D-stats/4mix tour of ancient Eurasia

This 4mix experiment is based on a series of statistics of the form D(Chimp,Reference_pop/Test_pop)(Mbuti,X), where X represents one of 9 ancient and present-day outgroups. The input data is available here. Feel free to try it yourself and post your models in the comments below.

Here's a Principal Component Analysis (PCA) based on the D-stats. As far as I can see, it makes very good sense. Click to enlarge.

See also...

Yamnaya = Khvalynsk + extra CHG + maybe something else

PC/nMonte open thread

CHG admixture in early western Anatolian farmers

Anatolian Neolithic farmer I0708 from the Mathieson et al. 2015 dataset belongs to Y-haplogroup J2a and is the most Caucasus-shifted of the early Anatolian farmers in my Principal Component Analysis (PCA) of West Eurasia (see below). This is unlikely to be a coincidence and provides strong evidence that at least some Neolithic farmers in western Anatolia harbored Caucasus Hunter-Gatherer (CHG) ancestry.

Note that the two CHG genomes sequenced to date courtesy of Jones et al. 2015, Kotias and Satsurblia, belonged to Y-haplogroups J and J2a. Moreover, J2 today shows peaks in frequency and diversity in and around the Caucasus. In other words, Y-haplogroup J, and in particular J2, appear to represent paternal signals of CHG admixture.

Unfortunately, it's not yet possible to demonstrate with formal tests beyond any doubt that I0708 has CHG admixture.

For instance, the D-stats below, in which a couple of the least Caucasus-shifted Anatolian farmers are Anatolia Neolithic1, while I0708 is Anatolia Neolithic2, fail to reach significance (Z=3). Please note, I ran the stats with the Amerindian and Siberian samples to test for Ancient North Eurasian (ANE) admixture, which appears to be a feature of CHG.

However, the results are all clearly positive, and might reach significance with higher quality data and/or a better reference than Anatolia Neolithic1.

Indeed, the subtle difference in ANE affinity between Anatolia Neolithic1 and Anatolia Neolithic2 is underlined by the D-stats below. Note that here Kotias shows significant signals of admixture when paired with Anatolia Neolithic1, but not when paired with Anatolia Neolithic2. This is despite the fact that Anatolia Neolithic2 is a higher coverage sequence (6.95x vs 2.66x) and offers more markers.

I0708 is unlikely to be the only early western Anatolian farmer with CHG/ANE admixture. The PCA above show that a couple of others are also pulling strongly towards the Caucasus. Indeed, all of the Anatolian and European Neolithic samples might harbor low levels of CHG ancestry. The problem with testing this idea at present is a lack of more basal Near Eastern ancient genomes from core areas of the Near East, like, say, the Levant.

Hopefully they're on their way, but in any case, it's almost certain now that CHG was already expanding west, and in all likelihood east, during the early Neolithic. This probably has some important implications for the peopling of West Eurasia and their linguistic affinities. Feel free to post what these might be in the comments.

Update 11/02/2016: I came up with new Anatolia Neolithic1 and Anatolia Neolithic2 sets using D-stats (by comparing each of the Anatolians to Kotias versus sample I0708). For a breakdown see here. Anatolia Neolithic2 now shows significant signals of admixture from Kotias, Dai, Surui and Han. This implies that it not only harbors CHG ancestry, but also ANE and East Asian-related admixtures.

Ancient Greeks and Romans may have imported a whole new genetic cline into Europe (or not)

Is anyone else thinking what I'm thinking? The Principal Component Analysis (PCA) below should be self-explanatory. But if you're having problems with the abbreviations and acronyms, consult the list of definitions here.

See also...

First Neolithic genomes from Greece

The enigmatic headless Romans from York

Greeks in a Longobard cemetery

The Poltavka outlier

Anyone who still thinks that Y-chromosome haplogroup R1a originated in South Asia should burn this map into their brains. It'll come in useful over the next few years as we learn from ancient DNA about the conquest of the Indian subcontinent, and indeed much of Asia, by pastoralists from the western Russian and Ukrainian steppes.

X marks the spot of the burial site of Poltavka sample I0432 from the Mathieson et al. 2015 dataset. This individual belongs to Y-chromosome haplogroup R1a-Z93(Z94+), which today accounts for well over 90% of the R1a lineages in Asia and peaks in frequency at over 60% in the northern parts of South Asia.

