The true source of the Steppe ancestry in modern Indians (continued)

In my previous post, I concluded that the steppe ancestry in Indians is most likely to have arrived after 1000 BCE via the East Iranian-speaking Śāka. the ~850 CE TKM_IA sample turned out to be the best steppe source among the options provided.

Since then, a particular commentator on Twitter, who is adamant about the ~1500 BCE invisible steppe invasion into India has been putting alternative viewpoints in support of the old theory favoured by Kurganists. 

In his first attempt, he argued for an Inner Asian Mountain Corridor (IAMC) specific ancestry (Aigyrzhal_BA from Kyrgystan 2000 BCE as the proxy).

@Astra_2310 @DevarajaIndra @razibkhan @agenetics1

Tl;dr I'm arguing we need bronze age Aigyrzhal, not just Sintashta, to model the steppe ancestry in Indians. Thoughts? https://t.co/7SRF1KwXWs

— Ummon Karpe | 𑀅𑀫𑀦 𑀓𑀭𑀧𑁂 💎 (@The_Equationist) December 13, 2022

He did not provide qpAdm output support to this claim. Since my previous article did not have Aigyrzhal_BA as a possible source (I did use Sarazm_En as a source which is a related population), I added it to the list of possible sources and retested a few of the targets. I believe that sunlight is the best disinfectant and hence all counters should be addressed satisfactorily to take the discussion further (unless they are outrageous).

None of the targets chosen showed any success with Aigyrzhal_BA, and TKM_IA was chosen as the best source in all cases. So this argument of his was easily dismissed. You can read my replies to his tweet with my results, and the file with the rotating model outputs for this particular exercise is accessible here.

In the next salvo, @The_Equationist proposed that the Loebanr_IA outlier woman from Swat valley has an ancestry which is better suited than Turkmenistan_IA to be the source of steppe in modern Indians. The attached file in his tweet has qpAdm results of a few modern Indians (Pandit, Tamil_Brahmin, Telugu_Brahmin, Haryana_Jat, Marathi_Vani) which compares the results for both options and proves his case. 

20) So what do the stats say? We can run the two models against each other to see which is a better source of steppe ancestry. It turns out that the woman in Loebanr is a better fit than Iron Age Turkmenistan for most modern Indian populations. https://t.co/jslL73JfrS

— Ummon Karpe | 𑀅𑀫𑀦 𑀓𑀭𑀧𑁂 💎 (@The_Equationist) December 15, 2022

 Let me first share a few minor problems that I see with that analysis so that it is sturdier in the future:

1. He has not made clear the version of the qpAdm tool used. Admixtools 2 for R is the latest version, but it is not tested enough and it does not give equal results as the standard, tested Admixtools 1 version on Linux/Mac. Furthermore, the results change drastically depending on the specific steps used (F2 extract first vs feeding the Geno data directly, size of the Geno file, etc). These are critical bugs, so I have stopped using Admixtools 2, and rely on the tested and stable Admixtools 1 version.

2. The number of SNPs of the private samples should be made clear (lower SNPs allow subpar models to pass as well).

3. Parameters used should be made clear (parameters setting - maxmiss? allsnps? inbreed?).

4. Sample ids and labels of all samples used should be made available.

All these are important so that one can try to replicate the results. This allows for a healthier debate. But all considered, this is nothing major because I can test these myself with my own settings.

A. SWAT Iron Age samples explained

Most of the Swat iron age (roughly dated from 1100-800 BCE) samples are extremely homogenous and form two clusters.

1. Swat_low steppe: A cluster of 10 odd samples which are not pulled towards BMAC/Sintashta as much as the Swat_IA cluster. These represent the local population ancestry before the steppe/BMAC impact on the region. Importantly, they are distinct from the Indus Periphery Samples from Shahr-Sokhta and Gonur but closer to Indus Periphery than the Swat_IA cluster is.

2. Swat_IA: The rest of the samples cluster tightly and show a higher steppe ancestry than the low steppe cluster.

I have analyzed the Swat samples in detail in this post. The likely steppe source for Swat_IA is something with the ancestry of Dali_MLBA, a population descended from ~80% Sintashta and ~20% IAMC ancestry (like Dali_EBA).

