Hello everybody. So, it's a real pleasure to be here in Copenhagen. The last time I was here was 10 years ago when everything was completely different. So I'm going to tell you about the work in ancient DNA that I've been privileged to be involved in for the last now 16 or actually even 19 years since this field really exploded beginning in 2010. So what happened in 2010 is that it first began possible to sequence DNA at a whole organism level, the whole genome level from ancient human remains and remains of other animals.
And it opened up worlds that we simply couldn't see before. So a measure of a new scientific instrument is that when you turn it to look at something that simply could not be seen before, it reveals things that are surprising and then challenge our previous understanding of how things are. So an example of this might be the invention of the microscope around 400 years ago. And when people like Anthony van Leeuwenhoek used it to look at cells, to look at pond water, would see microorganisms that were simply not imagined before. And in the same way, when ancient DNA is obtained from people who lived in the past, we can see how they're related to people living today or to each other and we learn things that we simply didn't even know about before.
So ancient DNA, the work starts with a human remain. So this is a photograph of a petrous bone which in around 2015 was discovered to have much more DNA than other parts of the body, maybe a hundred times more DNA on average. It's the inner ear. It carries part of the skull that has the inner ear and is a preferred source of DNA for extraction. And then in a clean room like the one we have at our institution, and one of the epicenters of this work is actually in Copenhagen as well, we get beneath the surface of these bones that might be contaminated by the people handling it. So the archaeologists who dig up the bones or the lab technicians who handle the bones.
And a whole variety of measures have been taken in order to protect this tiny old sample that might be thousands or tens of thousands of years old from contamination by the people who handle it. And this was evolved over time. So the rooms now have positive pressure to protect the samples from air coming in and being sucked in from outside where there might be a lot of DNA. And since they're so little, even a small amount can contaminate them. People wear these clean suits to prevent their own DNA from contaminating the sample. People wear gloves and double gloves again to prevent the DNA from being contaminated. People bleach surfaces with chemicals that will break up the DNA and make it impossible to sequence. People will take ultraviolet light and decontaminate the surfaces when people aren't in the room or put the human remain inside ultraviolet light to decontaminate the surface.
This combination of things has over time evolved a magic combination that people use in all of these ancient DNA laboratories to extract DNA from remains and it works. So we don't know which of these are important. Probably some of them are unimportant, but this combination of measures we take is now a very effective set of measures. And then we obtain a little bit of material that's particularly DNA rich. So this is the cochlea, so it's the organ of hearing within the skull, the spiral organ of hearing, and we take a very small amount, maybe 40 milligrams, maybe a 25th of the weight of a paperclip, and then we extract the DNA in an optimized solution that removes the protein and mineral content.
So it used to be the case that we didn't know how to do this as a community, but especially a group working in Germany, Matthias Meyer figured out a series of optimizations that have improved the extraction of DNA from these powders and those methods have spread around the world for doing that. It involves being able to retain the very tiny fragments that DNA is left in after it's been in a bone for thousands or tens of thousands of years. DNA sequence normally is in huge fragments of millions, tens of millions, hundreds of millions of DNA letters long. But by the time we get to it, it's only typically 20 or 30 or 40 or 50 DNA letters long. Tiny fragments. Is that all we can obtain? And you need an extraction methodology and a sequencing methodology that can pick up these tiny fragments.
And that's all been optimized. Also you actually need to run it through a chemistry that will remove some of the inhibiting substances that prevent the DNA reactions from going forward. So, it's this combination of things that has produced this miracle that means that now if we obtain DNA from say Denmark from 5,000 years ago, 60 or 70% of these samples will regularly obtain data that's as high quality as what you get almost in many ways when you send your spit sample to a personal ancestry testing company. It's really a miracle that this happens. When you try to extract DNA from an Egyptian mummy which looks very well preserved, it doesn't preserve well. It's a miracle about bone. The hydroxyapatite mineral in bone happens to be a binder for ancient DNA and it's preserved there.
And then we sequence it in one of these next generation sequencers that first came online in the late 2000s. So before this, when I started doing my scientific work, we used a different technology for sequencing that was much more expensive and specifically targeted a specific molecule, a specific section of the genome, one section at a time. This new sequencing technology that came online in the 2000s has dropped the cost of sequencing by about one millionfold very rapidly. And it made it possible to take these DNA samples that previously were very hard to sequence because the genome is huge. It's three billion DNA letters long. It's broken up into fragments, tens of bases in size. Most of the sequences you get are not from a human who had them but from the microbes who colonized the body when the person died. And because of that it was prohibitively expensive to sequence.
So there were interesting ancient DNA studies. People would target particular sections of the genome like the mitochondrial DNA a person gets from their mother and she gets from her mother and might sequence for example. Svante Pääbo led a team that sequenced the Neanderthal mitochondrial genome in 1997. But that was so difficult to do. It required very long sequence fragments. And what happened in the late 2000s, it became possible to use one of these new sequencers and just brute force sequence what comes out with this much less expensive technology.
So the talk that I'm giving is not about the deep integration of ancient DNA into archaeology, which is one of the most exciting things that's happening right now, and into biology and environmental sciences, which are so exciting and some of the things that are happening now. That's not what this talk is about, even though it could be what this talk is about. And the thing that I want you to come away with is how this particular technology is a type of new data that can falsify previous assumptions and can change our understanding of the world. And this is not even specific to ancient DNA. It's this argument and this thinking about when you have a new scientific instrument that makes it possible to probe something that was never probed before. If it's really a powerful new scientific instrument it will reveal horizons, vistas and features of the world that we didn't understand before, like a microscope and a telescope.
And I hope that by the end of this talk, if you don't already know as much about this field, that you'll understand that this is an exceedingly surprising type of data. In my own work I've come to this work with many, many biases which have again and again been falsified by the type of data that we happen to look at, so that at this point when we collect new data I'm very unconfident about what we'll find and mistrustful of my own biases. And so the first part of my talk is about how before ancient DNA we had a fundamentally wrong understanding of how modern humans relate to archaic humans.
So modern humans first appear in the skeletal record a few hundred thousand years ago. The oldest skeletons that are morphologically, in terms of their shape, thought to be modern humans are known from Ethiopia and North Africa two to 300,000 years ago. But also at the same time there were archaic humans living simultaneously. If you go back 70,000 years ago and you were able to put yourself in a time capsule and go into that time, we know that there's at least five groups at that time who are much more different from each other than anybody living today. 20,000 years later or maybe 30,000 years later, there was just one group which is the primary ancestors of everybody in this room today, anatomically modern humans. And those other four groups just disappeared.
Those other groups, we could think about them as the Denisovans, the Neanderthals, the northern Denisovans, the southern Denisovans, the Neanderthals, and also the Flores humans, and there were probably others, especially in Africa, maybe multiple others.
So in 2007, what happened and how I got involved in this field is Svante Pääbo, working in Germany, started getting genome-scale data from Neanderthals using these magical new sequencers and increasingly optimized technologies. And I got to be involved in this because at the time I was an expert in studying medical genetics and population mixture, which was obviously one of the questions one would want to try to analyze if one had data from archaic humans. So I got a lucky break. I was a fourth year assistant professor newly starting at Harvard and I was studying prostate cancer. And so what I was doing was looking in African-Americans.
