Population genomics of Bronze Age Eurasia (Allentoft et al. 2015)

[Coldmountains]

It could mean that the allotted ancestry that doesn't fit into Georgian and ENA, could be more similar to sintashta as a whole. Remember qpAdm parameters only go as far as 3 populations. It also has some other limitations, like it can't tell whether admixture goes from on group to another, if both populations admixed into one another. It can only tell f.ex if sintashta contributed to pashtoos, if it was a one way geneflow. But not in the situation where pashtoos had an ancestor that contributed to sintashta. Then the run itself becomes muddled.


Yes, but here is the thing. Sintashta, is somewhere around 80-90% east-euro/baltic like, with the remainder being Caucasian-like(Speaking in modern term populations). While pashtoos are something along the lines of 10-20% north-European like and 20% Caucasian, the remainder being typical for south-central-Asians and south-Asian populations. If pashtoos shared 63% of their test-relevant sequenced alleles with sintashta, they would literally share 40-50% of their alleles directly with NE Europeans. Which isn't the case.

Where was this run cited with androvono, if you don't mind citing it.

You raise important points and I think we are all confused by this results. But I wonder why the model for Pashtuns is one of the "best" models created so far and why there is 2x more Sintashta-related ancestry in Pashtuns than Georgian/Caucasian related if South Central Asians have not a large amount of Sintashta-derived ancestry. It is also hard to explain which ancestry was shared by Sintashta and non-IEs of South Central Asia if Pashtuns have most of their Sintashta-like ancestry from their non-IE ancestors. It can not be teal and neither Neolithic Farmer stuff so i wonder how else we can explain this results if we exclude significant genetic influx from Sintashta into South Central Asia. It is also not directly connected to ANE in my opinion because Pamiri have less ANE than Pashtuns and Kalash but according to this model they have most Sintashta-like ancestry.

[DMXX]

Not really, on a PCA with Europe, Near East, Siberia, South Asia and East Asia those qpAdm mixes would cluster close to where Tajiks, Kalash and Pashtuns do. This is because Dai, like presumably ASI, are extremely southern on such a PCA, while Northeast Siberians would be the northern extreme.

Here's a PCA on the Human Origins dataset where this shows.

https://drive.google.com/file/d/0B9o...TVk/edit?pli=1

These sorts of PCA's are entirely dependent on both the number of specified nodes and dimensions. Under 23andMe's three-way (European-African-Asian) PCA, a half Irish half Chinese individual will cluster in precisely the same region as Uzbeks or Uyghurs despite significant differences in either ADMIXTURE K>6 or qpAdm configurations with unrelated populations (e.g. Lithuanian, Thai, Telugu).

Citing PCA's as evidence of overall placement either way is a selective game.

[Raskolnikov]

It could mean that the allotted ancestry that doesn't fit into Georgian and ENA, could be more similar to sintashta as a whole. Remember qpAdm parameters only go as far as 3 populations. It also has some other limitations, like it can't tell whether admixture goes from on group to another, if both populations admixed into one another. It can only tell f.ex if sintashta contributed to pashtoos, if it was a one way geneflow. But not in the situation where pashtoos had an ancestor that contributed to sintashta. Then the run itself becomes muddled.

If only we had aDNA from BMAC and the Indus region we could finally put this to rest. I think a lot of the "Georgian" part is coming from an Andronovo+BMAC population. Is it likely that the pre-Aryan Indus population could have been mostly ASI?

Yes, but here is the thing. Sintashta, is somewhere around 80-90% east-euro/baltic like, with the remainder being Caucasian-like(Speaking in modern term populations). While pashtoos are something along the lines of 10-20% north-European like and 20% Caucasian, the remainder being typical for south-central-Asians and south-Asian populations. If pashtoos shared 63% of their test-relevant sequenced alleles with sintashta, they would literally share 40-50% of their alleles directly with NE Europeans. Which isn't the case.

I'm aware of this, but it could be just due to the way ADMIXTURE assigns clusters sometimes. The South-Central Asian cluster is very robust.

Where was this run cited with androvono, if you don't mind citing it.

