When did the last woolly mammoths go extinct?

About four thousand years ago — far more recently than almost anyone expects.

Woolly mammoths vanished from mainland Eurasia and North America at the end of the last ice age, roughly ten thousand years ago, and for most of the twentieth century that was assumed to be the end of the species. It was not. In 1993, Sergey Vartanyan and colleagues published radiocarbon dates in Nature showing that mammoths had survived on Wrangel Island, off the coast of north-east Siberia, for another six millennia.

That places the last of them inside recorded human civilisation. The Great Pyramid of Giza was completed around 2560 BCE. The sarsen circle at Stonehenge was raised around 2500 BCE. The first cities of Sumer were already old. Mammoths were alive through all of it, and for roughly five hundred years after the pyramid was finished.

It is worth stating plainly once, because it is a genuinely useful fact rather than a trick: the mammoth is not a creature of unimaginable deep time. It overlapped with literate, city-building humans by longer than the gap between us and the Roman Empire.

The island that trapped them

Wrangel Island sits about 140 kilometres off north-east Siberia and covers roughly 7,600 square kilometres. It is treeless tundra — cold, open, and for most of the last ice age not an island at all, but high ground on the exposed continental shelf.

When sea level rose at the end of the Pleistocene, around ten thousand years ago, the sea closed behind whatever happened to be standing on it. Nothing migrated there. Nothing chose it. A slice of a mainland population was cut off and left with 7,600 square kilometres of grazing and no way out.

Islands do strange things to the animals stranded on them, and we have covered another case in detail: Homo floresiensis, the small-bodied hominin of Flores. Isolation tends to reduce body size, narrow diet and shrink genetic variety, and the usual expectation is that a marooned population is living on borrowed time.

Were they dwarf mammoths?

No — and this correction has followed the story for thirty years. Vartanyan’s 1993 paper described the animals as “Holocene dwarf mammoths,” and the word stuck. Later analysis showed they were only modestly smaller than late Pleistocene mainland woolly mammoths. They were not a true dwarf species in the way the Channel Islands pygmy mammoth was, and the phrase should be retired.

They descended from just eight animals

The 2024 genome study put a number on the founding population, and the number is startling.

Every mammoth that lived on Wrangel Island across six thousand years descended from at most eight individuals.

That is an extreme bottleneck by any measure. A population founded by eight animals carries a fraction of the variation of the group it came from, and every animal born afterwards is related to every other. Ordinarily this is the opening of a well-known and unhappy story: inbreeding depression, harmful recessive mutations surfacing, fertility falling, and a slide toward an extinction that was effectively decided at the start.

What the genomes show instead is that the population grew to between 200 and 300 animals within about twenty generations — and then held roughly that size for six thousand years.

The “genomic meltdown” story, and where it came from

The inbreeding explanation was not invented online. It came out of careful published work, and it is worth being precise about what that work did and did not establish.

In March 2017, Rebekah Rogers and Montgomery Slatkin published “Excess of genomic defects in a woolly mammoth on Wrangel island” in PLOS Genetics. They compared two mammoth genomes: a mainland animal about 45,000 years old, and a Wrangel animal about 4,300 years old. The island genome carried a striking load of damaging mutations — lost olfactory receptor genes, reduced urinary protein genes, and changes that would have produced a translucent, silky “satin” coat. They described the population as undergoing a genomic meltdown.

In February 2020, Erin Fry, Sun Kim, Vincent Lynch and colleagues went further in Genome Biology and Evolution. They synthesised the mutated Wrangel versions of several genes, expressed them in living cells, and tested whether the proteins still worked. Several did not — genes involved in neurological development, male fertility, insulin signalling and the sense of smell. Lynch’s summary was widely quoted: the last mammoths “may have been pretty sick and unable to smell the flowers.”

Both studies were good science done with what existed. But two genomes is two data points, and a functional test in a cell line tells you what a protein does, not what a population does. Neither study could show the shape of the change over time — whether the mutation load was climbing toward a collapse, or sitting at a level the animals were tolerating perfectly well.

That is exactly the question twenty-one genomes could answer.

What twenty-one genomes actually showed

On 27 June 2024, Marianne Dehasque and colleagues at the Centre for Palaeogenetics in Stockholm published a temporal study in Cell. Instead of comparing a before and an after, they sequenced 21 mammoth genomes — 14 from Wrangel and 7 from the mainland — spanning the last 50,000 years of the species. That gave them something nobody had held before: a time series.

Four findings matter.

Genetic diversity declined, but very slowly. Across six thousand years of isolation, variation fell at a rate consistent with a small but stable population, not one in free fall.

The worst mutations were purged. Severely harmful variants were removed by selection over time — which is what happens in a small population that is functioning, because the animals carrying the worst mutations do not reproduce. Mildly harmful mutations did accumulate, as expected.

The population was demographically stable. Two to three hundred animals, holding, for six millennia.

There was no decline before the end. The extinction was abrupt.

Senior author Love Dalén put the conclusion bluntly: “it was probably just some random event that killed them off, and if that random event hadn’t happened, then we would still have mammoths today.”

Why the earlier work was not wrong so much as outnumbered

It matters how this correction is described. Rogers and Slatkin had two genomes; Dehasque and colleagues had twenty-one, distributed through time. Fry and Lynch tested gene function in cells, which is a real result about proteins and not a claim about a population’s survival. Nobody was careless and nobody was fraudulent. This is a story about how science self-corrects when the sample gets bigger — which is the system working, not failing.

