For a systems-level study of the transcriptome, we identified co-expressed genes using weighted gene co-expression network analysis (WGCNA) as described previously 42. Each tissue region was analyzed separately. All reliably detected genes were included in the network construction, and data from all treatments (feno, tesa, beza and saline) were used to detect co-expression patterns. Signed networks were constructed using R and custom functions available at http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/. The power was set to β=11, β=14 and β=7 for the amygdala, PFC and liver network, respectively, and minimum module size of 100, 150, and 100 genes for amygdala, PFC and liver, respectively 27. A dendrogram cut height of 0.99 was used for amygdala and PFC, and 0.995 was used for liver. Gene modules corresponding to branches of the dendrogram were labeled in unique colors. Genes whose profile failed to cluster in the network were labeled in grey. A module was treatment-responsive if it contained more genes with expression changes after PPAR agonist treatment than was expected by chance (hypergeometric p < 0.05).