African-Americans are a population of mixed ancestry in the last few hundred years with the primary ancestry for most people being West African due to the history of enslavement and slavery in the United States and the minority ancestry typically about 20% being European ancestry, sometimes a little Native American ancestry. And at the time it was super expensive to study genetic variation. It was thousands of US dollars to study a whole genome even at a few hundred positions in the genome to see whether people had one type of DNA or another type of DNA. And so it was very expensive to sequence the whole genome of anybody, a European or African. It was prohibitively expensive at the time.
And so we wanted to actually see were there risk factors in African-Americans that can contribute to the increased risk for this disease in African-Americans that was observed maybe one and a half or two fold more than European-Americans. And no environmental factors had been discovered. So people suspected it might be genetic. And so what we did is we had this hypothesis that maybe there's a genetic factor that occurs more commonly in people of African ancestry than in people of European ancestry. And we could find it by screening through the genome of African-Americans and painting the genome according to where in the genome African-American men with prostate cancer get their African and European ancestry from, whether from African and European ancestors, summing it up over everybody.
And on average, it should be 80% everywhere if people have 80% African ancestry. But our idea was that at places where African-American men who have prostate cancer, where the risk factor is, it will go up to 85 or 90%. Because given that you're an African man who has prostate cancer, probably you got African ancestry at that position. So we needed to develop a method to identify where these segments are and this method worked. And so we were able to identify such a segment. It was highly statistically unusual and it contains a risk factor near a known cancer risk gene that has seven independent genetic factors all more common in Africans than in Europeans and that together entirely explain the increased risk for prostate cancer in African-Americans.
I had also recently done work studying the ancestors of humans and chimpanzees and what we had shown by comparing human and chimpanzee genomes to the gorilla genome. So humans and chimpanzees are closest cousins and then the gorillas a bit more distantly related. Humans and chimpanzees we think separated 5 to 7 million years ago and gorillas maybe 2 million years before. By comparing them to each other we were able to show very clear evidence in my view that human and chimpanzee lineages initially separated and separated for a very long time and then hybridized and then split again and that either humans or chimpanzees descend from that event.
So I was someone who had studied population mixture in various ways and that was a skill that was needed for this analysis of the Neanderthal genome to try to understand how it was related. So Svante invited me to be part of an international consortium which contained all these amazing people, experts in different aspects that would be relevant to understanding this archaic genome and to learn about their relationships to modern humans was the thing that I was supposed to be involved in and was most interested in doing.
So to me it was clear that this was the most exciting data in the world in any field. Maybe I was wrong but to me I thought like this is where I want to go. This is what I wanted to do. And for the next seven years I spent more than half my time analyzing this data. I basically became a postdoc again working with the team in Germany. I went there 20 times or something like this. And it was my main focus for the next seven years until 2014 or so.
So the background to working on this was that the work that had happened in human genetics up to this point had consistently found evidence that modern humans outside of Africa, the genetic variation in non-Africans is a subset of that in Africans. And I for example linearly was part of this tradition. So my academic grandfather, my PhD supervisor's supervisor was Luca Cavalli-Sforza, who was part of the group of people who had demonstrated that modern human variation outside of Africa is a subset of that in Africa.
The data looked as if there had been no mixture from archaic humans who we know from archaeology like Neanderthals lived outside of Africa into non-Africans because it was a subset of that in Africans. Here's a paper from 2004 which analyzed the mitochondrial DNA available then. Mitochondrial DNA is this short DNA sequence maybe 16,000 DNA letters long so much smaller than the whole genome which is three billion base pairs long and it had been sequenced in Neanderthals and had been sequenced in many, many thousands of modern humans and we knew what the Neanderthal sequence was and it was an outgroup to all modern humans. There was no evidence of any interbreeding in modern humans.
And based on the patterns and mathematical models of what would happen if there was a spread of an invading population, modern humans, into a landscape full of archaic humans like Neanderthals, these authors argued that even if there was a small amount of interbreeding, modern humans would have been flooded by Neanderthal mitochondrial DNA and we would see it, but we don't see it. So they argued there had not been substantial interbreeding. That was one of several ways of looking at the data that made this clear.
So analyzing the Neanderthal data once we began to have substantial data, we got an incredibly strong signal that this was not true. And this is one of multiple lines of evidence that our consortium developed to try to probe this question. And since I was coming at this from a very strong bias that there was no mixture, I kept trying to make the results go away. So we got this result initially a couple of years before we published the paper and we just kept trying to look at the data. Maybe it's a bias and I'll show you the simplest line of evidence for mixture.
So what you see here is a cartoon of a chromosome. So your DNA is packaged into 47 packets. So 23 pairs of chromosomes 1 through 22, your sex chromosomes X and Y, and also the 47th packet which is your mitochondrial DNA. And what you see here is one of those packets, say your chromosome 3. And what we're looking at are the differences between different genome sequences. Your DNA, your genome is about three billion DNA letters long. The great majority of genome sequences at their positions are almost all identical. So if you line up the copy of chromosome 3 you get from your mom and the copy of chromosome 3 you get from your dad, they're 99.9% identical. There's only one in a thousand differences.
But since the genome is very long, about three billion base pairs long, that's about three million differences. That's a lot of differences. And you could use it to learn about history. So these differences arise through mutations, random changes that occur due to miscopying over time. And if there are more mutations separating your mother's and father's DNA, they descend from an ancestor longer ago where more mutations have had time to accumulate. So that's how you can learn about history through DNA.
So what we did is we looked through the genome sequence of a French person and through a Nigerian person from Africa and we looked at differences. So there's about three million of them. We looked at a few hundred thousand of them that we had high quality data on and we identified for example at this position the French person has a thymine, one of the four DNA letters. The Nigerian person has the guanine and we looked at which position was the newly arising mutation since separation from the primate ancestor. We used chimpanzees and gorillas to determine that.
And we then lifted the veil on the Neanderthal sequence and we said, "What does the Neanderthal have? Does it have the T or the G?" And if it's true that there's no interbreeding and West Africans and Europeans descend from a common ancestral population that separated earlier from Neanderthals, the matching rate should be exactly equal within statistical resolution. And you can ask that question, but instead there were about 92,000 matches to French people and about 84,000 matches to sub-Saharan Africans. This is a very significant excess statistically and there's no way this can occur by chance.
So this was true whether we used a Nigerian person or any other sub-Saharan African like a Khoisan person from southern Africa or a rainforest hunter-gatherer from central Africa. It was true whether we used any non-African, for example, a Han Chinese person or a Papuan or a Native American. It was a consistent difference between Africans and non-Africans that could only be explained by non-Africans being closer to Neanderthals than Africans are.
And we were able to show using other lines of evidence that the only way that this could happen is through interbreeding from Neanderthals into modern humans. One way to see this of many is that if you look at the fragments of DNA that are showing closeness between Neanderthals and non-Africans, they are separated if you count their mutations within less than 100,000 years, which is far too recent to be from some very distant population. It must be flow from Neanderthals into modern humans.
The next thing we did is we replaced the French person in this analysis by a second Neanderthal we had some data from and we saw how much bigger the skew got. The skew got 50 times bigger and that's because a French person is a 20th, 50th of the way to a Neanderthal and the correct interpretation of that is a French person has about 2% of their DNA deriving from Neanderthals.