The same method has rewritten human prehistory more than once. A finger bone from a Siberian cave turned out to belong to a population nobody knew existed — the Denisovans — and a two-centimetre splinter from the same cave turned out to be a first-generation hybrid of two human species.

They were not starving either

The other comfortable explanation is that the island simply ran out of food as the climate warmed. The chemistry says otherwise.

In 2019, Laura Arppe and colleagues published an isotopic study in Quaternary Science Reviews, working with the Universities of Helsinki and Tübingen and the Russian Academy of Sciences. Carbon and nitrogen isotopes in mammoth collagen record diet and nutritional stress. On the mainland, late populations show a clear isotopic shift before they disappear — the signature of animals under dietary pressure.

The Wrangel mammoths show no such shift. Their values stay steady through the warming at the start of the Holocene and remain steady for thousands of years afterwards. They were eating well until the end.

The same study did find one thing that changed. Sulphur and strontium values indicate intensified weathering of the island’s bedrock toward the end of the record, which may have affected drinking water quality. It is a real signal, it is not a cause of death, and it is the only environmental change anyone has detected in the run-up.

So what killed the last mammoths?

The honest answer is that nobody knows. There are four candidates, and not one of them has direct evidence behind it.

An extreme weather event. The specific scenario is a rain-on-snow icing event — rain falling on snowpack and freezing into a sealed crust that grazing animals cannot break through. Events like this still kill Arctic herbivores in large numbers today. For a population of two to three hundred animals on a single island, one bad winter could be enough. There is no evidence that one occurred.

Drinking water. The weathering signal in the Arppe data could indicate deteriorating water quality. There is no evidence it reached lethal levels.

Disease. A pathogen arriving in a genetically narrow population with no exposure history is an obvious candidate, and it leaves essentially no trace in the record.

Humans. The earliest known human site on Wrangel — Chertov Ovrag, with stone and ivory tools including a toggle harpoon for hunting sea mammals — dates to roughly 3,700 years ago, a few centuries after the most recent mammoth bone. No butchered mammoth bone and no kill site has been found anywhere on the island. Absence of evidence is not evidence of absence, and the archaeological record of a remote Arctic island is thin. But on the evidence that exists, people arrived at an island the mammoths had already left.

What a solved island extinction looks like

Set this against a case where we do know the answer. On St. Paul Island in Alaska, mammoths died out about 5,600 years ago, and Graham and colleagues pinned both date and cause in PNAS in 2016 using three independent methods: dung fungal spores in lake sediment, sedimentary DNA, and radiocarbon dates on fourteen specimens. The island had shrunk to about 110 square kilometres. The mammoths stripped the vegetation around the freshwater lake, erosion filled it in, and they ran out of water. The authors note that the animals “contributed to their own demise.”

That is what a solved island extinction looks like. Wrangel is the best-studied extinction in prehistory and it has nothing of the kind.

Why this correction matters right now

This is not a dispute about a dead animal. The 2017 paper drew an explicit general lesson: preserving a small isolated group, it argued, “is not sufficient to stop negative effects of inbreeding and genomic meltdown.”

That sentence is a conservation argument. It says that once a population is small and cut off, the genetic damage is already done and the outcome is written — so resources might be better spent elsewhere.

The 2024 data says close to the opposite. Those mammoths were viable. Eight founders became a functioning population that purged its worst mutations and held steady for six thousand years. Whatever ended them came from outside, and Dalén’s point is that without it, mammoths might still be here.

For the many species now living in small, fenced, isolated remnants, that is the difference between don’t bother and keep them alive and keep the weather off them.

H2: The silence at the end

There is one thing the genomes cannot tell us, and it is the part that stays with you.

A stable population does not know it is about to end. For six thousand years there were mammoths on Wrangel Island — grazing, calving, dying of ordinary causes, replacing themselves. Then, within a span too short for the record to resolve, there were none. No decline, no dwindling, no long sad fade. An island with mammoths on it, and then the same island.

We have twenty-one genomes, an isotopic record, a radiocarbon chronology and some of the best-dated bones in prehistory — and we still cannot say what happened in that gap.

Frequently asked questions

When did mammoths go extinct? Mainland woolly mammoths disappeared around 10,000 years ago, but the species survived on Wrangel Island in the Arctic Ocean until about 4,000 years ago — several centuries after the Great Pyramid of Giza was completed.

Were the Wrangel mammoths dwarf mammoths? No. The 1993 paper that announced them used the word “dwarf” and it stuck, but later analysis showed they were only modestly smaller than late Pleistocene mainland woolly mammoths. They were not a true dwarf species like the Channel Islands pygmy mammoth.

Did inbreeding kill the mammoths? No. A 2024 study of 21 genomes in Cell found that the Wrangel population purged its most harmful mutations and stayed demographically stable for roughly 6,000 years, with no decline before the extinction. The earlier “genomic meltdown” conclusion rested on two genomes.

Did humans kill the last mammoths? There is no evidence that they did. The earliest known human site on Wrangel dates to a few hundred years after the most recent mammoth bone, and no butchered mammoth bone or kill site has been found. It remains possible the archaeological record is simply too thin to show it.

How many mammoths lived on Wrangel Island? Between about 200 and 300 at any one time, for roughly six thousand years, all descended from at most eight founding animals.

What is the most likely cause of their extinction? No cause has direct evidence. The candidates are an extreme weather event such as rain-on-snow icing, declining drinking-water quality, disease, and humans. Anyone stating one of these confidently is going beyond the evidence.