It becomes clear that the ancestry profile of the Loebanr_o sample is not at all representative of the region. Rather, from the PCA it seems that Loebanr_o is an admixture between Swat_IA and TKM_IA in a roughly 50/50 manner (it falls in the middle of the Swat_IA - TKM_IA cline). I will prove this with formal stats later.

In the above PCA, you can also see that TKM_IA falls smack on the cline joining BMAC and Sintashta, whereas Loebanr_o is pulled towards Swat_IA and lies very close to the cline joining TKM_IA and Swat_IA.

From the below PCA, it also seems that the source of the steppe for Swat_IA is one which is heavier in BMAC than TKM_IA. We find this by joining the Swat_lowsteppe and Swat_IA cluster and extending it to meet the BMAC-Sintashta cline (at the pink circle). This probably tells us that the steppe/BMAC mixed source for Swat_IA was older than the date of TKM_IA since steppe contact to the south increased over time.

Above are the basic inferences that can be made by looking at the general ancestries of all these samples. 

Problem with using an Outlier as a source

The major issue is the usage of an outlier sample as a source. To put it in perspective, there are 101 samples from the Swat Valley Iron age in the aDNA database. Only one sample (two if you count another outlier with low quality) of these, Loebanr_outlier, shows this elevated steppe ancestry profile.

In my previous article, I did not include Loebanr_outlier as a source. I had also given the following caveat:

"5. A rotating model strategy checks each source for fit in an unbiased manner. However, the list of sources chosen introduces bias. Since there is a limit to the number of sources I can test (due to very high program runtimes) there could possibly be some Labels not present in this analysis which could be the true source. I have tried to ensure that the most relevant theorized sources are included here."

I did include Loebanr_IA (as a proxy for Swat_IA) and Aligrama_IA (as a proxy for Swat_lowsteppe) as potential sources in the previous article but they don't get selected. However, the exclusion of Loebanr_outlier had nothing to do with this caveat. I didn't include it because it is not good practice to use extreme outliers (~1/100 in Swat iron age) to model a population the size of India, since we do not know whether such a population was present in large enough numbers to cause an impact in the whole of India. If we do assume that such a population was present in large numbers, we do not know where they were present in those large numbers. Some educated guesses can be made, I will make them in some time. 

Having said all this, it is still worthwhile to attempt to model modern Indians with this sample, because if it does become successful it can give us valuable information about the kind of vector of steppe source in Indians.

B. Testing the models

You can skip to the next section if you want to miss the technical aspect of this article.

In this section, I will post the results of the various rotating qpAdm models that will be required for this article. I will use Admixtools 1 on Linux (qpAdm version 1520, qpfstats version 540). qpfstats parameters are set to allsnps: YES, inbreed: NO, scale: NO. qpfstats output will be used as input for qpadm models.

As usual, rotating models will be run with Fixed references and Rotating sources/references. When the sources are not being tested, they will be put in the reference list along with the fixed references. So all combinations of models for a target will have the same number of references.

Models with p-value > 0.05 AND positive admixture weights will be accepted. Passing models will be marked green, and failed models in red. If no models pass, the one with the highest p-value will be shaded yellow.

List of sources to be tested

Fixed Reference list

All the caveats of my previous post, apply to these qpAdm models as well.

B.1 Modelling of Targets

Targets to be tested

Most of the targets are the same as my previous post, but in this analysis, I added the RoopkundA samples from ~800 CE Roopkund lake, Uttarakhand. 4 samples with the highest steppe ancestry and with >200k SNPs were tested individually, and the rest of the 17 were clubbed together. In an earlier article, I analyzed the RoopkundA samples with distal qpAdm models. In this analysis, I will model them with proximal sources.

B.1.1 Roopkund_I2871

Indus_Periphery + Kangju provides the best model for I2871.

B.1.2 Roopkund I3346

Loebanr_o is the best steppe source for this sample.

