1Central Asia has played a key role in human history, not only as one of the first regions inhabited by humans outside of Africa, but also as the region connecting Africa and Eurasia on the one hand – for example with the Silk Road – and Siberia and Southern Eurasia on the other (Wells et al., 2001; Martínez-Cruz et al., 2011). The region’s location at the crossroads of migratory routes placed it at the hub of cultural, demographic and commercial exchanges, which translates into a long human presence with a history rich in migratory episodes and genetic and cultural admixtures, as confirmed by numerous studies of ancient DNA (Damgaard et al., 2018; de Barros Damgaard et al., 2018; Järve et al., 2019; Narasimhan et al., 2019; Gnecchi-Ruscone et al., 2021). The Bronze Age has been particularly studied through ancient DNA (Narasimhan et al., 2019), as it marks the rise in southern Central Asia of the Oxus civilisation, also called the Bactrio-Margian Archaeological Complex (BMAC). The BMAC is characterized by proto-urban cities, highly efficient irrigation techniques and a marked social hierarchy (Luneau, 2010). At the end of the Bronze Age, from about 1800 BCE, the Oxus civilization underwent major transformations in its final phase, marked by an overall impoverishment of its material culture and technology and the abandon of habitat sites and international trade, with the exception of contacts with the steppes of northern Central Asia (Bendezu-Sarmiento and Lhuillier, 2020; Bonora, 2021). During the period between 1800 and 1500 BC, the Andronovian type of culture took over, until the rise of the Yaz culture (Luneau, 2013; Lhuillier, 2019). After that, several eastward conquests by the Achaemenids, Greeks, Partho-Sassanids and Arabs occurred, as well as westward movements of various Asian peoples such as the Huns, Xiongnus, and Mongols, (Krader, 1966), before the region became a trading centre along the Silk Road. All these events produced great cultural diversity in the region. While this cultural diversity has been widely studied, few genetic studies have been made on the history of South Central Asian populations.
2Today, Central Asia is inhabited by populations making up a wide variety of ethnic groups: Tajiks and Yaghnobis, Turkmens, Uzbeks, Kazaks, Karakalpaks and Kirgiz (Marchi, 2017). These ethnic groups can be divided into two linguistic groups: the Indo-Iranian branch of the Indo-European language family, including Tajiks and Yaghnobis, and the Turkic-Mongolian branch of the Altaic language family (Soucek, 2000), which includes all the other populations (figure 1). Those speaking Turkic-Mongolian languages arrived in Central Asia more recently than the populations speaking Indo-Iranian languages (Johanson, 1998). Most interestingly, strong genetic dissimilarities between the two language groups have been shown, as well as differences in ancestral components (Martínez-Cruz et al., 2011; Marchi et al., 2018). Furthermore, different and specific patterns of social organization and subsistence can be associated with each of the two language groups (Marchi et al., 2017). The Indo-Iranian group consists of sedentary populations practicing agriculture, while the populations speaking Turkic-Mongolian languages are predominantly semi-nomadic and practice herding. Although all are patrilocal, the populations of the two groups differ in their social organization by their rules of alliance and their mode of filiation – cognatism for the Indo-Iranians, patrilineality for the Turko-Mongols.
Figure 1
Distribution of linguistic groups in Central Asia (after Marchi, 2017) |
Répartition des groupes linguistiques en Asie centrale (d’après Marchi, 2017)
3Overall, population genetics studies of the current diversity of Central Asia show that this region has been a zone of contact between two different and well differentiated population groups. On the one hand, a sedentary population, long established in the region and practicing agriculture, represented today by the Tajiks, the Yaghnobis and perhaps the Turkmen, although this group needs to be studied in more detail for evidence of its origin; on the other hand, one or more populations whose arrival in Central Asia is more recent, resulting from the movements of Eastern nomadic groups whose precise origin is still to be determined. Several questions remain unanswered: since when have Indo-Iranian language populations been present in Central Asia? How did they form? What exactly is the origin of the Turkmen group? What exactly is the origin of the populations who formed the Turko-Mongol groups? When did the admixture begin that gave rise to these latter groups?
4To answer these questions, we used detailed modern data from Central Asian individuals, produced in our laboratory, and combined these with a selection of published ancient genomes.
