Use of an empirical Bayes screening approach and gene ontology annotations to filter and interpret microarray data from the uteri of estrogen-treated mice.

Fertuck KC¹, Eckel JE², Gennings C², Zacharewski TR¹.

¹ Department of Biochemistry & Molecular Biology, National Food Safety & Toxicology Center, and Institute for Environmental Toxicology, Michigan State University, East Lansing, MI.
² Department of Biostatistics, Virginia Commonwealth University, Richmond, VA.

While microarrays can be a useful tool in the screening of pollutants, dietary components, and industrial products for (anti)estrogenic activity in estrogen-responsive tissues, important basic information is still lacking in several key areas. In particular, transcriptional responses to estrogen itself must be better understood, and methods for distinguishing significant responses from baseline or irreproducible signals must be refined. Mu11KSubA GeneChip data (6523 probe sets) was examined from uterine tissue of duplicate immature, ovariectomized C57BL/6 mice treated orally with 0.1 mg/kg ethynyl estradiol or vehicle for 2, 8, 12, or 24 hr, or 3x24 hr. A nonparametric Bayesian approach was employed to identify significantly changing responses (881 probe sets), followed by an ANOVA to distinguish treatment and treatment*time effects (392 probe sets) from time-only effects. The 392 responses passing both filtering steps were clustered into 8 generalized temporal patterns by k-means, and annotated with Gene Ontology (GO) descriptions where available (193 probe sets). GO terms appearing at high frequency in specific k-means clusters, and therefore associated with specific temporal expression patterns, were found to be valuable in interpreting the cellular and tissue responses occurring over time in the uterine tissue in response to estrogen, both confirming published findings and suggesting novel estrogen-induced responses. Knowledge of the specific processes that are activated by estrogen in the uterus, and the corresponding temporal distribution of these responses, will be very helpful in the design and interpretation of assays aimed at the identification and characterization of xenoestrogens. Furthermore, the filtering, clustering, and annotation strategy described here addresses common microarray study issues of analysis and interpretation.