Background

The horse (Equus ferus caballus) was one of the earliest domesticated species and has played numerous important roles in human societies: acting as a source of food, a means of transport, for draught and agricultural work, and for sport, hunting, and warfare [1]. Horse domestication is believed to have started in the western Asian steppes approximately 5,500 years ago, and quickly spread across the Eurasian continent, with herds being augmented by the recruitment of local wild horses [2]. Domestication in the Iberian Peninsula might have represented an additional independent episode, involving horses that survived in a steppe refuge following the reforestation of Central Europe during the Holocene [3].

The horse reference genome has provided fundamental genomic information on the equine lineage and has been used for improving the health and performance of horses [1, 4]. Horses exhibit 214 genetic traits and/or diseases that are similar to those of humans [5]. To date, several horse whole genomes have been sequenced and analyzed [4, 6]. In 2012, the first whole genome re-sequencing analysis was conducted on the Quarter Horse breed to identify novel genetic variants [4]. In 2013, divergence times among horse fossils, donkey, Przewalski's horse, and several domestic horses were estimated, together with their demographic history [6]. However, currently available whole genome sequences of modern horses only comprise western Eurasian breeds.

Over the centuries, more than 400 distinct horse breeds have been established by genetic selection for a wide number of desired phenotypic traits [7]. The Marwari (also known as Malani) horse is a rare breed from the Marwar region of India, and is one of six distinct horse breeds of India. They are believed to be descended from native Indian ponies, which were crossed with Arabian horses beginning around the 12th century, possibly with some Mongolian influence [8–10]. The Marwari horses were trained to perform complex prancing and leaping movements for ceremonial purposes [11, 12]. The Marwari population in India deteriorated in the early 1900s due to improper management of the breeding stock, and only a few thousand purebred Marwari horses remain [12].

Here, we report the first whole genome sequence of a male Marwari horse as one of the Asian breeds and characterize its genetic variations, including single nucleotide variations (SNVs), small insertions/deletions (indels), and copy number variations (CNVs). To investigate relationships among different horse breeds, we carried out a genome-wide comparative analysis using previously reported whole genome sequences of six western Eurasian breeds [4, 6], and single nucleotide polymorphism (SNP) array data of 729 horses from 32 worldwide breeds [13]. Our results provide insights into its genetic background and origin, and identify genotypes associated with the Marwari-specific phenotypes.

Results and discussion

Whole genome sequencing and variation detection

Genomic DNA was obtained from a blood sample of a male Marwari horse (17 years old) and was sequenced using an Illumina HiSeq2000 sequencer. A total of 112 Gb of paired-end sequence data were produced with a read length of 100 bp and insert sizes of 456 and 462 bp from two genomic libraries (Additional file 2: Figure S1, Figure S2). A total of 1,013,642,417 reads remained after filtering, and 993,802,097 reads were mapped to the horse reference genome (EquCab2.0 from the Ensembl database) with a mapping rate of 98.04%. (Additional file 2: Figure S3, Figure S4). A total of 133,091,136 reads were identified as duplicates and were removed from further analyses (Additional file 1: Table S1). To enhance the mapping quality, we applied the IndelRealigner algorithm to the de-duplicated reads. A total of 44,835,563 (5.2%) reads were realigned, and the average mapping quality increased from 53.11 to 53.16 (from 29.33 to 43.32 in the realigned reads). The whole genome sequences covered 95.6% of the reference genome at 10 × or greater depth.

To identify novel genomic sequences, we performed a de novo assembly using the unmapped reads (1.8 Gb) to the horse reference genome. A total of 120,159 contigs (24,781,670 bases in length and 227 bp of contig N50 size) were assembled. After mapping the contigs to the reference genome, we found that 25,614 contigs (4,855,119 bases in length and 196 bp of contig N50 size) did not match the reference sequences; indicating that they may be novel regions specific to the Marwari horse breed (Additional file 1: Table S2). To identify the biological functions of these novel regions, the un-matched contigs were further analyzed by BLAST searches using the NCBI protein database. However, none of the contigs significantly matched the known protein database (Additional file 2: Figure S5).

