Tracing the spread of Celtic languages using ancient genomics

Europe’s present-day linguistic landscape is dominated by Romance, Germanic and Slavic languages, which expanded relatively recently during the Iron Age^10,11,12,13. Before these events, Celtic languages were prominent across Europe^1,2,3: Lepontic has been documented in northwestern Italy (2600–2100 bp), Celtiberian in central-northeastern Iberia (2200–1900 bp) and Gaulish across present-day France, central Europe, northern Italy and the Balkans (2500–1400 bp)^14,15 (Fig. 1a). In the modern era, Celtic languages have retreated to the northwestern Atlantic fringe, where they persist as two groups: the Goidelic languages Irish, Manx and Scottish Gaelic; and the Brittonic languages Welsh, Cornish and Breton.

All Celtic languages diverged from an undocumented ancestral language, Proto-Celtic (Fig. 1b,c and Extended Data Fig. 1a). This language is defined by a series of linguistic innovations present in all Celtic languages, but absent from the other Indo-European languages, including closely related Italic and Lusitanian (Supplementary Note 5). Whereas there is widespread agreement that Indo-European dialects, including the one ancestral to Proto-Celtic, spread across Europe from the Pontic Steppe after 5000 bp (refs. ^16,17,18,19), the centre and timing of the subsequent dispersal of Celtic within Europe remain contentious⁴. Existing hypotheses generally fall within one of the following four models:

1. (1)

   The Bell Beaker model associates the Celtic split and dispersal with the Late Neolithic to Early Bronze Age Bell Beaker Complex (roughly 4750–4500 to 4000–3800 bp)^20,21,22,23, which spanned much of western and central Europe. Under this model, Celtic diverged largely in situ over the millennia following the Bell Beaker expansion. By contrast, the following three models place the formation of Proto-Celtic after 4000 bp.

2. (2)

   The Celtic-from-the-West model (Fig. 1a) claims that Proto-Celtic emerged from relatively undiversified Indo-European dialects within long-distance maritime networks of the Atlantic Bronze Age Culture (3300–2700 bp)²⁴, connecting Atlantic Iberia, parts of France, Ireland and Britain. It further posits a subsequent eastward expansion of a Gaulish–Brittonic–Goidelic Celtic subgroup into central Europe during the Hallstatt C period (2750–2600 bp), traced archaeologically through the distribution of Gündlingen swords²⁵.

3. (3)

   The Celtic-from-the-Centre model, in turn, posits that Proto-Celtic emerged neither along the Atlantic seaboard nor from eastern central Europe (next model). Instead, it situates this ancestral language in inland France and possibly northwestern Italy 4000–3000 bp, with a subsequent spread in many directions, including to Britain, Italy and Spain, after 3000 bp (refs. ^26,27,28) (Fig. 1a). This model rejects overly strong archaeological–linguistic correlations and is instead supported by a relative density of Celtic toponyms and personal names in France.

4. (4)

   Finally, the longstanding Celtic-from-the-East model places the formation of Proto-Celtic in eastern central Europe (Fig. 1a) and associates its diversification with the Late Bronze Age Urnfield Culture^29,30, the subsequent Hallstatt Culture and La Tène archaeological horizon^5,6,7,8, or a combination of these. The documentation of Lepontic Celtic inscriptions in the Urnfield-derived Alpine Golasecca Culture (2850–2350 bp) has been taken to support this model³⁰. In addition, under this model (as well as models (2) and (3)), the geographic distribution of the Iron Age La Tène Culture, extending from France to the Balkans, broadly coincides with the distribution of Gaulish and closely related Celtic dialects³¹.

These models differ in their proposed centres and chronologies for the formation and spread of Celtic and imply conflicting proposals of human mobility. We therefore test to what degree each model is supported by past population genomic data, similar to approaches that have proved to be illuminative in other historical-linguistic debates^{11,12,16,17,32,33,34,35}. We note, in this context, that the terms ‘spread’ and ‘population movements’ are used as purely descriptive terms that do not speak to the social mechanism driving them.

We expand on the results of a parallel study on the Germanic languages³², leveraging a filtered dataset of 4,587 imputed ancient human genomes that temporally and geographically encompasses the regions in which early Celtic languages have been historically documented (Fig. 1d,e and Supplementary Table 1.1). In addition, we assembled an auxiliary dataset including individuals from Urnfield contexts in Iberia, Celtic-associated contexts in Britain and southwest Germany, and Bronze Age northern Italy (Supplementary Table 1.4). We used identity-by-descent (IBD) mixture modelling and qpAdm, as well as an updated version of chromopainter together with Sourcefind (Methods), to distinguish population continuity from directional gene flow and to test whether proposed linguistic dispersals require additional, unsupported demographic events.

Current evidence suggests that the Indo-European dialect ancestral to the Celtic languages entered Europe along with a ‘massive migration’ from the Pontic Steppe^16,17. In western Europe, the earliest Steppe ancestry appears in people of the Bell Beaker Complex³⁶, whose cultural package in northwestern Europe includes Maritime Beaker influence from Iberia, and All-Over-Corded and All-Over-Ornamented Beaker influence from the Corded Ware region to the east³⁷. Western and eastern contributions are also represented genetically, with Bell Beaker individuals deriving ancestry in various proportions from western Neolithic and eastern Steppe sources. To test whether their Steppe ancestry persisted in the four proposed formative regions for Proto-Celtic, we performed IBD mixture modelling (Methods). In brief, a palette is created for every ancient individual in the dataset, representing the proportion of IBD segments of DNA shared with all 386 clusters in the dataset. The palette of each ‘target’ individual is then modelled as the best-fitting combination of the palettes of the ‘source’ groups, using non-negative least squares in the R package limSolve (v.1.5.6)³⁸. Full IBD mixture modelling results, including source-set definitions, spatial and temporal distributions, alternative source models and sensitivity analyses, are provided in Supplementary Note 1, Supplementary Figs. 1.1–1.58 and Supplementary Tables 1.1–1.6.

