Neandertal DNA

I've just submitted an abstract for a conference in the fall, with the title, "Immunogenetics of archaic humans."

Ten years ago, it would have been beyond imagining that this kind of science would be possible. Now, my graduate student Aaron Sams has been working directly with HLA and other immune system genes in ancient DNA sequences. It's pretty tough to work with the HLA region because of the low coverage of the ancient genomes and the high variation and repetitiveness of the HLA. But it is possible to find some of the basic human alleles in the ancient sequences, and those open the possibility of examining the coevolution of pathogens and human immunity in our recent evolution.

Turns out we're not alone: According to New Scientist, Peter Parham has also been looking at HLA in archaic humans: "Breeding with Neanderthals helped humans go global".

One allele, HLA-C*0702, is common in modern Europeans and Asians but never seen in Africans; Parham found it in the Neanderthal genome, suggesting it made its way into H. sapiens of non-African descent through interbreeding. HLA-A*11 had a similar story: it is mostly found in Asians and never in Africans, and Parham found it in the Denisovan genome, again suggesting its source was interbreeding outside of Africa.

HLA-A*11 is actually the most common allele of HLA-A in Papua New Guinea, the population that otherwise shows significant evidence of ancestry from a Denisova-like genome. However, I don't agree with the main idea of the article. The major human HLA alleles are evolutionarily ancient -- most of them predate the origins of modern human groups and are older than the founding of the Denisova-Neandertal populations. This is actually perhaps the worst region to look for evidence of interbreeding among these populations because the probability of incomplete lineage sorting (maintained by balancing selection) is very high.

As a case in point, HLA-A*11 is very common in Papua New Guinea, but it is also very common in north India and in China. These two areas otherwise show no significant evidence of Denisova ancestry. We might conclude that the HLA-A gene just has an unusually high level of introgression into Asian populations, not typical of the genome as a whole. That's certainly possible. But without finding any substantial number of derived mutations in the HLA-A*11 variant in the Denisova genome and in living Asians, it is hard to rule out that the sharing of HLA-A*11 in all these populations is just coincidence.

Of course, if the allele were absent in Africa, that would weigh in favor of the idea it is shared by Late Pleistocene interbreeding outside Africa. But HLA-A*11 is in Africa, just very rare. And it's in Europe. This is the kind of locus that is difficult to interpret: if it has any tiny disadvantage against malaria, for instance, its rarity in Africa is easily explained as a function of recent evolution, while its presence almost everywhere outside Africa would be no surprise even if there were never any interbreeding. This is not a case where the geographic distribution is an unusual coincidence -- it's present in Africa and relatively more common everywhere outside sub-Saharan Africa. So the distribution outside Africa cannot simply be explained by interbreeding with Denisovans -- not without selection -- leaving us stuck. Parham's hypothesis may be correct, but the data are really not sufficient to decide.

HLA-C*07:02 -- the one apparently mentioned in the story -- is all over sub-Saharan Africa at low frequencies. Allelefrequencies.net has a dozen entries for the frequency of HLA-C*07:02 in sub-Saharan Africa, they all have it at frequencies up to around 7 percent (except for the small (n

What about the question of hybrid vigor that the article raises? Is it possible that modern humans got HLA mojo from Neandertals and Denisovans?

While only 6 per cent of the non-African modern human genome comes from other hominins, the share of HLAs acquired during interbreeding is much higher. Half of European HLA-A alleles come from other hominins, says Parham, and that figure rises to 72 per cent for people in China, and over 90 per cent for those in Papua New Guinea.

I just don't think it's clear that these HLA alleles in humans have actually come from the archaic genomes.

We've tried to match these at more precise levels (in the HLA system, that would be four- or six-digit haplotypes) and have not found the quality of the data high enough to manage a close match. That leaves us with the most superficial classification, which isn't enough to argue that the present human types are derived from the archaic genomes. Incomplete lineage sorting remains a good explanation for the similarities. In fact, we're thinking it makes a nice case study of just how hard it is to work with these genomes, which have lower than 2x coverage. Just typing the Denisova genome requires an assumption about whether the individual was a homozygote or heterozygote across the locus -- an assumption that we can test easily with higher coverage, but not so much with 1x and many gaps. It also requires greater trust in the mapping quality of the reads than we probably should have. With those caveats, the match to HLA-A*11 is likely but not totally solid. Saying that HLA-A*11 in modern humans came from Denisovans is simply premature. And while I've focused here on HLA-A, this is also true of all the other loci. There's a tipping point at higher coverage where typing becomes more secure, and the archaic data are not there.