And so that was like an amazing finding, very surprising, but also amazing was the fact that there's actually as much Neanderthal ancestry in East Asians and also in Papuans from New Guinea and also in Native Americans as in Europeans. Neanderthals are a European group. They lived only in Europe and in some parts of central Asia but not in South Asia, not in East Asia, not in New Guinea. And yet all these other groups have as much Neanderthal ancestry and actually probably even more than Europeans who live in the homeland of the Neanderthals. So this too is how ancient DNA reveals a surprise.
The history seems to be that Neanderthals interbred with the ancestors of all modern humans outside of Africa and that somewhere perhaps in the Middle East and that mixed population then spread to all these different places. And while we know from other data that has been collected since that there was interbreeding in Europe, those groups were themselves replaced by the later movement from the Middle East that carried this ancestry.
So the next thing we did that I wanted to highlight was we looked at where in the genome the Neanderthal ancestry occurs. We can do this because it's actually a very characteristic signature in the genome of when Neanderthal bits occur for example in an East Asian person's genome. There's a few features of the data that are very characteristic. So for example, these mutations that are characteristic of Neanderthal DNA in non-Africans don't occur in sub-Saharan Africa because the interbreeding didn't happen. So you look for genetic variants that are absent in sub-Saharan Africa.
You also look for variants that are about 40,000 DNA letters long. Why? Well, that's because the mixture happened maybe 1,700 generations ago we now know. And when mixture happens between Neanderthals and modern humans, the first generation the offspring has one entire chromosome 3 from Neanderthals, from say their mom, one entire chromosome 3 from modern humans, say from their dad. And then meiosis, recombination, as you make sperm or eggs, splices up your genome, one or two cuts per generation, like a salad chopper. And every generation, it splices at one or two cuts. We know the rate.
And you just ask the question, how small should those segments be after 1,700 generations? It's about 40,000 base pairs. So, we look for islands of mutations that are absent in Africans that are about 40,000 DNA letters long and have the correct split time between Neanderthals and modern humans to be associated with an archaic human. And we identify about half the DNA that came in from Neanderthals. Others don't quite match that pattern and are missed by this dragnet.
When you apply this to a genome sequence of a non-African person, this is the result of such an analysis. You get a scan across a chromosome, for example, your chromosome 3. And there's places where the statistical methodology looking for these 40,000 base pair fragments becomes very confident that a Neanderthal sequence is there. And you can validate using multiple lines of evidence that most of these are correct and really there.
And so the next thing you can do after doing this analysis on many non-African people, for example, 100 Europeans, is you can pile up your Europeans and you could say how many people out of the hundred people we're looking at have Neanderthal DNA at that position. So you can measure that. So here above the line is the proportion of DNA from Neanderthals in Europeans and below the line is in East Asians. And for example at this position this is a place where 40% of East Asian DNA is from Neanderthals. So instead of the average of 2%, 40% of copies you see today are from Neanderthals.
There's also places that are deserts that have no Neanderthal ancestry. Some of them are quite big and these are places that are potentially places of negative selection where when Neanderthals and modern humans met and mixed there was intense natural selection to remove Neanderthal ancestry that perhaps was not advantageous in the modern human background. Maybe it was some genetic change that is helpful for adaptation or maybe it's some genetic change that contributed to infertility in the offspring. And so we see a landscape of Neanderthal ancestry where there are places of more and there are places of less. And that provides information about biology.
One of the most interesting patterns is that chromosome X, which is one of the two sex chromosomes and is the larger sex chromosome. Women have two copies, men have one copy, is depauperate. It has very few Neanderthal segments, much, much, much less than the rest of the genome. Another part of the genome that has much, much less Neanderthal are genes that are involved in sperm meiosis. And when you do the same analysis to brain-related genes or to liver-related genes, you don't see this particular deficiency.
So this characteristic combination of genes involved in sperm biology and genes on the X chromosome is a pattern seen in many, many species, not just humans, when there is a hybridization. So when you take two populations and split them for hundreds of thousands of years or the appropriate amount of time depending on the biology and generation time of the species and then put them back together, if they're far enough apart, what will happen is that the hybrids will start being infertile. Specifically, the males will start being infertile and the females will be fertile for much longer until their separation is far enough.
And the genetic changes that are responsible for this are concentrated, we know for many, many species for a variety of biological reasons, on the X chromosome and on the mitochondrial DNA and on sperm-related genes. So if you see today the result of a hybridization and there's a clearance of one of the two ancestries from the X chromosome and from sperm genes, it's telling you that the hybrid males in the mixed population had reduced fertility and there was rapid natural selection to remove those incompatibilities.
So we know that at the time Neanderthals and modern humans mixed they were likely far enough apart that there was substantial male infertility. And those male infertilities were resolved after a small number of generations of natural selection and the hybrid population produced many of the people living in this room today as their descendants. So you can see this from the pattern in the DNA.
What happened also in 2009 is a group of scientists and archaeologists working in Siberia in Denisova Cave in south-central Siberia came across a tiny, teeny bone which is a pinky bone of what we now know as a little girl. And this individual was identified by the Russian archaeologists as likely to come from a Neanderthal or a modern human. They didn't expect that to be anything else. But when Svante Pääbo's team sequenced the DNA, it turned out to be another thing altogether. There was no expectation that it would be.
But it actually turns out to be from a group that split from Neanderthals 500,000 years ago, a half million years ago, far more different than any groups living today. And that shared ancestral population of Neanderthals and Denisovans split from the ancestors of modern humans more than 700,000 years ago. So this was another of these crazy results that comes from ancient DNA. The archaeologists didn't expect to see another group like this at this point but the genetics forced our hand.
It's normal with Neanderthals, for which were discovered in the 1850s in Germany in the Neanderthal valley. We knew they were interesting. We had studied all their stone tools. We had studied their culture. We had many archaeological digs trying to characterize them and think about what they might have been like. So that was a case of a culture, a fossil in search of a genome. But with the Denisovan it was a genome in search of a fossil because all we had is this uninformative little pinky bone that was identified by excellent Russian archaeologists who are able to identify bones but nothing like a tooth or a skull or something like this.
So the next thing we did is we tried to estimate and understand how Denisovans are related to modern humans using the same toolkit we had developed for studying the relationship of Neanderthals to modern humans. And we didn't expect to see anything but we tried. And here's the most interesting statistic. When you compare the Denisovan to Chinese on the one hand and to New Guineans on the other and you do the same trick, you find that there's a much higher matching rate to New Guineans, people from this island which is a tropical island a little bit north of Australia, very, very far south from Siberia.
And this is the only group that we analyzed that had a very strong signal like this. So this is a crazy result because it meant that Denisovans interbred with the ancestors of New Guineans. So that's a crazy result, but not Chinese people to a substantial extent, not South Asians to a substantial extent, not Europeans, not Siberians.
The next thing we did is we estimated the amount of Denisovan ancestry in diverse people around the world using these methods. And in the pie charts that you see here it's an estimate of the proportion of Denisovan ancestry where what you see in New Guinea and Australia is the maximum 3 to 5% and the pie chart shows the fraction of that in different groups. And you see throughout South Asia and East Asia there is about a twentieth of that in New Guinea and Australia. In Europe there's almost none, in Africa there's almost none. It's a very different pattern from what is seen in Neanderthal DNA where the pattern is throughout non-Africans.
