Approximately 25% of the world sheep population comprises fat tail breeds that are grazed in a wide range of countries especially in Asia and northern parts of Africa. In these typically arid countries sheep breeding has an important place in the local economy and as a source of protein. Due to the overt and easily defined nature of a fat tail, the gene variants affecting the phenotypic expression of this trait are a topic of both theoretical and economic curiosity. The former from the perspective of the impact and nature of domestication and the latter because the trait is now markedly less important commercially due to improved forage availability and decreased price for the product.
In this research we have developed an initial selection map for the fat tail trait in sheep. The objective was to localize genomic regions potentially affecting the trait. This was undertaken by comparing fat and thin tail breeds using the newly available 50K SNPChip. The chip has already proven its ability to map causal mutations for traits showing Mendelian inheritance such as yellow fat  and microphthalmia .
It has been previously shown in cattle, that domestication, subsequent breed formation and artificial selection, leave detectable signatures of selection in several regions of the genome . Given the similarities in domestication time and selection between cattle and sheep, it is likely that the ovine genome contains similar signatures.
The Iranian breeds were chosen for this study, because they originated both near the centre of domestication  and near the first recorded archaeological evidence of fat tail sheep [3, 7]. It was felt this geographic proximity would reduce false positives due to bottlenecks and selection pressure due to alternative factors such as climate, disease and pasture types. However, we expected that some signals would still be spurious so we then independently validated the results using a different set of breeds obtained from the Ovine HapMap project. The latter project did not formally phenotype the individuals concerned so these breeds were classified by us solely on publicly available breed descriptions. It is important to note that the selection sweep approach is one of the few viable approaches for investigating the genomics of this trait given the poor level of historical recording and DNA sampling in the relevant breeds. It is also one of the most cost effective and powerful techniques albeit as previously stated also subject to potential false positive signals.
Several statistical approaches have been devised to detect evidence of selective sweeps. These different statistics detect different aspects of the pattern of variation left by selection of a beneficial mutation. Generally, the power of the tests also depends on the strength, duration and age of the selective sweep . Unlike FST, tests based on linkage disequilibrium like iHS  and XP-EHH  which are used primarily on human SNP data, where there are now millions of SNPs available, are dependent on SNP spacing and frequency as they are multi-marker tests. The 50,000 SNPs available for sheep may not provide enough information for these tests; it has been suggested in cattle that 150,000 evenly spaced SNPs would be required to study selective signatures in all parts of the bovine genome . Additionally, the power of these methods also depends on the ancestral allele information which is still available for only a portion of the SNPs on the ovine chip.
However, it should be considered that robust inference of recent positive selection using FST is complicated by the fact that the distribution of genetic variation effect due to selection can be difficult to distinguish from that which arises after certain demographic events. Some previous studies that have attempted to identify signatures of selection based on patterns of population differentiation have used simulations to obtain the expected distribution of FST under the assumption of no selection [24, 33, 34]. However, the simulated distribution of FST is strongly dependent on its underlying assumptions about population demographic history, which is rarely known with any degree of certainty [24, 37]. Despite this problem, examining several thousand loci using newly available SNP chips provides an exciting opportunity to distinguish between the effects of population structure, positive selection and ascertainment bias. Demographic events and ascertainment bias would be expected to alter patterns of FST across the whole genome in a similar way, while selection events would be expected to alter FST values only in selected and nearby loci [29–34].
In general, we found a low differentiation between populations, with a mean of 0.024 (SD = 0.035) for the Zel-Lori Bakhtiari data set and 0.027 ± 0.038 for the HapMap data set. This agrees with Kijas et al.  who, using 23 domestic breeds and two wild sheep species, revealed that sheep breeds showed generally low differentiation which is consistent with their short evolutionary history.
The general pattern of the signals across chromosomes was different in the HapMap data set and there were differences in the locations of the major peaks (data not shown). However, Barendse et al.  observed similar results when comparing Bovine Australian and HapMap data. Their windowed FST values between two data sets were correlated with r = 0.346, whereas this correlation in our data sets was r = 0.413. In general this correlation with different breeds may suggest that each study had identified signals of divergence particular to the genetic history of those breeds, and only some of which may be due to selection (as described by Barendse et al. ). This correlation for our candidate regions was improved to r = 0.963.