Anyway, I imagine that anyone typing HLA in whole genome data knows all this. The press account isn't going to go into the complexity, and I think it's worth noting the real difficulty of making inferences in this region of the genome on the archaic data. It's a tough problem and I've spoken to many human geneticists who thought we were foolhardy to start. But with the first information about the immune systems of archaic humans as the goal, you can see it's a worthwhile problem to tackle.

Shanti Pappu and colleagues [1] report on date estimates resulting from new excavations at the old site of Attarampakkam, India. The news element is that they date an Acheulean occurrence to as old as 1.5-1.6 million years ago. At the oldest, these dates would make the Acheulean in India equal in age to the earliest occurrences in Africa.

The dates themselves depend on the decay of cosmogenic nuclides in the artifacts themselves. This is a kind of exposure dating -- as the artifacts are exposed to cosmic rays at the Earth's surface, they build up radioactive isotopes of beryllium and aluminum (¹⁰Be and ²⁶Al), which have half-lifes of 1.39 million and 717,000 years, respectively. When they are buried deep underground, their exposure to cosmic rays stops, and the radioactive isotopes can only decay. Then the ratio of the two isotopes in the sample reflects the time since deep burial. But like other exposure methods, in practice this depends on a model of exposure time, burial speed, and radioactivity within the soil, which lends substantial uncertainty to the dates. The lower 95% confidence interval of each of the date estimates reported in the paper is still over a million years, leading to the minimal conclusion that the site is that age or older.

Robin Dennell has written an accompanying short essay that gives a broader view of the Acheulian in South Asia [2]. The essay includes a great paragraph summarizing the now-obsolete idea that Acheulean reached India only a half million years ago:

How does this new evidence affect our understanding of the South Asian Acheulian? Previously, the general consensus was that the Indian Acheulian was less than 0.6 to 0.5 Ma (5) and was thus much younger than that in the Levant (eastern Mediterranean). There, the earliest dates of 1.4 Ma, from ‘Ubeidiya in Israel, probably indicate a dispersal of hominins from Africa (6). A second influx of African immigrants is indicated by the discovery of African types of cleavers and hand axes at Gesher Benot Ya'aqov (GBY), in Israel, dated to 0.78 Ma (7). This evidence implied that the Acheulian dispersed eastward toward South Asia only several hundred millennia after it first appeared in the Levant. It also implied that the spread of Acheulian bifacial technologies into South Asia was broadly contemporaneous with its first appearance in Europe, where the earliest sites date from ∼0.5 to 0.6 Ma (8). Some have attributed this expansion of the Acheulian into South Asia and Europe to Homo heidelbergensis. This Middle Pleistocene type of hominin is known mostly from Europe, where it was first defined, but is also recognized by some (but not all) researchers at African sites such as Bodo, Ethiopia, and Kabwe, Zambia, and even at some sites in China (9).

The "Homo heidelbergensis" model is in such utter disarray right now, I'm not sure many paleoanthropologists have realized the full extent of the problems. You should know that I don't believe in Homo heidelbergensis, never have. A couple of months ago, I was discussing some of the issues about mutation rate estimation with a very prominent geneticist, and the conversation turned to Homo heidelbergensis. What a shock the Denisova sequence should have been to those itching to see a H. heidelbergensis incursion into Asia!

Notice however, the intrinsic nuttiness of archaeological interpretation. Oh, we have the first evidence for Acheulean in India around 600,000 years ago? Well, that's around the same age as the Bodo fossil from Ethiopia! What a coincidence! Maybe this new kind of hominin expanded from Africa and carried the Acheulean to India! And Sima de los Huesos is around 600,000 years old, too -- and there's a handax in the pit! My gosh, we need a name for those hominins!

Well, the nice thing about a hypothesis built on mere coincidence, is that it only takes one observation to falsify it. Million-year-old handaxes in India ought to do it, and how. That's the message of Dennell's essay, and the subtext of the paper by Pappu and colleagues. What I find interesting is the extent to which the fact was hinted by earlier discoveries in South Asia but hampered by weaknesses in stratigraphic control and dating. From Pappu and colleagues:

Sparse radiometric ages from sites in India have situated the Acheulian within the Middle Pleistocene, with a few dates suggesting an early Middle to Early Pleistocene age. However, these ages often exceed the limits of confidence of the methods used (2). They include an electron spin resonance (ESR) mean age of 1.27 ± 0.17 Ma, assuming linear U uptake, on two herbivore teeth from Isampur (23); an ESR age of ~0.8 Ma (lacking uncertainty envelopes) on calcrete from the Amarpura formation, Rajasthan (24), which has been correlated with the Acheulian site of Singi Talav (4); dates ranging from ~1.4 to 0.67 Ma for the tephra at Bori (Kukdi river) (25); and paleomagnetic measurements with evidence of reversals at the sites of Bori, Morgaon, Gandhigram, Andora, and Nevasa (26). However, the reliability of these ages has, in each case, been questioned on various grounds (5, 27, 28). Likewise, the age and stratigraphic position of artifacts and faunal remains from the Early Pleistocene Dhansi formation along the river Narmada are yet to be firmly established (29). Based on data from controlled excavations and two independent dating methods, our ages from Attirampakkam show that the Acheulian in India is older than previously thought. Evidence from other sites in South Asia should be reconsidered and redated.

