Tollense: Who Fought Europe's Oldest Battle? The DNA of the Bronze Age Warriors

In 1996 an amateur archaeologist walking the banks of the Tollense, a small river in Mecklenburg-Western Pomerania, noticed a human arm bone sticking out of the riverbank. Lodged in it was a flint arrowhead. Nearly thirty years of excavation later, that bone has become the founding clue of the largest Bronze Age battlefield ever found: more than 140 people, most of them young men, killed in what looks like a single violent episode about 3,250 years ago. No text records it. No king claims it. The only witnesses are the bones, the bronze, and the DNA.

So who were the combatants? Local farmers defending a river crossing? A raiding army from the south? Two coalitions from across Central Europe? Archaeology, isotopes and genetics have each given a partial answer, and they do not fully agree. In this article we walk through the evidence and then run our own Global25 analysis of the Tollense warriors, with every caveat on the table.

A battlefield no one expected

The Tollense valley is a flat, marshy corridor running north toward the Baltic. Excavations led by the State Authority for Culture and Heritage of Mecklenburg-Western Pomerania and the University of Greifswald have exposed only a small fraction of the site, yet the density of remains is astonishing. Most individuals are men between roughly 20 and 40. Many skeletons show fresh, unhealed injuries: arrow wounds, blunt force trauma to the skull, cuts from blades. Horse bones lie among the human remains, and two wooden clubs (one of ash, one shaped like a hammer) were preserved in the waterlogged sediment. The original study by Jantzen and colleagues set out the case in 2011: this was not a cemetery, it was a killing ground.

The location matters. The battle zone sits beside a wooden and stone causeway across the river, first built centuries earlier and apparently maintained into the battle period. A river crossing on a north to south route is exactly the kind of place worth fighting over, or worth ambushing a column on its way through.

Extrapolating from the excavated area, the excavators estimate that the dead may number many hundreds and the combatants several thousand. For northern Europe in the 13th century BC, a region with no cities and no states in the usual sense, that is a mobilisation on a scale nobody thought possible before Tollense.

The objects point south

Two recent studies shifted the debate toward long distance armies. In 2019, Uhlig and colleagues published a small cluster of 31 bronze objects found together on the battlefield: a knife, an awl, a chisel, scrap metal and three bronze cylinders that once fitted an organic container. Such boxes and kits have their best parallels in burials of southern Central Europe. The authors read it as the personal equipment of one warrior who had travelled from the south, lost or abandoned in the fighting.

Then in 2024 Inselmann and colleagues examined 64 arrowheads from the site (10 of flint, 54 of bronze). Some types are known locally. But several bronze types (their 2A, 4B and 4C) are absent from the northern region and match only southern parallels, above all in Bavaria and Moravia. Their conclusion: the battle probably pitted local groups against at least one incoming group from southern Central Europe, a scenario also supported by southern style dress pins and Bohemian type weapons from the site.

If the arrowheads are right, some of the dead should carry the genetic signature of Bohemia, Bavaria or the Middle Danube. That is a testable prediction, and it is the one we set out to test.

What the isotopes and genomes said

Isotopes came first. Price and colleagues analysed strontium, oxygen and carbon in the teeth and bones of Tollense individuals and concluded that at least some of them grew up in places different from the valley itself. Isotopes, however, can say that a person was not raised locally; they cannot easily say where they were raised instead.

The genomes arrived in 2020. Burger and colleagues generated data for 14 Tollense individuals using a targeted capture approach (a few megabases of neutral regions plus hundreds of trait associated loci, at a mean depth above 4x). Their headline finding was about milk: the lactase persistence allele was rare among the warriors, at an estimated frequency of about 7 percent, showing that the ability to digest milk as an adult was still uncommon in Bronze Age Europe and has been under strong selection ever since. On ancestry, they concluded that the Tollense individuals "conform to a sample from a single, unstructured population", close to Central and Northern Europeans, with no close relatives among them and two women in the sample.

That conclusion was challenged almost immediately. Davidski, the author of the Global25 dataset, argued on his blog that the warriors scatter visibly on principal component plots and do not look like a single homogeneous group. The Moriopoulos 2026 collection, which we use below, splits the Tollense samples into two main groups defined by Y chromosome lineage (an I2a group and an R1b group, five individuals each) plus three single individuals flagged as outliers, including WEZ56, a man carrying R1a.

So which is it? A single population, or a battlefield of strangers?

