WGCNA network analysis

The well-watered and drought time course datasets were filtered to remove any genes that did not reach an FPKM value of 10 in at least one time point in order to remove non-varying or low-abundance genes that introduce noise into the network analysis. Log2 normalized FPKM values were used to generate the co-expression networks using the WGCNA (RRID:SCR_003302) package in R (Team RC, 2016; Langfelder and Horvath, 2008; Langfelder and Horvath, 2012). Independent signed networks were constructed from the well-watered and drought time-course samples. An adjacency matrix was constructed using a soft threshold power of 16. Network interconnectedness was measured by calculating the topological overlap using the TOMdist function with a signed TOMType. Average hierarchical clustering using the hclust function was performed to group the genes based on the topological overlap dissimilarity measure (1-TOM) of their connection strengths. Network modules were identified using a dynamic tree cut algorithm with minimum cluster size of 30 and merging threshold function at 0.25. To visualize the expression profiles of the modules, the eigengene (first principal component) for each module was plotted using ggplot2 in R. To identify hub genes within the modules, the module membership (MM) for each gene was calculated based on the Pearson correlation between the expression level and the module eigengene. Genes within the module with the highest MM are highly connected within that module. To relate the physiology measurements with the network, the module eigengenes were correlated with the physiology data. Correlations were performed for each physiology trait separately using the mean values at each time point to associate the diel patterns between the physiology and eigengenes. To associate individual genes with the physiology we calculated Gene Significance (GS) as described in the WGCNA package as the absolute value of the correlation between gene expression and physiology across the time series.