As sources, we included groups of individuals from Bell Beaker-related and Corded Ware-related contexts, together with a series of earlier source groups (Supplementary Figs. 1.1 and 1.2 and Supplementary Note 1). For source individuals, we prioritized early Bell Beaker and Corded Ware individuals to measure the persistence of their ancestry through time. We also prioritized individuals with high Steppe ancestry and minimal Neolithic Farmer ancestries as sources, to allow more Farmer ancestry in later individuals to be detected. As we are interested specifically in the spread of Steppe ancestry, we refer to these ancestry sources as Bell Beaker-related and Corded Ware-related ancestry, while acknowledging that many Neolithic-related sources also contributed to Corded Ware and Bell Beaker populations^32,39. When including a single, early Steppe source (Yamnaya-related), we find Steppe ancestry appearing in eastern and northern Europe between 5000 bp and 4500 bp (Supplementary Fig. 1.31a), spreading to western and southern Europe between 4500 bp and 4000 bp, consistent with expectations^16,17,39. When including the two other sources with high Steppe ancestry (Bell Beaker-related and Corded Ware-related, Extended Data Fig. 2c), we find that Steppe ancestry of individuals in the Corded Ware region of northern and eastern Europe is now modelled with Corded Ware-related ancestry (Fig. 2, Supplementary Fig. 1.31b). In the Bell Beaker region of western, central and southern Europe between 4500 bp and 4000 bp, we find the Steppe ancestry of individuals modelled as Bell Beaker-related (Fig. 2), as previously demonstrated³². These ancestries remain dominant in the respective regions through to the Iron Age (Fig. 2). We note an influx of Corded Ware-related ancestry 4000–3600 bp in people of the Únětice Culture in central Europe (Extended Data Fig. 3), consistent with expectations⁴⁰. However, we find that Únětice people genetically contribute little to no ancestry to later people in the region, or to Celtic-speaking regions more broadly (Supplementary Note 1 and Supplementary Fig. 1.52). The absence of an influx of Corded Ware-related ancestry from northern Europe or further Yamnaya-related ancestry from eastern Europe, together with the persistence of Bell Beaker-related ancestry, is consistent with a Bell Beaker-mediated entry of the dialects ancestral to Indo-European languages spoken in the region during the Iron Age, including Celtic, and is thus key to evaluating all four models.

Next, we look to the region in which Steppe ancestry was incorporated during the formation of Bell Beaker-related populations. Across the Bell Beaker range, we find individuals with the highest proportions of Steppe ancestry to the northwest and lower proportions to the south, suggestive of a more northern origin (Supplementary Fig. 1.31a). When using the Bell Beaker-related source, we find that most individuals in Britain, Ireland, France and the Netherlands have higher proportions of Bell Beaker-related ancestry than in central Europe (Fig. 2 and Supplementary Fig. 1.41). This pattern remains even when using alternative Bell Beaker and Corded Ware sources from the Czech Republic (Supplementary Figs. 1.40 and 1.42). The expansion to Britain and Ireland from northwestern mainland Europe is also consistent with a previous study indicating that the Bell Beakers of the Lower Rhine formed through a mixture of Corded Ware and local Rhine–Meuse Neolithic ancestry, before spreading to Britain⁴¹. Our results indicate that this expansion also reached central and southern Europe. The linguistic identity of Bell Beaker groups cannot be directly inferred from genomic evidence. However, if Indo-European dialects were introduced to the region by this expansion, as previously proposed, it provides a potential vector for the Indo-European dialect ancestral to Celtic. Further sampling from northern France to Germany will help refine this northwestern region.

Over a millennium separates the presumed arrival of Indo-European language varieties in western Europe (roughly 4300 bp) from Iron Age Celtic-speaking populations (roughly 2500 bp) (Fig. 1a). Although northwestern Bell Beaker-related ancestry persisted, the extent to which the population structure within this region changed remains unclear. Evaluating the four models on Celtic origins therefore requires reconstructing the demographic processes that affected Bell Beaker-derived groups after roughly 4000 bp.

Detecting the movement of people during this time has so far proved difficult, due to the close genetic relation between all individuals postdating the arrival of Steppe ancestry. Here we apply two complementary approaches. First, we refine Farmer ancestry as a proxy for mobility by identifying the specific source region of European Farmer ancestry (set C3). This is informative about where the initial admixture occurred, allowing us to infer the region from which later movements spread. Second, we model the later movements directly, using Bronze Age sources from the southeastern and southwestern ends of the region in which Bell Beaker ancestry spread (set C4). The movement from a specific region should be reflected in both the Neolithic Farmer proxy and the Bronze Age source. Rather than attempting to pinpoint specific source populations, these proxies represent broader regional ancestries, allowing the likely spatiotemporal range of possible source populations to be inferred.

To identify variation in the specific sources of Farmer-related ancestry across the Bell Beaker region, we performed IBD mixture modelling (Supplementary Note 1). We included source individuals from IBD clusters associated with Neolithic Britain and Ireland (British–Irish Neo.-related), France and Iberia (Iberia–France Neo.-related), Italy (Italy Neo.-related) and the Czech Republic (Euro. Neo.-related), encompassing the proposed homelands for Celtic. In addition, we included a series of Neolithic-related groups, Bronze Age Anatolia (Anatolia BA-related), Polish Globular Amphora Culture (GAC-related) and Early Anatolian Farmers (E. Anatolia-related). The geographical distribution of these Neolithic-related clusters is plotted in Extended Data Fig. 2b. For the ancestries persisting after roughly 4000 bp (Supplementary Fig. 1.16), we performed spatiotemporal kriging to identify the regions in which these ancestries are found at the end of the Bell Beaker period. We find British–Irish Neo.-related ancestry in Britain and Ireland, Iberia–France Neo.-related ancestry in southern France and Iberia, Italy Neo.-related ancestry around the Adriatic and Anatolia BA-related ancestry in the Balkans (Fig. 3a and Extended Data Fig. 4). We confirmed the spatiotemporal results by plotting the raw results (Supplementary Fig. 1.32).

To model movements more directly, we included Bronze Age source individuals from around 4000 bp, a point in time that marks the end of the Bell Beaker period and predates almost all linguistic and archaeological estimates for the formation of Proto-Celtic (Extended Data Fig. 1). From southwestern Europe, we included individuals from Bronze Age southern France–northern Iberia (S. France–N. Iberia BA-related, Extended Data Fig. 2d), where Bell Beaker-related and the local Iberia–France Neo.-related ancestries are found (Figs. 2 and 3a). From southeastern Europe, where Bell Beaker-related, Italy Neo.-related and Anatolia BA-related ancestries are found (Figs. 2 and 3a), we included individuals from Bronze Age Hungary and Serbia (Hungary–Serbia BA-related, Extended Data Fig. 2d). On the basis of spatiotemporal kriging and plotting the raw mixture modelling proportions, we find these ancestries restricted to southwestern and southeastern Europe roughly 4000 bp (Fig. 4a and Extended Data Fig. 5).