Moreover, the dating of his burial site, 2925-2536 calBCE, suggests that he lived not long after the Z93 and Z94 mutations came into existence. That's because Z93 doesn't appear to be much older than 5,000 years based on full Y-chromosome sequence data (see here and here, including the comments).

So I0432 could well turn out to be a crucial piece in the puzzle of the peopling of South Asia.

Interestingly, this individual was flagged as an outlier in the Poltavka sample set by Mathieson et al., hence his other moniker: the Poltavka outlier. However, this wasn't because of any ancestry from South or even Central Asia. In fact, it was because he was too western.

Principal Component Analyses (PCA) featuring a wide range of present-day and ancient samples from Europe and Asia, like the one below, show that Poltavka outlier clusters further west than most Corded Ware individuals from Germany. Right click and open in a new tab to view full size.

In the past, using qpAdm, I modeled Poltavka outlier as 63.7% Yamnaya Samara and 36.3% German Middle Neolithic. This is probably not very far from the truth, but qpAdm offers a supervised mixture test in which the results are heavily reliant on the choice of outgroups, so I thought I'd revisit the issue with TreeMix, which allows an unsupervised analysis.

In a dataset including seven relatively high coverage Copper Age (CA), Early Bronze Age and Middle Neolithic (MN) European genomes, TreeMix picked out Poltavka outlier as the most likely sample to be admixed, showing a mixture edge of 33% from the base of the branch leading to the Iberian MN individual to that of Poltavka outlier.

This outcome is very similar to my qpAdm model, but it suggests an even more western source of admixture in Poltavka outlier. Could this admixture actually be from Iberia? I wouldn't discount this possibility, considering the presence of Bell Beaker communities, possibly of Atlantic or even Iberian origin, as far east as present-day Poland. Indeed, according to Cassidy et al. 2015, German Beakers show high affinity to MN and CA Iberians (see page 51 in the supp info here).

I double checked my TreeMix result with D-stats, and yep, when placed in a clade with Poltavka or Samara Yamnaya, Poltavka outlier shows the strongest signal of admixture from the Iberia MN individual.

At the same time, however, the signal from the Early Neolithic (EN) Iberian fails to reach significance (Z=<3), which suggests that, in fact, TreeMix and D-stats might be seeing the Iberia MN sample as the most attractive mixture source due to her high level of Western European hunter-gatherer (WHG) ancestry, which Poltavka outlier also has plenty of, rather than anything specific to Iberia.

In any case, it's clear enough that Poltavka outlier was the result of mixture between Yamnaya-related western steppe pastoralists and the descendants of Middle Neolithic Europeans with a high ratio of WHG ancestry. Where this admixture actually took place and which archaeological cultures were involved will have to be resolved with further sampling of ancient remains from Central and Eastern Europe.

However, it's already impossible to place the origin of Poltavka outlier anywhere in Asia, which suggests that both Z93 and Z94 are also from well inside the generally accepted borders of Europe.

This obviously has implications for the origins of the Indo-Iranians, because the widespread presence of these mutations in Asia gels very nicely with the idea, and indeed academic consensus, that Indo-Iranian languages expanded rapidly from the Eurasian steppe into Asia during the Bronze Age.

Considering that Poltavka outlier came from a Kurgan burial, and was therefore an individual of some social standing, he might be the direct ancestor of many millions of present-day Asians. If so, this won't be very difficult to prove in the near future as ancient DNA research revs up a few notches.

On a related note, apparently there's a paper on the way with ancient DNA results from Rakhigarhi, a Harappan site in Haryana, northern India (see here). As far as I know, the results will include Y-chromosome haplogroups of three males, but I don't think we'll see any decent genome-wide data at this stage. However, hopefully I'm wrong and the paper will come out with full ancient genomes.

Feel free to post your predictions in the comments. I'm tentatively expecting a couple of instances of J2 and maybe an L or H. Razib made basically the same prediction recently so I'm not being original. What I do know is that we won't see any R1a-Z93. The only way that might happen is if, say, someone coughed or sneezed on the Harappan remains.

Data source and reference...