And so there seems to have been an interbreeding event or multiple interbreeding events and we now know there's at least two or three interbreeding events. So there actually in East Asians there is interbreeding from a group closely related to the genome that was sequenced in Siberia but in New Guineans there's interbreeding with a different Denisovan genome that separated 350 to 400,000 years ago from the Siberian one. So there's also southern Denisovans and northern Denisovans and we can tell all of this from the ancient DNA and also from the bits of DNA that are left in modern people and are detectable.
We didn't have a fossil record until last year, but Xiaomei Fu working in Beijing succeeded in getting DNA from that calculus, the dental plaque in teeth from this individual, well more than 100,000 years old, called the Harbin skull, and also sequenced protein from it using protein sequencing and it's a match to the Denisovan. So we now begin to have a morphology of what the Denisovans were like.
So this is an amazing thing to discover that there's this whole group of people who are our cousins, as close to us in many ways as Neanderthals, and yet we simply didn't know existed before this technology came online.
Modern humans and archaic humans, what do you think they did when they encountered each other? Well, what we now know is that they mixed almost whenever they encountered each other. And the evidence for this is as follows. So in 2015 we got DNA from Oase Cave which is in Romania. And this individual genetically has huge chunks of Neanderthal ancestry.
Remember I told you that when interbreeding happens between two groups like Neanderthals and modern humans, the DNA chopping starts. There's one or two breaks per generation as the DNA between the mother and father is spliced and you send it to your offspring. And the number of chops of the Neanderthal DNA was only four to six generations. This person had a full Neanderthal four to six generations back in their family tree.
A few years ago, Mateja Hajdinjak working in Leipzig then sequenced several individuals from Bacho Kiro Cave even a little bit older, closer to 45,000 years old. And several of these individuals again had Neanderthals four to six generations back in their family tree. So when Neanderthals and modern humans were overlapping each other, they frequently interbred and if we get to sequence one of these individuals, we frequently see recent Neanderthal ancestry back in their family tree.
This was another amazing result from Svante Pääbo's lab. They screened thousands of little tiny uninformative fragments found in Denisova Cave for whether they were human. They used this protein sequencing. So they take a little bit of sample from each of these uninformative bone fragments and they use protein sequencing to see whether the collagen is typical of humans and they found just a few of these thousands of bones that were not animals, well we are animals, but specifically human.
And they sequenced or tried to sequence with DNA one of them and this one bone turned out to be a first-generation mixture of a Denisovan and a Neanderthal caught in the first generation. So this individual had a Denisovan father and a Neanderthal mother and it was a first-generation woman that was the product of that mixture. People of very, very divergent backgrounds in humans across the range met and mixed when they encountered each other.
So this is a summary of the model that we basically have struggled to produce over a decade and a half now and is sort of the standard model of human and archaic population relationships. It involves multiple features that I'm going to point to you right now. That Neanderthals and Denisovans stem from a common ancestral population several hundred thousand years ago, four or 500,000 years ago. That these split from the ancestors of modern humans maybe 6 to 7 to 800,000 years ago.
That there is a very divergent, well more than a million years, population that interbreeds with the ancestors of Denisovans a few percent. That modern humans, all modern humans, all the people in this room are the product of very deep population substructure in Africa with lineages that are very divergent and that came together a few hundred thousand years ago, possibly in proportions 20% 80% if there were two lineages.
And that there was another interbreeding event that we didn't know about in 2014 when many of these discoveries were made, but that was found two years later, which was about 5% of the DNA from Neanderthals comes from a modern human lineage. So this is the current standard model, but it can't be true. And the reason it can't be true is this. It maybe can be true, but it would be very weird if it was true. So I'll show you why it's hard to be true.
It has about this 5% gene flow from modern humans into Neanderthals a few hundred thousand years ago. But the mitochondrial DNA that you get from your mother and she gets from her mother and the Y chromosome that you get from your father and he gets from his father, they're split no more than 450,000 years ago very confidently, which is much, much less than this number which has now been updated to 700 to 800,000 years ago.
So it must have come through this 5% interbreeding event and then gone into Neanderthals and gone up to 100%. Both the mitochondrial DNA and the Y chromosome. There can't be two parts of the genome that are more different from each other, one on the entirely male line, another on the entirely female line. And yet this happened. The probability of this happening by chance is 5% squared, a very small number.
And so one wonders whether there's some subtle type of explanation that can produce this pattern. We still haven't found it as a community, but it's still evolving and could potentially be addressed with new ancient DNA or thinking harder.
So the second thing I wanted to tell you about in the theme of ancient DNA being very surprising and when you use it to look at a new type of data revealing things that are fundamentally different from how we expected is that before ancient DNA we had a fundamentally wrong understanding of the pattern by which human populations migrate and mix with each other. So I'm going to start by telling you about my own story generating this kind of data and it started with a frustrating faculty search at my institution in 2011.
So there was a department that was searching for a human evolutionary geneticist. That was the description of the job and I was brought on, I was at the medical school but they brought me on as a member of the search committee to look for this person. And we got the most amazing applicants for this position.
And so for example, one of these applicants was Johannes Krause, who was a sort of star postdoc working with Svante Pääbo who had for example been the person who had extracted the mitochondrial sequence from the Denisovans and had done all sorts of exciting things. He's now the director of the Max Planck Institute in Leipzig.
Another of these people was a star postdoc here in Copenhagen working with Eske Willerslev, Rasmus Nielsen, who had done some of the first sequencing for example of the Saq genome which was the first large-scale DNA sequence that had been reported in 2010 and also other really amazing work.
And then the people who directed the search committee said we are immediately disqualifying these people because ancient DNA is not relevant to human evolution. So I exploded and I walked out of the room and I said I can't be in this committee anymore and I said I'm going to do this myself.
And I didn't have the right kind of laboratory. I was working on prostate cancer and I just closed down that type of work. I had a researcher in my group who used to work in the Leipzig group with Svante Pääbo and other colleagues and Matthias Meyer. She had come to work on cancer and to do other things but she knew how to do these things. So we decided to try to build a clean room. We had no money for it.
But we knew what to do. The people in Leipzig helped us to build it. And our department converted a tiny 18 square meter cold room that was a disused cold room, used sort of like a walk-in refrigerator, into an ancient DNA suite. And we decided to try to generate data cheaply because we had no money.
And so the idea that we used was we were going to just take tricks from medical genetics, which was my background, instead of using this brute force approach that was being used by the well-funded laboratories like the one here in Copenhagen and the one in Leipzig, where people would look for interesting samples and then apply lots of sequencing power to them. That was enough research to do. We just didn't have the money to do that.
And also, we were convinced that large sample sizes was going to be key to discovering things that were important in the last 10,000 years, which was the period I wanted to query. And this approach of just brute forcing one or two or five samples just was not going to work.
So what we decided to do was to borrow a trick from medical genetics which was basically something called hybridization capture. So what you do is you synthesize a set of artificial baits to target the parts of the genome you were interested in. This technology had been developed for what's called exome sequencing. It was still expensive at the time to sequence the whole genome. So what you would do is just sequence the genes. The genes are only about 2% of the genome.
So you would use a DNA printer to print out the baits that just target the genes, 2% of the genome. You would wash your sample over these baits, they would stick and you would reduce the amount of sequencing you would do by two orders of magnitude, 100 times, 50 times, 30 times.