Much evidence already exists in the South Asian Acheulean that could be more accessible. The Acheulean in the region has been a long block of undifferentiated time, despite some very well-resolved sites. In addition to this much older dating for early Acheulean, India also has some of the youngest Acheulean assemblages anywhere -- for example, Haslam and colleagues [3] earlier this month reported on an Acheulean assemblage from around 130,000 years ago in northeastern India. That's long after the large biface tradition begins to give way to Middle Paleolithic and MSA toolkits in Europe and Africa.

On the topic of Denisova, Haslam and colleagues were writing before that genome was reported. But they did know about the Neandertal genetic results, including the evidence of Neandertal ancestry within India. Nevertheless, they assert a scenario in which the makers of earlier and later Acheulean in South Asia are the same biological population, without substantial gene flow from regions to the west, including the Neandertals.

Recent reports of the draft Neanderthal genome suggest that Neanderthals and H. sapiens likely did interbreed successfully soon after the latter had left Africa (Green et al., 2010), with the probable location of such contact to the west of India, in the Middle East. The southern limit of the Neanderthal range is unknown (Dennell and Roebroeks, 2005), but we emphasise that the continuity seen in the Middle Pleistocene South Asian technological record suggests that taxa derived from earlier hominin dispersals, and not Neanderthals, were the creators of the Indian Late Acheulean. Greater biological separation between dispersing humans and resident Indian hominins may have precluded viable genetic mixing (although see Liu et al., 2010 for an alternate view from East Asia), while similarities in certain technological strategies may have rendered cultural exchange a somewhat more likely occurrence.

Well, the Denisovans didn't have to live in India when the ancestors of Melanesians ran across them and intermarried. But Denisova and the Neandertal genomes now make it very likely that the inhabitants of South Asia were one or the other. And even if South Asians were yet a third group, as yet unattested from genomes, it is no longer credible to suppose that they were isolated from Europe or Africa for a million years previous. The tools just don't have that much to do with the populations.

References

When last we saw the Vi 33.16 X chromosome, I was wresting out its secrets by looking for SNP haplotypes shared by this Neandertal with the European and African samples from the HapMap ("Neandertal segments of X chromosomes"). Neandertal haplotypes in the CEU (Utah, European ancestry) sample, that are not also found in African samples, are candidate loci for Neandertal ancestry outside Africa.

In my earlier post, I pointed out some drawbacks and weaknesses of this simple approach. The SNPs have poorer power than sequence data, and we will miss relevant short haplotypes. Some Neandertal-derived alleles are probably present at low frequencies in Africa. Excluding rare African alleles will cause us to miss these cases. What we will find is a filtered set of Neandertal candidate loci, where we don't control the filter.

Finding these haplotypes lets us look at their frequencies within the European sample. As I pointed out, most of the Neandertal haplotypes in the CEU sample are rare, one or two copies. A handful are quite common, up to 30-40 copies in the sample. A good-sized set occurs in 5-10 copies.

We know from Green and colleagues' comparisons that at least three people outside of Africa have the same fraction of Neandertal ancestry -- one from France, one from China, and one from Papua New Guinea. But there's no reason to think they have inherited the same segments from Neandertals. The overall proportion of Neandertal ancestry is very slight, less than five percent. If five percent of loci were 100 percent Neandertal, then everyone would have the same Neandertal loci. But that's not the way they are distributed. Different individuals certainly have different Neandertal genes.

A rare allele in one sample is quite likely not to appear in geographically distant samples. So for many of the Neandertal haplotypes in the CEU sample, we shouldn't expect to see them in China. And, as you can tell from the figure below, that is in fact the case.

What you're looking at is a 3-D histogram of Neandertal candidate haplotypes in China and Europe. The number of copies in the CEU HapMap sample is on the X axis, the number of copies in the CHB HapMap sample on the Z axis, going back into the picture. From the leftmost corner, at the origin, going along the X axis is the set of haplotypes present in CEU but absent in China. As you can see, the most frequent outcome is one copy in either one sample or the other. This being a histogram, those are both lumped into the highest bar at the origin.

