Transcriptional profile (qPCR) response to treatment
The 50 selected candidate transcripts (hereafter “genes”) were tested
to determine which genes showed a transcription response to either of
the treatments. Two genes (cfap58 , ubr4 ) were dropped from
the analysis due to failure of PCR amplification for most of the
samples, thus 48 candidate genes were included for the rest of the
study. To reduce the number of
independent variables and to avoid over fitting the models, we used
Principal Component Analyses (PCA) on the qPCR data for the 48 selected
genes using “prcomp” (which is a part of the R statistical analysis
package) and factoextra package (1.0.7) (Kassambara and Mundt, 2017) in
R (version 4.1.1). Based on a threshold of Eigenvalue > 1,
and % variance explained > 2%, the first nine PC axes
were selected. We used Linear mixed models (LMM) (lmerTest package
(v3.1.3)) (Kuznetsova et al., 2017) in R with the selected PC axes to
test for the effect of treatment (fixed effect), and the random effects
of dam, sire, fish body weight, tank ID and chip effect, with all
interaction terms for fixed and random factors on gene transcription
patterns. Chip ID, body weight, dam, treatment×dam, treatment×sire
effects were nonsignificant before FDR correction and were removed from
the model. When any of the nine PCs were found to exhibit significant
effects with any of the independent variables (treatments, dam, sire,
body weight, tank ID, or chip effect), we examined the individual gene
transcription loading values. We used fviz_contrib within the
factoextra package (1.0.7) to identify genes with contributions to the
PC greater than expected (Kassambara and Mundt, 2017). The identified
genes were included in a second analysis that used LMM with the
ΔCT values for the selected genes and the same
independent variables (treatment, dam, sire, body weight, tank ID and
chip effect), including all interaction terms for fixed and random
factors. Nonsignificant factors (Chip ID, body weight, dam, and all
interactions) were removed from the model and the analysis was re-run.
Lastly, a sequential Bonferroni P value correction was applied
for multiple testing correction (Rice, 1989).