Sampling is now relatively dense across the region of interest during much of the Bronze and Iron Age, with the exception of the cremation-dominated Urnfield Culture (3200–2800 bp). We therefore first focus on the well-sampled peripheries: England and the Czech Republic, where dense sampling continues throughout the Bronze and Iron Ages, and France, where Iron Age sampling has been greatly improved³². To visualize the changes through time, we plot the IBD mixture modelling proportions as timelines for England, France and the Czech Republic (Figs. 3b and 4b).

In England, the prominent Farmer ancestries change through time (Fig. 3b). Before 4000 bp, individuals tend to have a slightly higher proportion of local British–Irish Neo.-related ancestry relative to the other Neolithic-related ancestries. From 3800 bp to 3200 bp, Iberia–France Neo.-related ancestry exceeds the local Neolithic ancestry (Fig. 3b), consistent with a “large-scale migration” from the mainland^34,42. Between 3200 bp and 2800 bp, we see a shift, in which the proportion of Italy Neo.-related ancestry has increased to similar proportions to that of Iberia–France Neo.-related, coinciding with the appearance of Anatolia BA-related ancestry. Bronze Age sources (Fig. 4b) show S. France–N. Iberia BA-related ancestry appearing in England between 4000 bp and 3200 bp, consistent with the appearance of Iberia–France Neo.-related ancestry. Between 3200 bp and 2800 bp, we see Hungary–Serbia BA-related ancestry increasing, becoming widespread by at least 2600 bp (Fig. 4b and Extended Data Fig. 6).

In France, we see a similar transition (Fig. 3b). Before 3500 bp, more local Iberia–France Neo.-related ancestry than Italy Neo.-related ancestry tends to be present. After 2800 bp, the relative proportions have reversed, so Italy Neo.-related ancestry is the highest, accompanied by Anatolia BA-related ancestry. When using the Bronze Age sources (Fig. 4b), we see an increase in the Hungary–Serbia BA-related ancestry and a decrease in the local S. France–N. Iberia BA-related ancestry at corresponding times. However, because of the lack of samples from the Late Bronze Age, the time of this transition cannot be directly measured. Nevertheless, the increased proportions of Italy Neo.-related (set C3) and Hungary–Serbia BA-related (set C4) ancestry between 3200 bp and 2800 bp in Britain (Figs. 3, 4) and Iberia (Supplementary Fig. 1.35) suggests that this ancestry was present in France by this time. We also observe in France a subtle increase in East North Sea (ENS)-related ancestry between 2800–2470 bp and 2470–2000 bp, corresponding to the transition from Hallstatt to La Tène in France (Supplementary Fig. 1.46). ENS-related ancestry is found throughout the Bell Beaker region and is found in the highest proportions in the Netherlands throughout the Bronze and Iron Ages (Supplementary Fig. 1.10). Although limited sampling between the Netherlands, western Germany and northern France means that a more precise source region for this ancestry cannot be assessed, it should be noted that the transition occurs after the spread of Celtic under all models. Nevertheless, the subtle increase in ENS-related ancestry in France is consistent with the integration of more northern ancestry by the La Tène period, as suggested by previous studies³⁵.

Further east, in the Czech Republic, an increase in Italy Neo.-related and Anatolia BA-related ancestry is observed by roughly 3300 bp in Tumulus-associated and Knovíz-associated individuals (Fig. 3b and Extended Data Fig. 3). Because Italy Neo.-related ancestry is also present east of the Adriatic Sea (Fig. 3a), this population movement need not originate in Italy. Consistent with the influence from the southeast inferred from the Neolithic ancestries, we detect an increase in Hungary–Serbia BA-related ancestry between 3600 bp and 3300 bp when including the Bronze Age sources (Fig. 4b). Unlike England, the Czech Republic shows no increase in Iberia–France Neo.-related ancestry between 4000 bp and 2600 bp that would reflect a movement from western Europe (Fig. 3b). We do, however, note an increase in S. France–N. Iberia BA-related ancestry from roughly 3300 bp (Fig. 4b). Combined, these results allow France and Iberia to be excluded as potential source regions from which the influx of ancestry came into the Czech Republic, and instead point to a movement from a region in which S. France–N. Iberia BA-related ancestry is present yet Iberia–France Neo.-related ancestry is absent, for example Italy (Supplementary Figs. 1.10 and 1.11) or central Europe (sets C3 and C4, Supplementary Fig. 1.37).

We interpret the persistent northwestern Bell Beaker-related ancestry together with shifts in local ancestry as resulting from several admixture processes between Bell Beaker-related populations and local Farmers across Europe, who subsequently moved within the region. Relative to the Celtic source regions proposed under linguistic models, we see movements to Britain from within the Celtic-from-the-West and Celtic-from-the-Centre source regions from 3800 bp, and from within the Celtic-from-the-East source regions from roughly 3200 bp.

The timing and distribution of the westward spread of Hungary–Serbia BA-related ancestry corresponds closely to the expansion of the Urnfield Culture (Supplementary Note 6). However, few genomes between 3200 bp and 2800 bp exist due to the Urnfield-associated practice of cremation. Including Urnfield context individuals from Iberia⁴³ as IBD mixture modelling targets reveals an influx of Hungary–Serbia BA-related ancestry by 3150 bp (set C4, Supplementary Fig. 1.35), consistent with previous suggestions of an influx of ancestry from central Europe^43,44. In the Czech Republic, individuals associated with the Urnfield subgroup of the Knovíz Culture are also represented in the dataset. Consistent with the results from the Urnfield context in Iberia, we find that the Knovíz individuals are modelled with high Hungary–Serbia BA-related ancestry (Extended Data Fig. 3). To test whether the Knovíz individuals represent Urnfield ancestry more broadly, we included Knovíz individuals as a source (set C5). By 2800 bp, when sampling improves, Knovíz-related ancestry is modelled in high proportions across western, southern and eastern central Europe in former Urnfield regions, complemented by more local sources (Extended Data Figs. 7 and 8). We note that in this study, ‘Knovíz-related ancestry’ is used as a proxy for a broader group rather than representing people specifically of the Knovíz Culture.

