The phenotypic consequences of inherited mutations and genetic engineering as well as resilience to systems perturbations and response to natural and artificial selection depend heavily on the genetic architecture of complex traits . This architecture includes features such as the number of genes, the nature of DNA sequence differences, the effects of dominance, the extent of epistasis, and the range of pleiotropic actions . But in most cases, we have little genome-wide sense of these features because most studies do not have adequate statistical power. However, new genetic resources, innovations in genotyping, sequencing and phenotyping technologies, and novel analytical methods now enable rigorous studies. Adequately powered studies have recently been reported in yeast, Arabidopsis, cotton, Drosophila, chickens, beetles and other species, with discovery of many complex trait genes (also known as quantitative trait loci, or QTLs) and their interactions [see for example refs. [3–7], but mammalian studies remain especially challenging [8, 9].
Chromosome substitution strains (CSSs), also known as consomic strains, provide a special opportunity to characterize the architecture of complex traits in model organisms . CSSs are made by introgressing individual chromosomes from a donor strain onto a host strain background. The first complete CSS panel in mammals was derived by replacing each chromosome in the C57BL/6J (B6) inbred strain with the corresponding chromosome from the A/J strain . Additional mammalian CSS panels have subsequently been made for mice  and rats . Although their genetic constitution is highly unusual, CSSs have many unique attributes for gene discovery, functional studies, and systems analysis [14–16]. In particular, by controlling the phenotypic noise of background genetic variation, CSSs have considerable power to identify QTLs and to characterize other genetic features such as epistasis and systems properties such as phenotypic buffering (so-called ‘ceiling and floor’ effects) that are often lost in the background noise of segregating populations [17–23].
With the C57BL/6J-ChrA/J CSS panel, we studied the architecture of complex traits in laboratory mice, focusing on gene number, the nature of their interactions, and related systems properties . This study found at least 466 QTLs among 90 blood, bone and metabolic traits. The picture of genetic architecture was definitive and unexpected. On average, 8 of the 22 CSSs in the mouse panel differed significantly from the host strain for each multigenic trait, with an average phenotypic effect of 76% of the parental difference. Epistasis was detected for 98% of the 41 multigenic traits where the parental strains also differed significantly. The median cumulative phenotypic effect was 803% (range 164% to 1,397%), demonstrating striking non-additivity across these 41 traits. In addition, chromosome substitution led to phenotypic changes that were with few exceptions in the direction of the phenotypic state of the donor strain. Finally, the genetics of the parental strains defined physiological boundaries that largely constrained the range of phenotypic variation. Similar results were found with a rat CSS panel , suggesting that these results may be a general feature of inbred strains of mammals, and perhaps of segregating crosses and natural populations, rather than idiosyncrasies of particular traits, strains or species.
Although our study provided new insights about complex traits, many important questions remain, such as the extent to which the close genetic similarity between the progenitor strains affected the results, and whether substituted chromosomes from more distantly related inbred strains would result in more extreme phenotypes and distinct genetic architectures. With increased divergence time, DNA sequence differences accumulate, affecting transcription control and protein functions, and as a consequence phenotypic variation typically increases. But extensive epistasis could buffer sequence differences and constrain the range of phenotypic variation. In the extreme case, genetic background often buffers the phenotypic consequences of loss-of-function mutations [16–19].
Progenitors of mouse CSS panels differ substantially in divergence times. The B6 and A/J strains were derived in the early 1900s from closely related founders among Castle’s mice  and have since been inbred for hundreds of generations (jaxmice.jax.org). Although these two strains have many distinct phenotypes (http://www.jax.org/phenome), they are highly similar genetically and their genomes are substantially from the Mus musculus domesticus subspecies and many segments of these genomes are identical by descent from a common founder population [25–27]. By contrast, two new CSS panels were recently reported where chromosomes from more distantly related inbred strains were used as a source of donor chromosomes for substitution onto a B6 inbred genetic background [12, 28]. PWD/Ph was derived from mice that belong to the Mus musculus musculus subspecies and that were trapped in the Czech Republic in 1972 . MSM/Ms belongs to the Mus musculus molossinus subspecies and was established from wild mice trapped in Mishima Japan in 1978 . It is estimated that Mus musculus musculus and Mus musculus domesticus diverged from their common ancestor around 350,000 to 500,000 years ago whereas the divergence time between Mus musculus molossinus and Mus musculus domesticus is estimated roughly around 1 million years [30–33]. Our study design takes advantage of the introgression of chromosomes from genetically distant strains onto the same host strain background as the C57BL/6J-ChrA/J CSSs.
In this report, we tested whether genetic divergence between host and donor progenitors for CSSs affected the genetic architecture of complex traits. This test was based on comparing a panel of traits for these two new panels of CSSs with results for the original C57BL/6J-ChrA/J panel. The A/J strain diverged from the B6 strain hundreds of generations ago, whereas the progenitors of the PWD and MSM strains diverged from B6 millions of generations ago . As a result, gene regulation and protein functions should differ substantially in the more divergent PWD and MSM strains, both of which show a ~4-fold higher rate of DNA sequence differences than the A/J versus B6 comparison [25–27]. For both panels, we found many large-effect QTLs and pervasive epistasis, as well as strong systems properties that together corroborated results of the original study [14, 15]. More importantly, these genetic and phenotypic properties were remarkably similar among the three CSS panels, suggesting that genetic divergence did not have a strong impact on measures of phenotypic variation and that systems properties resulting from pervasive epistasis limited the range of phenotypic variation.