2.6 ⎪ Statistical analyses
RNAseq analysis was performed using the genomic analysis tools available
through Galaxy (Afgan et al., 2018). Quality of RNAseq runs was
validated by FastQC and adapter sequences were clipped using FASTQ
(Gordon & Hannon 2017). Reads were mapped to the A. thalianareference genome (TAIR10) and preliminary differential expression
analysis was conducted using HISAT and StringTie (Pertea et al.2015). Differential expression analysis was conducted using DESeq2 as
well as the calculation of adjusted P- values, which limit high
false positive discovery rates due to multiple testing (Love, Huber &
Anders 2014). Data can be accessed on the Gene Expression Omnibus at
GSE154349. Log2 fold-changes were transformed with the
regularized log function to minimize variance caused by low expression
genes, then clustered and plotted using pheatmap. In pheatmap, each
sample was clustered on the horizontal axis based on the similarity of
its transcriptome to the 23 other transcriptomes. On the vertical axis,
individual genes were clustered based on the similarity of their
expression profile across the 24 samples to the expression profile of
other genes.
Comparisons of two means were evaluated via Student’s t tests and
comparisons of multiple means evaluated via one-way analysis of variance
(ANOVA) coupled with post hoc Tukey–Kramer Honestly Significant
Differences (HSD) tests. The effects of genotype (e.g., CBF1–3
deficiency) and growth conditions as well as genotype response to the
growth conditions for the IT (IT & it:cbf123 ) and SW (SW &
sw:cbf123 ) genetic backgrounds were each assessed via two-way
ANOVA. Nonlinear curves were generated using 3-parameter exponential and
4-parameter logistic models. All statistical analyses, excluding those
of RNAseq data, were conducted using JMP software (Pro 15.0.0; SAS
Institute Inc., Cary, NC, USA).