In Iron Age England, the highest ancestry proportion tends to be modelled as Bell Beaker-related, followed by Knovíz-related and S. France–N. Iberia BA-related (Extended Data Fig. 7). By contrast, on the mainland, the proportion of ancestry modelled as Bell Beaker-related tends to be low or absent, with the Steppe ancestry in these individuals better modelled by the other Bronze Age sources, that is, Knovíz-related, S. France–N. Iberia BA-related and Hungary–Serbia BA-related. When including more Bronze Age sources from England (set C6) and France (set C7), the increase in Knovíz-related ancestry by roughly 2800 bp remains apparent (Supplementary Figs. 1.19 and 1.20). Thus, compared with England, the effect of Knovíz-related ancestry is particularly high in France, Germany and the Czech Republic (Extended Data Fig. 7). In Austria, we note considerable diversity at the eponymous Hallstatt archaeological site, with individuals modelled with particularly high proportions of either Knovíz-related or Hungary–Serbia BA-related ancestry, more similar to Hungary, Slovenia and Slovakia, where the Hungary–Serbia BA-related ancestry is modelled in high proportions (Extended Data Fig. 7). In Iberia, we see Knovíz-related ancestry becoming present in almost all individuals after 3200 bp (set C5, Supplementary Fig. 1.35). Although a number of individuals before this point are modelled with small proportions of Knovíz-related ancestry in set C5, they lack Hungary–Serbia BA-related ancestry (set C4, Supplementary Fig. 1.35) expected for movements from central Europe. The results here indicate gene flow from eastern central Europe occurring between 3200 bp and 2800 bp, coinciding with the expansion of the Urnfield Culture. We note that the genetic effects diminish the further west they spread. Finally, we note in the Czech Republic a subtle increase of S. France–N. Iberia BA-related ancestry roughly 2400 bp, aligning with small proportions of Iberia–France Neo.-related ancestry in set C3 (Supplementary Fig. 1.54), corresponding to the beginning of the La Tène period.

Identifying which populations were involved in the formation of the Knovíz-related ancestry profile remains difficult due to the lack of samples between 3700 bp and 3300 bp. Compared with the earlier Únětice individuals of the Czech Republic, we see that Knovíz-related individuals have experienced an influx of Italy Neo.-related ancestry (found around the Adriatic Sea roughly 4000 bp) and Hungary–Serbia BA-related ancestry (found in the Balkans roughly 4000 bp) (Extended Data Fig. 3), suggestive of an influence from the southeast. We also see a subtle increase in ENS-related ancestry, which is found across the entire Bell Beaker range but occurs in the highest proportions in the Netherlands, suggestive of influence from central or northwest Europe. As a result, the specific populations and time of formation remain unclear, as does the possibility of some degree of local genetic continuity. However, two signals are clear and relevant to this study. First, we do not see evidence of Iberia–France Neo.-related ancestry in the Czech Republic from roughly 3300 bp to 2800 bp (Fig. 3b). The absence of this ancestry excludes western Europe as a potential source in the formation of Knovíz-related ancestry, which is of relevance to the evaluation of the Celtic-from-the-West and Celtic-from-the-Centre models. Second, we find that the Steppe ancestry in Knovíz individuals, as well as those across the entire Iron Age Celtic-speaking range, is modelled primarily as Bell Beaker-related rather than Corded Ware-related or extra Yamnaya-related ancestry (Fig. 2). In this historical period, Corded Ware-related ancestry is associated with Germanic-speaking populations in northern Europe³², and further Yamnaya-related ancestry (Supplementary Fig. 1.31b) with Greek-speaking populations in the Aegean³³. Thus, regardless of the precise timing and populations involved in the formation of the Knovíz-related ancestry, its Steppe ancestry derives from Bell Beaker populations, suggesting that the dialect ancestral to Celtic, too, can ultimately be traced back to Bell Beaker groups. Improved sampling from Bronze Age northern Italy and central Europe will probably provide further insight into the formation of Knovíz-related ancestry (Supplementary Note 1), potentially also explaining the linguistic ties between the Celtic and Italic branches of the Indo-European language family (Supplementary Note 5).

We tested the robustness of these results using qpAdm, chromopainter and Y haplogroups. Full qpAdm results are provided in Supplementary Note 2, Supplementary Figs. 2.1 and 2.2 and Supplementary Tables 2.1–2.4, and full chromopainting results in Supplementary Note 3, Supplementary Figs. 3.1–3.7 and Supplementary Tables 3.1 and 3.2. Y chromosome analyses are provided in Supplementary Note 4. These independent analyses supported the main demographic events inferred through IBD mixture modelling: gene flow into Britain after the Bell Beaker period, and into western Europe from further east at the end of the Bronze Age.

Past population genetics is increasingly invoked to evaluate competing hypotheses in historical linguistics when language change is assumed to result from population movements and resulting admixtures^{11,12,16,17,32,33,34,35}. Extending approaches from traditional archaeolinguistic syntheses^45,46,47, the present study uses genetic evidence to assess which linguistic models require the fewest extra assumptions about population movement, timing and language persistence (Methods). We stress, however, that linguistic change is not exclusively driven by demographic factors and therefore does not necessarily leave behind a population genetic signature^48,49.

With the above in mind, we set out to evaluate the four main linguistic models, based on different interpretations of historical-linguistic and archaeological data, for how Celtic could have acquired its historically documented range across the European continent. These are the Bell Beaker model; the Celtic-from-the-West model; the Celtic-from-the-Centre model and the Celtic-from-the-East model. Across the dataset, we identify three demographic processes that intersect differently with these four linguistic models (Fig. 5a,b): movement 1, the initial transmission of Steppe ancestry through the Bell Beaker region (4400–4000 bp); movement 2, population movements to Britain (3800–3200 bp) and movement 3, the westward spread of Knovíz-related ancestry from eastern central Europe (3200–2800 bp). We note that movements 1–3 are extended processes, not discrete single events.

Movement 1 aligns with the linguistic Bell Beaker model, which attributes the formation and spread of Celtic to Bronze Age Bell Beaker populations (roughly 4750–4500 to 4000–3800 bp). Under this model, movement 1 would represent the arrival of Celtic to the regions from which movements 2 and 3 occurred, with the later movements potentially reflecting the spread of a branch of Celtic. In the time following movement 1, we do not see an influx of Corded Ware-related or Yamnaya-related ancestry in the Bell Beaker region, suggesting Indo-European dialects in the region are Bell Beaker-mediated. If Celtic or its ancestral dialect arrived later, it probably did so with limited demographic movement.

