We analyzed whole-genome sequence data of 74 ancient humans (14, 15) (tables S1 to S3) ranging from the Mesolithic (~9000 BCE) to Medieval times, spanning ~5000 km across Eastern Europe, Central Asia, and Western Asia (Anatolia) (Fig. 1). Our genome data includes 3 Copper Age individuals (~3500 to 3300 BCE) from Botai in northern Kazakhstan (Botai_CA; 13.6X, 3.7X, and 3X coverage, respectively); 1 Early Bronze Age (~2900 BCE) Yamnaya sample from Karagash, Kazakhstan (16) (YamnayaKaragash_EBA; 25.2X); 1 Mesolithic (~9000 BCE) EHG from Sidelkino, Russia (SidelkinoEHG_ML; 2.9X); 2 Early/Middle Bronze Age (~2200 BCE) central steppe individuals (~4200 BP) (CentralSteppe_EMBA; 4.5X and 9.1X average coverage, respectively) from burials at Sholpan and Gregorievka that display cultural similarities to Yamnaya and Afanasievo (12); 19 individuals of the Bronze Age (~2500 to 2000 BCE) Okunevo culture of the Minusinsk Basin in the Altai region (Okunevo_EMBA; ~1X average coverage; 0.1 to 4.6X); 31 Baikal hunter-gatherer genomes (~1X average coverage; 0.2 to 4.5X) from the cis-Baikal region bordering on Mongolia and ranging in time from the Early Neolithic (~5200 to 4200 BCE; Baikal_EN) to the Early Bronze Age (~2200 to 1800 BCE; Baikal_EBA); 4 Copper Age individuals (~3300 to 3200 BCE; Namazga_CA; ~1X average coverage; 0.1 to 2.2X) from Kara-Depe and Geoksur in the Kopet Dag piedmont strip of Turkmenistan, affiliated with the period III cultural layers at Namazga-Depe (fig. S1), plus 1 Iron Age individual (Turkmenistan_IA; 2.5X) from Takhirbai in the same area dated to ~800 BCE; and 12 individuals from Central Turkey (figs. S2 to S4), spanning from the Early Bronze Age (~2200 BCE; Anatolia_EBA) to the Iron Age (~600 BCE; Anatolia_IA), and including 5 individuals from presumed Hittite-speaking settlements (~1600 BCE; Anatolia_MLBA), and 2 individuals dated to the Ottoman Empire (1500 CE; Anatolia_Ottoman; 0.3 to 0.9X). All the population labels including those referring to previously published ancient samples are listed in table S4 for contextualization. Additionally, we sequenced 41 high-coverage (30X) present-day Central Asian genomes, representing 17 self-declared ethnicities (fig. S5), and collected and genotyped 140 individuals from five IE-speaking populations in northern Pakistan.

Tests indicated that the contamination proportion of the data was negligible (14) (see table S1), and we removed related individuals from frequency-based statistics (fig. S6 and table S5). Our high-coverage Yamnaya genome from Karagash is consistent with previously published Yamnaya and Afanasievo genomes, and our Sidelkino genome is consistent with previously published EHG genomes, on the basis that there is no statistically significant deviation from 0 of D statistics of the form D(Test, Mbuti; SidelkinoEHG_ML, EHG) (fig. S7) or of the form D(Test, Mbuti; YamnayaKaragash_EBA, Yamnaya) (fig. S8; additional D statistics shown in figs. S9 to S12).

In the Early Bronze Age, ~3000 BCE, the Afanasievo culture was formed in the Altai region by people related to the Yamnaya, who migrated 3000 km across the central steppe from the western steppe (1) and are often identified as the ancestors of the IE-speaking Tocharians of first-millennium northwestern China (4, 6). At this time, the region they passed through was populated by horse hunter-herders (4, 10, 17), while further east the Baikal region hosted groups that had remained hunter-gatherers since the Paleolithic (18–22). Subsequently, the Okunevo culture replaced the Afanasievo culture. The genetic origins and relationships of these peoples have been largely unknown (23, 24).