Comparing the Marwari sequence to the reference genome, approximately 5.9 million SNVs and 0.6 million indels were identified (Table 1). Estimates of SNP rate and heterozygosity of the Marwari were similar to those of other horse breeds (Arabian, Icelandic, Norwegian Fjord, Quarter, Standardbred, and Thoroughbred) (Additional file 1: Table S3). We assessed the mutational frequency at the single nucleotide level in the Marwari and compared it to estimates from other breeds (Additional file 1 Table S4). Interestingly, we found that the prevalent mutation types were not consistent among horse breeds. The mutation spectrum of the Marwari was dominated by C>T (G>A) transitions; a pattern which was also observed in the Icelandic, Norwegian Fjord, and Quarter horses. Conversely, the genomes of the Arabian, Standardbred, and Thoroughbred horses were dominated by A>G (T>C) transitions. A significant association between the mutation spectrum and horse breed (p-value < 0.001) was found when we applied a chi-square test using SPSS [14] to statistically compare the differences in the mutation spectrums among the breeds.

Table 1 Variants in the Marwari horse genome.
Full size table

The Marwari genome consisted of 2,383,702 (40.2%) homozygous and 3,539,864 (59.8%) heterozygous SNVs (Table 1). Among them, 18,195 were found to be nonsynonymous SNVs (nsSNVs). When the Marwari variants were compared to those previously reported from the genomes of other breeds [4, 6] and the horse SNP database from the Broad Institute, 1,577,725 SNVs and 249,609 indels were novel variants. Of these, 4,716 variants (4,413 nsSNVs and 303 indels in coding regions) represented amino acid changes which were found in 2,770 genes (2,584 genes with nsSNVs, 279 genes with indels in coding regions, and 93 genes with nsSNVs and indels in coding regions simultaneously). To annotate the variants using well-known functional databases, human orthologs were retrieved from the Ensembl BioMart utility. A total of 1,970 of the 2,770 genes had human orthologs, and 1,896 genes were annotated using the DAVID Bioinformatics Resource 6.7 [15]. The genes with nsSNVs and/or indels in coding regions were highly enriched in olfactory functions (Additional file 1: Tables S5 and S6).

Copy number variations (CNVs) were identified using the R library "ReadDepth package" [16]. A total of 2,579 CNVs, including 869 gain and 1,710 loss blocks, were identified in the Marwari genome. The sizes ranged from 3 Kb to 6.43 Mb with an average length of 56 Kb. The CNV region (140 Mb in length) contained 2,504 genes which were duplicated (1,138 genes) or deleted (1,366 genes) (Additional file 1: Table S7). From the functional enrichment analysis, we found that the duplicated genes were enriched in olfactory function, whereas the deleted genes were enriched in immune regulation and metabolic processes (Additional file 1:Table S8, Table S9, Table S10, and Table S11).

Relatedness to other horse breeds

We constructed a phylogenetic tree using SNVs found in the whole genome data of the seven horse breeds (Arabian, Icelandic, Marwari, Norwegian Fjord, Quarter, Standardbred, and Thoroughbred) [4, 6]. We identified 11,377,736 nucleotide positions that were commonly found in the seven horse genomes. A total of 25,854 nucleotide positions were used for phylogenetic analysis after filtering for minor allele frequency (MAF), genotyping rate, and linkage disequilibrium (LD). We found that the Marwari horse is most closely related to the Arabian breed (Additional file 2: Figure S6), while the Icelandic horse and Norwegian Fjord were the most distinct from the other breeds, all of which are known to descend from Arabian horses [17, 18].

To further explore the relationships among breeds, we compared the Marwari horse genome data with SNP array data from 729 individual horses belonging to 32 domestic breeds [13]. A total of 54,330 nucleotide positions were shared across all individuals including the Marwari horse. After pruning as described above, 10,554 nucleotide positions were used for the comparative analyses. We calculated pairwise genetic distances and conducted multidimensional scaling (MDS) to visualize the relationships among the horse breeds (Figure 1). The Marwari horse fell together with Iberian-lineage breeds, such as the Andalusian, Mangalarga Paulista, Peruvian Paso, and Morgan horse breeds, all of which are known to have an Arabian ancestry [19–22]. Additionally, we found that the Marwari horse fell between Arabian and Mongolian horses, indicating their dual genetic influences on the Marwari horse as previously suggested [8–10].

Figure 1
figure 1

Multidimensional scaling plot derived from a Marwari horse and other horse breeds. Black arrow indicates the Marwari horse.