Movement 2 is consistent with the view that maritime connectivity of the Atlantic Bronze Age drove the formation of Celtic, as proposed by the Celtic-from-the-West model. It is furthermore geographically consistent with a hypothetical spread of Celtic from France (and possibly northern Italy), as claimed by the Celtic-from-the-Centre model. However, the latter specifically claims a spread after 3000 bp, whereas this movement occurred during the previous millennium, resulting in a chronological mismatch.

Movement 3 aligns with the Celtic-from-the-East model. This model draws on archaeological evidence indicating a westward expansion of populations deriving from the Urnfield Culture from eastern central Europe into France and northeastern Spain^50,51. We also note that, like movement 1, this movement affects all known Celtic-speaking areas of Europe. Archaeologically, the persistence of Knovíz-related ancestry into the Iron Age is consistent with the transition from the Late Urnfield to the Hallstatt Culture being seen as a reorganization of settlements rather than the arrival of a new population⁵² (Supplementary Note 6). In addition, under the Celtic-from-the-East model, the westward expansion of Hallstatt C (2750–2600 bp), from the Rhine into England and Ireland^53,54, then provides an extra potential vector for the spread of Celtic varieties, that is, Brittonic and Goidelic, to Britain.

Finally, all linguistic models converge on a Late Iron Age eastward spread of Gaulish-related dialects during the La Tène period (2400–2200 bp)⁵⁵, ultimately extending from France to northern Italy, central Europe and the Balkans. Consistent with previous studies³⁵, we indeed detect subtle genetic shifts (Supplementary Figs. 1.10 and 1.54) suggestive of a movement from the west during this period. It should be noted, however, that this genetic signal is more subtle than that of movements 1–3, and thus needs further confirmation.

For the four linguistic models, we identify relevant movements based on overlapping geographical ranges and assess their compatibility with the linguistic date estimates (Extended Data Table 1). When evaluated against the demographic movements identified here, the Bell Beaker model and the Celtic-from-the-East model require the fewest independent population dispersals and the least reliance on auxiliary assumptions to account for the historically attested distribution of Celtic. Whereas the Celtic-from-the-West model can be aligned with movement 2, it also requires the prolonged maintenance of Celtic linguistic identity across regions undergoing substantial subsequent admixture (movement 3).

To further assess the feasibility of these movements as vectors for the spread of Celtic, linguistic dating provides an extra constraint. The underlying linguistic date estimates and associated literature are summarized in Supplementary Note 5, Supplementary Fig. 5.1 and Supplementary Table 5.1. A plausible population vector for the spread of Celtic must fall after the initial Pre-Celtic linguistic divergence from other Indo-European branches, allowing time for the shared innovations that define Proto-Celtic to accumulate and before the subsequent diversification into the attested Celtic branches (Extended Data Fig. 1a). Most estimates for the divergence of Pre-Celtic overlap at 5000–3500 bp, with the highest density between 4500 bp and 3800 bp (Extended Data Fig. 1b). Estimates for the subsequent formation and fragmentation of Proto-Celtic fall mostly between 3600 bp and 2800 bp, with the highest overlap between 3200 bp and 3000 bp (Extended Data Fig. 1c). Finally, the divergence time of the Goidelic and Brittonic Celtic branches from each other has most estimates overlapping between 3500 bp and 2000 bp, with the highest overlap at 2800–2500 bp (Extended Data Fig. 1d).

Only movement 3 falls within the bounds provided by most quantitative estimates (the divergence of Pre-Celtic and the divergence into Goidelic and Brittonic), while also coinciding with the highest overlap of the qualitative estimates for the formation and fragmentation of Proto-Celtic (Fig. 5 and Extended Data Table 1), supporting the Celtic-from-the-East model. By contrast, movement 1 overlaps with many estimates for the divergence of Pre-Celtic, and both movements 1 and 2 predate most estimates for the formation and fragmentation of Proto-Celtic (Fig. 5 and Extended Data Table 1).

Given the present sampling density and available linguistic date estimates, we find the Celtic-from-the-East model to be the most parsimonious of the four predefined scenarios evaluated here (Methods). This preference does not rest on an assumption that linguistic spread must mirror demographic processes, but reflects that alternative models require further, at-present untested mechanisms such as prolonged linguistic continuity across major demographic turnover. At the same time, we note that human history does not necessarily follow the most parsimonious trajectory⁹, and the partial support for other models may result from a more complex divergence and spread, which cannot be excluded based on genomic data. If Bell Beaker groups with Steppe ancestry indeed transmitted Indo-European language varieties, it is likely that a continuum of related dialects, including those ancestral to Celtic, Italic and Lusitanian, were spoken across the region after 4400 bp. With this view, it is plausible that the people of movements 2 and 3 retained a degree of mutual intelligibility with the local population of Britain. Thus, the spread of Celtic may have involved the diffusion of Celtic linguistic features rather than a complete replacement of local languages: a complexity invisible to parsimony approaches. Although our findings do not definitively establish the homeland of the Celtic languages, they do constrain the demographic processes through which their formation and diffusion may be modelled. As a result, any revised or newly proposed models for the spread of the Celtic languages will need to reconcile with the demographic patterns identified in this study.

Dataset preparation

The 578 new ancient individuals included in the analyses here are part of a larger dataset of 712 ancient shotgun-sequenced genomes³². Details on the sampling, data generation, quality checks, site description and detailed metadata can be found in that study³². In brief, ancient samples were visually identified as suitable for processing. Samples were drilled, predigested⁵⁶, extracted¹⁷, built into libraries^57,58 and sequenced on Illumina HiSeq 4000 and NovaSeq 6000 platforms. Sequenced reads were mapped against the human reference genome build 37 (hs37d5) and authenticated based on the length of reads, damage patterns, endogenous rates and presence of contamination. All samples with a genomic coverage greater than 0.1× were imputed, and close relatives identified using ngsRelate (v.2)⁵⁹ were excluded for IBD-based analyses.

Genomes were imputed using Glimpse (v.1.1.1)⁶⁰ against the 1000 Genomes v.5 phase 3 dataset⁶¹. Shared IBD segments were called using IBDseq (v.r1206)⁶² with default parameters. Segments with a logarithm-of-the-odds (LOD) score less than three and hotspot regions³⁹ with excess sharing were excluded. A detailed description can be found in the accompanying paper³².