5This combined analysis allowed us:
61) to explore the links between the speakers of Indo-Iranian languages of Central Asia and the ancient populations of southern Central Asia, and to propose a model for the origin of the modern populations by dating the admixture that formed them;
2) to explore the origin of the Turkmen in relation to the history of Indo-Iranian speakers inferred from the first study;
3) to specify the populations that gave rise to the Iron Age individuals of Turkmenistan.
7We selected 3,269 published ancient human genomes from the Palaeolithic to the Middle Ages in Eurasia, including 126 ancient genomes from southern Central Asia (Damgaard et al., 2018; Narasimhan et al., 2019) for which DNA sequencing data were generated by whole-genome capture or hybridization techniques, from the v42.4 dataset available at https://reich.hms.harvard.edu/allen-ancient-dna-resource-aadr-downloadable-genotypes-present-day-and-ancient-dna-data. From this selection we retained unrelated individuals with more than 10,000 Single Nucleotide Polymorphisms (SNPs) covered on the 1,240 k chip. Ancient individuals from three recent publications on the Middle East (Skourtanioti et al., 2020) and the Mongolian steppe (Jeong et al., 2020; Gnecchi-Ruscone et al., 2021) were added.
8The data thus gathered do not document periods and regions uniformly. The Neolithic and Bronze Age are by far the best documented periods in the dataset for southern Central Asia, Iran, Anatolia and the Levant. The Eurasian Steppe (Western, Central and Eastern Steppe) and Europe are the regions for which the most genomes have been published, with a more homogeneous coverage of the different periods.
9The vast majority of published genomes from Central Asia are concentrated in the steppe area. The strong interest in this particular region is explained by the discovery that it is the source of major migrations that changed the European genetic landscape. Furthermore, these results came very early in the history of ancient genomics, with the first large-scale studies publishing hundreds of genome-wide data (Allentoft et al., 2015; Haak et al., 2015; Mathieson et al., 2015).
10To date, only two papers have published ancient genome-wide DNA data from southern Central Asia, as presented in figure 2. The paucity of data is largely attributable to the climate (hot and dry) and more generally to the conditions that make the preservation of ancient DNA very difficult in general. The first paper sequenced four genomes associated with the Namazga III period from the Geoksyur and Kara-Depe sites dating to the Chalcolithic (3,500 AEC) and one genome from the Tarkhibai 3 site (Turkmenistan_IA) dating to the Early Iron Age (Damgaard et al., 2018). The second paper published a large number of genomes from Turkmenistan, Tajikistan, Uzbekistan and Iran (figure 2). This covered the Neolithic and Chalcolithic periods with, among others, the site of Geoksyur and Tepe-Anau in Turkmenistan and the Sarazm site in Tajikistan. However, this paper specifically targeted the period of the Oxus civilization (Bronze Age, with the sites of Bustan, Gonur-Tepe, Dzharkutan, and Sappali-Tepe) and only provided three genomes from the Late Iron Age associated with the Kushan Empire, at the Ksirov site in Tajikistan (Narasimhan et al., 2019).
Figure 2
Locations and periods of genome-wide data available for South Central Asia. Points from a different period but from the same archaeological site have been moved slightly on the map for clarity |
Localisation et âge des données à l’échelle du génome disponibles en Asie centrale méridionale. Les points appartenant au même site archéologique mais à une période différente ont été légèrement décalés pour permettre une meilleure lecture
11To establish a framework for comparison, the ancient dataset was combined with 1,388 Eurasian, 109 Yoruba and 3 Mbuti individuals from two modern public datasets, from the SGDP project (Mallick et al., 2016) and the 1000 Genomes project (The 1000 Genomes Project Consortium et al., 2015). From the resulting dataset, two different datasets were assembled with the modern data from Central Asia generated in our laboratory:
12- The first, called the 300k-dataset, was assembled with the genome-wide capture data obtained in our laboratory including 527 individuals belonging to 18 Central Asia populations (Marchi et al., 2018). Among the 18 Central Asian populations, 4 are Indo-Iranian speakers (TJE), TJA and TAB are Tajiks, and TJY are Yaghnobis. The 300k-dataset includes 242,406 SNPs for 5,129 individuals altogether.