Our analysis: first, measure the noise

Before reading anything into outliers, we need to know how noisy these samples are. The Tollense genomes are low coverage in Global25 terms: the individually dated samples available to us sit at roughly 8 to 18 percent coverage of the Global25 SNP set. Low coverage samples scatter around their true position, and a single scattered sample can easily look like a foreigner.

To set an honest floor, we took every ancient population in the dated Global25 file that has at least five high coverage members (above 40 percent), computed the population mean from those, and then measured how far the population's own low coverage members (7 to 11 percent) fall from it. Across 176 such genomes, the median drift is 0.050, the 90th percentile 0.064 and the 95th percentile 0.076. In other words, a single sample at Tollense coverage can sit 0.05 away from its true population by chance alone, and one in ten will sit beyond 0.064.

This yardstick changes the story considerably:

• The I2a group and the R1b group are 0.040 apart. Each is an average of five noisy samples, so random scatter alone should put two such averages about 0.03 apart. The groups differ slightly more than chance would suggest, and in a consistent direction (see below), but the gap is small. Burger and colleagues' "unstructured" verdict is not crazy for the core of the sample.
• The female outlier sits 0.048 from the nearest group, squarely inside the noise band. On this evidence she should not be called an outlier at all.
• WEZ56, the R1a man, sits 0.077 away, just beyond the 95th percentile. Borderline, but probably real.
• The R1b outlier sits 0.093 away, well beyond anything coverage noise produces in our reference set. This is the clearest genetic stranger on the battlefield.

Where the warriors fall on the map of Bronze Age Europe

We built a principal component analysis from 23 Bronze Age populations spanning Scandinavia, the Baltic, Poland, Germany, Bohemia, the Alps and the Carpathian Basin, then projected the Tollense units onto it without letting them shape the axes. The first component (76 percent of the variance among these references) runs from the north-east (Baltic and Trzciniec populations, rich in steppe and hunter-gatherer ancestry) to the south (Carpathian and southern German populations, richer in early farmer ancestry). It is, roughly, a map.

Three things stand out.

The core of the Tollense dead sits in the middle of the map, beside the Lusatian Urnfield samples from Silesia, not beside the Nordic Bronze Age populations of Denmark and Norway, and not among the Bohemian and Bavarian groups either. The closest ancient populations to the R1b group are Lusatian Urnfield (Nowa Cerekwia, distance 0.027), Hungarian Urnfield (0.029), Bohemian Knoviz Urnfield (0.033) and the Urnfield population of the Unstrut valley in central Germany (0.033). The I2a group's nearest neighbours are the same Lusatian samples (0.029) and Trzciniec culture individuals from southern Poland who also carry I2a (0.029). In plain terms: the typical warrior looks like someone from the southern Baltic plain and the Oder basin, the heartland of the Urnfield world east of the Elbe.

WEZ56 is pulled to the north-east, toward the Trzciniec populations of Poland and western Ukraine (distances around 0.060) and, further out, the Baltic Bronze Age. Given his R1a lineage, that fits a man with roots somewhere east of the Oder.

The R1b outlier is pulled hard to the south, past the Bavarian and Bohemian groups, landing next to the Urnfield people of Neckarsulm in Württemberg and the Middle Bronze Age Vatya culture of Hungary. His nearest ancient matches are all southern: Unetice outliers from Lower Austria, Bronze Age central Italy, Late Bronze Age Slovenia, the Urnfield culture at Neckarsulm. If any sampled warrior came up the long road from the south, it is this man.

Deep ancestry: a north to south gradient

A three way model using Yamnaya steppe herders, Anatolian farmers and western hunter-gatherers is crude but robust (condition number 2.4, poles far apart). Its value here is that it places every group on the same scale.

The references line up almost perfectly by geography: Baltic and Polish Trzciniec populations at the top with the least farmer ancestry, Bohemia and Hungary at the bottom with the most. The Tollense core groups sit in the middle, close to the Lusatians. The R1b group carries slightly more farmer ancestry than the I2a group (41 against 36 percent), which is the direction of the small difference we measured earlier. The R1b outlier, at 60 percent farmer ancestry, is indistinguishable in this model from Middle Bronze Age Hungary. WEZ56 sits with Trzciniec.

One honest caution. The Tollense groups show 19 to 23 percent western hunter-gatherer ancestry, higher than most comparable references. Some of this may be real (the Lusatian and Trzciniec samples also run high), but coverage noise inflates fit distances and can shift small components, so we do not build any argument on it.