IBD mixture modelling

To identify fine-scale structure, we applied IBD mixture modelling³⁹. The use of non-negative least squares modelling has been routinely applied using chromopainter chunks shared between ancient source and target sets⁶³. However, recent studies have shown that using shared IBD segments between ancient individuals is effective at recovering fine-scale detail^32,39. We relied on the dataset and IBD clustering from the accompanying paper³². The cluster IDs for all individuals included can be found in Supplementary Table 1.1. The 4,587 ancient individuals in the dataset were determined to be suitable for IBD mixture modelling on the basis of a series of filtering steps for genomic coverage (0.1× for shotgun-sequenced data, 1× for 1,240,000 capture data), average genotype probability for the 1,240,000 capture sites (0.9) and the removal of close relatives. The final dataset included 690,211 sites that intersected with the 1000 Genomes panel and the 1,240 K sites, pass the 1000 Genomes mappability mask and had an imputation INFO score greater than 0.5.

For each of the 4,587 ancient individuals in the dataset, a palette was created representing the proportion of sharing with all other clusters, and with the other individuals in their own cluster. The proportions were normalized by the number of individuals in each cluster. To account for the fact that the sharing within the individual’s own cluster has n − 1 individuals, sharing with all other clusters excludes one individual at random. As a result, clusters with a single individual do not contribute to other individuals’ palettes. All segments greater than 1 cM, with a LOD score greater than 3, that did not originate from a hotspot of IBD sharing³⁹ were included in the palettes. IBD mixture modelling was then undertaken using these palettes, by modelling the target individual using non-negative least squares³⁸, as described in ref. ³⁹. Standard errors of the ancestry proportions were calculated using a weighted block jackknife, leaving out each chromosome.

A strength of this method is that target palettes are modelled as combinations of potential source palettes, meaning information from all other individuals in the dataset is incorporated into the modelling, rather than relying solely on the sharing of segments between the source and target individuals. When applied to densely sampled regions such as Holocene western Eurasia, sources and targets separated by 10,000 years continue to be modelled reliably³².

Spatiotemporal kriging

To understand the trends in ancestry through time, we performed spatiotemporal kriging (STAdmix, v.c72d845)⁶⁴ to interpolate IBD mixture modelling results for points in time and space where genetic data are not available. For the ancestry of particular relevance for set C2 (Corded Ware-related, Bell Beaker-related and European Farmer-related), set C3 (British–Irish Neo.-related, Iberia–France Neo.-related, Italy Neo.-related and Anatolia BA-related) and set C4 (S. France–N. Iberia BA-related and Hungary–Serbia BA-related), we fit spatiotemporal variograms by means of a metric covariance model⁶⁵.

qpAdm analyses

qpAdm analyses were undertaken on the 690,211 sites of the dataset described above, using admixtools2 (ref. ⁶⁶). Populations were placed into three sets: left_fixed, right_fixed or rotational (Supplementary Table 2.4). The ‘left_fixed’ population was included as a potential source in all qpAdm analyses. The ‘right_fixed’ population was included as outgroups in all analyses. In one population model, all ‘rotational’ populations were included as outgroups. For two population models, one ‘rotational’ population was placed together with the ‘left_fixed’ population as potential sources, and all others as outgroups, in all possible combinations. For three population models, two ‘rotational’ populations were placed together with the ‘left_fixed’ population as potential sources, and all others as outgroups, in all possible combinations.

Evaluating parsimony

In the present context, parsimony is understood as the minimization of independent population dispersals, chronological mismatches and auxiliary explanatory mechanisms (such as prolonged language persistence under repeated large-scale admixture) required to reconcile linguistic, archaeological and genomic evidence. Parsimony here functions as a comparative constraint on explanatory models rather than as an assumption about how prehistoric language spread must have occurred. Accordingly, parsimony is used here as a heuristic for model comparison rather than as a criterion of historical likelihood.

To evaluate whether a linguistic model is parsimonious with the genetic data, we assess whether a movement occurred in a direction and to a geographical extent consistent with the spread proposed by the linguistic model. In addition, we test whether the time of the movement is consistent with expectations based on linguistic dating. A plausible population vector for the spread of Celtic must fall after the initial Pre-Celtic linguistic divergence from other Indo-European branches, allowing time for the shared innovations that define Proto-Celtic to accumulate, and before the subsequent diversification into the attested Celtic branches.

The extent to which each model is consistent with the most relevant movement and further assumptions required to fit the model to the movement are detailed in Supplementary Table 1. The model that requires the fewest independent population movements, chronological mismatches and auxiliary assumptions is evaluated as most parsimonious.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Jørgensen, A. R. Celtic. In The Indo-European Language Family: A Phylogenetic Perspective (ed. Olander, T.) 135–151 (Cambridge Univ. Press, 2023).

2. Karl, R. & Stifter, D. The Celtic World: Critical Concepts in Historical Studies (Routledge, 2007).

3. Sims-Williams, P. The location of the Celts according to Hecataeus, Herodotus, and other Greek writers. Études Celt. 42, 7–32 (2016).

   Google Scholar 

4. Gibson, C. & Wodtko, D. S. The Background of the Celtic Languages: Theories from Archaeology and Linguistics (Univ. Wales, 2013).

5. Chadwick, N. K. The Celts (Penguin, 1970).

6. Powell, T. G. E. The Celts (Thames & Hudson, 1958).

7. Schmidt, K. H. History and culture of the Celts, draft plan of a comprehensive survey. In Geschichte und Kultur der Kelten (ed. Schmidt, K. H.) 14–24 (Carl Winter, 1986).

8. Schmidt, K. H. The Celts and the ethnogenesis of the Germanic people. Hist. Sprachforsch. 104, 129–152 (1991).

   Google Scholar 

9. Haber, M., Mezzavilla, M., Xue, Y. & Tyler-Smith, C. Ancient DNA and the rewriting of human history: be sparing with Occam’s razor. Genome Biol. 17, 1 (2016).