Sein posted this earlier in this thread.

Here are the oracle results for Andronovo:

[1,] "47.2% bhatia_harappa_2 + 52.8% lithuanian_behar_10" "4.9297"
[2,] "55.7% lithuanian_behar_10 + 44.3% sindhi_hgdp_24" "5.0829"
[3,] "51.5% lithuanian_behar_10 + 48.5% punjabi-jatt_harappa_8" "5.5154"
[4,] "55.3% lithuanian_behar_10 + 44.7% punjabi-arain_xing_25" "5.7863"
[5,] "53.4% haryana-jatt_harappa_5 + 46.6% lithuanian_behar_10" "5.7911"
[6,] "42.5% bhatia_harappa_2 + 57.5% russian_hgdp_25" "5.8461"
[7,] "47.9% burusho_hgdp_25 + 52.1% lithuanian_behar_10" "5.8778"
[8,] "60.5% russian_hgdp_25 + 39.5% sindhi_hgdp_24" "5.9062"
[9,] "51.6% lithuanian_behar_10 + 48.4% pathan_hgdp_23" "5.9656"
[10,] "47.4% kalash_hgdp_23 + 52.6% lithuanian_behar_10" "6.2188"

[DMXX]

You raise important points and I think we are all confused by this results. But I wonder why the model for Pashtuns is one of the "best" models created so far and why there is 2x more Sintashta-related ancestry in Pashtuns than Georgian/Caucasian related if South Central Asians have not a large amount of Sintashta-derived ancestry. It is also hard to explain which ancestry was shared by Sintashta and non-IEs of South Central Asia if Pashtuns have most of their Sintashta-like ancestry from their non-IE ancestors. It can not be teal and neither Neolithic Farmer stuff so i wonder how else we can explain this results if we exclude significant genetic influx from Sintashta into South Central Asia. It is also not directly connected to ANE in my opinion because Pamiri have less ANE than Pashtuns and Kalash but according to this model they have most Sintashta-like ancestry.

Forgive me if it sounds like I'm repeating myself even by a fraction, but I personally suspect it has something to do with our current conceptions of the genetic landscape present in South-Central Asia. The Dai capture ASI well apparently, but ASI surely couldn't have been the only ancestral component present in the region.

I'm compelled to think the 2/3 Sintashta score isn't entirely Sintashta-derived and simply means there's a lot of "steppe related stuff" among Afghans and Tajiks, which explains why "Georgian" didn't capture it particularly well. For all we know, the BMAC folk were primarily a mix of (simplified) the West Asian originating farmers (ENF) with accumulated waves of Eurasian hunter-gatherer-fishers (ANE) together with southern hunter-gatherers or Indus valley offshoot colonies spreading further north (ASI) coalescing into these successful oases agricultural communities (recall agricultural settlements provided excellent survival odds compared to hunter-gatherer subsistence). Sintashta arrives from the north and a mixture of the two resulting (Andronovo-BMAC), producing the immediate ancestors of those Iranic speakers which migrated into Iran. This model would conclude that modern Afghans and Tajiks aren't predominantly Sintashta as the qpAdm output indicates, but derive most of their ancestry from this Andronovo-BMAC combination, which is essentially the "actual" Iranic ancestry which Iranians and Kurds received from 1000 B.C. onwards.

If the above proves to be correct, it would certainly mean Afghans and Tajiks are ultimately mostly steppe derived but only a portion of it is truly via Sintashta. It would also satisfactorily explain why the Georgian score is only ~20%; this comprises the bulk of the ENF input that arrived into the BMAC. It's really the BMAC who're sorely needed to test whether any of these propositions are reality.

My own arbitrary guess at the actual Sintashta ancestry in Afghans and Tajiks is around 20-30%. Kurds and Iranians would fall between 5-15%. Turkmen overlapping between these two ranges.

[Edit]: One idea... Undertake qpAdm for Afghans and Tajiks using Sintashta, Dai and a composite of Georgians with MA-1 (a "zombie" conjectured BMAC).

