Our aim was to study differences in LD and haplotype variation between micro- and macrochromosomes, using a very high marker density across populations reflecting commercial diversity as well as fancy breeds and wild chicken. Chromosomal regions were chosen to represent the maximum range in size of macro- (GGA1 and GGA2) and microchromosomes (GGA26 and GGA27); the two microchromosomes were among the smallest well-assembled chromosomes available within the current genome build . Selection of SNPs was based solely on position (with the requirement of having 1 SNP every 2 kb), and thus systematic bias due to SNP selection was unlikely. Populations were chosen to reflect variation in the degree of polymorphism, and hence expected LD, to the widest possible extent, with the white egg layer at the lower end and Red Jungle Fowl at the upper end . Extent of LD in chicken has been studied before [2, 5, 6] but these studies were limited in numbers of markers, marker density, population sampling or sampling across chromosomes to accurately and comprehensively asses LD to the same degree as the present study.
Based on higher recombination rates in microchromosomes compared to macrochromosome differences in LD and haplotype variation were expected, but measures of these differences have not been previously reported. We found LD, HH, haploblock structure, and haplotype sharing all consistently lower for microchromosomes compared to macrochromosomes when measured using physical distance. A direct effect of recombination on these measures comes from changing the relationship from physical distance to genetic distance. From the fit of LD to the Sved equation , and assuming that Ne is the same for all chromosomes, the recombination rate was estimated to be on average 2.8 times higher at the microchromosomes (Table 2). This difference is less than the expected 4.5 times higher recombination rate for microchromosomes compared to macrochromosomes . The recombination frequency for the microchromosomes based on LD, therefore, appears to be systematically underestimated for all populations. While regional differences in recombination frequency are expected, currently no recombination map is available providing information at the scale of the present study (< 1 cM scale), not even for the macrochromosomes. For the smallest microchromosomes current recombination maps are even less detailed.
The inferred rate of 2.8× smaller recombination rate for macrochromosomes compared to microchromosomes, which is inconsistent with previous estimates (~4.5×, ), is due to a bias in the analysis from fitting the Sved equation across the same physical distance in micro- and macro chromosomes. LD at different distances has been shown to relate to effective population sizes at different numbers of generations by 1/(2 c), where c is the median distance between markers in Morgan . By performing local fits to the data, using SNP distance bins that are similar in genetic rather than physical distances, the systematic difference in Ne between micro- and macrochromosomes disappears. For most populations, past population sizes derived from both classes of chromosomes become quite similar when measured against genetic distance (Additional File 4).
Since Ne does not seem to deviate systematically once distances are properly corrected for differences in recombination rate, the main explanation for observed differences in heterozygosity, genotype differentiation, and derived allele frequencies in the microchromosomes is higher mutation rate. Higher heterozygosity is known to be positively correlated to recombination rate [16, 28], although the mechanism is not fully understood. We found derived allele frequency to be slightly higher on the microchromosomes, which suggests a higher evolutionary rate. A higher evolutionary rate for microchromosomes has been found before in a comparison between chicken and turkey macro- and microchromosomes . Higher levels of differentiation could result from increased background directional selection for higher GC content in the microchromosomes. The effect of directional selection would have the same effect as a smaller effective population size. Since there is no evidence for differences in Ne a higher mutation rate seems to be the best explanation for higher genetic differentiation on the microchromosomes.
The Sved equation assumes a static population size . However, the fact that the observed values or r2 (Figure 1) show more of a flat line compared to the expected values of r2 based on the fit to the global Sved equation is an indication of declining population size[22, 24]. Fits based on local inter-marker distance-bins reveal declining effective population sizes as shown in Additional File 4. Differences in LD and derived effective population sizes are largely consistent with known population histories, with white egg layers known to be more inbred than other breeds, while most of the commercial broiler lines are considered outbred [18, 30, 17]. Nevertheless, the dam broiler E5 has been a closed line for many generations (AV, unpublished results), which explains higher LD and HH in this population. The decline of effective sizes for the eight populations is consistent with earlier findings of substantial loss of allelic variation in domesticated chicken , reemphasizing the concern to maintain genetic diversity in this species.
In humans, markers diagnostic for haplotypes, so called tag SNPs, are often transferable between populations because of haplotype sharing and populations having common haploblock boundaries . In the chicken, haploblock boundaries show little overlap between populations, and haplotype sharing between populations is low. This difference between the two species could be the result of differences in demography, with the block-like structure of haplotype variation in humans being the result of population expansion in the past 10+ thousand of years originating from a population with an effective size of thousands to tens of thousands at most . Conversely, the present study finds evidence for population contraction in chicken, which is consistent with the relatively small number of long haplotypes and levels of haplotype homozygosity. These long current haplotypes are expected to be a mosaic of a much higher diversity of small past haplotypes, similar to what is observed in dogs . The ancient small haplotypes that make up today' s longer haplotypes, therefore, do not result in a very high r2 (unless inbreeding becomes very high and only a very small number of haplotypes remain). They do, however, result in high D' as the inbreeding erodes away many of the possible - and previously existing - haplotypes in a population. D' will more often result in high LD values when only part of all possible haplotypes are present compared to r2[32, 33]. Since the block construction methods applied here were based on D' it was not surprising to find considerable block structure - albeit often found in small blocks.
The block structure in the genomes of layers may be exploited to make genome-wide marker assays with 10,000 to 20,000 well chosen tag SNPs that would cover around 70% of the genome, supplemented by a similar number of SNPs to survey the remaining ~30%. Since block structure is mostly at the scale of < 10 kb for the more outbred broiler populations, and LD (measured as r2) near 0.2 at a similar scale, the number of informative SNPs would need to be > 100,000. However, as tag SNPs are probably not highly transferable between commercial populations, a general purpose assay might even need many more markers than 100 K.
Understanding sharing of haplotypes between populations is of further importance as it determines the success of transferring genetic parameters from one population to another . The present study confirms the findings of Andreescu et al  in that high overlap in haplotypes between broilers exists. However, it appears to only exist between closely related populations. Transferability of marker information between more distantly related populations may be problematic. For the microchromosomes, haplotype sharing is very small even among the broilers, showing that population-to-population transferability of marker information should be treated differently for micro- and macrochromosomes at the same physical scale. Since genotype differentiation is also systematically higher for the microchromosomes, differences in haplotype sharing are likely the result of both increased mutation rate and recombination frequency for the microchromosomes.