Targeted selection of experimental replicate groups is the likely driver of differential expression artifacts

The design of a scientific experiment forms the foundation for rigorous hypothesis testing and robust interpretations. In RNA-seq analyses where common model organisms are used to study the effects of gene perturbations, breeding schemes are typically designed to produce siblings with either mutant or wildtype alleles of a gene. Siblings are expected to possess reduced genetic variability, allowing the biological responses of a gene perturbation to be observed accurately, and above that of minimal background noise. However, as previously reported (Baer et al. 2024), inter- and intra-genetic variability of the parental subjects allows for the possibility of such between the progeny. What matters is not so much the genetic diversity, but more specifically the differences in polymorphic loci represented between replicate sample groups, which are constructed to perform comparative statistical tests. This is largely due to chance when selecting a subset of progeny at random to ultimately represent the RNA-sequencing samples. For many RNA-sequencing experiments this subset is often small due to minimise costs, and therefore the opportunity of skewing is likely. Additionally, when the subset of samples is not selected at random, particularly when defined based on genotype at a genetic locus, further skewing may artificially and unexpectedly be introduced. The end result is expression differences due to unequal representations of genetic features that influence gene expression, i.e. expression quantitative trai loci (eQTLs), which manifest as noise obscuring the desired signal and therefore confound interpretations. In this study we intentionally implemented an experimental design that opposes our advice from previous findings. We analyse the impacts of strain-specific differences in zebrafish, by outcrossing two different strains of zebrafish. Strain-specific differences were quantified by performing Differential Allelic Representation (DAR) analysis, consequently providing additional evidence that DAR contributes to false positives in differential expression analysis. We first assess the impact of DAR in isolation, where functional differences due to mutation are mitigated by equal representation within each group subject to comparison. We then extend our investigation by rearranging the same set of samples to form additional distinct experimental groups, allowing functional comparisons similar to our previous approach, but in the presence of elevated DAR. mRNA sequencing of 3 month old sibling zebrafish whole brains, consisting of 27 samples that were genotyped at two loci on chromosomes 14 and 17, with n=3 for each genotype combination. Genotypes at chromosome 14 are homozygous for Tubingen (TU) strain, homozygous for PK strain, and heterozygous TU/PK. Genotypes at chromosome 17 are heterozygous for the psen1 T428del allele modelling early onset familial Alzheimer’s disease, heterozygous for the psen1 W233fs allele modelling familial Acne Inversa, and wild type.