The Origin of a Coastal Indigenous Horse Breed in China Revealed by Genome-Wide SNP Data

The Jinjiang horse is a unique Chinese indigenous horse breed distributed in the southern coastal areas, but the ancestry of Jinjiang horses is not well understood. Here, we used Equine SNP70 Bead Array technology to genotype 301 horses representing 10 Chinese indigenous horse breeds, and we integrated the published genotyped data of 352 individuals from 14 foreign horse breeds to study the relationships between Jinjiang horses and horse breeds from around the world. Principal component analysis (PCA), linkage disequilibrium (LD), runs of homozygosity (ROH) analysis, and ancestry estimating methods were conducted to study the population relationships and the ancestral sources and genetic structure of Jinjiang horses. The results showed that there is no close relationship between foreign horse breeds and Jinjiang horses, and Jinjiang horses shared a similar genetic background with Baise horses. TreeMix analysis revealed that there was gene flow from Chakouyi horses to Jinjiang horses. The ancestry analysis showed that Baise horses and Chakouyi horses are the most closely related ancestors of Jinjiang horses. In conclusion, our results showed that Jinjiang horses have a native origin and that Baise horses and Chakouyi horses were key ancestral sources of Jinjiang horses. The study also suggested that ancient trade activities and the migration of human beings had important effects on indigenous horse breeds in China.


Introduction
There are over five million horses in China, the majority of which are indigenous horses that are distributed widely across the rural areas of China. The largest horse populations exist in the northern and southwestern provinces [1]. The Jinjiang horse is the only Chinese indigenous horse breed (CHB) located in the southeast coastal areas of Fujian Province, and it is named for the prefecture of the river and the city where Jinjiang horses are mainly distributed. Most Jinjiang horses have chestnut or bay coat colors, while a few show buckskin, black, or pinto colorations. The various coat colors stem from

Datasets
The dataset of worldwide horse populations consisted of the data collected in this study and previously published data, comprising a total of 653 animals from 24 horse breeds. The sampled horses of the 10 Chinese indigenous breeds in the present study were genotyped with the Illumina Equine SNP70 Bead Array. The previously published data of horses from 14 western breeds, genotyped with Illumina Equine SNP50 Bead Array, were from a study of Petersen et al [25]. There are 45,703 markers shared by the SNP70 and SNP50 arrays. Quality control of the raw data was carried out using PLINK v1.07 [27]. The following SNPs were discarded: (1) SNPs with Hardy-Weinberg Equilibrium p-value < 1 × 10 −5 ; (2) SNPs that were missing more than 10% of their genotype data; and (3) SNPs with a minor allele frequency less than 1%. Individuals with more than 10% missing genotyped data were also removed. The total SNPs left in the merged dataset of the

Datasets
The dataset of worldwide horse populations consisted of the data collected in this study and previously published data, comprising a total of 653 animals from 24 horse breeds. The sampled horses of the 10 Chinese indigenous breeds in the present study were genotyped with the Illumina Equine SNP70 Bead Array. The previously published data of horses from 14 western breeds, genotyped with Illumina Equine SNP50 Bead Array, were from a study of Petersen et al. [25]. There are 45,703 markers shared by the SNP70 and SNP50 arrays. Quality control of the raw data was carried out using PLINK v1.07 [27]. The following SNPs were discarded: (1) SNPs with Hardy-Weinberg Equilibrium p-value < 1 × 10 −5 ; (2) SNPs that were missing more than 10% of their genotype data; and (3) SNPs with a minor allele frequency less than 1%. Individuals with more than 10% missing genotyped data were also removed. The total SNPs left in the merged dataset of the worldwide horse populations were 39,187, while there were 57,597 filtered SNPs in the thinned data of the 10 Chinese indigenous breeds (Tables S1 and S2).

