Variance components were calculated using the VarComp procedure of SAS, declaring the variables Year, and replicates in the case of Brazil data, or Sites (year × site) in the global analysis, as fixed.
Experimental conditions that differed among replicates were treated as fixed effects in the regression, and included as interaction terms.
All other variables were tested performing an analysis of variance (ANOVA) with replicate and ploidy as fixed effects.
After normalisation probesets with intensity >100 in any of the experimental conditions were analysed using Bayesian linear modelling in the limma package [55], with replicate and treatment as fixed effects.
Challenge, day collection, and replicate were considered as fixed effects.
The experimental design equated to a four-way factorial analysis of variance (ANOVA) with thermal selection regime and cyanine dyes (Cy3, Cy5) in a flip dye design as fixed effects, replicated populations as a random factor nested in thermal selection regime, and slide (spotted microarray) as a random factor nested in thermal selection, dye, and replicate.
The study was performed from a design-based perspective, taking the populations as fixed and replicating the sampling procedure with 1000 Monte Carlo simulation runs.
In the model the treatment and the dye effects were treated as fixed, and the replicate and the array effects within replicate or treatment by dye effect as random.
NE values were compared among treatments using 'Treatment' (with the levels 'Control', 'SM'RM'RM' or 'Both mites') as fixed factor and with the factors 'Experimental replicate', 'Biological replicate' and 'Technical replicate' included as random factors in the model.
To select E2-responsive genes, each gene was subjected to a mixed-model ANOVA allowing for treatment, time, and the treatment by time interaction as fixed effects and replicate study as a random effect.
