It never ends with these Denisovans. More and more results. Filling the gaps in our knowledge.
Denisovan DNA found in cave on Tibetan Plateau:
For today’s Buddhist monks, Baishiya Karst Cave, 3200 meters high on the Tibetan Plateau, is holy. For the ancient Denisovans, extinct hominins known only from DNA, teeth, and bits of bone found in another cave 2800 kilometers away in Siberia, it was a favorite rest stop. Last year, researchers proposed that a jawbone found long ago in the Tibetan cave was Denisovan, based on its ancient proteins. But archaeologist Dongju Zhang of Lanzhou University and her team were on a quest for more definitive evidence. So in December, 2018, they began to dig. This week, Zhang’s team reports the first Denisovan ancient DNA found outside Denisova Cave: mitochondrial DNA gleaned not from fossils, but from the cave sediments themselves. Precise dates show the Denisovans took shelter in the cave 100,000 years and 60,000 years ago, and possibly as recently as 45,000 years ago when modern humans were flowing into Asia.
The result is pretty straightforward. Tens of thousands of years of occupation by “Denisovans” in highland Tibet.* The introgression into modern Tibetans, probably via a deeply basal “East Eurasian” Paleolithic lineage absorbed by rice farmers who migrated to the plateau in the last 10,000 years, makes more and more sense.
Related to this is a new paper on dogs, Ancient Dog DNA Shows Early Spread Around the Globe. I’ll have a piece out on this in Quillette shortly (as well as a podcast with the first author on my substack. But here’s one aspect: the Tibetan mastiff which has high altitude adaptation from Tibetan wolves doesn’t seem to show wolf admixture.
So what’s going on? Well, very small levels of genome-wide admixture can still bring in beneficial adaptive loci. Perhaps this is more common in humans than we might imagine.
* I put “Denisovans” in quotes because it is becoming clear East Eurasian hominins were a diverse group of populations with larger effective population sizes than West Eurasian hominins, Neanderthals. Denisovans are a clade, but one with many branches.
Somewhat related
Published Salkhit paper out – https://science.sciencemag.org/content/370/6516/579
qpGraphs in the supplement: https://imgur.com/a/KQtgFS2
Interesting to me in that they show the GoyetQ116 to share a very small amount of drift with the “ANE” lineage that leads to Mal’ta, Yana and contributes to Salkhit, after the “ANE” lineage splits off from the Kostenki14 lineage. Then GoyetQ116 has some geneflow from early East Eurasians, and then later West Eurasians in Upper Paleolithic Europe (possibly Sunghir, likely Vestonice cluster) are a mix of the GoyetQ116 and Kostenki14 related waves.
Seems to suggest that GoyetQ116 was actually a later / further east pop wave after K14 wave that exchanged genes for slightly longer with the proto-East Eurasian early splits, of which Tianyuan is an example that experienced no geneflow with early West Eurasians (like a Kostenki14 of the East), while Salkhit has much more extensive mixture with proto ANE.
That’s surprising to my intuition given that GoyetQ116 is only slightly later than K14 and sampled further west. But this seems to be how it is in their graphs.
Also, to reconfirm what I said when this was a preprint, makes me think that modern East Eurasians (and apparent early splits from them with no Onge related flow, e.g. Jomon!), who have no geneflow from early “ANE” splits that is present in Salkhit, must have been from a East Eurasian split that was located further South than Tianyuan/Salkhit, and wouldn’t have had opportunity for geneflow with these “ANE” / early West Eurasian branches. Then expanded north more recently with low enough levels of geneflow that it is statistically non-significant.
Need dna between Salkhit and late Upper Paleolithic to work out when this happens…
Re; Pontus’ comments on a human vector of a rapid radiation of domesticated dogs around 30-15kya (“How did that happen?” he said. “In ancient humans, we don’t really know of any human expansion that would have facilitated this, on the order of 15 to 30,000 years ago.”), need there even be one?
Maybe dogs that were “a little bit domesticated” spread themselves? There are human hunter gatherer camps about that they can scavenge from all over the Old World at this point, so maybe there is an adaptive radiation of semi-domesticated dogs that doesn’t require a specific human population movement…. Or maybe that’s crazy.
It is said the Neantherthals/Denisovans moved around in smaller tribes then the Dapiens. Did this increase speciation or heterogeneity in the overal population due to more inbred populations maintaining mutations and homogeneity within tribe more easily? Or the other way around because they needed to marry off to other tribes more readily to offset this?
@Matt
Could the dogs have been dispersed along northern Eurasian HG networks? Something like that was proposed for the more recent spread into melanesia.
I’m calling BS on the zero wolf ancestry statement. I suggest instead that they only tested a narrow set of breeds.