We found few genes in the regions of interest and no particular candidate genes related to fat deposition were identified. However, the current annotation of the cattle genome is not as comprehensive as in humans, so these areas cannot be dismissed as not containing any genes or regulatory elements. It has been observed that while some proposed candidates for selection have strong support in the form of a functional mutation with an identified phenotypic effect, often the functional target is unknown . As the cattle genome becomes more comprehensively annotated in these regions likely targets of selection may be identified.
A variety of alternative comparisons, with the various thin and fat tail breeds using the combined HapMap and Zel-Lori Bakhtiari data set, can be examined in addition to the comparisons presented here. In a comparison, when we compared some fat tail breeds together (Afshari and Lori Bakhtiari breeds), we found the FST peak in the same location on chromosome 2 region 1 which was significant based on Fisher's combined test. The orthologous area for this region (using Blat) on the cattle genome would be around 8:62,600,000-63,000,000 (Table 2). Several bovine studies have also detected signatures of selection in this region. MacEachern et al.  compared differences in allelic frequencies of Australian Angus and Holstein cattle at 7,500 SNPs. They reported a region with large differences among breeds at 61,300,000 to 62,500,000 bp on BTA8. Stella et al.  observed a significantly large signal of selection in the same location for the Holstein and Jersey breeds. Although neither study has reported any particular candidate gene for this observation, previously identified QTLs (Table 2) could suggest a candidate region for body composition and carcass traits in this chromosomal region.
Using median homozygosity plot in these regions, we identified that homozygosity has been increased on chromosome 5 and X in favour of fat tail breeds. This is consistent with a recessive mode of inheritance. Therefore future studies should track both the phenotype of this trait and the genotype status in these regions in F2 crosses to provide independent and causal evidence that these regions do in fact affect the fat tail trait and to verify its mode of inheritance.
Given that one of the most differentiated regions was located on chromosome X, this suggests sex linked differences could be present. In several unpublished experiments when a research institute in central Iran crossed wild sheep (male) with a couple of fat tail ewes, according to the reports all resulting lambs were thin tail. This suggests dominance of the thin tail phenotype and does not support a simple single gene located on sex chromosome alone. Similarly, when two European thin tail breeds were crossed with Iranian fat tail ewes, all of the more than 350 crossbred lambs of both sexes subsequently born were thin tail and no fat tail lambs were observed. This supports the previous observations and again does not support sex linked effects, unless some epistatic effects exist or any relevant single gene located on chromosome X should be located in a pseudo-autosomal regions.
In general, there is little published information related to tail fatness in crosses between native fat and thin tail sheep breeds and also the relatively small number of animals used in these studies makes it difficult to obtain reliable estimates pertaining to this question [10, 11, 13–18]. However, in several experiments measurements suggest crosses were intermediate between fat and thin tail sheep [11, 17]. This conflicts with the unpublished Iranian results and does raise the possibility that the results observed may depend on the fat and thin tail breeds used in the cross. In our opinion, more detail and larger scale experiments are needed to confirm the results on growth and carcass characteristics of crossbred lambs of different sexes. If such studies are undertaken the nature of inheritance on the X chromosome can be rather difficult to elucidate, especially if imprinting is also suspected .
A result which is irrelevant to the inheritance of the trait, but provides an insight into a possible mechanism of fat deposition in this organ, are the results of Gökdal et al.  who examined the effects of docking in fat tail breeds. The carcasses of the docked group contained more kidney, pelvic and internal fat than the intact lambs as well as a higher percentage of subcutaneous and intramuscular fat. The weights of the different carcass cuts of the docked lambs were also heavier that those of the intact group. However, there was little change in overall carcass composition suggesting that the genes affecting the fat tail phenotype are associated with the localization of fat stores to a regional depot rather than control of the overall level of fat deposition. This observation also may provide support to the suggestion that this trait is a result of human mediated selection as it is difficult to postulate any natural selection criterion favoring deposition in this area.
Finally, a search for comparable depots in other species especially grazing mammals identified Bacterian humps in camels and humps in bos indicus cattle breeds as possible analogous features that have been selected for in animals grazing in arid and tropical regions subjected to wide fluctuations in food supply. The closest equivalent structure present in all mammals is possibly the mammary gland with its associated mammary fad pad. The current work, while not able to answer these hypotheses about the origin of the fat tail and analogous structures in other species directly, does offer hope that these questions will soon be able to be answered by further experimentation.