Proximal models, and a revealing zero

Finally, we modelled each Tollense unit with four Bronze Age sources representing the four directions an army could have come from: the Nordic Bronze Age of Denmark (north), the Baltic Bronze Age of Latvia (north-east), the Urnfield culture of Bohemia (south) and the Vatya culture of Hungary (south-east).

The most striking result is a zero. The Nordic Bronze Age source is rejected almost completely, even though the Tollense valley lies at the southern edge of the Nordic Bronze Age cultural zone, less than 200 km from the Danish islands. Because a zero can simply mean the method cannot see a component, we ran a spike-in test: we added known amounts of Danish Nordic Bronze Age ancestry into the R1b group and re-ran the model. A 2 percent spike is invisible, a 5 percent spike is recovered as 2.5 percent, and 8 percent comes back as 5.6 percent. The honest detection floor is therefore about 5 percent. Swapping in other Nordic proxies (Late Bronze Age Denmark, Middle Bronze Age Denmark, Bronze Age Sweden) gives between 0 and 13 percent, never a major share. For the I2a group the model assigns 7 percent to Denmark, but the spike-in shows this group already sits at that level of background, so the true figure is not distinguishable from a small value.

We also checked the reverse question for the core groups: could they be carrying hidden southern Carpathian ancestry? Vatya scores zero for both. Spiking 5 percent of Vatya into either group returns 3.3 to 3.5 percent, so a Carpathian contribution larger than about 5 percent would have been visible. The southern signal is concentrated in the outliers, not spread thinly through the core.

These proximal models also confirm the picture from the PCA: the R1b outlier resolves as Vatya plus Knoviz with nothing from the north, WEZ56 is mainly Baltic, and the female outlier gives a mixed and poorly fitting answer, consistent with the noise she carries.

Their closest living relatives

Against modern Global25 population averages, the R1b group is closest to eastern Germans (0.034), then Germans from Hamburg and Erlangen, Austrians and Danes. The I2a group is closest to Swedes, Germans from Hamburg, Poles, Slovaks, eastern Germans, Czechs and the Sorbs of Lower Lusatia (all between 0.047 and 0.050). The distances are larger for the I2a group, partly because it is noisier, but the direction is clear: the core Tollense population left descendants, or at least close genetic cousins, in the same broad region, as Burger and colleagues also concluded.

So who fought at Tollense?

Putting the archaeology and the genetics side by side, a coherent picture emerges, with firm limits.

Most of the sampled dead were regional people. The two core groups fall within the genetic range of the Late Bronze Age populations between the Elbe and the Vistula, closest to Lusatian Urnfield samples and to the northern Urnfield world. They are neither Danish looking Nordic Bronze Age people nor Bohemian or Bavarian southerners. The slight difference between the I2a and R1b groups could reflect recruitment from neighbouring communities, but it is too small to call them two peoples.

At least one man looks genuinely southern. The R1b outlier's profile, with around 60 percent early farmer ancestry, matches the Middle Danube and southern German Urnfield populations, and it lies beyond the scatter that low coverage can create. That is exactly the kind of person the southern arrowheads and the southern warrior's kit predict. It is one individual, at modest coverage, so it is a consistent clue rather than a proof.

At least one man looks north-eastern. WEZ56 carries R1a and a Trzciniec or Baltic leaning profile. The battle at the crossing may have drawn in people from east of the Oder as well.

The famous "outlier" woman is probably not one. Her distance from the core falls inside ordinary coverage noise.

Two limits deserve emphasis. First, genetics cannot tell us who fought on which side. A local man could have marched with the southern army, and a southerner could have settled in the north a generation earlier. Second, 13 or 14 sampled individuals out of more than 140 recovered, and perhaps many hundreds killed, is a small window. If an incoming southern force was largely made up of Bohemian or Bavarian Urnfield people, they are strikingly underrepresented among the dead sampled so far, with at most one clear candidate. Either the southern contingent was small, or it mostly survived, or its dead lie in parts of the valley not yet excavated or sequenced. The next batch of genomes from Tollense could settle that question.

A footnote on milk

The finding that made headlines in 2020 deserves a final word. Only about 7 percent of the warriors' lactase alleles were the persistence variant that lets most northern Europeans drink milk today. These were cattle herders who, as adults, could not comfortably digest fresh milk. The rise of that allele to its present frequency across northern Europe happened mostly after Tollense, one of the fastest known cases of natural selection in human history.

Global25 coordinates used in this article

Tollense groupings from the Moriopoulos 2026 collection (Global25 by Davidski). Paste directly into the Vahaduo tool.