To address these issues, we characterized the genomic ancestry of the local Inner Asian populations around the time of the Yamnaya and Afanasievo expansion. Comparing our ancient samples to a range of present-day and ancient samples with principal components analysis (PCA), we find that the Botai_CA, CentralSteppe_EMBA, Okunevo_EMBA, and Baikal populations (Baikal_EN and Baikal_EBA) are distributed along a previously undescribed genetic cline. This cline extends from the EHG of the western steppe to the Bronze Age (~2000 to 1800 BCE) and Neolithic (~5200 to 4200 BCE) hunter-gatherers of Lake Baikal in Central Asia, which are located on the PCA plot close to modern East Asians and two Early Neolithic (~5700 BCE) Devil’s Gate samples (25) (Fig. 2 and fig. S13). In accordance with their position along the west-to-east gradient in the PCA, increased East Asian ancestry is evident in ADMIXTURE model-based clustering (Fig. 3 and figs. S14 and S15) and by D statistics for Sholpan and Gregorievka (CentralSteppe_EMBA) and Okunevo_EMBA, relative to Botai_CA and the Baikal_EN sample: D(Baikal_EN, Mbuti; Botai_CA, Okunevo_EMBA) = –0.025 Z = –12; D(Baikal_EN, Mbuti; Botai_CA, Sholpan) = –0.028 Z = –8.34; D(Baikal_EN, Mbuti; Botai_CA, Gregorievka) = –0.026 Z = –7.1. The position of this cline suggests that the central steppe Bronze Age populations all form a continuation of the Ancient North Eurasian (ANE) population, previously known from the 24,000-year-old Mal’ta (MA1), the 17,000-year-old AG-2 (26), and the ~14,700-year-old AG-3 (27) individuals from Siberia.

To investigate ancestral relationships between these populations, we used coalescent modeling with the momi (Moran Models for Inference) program (28) (Fig. 4, figs. S16 to S22, and tables S6 to S11). This exploits the full joint-site frequency spectrum and can separate genetic drift into divergence-time and population-size components, in comparison to PCA, admixture, and qpAdm approaches, which are based on pairwise covariances. We find that Botai_CA, CentralSteppe_EMBA, Okunevo_EMBA, and Baikal populations are deeply separated from other ancient and present-day populations and are best modeled as mixtures in different proportions of ANE ancestry and an Ancient East Asian (AEA) ancestry component represented by Baikal_EN, with mixing times dated to ~5000 BCE. Although some modern Siberian samples lie under the Baikal samples in Fig. 2A, these are separated out in a more limited PCA, involving just those populations and the ancient samples (fig. S23). Our momi model infers that the ANE lineage separated ~15,000 years ago in the Upper Paleolithic from the EHG lineage to the west, with no independent drift assigned to MA1. This suggests that MA1 may represent their common ancestor. Similarly, the AEA lineage to the east also separated ~15,000 years ago, with the component that leads to Baikal_EN and the AEA component of the steppe separating from the lineage leading to present-day East Asian populations represented by Han Chinese (figs. S19 to S21). The ANE and AEA lineages themselves are estimated as having separated approximately 40,000 years ago, relatively soon after the peopling of Eurasia by modern humans.

Because the ANE MA1 sample comes from the same cis-Baikal region as the AEA-derived Neolithic samples analyzed here, we document evidence for a population replacement between the Paleolithic and the Neolithic in this region. Furthermore, we observe a shift in genetic ancestry between the Early Neolithic (Baikal_EN) and the Late Neolithic/Bronze Age hunter-gatherers (Baikal_LNBA) (Fig. 2A), with the Baikal_LNBA cluster showing admixture from an ANE-related source. We estimate the ANE related ancestry in the Baikal_LNBA to be ~5 to 11% (qpAdm) (table S12) (2), using MA1 as a source of ANE, Baikal_EN as a source of AEA, and a set of six outgroups. However, neither MA1 nor any of the other steppe populations lie in the direction of Baikal_LNBA from Baikal_EN on the PCA plot (fig. S23). This suggests that the new ANE ancestry in Baikal_LNBA stems from an unsampled source. Given that this source may have harbored East Asian ancestry, the contribution may be larger than 10%.

These serial changes in the Baikal populations are reflected in Y-chromosome lineages (Fig. 5A, figs. S24 to S27, and tables S13 and S14). MA1 carries the R haplogroup, whereas the majority of Baikal_EN males belong to N lineages, which were widely distributed across Northern Eurasia (29), and the Baikal_LNBA males all carry Q haplogroups, as do most of the Okunevo_EMBA as well as some present-day Central Asians and Siberians. Mitochondrial haplogroups show less turnover (Fig. 5B and table S15), which could either indicate male-mediated admixture or reflect bottlenecks in the male population.