B.1.3 Roopkund I3406

Loebanr_IA_o is the best steppe source

B.1.4 Roopkund I6938

Loebanr_o is the best steppe source for this sample.

B.1.5 Roopkund A

This label contains 17 other samples from RoopkundA cluster except for the 4 high steppe samples that I already analyzed and the 2 Pallan-like samples.

No models pass, but Loebanr_o is the most promising steppe source.

B.1.6 Niyogi Telugu Brahmin

Gonur (BMAC), Loebanr_o as well as TKM_IA are acceptable external sources with similar p-values for their models.

B.1.7 Tyagi_UP

Both Loebanr_o and TKM_IA work, but the Loebanr outlier is to be preferred.

B.1.8 Telugu Brahmin

Brahmins from Vizag. Loebanr_o is preferred as the steppe source.

B.1.9 Rajput East UP

Loebanr_o is preferred.

B.1.10 Rajput from UP/MP border

Again, Loebanr_o is preferred.

B.1.11 Tamil Brahmins

Loebanr_o is preferred.

B.1.12 Saryuparin Dwivedi from UP

Dali_MLBA is preferred as the steppe source.

B.1.13 Kashmiri Pandit

Turkmenistan_IA is the preferred source of steppe ancestry.

B.2 Results of the Models

Setting aside the question of whether single outliers should be used to model large populations or not, it does seem the Loebanr_o sample is a better fit for steppe ancestry in modern Indians. We have 1 target for which Kangju works as a steppe source, 2 for which TKM_IA works, and 1 for Dali_MLBA; but 10/13 targets can be modelled best with Loebanr_o. In the next section, I will discuss this sample and its relevance.

C. Model for Loebanr_IA_o

For this, I will remove Kangju as it is much later in time than Loebanr_o and so can't be a source.

Result: Turkmenistan_IA + Loebanr_IA / Indus Periphery (SiS_BA2) / or Aligrama_IA is accepted. Sintashta + Loebanr_IA or SiS_BA2 is accepted.

If we combine the previous article with this one, in terms of pure technicality the hierarchy of the steppe source in Indians is

Loebanr_o > TKM_IA > Kangju > Sintashta / Dali_MLBA

But Loebanr_o itself seems to have descended from TKM_IA (~51% ancestry). The likely reason why Loebanr_o is preferred over TKM_IA is the combination of Indian ancestry along with TKM_IA in that sample. In the previous article, most targets were forced to choose a combination of a high AASI source (like Pallan or Irula, due to a lack of samples) and TKM_IA. But in reality, that would overstate the BMAC ancestry in the target (via Tkm_IA which is ~43% BMAC) when it really needed some lower AASI Indian ancestry as well.

So, (Irula + North Indian) + TKM_IA is what is needed for most of these Indians. Here, (Irula + North Indian) stands for a less AASI-shifted local population which we have not yet sampled. Surely all north Indians are not derived from an Irula-like high AASI population.

There is an additional reason why Sintashta + Loebanr_IA is not to be preferred as a model for Loebanr_o. In both articles, we see that Irula/Pallan + Loebanr_IA + Sintashta is not preferred in the 3-way models. Simply put, if Loebanr_o is removed from the competition, TKM_IA will always be chosen over Loebanr + Sintashta even in 3 source models. This confirms that Loebanr_o is indeed mediating TKM_IA-like ancestry.

Apart from having the highest p-value and working with an additional local source (pre-steppe ancestry of Aligrama_IA, Swat), Turkmenistan_IA is also to be preferred due to archaeological contacts with Swat. 

From Sarianidi (2001), emphasis is mine

The copper "double-teeth forks" from Yaz Tepe in Margiana (fig. 15) and Hissar III in the Iranian Khorasan (Schmidt 1937, pl. iv) offer further proof of this statement, and they closely resemble the analogous ones from Swat (EIEC, p. 559).

Also, 

According to some scholars (Mallory 1989; Sarianidi 1998a), there is no archaeological data testifying to the presence of the steppe Andronovo tribes on the Indian subcontinent. On the other hand, there is a lot of material proving the penetration of the BMAC type tribes into this area (Sarianidi 1998a: 153-158, fig.75).

