The results of the present study are consistent with the major conclusions obtained in our previous QTL analysis of this large intercross pedigree based on a less complete linkage map [2–4]. Despite the large phenotypic differences between the High-Weight and Low-Weight selected lines we were not able to detect any major QTL with a large marginal effect. Only five loci, Growth1, 4, 6, 9 and 10, reached genome-wide significance for at least one body-weight or growth trait and none explained more than 4.4% of the residual variance for a growth-related trait (Table 3). However, the importance of epistatic interaction as a mechanism contributing significantly to the remarkable response to selection was further strengthened because additional pairs of interacting loci were detected in the present study. We are currently using an Advance Intercross Line (AIL) to replicate and further explore interactions among these loci.
The HWS and LWS lines originated from the same base population that was formed by crossing seven partially inbred lines of White Plymouth Rock broiler chickens . That the results of the QTL analysis infer that no QTL with large individual effects segregated in the base population is consistent with the steady selection response in both directions that have been obtained during the 40 generations of divergent selection [1, 2]. This pattern is also consistent with a rather slow change in allele frequencies at QTL during the course of selection or a gradual release of selectable additive genetic variation from the epistatic loci .
The remarkable response to the divergent selection for juvenile body-weight in the HWS and LWS lines must reflect allele frequency changes between lines at those loci that have responded to selection. The chromosomal regions harboring such loci are expected to show reduced variation within lines and also a higher divergence between lines. Such selective sweeps are caused by hitchhiking of closely linked loci during selection . The size of a selective sweep depends on the number of generations that have passed since the QTL mutation occurred and the local recombination rate in the region. Thus, a recently derived mutation will be associated with a large haplotype block whereas an old mutation that well predates the initiation of the selection experiment is expected to be associated with a smaller haplotype block.
The outcome of the analysis of Fst and homozygosities based on the screening of 13,000 SNPs in the HWS and LWS lines are consistent with the results of the QTL analysis and the observed selection response. We found only minor differences in the average homozygosity as well as average Fst values within QTL intervals compared with the genome average (Table 4). The result suggests that selection has primarily been acting on standing genetic variation that existed well before the initiation of the selection experiment, which is supported by results of a recent simulation study exploring the potential role of epistatic interactions in response to directional selection . The causal mutation will occur in many haplotype combinations if it has been transmitted through a large number of meiotic events before it reaches a high allele frequency in the selected population and as a consequence the selective sweep will be short and not detected by an Fst analysis based on the rather sparse marker set used in this study (~1 polymorphic SNP/200 kb). The results suggest that the release of genetic variance due to epistatic interaction  is a more likely explanation for the long-term selection response in these lines rather than the occurrence of new mutations during the course of selection .
An indication of a selective sweep was, however, observed for Growth1 on chromosome 1 where there was a higher Fst value between lines than the genome average. The confidence interval for Growth1 was smaller than for the other growth QTL which facilitated the detection of footprints of selection. The homozygosity in this QTL interval for the HWS line was close to the genome average whereas the region showed almost complete fixation in the LWS line. The pattern suggests that while this locus has contributed to the selection for low growth in the LWS line it may have been selectively neutral in HWS line. The causal mutation(s) for this QTL is expected to be located within this region of high homozygosity. The region with near homozygosity in the LWS line spans from position 173.4 to 176.9 Mb on chicken chromosome 1, a region still too large and containing too many genes to pin-point strong positional candidate genes.
To be an exhaustive search for QTLs, a genome scan should be based on a marker set that covers the entire genome. The map used in the current study covered 93% of the assembled chicken genome a clear improvement to the ~80% coverage achieved with our previous microsatellite-based map. In particular, we have added markers on six microchromosomes that previously lacked markers. There are, however, still seven microchromosomes lacking markers in the current map. This is because they are missing in the genome assembly and no markers from these chromosomes have yet been reported. These missing microchromosomes are all small, on the order of five Mb or less but are expected to have a higher gene density than the macrochromosomes . It is apparent that parts of the chicken genome, including some microchromosomes, are difficult to clone in bacterial vectors [24, 25] and this is the reason why they are missing from the current genome assembly. Hopefully, the use of new sequencing technologies that do not require vector-based cloning will allow us to fill the holes in the current chicken genome assembly to eventually making a complete genome scan feasible.
If we compare our current QTL analysis based on an improved linkage map with our previous QTL analysis  there are as many as six previously reported loci (Growth2, Growth3, Growth8, Growth10, Growth11 and Growth13) that did not reach statistical significance in the present study. Our previous analysis included family as a fixed effect in the regression model whereas we decided to not include family as a fixed effect in the statistical model used in the present study. To test whether the new information provided by the improved linkage map or the change in regression model caused this discrepancy we reran the analysis including family as a fixed effect for those growth loci that we could not replicate. With a QTL model including Family as a fixed effect, the F-value associated with Growth2 increased from 5.4 to 6.3 and therefore reached the suggestive significance threshold. For Growth3, the F-value increased from 4.5 to 6.0 with family effect in the model. A more dramatic increase in significance was observed for Growth8 since the F-value increased from not reaching even the suggestive threshold to become genome-wide significant with an F-value of 9.3 when the Family effect was included in the model. It is very likely that these three loci (Growth2, 3 and 8) represents true QTL effects and at each locus the allele from the HWS line is increasing growth as expected from the line differences in growth. In contrast, for Growth10, 11 and 13 there was only a minor difference in statistical significance between the two models and in these cases it is more likely that the previous suggestive evidence for QTL at these positions in fact were false positives.