I’ve seen an introgression from a wild species “the dingo” into a herding dog line. Today the wild ancestor would have 100s of domesticated descendant and seemingly because the introduced traits were extremely beneficial.
Of course it’s possible that those genes will be selected against in the future. BUT cases of deliberate introgression of wild canine species into herding, guard, sled, and scent breeds were common. So it’s seems likely that some wild genes are beneficial.
mtdna from cave sediments really cool btw; it’s kind of “boring” from the general NGS autosomal perspective I usually have where mtdna is not as great for really assessing overall population relationships (one locus) but this kind of thing is great for characterising overall landscape of fauna in some region, beyond what archaeological record provides.
Couple other thoughts on dog paper:
1) anything in dog paper about the y and the mtdna?
2) kind of interesting that the European dog lineage that seems to replace all others in Europe (if this is not just a function of limits of their resolution), and then expand across the world is from Funnelbeaker. I am kind of reminded of how Strabo talked about how Britannia was known for exporting dogs, and this post by Anton Howes (https://antonhowes.substack.com/p/age-of-invention-observing-the-occident) talking about how later again Mughal traveller to Britain regarded the dogs as remarkable “they are taught to perform many wonderful and surprising things, which the common people of this country do not believe a word of”, and there is generally an idea in the modern consciousness of a somewhat different status of dogs in Europe. Perhaps there is a somewhat long history of dog breeding for improvement in Europe generally…?
@DaThang, yeah, a sort of trade in / theft of animals seems like a very plausible model too, I’m just suggesting an alternative.
@Peter, it seems like the claim is little introgression of bulk autosomal ancestry, not functional genes and traits. Maybe it could be that some early wolf crossbreeds were “so good” that they were systematically bred into waves of new dogs brought in, diluting the original wolf ancestry to very low levels?
Could be that the typical scenario with wolf breeding into dog lines was that some small set of wolves was bred into the dogs of a first founder population into a region, and then a lot of the wolf ancestry got bred out by new waves of dogs – which were probably tamer, more domesticated, had lots of beneficial alleles from larger population size – while retaining the adaptive variants with strong fitness advantage, for’ex the high altitude adaptation Razib mentions in Tibetan Mastiffs…
@Razib,
A little off topic, but, what do you think of multiple ‘Out of Southern Eurasia’ events to populate rest of Eurasia?
Using Vahaduo with G25 coordinates, I found out that Lithuanians, Ulchi and Aus. Aboriginals are all significantly closer to Onge than they are to each other. (Paniya is also roughly equidistant between Caucasians and East Asians.)
The fact that Ancient Southern Eurasians are intermediate between West Eurasians, East Asians and Aboriginals could mean they are the source population for these ‘peripheral Eurasian’ groups, which drifted further apart from each other because of ANA and archaic (Neandersovan) introgression.
Your opinions, please.
Thanks for the image in the first post Matt. The Salkhit paper makes early Gravettians and Aurignacians seem mostly like a mix of a Kostenki group and an otherwise very widespread Aurignacian group. I wonder where common west Eurasians (WHG included) fit into all of this.
A and B are preferred over C and D. According to A, the main ancestors of ANS, ANE and Aurignacians comes from the same group, while Kostenki is a completely different clade, Sunghir is mostly Kostenki with a little Aurignacian-like mix and Vestonice is a split mix between the two.
According to B, Gravettians, Aurignacians and early pre-Gravettian Europeans make one clade to the exclusion of the west Eurasian ancestors of ANS and ANE (this also happens to give ANS and ANE a little bit more west Eurasian ancestry). In this scenario, both Sunghir and Vestonice are near even mixes of para-Kostenki and para-Aurignacians, with Sunghir being just a little more of the former and Vestonice more of the latter.
I think that scenario A is more interesting, with two very distant west Eurasians making a clade to the exclusion of Kostenki 14 (and company) who were geographically located in between. It is also curious how it got a slightly lower z-score absolute value than scenario B.
@J Khan
G25 is not a suitable tool for that kind of thing; intermediate position mostly means doesn’t have a lot of drift shared with other modern populations.
What % Neanderthal ancestry do Andamanese like Onge have?
@Megalophias
With g25, Onge are at a distance of roughly 40 from West Eurasians whereas East Asians are at a distance of 60 from West Eurasians.
You said g25 is not suitable for this kind of thing — do you mean if we used some other tool we won’t see this significant difference?