Here's a detail of the area near the origin, turned upward so we're looking at almost an X-Z plot.

As we go down the X axis, you see there are many haplotypes with 3 or 4 copies in CEU and none in CHB. In fact, there are very few that have 3 copies in CEU and any in CHB -- many fewer altogether than occur in 3 copies in CEU and none at all in CHB. The ones that have 10 or so copies in both samples are, well, scarce.

This is very striking. China and Europe by and large have different Neandertal-derived haplotypes. Haplotypes from Neandertals that are common in Europe -- say, with more than two or three copies -- are mostly rare in China. And vice-versa; haplotypes that are common in CHB are rare in CEU.

Why should this be? Green and colleagues [1] hypothesized an early population mixture of Africans and Neandertals in West Asia, before that population dispersed throughout the rest of Eurasia. This hypothesis was meant to explain why China and Europe have the same proportion of Neandertal genes.

I think that is also consistent with the fact that China and Europe have different Neandertal genes. If the population mixture was followed by substantial genetic drift as the West Asian population dispersed in different geographic directions, drift would randomly increase the frequency of some haplotypes in one direction, others in the other direction. Europe and China would end up with the same proportion of Neandertal ancestry, but it would be distributed very differently among loci.

Next, we'll examine whether this pattern is the same for the rest of the chromosomes. Or maybe something even more interesting...

References

When I wrote about the Denisova genome late last year, I claimed that "A large-scale reorganization of the science of human origins is upon us."

I'm glad I had the sense to write that. A lot of people have pointed back to that quote over the last few months. Still, I know that the full implications of the Denisova and Neandertal genomes haven't really sunk in. "Large-scale reorganization" takes time.

A new paper by Brenna Henn and colleagues in PNAS [1] shows how the shifting landscape has caught many geneticists off their footing. Submitted before the Denisova genome, but long after the Neandertal, the paper is titled, "Hunter-gatherer genomic diversity suggests a southern African origin for modern humans". In today's landscape, with only one instance of the word "Neanderthal" in the paper, the conclusions are obviously incomplete.

The "southern African origins" conclusion of the paper comes out of a simple analysis that assumes that the best-fit maximum for genetic diversity (as assessed by linkage) is the most likely point of origin of the population. That would be true if the African population emerged by a series of founder effects from a single small ancestral population -- the "serial founder effect" model that I have criticized here before. But of course in 2011, we know that model is false, because it is predicated on a lack of ancient mixture with Neandertals or other populations. If the serial founder model can't work outside Africa, it certainly can't work inside Africa, where populations were larger and regionally diversified during by the beginning of the Late Pleistocene. Without that false assumption, the "southern African origin" evaporates. The primary observation, a cline of linkage disequilibrium within sub-Saharan Africa, can be explained with reference to mixture of populations without assuming an origin and expansion from one geographic location.

I don't want to criticize overmuch. Many ongoing research projects are casualties of our new knowledge of ancient genomics, and we'll see more papers like this before the fallout has settled. Simplistic founder models, acceptable only a year ago when these projects were conceived, are now unquestionably false. Ancient population mixture is the order of the day, and we don't have any simple, plug-in-the-data models to apply to data like these.

Instead, I want to consider the power of the data in this article to answer some fundamental questions about African population history. Henn and colleagues report on SNP genotyping of several Bushman groups from southern Africa and Sandawe and Hadza people from eastern Africa. These data are on the 550k SNP platform that was used by 23andMe before the recent increase to 1M SNPs. That means the data are comparable to many other studies. They are not entirely comparable with other samples of African genetic variation, and the authors cut the total number of SNPs down to the 55,000 that overlap among all the genotyping platforms used in their analysis. For this reason, the paper presents a genome-wide set of 55,000 SNPs across many African populations.

It's far from the perfect sample. I expect we'll be able to do much more with the full 550k dataset from the hunter-gatherer populations. The data have been made publicly available for download, and here we're already starting to investigate them.

Within the current paper there is a very useful analysis of the broader dataset using the ADMIXTURE software. ADMIXTURE assumes that the current samples represent a mixture of ancient populations that were more distinct than today's. I went through this algorithm with my students in class Wednesday and Friday, which I'm sure was an intimidating process to most of them. The math is not too conceptually daunting; it's just hard to conceptualize how all the possible interactions relate to gene frequencies when you are assuming more than a few putative ancestral populations. Razib Khan gives an impressive step-by-step guide to performing an ADMIXTURE analysis, including some of these samples.

