Draize Rabbit Eye Test Compatibility with Eye Irritation Thresholds in Humans: A Quantitative Structure-Activity Relationship Analysis

TOXICOLOGICAL SCIENCES 76, 384 –391 (2003) DOI: 10.1093/toxsci/kfg242 Draize Rabbit Eye Test Compatibility with Eye Irritation Thresholds in Humans: A Quantitative Structure-Activity Relationship Analysis Michael H. Abraham,* ,1 Mostafa Hassanisadi,* ,† Mehdi Jalali-Heravi,† Taravat Ghafourian,* ,‡ William. S. Cain,§ and J. Enrique Cometto-Muñiz§ *Department of Chemistry, University College London, London, England; †Department of Chemistry, Sharif University of Technology, Tehran, Iran; ‡School of Pharmacy, Tabriz University of Medical Sciences, Tabriz, Iran; and §Chemosensory Perception Laboratory, Department of Surgery (Otolaryngology), University of California at San Diego, La Jolla, CA 92093 Received July 24, 2003; accepted September 3, 2003 Draize rabbit eye test scores, as modified maximum average score (MMAS), for 68 pure bulk liquids were adjusted by the liquid-saturated vapor pressure P°. These 68 adjusted scores, as log (MMAS/P°), were shown to be completely equivalent to eye irritation thresholds (EIT), expressed as log (1/EIT), for 23 compounds in humans. Thus, for the first time the Draize eye test in rabbits for pure bulk liquids is shown to be perfectly compatible with eye irritation thresholds in humans. The total data set for 91 compounds was analyzed by the general solvation equation of Abraham. Values of log (MMAS/P°) or log (1/EIT) could be fitted to a five-parameter equation with R 2 ⴝ 0.936, SD ⴝ 0.433, AD ⴝ 0.000, and AAD ⴝ 0.340 over a range of 9.6 log units. When divided into a training set of 45 compounds, the corresponding equation could be used to predict the remaining 46 compounds in a test set with AD ⴝ – 0.037 and AAD ⴝ 0.345 log units. Thus, the 91-compound equation can now be used to predict further EIT values to around 0.4 log units. It is suggested that the mechanism of action in the Draize test and in the human EIT involves passive transfer of the compound to a biophase that is quite polar, is a strong hydrogen bond base, a moderate hydrogen bond acid, and quite hydrophobic. The biophase does not resemble water or plasma, but resembles an organic solvent such as N-methylformamide. The Draize rabbit eye test (Draize et al., 1944) is the only widely used assay for the effect of substances on the eye. In view of the scientific, ethical, and economic concerns over the Draize test (Wilhelmus, 2001), it is not surprising that alternatives to the Draize test have been examined and that various calculation procedures have been published. An in-depth study (Brantom et al., 1997) of numerous alternative assays has been carried out, but the conclusion was that none of them could be regarded as a valid replacement for the Draize test. On the other hand, it has been suggested (Spielmann et al., 1998) that 1 To whom correspondence should be addressed at University College London, Department of Chemistry, 20 Gordon Street, London WC1H 0AJ, UK. Fax: ⫹44 (20) 7679-7463. E-mail: . Toxicological Sciences 76(2), © Society of Toxicology 2003; all rights reserved. 384 a combination of two in vitro tests could be used to identify severe irritants. One challenge in finding alternatives to the Draize test is that the available data cover compounds in a variety of physical forms (i.e., liquids, solids, and aqueous solutions). The actual mechanism of irritation may well not be the same across these forms, and this would preclude any general alternative test or any general calculation. Fragmentation schemes for particular chemical or biological effects attempt to relate the effect to structural fragments of molecules. These may be functional groups or just parts of molecules, such as the CH 3 or CH 2 fragments. Then an effect is assigned to each fragment, and predictions are made by summation of the fragment effects in a given molecule. Such schemes for the estimation of eye irritation have been reported (Enslein, 1988; Klopman et al., 1993), but most of the data used by Enslein were not Draize scores. Although Klopman et al. (1993) used Draize scores, these were used in conjunction with other judgments; unfortunately, the full list of compounds studied is not available. Other workers have restricted their analyses to pure organic compounds. Principal components analysis (PCA) and neural networks (Barratt, 1995, 1997; Chamberlain and Barratt, 1995) have been used to discriminate between irritants and nonirritants with reasonable success. On the other hand, investigation of a similar data set using linear combinations of descriptors and PCA (Cronin et al., 1994) failed to generate any general linear correlation of modified Draize scores and failed to observe any marked distinction between irritants and nonirritants by PCA. The modified Draize scores were defined as MMAS divided by the molarity of the pure liquid; the latter is given by 1000 times the density of the pure liquid divided by the liquid molecular weight. The descriptors of the compounds in the best linear equation were ClogP, where P is a calculated wateroctanol partition coefficient, the lowest unoccupied molecular orbital (LUMO), and a connectivity index. Cronin et al. (1994) correctly pointed out that use of a physically heterogeneous set of compounds, i.e., pure liquids, solids, and aqueous solutions, would make it very difficult to obtain any useful structure- 385 DRAIZE RABBIT EYE TEST AND EFFECTS IN HUMANS activity relationship (SAR) and so restricted the analysis to pure bulk liquids. Kulkarni and Hopfinger (1999) obtained a reasonable relationship, but only for a very limited set of 18 compounds in a training set and five in a test set. Patlewicz et al. (2000) restricted their analysis to cationic surfactants, and for this set of compounds found a very good fit of observed and calculated Draize eye scores using a neural network. What is surprising is that such studies have been made before any substantial connection between results of the Draize test in rabbits and the effect of the corresponding substances in man has been established. In a comprehensive review of the Draize test, it was noted that the anatomy and biochemistry of the rabbit eye are not the same as those of the human eye and that there were numerous physiological reasons, including low tear production, blink frequency, and ocular surface area, that such a test on rabbits might not adequately predict human effects (Wilhelmus, 2001). York and Steiling (1998) stressed the need to validate the Draize test against controlled human eye data, but noted that “there are no adequate human data.” What comparisons have been made between the effects on rabbits and the effects on humans have been confined to consumer products that are a mixture of various chemicals. Freeburg et al. (1986) examined four such products and showed that the low-volume Draize test correlated with effects on the eyes of humans better than did the normal-volume Draize test. Allgood (1989) also matched the low-volume Draize test against human experience for four sham (...truncated)