[Raskolnikov]

I agree that the high figure for Sintashta is questionable and probably too high. We also know that Sintashta didn't directly admix with South-Central Asians but rather it's offshoot, Andronovo who in turn mixed with BMAC before heading south. Having said that I'd be surprised if the amount of Sintashta ancestry in South-Central Asians turned out to be significantly less than 50%, given just how good the qpAdm model is.

[Shaikorth]

These sorts of PCA's are entirely dependent on both the number of specified nodes and dimensions. Under 23andMe's three-way (European-African-Asian) PCA, a half Irish half Chinese individual will cluster in precisely the same region as Uzbeks or Uyghurs despite significant differences in either ADMIXTURE K>6 or qpAdm configurations with unrelated populations (e.g. Lithuanian, Thai, Telugu).

Citing PCA's as evidence of overall placement either way is a selective game.

That's true for other things like ADMIXTURE clustering as well though, the components very much depend on the number and type of samples in the run. A Sintashta-Caucasian-Dai mix based on the population averages of Eurogenes K13 or even the low-K ancient runs might not look like a Pashtun average, but that doesn't take into account the overlap that exists within ADMIXTURE clusters. The qpAdm fit is based on f4 ratios and I'd trust it over any ADMIXTURE or PCA results, but if either correlates with it that's good. Indeed an Irish-Chinese mix would not have the same best qpAdm fit as Uygurs or Hazaras.

The reason why I think it would be hard to improve on those qpAdm fits is that I trust the Haak et al method and qpAdm replicates it well. The S-C Asian fits are very good, much better than what was achieved with the ancient genomes in Haak et al dataset for them, and compare favourably to European fits:

Belarusian qpAdm fit (good for an European population)
Chisq 0,641 tail prob 0,887085 Yamnaya 0,474 EN 0,358 WHG 0,143 E-Asian 0,025
Best Haak et al Belarusian fit: Yamnaya 0.447 EN 0.388 WHG 0,135 Nganasan 0,03

It is of course true that the fit just reflects ancient ancestry and does not exclude the possibility that some of Pashtun Sintastha-like ancestry might have arrived before or after any Sintashta admixture, and same goes for Europe. But for the base components I see really no reason to doubt the results.

[Coldmountains]

That's true for other things like ADMIXTURE clustering as well though, the components very much depend on the number and type of samples in the run. A Sintashta-Caucasian-Dai mix based on the population averages of Eurogenes K13 or even the low-K ancient runs might not look like a Pashtun average, but that doesn't take into account the overlap that exists within ADMIXTURE clusters. The qpAdm fit is based on f4 ratios and I'd trust it over any ADMIXTURE or PCA results, but if either correlates with it that's good. Indeed an Irish-Chinese mix would not have the same best qpAdm fit as Uygurs or Hazaras.

The reason why I think it would be hard to improve on those qpAdm fits is that I trust the Haak et al method and qpAdm replicates it well. The S-C Asian fits are very good, much better than what was achieved with the ancient genomes in Haak et al dataset for them, and compare favourably to European fits:

Belarusian qpAdm fit (good for an European population)
Chisq 0,641 tail prob 0,887085 Yamnaya 0,474 EN 0,358 WHG 0,143 E-Asian 0,025
Best Haak et al Belarusian fit: Yamnaya 0.447 EN 0.388 WHG 0,135 Nganasan 0,03

It is of course true that the fit just reflects ancient ancestry and does not exclude the possibility that some of Pashtun Sintastha-like ancestry might have arrived before or after any Sintashta admixture, and same goes for Europe. But for the base components I see really no reason to doubt the results.