Population Divergence
To better infer the genetic structure of the Jinjiang horses and the 23 other worldwide horse breeds, we constructed a phylogenetic tree using the genome-wide high-density SNP data according to the following steps. First, the identical by state (IBS) distance matrix between individuals was generated by PLINK v1.07 [27] using the resulting 39,187 SNP sites. Second, based on the distance matrix, the neighbor joining (NJ) tree was constructed by MEGA v6 [28] and displayed by FigTree v1.4.0 [29]. For this analysis, we used donkey samples as an outgroup. We also performed principal components analysis (PCA) with the filtered 39,187 SNPs using the GCTA software (v1.24.2) [30]. The genetic relationship matrix and the covariance matrix were inferred from the PLINK format files (.ped and .map) with the parameters "-make -grm -pca 3." Then, the eigenvectors were defined based on the inferred covariance matrix, and the PCA biplot was plotted with ggplot2 (R Packages). The construction of the population structure was analyzed with the program Admixture v1.3 [31]. This program estimated the admixture proportions among the 24 horse breeds with the 39,187 SNPs. Nineteen scenarios (ranging from K = 2 to 20, Appendix A) were selected for genetic clustering with the parameters "major convergence criterion = 0.01". A cross-validation approach was used to determine the most likely number of populations (K) in the data. Additionally, we estimated the relationships between the Jinjiang horses and other 23 horse breeds in the dataset using TreeMix v1.13 [32] to determine the historical relationships between these breeds in terms of splits and migrations (mixtures). We used donkeys as the root population of the TreeMix analysis and set the SNP block size parameter to 10. TreeMix was run iteratively for values of the migration parameter (-m) between 1 and 8 (Appendix B). The f index, representing the fraction of the variance in the sample covariance matrix (Ŵ) calculated by the model covariance matrix (W), was used to identify the number of modeled migration events that best fit the data [32].
To study the relationship between the Jinjiang horse and the other CHBs, the 57,597 filtered SNPs were applied to conduct the subsequent analyses. A phylogenetic tree merely containing the Chinese horses was constructed with the same steps mentioned above. An analysis of population structure was carried out using the software Admixture v1.3 [31], which uses a model-based estimation of individual ancestry for a range of prior values of K defined by the user. The values of K were in the range from 2 to 12 (Appendix C) for accommodating a potential population structure within the breeds. The most suitable value of K was determined by the cross-validation approach. A principal components analysis was carried out using the GCTA software [30] with the same settings described above. A maximum likelihood tree of the 10 Chinese breeds was constructed using TreeMix v1.13 [32] based on the genome-wide allele frequency data. The TreeMix was run iteratively for values of the migration parameter from 1 to 5 (Appendix D), using the f index to identify the most suitable migration event model.

LD Decay
Linkage disequilibrium (LD) levels for Chinese horse populations were assessed by the genotype correlation coefficient (R 2 ) between any two loci (within and between different chromosomes) using PLINK v1.07 [27]. The parameters were set as "-blocks no -pheno -req -blocks -max -kb 10,000." Then, visualizations of LD decays among horse populations across the whole genome were generated using R scripts.

Identical by Descent Analyses
The data of 301 horse individuals genotyped with the 57,597 SNPs at the whole-genome level served as input for the identical by descent (IBD) detection. The frequencies of shared haplotypes between Jinjiang horses and each of the other CHBs were estimated with per 10,000 bp bins using IBDLD software (v3.37) [33]. The parameters were set as "-plinkbf int evolution -method GIBDLD -ploci 10 -nthreads 24 -step 0 -hiddenstates 3 -segment -length 10 -min 0.8". The calculation of normalized IBD (nIBD) between the Jinjiang horses and each of the other CHBs was as follows: nIBD = cIBD/tIBD, where cIBD = count of all haplotypes IBD between the Jinjiang horse and each of the other CHBs, and tIBD = total pairwise comparisons between the Jinjiang horse and each of the other CHBs.

Runs of Homozygosity and Genetic Diversity Analysis
Runs of homozygosity segments (ROH) were determined with an overlapping window approach implemented in PLINK v1.07 [27]. The final segments were called ROH if the minimum length of the homozygous segment was greater than 500 kb and comprised more than 50 homozygous SNPs, whereas one heterozygote and two missing genotypes were permitted within each segment. The analysis was based on the following settings: minimum SNP density was set to one SNP per 500 kb with a maximum gap length of 1000 kb. Three ROH parameters, genome length covered by ROH(S ROH ), number of ROH(N ROH ), and autozygosity (F ROH ) were computed.
Indices of genetic diversity, including observed heterozygosity (H O ) and expected heterozygosity (H E ), as well as the inbreeding coefficient (f), were also determined with PLINK v1.7 using the command -het.