- The second, called the 700k-dataset, was assembled with shotgun-sequenced individuals from only three Central Asian populations: 3 Yaghnobis (TJY), 19 Tajiks (TJE) and 24 Turkmen (TUR), forming a dataset of 716,743 SNPs and 4,648 individuals (Guarino-Vignon et al., 2022). This second assembly was used for analyses requiring more SNPs.
13To obtain an initial idea of the genetic diversity and proximity between the present-day and ancient populations of South Central Asia, a principal component analysis (PCA) was performed with smartpca (Patterson et al., 2006) on 1,388 present-day Eurasian individuals and 527 present- day Central Asian individuals from the 300k-dataset. The 3,269 ancient samples were projected onto the PCA thus calculated. An ADMIXTURE analysis (Alexander et al., 2009) was also performed for the 300k-dataset with a subset of 236,665 SNPs pruned for linkage disequilibrium (PLINK --indep-pairwise 200 25 0.4). The ADMIXTURE analyses were performed for a number of clusters (K) between 2 and 15, with 20 replicates for each K. The K with the best cross-validation score were selected.
14To go beyond simple data description and infer gene flow and admixture, a battery of statistical tools was used on the 700k-dataset (Guarino-Vignon et al., 2022). First, f3-admixture statistics were performed using qp3pop from the ADMIXTOOLS package (Patterson et al., 2012). These are used to test whether a target population can be modelled as the result of admixture between two test populations, and can be written as f3(source1, source2, target). D-statistics were calculated using the qpDstat program of the ADMIXTOOLS package (Patterson et al., 2012) to estimate the genetic contribution of a population (X), through admixture, to a population (Y) by comparison with a second population (Z) (Green et al., 2010). If the statistic D (Outgroup, X; Y, Z)>0, then X and Z are genetically closer than X and Y (and vice versa). To assess whether the D-statistics deviates significantly from zero, p-values calculated from the Z-score were corrected for repeated testing using the Benjamini and Hochberg method (Benjamini and Hochberg, 1995).
15qpAdm analyses with ADMIXTOOLS (Patterson et al., 2012) were conducted to find the best admixture model for Central Asian populations and for the Iron Age Turkmenistan individual, and to estimate the contribution of each source population.
16Finally, DATES software (Chintalapati et al., 2022) was used to estimate the date of the admixture events evidenced by qpAdm, by calculating the number of generations.
17To estimate sex-bias in the admixture, the proportions of the first source population’s contributions for the models selected by qpAdm for the X chromosome alone were calculated, then compared to the proportions obtained for the autosomes. Then the Z-score for the difference between the autosomes and the X chromosome was calculated as follows:
18where ρA and ρX are the proportions of the first source admixture calculated for the autosomes and X chromosome, respectively, and σA and σX are the standard errors (Mathieson et al., 2018). Thus, a negative Z-score means that there is less admixture with the first source for the autosomes than for the X chromosome, indicating that admixture with the first source was female-biased and thus that the second source had more of its males admixed.
19To get an initial idea of the genetic relationships between ancient and current populations, a PCA was performed. On axis 1, the Indo-Iranian group forms a gradient that extends to the south Siberian group. On axis 2, Indo-Iranian populations fall between the Steppe Group and the Ancient Iranian and Ancient South Central Asian Group (figure 3A). Specifically, the Yaghnobi cluster at the beginning of the gradient on axis 1 appears to show less Siberian ancestry than the Tajiks. The Iron Age individuals from southern Central Asia are very close to the Yaghnobi group, but further to the left, probably because they do not have south Siberian ancestry. The three Ksirov-Kushan individuals show substantial disparity.
20In the ADMIXTURE analysis (figure 3B), the Iron Age Turkmen individual shows a remarkably similar profile to the Yaghnobis, suggesting genetic continuity. The only difference is the absence of the Baikal (red) component in Iron Age South Central Asian individuals. As with the PCA, the three Ksirov-Kushan individuals show substantial disparity. Tajik individuals show a structure close to the Yaghnobi structure, with two major differences: they have more of the Baikal component (mean 7% in TJY against mean 14% in Tajiks; t-test p-value=2.10-16), and they have a new component – coloured in orange in figure 3 – which is maximized in South Asian groups (mean 8%). This ancestry is probably due to gene flow from South Asia that did not occur among the Yaghnobis.