So what we did working with Matthias Meyer is we targeted not all the genes but more than a million positions in the
genome that we know are variable have been studied in many many living people and that we could use to study relationships amongst people. And this made it very very inexpensive to do this type of work. Right now the cost of generating high quality data on a sample per sample and reagent cost is maybe $240 US. It became very very efficient to do this and we also quickly roboticized it. So doing this on 96 well plates and so we could for example in our laboratory screen about 400 samples per week and over a year a very large numbers of samples using this methodology. So as a result of this there's been a real boom in the amount of ancient DNA data driven also by brute force shotgun sequencing but largely by hybridization capture which really makes this a very efficient process.
So what happened is until 2015 really there were only a few dozen genomes with this sort of very careful work genome by genome working very hard and then hybridization capture was introduced and that's this big rise to many hundreds of genomes eventually more than a thousand and then what happened is there was a further takeoff as the methodology was roboticized with these liquid handling robots.
So a prevalent assumption until 2015 and which was really a reaction to the abuse of archaeology during the Nazi period for political reasons was that changes in the material culture of Europe 5,000 to 4,000 years ago were unlikely to be due to migration. And let me tell you a little bit about this history. So at the beginning of the 20th century, researchers, archaeologists like Gordon Childe and Gustaf Kossinna began noticing that certain layers in the European archaeological record were marked by the spreads of specific types of ceramics and other material remains across wide regions. In northern and eastern Europe, places like Denmark, there's the spread, maybe 4,500 to 5,000 years ago, of a phenomenon that's known as corded ware, a particular way of making pots and other artifacts that spreads over a vast region of Northern Europe and eventually even gets to the lower Rhine area in the Netherlands and spreads over a large part of Eastern Europe. And what they argued was the spread of this and also in the west of Europe, a phenomenon that closely followed it, the Bell Beaker phenomenon, were the movements of people spreading these pots.
And in Nazi Germany, this archaeology was weaponized to argue that these were the people who first spoke languages like Indo-European languages, which includes German. These were the place where the Aryans were and this was the natural homeland of the Germans, and therefore a claim to territory could be extended based on the historical territory over which the corded ware people lived. After the Second World War the reaction of the archaeological community to the abuse of archaeology in this way was to argue that you cannot argue from the spread of pots that people are spreading. It's pots not people. People can be copying ideas just like you will copy a cell phone even if you use cell phones in Korea even if it's not invented in Korea. And that was the orthodoxy until 2015.
But nobody had been able to look at it because nobody could collect DNA. People didn't know how these people were related to each other. But it wasn't possible to test this as I said until the advent of ancient DNA. So the background of this is the phenomenon in linguistics of the discovery that languages are connected to each other. One of the great language families of the world, there's many, is Indo-European languages. I'm speaking one. I'm speaking a Germanic language, English. Danish is another Germanic language. It is one of many languages that are closely related within the last 8,000 years or probably even the last 6,000 years, six or seven thousand years. This was discovered by a British colonial judge working in Kolkata in what's now India more than 200 years ago who had learned Greek and Latin in his schoolboy days and noticed that when he went to northern India he was interested in linguistics and ancient texts that the Sanskrit, the liturgical language of India, was related in the form of its grammar and many of its words to Latin and Greek and since that time it was identified that not just Latin, Greek and Sanskrit, Indic languages, but also Iranian, Armenian, Celtic languages, also Germanic languages, and also ancient languages that are no longer spoken but are known from inscriptions, parchments, and are Hittite, Anatolian languages in Turkey, known from tablets in Hattusa, the Hittite capital, or Tocharian in eastern China, Xinjiang, are all related closely within the last seven or six or eight thousand years. So, how could this be? The speculation was that they were once spoken, these languages, by a population that lived in the past and whose descendants or whose cultural influence has spread across this vast region of the world. Colin Renfrew, the great and late archaeologist, hypothesized that the way the spread happened was through the spread of farming. So a cataclysmic and transformational economic and cultural invention was farming which appears for the first time in the world 11 to 12,000 years ago in the Middle East and then spreads out of the Middle East into Europe 8,500 years ago, explodes into the Balkans and then spread within 2 to 3,000 years to the far reaches of Europe, to Britain for example by 6,000 years, to Scandinavia maybe by 5,000 years ago. Where it spreads culture is transformed. People live at much higher densities. They use many many different ceramics and so on and so this is the type of movement that could spread languages. He argued we know from studies of diverse societies that when people move languages move too, with exceptions, and that that's a powerful mechanism by which languages would spread. These same farmers also spread east, ultimately they spread across the Iranian plateau into the Indus Valley and into India and that would maybe be a mechanism by which those languages could be brought to India. So this was the dominant theory in archaeology until 2015 and was disproven by ancient DNA in the same way that for example Neanderthal interbreeding into modern humans was disproved. So I'm going to show you the evidence for this.
So what you're seeing here are about a thousand points on the left plotted in an XY plot where I haven't labeled the axes and they come from the locations here. That's Spain. That's Britain. That's Scandinavia. That's Turkey. And they come from the sampled locations here. And the way this plot is made is this is what's called a principal component analysis, by taking a table of data. So we're looking at about 600,000 positions in the DNA where people differ from each other. Some people might have a cytosine, one of the four letters of DNA. Some people might have an adenine. And so you should think of the data as a table with 600,000 rows corresponding to all the positions you've analyzed and a thousand columns corresponding to all the individuals that are being analyzed. And in each cell of the table is a 0, 1, or 2 depending on whether the person carries 0, 1, or 2 copies of the adenine or 0, 1 or 2 copies of the thymine. And then you multiply the table by itself and get a thousand by thousand table showing how closely related each person is to every other person. And then you operate on this table by some mathematical technique like principal component analysis that most efficiently separates people from each other based on their relatedness pattern, their thousand by thousand table. So this is principal component analysis and what you see on the x-axis is the most efficient way of separating them and on the y-axis is the second most efficient way of separating them. And what you see is this amazing pattern in Europe where Europeans are all over here. The Near Easterners are all over here. The northern Europeans are here. Like this might be Lithuanians or British people. The southern Europeans are here. These are Sardinians. The northern people from the Near East are here. These are people from Armenia and the Caucasus. The southern Europeans are here. These are people from the Levant and Arabia. And there's almost no one in between. The people who are in between are people of plausible, more recent contact between Europe and the Middle East, like people from Greece and people who are culturally Jewish.
So, how did this happen? Well, I'm going to now gray out these spots. Same dots, but now they're gray. And we're going to look at how the ancient people relate to the modern humans. So the game we're playing here is we have done this principal component analysis and the positions of each individual are something like 0.1 times the value 0, 1 or 2 at position one plus 2 times the value at position 2 minus 0.3 times the value at position 4 plus 2 times the value at position five all the way up to 600,000. You do a linear sum and that's how you get these positions on the plot. That's the first and the second principal component. So you've learned the patterns based on patterns of modern variation not using any of the ancient samples but you then could sequence the ancient genome and see where they fall. So this is what you see here and these are the hunter gatherers of Europe, the people who lived 8,500 years ago and you get an amazing pattern. So the hunter gatherers fall outside the variation of modern Europeans. The patterns are learned from modern Europeans but the hunter gatherers fall outside of that pattern. They're more differentiated from the Near Easterners in the direction that Europeans today are differentiated. The only possible explanation for this is that Europeans today are a mixture of people like this and people like this. You can prove this.