It is worth mentioning that Haak results for Europeans also dont look accurate in my opinion because Belarusians are significantly lower in Yamnaya ancestry according to them than Norwegians what does not fit with higher ANE/WHG/Teal and less EEF among Belarusians than among Norwegians. We should also not take the results for South Central Asians to literally in my opinion but even when they are not totally accurate they hardly can be totally off. I agree with DMXX and think that we should create a Georgian +MA-1 Zombie to filter out older steppe stuff. Pamiri Tajiks are according to this model 85% Sintashta-like + 15% Dai-like what sounds too much and they have no Georgian/Armenian-like stuff what also sounds not realistic for me. But it is obvious that Pamiri Tajiks will score highest in direct Indo-Iranian ancestry among modern Indo-Iranians. The model for Pashtuns looks quite good even when some of the Sintashta-like stuff has maybe an older Non-IE origin

[DMXX]

... But it is obvious that Pamiri Tajiks will score highest in direct Indo-Iranian ancestry among modern Indo-Iranians. The model for Pashtuns looks quite good even when some of the Sintashta-like stuff has maybe an older Non-IE origin

Completely agree. Certain aspects of the Pamiri language (NE Iranic) cluster indicate a high degree of continued localisation in the Pamirs no earlier than 1000 B.C.; I've interpreted that for a while as indication that they received additional admixture from the steppes following the BMAC's official disintegration shortly beforehand. There's mixed Andronovo-BMAC archaeological items shortly before this point in the Zerafshan area if I remember correctly, essentially en route from the BMAC core area towards the Pamirs.

If this line of reasoning is correct, it would mean that Pamiri Tajiks are both the closest extant group to Sintashta/Andronovo, but were more steppe-derived than the Andronovo-BMAC hybrid population that splintered to form Afghans, non-Pamiri Tajiks and (to a lesser degree) Iranians and Kurds. Essentially a "steppe enriched" modern variant of the Iron Age Iranic-speakers of South-Central Asia.

[Shaikorth]

To be fair Haak results for Europeans also dont look accurate in my opinion because Belarusians are lower in Yamnaya ancestry according to them than Norwegians what does not fit with higher ANE/WHG/Teal and less EEF among Belarusians than among Norwegians. Ukrainians are according to them lower in Yamnaya ancestry than Hungarians what does not make sense based on higher ANE/WHG among Ukrainians and much lower EEF. We should also not take the results for South Central Asians to literally in my opinion but even when they are not totally accurate they hardly can be totally off. I agree with DMXX and think that we should create a Georgian +MA-1 Zombie to filter out older steppe stuff.

ADMIXTURE component overlap is probably behind that Norwegian-Belarusian issue. The best Norwegian Haak fit is 47,5% Yamnaya, 48,2% EN, 4,2% Nganasan. Note that while their Yamnaya is higher, they have more EN than Belarusians and no WHG beyond what's hidden in Yamnaya and EN in that fit. That would overall make them a bit more southern and western than Belarusians (as they are on PCA's), and those results can be broken down into ADMIXTURE components in various ways which should account for discrepancies there.

In the qpAdm run (found in Mauri M's blog) the result in previous post was from, Belarusians were actually a bit more Yamnaya than Norwegians. However, for some reason that Yamnaya+WHG+EN+E-Asian fit was bad for Icelandic, Norwegians and Finns (chisq >6) compared to English, Slavic populations and Mordovians (chisq <4) while Haak's EN+WHG+Yamnaya+Nganasan fit was about equally good for them all so the difference may not be significant.

[Sein]

63% sintashta(remember sintashta are like 80-90% east-European like) would make pathans far more yamnaya, they'd cluster in eastern Europe, or with baskhirs or tatars. Not with tajiks and Tajik poimiris. Besides yamnaya are very different to androvono and sintashta. The reason why pashtoons score higher yamnaya is because yamnaya itself is half "teal". I'd like to see IBD and F4 statistics, before i'd take these results at face value. There just seems to be something wonky, about the results.

Besides I scored 19% yamnaya derived on K6, logically that would make me 30-35% sintashta. Which I highly doubt is the case.

Arbogan,

These are based on f-4 statistics!

If you think these results are incorrect, it would make the whole Haak et al. paper fundamentally misguided and incorrect.

Also, the program has Iranians at only 20% Sintashta, and Kurds at around 16%.

I know it is hard to shake off the picture painted via ADMIXTURE, but those are informal methods. For what it's worth, qpAdm represents the cutting edge, and involves a far more robust form of analysis versus ADMIXTURE and PCA. It would be unscientific to dismiss these results, they are what they are.