Reads aligned to known genes were analyzed by edgeR (Robinson, McCarthy, & Smyth, 2010); median alignments per sample was 20.7 M (minimum 17.7 M, maximum 24.1 M). Counts were normalized to counts per million reads (CPM) for each sample. Data were examined by Multidimentional Scaling in the edgeR package (Robinson et al., 2010) to detect outliers. Most samples clustered by experimental replicate, but one outlier was detected (Supplementary Figure 2), so all 3 samples from the replicate containing the outlier were removed from analyses, leaving 5 sets of 3 samples each: untreated, repeatedly treated (CIE), and repeatedly treated but withdrawn for 24 h (CIE+withdrawn). The MDS plot without the omitted samples (Supplementary Figure 3) indicates that there is a batch effect by replicate therefore analysis included replicate as a factor. To avoid analyzing genes that were not expressed or expressed at near background levels, only genes that had 2 or more CPM in at least 5 samples were retained for analysis (cf. (McClintick & Edenberg, 2006). Paired analysis was done based on experimental replicate: ethanol vs. controls, ethanol vs. ethanol
