Before there was Structure there was just structure. By this, I mean that population substructure has always been. The question is how we as humans shall characterize and visualize it in a manner which imparts some measure of wisdom and enlightenment. A simple fashion in which we can assess population substructure is to visualize the genetic distances across individuals or populations on a two dimensional plot. Another way which is quite popular is to represent the distance on a neighbor joining tree, as on the left. As you can see this is not always satisfying: dense trees with too many tips are often almost impossible to interpret beyond the most trivial inferences (though there is an aesthetic beauty in their feathery topology!). And where graphical representations such as neighbor-joining trees and MDS plots remove too much relevant information, cluttered FSTmatrices have the opposite problem. All the distance data is there in its glorious specific detail, but there’s very little Gestalt comprehension.
The Pith: New software which gives you a more fine-grained understanding of relationships between populations and individuals.
According to the reader survey >50 percent of you don’t know how to interpret PCA or model-based (e.g., ADMIXTURE) genetic plots, so I am a little hesitant to point to this new paper in PLoS Genetics, Inference of Population Structure using Dense Haplotype Data, as it extends the results of those earlier methods. But it’s an important paper, and at some point I’ll starting using their software. The “big picture” is that earlier methods left “some information on the table.” That’s partly due to the fact that they were developed (or in the case of PCA leveraged, as it’s a very general technique) in an era where very dense marker data sets were not available (today we’re shifting to full genome sequences in many cases!). The information left on the table would be haplotype structure. Genetic variation in a concrete form manifests as sequences along a line, many of them physically connected. These correlations of nearby variant markers represent haplotypes of great interest, because they are excellent clues to admixture or divergence events across populations. In contrast the older methods, were looking at variation from marker to marker, each in turn independently, which collapses some of the important genomic structure that we can now inspect (in fact, linkage disequilibrium due to these correlations can distort some of the results in the older methods, so you want to “thin” your marker set).
Let me make this concrete for you. On 23andMe you can see where your friends shake out on a PCA plot using the HGDP data set as a reference. What this means is that the HGDP data set is used to generate independent dimensions of genetic variation. As is the usual case in these analyses the largest dimension separates Africans from everyone else, and the second largest dimension separates Asians from Europeans and Africans. 23andMe customers are then projected upon this variation, so you can get a sense where you are positioned in the clusters. To the left is a zoom in on the section for Central/South Asians. You can see that one of my friends, highlighted with a green color, falls almost perfectly in the Uygur cluster. According to ancestry estimates my friend is 50 percent Asian and 50 percent European. The “representative” Uygur in the 23andMe chromosome painting gives about the same results. But these are total genome estimates. The historical nature of my friend’s admixture and that of the Uygur woman is very different, as one can see in the below figure.
Representatives of Szechuan and Shangdong cuisine
The Pith: The Han Chinese are genetically diverse, due to geographic scale of range, hybridization with other populations, and possibly local adaptation.
In the USA we often speak of “Chinese food.” This is rather peculiar because there isn’t any generic “Chinese cuisine.” Rather, there are regional cuisines, which share a broad family similarity. Similarly, American “Mexican food” and “Indian food” also have no true equivalent in Mexico or India (naturally the novel American culinary concoctions often exhibit biases in the regions from which they sample due to our preferences and connections; non-vegetarian Punjabi elements dominate over Udupi, while much authentic Mexican American food has a bias toward the northern states of that nation). But to a first approximation there is some sense in speaking of a general class of cuisine which exhibits a lot of internal structure and variation, so long as one understands that there is an important finer grain of categorization.
Some of the same applies to genetic categorizations. Consider two of the populations in the original HapMap, the Yoruba from Nigeria, and the Chinese from Beijing. There are ~30 million Yoruba, but over 1 billion Han Chinese! Even granting that the Yoruba seem excellent representatives of Sub-Saharan African genetic variation (not Bantu, but not far from the Bantu), there are still more Han Chinese than Sub-Saharan Africans (including the African Diaspora). So it’s nice that over the past few years there’s been a deep-dive into Han genetics. A new paper in the European Journal of Human Genetics focuses on the north-south difference among Han Chinese, using groups flanking them to their north and south as references, Natural positive selection and north–south genetic diversity in East Asia.
Some have asked what the point is in poking around African population structure when Tishkoff et al. and Henn et al. have done such a good job in terms of coverage. First, it is nice to run your own analyses so you can slice & dice to your preference, and not rely on the constrained menu provided by others. There’s value in home cooking; you can flavor to your taste. Second, you never know what data people might leave on your doorstep. I’ve received the genotypes of three Somalis. Nothing too surprising, a touch more Cushitic than the Ethiopians in Behar et al., but interesting nonetheless.
Also, you can see how ADMIXTURE tends to come to weird conclusions in certain circumstances. Below is a K = 12 run ~50,000 SNPs. I’ve included in a few Behar et al. and HGDP populations to the Henn et al. set, as well as pruned a lot of the African groups which seem redundant in terms of information. I’ve added a few geographically informative labels as well.
