Substack cometh, and lo it is good. (Pricing)

Cryptic Ashkenazi ancestry across Eastern Europe

One of the great things about the spread of ‘direct to consumer’ genomics is that it’s increasing sample size in countries where for various reasons there isn’t much coverage. It was brought to my attention that My Heritage DNA results have been analyzed by the company, and yielded the surprising result that Hungary has been most impacted by Ashkenazi Jewish admixture in the Diaspora. This is surprising since it is well known that the United States of America is home to the second-largest Jewish community in the world, with more than 90 percent of that being Ashkenazi.

The main issue here is the distinction between genetic and cultural definitions. Ashkenazi Jews are a coherent population genetic classification, emerging out of a series of admixtures during the medieval period as a strong endogamous group. This means that this community has a distinctive genetic profile, just as Finns or Cambodians have distinctive genetic profiles. But Ashkenazi Jews are also a cultural and religious entity. Because of social and cultural constraints imposed by Christian societies, Jews could leave their religious identity, but Christians could not become Jewish.

In the United States, the massive wave of Jewish migration occurred around 1900. This is not so many generations in the past, so not too many people have very distant Jewish ancestry. Additionally, anti-Semitism has been a more marginal factor on the American landscape, so Jewish ancestry has been less hidden (though not always).

The situation in Eastern Europe is very different. A massive wave of demographic expansion occurred among Ashkenazi Jews after 1500. In the 18th-century Jewish fertility was far greater than gentile fertility in Poland. This resulted in an increase in the Jewish proportion over time, but likely also assimilation of some Jews into Christian society. The “Jewish Enlightenment”, spanning the 100 years between 1780 and 1880, was also a period when massive defections occurred from the more integrated elements of the Central European Jewry. Moses Mendelssohn’s last male descendant to practice Judaism died in 1871, after one century of assimilation and conversion.

Overall, this result confirms what history would suggest to us. I believe if My Heritage DNA looks specifically at IBD tracts they will see that an early peak of admixture would center around 1830, during the height of the Jewish Enlightenment, in Central Europe. The admixture will be later further east in Europe, due to the later period of assimilation of Jews in those societies. In contrast, in the USA exogamy rates for Jews remained at 10% as late as 1960. Only in the past few generations have been risen to around 50% or more.

29 thoughts on “Cryptic Ashkenazi ancestry across Eastern Europe

  1. Hungary was in the Austro-Hungarian Empire (successor state to the Holy Roman Empire). After 1848, the Empire was willing to allow Jews who converted to join the civil service. This was a major incentive to conversion and quite a few Jews did that. I am not under the impression that the Haskalah or the Reform movement had much of an impact in Hungary.

    In Vienna, in the 20th century there was less persecution and more intermarriage. My wife’s family, which moved from Bohemia to Vienna in the late 19th Century had a couple of members like that.

    Friedrich Hayek’s aunt was Ludwig Wittgenstein’s grandmother, his only non Jewish ancestor. Frederich was the second smartest guy in the family, but the only one to win a Nobel Prize.

    The Nazis applied a brutal filter to the Empire’s Jewish Population. but, although they persecuted the children of mixed marriages in Austria, they were not sent to the camps. So, the children of converts survived preferentially.

  2. Looks like bad sampling or bad methodology, because none of the Central and East Euro samples from academic datasets show significant Jewish admix. There’s probably a handful that have minor Jewish admix.

    This is easy to show. Just run an IBD analysis with Jews and then a PCA to see how many European samples pull towards the Jewish cluster.

    Two French from the HGDP. One Iberian from 1000 Genomes. That’s pretty much it.

    https://1.bp.blogspot.com/-3eXmFdJMwm4/XWp93td4LBI/AAAAAAAAILM/Hf7EVnTS46ExHAaWjYCl13QX6uIEWlyzQCLcBGAs/s1600/Jews_IBD.png

    https://eurogenes.blogspot.com/2014/08/haplotype-based-pca-of-west-eurasia-and.html

  3. @Davidski

    I think that the MyHeritage folks admit that the their sample is biased:

    Dr. Staetsky suggests accounting for a degree of selectivity of MyHeritage users. Since commercial genetic testing is an activity where the most educated and well-to-do classes of a society will be over-represented

    I assume that this implies more urban populations as a result. On the other hand, there may also be some anti-urban bias in the academic databases, no? Also, have no idea how large these are.