I'm not in love with this analytical method -- there's no reality check on its assumptions. But its output can be informative about many aspects of population structure. Here are some first approximations:

1. The genetic diversification of African populations was once much greater than today. Razib Khan points out the homogenizing effect that agricultural populations have had on the African continent, particularly during and after the Bantu expansion. I think the current data suggest that earlier processes involving LSA hunter-gatherers also tended to homogenize populations.

For example, when eight initial clusters are assumed, the ADMIXTURE analysis constructs them in a way that most of the ancestors of today's Bushmen were in a population with a high degree of genetic divergence from the other seven ancestral populations. The F_ST between the Bushman ancestral population and others ranges from 0.1 (for forest pygmies) to a high of 0.25 (from Europeans). That estimate is nearly double the equivalent statistic in today's populations.

Again, we don't have to believe the assumptions underlying the ADMIXTURE algorithm, but it does highlight the basic partitioning of diversity in the African population. Today there is high diversity within African population samples, and some of that diversity can be traced back to populations of 100,000 years ago or more. Some of the diversity that once existed among these populations has now been spread within them instead. The populations got genetically closer over time.

A model of successive population expansions, bringing ancient populations genetically closer and closer together, is also what we may see in other places. As we have learned more about the mtDNA of ancient Europeans, it has become clear that successive expansions and migrations of people into Europe have radically reshaped the gene pool.

2. Click languages have no genealogical unity. Over the years, many linguists and anthropologists have proposed that Hadza, Sandawe, and Bushmen are closely related to each other, despite their geographic distance, because they all speak languages that use click sounds. No historical linguist has ever successfully demonstrated a system of sound changes or detailed correspondences among these languages, but people promoting the hypothesis seem immune to these kinds of facts.

The genetics show a very clear and ancient differentiation of these hunter-gatherer peoples. In the ADMIXTURE analysis, some of the largest genetic distances are among these peoples. By itself, that may not be surprising; these are the populations that have most evaded the homogenization that followed the spread of farming. The Hadza themselves are strikingly distinctive, and their genetics may reflect a history of small population size during the last several hundred years. The potential for genetic drift in this population was very high. Still, the genetic relations are just the opposite that would be expected if speakers of these click languages had shared a common origin.

Seems to me that this could have been the lede of the paper, if it had been written differently. A bit more exploration of the hunter-gatherer data (probably incorporating some haplotype-level analysis to give a better estimate of the ages of events) would demonstrate this point very well.

3. By the time we find "modern" humans in West Asia, the African population had long since diversified into regional populations. This is not news; the mtDNA evidence has suggested for several years that southern Africa and the remainder of sub-Saharan Africa were already regionally differentiated before 120,000 years ago. There have also been hints of this diversification from whole-genome evidence (including the supplement of the Neandertal genome paper last year). Here we have a clear indication that the regionality extends to every African hunter-gatherer population.

4. Hunter-gatherers have relatively little evidence for recent positive selection. The supplementary data of the current paper includes a short discussion of selection and a list of candidate loci in the hunter-gatherer samples. There is relatively little overlap in candidate regions for selection among these samples. Different genes have been selected in different populations, and not all that many of them. This is not surprising if the selection is relatively new -- the last 20,000 years or maybe more, given the distances and amount of historical population structure estimated for the data. It's also consistent with the demography of these populations. It will be interesting to check, but I would speculate that the signature of selection will on average appear older in these samples than in populations that have historically been agriculturalists.

5. Where's the Aterian? North Africa is relatively depauperate in variation in the large combined dataset. That may stem mostly from Holocene events, including the spread of West Asian populations across North Africa. But the low variation there doesn't readily fit the idea that an out-of-Africa dispersal of genes came from a North African source. I don't think the observations in the paper (centered around linkage disequilibrium with a very low SNP count) are enough to settle anything about this question, but I'd be nervous if I were busy trying to make the Aterian seem important to the modern human origins issue.

Bottom line

As interesting as these assertions look, I don't think that a lot of African prehistory is about to be rewritten. Obviously, geneticists need to get serious about reading some African archaeology. We already know that African regional populations were large and diverse during the Middle Stone Age, and that's a very good fit to the kind of genetic diversity we are seeing in these samples.

The barrier is Holocene population history. Agricultural populations grew, spread, mixed with and absorbed hunter-gatherers, and what we left are the shattered remnants of ancient African population structure. Linkage may be the most powerful way we have to consider historical hypotheses using these SNP data, but if we're going to rely on it we have to control for recent demography and selection.

And of course, it will be interesting to see a model that can integrate both Neandertal-African and within-African population histories. I don't really have a bang-up finish for this post, because there is immediately more work to be done with these data.

References