Observe below that there is a Fulani cluster. I think this is pretty much an artifact. At K = 7 the Fulani have a majority component which is modal in West Africa & Bantu speakers, and a minority component which is identical to the one modal in Mozabite Berbers from Algeria. The Mozabites reside in the far northern Sahara, and their modal component drops off as one goes east toward western Asia and the eastern Mediterranean. I suspect that what is showing up in ADMIXTURE is the ancient hybridization of the Fulani, and perhaps their demographic expansion from this core group. We have some glimmers of the prehistory of the Fulani, and no expectation for them to be such a distinctive cluster, so I naturally jump to these inferences. But it does make me reconsider the nature of the “Sandawe,” “Mbuti” or “San” clusters in ADMIXTURE. These populations are culturally distinctive in deep ways from their neighbors, so a reflexive inference one might make is that they’re “pure” ancient substrate groups which have been overlain and marginalized by their Bantu neighbors. But their prehistory is far murkier than the Fulani because of their geographical isolation, so there is far less to go on. These “ancient” isolated groups themselves may have gone through the same sort of distinctive recent ethnogenesis processes which we presume occurred with the Fulani (also, in the plot below the Biaka are pure; but in most of the bar plots they have a minor element which they share with their neighbors, probably due to greater admixture and interaction between western Pygmies and their Bantu neighbors than among the easter ones).
As you have begun interpreting the reference results, let me make a friendly warning: you have to keep in mind that most of the reference populations of ethnic groups are extremely limited in sample size (with only between 2 and 25 individuals) and from very obscure sources, and you should keep away from drawing conclusions about millions of people based on such limited number of individuals.
This seems a rather reasonable caution. But I don’t think such a vague piece of advice really adds any value. These sorts of caveats are contingent upon:
- The scope of the question being asked (i.e., how fine a grain is the variation you are attempting to measure going to be)
- The sample size
- The representativeness
- The thickness of the marker set (10 autosomal markers vs. 500,000 SNPs)
The pith: In this post I examine the most recent results from 23andMe for my family in the context of familial and regional (Bengal) history. I also use these results to offer up a framework for the ethnognesis of the eastern Bengali people within the last 1,000 years, and their relationship to other South Asian and Southeast Asian populations.
Since I received my 23andMe results last May I’ve been blogging about it a fair amount. In a recent post I inferred that perhaps I had a recent ancestor who was an ethnic Burman or some related group. My reasoning was that this explained a pattern of elevated matches on chromosomal segments with populations from southwest China in the HGDP data set. But now we have more than my genome to go on. This week I got the first V3 chip results from a sibling. And finally, yesterday the results from my parents came in. One thing that I immediately found interesting was my father’s mtDNA haplogroup assignment, G1a2. This came from his maternal grandmother, and as you can see it has a distribution which is mostly outside of South Asia. In case you care, I asked my father her background, and like my patrilineage she was a “Khan,” though an unrelated one (“Khan” is just an honorific). I received these results before the total genome assessment, and so initially assumed this confirmed my hunch that my father had some unknown recent ancestry of “eastern” provenance. But it turns out my hunch is probably wrong. In fact, my parents have about the same “eastern” proportion, with my mother slightly more! My expectation was that perhaps my mother would be around 25-30% “Asian,” and my father above 50%. The reality turns out that my father is 38%, and my mother 40%.
Image credit: f_mafra
Below are the “Ancestry Paintings” generated by 23andMe for my family (so far). What you see are the 22 non-sex chromosomes, which have two copies each, and assignments to “Asian,” “European,” and “African,” ancestry groups. The reference populations to generate these assignments come from the HapMap, the northern European sample of white Americans from Utah, Chinese from Beijing, Japanese from Tokyo, and ethnic Yoruba from Nigeria. What the assignment to one of these classes denotes is that that region of the genome is closest to that category in identity. It does not imply that your recent ancestry is European or Asian (African is probably a different matter, but there are many complaints about the results for African Americans and East Africans in the 23andMe forums). This caveat is especially important for South Asians, because we generally find that we’re ~75% European and ~25% Asian. All that means is that though most of our genetic affinity is with Europeans, a smaller fraction seems to resemble Asians more. Via “gene sharing” on 23andMe I can see that the Asian fraction varies from ~35% in South India and Sri Lanka, to ~10% in Pakistan and Punjab. This is not because South Indians have more East Asian ancestry than Punjabis. Rather, to a great extent the South Asian genome can be decomposed into two ancestral elements, one with a distant, but closer, affinity to populations of eastern Eurasia, and one with a close affinity to populations of western Eurasia. What some have termed “Ancient South Indians” (ASI) and “Ancient North Indians” (ANI). ASI ancestry, which is probably just a touch under 50% in South Asians overall, seems to shake out then as somewhat more Asian than European.* The fraction of ASI increases as one moves south and east in South Asia (and as one moves down the caste status ladder).
