RESUMEN
To illustrate the informative value of descriptive multivariate analysis in biochemical screening, we have analyzed several data matrices relating to the binding of steroids to the estrogen, progestin, androgen, glucocorticoid and mineralocorticoid receptors in different organs and species. We first compared dendrograms of steroid hormone receptors, that were obtained by an automatic hierarchical classification analysis of the binding data, to published phylogenetic trees of nuclear receptors based on amino-acid sequence analysis. The former classification describes the affiliations among the receptors as given by the binding specificity of a population of 187 steroids in a traditional cytosol binding assay (an indirect comparison of ligand binding sites); the latter describes the affiliations among the receptors as given by a comparison of selected primary sequences involved in ligand-dependent regulation of transactivation and dimerization. A similar hierarchical classification was also performed on the binding data of 62 steroids to myometrium cytosol from different species in order to show to what extent the progesterone-binding proteins in these species are affiliated. Hierarchical clustering methods classify each type of variable (receptor or steroid) independently. In order to be able to correlate both types of variable (receptors and steroids) on single-display graphs, it is necessary to resort to correspondence factorial analysis (CFA). CFA ranks the information content within the experimental system, highlighting major correlations and disclosing secondary correlations by eliminating redundant information and background noise. This multivariate method, applied to the analysis of published data, illustrated the particular specificity of estrogen binding in human vagina and raised the question of the nature of the binding protein in this tissue. Our examples are based on small data tables that can and have been analyzed de visu. However, it is certain that such descriptive multivariate techniques are indispensable for the analysis of large data banks not only to define structure-activity relationships but to estimate the degrees of affiliation among the biological variables being measured. Knowledge of such affiliations will help to organize available information in a context where the complexity of the biological systems under study is becoming increasingly apparent.