   Article  PubMed  PubMed Central  Google Scholar 

10. Stolarek, I. et al. Genetic history of east-central Europe in the first millennium ce. Genome Biol. 24, 173 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

11. Speidel, L. et al. High-resolution genomic history of early medieval Europe. Nature 637, 118–126 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

12. Gretzinger, J. et al. The Anglo-Saxon migration and the formation of the early English gene pool. Nature 610, 112–119 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

13. Antonio, M. L. et al. Stable population structure in Europe since the Iron Age, despite high mobility. eLife 13, e79714 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

14. MacAulay, D. The Celtic Languages (Cambridge Univ. Press, 1992).

15. Ball, M. J. & Muller, N. The Celtic Languages (Routledge, 2012).

16. Haak, W. et al. Massive migration from the steppe was a source for Indo-European languages in Europe. Nature 522, 207–211 (2015).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

17. Allentoft, M. E. et al. Population genomics of Bronze Age Eurasia. Nature 522, 167–172 (2015).

    Article  ADS  CAS  PubMed  Google Scholar 

18. Heggarty, P. et al. Language trees with sampled ancestors support a hybrid model for the origin of Indo-European languages. Science 381, eabg0818 (2023).

    Article  CAS  PubMed  Google Scholar 

19. Mallory, J. P. From the steppe to Ireland: the impact of aDNA research. In The Indo-European Puzzle Revisited (eds Kristiansen, K. et al.) 129–145 (Cambridge Univ. Press, 2023).

20. Brun, P. L’origine des Celtes. Communautés linguistiques et réseaux sociaux. In Celtes et Gaulois, l’Archéologie face à l’Histoire. 2 La Préhistoire des Celtes (ed. Vitali, D.) 29–44 (Collection Bibracte, 2006).

21. Gallay, A. Bell Beakers today. Pottery, people, culture, symbols in prehistoric Europe. In Proc. International Colloquium Riva del Garda 1998 (ed. Nicolis, F.) 41–57 (Provincia Autonoma di Trento, 2001).

22. Cunliffe, B. Setting the scene. In Exploring Celtic Origins (eds Cunliffe, B. & Koch, J. T.) 1–17 (Oxbow Books, 2018).

23. Cunliffe, B. W. Facing the Ocean: The Atlantic and its Peoples (Oxford Univ. Press, 2001).

24. Cunliffe, B. W. & Koch, J. T. Celtic from the West: Alternative Perspectives from Archaeology, Genetics, Language, and Literature (Oxbow Books, 2010).

25. Koch, J. T. Out of the flow and ebb of the European Bronze Age: heroes, Tartessos, and Celtic. In Celtic from the West 2: Rethinking the Bronze Age and the Arrival of Indo-European in Atlantic Europe (eds. Koch, J. T. & Cunliffe, B.) 101–146 (Oxbow Books, 2013).

26. Sims-Williams, P. An alternative to ‘Celtic from the East’ and ‘Celtic from the West’. Camb. Archaeol. J. 30, 511–529 (2020).

    Article  Google Scholar 

27. Schrijver, P. Sound change, the Italo-Celtic linguistic unity, and the Italian homeland of Celtic. In Celtic from the West 3 (eds Koch, J. T. & Cunliffe, B.) 489–502 (Oxbow, 2016).

28. Pope, R. Re-approaching Celts: origins, society, and social change. J. Archaeol. Res. 30, 1–67 (2022).

    Article  Google Scholar 

29. Childe, V. G. The Bronze Age (Cambridge Univ. Press, 1930).

30. de Hoz, J. Lepontic, Celtiberian, Gaulish and the archaeological evidence. Études Celt. 29, 223–240 (1992).

    Google Scholar 

31. Kristiansen, K. Europe Before History (Cambridge Univ. Press, 2000).

32. McColl, H. et al. Tracing the spread of Germanic languages using ancient genomics.Nature https://doi.org/10.1038/s41586-025-09064-1 (2026).

33. Yediay, F. E. et al. Ancient genomics support deep divergence between eastern and western Mediterranean Indo-European languages. Preprint at bioRxiv https://doi.org/10.1101/2024.12.02.626332 (2024).

34. Patterson, N. et al. Large-scale migration into Britain during the Middle to Late Bronze Age. Nature 601, 588–594 (2022).

    Article  ADS  CAS  PubMed  Google Scholar 

35. Gretzinger, J. et al. Evidence for dynastic succession among early Celtic elites in Central Europe. Nat. Hum. Behav. 8, 1467–1480 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

36. Olalde, I. et al. The Beaker phenomenon and the genomic transformation of northwest Europe. Nature 555, 190–196 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

37. Heyd, V. The mobility and migration revolution in 3rd millennium bc Europe. In Rethinking Migrations in Late Prehistoric Eurasia (eds Fernández-Götz, M. et al.) 41–62 (British Academy, 2022).

38. Soetaert, K., Van den Meersche, K. & van Oevelen, D. limSolve: Solving Linear Inverse Models. R package version 1.5.6 https://doi.org/10.32614/cran.package.limsolve (2008).

39. Allentoft, M. E. et al. Population genomics of post-glacial western Eurasia. Nature 625, 301–311 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

40. Papac, L. et al. Dynamic changes in genomic and social structures in third millennium bce central Europe. Sci. Adv. 7, eabi6941 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

41. Olalde, I. et al. Lasting Lower Rhine–Meuse forager ancestry shaped Bell Beaker expansion. Nature 652, 938–946 (2026).

42. Patterson, N. What genetics can say about Iron Age and Bronze Age Britain. In Presenting Counterpoints to the Dominant Terrestrial Narrative of European Prehistory (eds Koch, J. T. et al.) 187–190 (Oxbow Books, 2025).

43. Bretos Ezcurra, M. et al. Genomic insights from a final Bronze Age community buried in a collective tumulus in an Urnfield settlement in northeastern Iberia. Commun. Biol. 8, 1299 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

44. Olalde, I. et al. The genomic history of the Iberian Peninsula over the past 8000 years. Science 363, 1230–1234 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

45. Renfrew, C. Archaeology and Language: The Puzzle of Indo-European Origins (Jonathan Cape, 1987).

46. Mallory, J. P. In Search of the Indo-Europeans: Language, Archaeology and Myth (Thames & Hudson, 1989).

47. Anthony, D. W. The Horse, the Wheel, and Language: How Bronze-Age Riders from the Eurasian Steppes Shaped the Modern World (Princeton Univ. Press, 2007).