The next people are the farmers in Anatolia and Turkey and also in southern Europe and the Balkans and they fall in a completely different place in the plot. It's a mass migration from Anatolia into this area which is unrelated to the hunter gatherers and that happens after 8,500 years ago. And you don't see anybody like Europeans today 8,000, 7,000, 5, 6,000, 5,000 years ago. 5,000 years ago you see the appearance in the steppes north of the Black and Caspian seas, Ukraine and western Russia. This group is pictured here. This is a Yamnaya steppe pastoralist. This is from an extraordinary archaeological culture that appears for the first time 5,300 years ago. They're the first people to occupy and exploit the open steppelands which were previously devoid of people and people couldn't exploit them because there was no water that could be accessed. People in that part of the world would live along the Don River Valley or the Dnieper River Valley or the Volga River Valley but did not live in between despite all the biomass that was there to exploit as they couldn't get water there. But the Yamnaya took the recently invented wheel and they hitched it to wagons. They put them on wagons and then hitched animals to wagons and they took their supplies including water out into the open steppelands. And when this happened, all the previous cultures on each of the river valleys disappeared. You don't see the villages anymore. You don't see the pottery anymore characteristic of them. In fact, all villages disappear because these people are extremely mobile pastoralists. They just spread over a vast area all the way from Hungary in the west to Mongolia in the east. And you see a loss of cultural variation as a result of this phenomenon. And so this is what happens. And these are the people and they pile up in a different place on the plot. So you still don't see Europeans today. But then after 5,000 years ago, boom, you see Europe appear. So it's a very important period between 5,000 and 4,500 years ago. And what happens is these people mix with these people to form those people. So the Yamnaya form, they get to the edge of the steppe. They can't go further because their economics doesn't work anymore to go beyond the steppe. It's based on steppe exploitation. But they mix with European farmers and that mixed population in the form of corded ware expands further west and brings Yamnaya ancestry into Europe.
And so these results disprove Colin Renfrew's theory that the only way that languages could have spread into Europe is through the advent of farming. The argument was the only way a large-scale movement of people that would be large enough to bring languages could have moved into Europe was through something economically as transformational as farming once farmers were densely settled in Europe. How could anybody make an impact? It was like people in India today have been politically controlled by the Mughals from Afghanistan, by British people for many hundreds of years. But those populations made hardly any dent in India because it was a big established agricultural population. So how could there be a large enough movement of people to achieve a language shift at this time? But it happened. This happened. It happened somehow. The data tells you it happened. And in fact this is in many people in Europe today including in this part of Europe the most important ancestry component, these steppe pastoralists. Another large component is these early European farmers.
So Yamnaya descendants then started spreading further and further west and in a way that closely tracks the branching patterns of Indo-European languages. I'm going to show you two examples of this. The first is Britain where we have extraordinary and beautiful data. So Britain is occupied by hunter gatherers after about 14,000 years ago as the glaciers retreat from Britain. It's cleared of people by the last ice age. And then from 6,000 years ago, farmers from the continent start spreading into this region, into Britain, and it's a nearly 99% to 100% turnover of the population. The local hunter gatherers hardly contribute to the resulting population who are migrants from the continent.
This population lives for 1,500 years in Britain. The last big stones at Stonehenge go up at the end of this period built by the descendants of the farmers who arrived 6,000 years ago. And then within 50 or 100 years suddenly there is a huge transition, a 90 to 100% population transition where on the x-axis I'm plotting, the y-axis I'm plotting the proportion of ancestry of these people from the east. So there's a very sudden transition with a minimum of 90 to 100% replacement. It's possible that the replacement was somewhat more gradual and that we're just not seeing it over here. So because we have so few samples because there's a cremation bias, the people who built Stonehenge mostly burned their dead and so we don't get DNA very effectively. But we now think that's actually very sudden indeed and I'll tell you why in a bit. Here's Iberia, Spain, and Portugal in the same time period, 6,000 to 4,500 years ago. Again, the y-axis shows the proportion of ancestry from the east. And what you see is that it's not as dramatic, but still very substantial, where the proportion of ancestry is now 40% rather than 90% as in Britain. But the thing I've done in this plot is to color the dots based on their Y chromosomes. So there's empty circles which are people who don't have Y chromosomes, people with two X chromosomes. But the colored ones are people who are male and have a Y chromosome that we can label by whether the Y chromosome is characteristic of people from the steppe north of the Black and Caspian seas or northeastern Europe or characteristic of local European farmers and hunter gatherers. And what you see in Iberia is a really extraordinary pattern where after 4,000 years ago, there's a mixture of the people with eastern ancestry and local ancestry, only 40% eastern ancestry comes in. But you notice there's no blue left on the Y chromosome. What that means is that the local males completely failed to reproduce. So this must have been a cataclysmic event for the local males. Even though there was only a 40% gene flow into this population, the local males completely failed to reproduce. Now, we don't know whether this was due to violence or discrimination or female mate choice, but in any case, it was a cataclysmic event. And 4,000 years or 3,500 years later, the descendants of these Iberians came to the Americas. And if you look at Antioquia, Colombia today in South America, 95% of the Y chromosomes are from Europeans and 95% of the mitochondrial DNA are from Native Americans. We know what happened there. It was a case of exploitation, of violence, of genocide in various forms. We don't know what happened here which was a pre-literate society but DNA tells you about these types of events.
Work that we just published last week was sort of more about this, what happened at one time transect through the Netherlands and you could actually see the origin of this population over here that spreads into Britain. It happens in the lower Rhine area very specifically because it's a mixture of corded ware who I told you about before who are formed of Yamnaya ancestry and a local Eastern European farmer ancestry. They spread to the Rhine region, mixed with a very characteristic population that was only there with a very high European hunter gatherer component. That unique cocktail adopts a western culture known as Bell Beaker that was not carried by steppe pastoralist ancestry people but the corded ware people mixed with this group that adopted this western culture and then exploded everywhere into Britain very suddenly in a very quick transition in the Netherlands and then that moves into Britain and that's what happens.
So you can now use the spread of Yamnaya ancestry across regions as a tracer dye to track the spread of Indo-European languages. So you look at Yamnaya ancestry as a filled circle here. You can look at how it gets diluted through mixture with local populations as it spreads west and as it spreads east. And what you see is that as it moves into Europe it forms the corded group which is maybe three-quarters Yamnaya ancestry. It mixes further with Rhineland populations. It becomes about half ancestry. That's the Beaker population. It gets diluted further in Iberia. And then you see a retroflex movement back across the Ural Mountains and spreading across this region and punching through the mountains into India between 4,000 and 3,500 years ago. All documented with ancient DNA in a particular pattern which is very non-random. So this backward movement is actually exquisitely documented with ancient DNA. There's the corded ware spread. It forms in northeastern Europe like places like around here but then it flips back bringing Eastern European ancestry and this exactly correlates with the linguistically reconstructed relationships of languages. There is this very weird phenomenon where Indo-Iranian languages cluster with Baltic and Slavic languages that corresponds to the formation of this population here which then flexes back over this region and explains this archaeological movement of people documented with ancient DNA correlates to this weird linguistic connection. It's so striking that it can't be wrong.
