Ancient DNA across Europe: three ancestry components, 12,000 to 2,500 years ago
Three ancestry components in ancient genomes, from 12,000 to 2,500 years ago, in 50 seconds. Where there are few dated genomes, the colour fades out instead of being guessed.
What you are seeing
Each colour is one component of a statistical model of ancient genomes: hunter-gatherer-related (blue), Anatolian-farmer-related (green-yellow) and steppe-related (orange). A place's colour mixes them in proportion. These are components of a model, not peoples, nations or languages.
12,000 years ago the map is almost empty: only one patch, centred on the Iron Gates of the Danube, has enough dated genomes, and it is hunter-gatherer-related.
Anatolian-farmer-related ancestry first appears around the Sea of Marmara, mostly in northwest Anatolia, about 8,750 years ago in this model, reaches southeastern Europe about 8,000 and Britain about 6,000 years ago. Published dates: farming in northwest Anatolia by 8,550–8,450 years ago, an early farmer genome from Greece dated 8,388–8,214 years ago, and farmers in Britain from about 5,950–5,672 years ago.
An orange, steppe-related tint shows up in the east, and later in the north, long before the Bronze Age; in this model that is hunter-gatherer ancestry (see the limits below). In Europe west of the steppe, steppe-related ancestry arrives from the east between 5,000 and 4,500 years ago in this model (with the other thresholds tried that pass the same checks, the first date is 5,000–4,750 and the second 4,750–4,500 years ago). Published studies date its arrival in central Europe to about 4,500–4,450 years ago.
From about 4,500 years ago at least half of the lit land carries all three components, each at 15 % or more. The dataset then thins out: 252 genomes date from 4,500–4,000 years ago, only 7 from 3,000–2,500. The colour fades with them, and after 2,500 years ago there is no field at all.
How it was made
Ancestry proportions come from Racimo et al. 2020, who fitted an unsupervised admixture model (Ohana, four components) to ancient and present-day genomes; only ancient genomes are used here, 819 of them. The dates of 766 of them are refined with the Allen Ancient DNA Resource, and each of those is spread over its own date range. We interpolate the proportions ourselves, by kriging in space and time, into 49 maps, one every 250 years. Where too few genomes are near in space and time, the map is transparent. The terrain, the coast and the ice are the same reconstruction as the rest of the site; the land is drawn grey so that the three colours stay readable.
Limits of the model
- Three components are a simplification. The source model has a fourth, most common in North Africa and the Fertile Crescent; it is left out, and the 18 genomes that carry more than 25 % of it are not used.
- The components come from an unsupervised model, not from a formal test of named source populations; their names say in which ancient genomes each component is highest, nothing more.
- The steppe-related component is also carried by hunter-gatherers in eastern and northern Europe long before the Bronze Age; in the source model it stands in for eastern hunter-gatherer ancestry. That is why an orange tint shows in the east and the north early on.
- A smooth field cannot show sharp local boundaries or single movements of people; it shows the big picture where genomes are dense and nothing where they are not.
- Where and when a colour appears follows where and when dated genomes were found, so the model alone cannot separate the spread of an ancestry from the spread of the evidence for it. The directions it shows agree with the published studies listed below.
- Genomes are sparse before about 8,000 years ago (the source study itself discusses only the time after that), and after 2,500 years ago this dataset has only 1 ancient genome, not enough to draw the field.
- The map ends 2,500 years ago; it says nothing about present-day people.
- The dates are the model's; see the published dates above.
Known limitations recorded with the ancestry data (verbatim)
- the field is drawn only where enough independent genomes support it; the rest of the frame is deliberately transparent — on this bake 6.9 % of the grid on average, peaking at 22.2 %
- the youngest ancient genome in this corpus is younger than the oldest by far less than the window, so the most recent frames carry nothing — nothing after 2500 BP, 10 of 49 frames empty
- the acceptance check does not detect a latitude mirror about the frame centre: for a sample cloud nearly symmetric about it, the reflection is close to a self-map. It does reject a field built from the sampling geometry alone, which is the null that matters
- the parameter sweep and the leave-one-out check score the field WITHOUT the independent-source gate that the bake applies, so the support radius was chosen against a more generous coverage estimate than the shipped one
- two estimators were computed and they disagree most where the map is dark; the shipped one is the arithmetic mean of proportions
The interactive viewer is in private preview.
Video licence: CC BY-NC-SA 4.0, because it is derived from FABDEM. Share it with a link to this page; no commercial use.
Sources
- Ancestry proportions: Racimo, Woodbridge, Fyfe, Sikora, Sjögren, Kristiansen & Vander Linden 2020, PNAS, doi:10.1073/pnas.1920051117 (data: github.com/FerRacimo/STAdmix)
- Admixture model: Cheng, Mailund & Nielsen 2017, Bioinformatics, doi:10.1093/bioinformatics/btx098
- Dates: the Allen Ancient DNA Resource — Mallick et al. 2024, Scientific Data, doi:10.1038/s41597-024-03031-7; data: Mallick & Reich, Harvard Dataverse, v66.p1, doi:10.7910/DVN/FFIDCW (CC0 1.0); sample metadata via the Poseidon AADR Archive
- Steppe-related ancestry in central Europe: Haak et al. 2015, Nature, doi:10.1038/nature14317; Mathieson et al. 2018, Nature, doi:10.1038/nature25778
- Early farmers: Hofmanová et al. 2016, PNAS, doi:10.1073/pnas.1523951113; Brace et al. 2019, Nature Ecology & Evolution, doi:10.1038/s41559-019-0871-9
- Terrain and coast: FABDEM V1-2, GEBCO_2024, ICE-6G_C (VM5a) — see the first film's page
- Land fill where FABDEM has no data (Armenia, Azerbaijan and the adjoining area): Copernicus DEM GLO-90, doi:10.5270/ESA-c5d3d65; produced using Copernicus WorldDEM-90 © DLR e.V. 2010-2014 and © Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the European Union and ESA; all rights reserved