Differential Expression Analysis

The 24 resultant .CEL files were analyzed using the R statistical computing language and environment (18). The data quality of each microarray was assessed by examining the average background, percent of probe sets called present by the MAS5 detection call algorithm (19), and the 3’:5’ ratio for GAPDH and ACTIN. Additionally, to detect potential spatial artifacts resulting from sub-optimal hybridization conditions, probe level linear models were fit using the R Bioconductor package, “affyPLM”, and plots of the residuals were examined for each microarray (20, 21). Differential expression was assessed using the “affy” and “limma” Bioconductor packages (19, 22). Briefly, probesets were quantile normalized and processed by the Robust Multi-Array Average (RMA) algorithm (23) before the manufacturer’s control probesets and probesets considered “absent” in 20 or more arrays by the MAS5 algorithm were filtered from all arrays (24). Differential expression between treatment conditions was then assessed via moderated t-test adjusted for multiple hypotheses by the Benjamini & Hochberg method (25). The false discovery rate (FDR) was controlled so that only those probesets where q < 0.01 were deemed significant.

Hierarchical clustering was performed on the top 100 prolactin induced and top 100 prolactin inhibited genes (as defined by the expression fold change between the PRL⁻ / DMSO and PRL⁺ / DMSO microarrays) utilizing the GenePattern public server (26–28). Microarrays were clustered using a pairwise average-linkage method and Pearson correlation as the similarity metric; genes were not subjected to clustering and are presented as originally ordered in the .GCT file supplied to the module.

Stat5 target genes were defined as previously reported (29) and include genes flanked by the classic Stat5 palindromic repeat binding motif as well as those identified by Kang et al. utilizing Stat5 ChIP-seq on murine mammary tissues at parturition. After converting Mus musculus gene annotations to human gene symbols, 847 unique genes comprised the Stat5 target gene list used in this study.