21The following assumptions can be drawn from these analyses: modern Indo-Iranians are remarkably close genetically to Iron Age individuals, with limited gene flow from Baikal and, only for Tajiks, from South Asian groups.
Figure 3
Genetic structure of the current populations of Southern Central Asia compared to its ancient populations. A: Zoom on the PCA performed on the modern genomes with the ancient genomes projected. Only ancient individuals from Iran and South Central Asia are shown individually, the other ancient individuals are grouped by region by density lines. B: Result of ADMIXTURE analysis (K=9) for current and ancient South Central Asian populations (yellow rectangle) and for ancient populations maximizing the components found in Yaghnobis (TJY) and Tajiks. EMBA: Early Middle Bronze Age; MLBA: Middle Late Bronze Age; IA: Iron Age; BMAC: Bactrio-MargianArcheological Complex; CHG: Caucasian Hunter-Gatherers; EHG: Eastern Hunter- Gatherers; TJY: Yaghnobi; TJA/TJE/TAB: Tajik groups; TUR: Turkmen; UZB: Uzbek; KAZ: Kazakh; KIB: Kyrgyz; BOU: Buryat; HKS: Khakas |
Structure génétique des populations actuelles de l’Asie centrale méridionale par rapport aux populations anciennes. A : Zoom sur l’ACP réalisée sur les génomes modernes avec projection des génomes anciens. Seuls les individus anciens d’Iran et d’Asie centrale méridionale sont représentés individuellement, les autres individus anciens sont représentés par région par des lignes de densité. B : Résultat de l’analyse ADMIXTURE (K=9) pour les populations actuelles et anciennes d’Asie centrale du Sud (rectangle jaune) et pour les populations anciennes qui maximisent les composantes trouvées chez les Yaghnobis (TJY) et les Tadjiks.EMBA : Âge du Bronze ancien et moyen ; MLBA : Âge du Bronze moyen et tardif ; IA : Âge du Fer ; BMAC : Complexe archeologique bactrio-margien ; CHG : Chasseurs-Cueilleurs du Caucase ; EHG : Chasseurs-Cueilleurs orientaux ; TJY : Yaghnob ; TJA/TJE/TAB : Tadjik ; TUR : Turkmène ; UZB : Ouzbek ; KAZ : Kazakh ; KIB : Kyrgyz ; BOU : Bouryat ; HKS : Khakas
22To formally prove the assumptions made above, D-statistics were calculated with the form D (Outgroup, X; Turkmenistan_IA, Yaghnobis/Tajiks) with Mbuti as the outgroup, and all the ancient populations in our dataset as X. Because of the substantial disparity between the Ksirov-Kushan individuals, only the Iron Age Turkmen individual from Takhirbai 3, Turkmenistan_IA, was used as a representative of the Iron Age in South Central Asia. If the D-stat is positive for an ancient population, then it means that gene flow occurred between that ancient population and the Indo-Iranian group, with regard to the Iron Age individual.
23For the Yaghnobis, no positive D-stats were obtained, showing strong genetic continuity between the Iron Age and the Yaghnobis. For the Tajiks, positive D-stats (N=30) were found for populations of BHG or East Asian ancestry and for some South Asian populations (figure 4). These results show slightly less strong continuity with two gene flows among the Tajiks: one from Southern Siberia/Mongolia and one from South Asia.
