Before to the application of methods for detecting signatures of selection, we first assessed the phylogeny of the horse breeds to evaluate the phylogenetic position of DBP within the species. In agreement with previous studies, DBP was phylogenetically distant from other horse breeds while being relatively close to the Mongolian horse (MNG) [8, 9]. The distance patterns between horse breeds was also identified by PCA. However, THB and DBP had the greatest distance from each other, two German Warmblood breeds, Hanoverian (HNV) and Holsteiner (HLS), were not clearly distinguishable in either the phylogenetic tree and PCA, indicating the possibility of shared genetic components between the two breeds . Similar to the results from PCA, admixture analysis at K = 2 separated the DBP, Yakutian (YKT), MNG and THB horses from other populations, while at K = 3 the Standardbred (STB) horses separate from the remaining horse populations. Consistent with previous studies, we found that the DBP and THB populations are genetically homogeneous for all K values (K = 2 to K = 8) [4, 9].
In addition, LD decay analysis revealed a markedly lower level of LD across all genomic distances in DBPs than other breeds. The high LD in commercial breeds, especially for THB horses, could be a consequence of artificial selection for specific abilities, e.g. racing performance, in the breeding programs  while DBPs clearly have an ancient origin following a long-term natural selection. Also, computing the ROH per each breed showed the lower level in the DBP population compared to other horses. Here, we found high level of ROH for Sorraia (SOR) and THB populations, that is concordant with previous study in the six different horse breeds .
Potential independent of positive selection in the DBP population
Because of the small body size of DBP, that is notably less than average horse breeds, we focused specifically on the loci that may play more important roles in the rapid evolution of body size during the domestication process. Here, we used comparative genome analysis between DBP and THB breeds to identifying the genetic basis underlying the size variation among DBPs. In our broad spectrum analysis, we identified several previously reported genes that are associated with body size related traits. Highly significant candidate genes that may be potentially involved in body size traits are listed in Table 1. Within the regions showing extremely high values (top 0.01), both FST and XP-CLR methods showed BMP2 gene shared overlapped selection signatures as positively selected genes (PSGs) (Figs. 3A and 3B). BMP2, a bone formation-related gene, was found as one of the candidates on ECA5. This protein belongs to the TGF-β superfamily, which has diverse biological activities related to bone physiology and metabolism [17, 18]. Previous studies have found associations of the BMP-2 variants with bone and cardiac development, bone mineral density, as well as body size traits [19, 20]. Also in human, BMP-2 appears to be the most important BMP affecting the adult skeleton .
Another possible candidate gene, FGFR1, was found in one of the selection regions on ECA27 (top 1% cutoff for Fst and Pi methods) (Fig. 3A). FGFR1 is an important candidate gene that influences bone growth and skeletal development. Previous studies found that FGFR1 protein plays a critical role in formation of muscle and bone tissues [22, 23, 24, 25]. Considering the important function of FGFR1 in skeletal development, this gene is an important candidate for body size variation in mammalians.
Results from the detection of selection signatures revealed consistently high signal values in FST and XP-CLR analyses as well as high differences in allele frequencies for NELL1 gene, which is overlapped among candidate PSGs (Figs. 3A, 3B and 4). NELL1, encodes a mammalian cell-signaling protein (protein kinase C-b1, PKC-b1) that has been shown to regulate skeletal ossification [28, 29]. Overexpression of this gene in both human and mice induces craniosynostosis, the premature fusion of cranial sutures . Previous studies have shown that absence of Nell1 leads to decreased cell differentiation and cell proliferation in several organs such as heart, bone and cartilage tissue [30, 31]. Recently, an interstitial 11p14.1-p15.3 deletion involving the Nell-1 gene was also reported in associated with short stature in children . Moreover, it was demonstrated that the NELL-1 has potential roles as a bone-forming growth factor in sheep .
Body size is recognized as one of the most fundamental properties of an organism, affecting nearly all biological aspects. In the last decades, new insights from the genetic and physiological studies have refined our understanding of genetic basis of body size, as the target of positive selection in human and domesticated animals. Human body size is a polygenic trait affected by variants of numerous genes and their interactions with environmental factors. For example, hundreds of genetic variants, in at least 180 loci, with small effects, have impact on final human adult height . In contrast, several independent studies in domesticated animals have shown that changes in body size can be controlled by a few genes with large effects. For instance, it has been demonstrated that one specific haplotype defined by 20 SNPs spanning the recent selection sweep covering IGF-1 gene has a major effect on body size within all small dogs . A similar study has shown that one SNP within the strong linkage region of BMP10 gene explained around 22% of the overall body weight variance in five chicken lines . Also, one study on dairy and beef cattle revealed the variation in the average height can be controlled by only 10 genes in eight genomic regions .
So far, previous genome-wide association studies (GWAS) have revealed several QTLs for body size and stature in different horse breeds. Based on the standard additive model, Makvandi-Nejad et al. (2012)  identified four loci on the ECA3, 6, 9 and 11 that explained 83% of size variance in 48 horses, three each of eight large and eight small horse breeds. Using the same dataset of these 48 horses, a recent GWAS study involving both dominant and recessive mixed-model approaches as well as a genome-wide scan for signatures of selection based on the FST genetic differentiation and XP-CLR test, ANKRD1 gene was identified and validated as a novel candidate, explaining 7.98% of the genetic variance in body size of the American Miniature horse (AMH). Compared with the fixed status of all four loci identified by Makvandi-Nejad et al. (2012) , ANKRD1 gene could be applied in effective genotype-assisted selection for body size in AMH . In other independent studies, the differential SNPs in LCORL gene on ECA3 [10, 37, 38, 39, 40], ZFAT gene on ECA9 , TBX3 gene on ECA8  and LASP1 gene on ECA11  have also been shown to be strongly associated with body size traits in horses.
In this study, we have investigated the genetic basis underlying the body size variation in DBP. In our broad spectrum of analyses by three methods, we did not find any significant selection signal within or near genes which were previously identified as horse body size-related candidates. Instead, we observe that NELL1 gene likely played an important role in the evolution of the small stature of DBP, an ancient small pony that was evolved in the mountainous areas in southwestern China. In addition, some other loci on different chromosomes were also identified to be potentially involved in body development process. No evidence of directional selection for the detected genes in this study has been reported to date in other horse populations, suggesting that these genes have been probably selected independently for the short stature of DBPs.