Identification of candidate maternal-effect genes through comparison of multiple microarray data sets.

Transcriptional profiling by microarray hybridization has become a standard method to analyze global gene expression and has resulted in the availability of enormous amounts of experimental data. Given the number of different microarray platforms currently in use, it is critical to determine how reproducible results are from one platform to another. Additional variability may also arise from tissue collection and protocol differences among laboratories. In an effort to identify genes whose maternal mRNA pools are critical during preimplantation development, we compared published results of three independent studies of the mouse preimplantation embryo transcriptome, each performed in a different laboratory using different microarray platforms. We searched the combined data set for genes whose expression patterns were consistent among the three experiments. Querying for presence or absence at single developmental windows indicates that between 52% and 60% of genes are in agreement among the three experiments. Searching for expression patterns across three developmental windows (oocyte + 1-cell, 2- through 8-cell, and blastocyst stage) revealed approximately 33% agreement among the three experiments, although the majority of these genes were either always present or always absent. Using this approach, we identified 51 genes with a predicted expression pattern of maternal RNA only (not present during 2-cell through 8-cell or at the blastocyst stage). RT-PCR validation indicates 37 (72%) of these candidates have the microarray-predicted expression pattern and represent candidate maternal-effect genes. Based on our analysis, we conclude that data mining microarray experiments in this way greatly enhances candidate gene expression pattern accuracy.