The deep population structure among the local populations in Inner Asia around the Copper Age/Bronze Age transition is in line with distinct origins of central steppe hunter-herders related to Botai of the central steppe and those related to Altaian hunter-gatherers of the eastern steppe (30). Furthermore, this population structure, which is best described as part of the ANE metapopulation, persisted within Inner Asia from the Upper Paleolithic to the end of the Early Bronze Age. In the Baikal region, the results show that at least two genetic shifts occurred: first, a complete population replacement of the Upper Paleolithic hunter-gatherers belonging to the ANE by Early Neolithic communities of Ancient East Asian ancestry, and second, an admixture event between the latter and additional members of the ANE clade, occurring during the 1500-year period that separates the Neolithic from the Early Bronze Age. These genetic shifts complement previously observed severe cultural changes in the Baikal region (18–22).

The earliest unambiguous evidence for horse husbandry is from the Copper Age Botai hunter-herder culture of the central steppe in Northern Kazakhstan ~3500 to 3000 BCE (5, 10, 23, 31–33). There was extensive debate over whether Botai horses were hunted or herded (33), but more recent studies have evidenced harnessing and milking (10, 17), the presence of likely corrals, and genetic domestication selection at the horse TRPM1 coat-color locus (32). Although horse husbandry has been demonstrated at Botai, it is also now clear from genetic studies that this was not the source of modern domestic horse stock (32). Some have suggested that the Botai were local hunter-gatherers who learned horse husbandry from an early eastward spread of western pastoralists, such as the Copper Age herders buried at Khvalynsk (~5150 to 3950 BCE), closely related to Yamnaya and Afanasievo (17). Others have suggested an in situ transition from the local hunter-gatherer community (5).

We therefore examined the genetic relationship between Yamnaya and Botai. First, we note that whereas Yamnaya is best modeled as an approximately equal mix of EHG and Caucasian HG ancestry and that the earlier Khvalynsk samples from the same area also show Caucasian ancestry, the Botai_CA samples show no signs of admixture with a Caucasian source (fig. S14). Similarly, while the Botai_CA have some Ancient East Asian ancestry, there is no sign of this in Khvalynsk or Yamnaya. Our momi model (Fig. 4) suggests that, although YamnayaKaragash_EBA shared ANE ancestry with Botai_CA from MA1 through EHG, their lineages diverge ~15,000 years ago in the Paleolithic. According to a parametric bootstrap, the amount of gene flow between YamnayaKaragash_EBA and Botai_CA inferred using the sample frequency spectrum (SFS) was not significantly different from 0 (P = 0.18 using 300 parametric bootstraps under a null model without admixture) (fig. S18). Additionally, the best-fitting SFS model without any recent gene flow fits the ratio of ABBA-BABA counts for (SidelkinoEHG_ML, YamnayaKaragash_EBA; Botai_CA, AncestralAllele), with Z = 0.45 using a block jackknife for this statistic. Consistent with this, a simple qpGraph model without direct gene flow between Botai_CA and Yamnaya, but with shared EHG-related ancestry between them, fits all f4 statistics (fig. S28), and qpAdm (2) successfully fits models for Yamnaya ancestry without any Botai_CA contribution (table S12).

The separation between Botai and Yamnaya is further reinforced by a lack of overlap in Y-chromosomal lineages (Fig. 5A). Although our YamnayaKaragash_EBA sample carries the R1b1a2a2c1 lineage seen in other Yamnaya and present-day Eastern Europeans, one of the two Botai_CA males belongs to the basal N lineage, whose subclades have a predominantly Northern Eurasian distribution, whereas the second carries the R1b1a1 haplogroup, restricted almost exclusively to Central Asian and Siberian populations (34). Neither of these Botai lineages has been observed among Yamnaya males (table S13 and fig. S25).

Using ChromoPainter (35) (figs. S29 to S32) and rare variant sharing (36) (figs. S33 to S35), we also identify a disparity in affinities with present-day populations between our high-coverage Yamnaya and Botai genomes. Consistent with previous results (1, 2), we observe a contribution from YamnayaKaragash_EBA to present-day Europeans. Conversely, Botai_CA shows greater affinity to Central Asian, Siberian, and Native American populations, coupled with some sharing with northeastern European groups at a lower level than that for Yamnaya, due to their ANE ancestry.