In my post below I quoted my interview L. L. Cavalli-Sforza because I think it gets to the heart of some confusions which have emerged since the finding that most variation on any given locus is found within populations, rather than between them. The standard figure is that 85% of genetic variance is within continental races, and 15% is between them. You can see some Fst values on Wikipedia to get an intuition. Concretely, at a given locus X in population 1 the frequency of allele A may be 40%, while in population 2 it may be 45%. Obviously the populations differ, but the small difference is not going to be very informative of population substructure when most of the difference is within populations.
But there are loci which are much more informative. Interestingly, one controls variation on a trait which you are familiar with, skin color (unless you happen to lack vision). A large fraction (on the order of 25-40%) of the between population variance in the complexion of Africans and Europeans can be predicted by substitution on one SNP in the gene SLC24A5. The substitution has a major phenotypic effect, and, exhibits a great deal of between population variation. One variant is nearly fixed in Europeans, and another is nearly fixed in Africans. In other words the component of genetic variance on this trait that is between population is nearly 100%, not 15%. This illustrates that the 15% value was an average across the genome, and in fact there are significant differences on the genetic level which can be ancestrally informative. You can take this to the next level: increase the number of ancestrally informative markers to obtain a fine-grained picture of population structure. In the illustration above the top panel shows the frequencies at the SNP mentioned earlier on SLC24A5. The second panel shows variation at another SNP controlling skin color, SLC45A2. This second SNP is useful in separating South and Central Asians from Europeans and Middle Easterners, if not perfectly so. In other words, the more markers you have, the better your resolution of inter-population difference. This is why I found the following comment very interesting:
After linking to Marnie Dunsmore’s blog on the Neolithic expansion, and reading Peter Bellwood’s First Farmers, I’ve been thinking a bit on how we might integrate some models of the rise and spread of agriculture with the new genomic findings. Bellwood’s thesis basically seems to be that the contemporary world pattern of expansive macro-language families (e.g., Indo-European, Sino-Tibetan, Afro-Asiatic, etc.) are shadows of the rapid demographic expansions in prehistory of farmers. In particular, hoe-farmers rapidly pushing into virgin lands. First Farmers was published in 2005, and so it had access mostly to mtDNA and Y chromosomal studies. Today we have a richer data set, from hundreds of thousands of markers per person, to mtDNA and Y chromosomal results from ancient DNA. I would argue that the new findings tend to reinforce the plausibility of Bellwood’s thesis somewhat.
The primary datum I want to enter into the record in this post, which was news to me, is this: the island of Cyprus seems to have been first settled (at least in anything but trivial numbers) by Neolithic populations from mainland Southwest Asia.* In fact, the first farmers in Cyprus perfectly replicated the physical culture of the nearby mainland in toto. This implies that the genetic heritage of modern Cypriots is probably attributable in the whole to expansions of farmers from Southwest Asia. With this in mind let’s look at Dienekes’ Dodecad results at K = 10 for Eurasian populations (I’ve reedited a bit):
One of the more popular posts on this weblog (going by StumbleUpon and search engine referrers) focuses on genetic variation in Europe as a function of geography. In some ways the results are common sense; populations closer to each other are more genetically related. Why not? Historically people have married their neighbors and so gene flow is often well modeled as isolation by distance. The scientific rationale for these studies is to smoke out population stratification in medical genetics research programs which attempt to find associations between genes and particular diseases. By population stratification I mean the fact that different populations will naturally have different gene frequencies, and if those populations exhibit different frequencies of the disease/trait under investigation then one may have to deal with spurious correlations. If, for example, your study population includes many people of African and European descent, presumably cautious researchers would immediately by aware of this problem and attempt to take it into account. But what about populations which are genetically closer, or whose genetic difference may not be so well manifest in physical characteristics which might clue you in to the issue of stratification?
That’s why the sorts of results which might seem common sense in the aggregate are useful. One can ask questions as to the genetic closeness of Irish and English, or Irish and Spanish, in a rigorous sense. In the United States research programs which are constrained to white cases and controls may hide population stratification because of the ethnic diversity of the American population. A primary motivation for studies of Jewish genetics are the cluster of “Jewish diseases” which are common within that population. In our age it is fashionable to focus on what binds us together as a species, but genetic differences matter a great deal. Ask the parents of multiracial children who require bone marrow transplants.
A new paper in Human Heredity examines a large sample of five European populations, and goes over the between population allele frequency differences with a fine tooth comb. Genetic Differences between Five European Populations:
A follow up to the post below, see John Hawks, Selection’s genome-wide effect on population differentiation and p-ter’s Natural selection and recombination. As I said, it’s a dense paper, and I didn’t touch on many issues.