Figure 4
Genetic structure of the current populations of Southern Central Asia compared to its ancient populations. A: Zoom on the PCA performed on the modern genomes with the ancient genomes projected. Only ancient individuals from Iran and South Central Asia are shown individually, the other ancient individuals are grouped by region by density lines. B: Result of ADMIXTURE analysis (K=9) for current and ancient South Central Asian populations (yellow rectangle) and for ancient populations maximizing the components found in Yaghnobis (TJY) and Tajiks. EMBA: Early Middle Bronze Age; MLBA: Middle Late Bronze Age; IA: Iron Age; BMAC: Bactrio-MargianArcheological Complex; CHG: Caucasian Hunter-Gatherers; EHG: Eastern Hunter- Gatherers; TJY: Yaghnobi; TJA/TJE/TAB: Tajik groups; TUR: Turkmen; UZB: Uzbek; KAZ: Kazakh; KIB: Kyrgyz; BOU: Buryat; HKS: Khakas |
Flux génétique dans les populations indo-iraniennes et turkmène depuis l’Âge du Fer. Les D-statistiques positives (seules les valeurs significatives sont montrées, i.e. Z>3) de la forme D(Mbuti, population ancienne ; Turkmenistan_IA, TJY/TJE/TUR) corrigées pour les tests répétitifs sont présentées. Une D-statistique positive démontre qu’un flux génétique s’est produit de la population ancienne vers la population indo-iranienne ou la population turkmène par rapport au Turkmenistan_IA. Aucune D-statistique significativement positive pour TJY (Yaghnobi) n’a été trouvée (non montré). La D-statistique est indiquée avec ±3 erreurs standard. TJE : Tadjik ; TUR : Turkmen ; M : Mésolithique ; N : Néolithique ; LN : Néolithique tardif ; EN : Énéolithique ; EBA : Âge du Bronze ancien ; MLBA : Âge du Bronze moyen et tardif ; IA : Âge du Fer ; _o signifie que les individus sont différents génétiquement du groupe principal associe à leur population
24We used qpAdm to model admixture proportions. To test which populations best fit our model, the rotating method was used and all combinations with a p-value <0.01 were excluded. For the Yaghnobi population, two- and three-population admixture models were obtained, still involving about ~88-93% Turkmenistan_IA and ~7-12% Xiongnu, which represents a proxy population of Baikal or North Asian ancestry. Tajiks were modelled and, unlike the Yaghnobis, were unable to fit a two-population admixture model, but only a three-population model (p-value=0.49). This model involved about 17% Xiongnu, between 60 and 75% Turkmenistan_IA, and between 8 and 18% of a South Asian individual (Indian_GreatAndaman_100BP) (Moreno-Mayar et al., 2018) representing ancient south Asian ancestry.
25We also investigated whether the admixture that formed the Indo-Iranian language populations exhibited any sex bias, i.e., whether more males, more females or an equal number of both from the Xiongnu contributed to the admixture. A negative Z-score would mean that there was less admixture with Turkmenistan_IA on the autosomes than on the X chromosome, indicating that admixture with Turkmenistan_IA was female-biased and thus the Xiongnu-like population had more admixed males. For TJE a Z-score of -0.13 was obtained, which is a not a high value (Mathieson et al., 2018; Mittnik et al., 2019), indicating a rather small bias and a fairly balanced admixture between males and females in both populations. For modern Turkmenistan, a Xiongnu + Turkmenistan_IA model was tested for autosomes and for the X chromosome. A sex-bias Z-score of -0.26 was obtained, which is higher than for Tajiks. It could be that the admixture that gave rise to the Turko-Mongols was less balanced and that this shows up in the Turkmen who are slightly mixed with these populations.
26Finally, DATES estimated the date of the admixture events described here at about one thousand years ago for the Yaghnobis (35±15 generations thus ~1,019±447 years ago considering 29 years per generation [Patterson et al., 2012]) and five hundred years for the Tajiks (~546 ±138 years ago [18.8±4.7 generations] to ~907±617 years ago [31.2±21.3 generations]).