Full size image

We applied the STRUCTURE program [23, 24] to estimate the genetic composition of the Asian horse breeds including the Marwari horse. For K = 2 groups, the Arabian horses were strongly separated from Mongolian horses, and the genetic composition of the Marwari horse was composed of alleles clustering with both the Mongolian horse (65.8%) and the Arabian horse (34.2%) (Figure 2). Other Asian breeds (Akhal Teke, Caspian, and Tuva) also showed genetic admixture between Arabian and Mongolian horses. From K = 3 to K = 5, the Marwari had high genetic components of both Arabian and Mongolian horses, whereas Akhal Teke and Caspian horses were mostly assigned to other clusters. These results indicate that the Marwari is genetically closely related to the Arabian and Mongolian horses. It is unclear whether the latter relationship represents direct genetic input from Mongolian horses or whether these horses are the closest population to the Indian ponies from which the Marwari is thought to have descended [8–10]. Further analysis including Indian ponies and Marwari horses will be required to distinguish the relative importance of these two scenarios, which are not mutually exclusive.

Figure 2
figure 2

STRUCTURE analysis using Marwari and Asian horse breeds. For all K values, the Marwari has genetic affinities to both Arabian (blue) and Mongolian horses (orange).

Full size image

Phenotype association of the identified variants

To provide insight into the unique Marwari phenotypes, we investigated amino acid changes specific to this breed compared to those of other breeds (Arabian, Icelandic, Norwegian Fjord, Quarter, Standardbred, and Thoroughbred). A total of 343 amino acid changes in 236 genes were unique to the Marwari horse. Among the 236 genes, 75 genes included one or more amino acid changes predicted by the PolyPhen2 program to alter protein function [25] (Additional file 1: Table S12). Interestingly, the teashirt zinc finger family member 1 (TSHZ1) gene had a homozygous p.Ala344>Val variant (Figure 3). TSHZ1 is involved in transcriptional regulation of developmental processes and is associated with congenital aural atresia in humans, a malformation of the ear occurring in approximately 1 in 10,000 births [26, 27]. Additionally, TSHZ1-deficient mice show malformations in the middle ear components [28]. Therefore, the A334V amino acid change in TSHZ1 is a strong candidate as the genetic factor responsible for the inward-turning ear tips characteristic of the Marwari breed. A future genomic comparison with the Kathiawari horse, which also has inward-turning ear tips, might support to this prediction.

Figure 3
figure 3

Partial alignment of TSHZ1 amino acid sequences among horse breeds and vertebrate species. Red rectangle indicates a Marwari horse-specific amino acid change (A344V). Gray and blue rectangles indicate a Ser-rich region and Zinc fingers, respectively.

Full size image

After annotating the Marwari variants for their known disease and trait information [26–55] (Table 2), we found that this breed has a homozygous variant for the g.27991841A>G mutation in the SCL26A2 gene, which causes autosomal recessive chondrodysplasia in equine. Other variants were associated with racing endurance in Thoroughbred horses (g.32772871C>T in COX4I1, g.40279726C>T in ACN9), horse size (g.81481065C>T in HMGA2, g.23259732G>A in LASP1), and pattern of locomotion (g.22999655C>A in DMRT3).

Table 2 Genetic variants for known traits and diseases.
Full size table

Selection in the equid lineage

We assessed the signatures of selection in the equid lineage using the d N /d S (nonsynonymous substitutions per nonsynonymous site to synonymous substitutions per synonymous site) ratio [56]. A consensus horse (equid) sequence was constructed by integrating all of the available breed genomes (Arabian, Icelandic, Marwari, Norwegian Fjord, Quarter, Standardbred, and Thoroughbred) in an attempt to remove breed specificity and to include an Asian breed component via the central Asian heritage of the Marwari (in contrast to the western Eurasian breeds for which whole genomes had been previously sequenced). A total of 7,711 out of 22,305 genes in the horse reference genome were substituted by the consensus sequences. Using the protein sequences of seven non-horse genomes (camel, pig, cow, minke whale, dog, mouse, and human), 5,459 orthologous gene families were constructed using OrthoMCL [57]. Using alignments of these gene families to estimate d N /d S , we identified 188 genes under selection in the horse genome (Additional file 1: Table S13). The selected genes were particularly enriched in immune response (immune effector process, leukocyte mediated immunity, positive regulation of immune system process, and defense response) and possible motor ability (T-tubule, muscle contraction, and regulation of heart contraction) functions (Additional file 1: Table S14). Over evolutionary time, the horse has developed increased speed and its musculature has become specialized for efficient strides [58, 59]. It is therefore possible that the motor activity-associated genes we identified to be under positive selection have contributed to the muscular efficiency seen in modern horses.