    My own experience with MyHeritage is that their Ashkenazi “target” is much stricter than the other DTC.

    I’m 99.6% Ashkenazi at 23andMe and only 89% under MyHeritage (FTDNA splits the differnece at 94%).

  4. @Eric

    Yep, that’s one possible confounding factor, another one might be that people who suspect Jewish ancestry are more likely to order a test from a company advertising a Jewish ancestry test.

    Apart from that I don’t think any commercial tests really take into account the fact that Ashkenazi Jews aren’t pure, but rather a complex recent mixture with low level but significant Central and East Euro ancestry, which can reflect back as Jewish admix in many Central and East Euros.

    If MyHeritage is serious about this, then it should try to replicate its results by sampling random individuals from some of these countries, like Hungary, and also running different types of tests, especially based on haplotypes.

  5. I suspect this will take large scale bio bank types samples to solve. Sample them all and let the data sort it out…

  6. Sampling may be questionable… National Geographic Geno 2.0 claimed a 6% contribution to “Polish” from the Jewish Diaspora. I don’t know how reliable that estimate was either but I have found a roughly 8.2% inheritance among Jews from Eastern Europe using ChromoPainter vs 13.8% pairing of Jews to Eastern Europeans using Beagle 4.1*, which would indicate to me that gene flow was more Jewish-to-Gentile in Eastern Europe. This may be related to the Frankist movement {Jacob Frank} in Poland which brought tens of thousands of Jewish converts to Catholicism during the late 18th Century. I would speculate that this admixture from ~ 6 generations ago would not lead to great percentages of Ashkenazi admixture among Poles today; unlike, perhaps, Hungary with a more recent “convsero” WW2 admixture…

    * Dienekes had gotten similar results with FastIBD in 2012

  7. @Davidski

    You are right — family stories are a major confounding factor to the sample. Not sure I agree with you about the size of the C&E Euro inflow.

    I have subscribed to the bottleneck theory of Ashkenazim genesis since I read about it. 1/2 Levantine and 1/2 southern Euro until bottleneck about 700 years ago then some low C&E (5%) Euro flow.

    I’m not too wedded to this — seems like its our best guess at this point. More ancient anchors to measure this would be helpful, otherwise it seems like were are left relying on the quality of our priors. I’m all for digging up older graves, but I’m in the minority on this…

    I do think, at least wrt modern East Euro populations, most of the flow has been out not in. This is based anecdotally on my family and friends (I was born in Soviet Ukraine). Very endogamous until the late Soviet period, then a lot of admixture.

    I also disagree somewhat with Razib here — the peak of admixture is a function of location. I don’t think much happened in the lands of the Russian Empire until the late 20th century.

  8. Razib, apologies for adding another reply to an Ashkenazi related post which usually generate inordinate number of comments. But I think it hasn’t been pointed out yet.

    The admixture they are tracking at MyHeritage dates back to the XX century, and has nothing with the assimilation tendencies of the XVIII-XIX c. Indeed, as they explain, the majority of the customers they tagged have 25%+ AJ ancestry (a parent or a grandparent was of fully Ashkenazi heritage), with the rest having 10% to 25% (typically a great-grandparent). And it isn’t like MyHeritage has a good track record with detecting <10% admixtures anyway (their selling point is that they can tell apart really genetically similar populations, using an high number of reference populations, which comes at a cost of poor separation of the ancestral components at lower percentages).

