Mitochondrial DNA and the origins of the domestic horse

We pooled the mtDNA sequences, covering nps 15,494–15,740, of domestic horses (626 sequences), Przewalski's horses (14 sequences), 1,000- to 2,000-y-old samples from archaeological sites in southern Sweden and Estonia (4 sequences) and 12,000- to 28,000-y old Alaskan permafrost remains (8 sequences), a total of 652 sequences (93 mtDNA types). We reconstructed the unrooted evolutionary network (Fig. 2) and then identified the root of the network by using the most closely related equid outgroups, namely Equus grevyi, Equus kiang, and Equus hemionus. The root turned out to be the central horse node A6, which implies that the Alaskan fossil equids are within the Eurasian mtDNA variation. Vilà et al. (3) placed the Alaskan fossil mtDNA as a sister branch to Eurasian mtDNA, but this discrepancy is explained by their choice of an ass (Equus africanus) as an outgroup: the ass's mtDNA (23) is divergent among equids and appears to have undergone a parallel mutation at np 15,703, influencing the root placement. To calculate phylogenetic time estimates, we postulated within the network one most parsimonious tree (shown in bold in Fig. 2) of the 641 modern horses; that is, we disregarded the ancient Alaskan and Scandinavian horses. We dated the mtDNA root type to be between 342,000 ± 74,000 and 1,198,000 ± 260,000 y old (Table 1). The large uncertainty is caused by the poor calibration of the mtDNA mutation rate (see Materials and Methods). This root type is well distinguished from any other modern equid root type (7) and multifurcates into about 13 horse mtDNA branches. This means 300,000 y approximately represents the latest possible date for the first modern caballine horse.

The most striking feature of the mtDNA network (Fig. 2) is the presence of 17 very frequent mtDNA types, most of which are old enough to have developed a star-like branching structure (a phylogenetic cluster). Each cluster comprising over 2% of the sample is listed in Table 1 along with the corresponding age estimates. For sequence motifs defining each cluster, see Table 2. The clusters are historically interesting in that the ancestral mtDNA type of each has had greater breeding success compared with the other ancestral mtDNA types in the network, which are poorly represented or even extinct. To assess whether domestication, climate, or other factors have led to the success of these types, it is necessary to determine where and when they arose. However, the uncertainty in the mtDNA mutation rate calibration and the large standard deviation for each cluster (Table 1) make it impossible to give an accurate date for the expansion of any particular cluster.

Several clusters are especially interesting from the point of view of geographic distribution and breed. To exclude modern interbreeding as far as possible, we considered in our geographic analysis only horses with a documented geographic origin of their maternal ancestry that corresponds to the origin of the respective breed. For example, if the earliest documented maternal ancestor of a horse sample was Andalusian by breed, but born in Germany, then it was excluded, even though the Spanish origin of Andalusians is not in dispute. Applying this criterion, we included for the geographic analysis only 331 of the 652 horses.

The clearest association between cluster and breed is evidenced by cluster C1 (n = 48): in our sample, it is geographically restricted to central Europe, the British Isles, and Scandinavia, including Iceland (Fig. 1). A total of 17 of 19 documented horses with C1 are northern European ponies (Exmoor, Fjord, Icelandic, and Scottish Highland). Additionally, 14 of 27 undocumented horses (3) with C1 are ponies, including Connemara ponies. The cluster is younger than perhaps 8,000 y (Table 1), but definitely older than 1,500 y, because C1 was also found in two ancient Viking horses. Furthermore, mtDNA cluster E (n = 16) consists entirely of Icelandic, Shetland, and Fjord ponies. Taken together, this suggests a common late glacial or postglacial origin for these pony breeds.

D1 is another geographically striking mtDNA cluster (Fig. 3). It is widespread in our sample, but with a frequency maximum in Iberian (Andalusian and Lusitano) and North African horses (Barbs). As we might expect from the historical presence of the Spanish in North America, a relatively high percentage of American mustangs (31%) also belong to this cluster. It is noteworthy that only a small proportion of the Arabs belong in this cluster (around 5%), regardless of whether we define Arabs by breed or by geography. This genetic result is in accordance with the phenotype; Barb conformation is very similar to that of Iberians, whereas the phenotype of the Arabs is quite distinct from that of Iberians and Barbs (24).

Relatively recent bottlenecks are also reflected in the mitochondria, namely in the Senner, Sorraia, and Dülmener. The first two are locally regarded as indigenous.

Senner horses are a small population near Paderborn, Germany. This population is very small and inbred regarding its maternal line, because all horses descend (21 to 24 generations ago) from the mare David, born 1725 (25). mtDNA cluster G, in which we found all sampled Senners, is very rare in other breeds (2 of 94 Arabs, 1 of 24 Rhineland heavy draft, none in others).

The Sorraias originate from a small group of 7 mares and 4 stallions obtained by R. d'Andrade near Coruche, Portugal, in around 1930, after he had seen a phenotypically identical wild population there in 1920, which was distinct from the local riding horse. According to d'Andrade (26), 5 of the 7 mares passed down mtDNA lineages. All 18 sampled Sorraias have either of two A1 mtDNA types (61% A1 root type, 39% ancestral JSO41 type), which are quite unrelated to the D1 type predominant in the other Iberians.