In@J Khan, Megalophias is right; David didn’t design the G25 to include the Onge’s specific drift (e.g. probably their coordinates are projected by him, and/or he has strongly undersampled them intentionally when creating the PCA) and so it underestimates their differentiation. (This is OK for the purposes he designed it for, as the Onge are so far a clade to all the other world populations for which G25 is mainly used for ancestry estimation). In reality, under direct measurement of genetic distance by Fst, Onge are quite a bit more differentiated from West Eurasians than East Eurasians (more drift) and phylogenetically speaking are equally differentiated.
I will post a G25 distance^2 vs Fst cross-plot that shows this directly if you would like.
@jkhan: Fst vs G25 Euclidean Distance ^2 crossplot: https://imgur.com/a/c9wSkoC
The labels positioned at the centroid (average) of population pairs involving the label as one of the populations in the comparison.
You can see that there is a pretty good linear relationship between the G25 Euclidean Distance ^2 and Fst for these modern populations (this is in line with Florian Prive’s recent paper showing the relationship between Fst and euclidean distance on PCA, and other previous ones comparing PCA results and Fst).
But! Onge are an outlier in the comparisons, with their Fst differentiation not represented in G25 distance. This is because David’s just not made the G25 to reflect that (and this is arguably fine for his purposes and he’s no doubt well aware of it, just something for users to be aware of!).
In general, looking at distances from high quality, high dimensions, well projected PCA (like the G25, and in general David has pretty good practices), can be a good substitute for Fst in some instances, for some ancient samples in particular.
These are when the population is thought to have contributed to later populations and so its projected position is more likely to well reflects its overall drift, and where conversely the Fst itself might be likely to be strongly affected by damage, real very recent founder effects (which you’re uninterested in), and batch effects (like shotgun vs capture arrays) that might inflate (or deflate) Fst. There are some highly damaged samples where Fst is particularly misleading (and I personally suspect it is generally somewhat slightly inflated in many ancient sets, like the WHG).
But when it comes to ancient populations which are not well represented in modern people, it can also be misleading – for an example if you look at the G25 distance of the Upper Paleolithic Sunghir population from others, its pretty low, but in reality the Fst is actually quite high (and this is because Sunghir’s unique drifts are not well represented in modern people, as they either didn’t contribute to later people, or did so after a lot of admixture which homogenized out their specific drift).
In the same way, if you have two modern populations and you have high quality, comparable data, and you want some understanding of their level of differentiation, it is better to just look at the Fst directly, without projecting them both onto a secondary PCA first.
This is particularly true any differences that might not be reflected on a PCA, like the one above for Onge, but also fine-scale differentiation within a region (for instance, fine scale differentiation between subregions within a continent that might not be well represented in a PCA’s dimensions, even a relatively high dimension one like G25).
So yeah, overall, unfortunately, the distance of Onge on G25 isn’t telling you something unique about the Onge’s, and wider South Eurasians’, position relative to West Eurasians and East Asians, it’s just telling you that the Onge’s genetic drift is not v well represented in G25.
That was an excellent explanation, thanks Matt.
@jkhan: No problem; that’s all as I understand it / guess it, hopefully accurately. You could email/PM David to double check it. I don’t think he’s big into going into detail of how he constructs and projects PCA on public websites, but he might confirm privately. Just to underscore and repeat myself a bit, and I know you’re not saying this, not because he’s done anything wrong in making the PCA – it’s just that there will be these inevitable tradeoffs in creating a finite dimension PCA, and where some of populations being examined are projected. He’s built it to be informative for most of the uses its userbase will put it to, and that’s probably better for that (than having dimensions among his 25 that are essentially empty space that just contain the Onge at a distance from all other samples). And in some instances of ancient dna it is useful to look at these distances get a sense check if Fst scores calculated seem to be out of whack. But yeah, in many instances if it seems that a PCA is producing an unexplained low distance or sharply different distance to Fst, these things are probably responsible.
@DaThang, good summary of those qpAdm trees, I think you’ve pointed out some things I missed by glancing at the figures tbh (probably didn’t take enough time to digest them properly).
In terms of WHG, seems to me that we’ll probably need more sampling of Dzudzuana samples, which may have contributed to Late Upper Paleolithic / Mesolithic Europeans, to understand that.
That one sample in the preprint seems have that unusual characteristic of having relatively high affinity to Ust Ishim compared to Anatolian Early Neolithic, which may indicate low Basal Eurasian, while also of possibly being able to provide the Near Eastern affinity to LUP / Meso which Early UP lack, without inflating the Basal Eurasian. That seems to make it promising to explain the LUP Europeans.
It seems reasonably plausible to me that the LUP are a combination of the Vestonice / Gravettian group (with middle levels of Aurignacian (GoyetQ116) and K clade) and Dzudzuana, probably from SE / SC European refugia, without much additional contribution from SW European refugia (Magdalenian / El Miron group), where the Aurignacian ancestry seems higher… But then *some* contribution of Magdalenian group into some of a specific “WHG” subset of LUP / Meso Europeans found mainly in Western Europe. Probably some North Eurasian group geneflow, mainly E Europe, too.
