2.6 Microarray data analysis
Data processing and statistical analysis were carried out with the
Bioconductor library limma (Ritchie et al. 2015). Background
correction and normalization were performed using the “normexp” and
quantile methods, respectively. We only considered probes whose
intensity was more than 10% above background on at least one
genotype/treatment combination. An empirical Bayes method with moderatedt -statistic was employed for the determination of the genes with
statistically significant changes, whereas the Benjamini and Hochberg’s
method was used to control false discovery rates (FDR). Differentially
expressed (DE) genes were identified from pairwise comparisons when FDR
< 0.05 and fold-change (FC) was > 2 or
< 0.5. Pathway over-representation analyses between lines or
treatment comparisons were performed with PageMan (Usadel et al.2006) using Fisher’s exact test with Bonferroni correction (FDR
< 0.05).
Based on the results of the multiple comparison test described above,
genes were defined as induced (FC > 2 and FDR <
0.05), repressed (FC < 0.5 and FDR < 0.05) or
unaffected in each of the four pairwise comparison combinations
(Stpfld 252 vs . WT under control conditions,Stpfld 252 vs . WT under drought, drought vs . control
in Stpfld 252 line, drought vs. control in WT line). An ad
hoc -made R script was used to group the genes sharing the same results
in the four pairwise comparisons, and thus defining the same cluster.
Graphics representing the resulting clusters were prepared with the R
library ggplot2 (Wickham 2009). Pathway over-representation in each
cluster was determined as described above. Mapman ontology was used for
functional annotation (Thimm et al. 2004) employing a mapping
file updated in May 2018 (stu_Agilent_4x44k_2018-05-25_mapping.txt)
from the GoMapMan website resource
(www.gomapman.org; Ramšaket al. 2014).