December 12, 2012

Efficient moment-based inference of admixture parameters and sources of gene flow (Lipson et al. 2012)

My reading list keeps getting longer as another paper referenced by Loh et al. has now appeared on the arXiv, a day after the new Moorjani et al. paper on Romani origins. A number of papers from many of the same co-authors have appeared over the span of a couple of months, all of them containing interesting technical discussion on admixture parameter estimation, so perhaps this is a good place to make a list of them for easy reference:
This series of papers builds on earlier work, which can be found in the following:
The software introduced in the current paper (Lipson et al. 2012) can be found in the MixMapper page, and according to its description it is similar in spirit to the TreeMix software. Hopefully I'll be able to try it out for myself.

The most interesting thing about the current paper is, of course, its detection, to a lesser extent of the same "North Eurasian" ancestry found in northern Europeans by Patterson et al. also in Sardinians and Basques.

Sardinians did not appear to have such ancestry on the basis of the f3-statistic, but this might have been a consequence of the fact that they were the "least unadmixed" of the Europeans, so any application of f3(Sardinian; X, Amerindian) would not have given a negative result, because there does not exist any X less mixed with this Amerindian-like "North Eurasian" element than Sardinians.

Also, the ALDER paper seems not to have been able to date this type of admixture because of its antiquity. I have tried myself using a 1-ref approach on Sardinians (using Sardinians and various other "eastern" populations as possible contributors) but without success. So, it will be interesting to read how this type of ancestry was detected in the current paper. Any further comments will be posted in this space as updates.

UPDATE I: On the left you can see the model proposed for Europe. A first observation is the absence of a primate outgroup, or indeed of representatives of African hunter-gatherers. This makes sense in the context of this paper, since all African hunter-gatherers have been shown now to have admixture from African farmers, so they cannot be used for the "scaffold" tree, as they are not unadmixed.

However, their type of admixture differs from the admixture found in all other populations. For example, Europeans are a mixture of "Ancient Western Eurasians" and a group related to "Ancient Northern Eurasians". African hunter-gatherers, on the other hand, are a mixture between a group related to the Mandenka-Yoruba clade, and (potentially diverse) sets of "Palaeoafricans". The latter are an outgroup to the rest of mankind, and as such admixture with them cannot be represented in this model; consequently Yoruba assume by default a position of unadmixed outgroup to the rest of mankind, a position which -for reasons mentioned before in this blog- I believe is not correct. What effect this might have on the rest of the tree is not yet clear to me.

arXiv:1212.2555 [q-bio.PE]

Efficient moment-based inference of admixture parameters and sources of gene flow

Mark Lipson, Po-Ru Loh, Alex Levin, David Reich, Nick Patterson, Bonnie Berger

(Submitted on 11 Dec 2012)

The recent explosion in available genetic data has led to significant advances in understanding the demographic histories of and relationships among human populations. It is still a challenge, however, to infer reliable parameter values for complicated models involving many populations. Here we present MixMapper, an efficient, interactive method for constructing phylogenetic trees including admixture events using single nucleotide polymorphism (SNP) genotype data. MixMapper implements a novel two-phase approach to admixture inference using moment statistics, first building an unadmixed scaffold tree and then adding admixed populations by solving systems of equations that express allele frequency divergences in terms of mixture parameters. Importantly, all features of the tree, including topology, sources of gene flow, branch lengths, and mixture proportions, are optimized automatically from the data and include estimates of statistical uncertainty. MixMapper also uses a new method to express branch lengths in easily interpretable drift units. We apply MixMapper to recently published data for HGDP individuals genotyped on a SNP array designed especially for use in population genetics studies, obtaining confident results for 30 populations, 20 of them admixed. Notably, we confirm a signal of ancient admixture in European populations---including previously undetected admixture in Sardinians and Basques---involving a proportion of 20-40% ancient northern Eurasian ancestry.

Link

38 comments:

eurologist said...

Makes a lot of sense. The drift length becomes more uniform, as well.

It is interesting how early and central Europeans are placed - but I am still skeptical, and prefer a model where much of Europe instead originates in Pakistan/ NW India - based on existing Neanderthal presence and on uniparental signatures.

Matt said...

The models here are quite different to any which have come before:

- The proto-North Eurasian/American element contributes to the entire Middle Eastern/European group, not only to the European branch (Figure 3).

- The contributions to the European populations by the North Eurasian component (Table 1) seem to be more related to longitude than latitude, especially once leaving aside Russians and Adygei (who have recent mixture), e.g. French have less than Italians and Tuscans, Orcadians slightly less than Basques.

- Is it interesting that the tree does not reproduce Rasmussen et al's result of an ancient split between a proto-Papuan/Melanesian and a Eurasian group, with subsequent mixture donation from/between the East Asian branch of the Eurasian group and the proto-Papuan/Melanesian branch?

Again, how annoying that this HGDP panel lacks South Asian samples....

Václav Hrdonka said...

Very interesting. It make sense. When considering graph on the picture A, I think Kalash and Burusho people would be placed very close to the root of EastAsia and Ancient Northen Euroasia and thus also close to Sardians. Maybe it would be interesting to add them into the graph.

hamarfox said...

Well, the picture's beginning to make a little more sense now. With numbers this high, it's difficult to maintain that the element is inherently Mongoloid, unless someone wants to take up the challenge of arguing Basques and Sardinians are close to a quarter such ancestry.