27Through comparison with ancient data, present-day Indo-Iranian populations in southern Central Asia were traced back to the Iron Age, with only a slight episode of admixture, providing further evidence for their settlement in the region long before Turkic-Mongolian-speaking populations. This conclusion is consistent with genetic studies based solely on modern DNA (Martínez-Cruz et al., 2011; Palstra et al., 2015) but also with historical (Grousset, 1965) and archaeological (F. Brunet, 1998) studies that had proposed this hypothesis much earlier. The small amount of gene flow from a population of BHG ancestry dating back about 1,000 years that has been highlighted suggests a recent wave of westward migration from the Altai Mountains. The hypothetical migrating population would have been composed mainly of an Ancient Northeast Asian component (ANA, characteristic of the Neolithic BHGs, but also of individuals from the Neolithic of Eastern Russia) and a very small part of the Ancient North Eurasian component (ANE, which is found in present-day Amerindian populations and in ancient individuals from Siberian sites). In this regard, some Xiongnu individuals present a genetic composition that fits this profile (Jeong et al., 2020), but not all of them as Xiongnus exhibit high intragroup diversity (Keyser et al., 2020; Jeong et al., 2020). Nevertheless, Xiongnu migrations are much older than the inferred date of admixture. The analysis of individuals from the early Middle Ages in Mongolia shows that they present a new genetic profile, Ancient East Asian (AEA), which is associated with individuals from East Asia (e.g. Han). However, the AEA component is not found in the Yaghnobis and Tajiks – except in small quantities in the TAB population – which implies that the admixture that gave rise to the Indo-Iranians of Central Asia involved individuals without this component who were therefore relatively old.
28The fine-scale analysis of the ADMIXTURE shows that TABs, Tajiks sampled in the large city of Bukhara in Uzbekistan and Turkic-Mongolian language populations, have an AEA component. A possible scenario is that there were at least two waves of admixture:
29- A first wave of admixture that formed the basal genetic diversity of Turkic-Mongolian and Indo-Iranian speakers, involving fairly ancient eastern steppe nomads with a strong ANA component but lacking an AEA component, in the late Iron Age or during antiquity (Xiongnu or Huns, for example). This admixture is present in small proportions in the remaining Indo-Iranian language populations, but is, in the rest of the region, of great magnitude with very high rates of admixture, forming the basal diversity of Turkic-Mongolian language populations and probably replacing the locally spoken languages with languages of the Turkic-Mongolian family.
- A second wave of admixture involving nomadic groups from East Asia or Mongolia with a high proportion of the AEA component (like the Mongolian invasions), which admixed slightly and only with Turkic-Mongolian language populations, not impacting Indo-Iranian speakers.
30The presence of an AEA component in TABs could be related to their location: the TAB label covers individuals living in a large Uzbek city (Bukhara, population over 200,000 inhabitants). These individuals would be more likely to have a slight admixture with the local Turkic- Mongolian language populations, which would explain the presence of the AEA component.
31This scenario is quite consistent with the conclusions of Martinez-Cruz et al. (2011) who also observed the low impact of westward invasions (Huns, Mongols) on Indo- Iranian groups in Central Asia. The scenario of early admixture is also consistent with the hypotheses of several papers on the origin of Turko-Mongols in Central Asia, which propose that they originated from an ancestral group of Turkic speakers in the Altai region (Li et al., 2009; Martínez-Cruz et al., 2011) whose admixture has been dated to around the 10th-14th centuries (Yunusbayev et al., 2015). Yet linguistic studies have also put forward the hypothesis that the Xiongnus spoke a Turkic language (or languages?) as their primary language (Wright, 1999), and the second miscegenation correlates quite well with linguistic studies of the impact of Mongols on Turkic-Mongolian and Indo-Iranian speakers (Doerfer, 1963). Furthermore, this partial interbreeding with mediaeval eastern nomadic groups, almost exclusively involving Turko-Mongol populations, is consistent with the fact that Turks and Mongols share similar cultural traditions and lifestyles, which may have facilitated intermarriage between groups.
32With the PCA, the Turkmen population is placed in the Indo-Iranian cluster (figure 5A) and with the ADMIXTURE analysis it presents a profile significantly closer to the TABs (urban Tajiks) than to the other Turkic-Mongolian language populations, with, like the TABs, a small proportion of an East Asian ancestry (figure 5B). Indeed, all Turko-Mongol populations in Central Asia except Turkmens show a significantly large amount of the Baikal component (mean value: 50%) (t-test, p-value<2.10-16), while the Baikal component among Turkmens (mean value: 22%) is closer to the proportion among Tajiks (mean value: 15%).