Mathieson et al., Genome-wide patterns of selection in 230 ancient Eurasians, Nature, 528, 499–503 (24 December 2015), doi:10.1038/nature16152

See also...

The beast among Y-haplogroups

Next year the Armenian Plateau hypothesis will collapse

It's been a great year for population genetics and paleogenomics, and also for this blog. I ran a lot of analyses in 2015 and managed to make a few discoveries that were subsequently confirmed, or at least, backed up by academia. For instance:

- first to show with ancient genomes that the Anglo-Saxons made a significant genetic impact on England. See here. Eventually confirmed here.

- first to show that the southern admixture in the Yamnaya pastoralists of the Early Bronze Age steppe was Georgian-related rather than Armenian-related. See here. Confirmed here.

- first to show that Anatolian Neolithic farmers were very similar to European Neolithic farmers, and lacked Ancient North Eurasian (ANE) ancestry. See here. Confirmed here.

- first to show using ancient DNA and formal statistics that South Asia experienced massive gene flow originating in Late Neolithic/Bronze Age Europe. See here. Backed up with my help here.

The fact that Caucasus hunter-gatherers (CHG) like Kotias are essentially an ideal fit for the southern ancestry in the Yamnaya is a big problem for the Armenian Plateau Indo-European homeland hypothesis. This TreeMix graph shows why.

Basically, it looks like the Kotias-related ancestry in the Yamnaya came from the North Caucasus, rather than any place closer to the Near East than Georgia. Unless, of course, the southern Caucasus was populated by unadmixed CHG right until the 4th Millennium BC, when the hypothetical Proto-Indo-Europeans from the Armenian Plateau set off on their journey to Northern Europe around the Caspian Sea. But let's be honest, that's extremely unlikely.

Indeed, I expect that next year we'll see the first Neolithic and Copper Age samples from Armenia and/or surrounds, and even though they will be in large part CHG, they'll be nowhere near unadmixed. This will essentially kill the Armenian Plateau hypothesis, and thus leave the Kurgan or steppe hypothesis as the only plausible choice.

In any case, 2016 will probably be the year when ancient DNA helps to settle the Indo-European homeland question once and for all. So get ready for more ancient DNA from the steppe, but also, among others, from Mesolithic and Neolithic Iran, Mycenaean Greece and the Maikop Culture of the North Caucasus. I'm also pretty sure that the Varna man with the golden codpiece will make an appearance in a paper about Neolithic and Copper Age Bulgaria. Bring it on!

At least three genetically distinct Indo-European migrations into South Asia

First came the Indo-Aryans, probably in a couple of waves. Historical linguistics and archeology tell us that they originated on the Trans-Urals steppe in the Sintashta-Andronovo horizon, and pushed south around 2,000 BC to establish themselves as the ruling elite over Central Asian agriculturalists, who were probably in large part of West Asian origin.

There are multiple lines of genetic evidence suggesting that this is indeed what happened, which I discussed in detail in several earlier blog posts, like here.

But arguably the easiest way to show it is with D-stats of the form D(Indo-Aryan,Southeast_Asian; X,Outgroup), where the Indo-Aryans are the Kalash, a population isolate from the Hindu Kush with a relatively low level of extra-West Eurasian admixture and speaking an archaic form of Indo-Aryan. The Southeast Asians are the Dai from southern China, one of the best proxies for the South and East Asian admixture in the Kalash, while X represents a wide variety of present-day and ancient populations in my dataset. The top five D-stats, each based on well over 500K SNPs, are listed below:

Kalash Dai Kotias Ju_hoan_North 0.0684 22.704
Kalash Dai Sintashta Ju_hoan_North 0.0632 25.036
Kalash Dai Georgian Ju_hoan_North 0.0625 30.991
Kalash Dai Afanasievo Ju_hoan_North 0.0612 24.496
Kalash Dai Yamnaya_Samara Ju_hoan_North 0.0611 27.97

Really cool results. Obviously, Kotias is the recently published Caucasus hunter-gatherer (CHG) genome. The Kalash appear to carry the highest level of Kotias-related ancestry among present-day populations, which they probably acquired from both the Central Asian agriculturists and Indo-Aryan invaders. At the same time, however, Georgians show the highest affinity to Kotias because they harbor less extra-West Eurasian admixture.

