Principal Component Analysis of Tissue Scores from Substances used in the COLIPA Eye Irritation Validation Study

Principal component analyses (PCA) have been carried out on the tissue scores from Draize eye irritation tests on the 55 formulations and chemical ingredients included in the COLIPA Eye Irritation Validation Study. A PCA was carried out on the tissue scores 24, 48 and 72 hours after instillation of the substances. The first Principal Component (PC I) explained 77% of the total variation in the tissues scores and showed a high negative correlation (r=-0.971) with the scores used to derive the Modified Maximum Average Score (MMAS). The second component (PC II) explained 7% of the total variability and contrasted corneal and iris damage with conjunctival damage as in a similar analysis carried out previously on the ECETOC databank. The third component (PCIII), while only explaining about 3% of the variability, identified individuals treated with formulations that were observed to have low corneal opacity but large corneal area scores. This may represent some particular manner of scoring at the laboratory administering the Draize test or a specific effect of some formulations. A further PCA was carried out on tissue scores from observations at 1hr to 21 days. PC I in this analysis explained 62% of the variability and there was a high negative correlation with the sum of all the tissue scores, while PC II explained 14% of the variability and contrasted damage up to 72 hours with damage after 72 hours. A number of formulations were identified with relatively low MMAS scores but tissue damage that persisted. PCA analysis is thus shown to be a powerful method for exploring complex datasets and for identification of outliers and subgroups. It has shown that the MMAS score captures most of the information on tissue scores in the first 72 hours following exposure, and it is unlikely to be of any advantage in using individual tissue scores for comparisons with alternative tests. The relationship of the classifications schemes used by three alternative methods in the COLIPA study with the results of the PCA were investigated and the implications of the effect of persistence of tissue damage for various classifications schemes visualized.