    It isn't quite correct to say that they market themselves as "Ashkenazi DNA experts". Rather, they are the only major lab specializing in NON-Ashkenazi Jews (Sephardi, Misrahi, Yemeni). These are in demand among the Ashkenazi Jews seeking their legendary Sephardi ancestors. Since Sephardi and Ashkenazi are so genetically similar, on the one hand, and internally diverse, on the other hand, most Ashkenazi customers of theirs get a few percents of their DNA assigned to Sephardi and/or Iberian components, and feel vindicated. Needless to say, their identification of the actual Sephardi ancestry in the related Ashkenazim isn't reliable at <20% of the DNA content (works fine with parents or grandparents, which is also cool, of course)

  9. Sgt rightly noted the conversions of thousands of Polish-Ukrainian Jews to Roman Catholicism in the 1700s. I have confirmed the existence of low levels of Ashkenazic Jewish admixture in southeastern Poles, Rusyns from Slovakia, and western Ukrainians of the Christian faith. I also found tiny amounts of Sephardic Jewish and Italki Jewish ancestry in a few Poles.

    For instance, a southeastern Pole scored 2.1% Ashkenazi in 23andMe and, later, I found 5 Jewish segments in his autosomal DNA: 4 Ashkenazic segments matching large clusters of Ashkenazim plus 1 Sephardic segment shared with both Ashkenazim and a New Mexican Hispano. I checked the ethnicity paintings of most of the segments and found that they consisted of Middle Eastern and Mediterranean elements rather than Slavic components.

    I also matched up a Southern Italian Catholic (roots from Calabria and Campania) with a segment shared with not only Ashkenazic autosomal matches but also an autosomal match who was mostly Polish Christian + a small part Ashkenazi Jewish. (Sephardic Jews or Sephardic Conversos don’t share this segment; that’s why I believe its ancestor was Italki.)

    I found a person with majority Ukrainian ancestry plus about 7.6% Ashkenazic DNA (per Eurogenes’ Jtest’s oracle’s mixed mode estimate) carrying a Sephardic segment shared by clusters of Ashkenazic Jews, Northeastern Mexican Catholics, and non-Mexican Latin American Hispanics.

    Several other Poles and Ukrainians I worked with who had Middle Eastern type segments had only Ashkenazic cousin matches without any matches of a Sephardic or Italki nature.

    Uniparental evidence occasionally backs this up. The distinctly Ashkenazic mtDNA haplogroup K1a1b1a has been found occasionally in Poles, and the Ashkenazic-Karaite Y-DNA haplogroup G-P303 has been found in a Rusyn family from Slovakia who also scored low Ashkenazic percentages in their autosomal DNA tests.

  10. Interesting discussion. I’ve got to do one of these DTC DNA tests and see how much pct non-Ashkenazi I am 🙂

    The story about My Heritage DNA and Jewish ancestry in Hungary was picked up by the popular Jewish press as something like “Hungarians have the most Jewish genes”. It’s hard to convey this stuff accurately in the popular press.

    Hungary as it exists now, of course, is smaller than the region that was referred to by that name during the Imperial era. Anecdotally, there is a distinction between the “Germanized” Hungarian Jews of Budapest, Pressburg, etc. (who were highly assimilated and basically Viennese in their outlook) and the more traditional or “backward” Jews who lived in areas that are, to a great extent, now in Romania. Orthodox Jews in Israel might still today refer to communities of ultra-traditional “Hungarians” whose ancestors lived in towns that are now part of Romania. (Included among these is the family of the famous writer Elie Wiesel)

    The fate of the Hungarian Jews in WW2 was particularly brutal. The deportations were in full-swing there for only about a year (after the German takeover in the spring of 1944) but managed to kill about a half or two-thirds of the country’s Jews during that time.

  11. What about cryptic Sephardic ancestry in US non-Hispanic populations? MyHeritage showed me having 10.3% North African Sephardic ancestry.

  12. This is a question of some personal interest to me.

    I have southeastern Polish (Christian) heritage on my father’s side which both AncestryDNA and 23andMe say includes trace Ashkenazic ancestry at the 1% and 0.3% levels, respectively, with several Ashkenazic cousin matches. MyHeritage on the other hand shows 0% Ashkenazic but 1.1% Middle Eastern. These results were surprising, and was unsure how to interpret them in the context of what I know of family history.

    Given the small percentages, I was tempted to dismiss them as an error in their models, maybe showing Slavic ancestry in their Ashkenazic reference populations rather than vice versa, or even just noise in the data. However, if as Kevin Brook says above this is a frequent and robustly supported result among southeastern Poles and western Ukrainians, then perhaps it’s real after all.

