Controlling litter effects to enhance rigor and reproducibility with rodent models of neurodevelopmental disorders

Abstract

Research with rodents is crucial for expanding our understanding of genetic and environmental risk factors for neurodevelopmental disorders (NDD). However, there is growing concern about the number of animal studies that are difficult to replicate, potentially undermining the validity of results. These concerns have prompted funding agencies and academic journals to implement more rigorous standards in an effort to increase reproducibility in research. However, these standards fail to address a major source of variability in rodent research brought on by the "litter effect," the fact that rodents from the same litter are phenotypically more similar to one other than rodents from different litters of the same strain. We show that the litter effect accounts for 30-60% of the variability associated with commonly studied phenotypes, including brain, placenta, and body weight. Moreover, we show how failure to control for litter-to-litter variation can mask a phenotype in Chd8^V986*/+ mice that model haploinsufficiency of CHD8, a high-confidence autism gene. Thus, if not properly controlled, the litter effect has the potential to negatively influence rigor and reproducibility of NDD research. While efforts have been made to educate scientists on the importance of controlling for litter effects in previous publications, our analysis of the recent literature (2015-2020) shows that the vast majority of NDD studies focused on genetic risks, including mutant mouse studies, and environmental risks, such as air pollution and valproic acid exposure, do not correct for litter effects or report information on the number of litters used. We outline best practices to help scientists minimize the impact of litter-to-litter variability and to enhance rigor and reproducibility in future NDD studies using rodent models.

Keywords: Animal models; Litter effect; Neurodevelopmental disorders; Rigor and reproducibility.

Fig. 1

Body weight, brain weight, and placental weight of wild-type and Chd8^V986*/+ mutant mice at E15.5. The variation across different litters is greater than the variation within a litter

Fig. 2

In treatment studies, pregnant females, and hence whole litters, are exposed to a substance (a, c, e). Litters are the experimental unit because they are randomized to the treatment. Offspring within a litter will be more similar to each other than to offspring from different litters and should be treated as subsamples or technical replicates. Similarly, in studies with mutant animals, individual animals coming from the same litter share prenatal and postnatal environments that render them more similar to each other than to animals from different litters (b, d, f). Most studies incorrectly ignore the litter and instead apply standard statistical tests to multiple animals per litter, thus failing to correct or control for relatively large differences between litters. Appropriate analysis can be conducted by a and b only using one animal of a given genotype and sex per litter (randomly selected), c and d using more than one animal per litter and averaging their values, or e and f using multiple animals per litter and applying a mixed-effects model for analysis. The mixed-effects model allows for each offspring to be used as a unit of analysis and treats the litter as a random variable in the ANOVA. Even though each offspring is used as a sample, the litter is still the statistical unit for exposure studies, as the dam was randomly assigned to the treatment condition. Sex of the offspring should be balanced across groups. It may be necessary to study male and female neonates separately, in which case a total of 1 male and 1 female per litter (and genotype if relevant), can be sampled

Fig. 3

a When litter effect is ignored, the variation in body, brain, and placental weights of wild-type (WT) and Chd8^V986*/+ mutant mice at E15.5 varies greatly. b Adjusting for the litter effect removes unexplained variation in the data and has the potential to unmask significant differences between groups