When the history of our ancestry is written, the fact that we’ve interbred with some of our closest relatives, the Neanderthals and Denisovans, will have a central role. And it will be tempting to write it as a very tidy story: Once we got genomes from these other groups, it was possible to identify the sequences in our genomes that we shared with them.
But in reality, as scientists started poking large enough collections of data, there were regular hints of some strange ancestry in our genomes. It was hard to pin down, though, at least in part because the 2 percent on average of Neanderthal DNA found in many populations does not guarantee that any two individuals will have the same 2 percent. So having the genomes of those two groups made sense of some things researchers had already been seeing.
But knowing what we do about Neanderthal and Denisovan DNA is now allowing researchers to answer a somewhat different question: Is there anything else? Using recently developed analytical techniques, they find evidence of a third lineage that we apparently interbred with before any modern humans left Africa. Again, there were hints of this earlier, but so far, there’s been no genome from a modern human relative to help us understand the details—the source of this DNA remains a “ghost lineage.”
Old, and yet young
The new work, done by a group largely based at Berkeley, relies on developments from elsewhere in the field of genomic analysis. Any site in a given person’s genome is the product of a mixture of common descent and random mutations, and its relationship to its neighbors can be mixed up by recombination, when pairs of chromosomes swap segments of DNA.
With enough genomic data, computers can be used to reconstruct what are called ancestral recombination graphs that try to reconstruct this history. For each base in the genome, ancestral recombination graphs estimate its history: How many generations back that particular base first appeared in the genome and when it has been involved with recombinations. Because of the randomness of some of these things and complexities like deletions, many of the individual inferences about history will be wrong. But those are likely to be the exceptions, and the average picture across the genome’s three billion bases should be informative.
The Berkeley team made a few inferences about what these ancestral recombination graphs should look like in cases where a separate lineage contributed DNA to modern humans (a process called “introgression”). One is that there should be a cluster of sequences that look consistently old, since they shouldn’t have as many of the same variants that the modern human genomes have picked up while the lineages were separate.
Normally, sequences that have been around for a while have more chances to be involved in a recombination. But these sequences were reintroduced to the human genome later in our history, so recombination should be far less frequent relative to a genome that’s been in the modern human lineage the whole time.
So any part of the genome that introgressed from a separate lineage should have two properties: Many of its bases should look “old” in the sense of how far back their common ancestry can be traced, yet they should look “young” in terms of how much recombination has taken place. So the researchers developed a software tool they call TRACE to look for these sequences.
Like a ghost
To test the tool, they had convenient examples: the Neanderthal and Denisovan sequences we’ve already identified in the human genome. If TRACE couldn’t pick those out, it wouldn’t find anything else useful. The development of ancestral recombination graphs is still a work in progress, so the researchers used two different tools to generate them and stuck with the one that produced the best results. The combination produced a very low false discovery rate (less than a quarter of a percent) while having an accuracy of over 90 percent.
In another demonstration, the researchers performed the analysis on African populations, which only received Neanderthal and Denisovan DNA when individuals from Eurasia migrated back. TRACE found only 0.1 percent of ancestry from these archaic lineages in these African genomes, consistent with its low false error rate.
Given about 500 modern human genomes, TRACE pulled out the expected Neanderthal and Denisovan segments. But it also revealed a lot of DNA from a ghost lineage. The percentage was small, at about 0.5 to 1.1 percent of current human genomes. But collectively, they covered nearly 1.5 billion bases. For context, the full human genome is three billion bases.
It was present in all modern human populations, indicating that the interbreeding occurred before the out-of-Africa expansion. But African populations carry more diverse segments and some that are distinct, a consequence of some of the diversity that was lost when only a fraction of the total population expanded into Eurasia. In fact, the researchers could identify about 100 areas of the genome that lack ghost lineage DNA entirely in non-African populations.
Based on the last common ancestor of the sequences, the researchers estimate that the lineage last shared a common ancestor with modern humans over 800,000 years ago, making it about the same age as the split with the ancestor of Neanderthals and Denisovans. The segments in the human genome are, on average, shorter than Neanderthal and Denisovan segments. This means they’ve been around in the modern human genome longer, which is consistent with them having arrived before modern humans left Africa.
The modern human genome has areas, called “deserts,” that lack Neanderthal and Denisovan DNA entirely. That has led to the suggestion that the human genome can’t tolerate dramatically different variants in these areas. But the ghost lineage DNA shows up in both Neanderthal and Denisovan deserts. So there seems to be something actively problematic with the DNA from those lineages in these regions.
Super Archaic
Something else suggested by other data is that the Denisovan population had also undergone a separate episode of interbreeding, this with a far older lineage. This “super archaic” lineage appears to have split from modern humans far earlier, suggesting it came from Homo erectus or another human ancestor. Because some populations—primarily in Southeast Asia and Oceania—have relatively high levels of Denisovan DNA, it’s possible that some of this superarchaic DNA found its way into our genomes.
The team went looking for it, searching for DNA that looked considerably older than the three known lineages. There’s not much—only about 0.3 percent of the Denisovan DNA in genomes from Oceania can be traced back to the superarchaic lineage. So although a tiny fraction of the genomes in some modern humans comes from this introgression, it’s definitely out there. And it apparently comes from a lineage that branched off from ours nearly 1.8 million years ago.
The work doesn’t really get into the significance of these archaic sequences. They’re less common in the neighborhood of genes, but some genes nearby are often involved in metabolism and immune function. It’s not clear whether these ancient variants are adaptive in any way, though. It’s also impossible to say much about the ghost lineage beyond the fact that our ancestors encountered it in Africa.
Still, we now have samples of a fair amount of its DNA. And there’s always a chance that ancient DNA or archeology could ultimately tell us a bit more about what these ghosts might have been.
Science, 2026. DOI: 10.1126/science.aef8874 ( About DOIs).