Further toward the Altai, the genomes of two CentralSteppe_EMBA women, who were buried in Afanasievo-like pit graves, revealed them to be representatives of an unadmixed Inner Asian ANE-related group, almost indistinguishable from the Okunevo_EMBA of the Minusinsk Basin north of the Altai through D statistics (fig. S11). This lack of genetic and cultural congruence may be relevant to the interpretation of Afanasievo-type graves elsewhere in Central Asia and Mongolia (37). However, in contrast to the lack of identifiable admixture from Yamnaya and Afanasievo in the CentralSteppe_EMBA, there is an admixture signal of 10 to 20% Yamnaya and Afanasievo in the Okunevo_EMBA samples (fig. S21), consistent with evidence of western steppe influence. This signal is not seen on the X chromosome (qpAdm P value for admixture on X 0.33 compared to 0.02 for autosomes), suggesting a male-derived admixture, also consistent with the fact that 1 of 10 Okunevo_EMBA males carries a R1b1a2a2 Y chromosome related to those found in western pastoralists (Fig. 5). In contrast, there is no evidence of western steppe admixture among the more eastern Baikal region Bronze Age (~2200 to 1800 BCE) samples (fig. S14).

The lack of evidence of admixture between Botai horse herders and western steppe pastoralists is consistent with these latter migrating through the central steppe but not settling until they reached the Altai to the east (4). Notably, this lack of admixture suggests that horses were domesticated by hunter-gatherers not previously familiar with farming, as were the cases for dogs (38) and reindeer (39). Domestication of the horse thus may best parallel that of the reindeer, a food animal that can be milked and ridden, which has been proposed to be domesticated by hunters via the “prey path” (40); indeed, anthropologists note similarities in cosmological beliefs between hunters and reindeer herders (41). In contrast, most animal domestications were achieved by settled agriculturalists (5).

The presence of Western Eurasian ancestry in many present-day South Asian populations south of the central steppe has been used to argue for gene flow from Early Bronze Age (~3000 to 2500 BCE) western steppe pastoralists into the region (42, 43). However, direct influence of Yamnaya or related cultures of that period is not visible in the archaeological record, except perhaps for a single burial mound in Sarazm in present-day Tajikistan of contested age (44, 45). Additionally, linguistic reconstruction of protoculture coupled with the archaeological chronology evidences a Late (~2300 to 1200 BCE) rather than Early Bronze Age (~3000 to 2500 BCE) arrival of the Indo-Iranian languages into South Asia (16, 45, 46). Thus, debate persists as to how and when Western Eurasian genetic signatures and IE languages reached South Asia.

To address these issues, we investigated whether the source of the Western Eurasian signal in South Asians could derive from sources other than Yamnaya and Afanasievo (Fig. 1). Both Early Bronze Age (~3000 to 2500 BCE) steppe pastoralists Yamnaya and Afanasievo and Late Bronze Age (~2300 to 1200 BCE) Sintashta and Andronovo carry substantial amounts of EHG and CHG ancestry (1, 2, 7), but the latter group can be distinguished by a genetic component acquired through admixture with European Neolithic farmers during the formation of the Corded Ware complex (1, 2), reflecting a secondary push from Europe to the east through the forest-steppe zone.

We characterized a set of four south Turkmenistan samples from Namazga period III (~3300 BCE). In our PCA analysis, the Namazga_CA individuals were placed in an intermediate position between Iran Neolithic and western steppe clusters (Fig. 2). Consistent with this, we find that the Namazga_CA individuals carry a significantly larger fraction of EHG-related ancestry than Neolithic skeletal material from Iran [D(EHG, Mbuti; Namazga_CA, Iran_N) Z = 4.49], and we are not able to reject a two-population qpAdm model in which Namazga_CA ancestry was derived from a mixture of Neolithic Iranians and EHG (~21%) (P = 0.49).