And,

The Swat Valley is unanimously considered to be the only suitable way through which the Indo-Aryans could have reached the Indus Valley. In such a case Swat may be looked upon as an intermediate point on the joumey that Indo-Aryan tribes took when migrating from the BMAC to the sites of the Harappan civilization.It has already been mentioned that Swat materials find their closest analogies in Central Asia and North Iran... (Mallory 1989)

In my previous article, I have also provided additional archaeological evidence connecting the Yaz cultures to sites (900-600 BCE) in modern Pakistan like the Bannu basin (Petrie and Magee, 2005). Parpola (2012) identifies the Yaz-1-related people are proto-Saka Dasyus of the Rig Veda.

As I mentioned before, the Loebanr_outlier sample has excess steppe ancestry unseen in any other sample from the region. The closest samples in time and space which could have been the source of the steppe ancestry would be TKM_IA (~850 BCE) or Kashkarchi_BA (~1100 BCE Uzbekistan, ancestry is very similar to Sintashta, almost pure steppe_MLBA profile). We are missing samples from North Afghanistan (BMAC culture at Dashly 3 in North Afghanistan) which would have provided important clues about the path taken to Swat valley.

Nevertheless, the date of the Loebanr_outlier sample is 1107-924 cal BCE. Her mtDNA Hg is T1a1 which is definitely from the steppe, indicating that she had a steppe-ancestry-rich mother but likely a local father. Narasimhan et al 2019 in their archaeological supplement put the material in her grave as belonging to the 'middle/late phase'.

The assemblage includes 20 vessels and miniature vessels, beads of semi-precious stones, an iron pin and a spindle whorl. The chronology of the pottery is associated to the middle/late phase of the Swat Protohistoric Graves assemblages.

Loebanr sample dates

The spindle whorl is a common BMAC artefact (Lynne Rouse, 2015)

Given the calibrated dates of all the dated samples from Loebanr (at 95% Confidence), it is likely that the Loebanr_o woman lived closer to 920 BCE than 1000 BCE. This is because the other samples with middle/late phase material in their graves are dated closer to 900 BCE than 1000 BCE (928, 897, 906, and 830 BCE being the early limit at 95% CI). It is also to be noted that such a steppe-enriched individual is not present in the oldest phase of the site. For reference, TKM_IA is from the Yaz-II site of Takhirbai 3. The date of the sample is 916-796 cal BCE at 95% confidence. ie ~850 BCE.

All of these facts combined tell us that this woman was likely the child of a migrant mother from the Yaz culture (or a culture with a similar genetic profile) and a local father from Swat. This outlier sample is probably one of the first of many Saka-related migrants into the periphery of India.

D. Responding to @The_Equationist's Arguments made in his thread

I will address him point by point starting with 3)

"Most of India would have been filled with ASI people during the Iron Age - aside from some groups such as Tamil Scheduled Caste with elevated IVC ancestry, even in North India in the regions where the IVC ruled, the IVC ancestry had become heavily diluted by hunter-gatherers."

This is a misrepresentation of what I say. First, I don't use the word 'Hunter-Gatherers'. Perhaps he equates Irula or Pallan-like ancestry with hunter-gatherers (which they are not). Maybe he thinks of AASI ancestry as belonging to literal Onge people, who knows.

Second, I gave various caveats in my previous article stating that the number of useable samples from India is low, and so we have to make do with what we have. For eg, I wrote

4. There is a close to a 100% chance that the 5 local sources do not cover the variance within the pre-steppe population of India. So, regardless of what the passing models say, with new data and more choices, the working models will keep changing. Models should not be taken too literally eg. Irula being chosen as a successful source does not literally mean that the Irula ancestors admixed with the other sources in the North.

I wrote this for Niyogi, where I clearly mean that a northern shifted population must be present in North India.

He can also be modelled without steppe (Irula + Bmac, which may also be a proxy for a northern shifted local pre-steppe population without actual BMAC ancestry).