It's complicated though. We can also see spreads of cultures without genes. This is a paper from last year on the Carthaginians, the people who were defeated eventually by the Romans in three Punic wars. And what we did is we collected DNA from individuals who lived in this red area where there were Carthaginian colonies including in Carthage. These people spoke a Phoenician language that had first been spoken about 3,000 years ago in places like Sidon and Tyre. Carthage is a Sidonian colony. And they worshipped Phoenician gods. And so a question was how much DNA do they have from the Phoenicians? So when we got DNA from these people, we see this amazing pattern. So if you look at Phoenicians from this area, they cluster with previous Canaanite populations from that area just in the same place. Blue and red are those two groups I'm talking about. But if you look at the Punic people, the Carthaginians, you see they're in yellow and they just don't overlap. There's almost no contribution of the Phoenicians to the Carthaginians despite them speaking the same language and using the same religion. Instead, the primary ancestry of Carthaginians is Greek, probably via Sicily. And there's also some contribution from North Africa, which is where Carthage was. And it's the same mix whether you're in Iberia or North Africa or in Sardinia or in Sicily itself. And Carthaginians were genetically largely derived from the same source as their Greek commercial rivals, forcing even authors of our own paper to change their assumptions about this topic, who argued that there was probably a genetic or a biological connection between these people. It seems that somehow people from Sicily or this area who were Greek somehow switched to the Carthaginian way of life, maybe by intermarrying in lots of ways or maybe just thinking it was a better way to live than the Greek colonies. And that Greek expansion actually manifested itself both under Greek cultural guise but also under Phoenician cultural guise, genetically a completely different population.
So, here's some more examples of this. There's this assumption about, you know, the quote races that you see in the world today, like white people, Caucasian people. And if you go back in time, maybe you imagine these people existed in the past. But if you look in ancient DNA from people 8,000 years ago in Europe and Western Eurasia, there's four groups that are each as different from each other as Europeans and East Asians today. So if you were to go back to that time and look at the fault lines about how people relate to each other and categorize people, it just wouldn't break down as people break it down together. And it's not that any of these groups, these European hunter gatherers from the west, these European hunter gatherers from the east, these Levantine farmers or these Iranian farmers and food producers, it's not that any of them disappeared. It's not that any of them were pushed out. What we know happened is that in the Bronze Age, like times when we see the Yamnaya, they just mixed and people today from this region just mixed. And the reason there's low differentiation today is they mix together in different ways. The Yamnaya mixing with the European farmers is an example of that. The Yamnaya themselves are mixed of earlier groups. And so it's a mixture process that changed these patterns.
Here is another example. If you look in the southwest Pacific, not associated with Indo-Europeans at all, the first people of this region, for example Vanuatu in the southwest Pacific, there were no people in this area until 3,000 years ago, even though the adjacent Solomon Islands and New Britain and the Bismarck Archipelago had people for 30 or 40,000 years. But people in probably Taiwan figured out how to sail across open waters for the first time about 5,000 years ago and started using these outrigger canoes to spread across these islands and then eventually got to these places. Today people in Vanuatu and in New Caledonia have about 90% ancestry from New Guinea, from Papuans. And so the guess was that these first people would have had similarly a lot of New Guinean ancestry, but when we sequence them, they're almost entirely East Asian. So the first spread is almost entirely East Asian. And then today there was 90% Papuan ancestry. So it was similar to what happened in Europe. We knew that something must have happened between the earlier period like the European farmers and today when there's a new ancestry type. And so we did these studies to identify what happened in the middle. And what you see is that about 500 years after the first Lapita expansion, there's a huge pulse of primarily male Papuan ancestry, it's one of the most sex biased events I've ever seen, into this population. It floods the population with Papuan ancestry from the male side. And this mixed population ends up with about 90% Papuan ancestry. It happens as part of the same culture. So the first people are making pots known as Lapita. They spread into Vanuatu and New Caledonia with this Lapita culture and then Lapita culture is also practiced on the shore of this big island of New Guinea and is adopted by local New Guineans. These people bearing Lapita cultures but genetically extremely different then flood into the southwest Pacific and bring the Papuan ancestry.
So the third, the big perspective change from ancient DNA studies is that people living today are almost never the descendants of the people who lived in the same place thousands of years before. Everybody's mixed. No one is or really could ever be pure. And when we collect ancient DNA from previously unsampled times and places, I focused on Europe, but we see this in many places in Africa. We see this in East Asia and many places. We find that there are these ghost populations, major human lineages that simply don't exist anymore in unmixed form but were widespread at the time and that large-scale human movements have occurred with multiple time layers. You may think that the present day with African-American and Latino-American populations mixing in the last few hundred years due to big industrial changes and transoceanic travel that we're living in an unusual time of history. And in some ways we are. But from the perspective of mixing populations, we're not. We don't know about the mixing of people that led to South Asian populations. But we now know with genetic data that that happened profoundly between 4,000 to 2,000 years ago. All of the events I'm talking about in Europe are results of massive mixtures. We see massive mixtures in Africa. We see massive mixtures in East Asia. We see even massive mixtures in parts of the Americas.
So the third thing and final thing I'm going to tell you about is that before ancient DNA we had a fundamentally wrong understanding that natural selection has been quiescent in the last 10,000 years. So here's a paper from 2009 which is a very influential paper which had the title 'Classic selective sweeps were rare in recent human evolution.' So the observation was that if you look at genetic changes that are non-synonymous changes that change the protein sequences of genes, around such changes if there had been a selective sweep, a newly arising mutation that then jumps to high frequency because it's advantageous, there should be removal of variation in that vicinity because only one variant survived. You see this in fruit flies where there's natural selection sweeps, there's a divot of variation around non-synonymous sites that you don't see in synonymous sites that don't change the coding sequence of genes, but you don't see it in humans and based on this they argue that there's just not a lot of selective sweeps in our species over the last few hundred thousand years when you would expect to see this signal.
So based on this and also based on observations like if you compare Europeans and East Asians there's no places that are 100% different in frequency across populations as you would expect if there were selective sweeps on either the European side or the East Asian side. People argue that this has just not been common in evolution, that maybe somehow we had reached an adaptive optimum a few hundred thousand years ago and there just hasn't been much selection. And since that time, here's a blog post that appeared in late 2024 that argued that said that really we're not seeing selective sweeps even in the ancient DNA data of the time. So ancient DNA studies from 2015 to 2024 found only a handful of mutations that changed too rapidly to be due to chance. So we published a study in 2015 with a few hundred samples that looked to see genetic places that changed rapidly or substantially under the pressure of natural selection. And we found 12 such positions in a paper led by Iain Mathieson.
And all the work in the subsequent years really didn't find many more even though this huge increase in sample size. So for example, there was a paper led by the Copenhagen group that came out in 2024 that identified 21 genome-wide significant signals of natural selection. But in some ways that's very disappointing because the hope was that as sample sizes explode, we're going to find hundreds of positions and we're going to be able to study the patterns by which selection acted over time. And in some ways now we just have some anecdotal examples.
So what we focused on was to try to improve this in three ways. So one way we improved this in work led by Ali Akbari was to improve a statistical method. So we developed a method where the goal was not to look at the beginning and the end but instead to look to see whether natural selection was blowing in the same direction at all times in all places at each position. We also massively increased the amount of data. So this one study reports more than 10,000 individuals with high quality ancient DNA data. And so the combined data set is about 1,600 people from this part of the world spread over the last 15,000 years. And then finally we improved data quality by using statistical methods of imputation and other techniques. When we applied it to the data, we find 479 positions with 99% probability or more of being real. So this was a totally shocking result. It was so different from the results we had before. We were certain it was wrong. We had this result already three years ago and we just spent the next few years trying to make this result go away, but it just kept getting stronger and stronger. So how to reconcile our results with the lack of evidence for selective sweeps over hundreds of thousands of years?