  13. I meant to add that the Slovakian Rusyns with Ashkenazic percentages correspondingly have Ashkenazic autosomal cousin matches.

  14. I think i tried posting before, please let me know if it broke some rules. I had just two things to point out. Firstly, more than half of the Ashkenazi admixed Hungarians in the report had VERY recent admixture (a parent or a grandparent of Ashkenazi heritage), and most of the rest had an Ashkenazi great grandparent. Wouldn’t it be too late in history to tell anything about the assimilation in the XIX .. XVIII c.?
    Secondly, Myheritage markets itself as an expert in non-Ashkenazi Jewish ancestry, especially Sephardi. Many people of Eastern European origin have legends of the ancestors coming from the Medieval Spain, from the lineages of famous Jewish sages of yore. Many hope that Myheritage testing will validate their legends…

  15. From the results I have seen, which is not that many, I got the impression that 23andme is by far the best for estimating Ashkenazi ancestry, while some other tests constantly over- or underestimate this.
    23andme is not always superiour, but their ability to estimate Ashkenazi is, for whatever reason.

    One of the simplest ways is to check in people which background is investigated and proven. F.e. A typical one quarter AJ gets between 23-26 percent regulatarly. Same people in other tests get odd results with percentages add or lost in an incomprehensible way.

    So I would trust 23andme more on that issue. Or did the other companies improve their accuracy?

  16. RE: 23andme vs My heritage on Sephardi

    My guess is that the relative performance has to do with each company’s base. I’m making the assumption that each company’s models were parametrized on their initial customer base.

    MyHeritage is an Israeli company — Israel is home to pretty much all of the world’s Sephardic population. 23andme is American and American Jews are overwhelmingly Ashkenazi. 23 does not even have a Sephardi category.

    It’s also worth remembering that the Ashkenazis and North African Sephardim run in a relatively continuous cline from Sicilians to BedouinA (based on a Reich PCA).

    A model fit on this data with both sets of data would presumably allocate more “marginal” Ashkeazis to some Sephardic. So, for example, I have 99.6% Ash on 23andme while MyHeritage yields 5% Sephardi. To my knowledge, however, there is no recent Sephardic heritage in my family.

  17. @Obs.

    No, there was about 5% Finnish and 5% Middle Eastern in MyHeritage. I also have no knowledge of Finnish or non-Jewish heritage. Feels like these were residuals necessary to make the other parts of the model fit.

    Also, I say “about” b/c I don’t want to log in from work and get the precise info.

  18. Slight correction to the above — downlaoded the app.

    It’s 2% Finnish, 4% “West Asian” and 4% “Sephardic Jewish – North African”

    MyHeritage has separate categories for

    Ashkenazi
    Sephardi
    Mizrahi (Iraq/Iran)
    Ethiopian Jew
    Yemenite Jew

    Also, separate categories for general West Asian and Middle Eastern…

  19. What D-T-C companies try to do with reliability is tell you what your grandparents were; and delineate some ancestry as far back as 6 generations. What none of them are interested in doing is explaining Ashkenazi ethnogenesis — for that “Davidski” does a better job.

    I compared IBD Total cM between current “Sephardic” samples (a remnant peoples actually) with Ashkenazi samples AND a 3rd party. Ashkenazi shared more IBD with Lithuanians and less with North Africans than Sephardic. But (seemingly counter-intuitive) Ashkenazi shared more IBD with French Basque and were very close to even with samples from Spain (with Ashkenazi holding slightly more IBD cM with Spain). This leads me to believe that Kevin Brooks is on the money with his narrative of Jews fleeing Iberia 500 years ago and those “Sephardic” individuals that remained or returned to Judaism often gravitated to Western and Eastern Europe becoming part of the Ashkenazi mix.

  20. Eric, thats exactly what I meant, they dont get it right with their calculation. I wonder if 23andme used a segment shared approach of a different kind.
    Because with 1000 AJ samples you can recombine every AJ persons genome. That’s not possible with most other people in Europe.