I wrote this for Telugu Brahmin

The combination of Irula + a 2nd (lower AASI) local source tells us that the steppe source admixed with a population which was more northern in ancestry than Irula.

Clearly, I am aware that northern populations exist that had less AASI than Irula. We just don't have samples from them. I did provide lightly AASI shifted sources - Indus Periphery, Loebanr_IA, Aligrama_IA, but they don't get selected as sources, only Roopkund_Pallan or Irula do.

Points 4) to 6) are more or less accurate representations of my view.

About 7) - Historical reality does not speak for itself. Many people torture the evidence to prove something which is invisible. Such as the 1500BCE invasion of Punjab and Haryana by steppe folks which can't be seen in archaeology, and is not mentioned in the earliest literature. In the case of India however, the fact that there is so little evidence of Andronovo interaction in the core Vedic region, that it is hard to torture what is not available.

He writes

8) In this respect however, when it comes to this Saka migration theory, the sociopolitical implications would be quite dire.

As if the implication of the male-mediated steppe migration/invasion theory has some extremely rosy and pleasant consequence. Though the consequence is not my concern, only the truth is.

 

9) Saka migrants and invaders would have inserted themselves as rulers right in the middle of the blossoming of classical Indian society, with their descendants being privileged for thousands of years over the natives who composed the Vedas and founded the Kuru Panchala kingdom.

I do not know whose theory he is talking about, must be something he made up in his mind. I talk of an elite mating network, between elites of India and foreigners, largely female-mediated. The Loebanr_o woman herself is such an example. Her mother was likely a steppe-born T1a1 descendant and the father was a local from Swat. If he agrees that Loebanr_outlier or swat, in general, was the mediating population, then he must accept that the steppe ancestry there is female-mediated.

Furthermore, as I have written about many times, the Andronovo or Yaz-related R1a-Z2124+ has a minor presence in India (~5% frequency overall, but higher in NW, check article) and is present in even lower frequencies in the east and central India. We already have 2 recorded examples of foreigners giving their daughters to Indian kings - Greek Seleucus and Saka Rudradāman I. 

I also presented the case of the Maga Brahmins, who were foreigners and may have eventually assimilated into the Brahmin fold and intermarried, thereby mediating some steppe ancestry.

10) is his fantasy, nothing to do with what I have written or my views. Why would eastern Iranian-origin rulers impose Paninian Sanskrit on people, I do not know. Regarding Buddha being a Scythian - well he already believes him to have steppe ancestry, so what's the big deal with Saka ancestry? Is there something I am missing?

11) and 12) - Historical records don't speak of a massive Saka influence but they sure do speak of macho steppe warriors destroying IVC around 1500 BCE and upturning the whole culture right? He should provide just one archaeological evidence of his claim. At least the Saka contact was long-lasting and well-attested in literature, epigraphy, numismatics, and archaeology.

13) - He is stuck on Dravidian tribals for some reason, I have covered it already. Wonder if he had similar issues when Narasimhan modelled Indians using Onge.

14) - 17) details his pet theory. He can have that, but of course, he has no evidence to prove any of it.

 

19) is where he takes a bigger leap than Michael Jordan

For example, one woman buried in 10th/9th century BCE Swat culture in Loebanr genetically resembles an already steppe-admixed ANI-rich person. And we know she’s a transplant from India and not from the steppe because her ancestry has some AASI in it.

She's a transplant from India and not from the steppe because of AASI? Her mtDNA T1a1 is literally from the steppe. Her autosomal profile is Loebanr_IA + TKM_IA. She has nothing to do with any region south of Swat. There is not a single person in modern India with that high a steppe ancestry. She absolutely cannot be a migrant from deeper inside India. There is also not a single ancient sample from anywhere close to India with that high a steppe ancestry. So unless India is a deep fountain of steppe ancestry, he is wrong.

She was the child of a local Swat valley man and a woman who was likely a migrant from the Yaz culture.

20) - I'm in agreement. Loebanr_o fits better than the Iron age Turkmenistan sample as a steppe source. However, Loebanr_o herself is descended partly from TKM_IA (~51%) and they're both from around the same time period of ~900 BCE. Tomato/tamaato.