It reinforces my initital notion that this element is a deeply entrenched and constitutive factor in the formation of modern European types and that archaic ancestry can't always be explained in terms of the features of modern races. It also helps to explain how Europeans remain 'glued' to one another in PCAs and Fst figures: all have this element in common, except some have slightly more than others.

It also accords with physical anthropology: notably the fact that the only qualitative difference in Europe in regards to East Eurasian types is between East and West, not North and south as such. The reason is likely that in regard to the old East Eurasian element, it's everywhere, with only slight quantitative differences within Europe. On the other hand, there are qualitative differences (i.e.in terms of modern Siberian admixture) between East and West, hence the types noted by physical anth.

---------

Matt, West Asians sharing this element with Europeans goes a long way to explaining the relatively close clustering of West Asians with Europeans. If they lacked it, it's hard to see why, with up to 40% genetic alienhood, they'd be so close/similar to Europeans in PCAs/Fst. I'd say the Southern component is the most purely West Eurasian, but modern examples of the Southern component are too heavily SSA admixed to serve as good proxies.

Matt said...

With numbers this high, it's difficult to maintain that the element is inherently Mongoloid, unless someone wants to take up the challenge of arguing Basques and Sardinians are close to a quarter such ancestry.

That's an interesting point.

If we're reading the graph in Figure 3 as accurate, the divergence of the Native American/North Eurasian branch from the East-Southeast Asian (Japanese-Dai) branch predates the contribution from the Native American/North Eurasian branch to the European/Middle Eastern branch.

Presumably whatever Mongoloid features are shared by the Native Americans (e.g. Karitiana) and East-Southeast Asian* (e.g. Japanese) would have been in place prior to their divergence (which remember is prior to the contribution to the European/Middle Eastern branch) or occur due to convergent evolution. Convergent evolution seems like it would not be the null hypothesis in this situation, although possible (it seems like the East-Southeast Asian branch would have inhabited a different Eurasian zone than the Native American/North Eurasian branch [or else why would population structure be present in the hunter gatherer era?], and it is difficult for me to see what shared selective pressures these two different zones could share so strongly which would also not be present in West Eurasia).

Reconciling this with the fossil record is difficult, , like you say (i.e, there is not, I think, a sign of populations in West Eurasia who were extremely more anti-Mongoloid than present day West Eurasians in the West Eurasian fossil record, and who could plausibly have mixed with a Mongoloid group to generate a West Eurasian-ish phenotype)..

*North Asian populations seem more complex under this model, as even the Yakut has a estimated contribution of around 40% East-Southeast Asian and they have the highest % contribution from the Native American/North Eurasian branch of all the extant North Eurasian populations tested.

Onur said...

It is not yet so clear how we should interpret the MixMapper, ADMIXTOOLS and TreeMix results of modern-day Europeans and Caucasoids in general. All I can say for now is that the ADMIXTOOLS and TreeMix results have more concurrence with each other as well as with physical anthropology.

But, as Matt points out, it is very clear that the line ancestral to the East-Southeast Asian branch and the Native American/North Eurasian branch must have already belonged to the Mongoloid physical type. Alternative explanations are implausible, as the Mongoloid physical type is too complex to have emerged more than once through convergent evolution. The proportions of the Native American/North Eurasian type ancestry in various Caucasoid populations may be open to debate, but, if the ADMIXTOOLS, TreeMix and MixMapper result of non-negligible levels of Native American/North Eurasian type ancestry in most modern-day Caucasoid populations is to be trusted, it is certain that the Native American/North Eurasian type population in question belonged to the Mongoloid physical type.

hamarfox said...

It is not yet so clear how we should interpret the MixMapper, ADMIXTOOLS and TreeMix results of modern-day Europeans and Caucasoids in general. All I can say for now is that the ADMIXTOOLS and TreeMix results have more concurrence with each other as well as with physical anthropology.

Well, we've discussed this in the past. These latest results are unintuitive indeed, because modern Caucasoids wouldn't be the result of admixture between modern Amerindians (especially when such admixture is as high as 25% or more) and any hypothetical ultra-Caucasoid, lacking such admixture, that we can imagine. However, since Western Europeans also wouldn't result, intuitively, from admixture between said populations (or with Sardinians as surrogate for 'pure West Eurasians'), the intuitive discordance exists with both sets of findings, and therefore these new results don't jar with phys anth. any more than did previous ones.

Also consider that it would be, as far as I understand, difficult (or impossible) for all modern West Eurasians to cluster together on PCA analyses if modern 'West Eurasians' lacked significant portions of their lineage in common along which to share drift. In short, this deep ancestral difference would change the tangent of drift of Europeans and ME populations in relation to one another. We'd, in this case, have to consider that drift 'ignored' 20-40% of Europeans' ancestry to continue drifting in the same direction as West Asians, and then that somehow Europeans actually 'overtook' other West Eurasians in drift away from modern East Eurasian populations, and along essentially the same trajectory.

Similar fundamental ancestral contributions to all modern West Eurasians of ancient West and North Eurasian populations, plus minor/moderate modern contributions to certain populations (e.g. Siberian in Finns and Russians, ASI in the Middle East, some SSA here and there etc.) brings most genetic measures and physical anthropology into greater harmony, IMO.

Dienekes' ADMIXTOOLS work is excellent, btw. I'm sure he nailed the ordering of most-to-least of this element in Europe, though keeping in mind that the D-stats for any population are heavily influenced by recent East Asian and SSA. What these findings do, IMO, is broaden the scope of the overall process -- though lacking the finesse of a more concentrated analysis, hence the failure to reproduce Dienekes' ordering of populations exactly.






princenuadha said...

@onur

Couldn't this North Eurasian branch be minimally mongoliod while the native Americans are more mongoliod because of mixture with the eastern lineage post north eurasian divergence?

In other words, the native Americans are the result of this north eurasian, shared with westerners, and a more east eurasian type.

hamarfox said...

But, as Matt points out, it is very clear that the line ancestral to the East-Southeast Asian branch and the Native American/North Eurasian branch must have already belonged to the Mongoloid physical type. Alternative explanations are implausible, as the Mongoloid physical type is too complex to have emerged more than once through convergent evolution. The proportions of the Native American/North Eurasian type ancestry in various Caucasoid populations may be open to debate, but, if the ADMIXTOOLS, TreeMix and MixMapper result of non-negligible levels of Native American/North Eurasian type ancestry in most modern-day Caucasoid populations is to be trusted, it is certain that the Native American/North Eurasian type population in question belonged to the Mongoloid physical type.

This is sound logic, I agree. But the multiple waves of settlement of the Americas (not detected in this paper) could explain why North Eurasian populations (in the Americas) increasingly became East Eurasian (as distinct from North Eurasian for the sake of argument) in appearance. The plurality of physical types in the Americas lends itself to such an interpretation.

But there's also another possibility, I think. A silly analogy: Imagine you have two wobbly wind-up toys. You set them off from the same position (a common proto-Mongoloid ancestor) and then let them free. They wobble about, one goes faster than the other, they both veer off a linear path a bit here and there. But while they end up in slightly different places, it's clear that they both went in the same direction and ended up generally in the same place.

But would it be correct to say that because they both ended up in roughly the same place that they must have started the 'race' in the same spot they began, because the alternative is that they started off in random places and coincidentally ended up in roughly the same place? No. They both realised tendencies inherent to both at the beginning (i.e. the direction they started from) over time.

Now let's say each toy gets heavier with distance (i.e. features become more extreme). Let's also say there's a third toy (the ancestor of West Eurasians). He starts off in the same direction, but before he gets too far he's plucked from the race by somebody (the West Eurasian gene pool) and twisted in a new direction, which is easy, because the toy hasn't been in the race long enough to become particularly heavy.

He ends up in a different place now, since the toy's trajectory was completely changed (by introduction into another, proportionally slightly larger population). Now it gets heavier and heavier in a fresh direction such that at the end of the race, the 'weight' it gained in the race with the original ducks pales in contrast to the weight it accrued in the second race.

The fact that Amerindians and East Asians may have 'extreme' features, and that these may be extreme in similar ways doesn't necessarily mean that the features were always present in the common ancestor, but only the incipient tendencies of such features, and that these features were countered or coopted uniquely upon introduction with a physically not-quite-known ancient West Eurasian.

princenuadha said...

So the split between "ancient west eurasians" and the path that includes the eventual east asians was sooner that the split between the path of eventual papuans and the path including eventual asians! Very interesting considering how distant papuans are.

Anyways, its very cool that we have finally gotten insight into the ancient European admixture that was talked about a year ago.

Matt said...

Onur, one thing to bear in mind that the team here have confirmed the admixture in Sardinians and Basques using ADMIXTOOLS f4 statistics, with a proportion that is in agreement. See Table S1.

"To validate the mixture proportions estimated by MixMapper for Sardinian and Basque, we applied f4 ratio estimation. The fraction of “ancient northern Eurasian” ancestry was estimated as = f4(Papuan, Asian; Yoruba, European) / f4(Papuan, Asian; Yoruba, American), where the European population is Sardinian or Basque, Asian is Dai or Lahu, and American is Karitiana or Suru´ı."

So I would be wary about viewing it as a ADMIXTOOLS & Treemix "vs" Mixmapper scenario.

Dienekes (although he is obviously under no obligation to do so) might be able to comment on why their method of f4 ratio estimation, has found this signal while his f4 statistics and ratios were unproductive for this.

Unfortunately, I don't understand the methods well enough to do so. It's not clear to me why this is the case at all - and of course the MixMapper ancestry proportions don't really support a Sardinian / Basque as less admixed scenario.

Regarding the timing, although I did say that, the authors themselves do note -
"At first glance, this inferred admixture might appear improbable on geographical and chronological grounds, but importantly, the two ancestral branch positions do not represent the mixing populations themselves. Rather, there may be substantial drift from the best-fit branches to the true mixing populations, indicated as branch lengths a, b, and c in Figure 4A."

hamarfox said...

Just to be a little clearer, the 'weight' in my analogy was a metaphor for differentiation, and is thus only relative to other populations. The weight would be set to zero each time two populations split from a common ancestor. The heavier the weight two populations are in relation to each other (differentiated from a common ancestor), the harder it is for the characteristics of the one to disappear into those of the other upon a second contact. North and West Eurasians were lighter in relation to each other then than now.

hamarfox said...

Just to be a little clearer, the 'weight' in my analogy was a metaphor for differentiation, and is thus only relative to other populations. The weight would be set to zero each time two populations split from a common ancestor. The heavier the weight two populations are in relation to each other (differentiated from a common ancestor), the harder it is for the characteristics of the one to disappear into those of the other upon a second contact. North and West Eurasians were lighter in relation to each other then than now.

Onur said...

Hamar and Nuadha,

According to Matt's calculations based on the results of the Lipson et al. paper and taking into account Negroid admixture, even Arabic-speaking West Asians have Native American/North Eurasian type ancestry in levels little below those of Sardinians and Basques and North Africans have such ancestry in levels below, but not much below, those of Arabic-speaking West Asians (the difference is obviously due to the elevated levels of Negroid ancestry in North Africans compared to Arabic-speaking West Asians):

http://eurogenes.blogspot.com/2012/12/europeans-are-three-way-mix-with.html?showComment=1355602786428#c1349641342071295415

That means that Sardinians and Basques have Native American/North Eurasian type ancestry in virtually equal levels to those of non-recent Mongoloid-admixed and non-Negroid-admixed West Asians. Non-Negroid admixed modern Caucasoids have a minimum of about 25% (the level of Sardinians) Native American/North Eurasian type ancestry according to the results of the Lipson et al. paper. So what we term as the ideal Caucasoid based on modern-day populations may actually be about 3/4 descended from an ultra-Caucasoid race, which no longer exists in pure form, and about 1/4 descended from the Native American/North Eurasian variety of the Mongoloid race. This does not conflict with physical anthropology, as the ideal Caucasoid physical type was defined based on modern-day Caucasoids.

Matt,

My criticism of the MixMapper results of Europeans has nothing to do with the highness of the levels of Native American/North Eurasian type ancestry but rather to do with the intra-European differences in the levels of such ancestry. For instance, according to the MixMapper results Tuscans have more Native American/North Eurasian type ancestry than the French and Basques, who in turn have more such ancestry than that of Orcadians, who in turn have such ancestry in equal amount to that of Sardinians, which is the opposite of the ADMIXTOOLS and TreeMix results and conflicts with physical anthropology. It seems MixMapper has some problems in calculating intracontinental differences in ancestry.

Matt said...

My criticism of the MixMapper results of Europeans has nothing to do with the highness of the levels of Native American/North Eurasian type ancestry

Sure, I didn't think that was the case.

which is the opposite of the ADMIXTOOLS and TreeMix results and conflicts with physical anthropology

I am only making the point that it's not "Oh, MixMapper is saying this thing that is in contradiction to Admixtools and TreeMix" because the authors have user Admixtools in the paper to actually confirm their result of the level in Sardinians.

It is a shame they didn't rerun this on all their European populations though, to confirm levels - we know the Sardinian level they have found through MixMappers is matched by Admixtools, but...

I would just present this more like a current Admixtools analysis + MixMapper says one thing and older Admixtools analyses say another.

I actually don't remember TreeMix confirmed any contributions from a Native American / North Eurasian population to any European populations in a tree... TreeMix found directionality from Orcadian to Native American, and even when Dienekes examined his ADMIXTURE components, the direction was still North European component -> Siberian component.

I still don't know enough about the trends in intra-European facial and cranial shape to feel confident about that, particularly what you get when you leave out recently admixed Northern Europeans. I am only really properly aware of a broad trend to increased brachycephalisation and facial width in North and South Eastern Europe and Turkey relative to Western Europe. I don't think there is any widespread consensus that Irish (for example) have higher cranio-facial similarities to East Asians than Basques do.

hamarfox said...

So what we term as the ideal Caucasoid based on modern-day populations may actually be about 3/4 descended from an ultra-Caucasoid race, which no longer exists in pure form, and about 1/4 descended from the Native American/North Eurasian variety of the Mongoloid race. This does not conflict with physical anthropology, as the ideal Caucasoid physical type was defined based on modern-day Caucasoids.

You make a fair point, and one that I myself have considered. However, this (perhaps along with the notion that West Africans may be significantly out-of-Africa admixed) ranks as the least intuitive population admixture we know of. The Mongoloid/Australasian mixes of SE Asia, the ASI/ANI admixtures of India, the SSA/West Eurasian mixtures of North Africa, and the Siberian/West Eurasian mixtures of parts of NE Europe, for example, are all fairly obvious and were spotted by physical anthropology long before population genetics confirmed the mixtures.

The point is that, referring back to my analogy, all of those populations had gained sufficient weight relative to one another for neither to really dominate or co-opt the other, while the much earlier admixtures of this W/N Eurasian mixture seems to have occurred at a stage of relatively minimal differentiation of N and W Eurasians from their common out of Africa ancestor, and the bulk of the N Eurasian component that was absorbed into the early West Eurasians underwent more evolution in step with the West Eurasians it was now a component of than it did with the early proto-Mongoloids.

This same process didn't occur with the other, more recent mixes mentioned that are much more intuitive, presumably because in those cases the two components 'travelled' for a shorter time together than they travelled apart (since their common out of Africa ancestor).

This is my theory as to why it's not obvious, since regardless how a pure West Eurasian may look, it shouldn't stop the 25% Mongoloid (if such it is) being obvious.

hamarfox said...

My criticism of the MixMapper results of Europeans has nothing to do with the highness of the levels of Native American/North Eurasian type ancestry but rather to do with the intra-European differences in the levels of such ancestry. For instance, according to the MixMapper results Tuscans have more Native American/North Eurasian type ancestry than the French and Basques, who in turn have more such ancestry than that of Orcadians, who in turn have such ancestry in equal amount to that of Sardinians, which is the opposite of the ADMIXTOOLS and TreeMix results and conflicts with physical anthropology. It seems MixMapper has some problems in calculating intracontinental differences in ancestry.

Orcadians likely have more of the element by virtue of having a greater proportion of mesolithic European ancestry compared to the other populations listed, and then the French come next, and so on, but what in physical anthropology places Orcadians and especially French (and I say 'especially' since it's conceivable that Orcadians have a few Siberian influences by way of Scandinavia -- though none that I've seen) are more Asiatic than Basques or Sardinians?

Onur said...

Matt,

TreeMix did indeed show northern Mongoloid (i.e., Native American/North Eurasian variety) ancestry in modern-day Caucasoids (especially Europeans) with pretty similar results to those of ADMIXTOOLS:

http://dienekes.blogspot.com/2012/08/4-population-test-and-east-eurasian.html
http://dienekes.blogspot.com/2012/08/east-eurasian-like-admixture-in.html
http://dienekes.blogspot.com/2012/08/east-eurasian-like-ancestry-in-northern.html

Compare with the ADMIXTOOLS qpF4Ratio test results:

http://dienekes.blogspot.com/2012/09/quantifying-karitiana-like-admixture-in.html

Hamar and Matt,

Southern Europeans show more distinctively Caucasoid physical traits than both northwestern and northeastern Europeans. This can be deduced even by casual observation. As for the French, especially the southern French, they are transitional between northern and southern Europeans.

eurologist said...

Concerning the "Mongoloid admixture conundrum", I would like to make a few points.

Firstly, Native Americans are not homogenous. In much of America, Mongoloid features are rather sparse - especially in parts of the South but also in Middle- and North America. Next are Na-Dene speakers, who seem to have a higher rate but still highly variable Mongoloid features. Inuits top it off. Note that the Surui and Karitiana used in this study are from Brazil, some of who could pass as Pacific Islanders or generic US "Hispanics."

So, this is consistent with a model where ancient NE Asians -if at all- only had emerging Mongoloid features in part of the population. Climate change and later northward expansion of full-fledged Mongoloid populations then displaced or absorbed most of the ancient NE Asian population except in the far west and far east (Beringia). Such a cline in NE Siberia and Beringia would then be reflected in the populating of the Americas.

In fact, in this paper, extant N Asians are indeed a mixture of ancient NE Asians and more recent (presumably fully Mongoloid) NE Asians, as would be expected from the above model.

Also note, for all Europeans (and additional West Eurasians) to have rather similar ancient NE Asian contribution, this admixture must have taken place in the 45,000 - 28,000 ya time frame - definitely before the beginning of the Gravettian. A IMO good scenario would involve people back-migrating from NW India and out of Pakistan/Afghanistan into the Caspian/Pontic steppes, and there mixing with ancient NE Asians who had migrated to the same area via the Altai. If they both followed grazing animals and expanded west at about the same time (advantageous climate), there would have been little competition, and a peaceful joining is imaginable in a land that then seemed to have endless resources.

Matt said...

It occurs to me that a useful sense check against timing might be uniparental phylogeny.

I don't know much about that, but for anyone who is an expert, does any of this make any sense when matched against this?

eurologist said...

I would like to add to the above that there are numerous Upper Paleolithic sites dated in the 40,000 - 45,000 ya range in Siberia, later extending all the way to the arctic coast line (Yana RHS site, ~32,000 ya).

So, flow of such people westward between the earliest arrival of AMHs in Europe (~43,000 ya) and the Gravettian seems more than plausible.

When I picture "ultra Caucasian" early Europeans, some Baltic or German faces come to mind: image-google, e.g., Kristina Schröder, Maria Höfl-Riesch, and Franka Potente. I think you could admix 25% proto-Mongolian into that and come up with modern Caucasians, no problem. (Also, is there something wrong with me that I find the above rather attractive?) ;)

So the split between "ancient west eurasians" and the path that includes the eventual east asians was sooner that the split between the path of eventual papuans and the path including eventual asians! Very interesting considering how distant papuans are.

princenuadha,

Fig. 3b says otherwise, at least partially (Papuans are the first to split off on the "solid tree." If all of Eurasia got populated at once, it makes sense that people in the Pakistan/W India area split before Papuans. So, one branch of the later-mixed Caucasians split off very early.

However, the picture might look yet a little different had the authors included Indian populations. Perhaps there was more than one migration, in which case Papuans might be admixed from a truly early migration (not captured here) and the "main event." This could possibly be tested by including a different (and smaller) set of populations (e.g., Africa, W India, N Asia, and Papua) and not fixing Papua on the tree a priori.

hamarfox said...

Matt,

It is a shame they didn't rerun this on all their European populations though, to confirm levels - we know the Sardinian level they have found through MixMappers is matched by Admixtools, but...

Indeed. According to previous measures, Basques had an excess of around 10% of the element vs. Sardinians, while now they're roughly equal. It's hard to say now whether other populations will be set back ~10% in reference to Sardinians. Certainly, it's hard to reconcile the Orcadian excess (vs. Sardinians) of ~18%, in some calculations, with the near equality in levels found by MixMapper.

PCAs support intra-European near-equality of the ancient element, since smaller amounts of recent admixture create greater shift than the (largest) inferred differences in ancient levels within Europe (compare differences in Asian shift between SW and NW Europeans vs. NW and NE Europeans: http://tinyurl.com/breuf4u).

However, given that recent admixtures involve populations presumably more differentiated from modern Europeans than ancient North Eurasians were from ancient West Eurasians, it's not an exact science. Drift is also possible, though that would likely result in exaggerating differences within Europe (thus making the argument for initial near-parity in ancient N Eurasian levels stronger).

I still don't know enough about the trends in intra-European facial and cranial shape to feel confident about that, particularly what you get when you leave out recently admixed Northern Europeans.

I'd formerly entertained that Coon's UP types fit the criteria for describing the distribution of this ancient N. Eurasian element in Europe: they were present everywhere, though more common in the north than in the south. They were also pretty much Europe-specific (though there's a continuity into North Africa), and hence explained why the signal wasn't really found beyond Europe.

But with these new findings, that now seems less tenable. At best, we probably only have a slight excess of the signal in mesolithic Europeans compared with other West Eurasians.

Regardless, it's without question that recent Asian admixture has introduced different ranges and frequencies of traits than were introduced by these ancient N. Eurasians. Even faintly East Eurasian phenotypes, such as Carly Rae Jepsen, are by and large absent in Western Europe (excluding Scandinavia, obviously).

I am only really properly aware of a broad trend to increased brachycephalisation and facial width in North and South Eastern Europe and Turkey relative to Western Europe. I don't think there is any widespread consensus that Irish (for example) have higher cranio-facial similarities to East Asians than Basques do.

Agreed. The non-typical Caucasoids (in Onur's opinion) of mesolithic-surviving Irish look like Brendan Gleeson. I don't see anything intuitively Mongoloid, or even proto-Mongoloid about that.

hamarfox said...

When I picture "ultra Caucasian" early Europeans, some Baltic or German faces come to mind: image-google, e.g., Kristina Schröder, Maria Höfl-Riesch, and Franka Potente. I think you could admix 25% proto-Mongolian into that and come up with modern Caucasians, no problem. (Also, is there something wrong with me that I find the above rather attractive?) ;)

Honestly, to my (British) eyes, all three look a bit East Eurasian shifted to me. It's a certain look I see in central Europeans that I don't see elsewhere, certainly not anywhere too far west. I always reasoned that since those traits are present in populations that aren't significantly Asian shifted relative to my own, it's not actually Asian influenced, and instead just a local central European type, but I definitely don't see them as a less Asiatic type, compared to which the average European (even NW European) is (proto-)Mongoloid-shifted. To me, a Caucasoid European is Andy Garcia (Cuban born, but obviously not mixed) or, when more northern, Mark strong or Jenson Button. All three, however, have some degree of mesolithic traits.

I imagine the admixture predates the emergence of Europeans or 'white people', and most/all physical types we're familiar with in modern west Eurasia were likely painted on a canvas that was already a patchwork of very early N. and W. Eurasian. This explains why there's no intuitive physical shift 'eastward' between any two West European populations, regardless of levels between them of the ancient element, while tiny ~2% levels of recent East Eurasian admixture in central and eastern Europeans are easily discernible to a NW European.

eurologist said...

hamarfox,

As I mentioned above, I also think that we are talking about a very early admixture here between proto-populations ~30,000 - 40,000 ya. Your examples seem Mediterranean-admixed to me and also somewhat extreme (e.g., Scottish-Irish) NW drifted. It is probably almost futile trying to demonstrate a proto-Caucasian face in extant populations (although they may very well exist).

Also note that he faces I linked to are pretty much opposite to what traditional Mongoloid features are - who have: broad/wide faces, low cheek bones, shallow and wide nasal bridge, small eyes, epicanthic fold, etc.

hamarfox said...

As I mentioned above, I also think that we are talking about a very early admixture here between proto-populations ~30,000 - 40,000 ya.

I hope my last comment didn't come across as my disagreeing with most of what you've said. A lot of the same thoughts have occurred to me, including the possible significance of the formerly much wider distribution of the European upper paleolithic population. But a major check for me against that line of interpretation is the now known West Eurasia-wide presence of this ancient N. Eurasian element, and in approximately even levels for all non-recently-admixed populations. It's almost analogous to the Neadertal problem: it's fairly even for all out of Africa populations, suggesting a major admixture event before the splitting up of OoA populations from a last common ancestor, but with some minor subsequent admixture events occurring in pockets.

It seems that mesolithic Europeans had the element at slightly higher levels than, say, the source population(s) of neolithic immigrants to Europe, and this probably explains the fairly slight surplus of the element in European populations that are proportionally more mesolithic in descent. But -- and correct me if I'm wrong -- I don't think there's been too much genetic exchange between European UPs and West Eurasians beyond Europe, except maybe into North Africa and the Caucasus region, so, for now, I'm thinking the admixture occurred closer to the root of all West Eurasian populations.

Your examples seem Mediterranean-admixed to me and also somewhat extreme (e.g., Scottish-Irish) NW drifted. It is probably almost futile trying to demonstrate a proto-Caucasian face in extant populations (although they may very well exist).

Well, I sort of agree with Onur that 'typical' Caucasoids are the gracile types ultimately from the fertile crescent, and likely introduced to Europe during the neolithic, as against the more brachycephalic mesolithic natives. I don't agree with him, however, that deviation from 'typically Caucasoid' necessarily implies Mongoloid admixture. So that's really why I chose the examples I did, rather than because I think other physical expressions are inherently less West Eurasian.

As for looking for the effects of said admixture in modern European (and MENA) populations, I'd say, since we don't know exactly how anciently the mixture occurred, it's possible modern Europeans look more different from their N. Eurasian ancestors than the unadmixed W. Eurasians did from the N. Eurasians they interbred with, considering the multitude of forms that bloomed from an OoA common ancestral group, possibly within the same time frame as has passed sine the W/N Eurasian mixture in question, and the tendency for this to occur over long periods of relative isolation.

hamarfox said...

Also note that he faces I linked to are pretty much opposite to what traditional Mongoloid features are - who have: broad/wide faces, low cheek bones, shallow and wide nasal bridge, small eyes, epicanthic fold, etc.

It's interesting, because I'd noticed the same appearance in some Germans, and I'd always thought to myself that, though most Germans and Britons are essentially interchangeable in appearance, this particular combination of traits was something specifically German, so I suppose those examples you used conflicted with my intuition.

Still, I contend that the average German looks at least no more 'E. Eurasian' than the examples you used. It's a good idea, I think, to test our notions of what is and isn't natively European (a notion surely all Europeans have) against samples from groups most insulated from non-European/West Eurasian geneflow.

These Basques represent my intuition of what unadmixed Europeans essentially look/looked like (though, of course, there's going to be a bias in favour of whatever region of Europe you select for comparison, in this case SW Europe):

http://tinyurl.com/d6axbde

http://tinyurl.com/d96cvyr

Not pigmentally, but metrically, I think they represent Europeans with mesolithic and neolithic ancestry, but as little N. Eurasian (modern or ancient) as any extant W. Eurasian group.

Matt said...

I don't know about a hyper-Caucasoid face, exactly, but taking a (I think mostly) NW European composite female face (the first link below) and exaggerating about 25% in a direction away from an East Asian composite female face (the second link below) using the tools at http://faceresearch.org/demos/transform yields the third composite face.

1. http://faceresearch.org/demos/images/avg_res/ethnicity/female/eastasian
2. http://faceresearch.org/demos/images/avg_res/ethnicity/female/white
3. http://oi50.tinypic.com/xprmuo.jpg

As applied to a couple of real West Eurasian female face

- http://faceresearch.org/demos/images/avg_res/individuals/124_03
- http://oi48.tinypic.com/19rwm.jpg

and

- http://faceresearch.org/demos/images/avg_res/individuals/001_03
- http://oi45.tinypic.com/2p1xe.jpg

The changes seem comparatively subtle to a modern eye (e.g. compared to the difference between the average face from even intra-European regions), leaving aside that probably neither of these faces is exactly going to be a proxy for the ancient populations. There are at the very least differences between the "pure" North Eurasian postulated pop and present day North Eurasians, let alone "pure" East Asians, and that a lot of the pigmentation evolution at least would obviously has happened subsequent to any gene flow - the East Asian pigmentation lightening genes are not present in NW Europeans or populations who have not recently mixed with East or North Asians.

eurologist said...

hamarfox,

A related question is whether typical Baltic features are a drift that occurred in that area, perhaps retaining more ancient NE Asian contribution, or from Uralic/ N Asian admixture, or a combination of the two. Baltic features are quite common in Germany, as well, even away from the Baltic:

http://upload.wikimedia.org/wikipedia/commons/4/48/Alexandra_Popp_2011.jpg

So, people with such features (and often epicanthic folds) certainly look more Eastern - but it is not clear to me what the primary cause is (of the list above).


Matt,

Very interesting - and truly subtle. The differences I can perhaps discern are into the direction of more robust, bigger chin, higher and more pronounced cheek bones, and wider and bigger eyes - all features of the faces I selected above. Also, Mongoloid features are largely essentially childlike retained (neoteny). If you have time, perhaps you could redo your experiment with a European who looks already more robust and less "roundish", and a Mongoloid face that looks less southern Asian, such as these (I know, they may not be particularly useful - just examples):

http://transformsiberia.files.wordpress.com/2010/10/dsc_0062.jpg

http://idugan.com/images/moilkhon/07.jpg

Onur said...

Matt, your composites are based on existing humans. They can not inform us about the physical appearance of no longer existent human types such as the hypothetical ultra-Caucasoids.

Matt said...

eurologist: Mongoloid features are largely essentially childlike retained (neoteny).

Yeah, I'm agnostic on this. Seems kind of plausible. Many features such as a relatively broad face, relatively flat maxilla surface, small/narrow mouth and maybe relatively small mandible (compared to maxilla) are shared between children relative to adults and Mongoloids relative to Caucasoids.

Not true for everything though, e.g. maxilla internal features, in terms of the typically relatively narrow and small orbits and tall narrow nose of Mongoloids vs relatively broad orbits, large orbits and broad, short nose of children are opposite.

Neoteny also seems to apply more to facial shape than facial size (maxilla + mandible) relative to the neurocranium, which looks typical.

If you have time, perhaps you could redo your experiment with a European who looks already more robust and less "roundish", and a Mongoloid face that looks less southern Asian

Not really got time, but if you want to track some composites, I'm sure there are relevant composites of Mongolians and the like online, and the software is actually very easy to use (basically spend about 5 mins placing landmarks on a couple of faces)!

eurologist said...

Matt, your composites are based on existing humans. They can not inform us about the physical appearance of no longer existent human types such as the hypothetical ultra-Caucasoids.

Onur,
I disagree: those experiments certainly suggest that a 25% admixture of proto-Mongoloids may have left rather subtle signatures. In the real world, in California you can see many people who are 25% (of course extant) NE Asian and 75% European, and with some it is evident, with others it is not noticeable (which is about what you would expect). Much less diverged proto-Mongoloid into proto-Caucasian truly may have left few signatures, but they would have widened the gene pool and thus allowed drift and adaptation to later form more different types per region.

Matt: sure, cold adaptation seems also at play. You can also see that in the rounder heads and thicker skulls and hypodermis of people with Baltic features.

eurologist said...

Matt,

I played around with the stored images at http://faceresearch.org/demos/transform . I used the averaged Asian (#3) in the "-" field (bottom left), and the pretty one [insert movie reference] #5 individual and the composite white #5 alternatively in the "+" (bottom right) and "to transform" (top left) quadrants. Using something like 50%-80% on the shape and color gives the more robust Caucasoid one would expect, rather similar to the examples I provided earlier.

I am no longer sure that the percentage scales anything like admixture, though.

Onur said...

Onur,
I disagree: those experiments certainly suggest that a 25% admixture of proto-Mongoloids may have left rather subtle signatures. In the real world, in California you can see many people who are 25% (of course extant) NE Asian and 75% European, and with some it is evident, with others it is not noticeable (which is about what you would expect). Much less diverged proto-Mongoloid into proto-Caucasian truly may have left few signatures, but they would have widened the gene pool and thus allowed drift and adaptation to later form more different types per region.


The inferred ancient Mongoloid admixture in all extant Caucasoids is not very ancient, as it is inferred to be from a time after the split of Mongoloids into the Native American/North Eurasian and East-Southeast Asian branches, thus from a time Caucasoids and Mongoloids were already well diverged both genetically and morphologically. Also, the composite maker tool in Matt's links uses extant populations and ignores all the morphological changes that have occurred over the past millennia. You say that a quarter extant Mongoloid admixture in extant Caucasoids is sometimes noticeable and sometimes not, but in all of the composites that Matt provided as examples to the hypothetical ultra-Caucasoids there is no noticeable difference from extant Caucasoids. That is implausible.

Matt said...

in all of the composites that Matt provided as examples to the hypothetical ultra-Caucasoids there is no noticeable difference from extant Caucasoids. That is implausible.

Well, there are noticeable differences - eurologist has described them. The third image I generated in the first set is the first (NW European Caucasoid composite) image transformed away (by the face transformer, not really a composite maker) from the second image (East Asian composite) by -25%. It just is - whatever you think about how similar it looks.

The same algorithm that transforms this face a) http://faceresearch.org/demos/images/avg_res/individuals/124_03 to this b) http://oi45.tinypic.com/16ifywo.jpg also transforms a) to c) http://oi47.tinypic.com/24ymn3l.jpg.
The changes a)->b) and a)->c) are of the same magnitude and along the same vector (in opposite direction). If they don't look of the same magnitude to the beholder....

I fully agree that the transformed composite gives us only a weak idea about how a hypothetical pre-North Eurasian gene flow ancient population would look.

Firstly, because these are evolutionary changes specific to and shared between populations, as I've acknowledged (which is persuasive at least in pigmentation).

Secondly, because these are only related proxies even if we disregard the first point.
Also because of composites and transforms do only give a weak idea about how populations vary in their look - populations do not only vary in terms of their mean consensus face, but in terms of their patterns of variation and covariation in traits.

For example, Europeans often vary in hair and eye colour, other populations largely don't. East Asians vary in the presence of the epicanthic fold, other populations largely don't. South Asians probably vary more in terms of skin colour than Europeans, even locally. Some populations have more variation in BMI and size than others, etc.

E.g. transforming a blond haired European by the +100% of the difference between a European average and East Asian average does not generate a person which plausibly could exist in an real world Asian population. This is true for a whole host of traits, albeit probably more weakly (but with cumulatively large effect).

(and this kind of difference in patterns of variation is a measure of population differentiation - in theory two populations who cohere to relatively similar composite faces could be more different than two populations who cohere to very different composite faces, if the patterns of intrapopulation variation are more different between the first pair).

And finally also genetic mixing is not blending or transforming, and there may be dominance effects which would not be captured by composites. Traits are not all additive genetic variance.

So it's just a little example to take with a grain of salt, based on the differences between the extant populations.

Onur said...

Well, there are noticeable differences - eurologist has described them. The third image I generated in the first set is the first (NW European Caucasoid composite) image transformed away (by the face transformer, not really a composite maker) from the second image (East Asian composite) by -25%. It just is - whatever you think about how similar it looks.

I know. In my previous post I said, "in all of the composites that Matt provided as examples to the hypothetical ultra-Caucasoids there is no noticeable difference from extant Caucasoids", not from their untransformed Caucasoid forms. The thing is, all the transformed Caucasoid composite forms that you generated to be examples to the appearance of the hypothetical ultra-Caucasoids look typical extant Caucasoid, just like all the untransformed Caucasoid composite forms. That is what I called implausible in light of the facts that I mentioned in my previous post and you now elaborate in much of the rest of your post that I quote.

eurologist said...

Onur,

Are you sure you have looked at all the images or the ones I guided to (it takes only a few seconds to repeat the steps I outlined)? There is no doubt that at 50-80% in the tool (which, as I stated earlier, surely does not equate admixture in a linear way) you start seeing faces that are way more robust than the 95th-99th percentile of extant Caucasians. Similar to the images I posted earlier.

Thing is, among hundreds of millions of people, you would expect such extreme faces and signatures to be part of the extant population, as I mentioned earlier.

Of course, all this does not say very much except that proto-Caucasians may have been on the robust side with large and widely-spaced eyes (which there is evidence of), and admixture of proto-NE Asians would have reduced this on average but also would have widened the gene pool such that extant populations are broader and also may have drifted and adapted in either direction in certain localities.

Onur said...

Eurologist,

I have exercised a bit with the face transformer and, yes, I have repeated your and Matt's exercises. But I, and Matt, have also already addressed your claims and replied to them adequately, so there is no need to repeat my answer.

Seinundzeit said...

"The models here are quite different to any which have come before:

- The proto-North Eurasian/American element contributes to the entire Middle Eastern/European group, not only to the European branch (Figure 3)."

This is incorrect. I talked to David Reich, and he said they think Middle Eastern populations(such as the Bedouin)lack this North Eurasian-Amerindian admixture.