After the Indo-Aryans came the Iranians, in all likelihood also from the steppe. They were either an offshoot of Sintashta-Andronovo or the more westerly Srubnaya Culture. I'd say the D-stats below, of the form D(Eastern_Iranian,Southeast_Asian)(X,Outgroup), are inconclusive, because the differences are small, and the outcome possibly affected by the methodology and/or sampling bias.

Tajik_Shugnan Dai Sintashta Ju_hoan_North 0.0716 26.427
Tajik_Shugnan Dai Poltavka Ju_hoan_North 0.0695 25.234
Tajik_Shugnan Dai Afanasievo Ju_hoan_North 0.0691 24.703
Tajik_Shugnan Dai Srubnaya Ju_hoan_North 0.069 28.266
Tajik_Shugnan Dai Corded_Ware_Germany Ju_hoan_North 0.0684 27.328

But again, the top five results make a lot of sense in the context of historical linguistics and archeology. By the way, Tajik Shugnans are a population isolate in the Pamir Mountains, like the Kalash with low level extra-West Eurasian admixture, and thus likely to be among the best available reference groups for early Eastern Iranians.

Interestingly, based on that list the Shugnans look more European than the Kalash. In large part this might be a reflection of the sharp rise in the level of European-specific Western hunter-gatherer (WHG) admixture on the steppe during the Middle Bronze Age, probably caused by population movements originating at the western edge of the steppe and/or in East Central Europe.

As far as I can tell, the fact that the Shugnans and Kalash have around the same level of extra-West Eurasian admixture means that I can try to hone in on the differences between their steppe-derived ancestry with D-stats of the form D(Kalash,Tajik_Shugnan)(Kotias,X). The top result seems to confirm my hunch, because Loschbour is, of course, a Western hunter-gatherer.

Loschbour 0.0149 3.874
Basque_Spanish 0.0113 4.232
Anatolia_Neolithic 0.0112 4.257
Karelia_HG 0.0105 3.005
Poltavka 0.01 3.539
Corded_Ware_Germany 0.0099 3.734
Afanasievo 0.0094 3.213
Srubnaya 0.0094 3.538
Yamnaya_Kalmykia 0.0091 3.362
Albanian 0.0088 3.419
Altai_IA 0.0088 3.087
Sintashta 0.0088 3.146
Greek 0.0076 3.094

Full output available here

More recently, during historic times, large parts of northern South Asia were settled by the Balochi, a Western Iranian people from the South Caspian region, whose ancestors were probably Indo-Europeanized a couple millennia earlier by Proto-Iranians from the steppe moving west across the Iranian Plateau. D-stats comparing the Balochi to the Kalash and Shugnans, respectively, clearly reflect the Near Eastern origins of the Balochi.

BedouinB 0.0104 6.151
Anatolia_Neolithic 0.0094 5.495
Druze 0.0084 5.228
Cypriot 0.0082 4.839
Syrian 0.0079 4.714
Armenian 0.0063 3.935
Satsurblia 0.0059 2.472
Georgian 0.0055 3.443
Iranian 0.0055 3.345
Abkhasian 0.0053 3.279
Greek 0.0052 3.166

...

Okunevo -0.0081 -3.552
Karelia_HG -0.0104 -4.666

Full output available here

Satsurblia 0.007 2.078
BedouinB 0.0051 2.277
...

Basque_Spanish -0.0073 -3.156
Mezhovskaya -0.0085 -3.045
Altai_IA -0.0092 -3.677
Scythian_IA -0.0092 -3.108
Yamnaya_Samara -0.0095 -4.092
Karitiana -0.0098 -3.501
Karasuk -0.0099 -4.322
Andronovo -0.01 -4.09
Sintashta -0.01 -3.951
Corded_Ware_Germany -0.0102 -4.34
Srubnaya -0.0106 -4.605
Yamnaya_Kalmykia -0.011 -4.511
MA1 -0.0118 -3.691
Okunevo -0.0122 -3.783
Poltavka -0.0125 -5.043
Afanasievo -0.0136 -5.235
Loschbour -0.0148 -4.201
Karelia_HG -0.0208 -6.537

Full output available here

In this analysis I used ancient samples from the recently published Jones et al. and Mathieson et al. studies, available on request from the authors and at the Reich lab website here, respectively. The present-day samples are from the Human Origins dataset, also available at the Reich lab website.

