Chr 1, which was also suggestive in the original panel but dropped below the suggestive level in the recent BXD RI set. This could be due simply to sample size, which for RI panels is still quite low compared with other mapping populations. Therefore, mapping results are not robust to the removal of 50% of cases as was done in our analysis. However, the different results for different sub-populations more likely to illustrate the manner in which systematic differences in the new and old RI populations may conditionally affect QTL detection. Although these differences were not detected in the anova modeling as a main effect of sub-population, it is quite conceivable that these results reflect a slice through an epistatic interaction conditioned on loci that are fixed in one population or another. Our results suggest that this may even interact with sex, or that sex-specific loci may mask the detection of other QTL. Simulations can be used to determine whether the effect is due to low strain numbers or the actual composition of the population. It may prove necessary to include sub-population in the mapping model or make use of multiple-QTL modeling when using the entire combined BXD panel. Epistatic
