Celtic Did Not Start on the Atlantic Coast: 4,587 Ancient Genomes Point to Eastern Central Europe

Today Celtic survives only in a handful of languages — Irish, Scottish Gaelic, Welsh and Breton — confined to a thin strip of Europe’s Atlantic seaboard. Yet between three and two thousand years ago it was spoken across most of the continent, from the Iberian Peninsula and Gaul to Britain, the northern Alps and northern Italy.

How that vast range shrank to a coastal fringe is one of the oldest puzzles in historical linguistics. Four competing models have long coexisted: dispersal with Bell Beaker populations, a western Atlantic homeland, a central European homeland, and an eastern central European homeland. Earlier ancient DNA work covered only single regions — Britain, Iberia, southwestern Germany — and so could never compare the models within one dataset.

On 4 November 2026 Nature published online “Tracing the spread of Celtic languages using ancient genomics” (DOI 10.1038/s41586-026-11198-9, open access), led by the Globe Institute’s GeoGenetics Centre at the University of Copenhagen with partners at the universities of Gothenburg, Leiden and Cambridge. The team assembled genome-wide data from 4,587 ancient individuals of Bronze and Iron Age Europe: 712 newly sequenced and the rest drawn from 4,009 already published genomes. Hugh McColl is first author; corresponding authors include Guus Kroonen, Kristian Kristiansen, Martin Sikora and Eske Willerslev.

Methodologically the study rests on identity-by-descent (IBD) mixture modelling, supported by qpAdm tests of ancestry proportions, updated chromopainter and Sourcefind runs that separate population continuity from directional gene flow, and spatiotemporal kriging in STAdmix to project ancestry proportions onto a time–space grid. Proportions were estimated by non-negative least squares, with weighted block jackknife errors. On the linguistic side, the source regions and date ranges implied by each of the four models were compared directly with quantitative language-date estimates for Pre-Celtic divergence, Proto-Celtic formation and fragmentation, and the Goidelic–Brittonic split.

The first confirmed thread is steppe ancestry. Around 4,400–4,000 BP, steppe components tied to Bell Beaker populations entered the regions that later spoke Celtic and persisted into the Iron Age. Corded Ware-related ancestry stayed mostly in the north and east; a single pulse reached the Únětice area but was barely transmitted onward. This identifies Bell Beaker-related people as the more likely vector for Indo-European speech entering western Europe — but that step explains Indo-European as a whole, not the Celtic branch.

The second thread belongs to the later Middle Bronze Age. Around 3,800–3,200 BP, Iberia–France Neolithic ancestry rose sharply in England, pointing to a south-west European movement into Britain; France flipped at the same time, while the Czech Republic shows almost none of that signal. This links Britain to the European south-west, but still does not locate the Celtic homeland.

It is the third thread that turns the answer eastward. The team identified ancestry associated with Knoviz, a branch of the eastern central European Urnfield culture, which formed in the Carpathian Basin around 4,000–3,200 BP. That ancestry expanded westward between 3,200 and 2,800 BP into France, Britain, north-eastern Iberia and Italy, persisted into the Hallstatt culture of France, Germany and Austria, and continued into La Tène. In dates: Britain by 2,800 BP, Iberia by 2,500 BP.

Set against the linguistic timeline, the conclusion is sharp. Only the third movement falls simultaneously inside most of the quantitative boundaries and onto the dense region of Proto-Celtic estimates. For the main component of all attested Celtic languages, eastern central Europe is therefore the more parsimonious centre of spread than western Europe or the Atlantic coast — a correction to the long-standing image of Celtic as an Atlantic-native speech.

The paper is equally candid about its limits. First, the authors write plainly that history need not be parsimonious; the simplest model is not the historical fact. Second, late Bronze Age samples from France — precisely the critical stretch of the third movement — remain scarce, so the sampling gap is still a break in the chain. Third, ancient DNA tracks the movement of people and the proportions of ancestry; it cannot prove that a given population spoke a given language. Language attribution requires archaeology, linguistics and genetics to converge, and no single line of evidence will do.

A companion study in the same issue of Nature applies the same framework to the spread of Germanic languages. Together the two papers place the language–people–gene triangle under test across the full time and space of Bronze and Iron Age Europe.

For lineage research in East Asia the framework is portable. The spread of Sino-Tibetan, Kra-Dai, Hmong-Mien and Austronesian languages can be examined with the same tools: which language expansions travelled with large-scale population movement, and which were carried by small groups whose speech was then adopted by a local majority. The East Asian population history that EAGGA follows needs exactly this caution — treating hypotheses about language, archaeological dates and genomic evidence side by side, never equating genes with language, and never reading a genealogical record as a biological fact.

The photograph shows Lake Hallstatt in the Salzkammergut region of Upper Austria. The Hallstatt culture (roughly 850–450 BCE) takes its name from this site and is one of the Iron Age cultures into which the eastern central European ancestry tracked in this study expanded. Image: Melancong.my