Basques are not simply a fusion of Iberian hunter-gatherers and early farmers

I thought I'd revisit the issue of Basque origins with my new Principal Component Analysis (PCA) of West Eurasian genetic variation. The useful thing about this PCA is that it gets around two problems that routinely affect PCA featuring ancient samples: projection bias, otherwise known as shrinkage, and exaggerated outcomes for individuals with high counts of homozygous genotypes.

A couple of recent papers argued that Basques were the direct descendants of local hunter-gatherers and early Neolithic farmers who arrived in Iberia from the eastern Mediterranean. This is probably correct for the most part, but it doesn't tell the whole story.

On the PCA above, Basques are quite distinct from Early Neolithic, Middle Neolithic and Copper Age Iberians (marked Iberia_EN, Iberia_MN and Iberia_CA, respectively), because they are significantly more eastern. In fact, they cluster with the only Bronze Age Iberian on the plot (Iberia_BA), which is the same individual that I found to harbor steppe-related ancestry (see here).

Thus, the story told by the PCA is that Basques are the progeny of Bronze Age Iberians, who, unlike their Copper Age predecessors, experienced a pulse of steppe-related admixture from the east.

Formal statistics back this up. For instance, here's a quote from the recently revised Mathieson et al. preprint:

However, the statistic f4(Basque, Iberia_Chalcolithic; Yamnaya_Samara,Chimp)=0.00168 is significantly positive (Z=8.1), as is the statistic f4(Spanish, Iberia_Chalcolithic; Yamnaya_Samara, Chimp)= 0.00092 (Z=4.6). This indicates that steppe ancestry occurs in present-day southwestern European populations, and that even the Basques cannot be considered as mixtures of early farmers and hunter-gatherers without it (4).

The key question now is who brought the steppe-related ancestry to Basque country. Were they Indo-Europeans or speakers of Proto-Basque? Also, did they actually come from the steppe, or somewhere nearby, like the Carpathian Basin?

The reason I mention the Carpathian Basin is because, as per the PCA, Basques more or less cluster between Copper Age Iberians and some of the Bronze Age Hungarians (marked Hungary_BA). But this is just one possibility, and I'm not sure at this stage how plausible it looks with, say, formal statistics.

In this analysis I used samples from the Allentoft et al., Gunther et al., Haak et al. and Lazaridis et al. datasets, all of which are publicly available. The latter two are found at the Reich Lab site here. If you're confused by some of the acronyms in the PCA key, see here.

New PCA format

From now on, every time a new dataset of ancient West Eurasian samples is made available online, I'll run it in a Principal Component Analysis (PCA) like this:

Please note that the plots above include the majority of recently published ancient samples, and yet they are not affected by projection bias, otherwise known as shrinkage. If you're confused by some of the acronyms in the PCA key, see here.

See also...

Basques are not simply a fusion of Iberian hunter-gatherers and early farmers

Uralian genes

Below is a graph based on a couple of D-statistics. Note that out of the five Uralic-speaking groups, only Hungarians remain under the upward-sloping red line, clustering among Indo-European-speakers. On the other hand, the Indo-European-speaking Russians from Kargopol cluster with Uralic speakers. However, Russia is a complex affair in this context, because much of the north and east of European Russia was Uralic-speaking until very recently. The relevant D-stats are available here.

See also...

Ancient genomes from NE Europe suggest the tandem spread of Siberian admixture and Uralic languages into the region >3,500 ya

A multidimensional approach

This is arguably the most interesting Principal Component Analysis (PCA) I've run to date. Note that overall the ancient steppe genomes appear to be the crucial link in the Indo-European chain, basically bridging the gap between European and South Asian Indo-European-speakers. A plot with all of the samples labeled individually can be downloaded here. If you have any questions about the methodology, just ask in the comments.

In this analysis I used samples from the Allentoft et al., Haak et al. and Lazaridis et al. datasets, all of which are publicly available. The latter two are found at the Reich Lab site here.

See also...

No significant genetic substructures within (eastern) Yamnaya