Conclusion

Our study provides the first whole genome sequences and analyses of the Marwari, an Asian horse breed. Comparing the Marwari genome to the horse reference genome, approximately 5.9 million SNVs and 0.6 million indels, including 4,716 variants that cause amino acid changes, were identified. We found a clear Arabian and Mongolian component in the Marwari genome, although further work is needed to confirm whether modern Marwari horses also descended from Indian ponies. We analyzed the Marwari variants and found a candidate SNV determining its characteristic inward-turning ear tips. Additionally, we investigated selection in the horse genome through comparisons with other mammalian genomes. By creating a consensus sequence that included information on an Asian breed, we found a number of genetic signatures of selection, providing insights into possible evolutionary and environmental adaptations in the equid lineage. The whole genome sequencing data from the Marwari horse provides a rich and diverse genomic resource that can be used to improve our understanding of animal domestication and will likely be useful in future studies of phenotypes and disease.

Methods

Sample preparation and whole genome sequencing

Genomic DNA was extracted from the blood of a 17 year old male Marwari horse with the XcelGene Blood gDNA Mini Kit (Xcelris Labs Ltd, Gujarat, India) following the manufacturer's protocol. Two genomic libraries with insert sizes of 456 and 462 bp were constructed at Theragen BiO Institute (TBI), TheragenEtex, Korea. The genomic DNA was sheared using Covaris S series (Covaris, MS, USA). The sheared DNA was end-repaired, A-tailed, and ligated to paired-end adapters, according to the manufacturer's protocol (Truseq DNA Sample Prep Kit v2, Illumina, San Diego, CA, USA). Adapter-ligated fragments were then size selected on a 2% Agarose gel, and the 520-620 bp band was extracted. Gel extraction and column purification were performed using the MinElute Gel Extraction Kit (Qiagen, CA, USA) following the manufacturer's protocol. The ligated DNA fragments containing adapter sequences were enhanced via PCR using adapter-specific primers. Library quality and concentration were determined using an Agilent 2100 BioAnalyzer (Agilent, CA, USA). The libraries were quantified using a KAPA library quantification kit (Kapa Biosystems, MA, USA) according to Illumina's library quantification protocol. Based on the qPCR quantification, the libraries were normalized to 2 nM and denatured using 0.1 N NaOH. Cluster amplification of denatured templates was performed in flow cells according to the manufacturer's protocol (Illumina, CA, USA). Flow cells were paired-end sequenced (2 × 100 bp) on an Illumina HiSeq2000 using HiSeq Sequencing kits. A base-calling pipeline (Sequencing Control Software, Illumina) was used to process the raw fluorescent images and the called sequences.

Filtering and mapping processes

Before the mapping step, raw reads were filtered using NGS QC toolkit version 2.3 (cutoff read length for high quality, 70%; cutoff quality score, 20) [60]. After the filtering step, clean reads were mapped to the horse reference genome (Ensembl EquCab2.0, release 72) [1] with BWA version 0.7.5a [61] with minimum seed length (-k 15) and Mark shorter split hits as secondary (-M). We realigned the reads using the GATK [62] IndelRealigner algorithm to enhance the mapping quality, and marked duplicate reads using MarkDuplicates from picard-tools version 1.92 (http://broadinstitute.github.io/picard/).

De novoassembly of unmapped reads

We extracted unmapped sequences from aligned Marwari BAM files. To find Marwari specific genomic regions, we assembled unmapped reads using SOAPdenovo2 [63] with "all" mode and multiple K values (ranged from 23 to 63). A total of 120,159 contigs were obtained, and N50 length was 227bp. To identify whether these contigs are in non-reference regions, we aligned contigs to the horse reference genome. A total of 25,614 contigs were not aligned to the reference genome. The non-reference sequences were further analyzed by BLAST to NCBI protein and DNA sequence databases with the criteria E≤10-5 and identity ≥ 70%.

Variant detection and annotation

Putative variant calls were made using the SAMtools version 0.1.16 [64] mpileup command. In this step, we used the -E option to minimize the noise resulting from pairwise read alignments, and the -A option to use regardless of insert size constraint and/or orientation within pairs. Variants were called using bcftools and then filtered using vcfutils varFilter (minimal depth of 8, maximal depth of 100, Phred scores of SNP call ≥ 30, and no indel present within a 2 bp window) as previously reported [6]. SnpEff [65] was used to annotate the variants. To find unique variants for the Marwari horse, SNVs and small indels were further compared with the horse SNP database that was identified by the Horse Genome Project (http://www.broadinstitute.org/mammals/horse/snp), and other previously reported horse breed genomes [4, 6]. Copy number variants based on the differences in sequencing depths were detected using R library "ReadDepth package" with default options. The ReadDepth calculated the thresholds for copy number gain (2.642) and loss (1.380).