33Furthermore, the identification of an “affinity profile” with all ancient populations in the database through the f3-outgroup (figure 5C-D) shows that the Turkmen profile correlates much better with the Tajik profile (figure 5C) than with the Kazakh profile (figure 5D), which is the typical Turko-Mongol profile from Central Asia. Finally, qpAdm show that the Turkmen can be modelled as resulting from admixture between Tajiks, up to about 90%, and a population with both BHG and East Asian ancestry up to 10%. This admixture is dated by DATES at about 687±100 years ago. All these results show that the Turkmen are probably an Indo- Iranian population that recently changed its language and cultural practice with no significant population replacement.
Figure 5
The Turkmen population belongs to the Indo-Iranian cluster. A: PCA performed on the modern genomes with the ancient genomes projected. Only ancient individuals from Iran and South Central Asia are shown individually, the other ancient individuals are grouped by region by density lines. Turkmens (TUR) are highlighted in dark blue and other Turkic-Mongolian speaking populations (UZB/KAZ/AKZ/KIB/HKS/KEK) in pale blue. B: Result of ADMIXTURE analysis (K=9) showing present-day Indo-Iranian, Turkmen and Turko-Mongol populations. TJA & THE: rural Tajiks; TAB: urban Tajiks. C& D: Turkmen affinity with Tajiks rather than with Turko-Mongol groups shown by f3 statistics of the form f3 (Mbuti; TUR/TJA/AKZ, Ancient population). C: Outgroup f3-statistics for Turkmen and for Tajiks (TJA) plotted against each other. D:Outgroup f3-statistics for Turkmen and for Kazakhs (AKZ), belonging to the Turko-Mongol group, plotted against each other. EMBA: Early Middle Bronze Age; MLBA: Middle Late Bronze Age (modified after Guarino-Vignon et al., 2022) |
La population turkmène appartient au groupe des Indo-Iraniens. A : ACP réalisée sur les génomes modernes avec projection des génomes anciens. Seuls les individus anciens d’Iran et d’Asie centrale du Sud sont représentés individuellement, les autres individus anciens sont visualisés par des lignes de densité par région. Les Turkmènes sont en bleu foncé et les autres populations turco-mongoles en bleu pâle. B : Résultat de l’analyse ADMIXTURE (K=9) avec les Indo-Iraniens, les Turkmènes et les populations turco-mongoles actuels. TJA/TJE : Tadjiks ruraux ; TAB : Tadjiks urbains C&D : Affinité des Turkmènes avec les Tadjiks plutôt qu’avec les groupes turco-mongols, mise en évidence par les statistiques f3 de la forme f3(Mbuti ; TUR/TJA/AKZ, population ancienne).C : Statistiques f3 pour les Turkmènes et pour les Tadjiks (TJA) tracées les unes par rapport aux autres. D : Statistiques f3 pour les Turkmènes et les Kazakhs (AKZ), appartenant au groupe turco-mongol, tracées les unes par rapport aux autres. EMBA : Âge du Bronze ancien et moyen ; MLBA : Âge du Bronze moyen et tardif (modifié d’après Guarino-Vignon et al., 2022)
34This population is a notable example of a population changing their language and cultural practices without any substantial change in its genetic ancestry. This pattern does not correspond to the dominant mechanism in Central Asia, which features strong demic diffusion associated with the arrival of Turkic-Mongolian languages, but fits into the global pattern of Turkic language diffusion (Yunusbayev et al., 2015; Triska et al., 2017): The Turkic language peoples found throughout Eurasia are the result of several nomadic migrations from Siberia through Central Asia to Eastern Europe and the Middle East, which took place over a long period from the fifth to the sixteenth centuries (Yunusbayev et al., 2015; Triska et al., 2017). In regions other than Central Asia, several studies have shown that Turkic language peoples genetically resemble their geographic neighbours, with no clear genetic signal to distinguish them (Yunusbayev et al., 2015; Triska et al., 2017). This supports a model of language replacement by elite dominance rather than demic diffusion.
