Was the Maikop culture the PIE homeland?

From EBA to MBA
We report new genome-wide data from ten individuals associated with the Yamnaya cultural complex that we refer to as Yamnaya_North_Caucasus (Yamnaya_NC; 3300–2800 BC; Supplementary Table 1). They broadly fall on the EHG–CHG cline of the Steppe groups in PC space (Fig. 3a, Extended Data Fig. 1 and Supplementary Table 14), forming a tight cluster with published data from the Black Sea, Samara and North Caucasus regions. f4-statistics confirm their close genetic similarity, albeit with subtle geographic differences. Responding to previous studies that explored the regions of contact between Yamnaya pastoralists and farming groups5,20,21, we tested a series of possible two-way qpAdm models. Using various Steppe groups as the baseline ancestry and Cucuteni–Trypillia and Globular Amphora (CTC-GAC) or Maykop_main individuals as a second ‘farming-associated’ source, only western Ukraine_Yamnaya can be modelled as a two-way mixture of Steppe_Eneolithic and CTC-GAC, whereas these models are rejected for Yamnaya_NC and Yamnaya_Samara individuals (Supplementary Table 15). However, adding Ukraine_Neolithic and Mesolithic as a third source improved the model fit for almost all groups. Thus, we can model Yamnaya_NC as a three-way mixture of the local proximal sources Steppe_Eneolithic, Maykop and Ukraine_Neolithic (Fig. 3c), although models with CTC-GAC as alternative source(s) are also supported.



a,b, PCA of newly produced ancient individuals (black outline) and individuals from previous publications (no outline) from the third millennium BC (a), and the second millennium BC (b), projected onto 102 modern-day populations (grey dots). The dashed arrows represent observed mixture clines between the Caucasus and Steppe groups and re-emerging gene flow from the northeast. The correspondent labels and groupings are listed in Supplementary Table 5. c,d, Sankey diagram of genetic ancestry modelling for third millennium BC (c) and second millennium BC (d) individuals from this study based on temporally and geographically proximal sources. The admixture proportions (as percentages) are indicated on each ancestry flow, with sources on the left and target populations on the right, and P values for each model in brackets under the population names (Supplementary Tables 15 and 17). The suffixes in the group labels present archaeological time periods and geographical regions: MLBA, Middle–Late BA; BIA, Bronze–Iron Age; IA, Iron Age.

Final MBA and LBA
The final MBA (2200–1650 BC), represented by the post-Catacomb cultural horizon22 and LBA (1800–1200/1000 BC)23 phases, marks another period of increased population interaction and transformation, as evidenced by 33 new individuals from this period. The PC space previously occupied by Steppe cluster individuals is now largely void, with only four individuals of different genetic ancestries (KVO009, KNK006, ESY007 and ESY009) falling in this position. All other kurgan burials in the central steppe zone are shifted towards the Caucasus cluster owing to significantly increased affinity to South Caucasus populations (Z = 3.352; Fig. 3b and Supplementary Table 16). Consequently, these post-Catacomb individuals can be modelled successfully as a mixture of preceding Catacomb (79%) and Kura–Araxes (21%) groups (Fig. 3d and Supplementary Table 17).

Among contemporaneous Steppe populations, individuals of the Lola culture (n = 9) represent the predominant ancestry pattern, which falls close in PC space to earlier Steppe_Maykop individuals5 (Fig. 3b and Extended Data Fig. 2). We separate them into two groups (Lola_1 and Lola_2), on the basis of differing amounts of ANE ancestry indicated by PCA and f4-statistics (Supplementary Table 16). We next tested whether Lola represented a continuation of Steppe_Maykop ancestry that had returned to the North Caucasus, but find that models with Steppe_Maykop as a single ancestry source are rejected, whereas two-way mixtures of Steppe_Maykop and either NCC or Catacomb-associated ancestry are supported (Supplementary Table 17). Considering the 2,000-year time gap between Steppe_Maykop and Lola, we also tested multiple qpAdm models with North Caucasus MBA Steppe groups (for example, Catacomb or NCC) as the local substrate and central steppe Early and Middle BA11 groups as a non-local source of ANE ancestry,and find well-fitted models with Kazakhstan Kumsay EBA. The individual KVO013 from the Northwest Caucasus falls close in PC space to BA Srubnaya individuals from the eastern European forest steppe (Fig. 3b) and can be modelled as a mixture of preceding BA Sintashta11 and Steppe groups (Fig. 3d and Supplementary Table 17). The LBA Prescythian_steppe individuals ESY006, ESY007, ESY009 and KNK006 represent the last signal of local Steppe ancestry in the North Caucasus (Fig. 3b). However, f4-symmetry tests show that these individuals carry traces of ANE ancestry similar to Lola and Steppe_Maykop, and can be modelled as a two-way mixture of Srubnaya and Lola_1 ancestry (Supplementary Tables 16 and 17).

The final MBA and LBA individuals from the Caucasus cluster are markedly shifted upwards on PC2 towards the Steppe cluster (Fig. 3b). This marks the first time in which ancient individuals fall within the same PC space as present-day populations. Individuals from the western and eastern Caucasus highlands are spread along PC2, implying varying levels of admixture with the Steppe cluster (Fig. 3b and Extended Data Fig. 2). The individuals KVO008 (MBA_Caucasus) and PUT001 (Arkhon) carry similar ancestry profiles, suggesting that a convergence of both clusters might not have been restricted to the Caucasus highlands. Using f4-statistics, we find that most MBA and LBA individuals show an affinity to Steppe groups (Supplementary Table 16). Further, compared to the MBA_Highland_east group, KVO008 and PUT001 show an even higher genetic affinity to EHG–WSHG and BA Steppe groups, and both can be modelled successfully with Kura–Araxes as a proxy for Caucasus ancestry and BA Steppe groups as sources (Supplementary Tables 16 and 17). This two-way model is also supported for MBA and LBA eastern highland individuals, albeit with a higher proportion of Kura–Araxes-related ancestry, but can also be modelled with the preceding MBA_Highland_east group as a single source (Supplementary Table 17), implying that this gene flow had occurred already before the final MBA. Moreover, we observe a genetic cline from east to west, suggesting geographic structure in LBA highlander ancestry (Fig. 3b). Here, individuals from Shushuk and Marchenkova Gora in the northwest share more genetic drift with BA Srubnaya and Steppe groups compared to eastern individuals from Ginchi and Gatyn-Kale, which is consistent with the finding of the Srubnaya-associated individual KVO013 from the Northwest Caucasus (Figs. 1a and 3b). Western highlanders also show a greater genetic affinity with Maykop individuals, whereas eastern highlanders are more similar to Kura–Araxes individuals (Supplementary Table 16). Using Maykop and Kura–Araxes groups as the respective locally preceding ancestry source and Srubnaya as the second source resulted in well-fitted models (Supplementary Table 17). The five LBA individuals from the site of Lernakert in Armenia (1411–1266 cal BC) occupy a similar position in the PCA as the published MBA and LBA individuals from other sites in Armenia, suggesting local genetic continuity20,24. Using Armenia_MBA as the local baseline in f4-statistics and qpAdm, we find support for a single local source.