On dog paper again, I wonder if having this understanding of the phylogenetic structure of early dogs now allows us to reconstruct the shape skull and body of the earliest domesticated dogs?
There’s a lot of high resolution data about dog skull and body shape (and even brain volume and shape – https://www.jneurosci.org/content/39/39/7748), so it seems like once you have a fairly good tree structure and you can then weight how much importance you place on each existing lineage, you can lump it into models like these ones to reconstruct the Homo Sapiens LCA – https://www.nature.com/articles/s41467-019-11213-w#Sec6
To go on a bit of a tangent to an above comment about Fst and distances on PCA because I’m at a loose end and someone might find it interesting, I said above: “I personally suspect (Fst) is generally somewhat slightly inflated in many ancient sets, like the WHG”.
Here’s a quick example of why I think that may be so.
Previously I said that the squared distance in Global 25 correlates fairly well with Fst, for most modern sets I looked I could compare, with some outliers which have high levels of their own drift: https://imgur.com/a/DhcR8kS
Remove the outliers (which I’ve set as in this set as Igorot, Kalash, Ket, Onge, Saami), and the relationship is even stronger: https://imgur.com/a/xv3Nrou
You can then get a good power equation predicting Fst from G25 distance: https://imgur.com/a/1DUcnat (R^2 of 0.98 for 3542 Fst for 60 populations; Equation: 0.258*Global_25_Distance^1.62).
Now what happens when I plug ancient Global 25 distances into that equation to estimate Fst and compare to direct estimate of Fst? See: https://imgur.com/a/EuG7P0x
The residuals can be fairly substantial for some comparisons, at the most extreme, Fst for Karelia_HG:WHG is 0.105 directly, but only 0.041 estimated for the G25 distance, while CHG:Iran_N is 0.058 directly but only 0.017 estimated for the G25 distance (substantial but only comparable to present day differentiation between Caucasus and North Europe, rather than 3x as substantial). The residuals are more modest comparing Yamnaya:Anatolia_N at 0.055 vs 0.044 or WHG:Anatolia_N at 0.103 vs 0.073, or Yamnaya:Globular_Amphora_Farmer at 0.043 vs 0.034. Comparisons between modern and ancient populations drop a bit; Globular_Amphora/Yamnaya:Modern North European is about 0.02 vs 0.01 (about as differentiated as Lithuanians and Central Italians today).
This is all dependent on G25 having about the right level of projection for ancient samples though – you could argue that the reason that there is a slight mismatch is because the G25 PCA under projects ancients, making them too close to present day populations and each other. And there might be real founder effects in these population that aren’t captured in those PCA. But I think it is an interesting experiment.
Hey matt, the tagliente paper modeled later WHGs (villabruna 1) as either being mostly tagliente + 10% vestonice or mostly tagliente + 5% Goyet. But according to model A from the salkhit paper, these 2 scenarios are pretty much the same thing (vestonice being 50% goyet). Especially if tagliente nearly entirely belongs to the kostenki/para-sunghir clade of west Eurasians (from the ‘A’ model only since I am only considering it, not B).
So in this scenario, Vestonice has half kostenki/para sunghir ancestry while tagliente as a cousin is almost entirely within the kostenki/para sunghir group (probably more than even sunghir since sunghir may have some goyet ancestry).
Though I suspect that while tagliente belongs to that general broad kostenki type clade, I still think that it isn’t descended from the known European Gravetitan cluster (even with their Goyet ancestry excluded) and was only a cousin to them in the same clad to the exclusion of Aurignacian and Siberian West Eurasians***.
***this does not however rule out minute ane input in tagliente (mtdna u4’9), it would still be mostly/almost entirely kosteni/para sunghir like.
tl;dr the kostenki (let’s call it cwe) group produces a line leading towards tagliente and a separate line leading towards gravettians. The line leading toward gravettians mixes with the goyet group to make gravettians and the tagliente line gets minor ane to result in tagliente.
What do you think about this?
@DaThang, forgot about that paper! Unless I’m reading wrong (paper or your comment), don’t their Extended Fig 4 and 5 models all have Tagliente / Villabruna weakly form a clade with GoyetQ116 to the exclusion of K14?
It did like the older dzudzuana preprint which did the same and had ane on the kostenki side on top of that, but the salkhit paper has a different arrangement. Not only that but I haven’t seen earlier papers model Sunghir with some goyet admixture. So there are some new approaches in the recent paper.