35Finally, we explored the identity of the steppe population which admixed with BMAC to form the Iron Age population of southern Central Asia (Turkmenistan_IA individual). D-stats, testing whether there is preferential gene flow from a steppe population compared to other steppe populations of the form D (Outgroup, Turkmenistan_IA; Steppe1, Steppe2), were only significant when steppe populations with a clear proximity to BHG or East Asia were contrasted with a population without such proximity, with D-stats showing gene flow from the steppe population without BHG proximity to Turkmenistan_IA. These results indicate that the steppe population involved in the admixture that gave rise to Turkmenistan_IA did not have a component inherited from the BHGs or East Asia. Among qpAdm admixture models involving BMAC on the one hand and about ten different steppe populations on the other, only one model was retained. The only valid model proposed an admixture between BMAC and a population composed of four Andronovo individuals from the Kytmanovo site in the eastern part of the Central Steppe (43% BMAC and 57% Andronovo, p-value=0.31).
36The D-stats analyses show that the Steppe group that admixed with BMAC does not have a BHG component, which is consistent with archaeological (Bendezu-Sarmiento et al., 2007) and genetic (Jeong et al., 2020) findings that the BHG component did not arrive in the Central Steppe until the late Iron Age. The best proxy for admixture with BMAC is the subpopulation formed by four individuals all belonging to Kytmanovo (Allentoft et al., 2015; Rasmussen et al., 2015), north-east Central Asia and Russia. These individuals have a very similar genetic profile to that of individuals from Sintashta, near the Caspian Sea. The geographical distance of these individuals from southern Central Asia does not seem particularly problematic insofar as the existence of a group closer to southern Central Asia, not sequenced to date, and genetically close to the Kytmanovo individuals, is plausible. This assumption seems even less unrealistic in the context of the central steppes where genetic and geographical proximity are not correlated. Indeed, the screening of the individuals associated with the Andronovo horizon in the data shows that they formed a moderately heterogeneous group, with genetic proximities that do not necessarily correlate with geographical or cultural proximity. For example, Srubnaya-Alakulskaya individuals are more closely related to individuals grouped under the precise Andronovo label than Srubnaya from the Samara region (Krzewińska et al., 2018).
37This study sheds new light on Central Asia’s history. Among other findings, our analyses of ancient DNA showed that the Yaghnobis, and to a lesser extent the Tajiks, exhibit remarkable genetic continuity with Iron Age individuals from southern Central Asia. The Yaghnobis received a small gene flow of about 7% from an East Asian population with a genetic background closer to the BHGs, and therefore different from the historical populations from East Asia who have a strong East Asian component and only a weak BHG component. Tajiks, for their part, received an increase in gene flow from the BHG-like population, and our study shows that they also received another gene flow from a South Asian population. This study displays the power of ancient DNA analysis: the combined analysis of present-day southern Central Asian populations and ancient populations produced a more precise and detailed understanding of the population history of Central Asia than the studies already conducted on modern DNA only (Chaix et al., 2008; Heyer et al., 2009; Martínez-Cruz et al., 2011; Palstra et al., 2015; Thouzeau et al., 2017; Marchi et al., 2018). On the one hand, it confirmed well known hypotheses such as the older local origin of Indo-Iranians (Martínez-Cruz et al., 2011; Palstra et al., 2015). On the other hand, and more excitingly and intriguingly, our study unveiled and dated previously unknown admixture events, for example with the BHG-like population or the South Asian population. Ancient DNA also allowed us to explore the Iron Age population of southern Central Asia, the basis of the Indo-Iranian ancestry, which emerged in the early Iron Age from admixture between BMAC and a Bronze Age population close to the Andronovo individuals from the Kytmanovo site. Finally, the genetic proximity of Turkmen to Indo-Iranian speaking individuals has been confirmed, showing that they come from the same genetic background, with only a weak recent gene flow from East Asia. This linguistic and cultural change without any demic diffusion makes the Turkmen an exception in the region where the arrival of the Turkic- Mongolian languages was accompanied by a significant replacement of populations. But Turkmen are part of the classic dynamic of the diffusion of Turkic languages in the other regions of Eurasia, which is barely associated with significant gene flow.
Acknowledgements: First, I would like to express my gratitude to the Société d’Anthropologie de Paris, which did me the honour of awarding me its 2021 PhD prize. I would also like to thank my thesis directors: E. Heyer and C. Bon for their wonderful mentoring. I would like to thank Ludovic Orlando, Laure Ségurel and Paul Verdu for their fruitful comments about this study, as well as Tatyana Hegay, who made sampling in Central Asia possible.