Living people may carry small amounts of DNA from two human lineages that have never been matched genetically to a known fossil population, according to a new reconstruction of human ancestry. The result does not come from DNA extracted from a newly discovered bone. It comes from family trees hidden inside hundreds of modern genomes.

Yulin Zhang and colleagues at the University of California, Berkeley, describe their method, called TRACE, in a peer-reviewed paper in Science. The researchers analyzed 503 phased whole-genome sequences from five populations in the 1000 Genomes Project: 91 British, 103 Han Chinese, 102 Indian Telugu, 108 Yoruba and 99 Luhya genomes. They separately examined 92 high-coverage Oceanian genomes, including people from Papua New Guinea, Vanuatu and the Santa Cruz Islands.

The central claim needs a timing qualifier. A UC Berkeley account of the study estimates that the first unidentified lineage split from ours about 800,000 years ago. The final paper describes that split more cautiously as occurring more than 500,000 years ago. The dates come from a model of genetic divergence, not from directly dated fossils, so 800,000 years is an estimate rather than a fixed boundary.

This is one study, not settled consensus.

What TRACE reconstructs from living genomes

A chromosome does not have one unchanging family tree. Recombination shuffles DNA in every generation, leaving adjacent sections with different genealogies. Geneticists can represent that mosaic as an ancestral recombination graph, or ARG, which links local family trees across the genome.

The team first inferred ARGs with a program called SINGER. TRACE then searched them for a particular pairing: branches deep enough to reflect a population that had long been separated from the ancestors of the sampled people, attached to inherited segments long enough to suggest that the DNA reentered their ancestry through later interbreeding.

Unlike methods that begin with a Neanderthal or Denisovan reference genome, TRACE can flag an ancestry source for which no ancient genome is available. It also does not require an external outgroup to decide which sequence is ancestral. Those features make it useful for possible donor populations known only through the traces they left in living descendants.

In simulations where the true ancestral graph was known, TRACE reached 92 percent accuracy and 71 percent recall. When the graph had to be reconstructed from sequence data, as it must be with real genomes, the researchers added filters designed to keep the false-discovery rate below 10 percent. That validation shows that the method can recover a simulated signal under tested conditions. It cannot guarantee that the demographic scenario used to explain every real signal is unique.

A ghost lineage that mixed in Africa

Across the five 1000 Genomes populations, TRACE attributed an average of about 0.49 to 1.1 percent of each genome to an unidentified, deeply diverged ancestry source. The signal appeared in both African and non-African samples. The authors therefore place the interbreeding in Africa before the final major migration that contributed most of the ancestry of present-day people outside Africa.

The Berkeley release says the gene flow occurred before about 50,000 years ago and puts the population split at roughly 800,000 years ago. The paper’s broader description, more than 500,000 years ago, better reflects the uncertainty in the underlying inference. Mutation rates, assumed generation times, ancestral population sizes and the demographic model all influence a split-date estimate.

“Ghost” is a technical description, not the name of a species. Population geneticists use it for an ancestry source inferred from the DNA of descendants when there is no matching reference genome from the donor itself.

The researchers also found ghost ancestry in genomic regions considered deserts of known Neanderthal and Denisovan ancestry. That result argues against the first signal being only a mislabeled remnant of either well-sampled archaic group. It does not reveal whether the source was one stable population or several related populations with a complex history.

The older signal appears to have traveled through Denisovans

The second lineage appears one step further removed from living humans. In the 92 Oceanian genomes, unusually deep segments were enriched within regions already identified as Denisovan-derived. The same enrichment was not seen to a comparable degree in Neanderthal-derived regions. The authors interpret the asymmetry as evidence that a much older population mixed with Denisovans, who later carried a fraction of that DNA into modern human populations.

The reconstructed branch for this “super-archaic” source split from the lineage leading to living humans around 1.8 million years ago. The Berkeley account places its contact with Denisovans more than 200,000 years ago. These figures describe different events. The first is the separation of ancestral populations; the second is later gene flow between descendants of those populations.

Oceanian genomes are especially informative because they retain more Denisovan ancestry than most other living populations. In this sample, TRACE assigned averages of about 0.73 percent of the genome to Neanderthal ancestry, 0.66 percent to Denisovan ancestry and 0.33 percent to the unidentified deep source. The authors caution that TRACE underestimated known archaic ancestry in simulations, so those shares should not be treated as complete inventories.

Denisovan inheritance already has documented effects in living people, including the EPAS1 high-altitude variant common among Tibetans. The new analysis asks a different question: whether some DNA classified as Denisovan also contains a still older layer inherited by Denisovans before their encounters with modern humans.

Why neither signal has a fossil name

A genetic branch is not a species identification. It can indicate that two ancestral populations remained separated long enough to accumulate differences, and that their descendants later exchanged DNA. On its own, it cannot connect that ancestry to a particular skull, archaeological site or taxonomic label.

The researchers discuss Middle Pleistocene African hominins sometimes grouped under Homo heidelbergensis as one possibility for the first donor. They also note that a lineage related to Homo erectus could be compatible with the 1.8-million-year split. These are hypotheses based on timing and geography, not genetic matches.

That distinction matters because fossil classifications themselves are disputed. Similar-looking remains may not represent one interbreeding population, while genetically distinct populations can leave bones that are difficult to tell apart. Most fossils old enough to test the proposed identities have not preserved recoverable DNA, especially in warm regions.

The skull shown with this article is therefore a representative museum image, not a specimen linked to either inferred lineage.

Why the ancestry percentages remain model-dependent

The broad idea of unidentified archaic ancestry in Africa predates TRACE. In 2020, Arun Durvasula and Sriram Sankararaman reported in Science Advances that four West African populations could derive 2 to 19 percent of their ancestry from an archaic source. Their model placed that source population’s split from modern human ancestors between about 360,000 and 1.02 million years ago.

TRACE’s estimate of 0.49 to 1.1 percent is much smaller. It should not be read as a simple replacement measurement. The two studies use different data sets, demographic assumptions and signals, and they may not be counting precisely the same historical contribution. Long-standing population structure within Africa can also create patterns that resemble admixture from a separate archaic group.

The new study’s strongest checks are its ability to recover known Neanderthal and Denisovan ancestry and its performance on simulated histories. Its full methods manuscript explains the graph reconstruction, simulations and filtering in greater detail. Yet a method can reliably detect an unusual genealogy while leaving more than one plausible history for how that genealogy arose.

For that reason, the most durable result is not a precise species label or percentage. It is the evidence that parts of modern human ancestry may pass through branches deeper and more interconnected than the available ancient genomes can directly reveal.

What could test the reconstruction next

Broader sampling of living populations, particularly groups underrepresented in genomic databases, could test whether the first signal follows the geographic pattern expected from one pre-migration admixture event. More high-quality Denisovan genomes could show whether the older contribution is consistently nested within Denisovan ancestry or reached modern humans by additional routes.

A secure fossil identification would require a molecular bridge. DNA is unlikely to survive at many African and Asian sites relevant to events hundreds of thousands of years old, but ancient proteins may persist in some remains after DNA is lost. New fossils with preserved biomolecules could compare the inferred branches with named populations rather than relying on dates and geography alone.

Until then, the number of interbreeding events, where they occurred and which fossil populations took part remain open questions for the next genomic and paleontological tests.