    Its almost as if they use a different calculation for AJ, since other peoples results might follow a different direction.
    But I don’t know, just my impression.

  21. @Sgt What D-T-C companies try to do with reliability is tell you what your grandparents were
    @Eric K I have 99.6% Ash on 23andme while MyHeritage yields 5% Sephardi

    When the genetic composition of the different ancestors is so radically different that even a tiny segment of DNA yells about its origin (“I came from a Neanderthal!!!”), then it’s possible to reliably assign ancestry to a small fraction of one’s genome.

    But when the putative ancestors have as much in common in the DNA as Sephardi / Ashkenazi Jews, then it takes a fairly large ancestral contribution to the descendant’s genome to assign the fractions reliably. And as Sgt mentions, the customary yardstick is the ability to identify a grandparent. MyHeritage capitalizes on genetically similar, and internally diverse, pairs of reference sets to separate Sicilian from peninsular Italian, Greek from Slavic Balkan, or indeed Sephardi from Ashkenazi DNA at the grandparent level.

    The flip side, however, is the inevitable noise in ancestral fraction assignments between the similar ancestral groups, as strong as 10%. But of course those of us who want to confirm uninformed family legends may be all too happy to be told that 5% or 10% of their Ashkenazi DNA “is” Sephardi / Iberian / Mid-Eastern.

    The situation is even more stark with the obsolete j-test of Eurogenes Project. Its creator warns that the j-test doesn’t even have the power to identify a Jewish grandparent; the noise is too strong to use it beyond the parents’ level. But the j-test, too, makes people happy when their family legend tells them that they should have some “Jewish DNA”, yet the better tests fail to see any…

  22. The J-test is completely unreliable in my opinion. I know of false positives/negatives. 23andme on the other hand seems to get AJ always right to the point.

  23. At the time Eurogenes came out with the J-Test (2012) 23andme had more limited reference populations to work with and they were identifying individual customers as 87-94% Ashkenazi based on ~28% of their genome. Razib can probably speak to this better than I but it appears that regardless of the underlying segment of ancestry (Levantine, West-Asian, European or Exotic) if these segments were held by other Ashkenazi individuals -by the algorithm of “Most Likely”- this was who you were. As 23andMe’s customer data base swelled with self-reported ancestry the confidence levels increased.

    I believe that MyHeritage relies on a newer version of Joe Pickrel’s supervised version of Structure (perhaps with some modification based on current IBD-??) … Nonetheless, the “Ashkenazi” signature dates to no earlier than the 10th Century. Beyond this point the Alder curve gets very noisy trying to identify a single admixture event.

    I would speculate that 10th Century Jews from Iberia, France, Italy and North-Africa were panmictic (and well-traveled) and Fst differences we see today between Behar’s 19 samples of “Sephardic” from Turkey and Bulgaria with Ashkenazi samples are largely a result of drift.

  24. 23andme didnt just add more samples, they completely altered the calculation.
    The old AJ calculator worked in a completely different way.
    In the white paper 23andme estimates the accuracy for AJ almost on the same level as European vs non-European and thats what they deliver.

    Only Finns come close to that according to them for obvious reasons. Before the latest updates U saw more false Finn positives than AJ though, but they got rid of that largely since then. So their Finn radar improved as well.

  25. I would speculate that 10th Century Jews from Iberia, France, Italy and North-Africa were panmictic
    Is it really necessary to posit panmixia to explain low Fst? A more parsimonious explanation would be that various Jewish groups in Southern Europe formed as a roughly 50:50, sex-biased admixture between ancestors of Eastern and Western Mediterranean origin.

    Strong drift in the Ashkenazi branches (or, rather, most likely in its most populous sub-branches out East) is well documented but it’s hard to leverage its effects to reliably assign ancestral fractions at low percentages. Phasing / haplotype-based ancestry assignment methods like 23andMe work more reliably owing to much more dramatic drift effects on the local haplotype frequency spectra. The reasons why MyHeritage algorithms work at all may be related not to the drift per se, but to an additional minor Balto-Slavic admixture received by the ancestors of the Eastern Askenazim in XVI-XVIIth centuries…

Comments are closed.