23) - "The mediating source was already in India at this time". 

Already at the periphery of India, to be precise. And yes, this is great proof that by 900 BCE the Saka-like ancestry of the Yaz people was entering India. Later, between 900-600 BCE, Yaz related material culture can be found at Bannu-Akra sites as per Petrie & Magee, 2012. How many people with ancestry like hers migrated south, is the question.

CONCLUSION

I must thank @The_Equationist for this. I had not thought of using the Swat outlier to model modern Indians and had not focused much on that sample. After all, a single outlier in an ordinary burial context among 100 can't convey much unambiguous information. But in this case, given the good fits it provides, I think we have the first link between the Yaz culture and the periphery of India post-1000 BCE.

Narasimhan et al 2019 concluded Central_Steppe_MLBA ancestry in Indians. My previous post narrowed the proximal source further to Turkmenistan_IA. This article pinpoints it further to a Loebanr_IA_outlier type population? 

More steppe-heavy samples like these will surely help us find the answer.

SUPPLEMENTARY

Rotating model results and Sample Id list is in this google sheet

REFERENCES

Ashish, K. (2021). R1a L657 in Roopkund lake, Uttarakhand from 800CE, but no Steppe. The Archaeogenetics Blog. https://a-genetics.blogspot.com/2021/12/l657-roopkund.html

Ashish, K. (2022). The true source of the steppe ancestry in modern Indians. The Archaeogenetics Blog. https://a-genetics.blogspot.com/2022/12/steppe-source-in-indians.html

Ashish, K. (2022). Update: IVC & Swat Valley Genetics, Bonus - Kashmiri Pandit Ancestry. The Archaeogenetics Blog. https://a-genetics.blogspot.com/2022/08/ivcandswat.html

Ashish, K. (2022). R1a Explained. The Archaeogenetics Blog. https://a-genetics.blogspot.com/2022/10/r1a-explained.html

Kuzʹmina, Elena Efimovna. The origin of the Indo-Iranians. Vol. 3. Brill, 2007.

Magee, Peter, et al. “The Achaemenid Empire in South Asia and Recent Excavations in Akra in Northwest Pakistan.” American Journal of Archaeology, vol. 109, no. 4, 2005, pp. 711–41. JSTOR, http://www.jstor.org/stable/40025695. Accessed 16 Dec. 2022.

Mallory, James P. "The Indo-Europeans: language, archaeology and myth." (1989).

Narasimhan, Vagheesh M., et al. "The formation of human populations in South and Central Asia." Science 365.6457 (2019): eaat7487.

Olivieri, Luca Maria, et al. "A new revised chronology and cultural sequence of the Swat valley, Khyber Pakhtunkhwa (Pakistan) in the light of current excavations at Barikot (Bir-kot-ghwandai)." Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 456 (2019): 148-156.

Parpola, Asko. "The Dāsas of the Ṛgveda as Proto-Sakas of the Yaz I-related cultures: With a revised model for the protohistory of Indo-Iranian speakers." Archaeology and language: Indo-European studies presented to James P. Mallory. Institute for the Study of Man, 2012. 221-264.

Petrie, Cameron A., and Peter Magee. "The Achaemenid expansion to the Indus and Alexander’s invasion of north-west South Asia." Iranian Journal of Archaeological Studies 2.1 (2012): 1-25.

Rouse, Lynne Marie. A line in the sand: archaeological evidence for the interactions of settled farmers and mobile pastoralists in the Late Bronze Age (1950–1500 BC) Murghab alluvial fan, Turkmenistan. Washington University in St. Louis, 2015.

Sarianidi, Viktor. "The Indo-Iranian problem in the light of the latest excavations in Margiana." Studia Orientalia Electronica 94 (2001): 417-442.

Zahir, Muhammad. "The “Gandhara Grave Culture” New Perspectives on Protohistoric Cemeteries in Northern and Northwestern Pakistan." A Companion to South Asia in the Past (2016): 274-293.