treated samples that were similarly randomized to individual and treatment across arrays. Complementary RNA synthesis, cleanup, and hybridization were as per the manufacturer’s instructions. Arrays were read, and data were exported using BeadStudio software (Illumina). Samples were subsequently processed in R using the R packages lumi, limma, and ComBat (55–57). Principal component analysis (PCA) was performed to identify samples within each treatment cohort with outlier expression for removal from further analyses. After QC and removal of duplicates and sample outliers, 1438 independent samples were suitable for further analysis. Raw data were transformed and normalized using robust spline normalization within lumi. Samples were then corrected for batch effects using ComBat. A linear model incorporating data from 59 untreated, incubated samples and 421 untreated, nonincubated samples was used to estimate the effect of incubation on gene expression. This was then regressed from all treated samples to minimize the influence of incubator effects on the analysis. Differential gene expression analysis was performed with the limma package. Pathway analysis was performed with IPA (Ingenuity Systems) to define gene networks, significant upstream regulators, and relationships with canonical signaling pathways. For network analysis of trans gene lists, a global molecular network based on the IP knowledge