Although CHG contributed both to Copper Age steppe individuals (e.g., Khvalynsk, ~5150 to 3950 BCE) and substantially to Early Bronze Age (~3000 to 2500 BCE) steppe Yamnaya and Afanasievo (1, 2, 7, 47), we do not find evidence of CHG-specific ancestry in Namazga. Despite the adjacent placement of CHG and Namazga_CA on the PCA plot, D(CHG, Mbuti; Namazga_CA, Iran_N) does not deviate significantly from 0 (Z = 1.65), in agreement with ADMIXTURE results (Fig. 3 and fig. S14). Moreover, a three-population qpAdm model using Iran Neolithic, EHG, and CHG as sources yields a negative admixture coefficient for CHG. This suggests that while we cannot totally reject a minor presence of CHG ancestry, steppe-related admixture most likely arrived in the Namazga population before the Copper Age or from unadmixed sources related to EHG. This is consistent with the upper temporal boundary provided by the date of the Namazga_CA samples (~3300 BCE). In contrast, the Iron Age (~900 to 200 BCE) individual from the same region as Namazga (sample DA382, labeled Turkmenistan_IA) is closer to the steppe cluster in the PCA plot and does have CHG-specific ancestry. However, it also has European farmer–related ancestry typical of Late Bronze Age (~2300 to 1200 BCE) steppe populations (1–3, 47) [D(Neolithic European, Mbuti; Namazga_CA, Turkmenistan_IA) Z = -4.04], suggesting that it received admixture from Late (~2300 to 1200 BCE) rather than Early Bronze Age (~3000 to 2500 BCE) steppe populations.

In a PCA focused on South Asia (Fig. 2B), the first dimension corresponds approximately to west-east and the second dimension to north-south. Near the lower right are the Andamanese Onge, previously used to represent the Ancient South Asian component (12, 42). Contemporary South Asian populations are placed along both east-west and north-south gradients, reflecting the presence of three major ancestry components in South Asia deriving from West Eurasians, South Asians, and East Asians. Because the Namazga_CA individuals appear at one end of the West Eurasian/South Asian axis, and given their geographical proximity to South Asia, we tested this group as a potential source in a set of qpAdm models for the South Asian populations (Fig. 6).

We are not able to reject a two-population qpAdm model using Namazga_CA and Onge for nine modern southern and predominantly Dravidian-speaking populations (Fig. 6, fig. S36, and tables S16 and S17). In contrast, for seven other populations belonging to the northernmost Indic- and Iranian-speaking groups, this two-population model is rejected, but not a three-population model including an additional Late Bronze Age (~2300 to 1200 BCE) steppe source. Last, for seven southeastern Asian populations, six of which were Tibeto-Burman or Austro-Asiatic speakers, the three-population model with Late Bronze Age (~2300 to 1200 BCE) steppe ancestry was rejected, but not a model in which Late Bronze Age (~2300 to 1200 BCE) steppe ancestry was replaced with an East Asian ancestry source, as represented by the Late Iron Age (~200 BCE to 100 CE) Xiongnu (Xiongnu_IA) nomads from Mongolia (3). Interestingly, for two northern groups, the only tested model we could not reject included the Iron Age (~900 to 200 BCE) individual (Turkmenistan_IA) from the Zarafshan Mountains and the Xiongnu_IA as sources. These findings are consistent with the positions of the populations in PCA space (Fig. 2B) and are further supported by ADMIXTURE analysis (Fig. 3), with two minor exceptions: In both the Iyer and the Pakistani Gujar, we observe a minor presence of the Late Bronze Age (~2300 to 1200 BCE) steppe ancestry component (fig. S14) not detected by the qpAdm approach. Additionally, we document admixture along the West Eurasian and East Asian clines of all South Asian populations using D statistics (fig. S37).

Thus, we find that ancestries deriving from four major separate sources fully reconcile the population history of present-day South Asians (Figs. 3 and 6), one anciently South Asian, one from Namazga or a related population, a third from Late Bronze Age (~2300 to 1200 BCE) steppe pastoralists, and one from East Asia. They account for western ancestry in some Dravidian populations that lack CHG-specific ancestry while also fitting the observation that whenever there is CHG-specific ancestry and considerable EHG ancestry, there is also European Neolithic ancestry (Fig. 3). This implicates Late Bronze Age (~2300 to 1200 BCE) steppe rather than Early Bronze Age (~3000 to 2500 BCE) Yamnaya and Afanasievo admixture into South Asia. The proposal that the IE steppe ancestry arrived in the Late Bronze Age (~2300 to 1200 BCE) is also more consistent with archaeological and linguistic chronology (44, 45, 48, 49). Thus, it seems that the Yamnaya- and Afanasievo-related migrations did not have a direct genetic impact in South Asia.