German Dülmener ponies come from an area where, in AD 1316, “wild” horses were recorded (25). They are supposed to be a mixed breed, but surprisingly, 9 of 10 Dülmeners in our sample belong in cluster D3. Two of these 9 D3 types have derived mutations, suggesting an ancient origin.

Cluster A2, which is restricted to the Przewalski's horses, is problematical. According to studbook records, four mitochondrial lineages survived, and a sequencing study (4) showed that only two mtDNA types exist in these four lineages. However, in the Przewalski's horses' sequences we found three mtDNA types, the third in the data of Ishida et al. (10). Nevertheless, all three Przewalski's horse's mtDNA types are closely related and, in agreement with Ishida et al. (10), we confirm that these types are not found in any other breed.

For the geographic distribution of all clusters, see the supporting information on the PNAS web site.

We used the mtDNA mutation rate to determine the minimum number of mares domesticated from the wild. Excluding the 3 Przewalski's horse mtDNA types, there are 81 domestic horse mtDNA types within the analyzed mtDNA region. Of these domestic mtDNA types, 43 minor (noncluster) types are only one mutation distant from their immediate ancestors (Fig. 2). Using extremely conservative assumptions (see Materials and Methods), it follows that a maximum of 4.4 of the 43 mtDNA types postdate domestication. In other words, 77 successfully breeding mares is a minimum estimate for the number of wild mares ever domesticated. These 77 ancestral types are nearly as diverse as the entire current horse mtDNA pool, and an order of magnitude more diverse than the monophyletic Alaskan wild horse mtDNA or the Przewalski wild horse mtDNA (Table 1).

Does the greater diversity in domesticated horse mtDNA vis-à-vis wild Alaskan and Przewalski‘s horse mtDNA indicate that more than one wild horse population was recruited for domestication, or alternatively, does the difference in diversities imply that the wild Alaskan and Przewalski‘s horses have undergone genetic bottlenecks unrepresentative of ancient wild horses? The former possibility appears more likely; even though Alaskan Pleistocene samples are spread over a time window of 16,000 y, most of them cluster monophyletically (3). As for the Przewalski’s horses, in which only closely related mtDNA types were found, it has been claimed that this low diversity is caused by a recent drastic population decrease (4). This appears unlikely because three of four maternal lines, representing at least two captures separated by 45 y, show an identical mtDNA type. The Przewalski's horse indeed experienced a bottleneck situation, but well-documented bottlenecks in other populations (e.g., Exmoor ponies, which collapsed to 46 mares in the 1940s) have not resulted in such a decrease of mtDNA types. The geographically specific mtDNA clusters which we have found, for example in northern European ponies, agree with a scenario of recruitment of wild mares for domestication from geographically different areas. More evidence might be seen in (controversial) archaeological claims identifying “pony” and “warmblood-like” prehistoric horse phenotypes during and after the Ice Age in Europe (27, 28). Recent research indicated that cattle (29, 30, 31) and goats (32) have had much fewer founder types than horses, but were domesticated more than once. This observation further supports the conclusion that horses were domesticated from several wild populations (2, 3), which individually had a lower genetic diversity.

Assuming our interpretation of multiple genetic horse origins is correct, does it follow that the technique of horse domestication was developed independently by different human communities in different places? From an archaeological and ethological point of view, a single origin of the required human expertise cannot be ruled out. Modern breeding of the wild Przewalski's horse initially encountered problems such as pacing, excessive aggression, impotence, and infanticide (33), leading the Przewalski's horse to the brink of extinction. The Przewalski's horse is not ancestral to domestic horses, but if their wild ancestors were similarly intractable, it is unlikely that the technique was mastered many times independently during prehistory. The ease of domestic horse breeding today may be the genetic consequence of selections of particularly amenable beasts some thousands of years ago (34, 35). Furthermore, if domestication had arisen independently multiple times, one would expect to find archaeological evidence for domestication at very different times and places. This may not be the case. Although there are claims for horse domestication as early as 4500 BC for Iberia and the Eurasian steppe, the earliest undisputed evidence are chariot burials dating to ≈2000 BC from Krivoe Ozero (Sintashta-Petrovka culture) on the Ural steppe (36, 37, 38). Burial, textual, and/or iconographic evidence shows that by 1250 BC, chariots were widespread from Greece to China (37, 39, 40). Such an expansion may suggest a diffusion of the knowledge of horse breeding (34), and possibly a concomitant spread of horses themselves, originally localized both temporally and spacially. In this reading of the archaeological record, the knowledge and the initially domesticated horses themselves would have spread, with local mares incorporated en route, forming our regional mtDNA clusters.

Further resolution of ancient horse domestication may best be achieved by better calibration of the equid mtDNA mutation rate using whole mtDNA and more fossil data. This will allow molecular genetic data to discriminate between the genetic events occurring in the Holocene (e.g., 2000 BC) and those in the Pleistocene. We would claim that a finer resolution of domestic horse origins will be achieved within the analytical context outlined here.

Data deposition: The sequences reported in this paper have been deposited in the GenBank database (accession nos. –).

We thank S. Baker, M. Bowling, V. Eisenmann, L. Kaagan, K.-I. Kim, H. Macgregor, E. A. Oakenfull, and C. Vilà for additional information on their publications, I. Wilson for amending batwing for mtDNA, and the breeders for their assistance with collecting the samples. We are indebted to D. Bradley (Trinity College, Dublin) and J. Mallory (Queen's University, Belfast) for critically evaluating a previous version of the manuscript.