Phylogenetic tree construction

Genotype data were extracted from a total of 11,377,736 single nucleotide positions, which were shared and sufficiently covered regions (> 8 × depth), in the seven horse whole genome data (Arabian, Icelandic, Marwari, Norwegian Fjord, Quarter, Standardbred, and Thoroughbred) [4, 6]. The genotyping data were merged, and then filtered to remove those SNP with a genotyping rate of < 0.05 and allele frequency > 0.2 using PLINK [66]. SNPs that were in linkage disequilibrium (LD) were also removed: the merged files were pruned for r2 < 0.1 in PLINK, considering 100 SNP windows and moving 25 SNPs per set (-indep-pairwise 100 25). After the filtering and pruning process, 25,854 SNPs remained and were used for the phylogenetic analysis. RAxML version 7.28-ALPHA [67] was used to generate the parsimony starting trees, and RAxML-Light version 1.0.9 [68] was used to carry out tree inference with the GTRGAMMA model of nucleotide substitutions. A total of 100 bootstrap trees were generated for each phylogeny. The resulting tree was drawn by MEGA6 [69].

MDS and population structure analyses

Equine SNP array data of 729 individuals belonging to 32 horse breeds were obtained from a previous report [13]. The Marwari horse data used in this analysis were selected from 54,330 nucleotide positions that were derived from the SNP array data. The SNP array and Marwari data were filtered and pruned to remove SNPs with the same cutoffs described above, except that the MAF option was set to --maf < 0.05. A total of 10,554 single nucleotide positions were used for the following comparative analyses.

The MDS plot was drawn in R [70] using the "MASS" library and "canberra" distance metric. STRUCTURE version 2.3.4 [23, 24] was used to cluster Asian breeds based on genetic similarity, investigating K values from 2 to 5. Each run for a given K value consisted of a 15,000 steps burn-in and 35,000 MCMC repetitions. We applied a default admixture model and a default assumption that allele frequencies were correlated. The convergence of STRUCTURE runs was evaluated by the equilibrium of alpha. Individual and population clump files were produced with Structure Harvester [71] and visualized in Distruct1.1 [72].

Orthologous gene family

Protein sequences of cow (Bos taurus), dog (Canis familiaris), human (Homo sapiens), mouse (Mus musculus), and pig (Sus scrofa) were downloaded from the Ensembl database version 72. Protein sequences of minke whale (Balaenoptera acutorostrata) [73] and camel (Camelus bactrianus) [74] were obtained from the original publications. A total of eight species were used to identify orthologous gene clusters with OrthoMCL 2.0.9. Pairwise sequence similarities between all protein sequences were calculated using BLASTP with an e-value cutoff of 1E-05. On the basis of the BLASTP results, OrthoMCL was used to perform a Markov clustering algorithm with inflation value (-I) of 1.5. The OrthoMCL was run with an e-value exponent cutoff of -5 and percent match cutoff of 50%. In total, 5,501 orthologous groups were shared by all eight species. The representative sequences for each gene cluster were selected using the longest horse transcript and the corresponding protein sequences of the other species. BLASTP searches (E-value 1E-5 cutoff) between horse and all the other species were used in this process. Finally, we identified 5,459 1:1:1:1:1:1:1:1 orthologs.

Molecular evolutionary analysis

The phylogenetic tree was constructed from 5,459 single copy ortholog genes. CODEML in PAML 4.5 [75] was used to estimate the d N /d S ratio, where d N indicates nonsynonymous substitution rate and d S indicates synonymous substitution rate. The d N /d S ratio along the horse branch (free-ratio molel) and d N /d S ratio for all branches (one-ratio model) were calculated as the branch model. We also applied the branch-site model to further examine potential positive selection [76]. The LRTs (likelihood ratio tests) were applied to assess statistical significance of the branch-site model. We supposed that positively selected genes are that of having a higher d N /d S ratio with the free-ratios model than that with the one-ratio model and having p-value < 0.05 from branch-site model.

Availability of supporting data

Whole genome sequence data was deposited in the SRA database at NCBI with Biosample accession numbers SAMN02767683. SRA of whole genome sequencing can be accessed via reference numbers SRX535352. The data can also be accessed through BioProject accession number PRJNA246445 for the whole genome sequence data.