So one of the answers which we're able to prove with the data is that we're in a period of intensified evolution at least in Europe and Western Eurasia. So one way you can show this is by looking at the genetic changes that are changing rapidly in frequency and see whether they're concentrated in traits that are associated with particular phenomena like immune traits or metabolic traits or behavioral traits. And what we see is that specifically for immune traits there's an extreme overrepresentation of selection signals. Immune traits are ones that are identified by just studies of variation in modern people. And also there's an enhancement of metabolic traits like traits like diabetes risk factors or so on. And when you look at this and you test whether the enrichment changes over time comparing the last 5,000 years to the previous time what you see is there's evidence for acceleration. There's in other words the intensity of the enrichment is increased in the last 5,000 years compared to the previous 5,000 years. So what this means is that there's intensifying evolution. So there's some kind of cultural or environmental or lifestyle change that is becoming more and more extreme and that is resulting in an acceleration of these patterns. So that's one reason why classic selective sweeps have been rare over a time period of hundreds of thousands of years. That's because there hasn't been such a dramatic cultural and environmental and lifestyle change over much of that period. We put populations into an entirely new environment practicing agriculture, living at much higher densities, living next to animals with different types of economic patterns.
The other pattern is that selection has fluctuated over time. It's not only become more intense, but it's changed over time. One way to see this is you can estimate the age of any genetic change that we find of these 479 changes that are affected by selection. And we can also, based on patterns of correlation to neighboring positions, estimate its age if the mutation newly arose and then rocketed up to high frequency under constant selection. And about a third of these mutations are the same as you would expect if they look like they've had the same on an x equals y line as if they had immediately been subject to positive selection when they arose. But two-thirds of them are over here. And the red dots are ones where the ancestral mutation, the old one, has actually been increasing in frequency in the last 2,000 years. So these are cases of selection reversal where most of these variants have had fluctuating selection on them. And the reason that you don't see sweeps is they went up and they went down and they went up and they went down because they were advantageous at different time periods.
So here's some examples of individual variants. So here are two variants that are major risk factors for celiac disease. This is gluten insensitivity. So these are, you can see how much data there is. This is confidence intervals for the frequency at different times in the past. You see there's very little error in these estimates because there's 16,000 samples. So this is the European hunter gatherers. This is the European farmers. These are the steppe pastoralists. Then it rockets up in frequency after 4,000 years ago. And this other variant somewhere else in the genome also rockets up in frequency after 4,000 years ago. You may ask why a variant that makes you have gluten insensitivity at a time when people are eating wheat is increasing in frequency and potentially speculatively it's protecting against some kind of pathogen or something like this. Here's another variant that was actually already identified in 2021 by a group in Paris as having decreased in frequency in the last few thousand years based on ancient DNA analysis. But they didn't yet have enough data to know that it had increased in frequency under the pressure of selection in the previous 4,000 years. So this variant, which is the strongest variant risk factor for tuberculosis, clinically severe tuberculosis, seems to have increased dramatically in frequency between 8,000 years and 2 to 3,000 years ago. And then the selection reversed potentially maybe speculatively because tuberculosis became endemic.
You can also look at combinations of genetic changes that affect traits. So from huge studies of modern people like hundreds of thousands of white British people living today, you can carry out a genome-wide association study and look at 10 million positions across the genome and see how they affect a trait like risk for type 2 diabetes. So each variant in the genome, you test whether it makes type 2 diabetes risk higher or type 2 diabetes risk lower. And you find thousands, literally thousands of independent positions that each affect risk for type 2 diabetes. A little higher, a little lower. And you can sum up their effects and make what's called a polygenic predictor that predicts your risk of type 2 diabetes. And this is true for height. And this is true for hypertension, blood pressure, and all these different traits, different types of cancers. So we can ask whether this polygenic predictor has fluctuated more than you would expect by chance in the absence of selection. So that's what we did. And when we do this for type 2 diabetes risk, we run this test on about 550 traits and a hundred of them show overwhelmingly significant evidence of directional selection for the trait.
So when we run this what we see is that type 2 diabetes risk predictor has decreased by a full standard deviation on the scale of modern variation over 10,000 years. So this doesn't mean that type 2 diabetes is less common today than it was in the past. This does not mean that because there's other cultural and risk factors. For example, obesity is much higher today than probably it was in the past and is an important risk factor for type 2 diabetes as well. But it means there was selection against risk factors for diabetes. Similarly, if you look at risk for psychosis such as schizophrenia and bipolar disease, there is about a one standard deviation decrease again over this time period, which means that there's selection against genetic variants that contribute to psychosis in this time period. Again, what this meant in the past, we don't know if psychosis as it manifests in white British people in industrialized society today may not mean the same thing 7,000 years ago in a shamanic background.
And then finally we also see effects associated with behavioral traits like the number of years of schooling is amazingly somewhat predicted by genetic data and this trait increases by something more about like six standard deviations over 10,000 years and it's a highly significant effect. So a caveat is that there was no school more than 500 years ago for most people and what this would have meant in the past and what behavioral traits were selected we don't know. The same traits are correlated to intelligence measures on IQ tests which of course also didn't exist in the past. The age at which women have sex for the first time, the number of children, the age and so on. Some very different traits and it's not at all clear what the selection is for.
So we were worried that these selection signals might be an artifact of insufficiently corrected population structure which was previously known to create false positives. So we did a number of analyses. The most striking one was we measured the effect sizes on for example years of education in Chinese people and saw how they correlated to genetic trajectories over time in European people. And you see a huge high correlation. There's really almost no way we could see that could be due to chance. We finally carried out a scan in East Asians in about 1,800 East Asians. It's a much smaller scan than in Europeans. Instead of finding 479, we find about 40 positions that are genome-wide significant. But what we can do is look at the enrichments. Remember I told you that immune traits are about four-fold enriched in West Eurasians. And you see a very similar pattern in East Eurasians and West Eurasians. So it looks like the natural selection over this time period is operating in a very similar way in these two groups. If you look at these 550 polygenic traits, they're highly highly correlated with exceptions especially for pigmentation over these two groups suggesting that East Asians and West Eurasians have reacted in parallel ways in these polygenic traits due to the introduction of agriculture and spread of higher densities and encounters with animals.
So this is the end of my talk. I gave you three examples of how ancient DNA has really forced our hand and produced surprising surprises challenging understandings of the world in deep ways. Before ancient DNA, we had a fundamentally wrong understanding of how modern humans relate to archaic humans. Before ancient DNA, we had a fundamentally wrong understanding of how human populations migrate and mix. And before ancient DNA, we had a fundamentally wrong understanding that natural selection has been quiescent in the last 10,000 years. Thanks. These are the people in my research group. This work was done in collaboration with many amazing people who made critical contributions to these projects. I'm only picturing the ones in my own research team who are right now, but people from my own group in the past and from many other groups often made critical contributions. I don't actually know how much time I talked because the clock wasn't really very visible. But I'm happy to chat or take questions. Thanks.