Formal Test of Ancestor Admixture
The genome-wide most likely ancestry of Jinjiang horses was estimated using the f4 Ratio Estimation method of the ADMIXTOOLS Software Package [34] with default parameters. Applying f4 ratio estimation, we investigated the ratio of f4 (A, O; X, C)/f4 (A, O; B, C). The population X is an admixture of populations B and C. We calculated this ratio using the Naqu horse as A, the Baise horse as B, the other CHBs as C, the Jinjiang horse as X, and the donkey as O.

Estimating the Ancestry Proportion of the Jinjiang Horse
To infer the potentially genetic proportion of ancestral sources of Jinjiang horses from other horses, we estimated the admixture proportion using the LD based admixture inference algorithm implemented in ALDER v1.03 [35]. Based on the allele frequency difference in ancestral populations, the algorithm computes SNP correlations in an admixed target population and weights the correlations. We used the Baise horse and the Chakouyi horse as the reference populations and the Jinjiang horse as the target population. We created the weighted LD curves for the tests with the parameters "mincount = 4, binsize = 0.0005, maxdis = 0.5, fast_snp_read = NO."

Genetic Relationships between the Jinjiang Horses and Worldwide Horse Breeds
The genotype data of the Chinese indigenous horses from the 10 breeds, which were generated in the present study, as well as that of the 14 breeds of foreign horses from the previous study were used in the phylogenic analysis. After quality control, a total of 580 individuals with 39,187 SNP positions were involved in the subsequent analyses (Table S1). A neighbor-joining tree was constructed using the worldwide horses with the filtered data ( Figure 2A). Major clades of the tree showed breed groups, including the group of CHBs, the breeds which were recently admixed with or partly derived from Thoroughbreds (Quarter horse, Hanoverian horse), the Middle Eastern breeds (Akhal-Teke and Arabian horses), the group consisting of the Caspian horse and the Tuva horse, and the group containing the Shetland horses, Andalusian horse, and Morgan horse. The Belgian horse, Percheron horse, and Mongolian horse clustered together, which indicated that they have a close relationship, especially between the Daan horse and the Inner Mongolian horse. The Jinjiang horses were assigned into two clusters, one of which was close to the Chakouyi horses and another clustered with the Southwest breed group consisting of the Tengchong horses, the Lijiang horses, the Baise horses, and the Zhaotong horses.
Based on genetic co-ancestry analyses [36], we partitioned all individuals into known groups by varying the number of presumed ancestral populations (Appendix A, K ranging from 2 to 20). The cross validation (CV) statistics were used to choose the most suitable number of clusters, whose lowest value was K = 15 (Appendix A). When K was 15, the results showed that CHBs were derived from similar genetic backgrounds. The results revealed an evident admixture between the Daan horse and the draft horse breeds (i.e. the Belgian horse and the Percheron horse ( Figure 2B)). The ternary principal components analysis (PCA) revealed similar results with those of the ancestry estimation, while also showing closer genetic relationships among the Chinese horse breeds and evidence that some breeds in North China are closely related to the heavy horse breeds ( Figure 2C).
In addition to PCA and clustering analyses, the TreeMix algorithm was used to place the Jinjiang horse on a tree based on a maximum likelihood estimation approach. First, the relationships between the Jinjiang horse and other worldwide horse populations were estimated. The f index representing the fraction of the variance in the sample covariance matrix (Ŵ) was calculated with the model covariance matrix (W), as a function of modeling the number of modeled migration events. More migration edges did not seem to further increase the variance explained by the phylogenetic model, as the f index reached an asymptote above 6 migration edges ( Figure 2D, inset), so m = 6 was chosen as optimum migration mode in this dataset. The results showed some well-known gene flows, such as that from Thoroughbreds to Morgan horses, and from Arabian horses to Caspian horses, and several other migration events, which mainly happened between foreign breeds. Notably, the migration from foreign draft horse breeds to the Daan horses was the only gene flow detected between Chinese and foreign horses. The results showed that there were no migration events between Jinjiang horses and overseas populations, including Arabian horses ( Figure 2D).