48. Campbell, L. Do languages and genes correlate? Lang. Dyn. Chang. 5, 202–226 (2015).

    Article  Google Scholar 

49. Barbieri, C. et al. A global analysis of matches and mismatches between human genetic and linguistic histories. Proc. Natl Acad. Sci. USA 119, e2122084119 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

50. Müller-Karpe, H. Beiträge zur Chronologie der Urnenfelderzeit nördlich und südlich der Alpen (de Gruyter, 1959).

51. Sørensen, M. L. S. & Rebay-Salisbury, K. A brief history of urns, urnfields, and burial in the Urnfield culture. In Death and the Body in Bronze Age Europe: From Inhumation to Cremation (eds Sørensen, M. L. S. & Rebay-Salisbury, K.) 15–35 (Cambridge Univ. Press, 2023).

52. Sienkewicz, T. J. Hallstatt Culture. Research Starters in Anthropology (EBSCO, 2022).

53. Gerloff, S. Hallstatt fascination: ‘Hallstatt’ buckets, swords and chapes from Britain and Ireland. In From Megaliths to Metals: Essays in Honour of George Eogan (eds Roche, H. et al.) 124–154 (Oxbow Books, 2004).

54. Fontijn, D. & Fokkens, H. The Emergence of Early Iron Age ‘Chieftains’ Graves’ in the Southern Netherlands: reconsidering transformations in burial and depositional practices. In The Earlier Iron Age in Britain and the Near Continent (eds Haselgrove, C. & Pope, R.) 354–374 (Oxbow Books, 2007).

55. Sims-Williams, P. Celtic. In The Indo-European Languages (eds Kapović, M. et al.) 352–386 (Routledge, 2017).

56. Damgaard, P. B. et al. Improving access to endogenous DNA in ancient bones and teeth. Sci. Rep. 5, 11184 (2015).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

57. Meyer, M. & Kircher, M. Illumina sequencing library preparation for highly multiplexed target capture and sequencing. Cold Spring Harb. Protoc. 2010, db.prot5448 (2010).

58. Gansauge, M.-T., Aximu-Petri, A., Nagel, S. & Meyer, M. Manual and automated preparation of single-stranded DNA libraries for the sequencing of DNA from ancient biological remains and other sources of highly degraded DNA. Nat. Protoc. 15, 2279–2300 (2020).

    Article  CAS  PubMed  Google Scholar 

59. Hanghøj, K., Moltke, I., Andersen, P. A., Manica, A. & Korneliussen, T. S. Fast and accurate relatedness estimation from high-throughput sequencing data in the presence of inbreeding. GigaScience 8, giz034 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

60. Rubinacci, S., Ribeiro, D. M., Hofmeister, R. J. & Delaneau, O. Efficient phasing and imputation of low-coverage sequencing data using large reference panels. Nat. Genet. 53, 120–126 (2021).

    Article  CAS  PubMed  Google Scholar 

61. The 1000 Genomes Project Consortium. A global reference for human genetic variation. Nature 526, 68–74 (2015).

    Article  Google Scholar 

62. Browning, B. L. & Browning, S. R. Detecting identity by descent and estimating genotype error rates in sequence data. Am. J. Hum. Genet. 93, 840–851 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

63. Lawson, D. J., Hellenthal, G., Myers, S. & Falush, D. Inference of population structure using dense haplotype data. PLoS Genet. 8, e1002453 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

64. Racimo, F. et al. The spatiotemporal spread of human migrations during the European Holocene. Proc. Natl Acad. Sci. USA 117, 8989–9000 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

65. Pebesma, E. & Graeler, B. gstat: spatial and spatio-temporal geostatistical modelling, prediction and simulation. R package version 2.1.2 https://doi.org/10.32614/cran.package.gstat (2003).

66. Maier, R. et al. On the limits of fitting complex models of population history to f-statistics. eLife 12, e85492 https://doi.org/10.7554/eLife.85492 (2023).

Download references

Author notes

1. These authors contributed equally: Hugh McColl, Guus Kroonen

Authors and Affiliations

1. The Lundbeck Foundation GeoGenetics Centre, Globe Institute, University of Copenhagen, Copenhagen, Denmark

   Hugh McColl, Thomaz Pinotti, William Barrie, Kristian Kristiansen, Martin Sikora & Eske Willerslev

2. Centre for Ancient Environmental Genomics, Globe Institute, University of Copenhagen, Copenhagen, Denmark

   Hugh McColl, Thomaz Pinotti, Kristian Kristiansen, Martin Sikora & Eske Willerslev

3. Department of Historical Studies, University of Gothenburg, Gothenburg, Sweden

   Hugh McColl, Johan Ling & Kristian Kristiansen

4. Leiden University Centre for Linguistics, Leiden University, Leiden, The Netherlands

   Guus Kroonen

5. Department of Nordic Studies and Linguistics, University of Copenhagen, Copenhagen, Denmark

   Guus Kroonen

6. Laboratório de Biodiversidade e Evolução Molecular (LBEM), Universidade Federal de Minas Gerais, Belo Horizonte, Brazil

   Thomaz Pinotti

7. Department of Genetics, University of Cambridge, Cambridge, UK

   William Barrie & Eske Willerslev

8. University of Wales Centre for Advanced Welsh and Celtic Studies, Aberystwyth, UK

   John Koch

9. Panthéon-Sorbonne University of Paris I, Paris, France

   Jean-Paul Demoule

10. Unité Mixte de Recherche 8215 Trajectoires, Institut national des sciences humaines et sociales, Centre national de la recherche scientifique (CNRS), Paris, France

    Jean-Paul Demoule

Contributions

E.W. initiated the study. H.M., G.K. and E.W. led the study. H.M., G.K., M.S. and E.W. conceptualized the study. K.K., M.S. and E.W. supervised the research. G.K., J.K., J.L., K.K., M.S. and E.W. acquired funding for research. H.M., T.P., W.B. and M.S. were involved in data analysis. H.M., G.K., J.K., J.L., K.K. and E.W. drafted the main text. H.M., G.K., T.P., W.B., J.K., J.L., K.K. and M.S. drafted supplementary notes and materials. H.M., G.K., T.P., W.B., J.K., J.L., J.-P.D., K.K., M.S. and E.W. were involved in reviewing and editing drafts.

Corresponding authors

Correspondence to Hugh McColl, Guus Kroonen, Kristian Kristiansen or Eske Willerslev.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks Rachel Pope, David Stifter and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Reporting Summary (download PDF )

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions