animals, 1998, 26, 811-820.

Chamberlain, M., Gad, S. C., Gautheron, P. and Prinsen, M. K.: Organotypic models for the assessment/prediction of ocular irritation. IRAG Working Group 1. Food and Chemical Toxicology, 1997, 35, 23-37.

Curren, R. D., Sina, J. F., Feder, P., Kruszewski, F. H., Osborne, R. and Régnier, J.-F.: Other assays. IRAG Working Group 5. Food and Chemical Toxicology, 1997, 35, 127-158.

deSilva, O.: Alternatives to eye irritation evaluation - the industry perspective. From: Lisansky, S. G., Macmillan, R. and Dupuis, J. (eds.): Alternatives to animal testing. Proceedings of an international scientific conference organised by the European Cosmetics Industry, Brussels, Belgium 1995. CPL Press, Newbury, UK, 1996, pp. 67-73.

Doucet, O., Lanvin, M. and Zastrow, L.: A new in vitro human epithelial model for assessing the eye irritating potential of formulated cosmetic products. In Vitro & Molecular Toxicology, 1998, 11, 273-283.

ECETOC (European Centre for Ecotoxicology and Toxicology of Chemicals) (1998). Eye Irritation: Reference chemicals data bank (second edition). Technical Report No. 48 (2). Brussels.

Espersen, R. J., Olsen, P., Nicolaisen, G., Jensen, B. L. and Rasmussen, E. S.: Assessment of recovery from ocular irritancy using a human tissue equivalent model. Toxicology in Vitro, 1997, 11, 81-88.

EU (1993): Directive 93/35/EEC. Official Journal of the European Communities, L 151/32, 23.6.1993.

EU Commission (1995): Development, validation and legal acceptance of alternative methods in the field of cosmetic products.

EU Commission (1996 a) Classification, packaging and labelling of dangerous substances in the European Union. Opdated version of Directive 67/548/EEC.

EU Commission (1996 b): Development, validation and legal acceptance of alternative methods in the field of cosmetic products.

Gettings, S. D., Lordo, R. A., Demetrulias, J., Feder, P. I., and Hintze, K. L.: Comparison of low-volume, Draize and in vitro eye irritation test data. I. Hydroalcoholoc formulations. Food and Chemical Toxicology,1996a 34, 737-749.

Gettings, S. D., Lordo, R. A., Hintze, K. L. et al.: The CTFA evaluation of alternatives program. An evaluation of in vitro alternatives to the Draize primary eye irritation test. (Phase III). Surfactant based formulations. Food and Chemical Toxicology, (1996b), 34, 79-117.

Gettings, S. D., Lordo, R. A., Feder, P. I., and Hintze, K. L.: Comparison of low-volume, Draize and in vitro eye irritation test data. II. Oil/water emulsions. Food and Chemical Toxicology, (1998), 36, 47-59.

Gupta, K. C., Chambers, W. A., Green, S. et al.: An eye irritation test protocol and an evaluation and classification scheme. Food and Chemical Toxicology, 1993, 31, 117-121.

Harbell, J. W. and Curren, R.: Report from the bovine corneal opacity and permaebility technical workshop, November 3-4, 1997, Gaithersburg, MD. Assay performance. In Vitro & Molecular Toxicology, 1998, 11, 337-345.

Harbell, J. W., Koontz, S. W., Lewis, R. W., Lovell, D. and Acosta, D.: Cell cytotoxicity assays. IRAG Working Group 4. Food and Chemical Toxicology, 1997, 35, 79-126.

Hatao, M., Murakami, N. Sakamoto, K. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (4) Haemoglobin denaturation test. Toxicology in Vitro, 1999, 13, 125-137.

Hagino, S., Kinoshita, S., Tani, N. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (2) Chorioallantoic membrane tests. Toxicology in Vitro, 1999, 13, 99-113.

Klausner, M., Sennott, H. A., Breyfogle, B., Makwana, A., and Kubilis, J.: The EpiOcular prediction model: a reproducible in vitro means of assessing ocular irritancy potential. The Toxicologist, 1999, 48, 336.

Kurishita, A. Katoh, T., Ohsawa, H. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (5) SKIN^2TMZK1100 and tissue equivalent assay. Toxicology in Vitro, 1999, 13, 139-151.

Matsukawa, K., Masuda, K., Kakishima, H. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (11) EYTEX^TM. Toxicology in Vitro, 1999, 13, 209-217.

Ohno, Y., Kaneko, T., Inoue, T. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (1) Overview of the validation study and Draize scores for the evaluation of the tests. Toxicology in Vitro, 1999, 13, 73-98.

Ohuchi, J., Kasai, Y., Sakamoto, K. et al.: Interlaboratory validation of in vitro eye irritation tests for cosmetic ingredients (6) Evaluation of MATREX^TM. Toxicology in Vitro, 1999, 13, 153-162.

Osborne, R., Perkins, M. A. and Roberts, D. A.: Development and intralaboratory evaluation of an in vitro human cell-based test to aid ocular irritancy assessments. Fundamental and Applied Toxicology, 1995, 28, 139-153.

Purchase, I. F. H.: Prospects for reduction and replacement alternatives in regulatory toxicology. Toxicology in Vitro, 1997, 313-319.

Rasmussen, E. S: Prospects for use of in vitro methods for assessment of human safety. National Food Agency of Denmark, 1995.

Rasmussen, E. S.: Use of reconstructed human tissue models in toxicological and pharmacological studies. Dansk Veterinærtidsskrift, 1/11, 1996, 943-948.

Sheasgreen, J. E., Kubilus, J., Sennott, H., Ogle, P. and Klausner, M.: Reproducibility and correlation of EpiOcular ^TM, a three-dimensional tissue culture model of human corneal epithelium. Alternatives to Laboratory Animals, 1996, 24, Special Issue, 284.

Sina, J. F. and Gautheron, P.: Report from the bovine corneal opacity and permeability technical workshop, November 3-4, 1997, Gaithersburg, MD. An historical perspective. In Vitro & Molecular Toxicology, 1998, 11, 316-326.

Southee, J. A., McPherson, J. P., Osborne, R., Carr, G. J. and Rasmussen, E.: The performance of the tissue equivalent assay using the SKIN^2TM ZK1200 model in the COLIPA international validation study on alternatives to the Draize eye irritation test. Toxicology in Vitro, 1999, 13. 355-373.

Spielmann, H., Kalweit, S., Liebsch, M. et al.: Validation study of alternatives to the Draize eye irritation test in Germany: cytotoxicity testing and HET-CAM test with 136 industrial chemicals. Toxicology in Vitro, 1993, 7, 505-510.

Spielmann, H., Liebsch, M., Kalweit, S. et al.: Results of a validation study in Germany of two in vitro alternatives to the Draize eye irritation test, the HET-CAM test and the 3T3 NRU cytotoxicity test. Alternatives to Laboratory Animals, 1996, 24, 741-858.

Spielmann, H., Liebsch, M., Moldenhauer, F. et al.: CAM based assays. IRAG Working Group 2. Food and Chemical Toxicology, 1997, 35, 39-66.

Stern, M., Klausner, M., Alvarado, R., Renskers, K., and Dickens, M.: Evaluation of the EpiOcular ^TM tissue model as an alternative to the Draize eye irritation test. Toxicology in Vitro, 1998, 12, 455-461.

van de Sandt, J. J. M. and Rutten, A. A. J. J. L.: Differential effects of chemical irritants in rabbit and human skin organ cultures. Toxicology in Vitro, 1995, 9, 157-168.

Weil, C. S. and Scala, R. A.: Study of intra- and interlaboratory variability in the results of rabbit eye and skin irritation tests. Toxicology and Applied Pharmacology, 1971, 19, 276-360.