Population Genetic Relationships between the Jinjiang Horse and Other Chinese Indigenous Horse Breeds
Because the Jinjiang horse is solely genetically related to other CHBs according to the results of the analyses of the worldwide horse data, the subsequent studies were focused on the 10 CHBs to investigate the genetic relationships between the Jinjiang horse and other Chinese horse breeds in the following analyses using the 57,597 SNPs that passed quality filtration. A neighbor joining (NJ) tree only containing Chinese indigenous horses showed that Jinjiang horses form four branches, with most Jinjiang horses following their geographical distributions and clustering with horse breeds from Southwest China, while a few individuals of the breed are relevant to horse breeds in North China ( Figure 3A).
Artificial selection tends to reduce genetic diversity, and commercial breeds have undergone stronger artificial selection than local breeds have. The impact of artificial selection could also be detected in the genome LD levels in each population, as artificial selection can facilitate the increase of LD within a population [37]. The result of LD analyses showed that Jinjiang horses have the lowest level of LD, which indicates that they have not been subjected to intensive selection ( Figure 3B).
The results from both analysis of ADMIXTURE (K = 4, Figure 3C) and PCA ( Figure 3D) showed that Jinjiang horses have closer relationships with horse breeds from South China, especially Baise horses. The results also indicated a differentiation between horse populations from South China and North China. A TreeMix analysis of the Chinese indigenous breeds was performed by setting the migration edges at 3, as there was no further increase of the variation explained when the value of migration edges was greater than 3 ( Figure 3E, inset). The results showed that there was a significant gene flow from the Chakouyi horse to the Jinjiang horse. In addition, migration events from the Naqu horse to the Chakouyi horse, and from the Northern horses to the Zhaotong horses, were also observed in the results ( Figure 3E).

Runs of Homozygosity and the Genetic Diversity of the Worldwide Horse Breeds
F ROH can quantify more recent inbreeding. In the horse populations studied, the highest F ROH was observed in the Andalusian horses (mean F ROH = 0.1022), followed by the Thoroughbreds (mean F ROH = 0.0990). The lowest value was obtained for the Chakouyi horses (mean F ROH = 0.0060), and the Naqu population also had very low F ROH values.
The observed heterozygosity of the horse breeds ranged from 0.2955 to 0.2748 (Table 1). The Morgan horses and Jinjiang horses showed relatively high heterozygosity, and the lowest value was observed in the Akhal Teke horse population. The expected heterozygosity of the Belgian horses and the Arabian horses was slightly higher in the horse populations studied (Table S1).
The Jinjiang horse population showed medium values of the ROH parameters compared with other Chinese indigenous horse breeds. Although only a moderate degree of inbreeding occurred in the Jinjiang horse population in the past, based on the estimated value of F ROH , most of the individuals in the population possess the ROHs, and the inbreeding coefficient (f) of the breed indicated that Jinjiang horses currently may have the highest inbreeding level among the Chinese horse breeds studied.
Compared with the Chinese indigenous horse breeds, the foreign breeds, which have been subjected to highly selective breeding, such as Thoroughbreds, Andalusian horses, Arabian horses, Morgans, Shetland ponies, and Akhal_Teke horses, have higher N ROH and F ROH . The results are consistent to the breeding history of these horse breeds.

Identical by Descent Analyses of the Jinjiang Horse
The results of IBD analysis showed that Jinjiang horses are closely related to horse breeds in Southwest China (Figure 4). The Baise horse showed a significantly higher shared IBD with the Jinjiang horse compared to other Chinese breeds. The Naqu horse has the smallest IBD segment length shared with the Jinjiang horse.
The results of IBD analysis showed that Jinjiang horses are closely related to horse breeds in Southwest China (Figure 4). The Baise horse showed a significantly higher shared IBD with the Jinjiang horse compared to other Chinese breeds. The Naqu horse has the smallest IBD segment length shared with the Jinjiang horse.

Identifying Signatures of Selection in the Jinjiang Horse Population with Runs of Homozygosity Analysis
Long consecutive homozygous genotype segments, i.e. ROHs, result when parents transmit identical haplotypes, which can be used for estimating autozygosity. Thus, we used PLINK software to compute the number of ROHs. Through investigating the distribution of ROHs across the genome, we detected the highest number of ROHs on ECA 11 of Jinjiang horses ( Figure 5). An overview of the known genes in these regions (analyzed at the DAVID website: https://david.ncifcrf.gov/) is provided in Table 2. Among the genes, SPAG9 [38][39][40], NME1 [41,42], and NME2 [43,44] are involved in regulating the proliferation of cancer cells. MBDT1 is a gene regulating the development of

Identifying Signatures of Selection in the Jinjiang Horse Population with Runs of Homozygosity Analysis
Long consecutive homozygous genotype segments, i.e. ROHs, result when parents transmit identical haplotypes, which can be used for estimating autozygosity. Thus, we used PLINK software to compute the number of ROHs. Through investigating the distribution of ROHs across the genome, we detected the highest number of ROHs on ECA 11 of Jinjiang horses ( Figure 5). An overview of the known genes in these regions (analyzed at the DAVID website: https://david.ncifcrf.gov/) is provided in Table 2. Among the genes, SPAG9 [38][39][40], NME1 [41,42], and NME2 [43,44] are involved in regulating the proliferation of cancer cells. MBDT1 is a gene regulating the development of embryonic skeletal system [45], UTP18 is a small subunit processome component [46], and CA10 is a zinc enzyme [47].

Estimation of Possible Ancestry with a Formal Test of ADMIXTURE
Although above studies suggest that the Jinjiang horse has a diverse native-Chinese origin, the possible ancestry of Jinjiang Horses still remains unknown. Therefore, the f4 Ratio Estimation algorithm (f4 ratio = f4 (A, O; X, C)/f4 (A, O; B, C)) from ADMIXTOOLS Software Package [34] were used to conduct the further analysis. Baise horses (C) and Naqu horses (A) served as reference populations, and the Jinjiang horse is the target population (X) in f4 Ratio Estimation. In the output of the analysis, the results are positive only when the value of f4 ratio is above zero, and the greater the f4 ratio, the more likely it is that the B breed and Baise horses are the ancestors of Jinjiang horses. The results showed that Baise horses and Chakouyi horses are the most likely ancestors of Jinjiang horses (Table 3).

Estimating the Proportion of Admixture
ALDER v1.03 software [35] was applied to estimate the proportions of Baise horse and Chakouyi horse in the ancestral sources of the Jinjiang horse. The Alder analysis, which is based on the exponential decay rate of admixture LD, was used to estimate the proportion of the gene flows between Jinjiang horses and the ancestral breeds. Our results showed that the proportion of Baise horses in the ancestry of Jinjiang horses is approximately 47%, and the Chakouyi horses account for approximately 20% of the ancestral sources.

The Relationship between Jinjiang Horses and Worldwide Horse Populations
The earliest records about horse raising in Jinjiang of Fujian Province was in the Tang dynasty (AD 618-AD 907), which was much later than that of North China, where the remains of the earliest domestic horses in China were found. It was also later than the most horse-raising areas of Southwest China. It is unclear from where the early horses in Jinjiang were introduced. Because Quanzhou city was the most important port in the Marine Silk Road and there were frequent and prosperous trading activities between Quanzhou and the Arab world [48], it was hypothesized that some foreign horses, especially the horses from ancient Arabian countries, might contribute to the ancestral sources of Jinjiang horses [49]. Thus, it is essential to investigate the genetic links between Jinjiang horses and the main foreign horse breeds.
To study the relationship of the Jinjiang horses and the worldwide horse populations, we created a merged dataset of 580 Chinese and foreign horses from 10 CHBs and 14 foreign horse breeds with 39,187 autosome SNPs, which were used to construct phylogenetic NJ trees (five donkeys used as outgroup). It was indicated that the Jinjiang horses have close relationship with other CHBs but did not show direct genetic link to any foreign horse breeds. Admixture analysis results also showed that the Jinjiang horse has the similar ancestral background to other CHBs, which is obviously different from foreign horse breeds. PCA and TreeMix analyses further confirmed the results of the clustering analysis and showed the close genetic relationship between Jinjiang horses and the other Chinese indigenous breeds, and there was no gene flow between Jinjiang horse and foreign horse breeds. These results showed that there is no significant genetic association between Jinjiang horses and foreign horses, and the Jinjiang horse was not influenced by the foreign breeds in history. Some of our results also showed that there is a close genetic relationship between European heavy horses and northern Chinese horses, especially with Daan horses, which was also confirmed in the TreeMix analysis. These phylogenetic results are in accordance with the breeding history of Daan horses, which had been bred by hybridizing the mares from local breeds with the sires of European heavy horses [50]. The genetic structure of the northern Chinese horses is similar to the Tuva horses and Caspian horses, which suggests the possible genetic links between horse populations of north China and the breeds in Eurasian steppes and supports the conclusion of our previous studies [11]. We also found evidence of genetic introgression from Arabian breed to Caspian horse, which is also consistent to the breeding records of Caspian horses [51]. The analysis also revealed known gene flows between horse breeds, such as Thoroughbreds and Morgan horses [25]. In summary, Jinjiang horses have a close relationship with other CHBs, and they are not closely related to foreign horses, including Arabian horses.

Native Origin of Jinjiang Horses
Because Jinjiang horses are only closely related to Chinese indigenous horses, we focused on the Chinese indigenous horses. Thus, the dataset of the 10 Chinese horse breeds (n = 234 and 57,597 SNPs) was involved in the subsequent analyses. The phylogenetic NJ tree (plus five donkeys as an outgroup) showed that most of the Jinjiang horses have a close relationship with the horse breeds in Southwest China, while a few individuals of the breed are relevant to the northern horses, which is attributed to the low selection imposed on the indigenous breed. Some stallions, which have no significant blood relationship, were used in breeding activities and were the founders of the subpopulations. These results are also in accord with the output of the LD analysis. The lowest LD level of Jinjiang horses suggested that Jinjiang horses did not experience intensive selection and that the genetic components of their ancestral sources may be largely preserved. The PCA and Admixture analysis confirmed the results of the phylogenetic NJ tree about the close genetic links between Jinjiang horses and the southwestern horse populations. The results of estimating genome-wide IBD sharing indicated that Jinjiang horses are most closely related to Baise horses among the southwest Chinese populations. Our study also revealed an evident admixture of Chakouyi horses and Jinjiang horses, suggesting a complicated history of gene flow from Chakouyi horses to Jinjiang horses. Further analysis revealed that Baise horses and Chakouyi horses are two of the most important ancestors of Jinjiang horses, and they account for approximately 47% and 20% of the ancestral sources of Jinjiang horses, respectively. Thus, the ancestry of Jinjiang horses is the mixture of Southwest and North Chinese horses. It is far away from the distributing areas of Baise horses (Baise of Guangxi Province) or Chakouyi horses (Tianzhu of Gansu Province) to those of Jinjiang horses (Jinjiang of Fijian Province)-i.e., over thousands of kilometers. The gene flows from Baise horse and Chakouyi horse were inevitably related with the historical trades and migrations of human beings, which may give reasonable explanation about the long-distance migrations of the horses.

The Migrating Routes of the Ancestral Horses
The ancient Tea-Horse Road and Marine Silk Road in China may have a significant impact on the formation of the Jinjiang horse. From the Tang Dynasty (AD 618-AD 907) to the Qing Dynasty (AD 1636-AD 1912), the ancient Tea-Horse Road traversed through Yunnan province and Sichuan province to Tibet, which primarily aimed to trade tea from the tea-producing areas for horses from Tibet and other regions of horse-raising, but eventually it developed into arteries for trade in the mountainous regions of Southwest China, not only limited to Tea-Horse trading activities. In the mountainous ancient Tea-Horse Road, packhorses were used to cross difficult terrain, and horse-packing was the main means of transport for goods, while the businessmen who managed the horse-packing trade were called horse caravans. Guangxi Province, where high-quality tea is also produced, is at the eastern end of the ancient Tea-Horse Road. The ancient Heng-shan-zhai city of Guangxi Province, adjacent to Baise, had been an important and prosperous trade center in the Song Dynasty (AD 960-AD 1297). The ancient trade and the Tea-Horse Road facilitated the migrations and exchanges of horses in these areas [52]. The results of our study also confirmed the impact of the trade on the horse populations in Southwest China, which showed there are close genetic links among the Baise horse, the Tengchong horse, the Lijiang horse, and even the Naqu horse.
The Baise horse is an important horse breed in Guangxi Province. Because Guangxi Province was a major transportation and trade area (especially in the Song Dynasty), which connected inland provinces (such as Yunnan Province and Guizhou Province) to the coastal provinces, including Guangdong and Fujian where important ancient ports were located, the Baise horse had a profound influence on other horse breeds in the whole southwest region [53]. The Marine Silk Road was formed in the Tang Dynasty and bloomed in the Song Dynasty, starting from the southeastern coastal cities, Quanzhou (overlapping with the region where Jinjiang horses were distributed) in Fujian Province and Guangzhou in Guangdong Province, and passing through the sea route to West Asia and the Arab world. As the most prosperous port in the world at that time, Quanzhou attracted immigrating human beings and was the core city for the trading commodities of the surrounding areas. Quanzhou had established close ties with Guangxi Province through Guangdong Province [54]. In this way, the ancient Tea-Horse Road and the Marine Silk Road were also linked together through Guangdong Province and formed the passage through which the Baise horse and the Chakouyi horse migrated to Fujian Province.
Chakouyi horses are mainly distributed in the Gansu Province. The province is at the northern end of the Tibet-Yi Corridor, which represented the long routes (over 1000 km) along the north-south oriented rivers and valleys in the boundary of the southwest and northwest provinces of China. By following the corridor, people could migrate from the northwestern provinces to Southwest China, including Yunnan Province, Sichuan Province, and Tibet, all of which are areas along the ancient Tea-Horse Road [55]. Ancient people and their horses may have spread into the southern provinces through the Tibet-Yi Corridor and the ancient Tea-Horse Road, eventually affecting the indigenous horses in Fujian Province.
The hypothesized routes by which Baise horses and Chakouyi horses migrated to Fujian Province are shown below ( Figure 6). Therefore, the formation of the Jinjiang Horse was closely related to ancient trading activities and the migration of human beings.

The Conservation of Jinjiang Horses
After the population of Jinjiang horses continuously decreased in the past several decades, Jinjiang horses have become an endangered horse breed with a small population size and the conservation of the breed is now a pressing matter to carry out.
From this study, the population structure and the genetic relationship of Jinjiang horses were elucidated, and the high level of inbreeding was also revealed with the analysis of ROH and the genetic diversity. The subpopulations of the breed could be used as main lines for further breeding programs and conservation. Moreover, the results regarding the genetic relationship among the individuals of the breeds also provide useful information for establishing sound mating systems for the population, thereby avoiding further inbreeding depressions. The ancestry sources revealed in the study can also be introduced to rejuvenate the Jinjiang horse. Although obvious association between the function of genes in the selected segments and the specific traits of Jinjiang horses was not found, the results provide valuable information for further studies to illustrate the signatures of selection, which offer meaningful information for the conservation of the breed. the population, thereby avoiding further inbreeding depressions. The ancestry sources revealed in the study can also be introduced to rejuvenate the Jinjiang horse. Although obvious association between the function of genes in the selected segments and the specific traits of Jinjiang horses was not found, the results provide valuable information for further studies to illustrate the signatures of selection, which offer meaningful information for the conservation of the breed. Figure 6. The diagram for the migration routes of the ancestral sources of Jinjiang horses. The broken blue lines indicate the Tibetan-Yi Corridor; the broken green lines, the Ancient Tea-Horse Road; the broken black lines, the Marine Silk Road (all of these ancient roads or corridors were simplified); the red heavy lines, the migrating routes of the Chakouyi horses; the purple heavy lines, the routes of the Baise horses; the black text, the names of the provinces; and the red text, the names of the horse breeds. (The map was figured with R Packages).

Conclusions
The results of our study suggest that Jinjiang horses originated from other Chinese indigenous breeds. The Baise horse and Chakouyi horse are two of the main ancestral sources of the Jinjiang Figure 6. The diagram for the migration routes of the ancestral sources of Jinjiang horses. The broken blue lines indicate the Tibetan-Yi Corridor; the broken green lines, the Ancient Tea-Horse Road; the broken black lines, the Marine Silk Road (all of these ancient roads or corridors were simplified); the red heavy lines, the migrating routes of the Chakouyi horses; the purple heavy lines, the routes of the Baise horses; the black text, the names of the provinces; and the red text, the names of the horse breeds. (The map was figured with R Packages).

Conclusions
The results of our study suggest that Jinjiang horses originated from other Chinese indigenous breeds. The Baise horse and Chakouyi horse are two of the main ancestral sources of the Jinjiang horse. Ancient trading activities and the migration of human beings had deep impacts on the formation of the coastal horse breed.
The Jinjiang horse has been bred from their ancestors and is adapted to coastal southeast China, and it is essential to take action to conserve the indigenous breed and develop a new breeding strategy to improve the equestrian traits of the breed, through which it can eventually be developed into a high-quality saddle horse breed.