Finally, we consider the evidence for Bronze Age steppe genetic contributions in West Asia. There are conflicting models for the earliest dispersal of IE languages into Anatolia (4, 50). The now extinct Bronze Age Anatolian language group represents the earliest historically attested branch of the IE language family and is linguistically held to be the first branch to have split off from PIE (51, 52, 53). One key question is whether Proto-Anatolian is a direct linguistic descendant of the hypothesized Yamnaya PIE language or whether Proto-Anatolian and the PIE language spoken by Yamnaya were branches of a more ancient language ancestral to both (49, 53). Another key question relates to whether Proto-Anatolian speakers entered Anatolia as a result of a Copper Age western steppe migration (~5000 to 3000 BCE) involving movement of groups through the Balkans into Northwest Anatolia (4, 54, 55) or a Caucasian route that links language dispersal to intensified north-south population contacts facilitated by the trans-Caucasian Maykop culture ~3700 to 3000 BCE (50, 54).

Ancient DNA findings suggest extensive population contact between the Caucasus and the steppe during the Copper Age (~5000 to 3000 BCE) (1, 2, 42). Particularly, the first identified presence of Caucasian genomic ancestry in steppe populations is through the Khvalynsk burials (2, 47) and that of steppe ancestry in the Caucasus is through Armenian Copper Age individuals (42). These admixture processes likely gave rise to the ancestry that later became typical of the Yamnaya pastoralists (7), whose IE language may have evolved under the influence of a Caucasian language, possibly from the Maykop culture (50, 56). This scenario is consistent with both the Copper Age steppe (4) and the Caucasian models for the origin of the Proto-Anatolian language (57).

PCA (Fig. 2B) indicates that all the Anatolian genome sequences from the Early Bronze Age (~2200 BCE) and Late Bronze Age (~1600 BCE) cluster with a previously sequenced Copper Age (~3900 to 3700 BCE) individual from Northwestern Anatolia and lie between Anatolian Neolithic (Anatolia_N) samples and CHG samples but not between Anatolia_N and EHG samples. A test of the form D(CHG, Mbuti; Anatolia_EBA, Anatolia_N) shows that these individuals share more alleles with CHG than Neolithic Anatolians do (Z = 3.95), and we are not able to reject a two-population qpAdm model in which these groups derive ~60% of their ancestry from Anatolian farmers and ~40% from CHG-related ancestry (P = 0.5). This signal is not driven by Neolithic Iranian ancestry, because the result of a similar test of the form D(Iran_N, Mbuti; Anatolia_EBA, Anatolia_N) does not deviate from zero (Z = 1.02). Taken together with recent findings of CHG ancestry on Crete (58), our results support a widespread CHG-related gene flow, not only into Central Anatolia but also into the areas surrounding the Black Sea and Crete. The latter are not believed to have been influenced by steppe-related migrations and may thus correspond to a shared archaeological horizon of trade and innovation in metallurgy (59).

Importantly, a test of the form D(EHG, Mbuti; Anatolia_EBA, Anatolia_MLBA) supports that the Central Anatolian gene pools, including those sampled from settlements thought to have been inhabited by Hittite speakers, were not affected by steppe populations during the Early and Middle Bronze Age (Z = –1.83). Both of these findings are further confirmed by results from clustering analysis (Fig. 3). The CHG-specific ancestry and the absence of EHG-related ancestry in Bronze Age Anatolia would be in accordance with intense cultural interactions between populations in the Caucasus and Anatolia observed during the late fifth millennium BCE that seem to come to an end in the first half of the fourth millennium BCE with the village-based egalitarian Kura-Araxes’ society (60, 61), thus preceding the emergence and dispersal of Proto-Anatolian.

Our results indicate that the early spread of IE languages into Anatolia was not associated with any large-scale steppe-related migration, as previously suggested (62). Additionally, and in agreement with the later historical record of the region (63), we find no correlation between genetic ancestry and exclusive ethnic or political identities among the populations of Bronze Age Central Anatolia, as has previously been hypothesized (64).