Germany_BA_Tollense_(I2a_Group)_(n=5),0.124523,0.133034,0.080327,0.065181,0.058288,0.017291,0.009119,0.012230,0.015135,-0.000984,-0.004222,-0.008603,0.006987,-0.005973,-0.002199,0.002811,0.004955,0.003801,0.007517,0.011831,0.000823,0.003240,-0.004881,-0.015954,0.006945
Germany_BA_Tollense_(R1b_Group)_(n=5),0.123384,0.145627,0.070974,0.052520,0.043762,0.008925,0.008319,0.003738,0.017507,0.006561,-0.009029,0.002877,-0.008236,0.003551,0.003339,0.001644,-0.000339,0.002128,0.001885,0.004702,0.004542,-0.001830,-0.004338,-0.011158,-0.001006
Germany_BA_Tollense_(R1b_Group)_o_(n=1),0.133173,0.154360,0.053174,0.004522,0.060627,0.002510,-0.004700,-0.011538,0.013703,0.042279,0.004222,0.012139,-0.026908,-0.013900,0.010043,-0.034606,0.010431,-0.017610,-0.005656,-0.004752,-0.006613,-0.002473,-0.018857,0.012893,0.007544
Germany_BA_Tollense_(R1a_Outlier)_(n=1),0.147970,0.145221,0.085606,0.088502,0.034468,0.035419,0.015981,0.013846,-0.006545,-0.008930,-0.009094,-0.004046,-0.002230,0.035782,0.003936,0.004110,0.022426,0.000127,-0.011187,0.007003,-0.010981,-0.003215,0.012695,-0.018557,-0.005149
Germany_BA_Tollense_(Female_Outlier)_(n=1),0.127482,0.152329,0.061848,0.034884,0.048317,0.001394,-0.012456,0.002077,0.020248,0.019317,-0.000162,-0.007943,-0.002676,0.022983,0.001357,-0.001724,-0.009779,0.008615,-0.006285,-0.000625,0.007736,-0.004451,-0.008751,-0.027353,-0.004910

Methods in brief

Global25 scaled coordinates were merged with first occurrence priority from Davidski's population averages, modern averages, the Moriopoulos 2026 collection (all averages, no sims) and the dated individual ancients file. Distances are Euclidean over 25 dimensions. NNLS models use a sum to one constraint enforced by a heavily weighted extra equation, with results normalised. Condition numbers and pole separations are reported with each model. Spike-in tests add a known fraction of a source to the target and re-fit to determine the detection floor. The noise yardstick uses populations with at least five members above 40 percent coverage and measures the drift of their 7 to 11 percent coverage members from the high coverage mean. G25 results are a proxy; where published qpAdm or f-statistic results exist, they take precedence.

References

1. Genomics Burger J, Link V, Blöcher J, et al. (2020). Low prevalence of lactase persistence in Bronze Age Europe indicates ongoing strong selection over the last 3,000 years. Current Biology 30(21): 4307 to 4315. doi.org/10.1016/j.cub.2020.08.033
2. Archaeology Jantzen D, Brinker U, Orschiedt J, et al. (2011). A Bronze Age battlefield? Weapons and trauma in the Tollense Valley, north-eastern Germany. Antiquity 85(328): 417 to 433. doi.org/10.1017/S0003598X00067843
3. Isotopes Price TD, Frei R, Brinker U, Lidke G, Terberger T, Frei KM, Jantzen D (2019). Multi-isotope proveniencing of human remains from a Bronze Age battlefield in the Tollense Valley in northeast Germany. Archaeological and Anthropological Sciences 11: 33 to 49. doi.org/10.1007/s12520-017-0529-y
4. Archaeology Uhlig T, Krüger J, Lidke G, et al. (2019). Lost in combat? A scrap metal find from the Bronze Age battlefield site at Tollense. Antiquity 93(371): 1211 to 1230. doi.org/10.15184/aqy.2019.137
5. Archaeology Inselmann L, Krüger J, Schopper F, Rahmstorf L, Terberger T (2024). Warriors from the south? Arrowheads from the Tollense Valley and Central Europe. Antiquity 98(401): 1252 to 1270. doi.org/10.15184/aqy.2024.140
6. Commentary Davidski (2020). Warriors from at least two different populations fought in the Tollense Valley battle. Eurogenes Blog. eurogenes.blogspot.com
7. Data Global25 PCA coordinates (Davidski) and the Moriopoulos 2026 collection of Global25 averages.