Direct detection of natural selection in Bronze Age Britain:
We developed a novel method for efficiently estimating time-varying selection coefficients from genome-wide ancient DNA data. In simulations, our method accurately recovers selective trajectories, and is robust to mis-specification of population size. We applied it to a large dataset of ancient and present-day human genomes from Britain, and identified seven loci with genome-wide significant evidence of selection in the past 4500 years. Almost all of them are related to increased vitamin D or calcium levels, and we conclude that lack of vitamin D and consequent low calcium was consistently the most important selective pressure in Britain since the Bronze Age. However, the strength of selection on individual loci varied substantially over time, suggesting that cultural or environmental factors moderated the genetic response to this pressure. Of 28 complex anthropometric and metabolic traits, skin pigmentation was the only one with significant evidence of polygenic selection, further underscoring the importance of phenotypes related to vitamin D. Our approach illustrates the power of ancient DNA to characterize selection in human populations and illuminates the recent evolutionary history of Britain.
If you read this blog closely over the years you will know that I’ve noticed that the ancient DNA has yielded the general result that Bronze Age Europeans were somewhat darker in complexion than their modern descendants. This is without mass overall population genetic change in many areas. This is not always statistically significant, but you can tell a likelihood before something hits p = 0.05.
The ancient DNA temporal transects are pretty good for Britain. The best in Europe. And the above preprint seems to present a strong result of depigmentation over the last 4,500 years in Britain using powerful new methods drawing on ancient DNA sources. I strongly suspect this generalizes to much of Northern Europe.
there was a paper from 15 years ago that showed that andronovo ppl from the altai in siberia were basically colored like modern slavs. i am very very curious now whether this was contamination of some sort by russia lab workers.
From the British Isles I imagine it also helps this thesis that the Continent has been cut off for centuries at a time, when the Brits weren’t getting infused with their DNA. I understand that by now we have a decent handle on those periods of mutual isolation.
Meaning: the only genetic changes over that time will be drift and selection, in-house.
they state it explicitly in the paper
What is the speculative cause for a lack of vitamin D that would encourage such selection? What changed in their diets? Or were EEF missing health amounts too?
@razib, is there any chance you could ELI4 on what the method is and why it’s important here?
Being a sort of simple-minded person, I’d generally just look at graphs of allele frequency or polygenic score in a time series (like this one from Patterson 2021 on lactase – https://mobile.twitter.com/edhollox/status/1479463826709889026), and be all, like, duh it’s selection cos what else could it be? But what’s this paper doing that goes beyond that approach?
What’s the paper telling us that we would not know from slapping a polygenic skin colour score on all the transect samples and then fitting a curve?
@Otanes – It is spelled out in the discussion in the paper. Region of low solar radiation, so reduced Vitamin D synthesis from sunlight on the skin, and an agricultural (grain) diet poor in sources of dietary Vitamin D (and also Calcium). You need Vitamin D to process Calcium. Milk gives you both, so this could be one of the reasons for selection for lactase persistence after weaning as well.
All of that is such common knowledge (or I thought it was) that it seems possible I could be missing something in your question.
@Otanes, yes, it does seem like a puzzle why there aren’t similarly strong frequency changes in the first 3000 years of farming in Europe compared to post 4500 YBP? Could be that new diseases after neolithic increased selective pressure for Vit D + calcium, esp if the diseases hit bones, teeth?
Though that’s why it surprises me that they don’t identify that any of anthropometric traits themselves are under selection. If calcium intake is selected for strong or long bones, then wouldn’t the phenotype be under direct selection too? Unless those all had trade offs. I’m not sure I’m ready to give up on height being under selection when it seems robust to many methods – e.g. the SDS method, and also change in samples where post-BA Europeans generally seem > the product of sampled LN EEF and steppe people in the polygenic score (and measured long bones). Surely that is in-credible to believe to be due to drift?
I think this gives it to you:
https://globalsolaratlas.info/map
EEFs were getting more solar radiation during the first 3,000 years.
@John, that’s an argument. That, if I’m understanding you right, the EEFs were mainly evolving in Southern Europe, where they weren’t exposed to that selective pressure, and hadn’t been in Northern Europe for very long, where they would be selected for depigmentation.
But there’s some (still quite limited) evidence I’ve seen that could argue against that. There’s a fairly good Roman Imperial Era transect of samples, and I think we can assume that their ancestors were mostly living in either Italy or further South and East (based on the genetic models which work).
At SMBE last year Alba Refoyo-Martinez showed some figures which indicated on the PRS scores on pigmentation (lighter skin) for ancient samples: https://imgur.com/a/PFjGk40 (Caution: Not yet even a preprint!).
So we can see that that 1) post-Bronze Age Europeans generally seem to have lower scores than the product of EEF and Yamnaya would suggest, 2) there’s no apparent obvious general trend to lower PRS scores along the Anatolian Early Neolithic->Middle Neolithic European trend line. But also 3) the samples that represent Roman Italy also seem to have lower PRS than Yamnaya or MN Europeans (and this is tentatively so for the very few Spanish Bronze Age samples on the plot with steppe ancestry). Although less so than the very clear signal in Northern and Central Europe.
That seems a bit harder to square with an idea that selection for lighter pigmentation was about a different environment for farming, a cloudier one where EEF hadn’t been living for very long? If it seems also to happen for people whose ancestors stayed in that environment, or were living in an even sunnier one.
(We also know that the SLC45A2 variant that seems to be sweeping in Britain here is also about fixed in Tuscans and Northern Italy and rose in frequency at some point after the Middle Neolithic, and that it seems that so far from ancient dna, the Bronze Age immigrants into Italy with Steppe ancestry didn’t seem to have a higher frequency than the Copper Age populations who came before them. There was no systematic difference in frequency for this variant between Anatolian Early Neolithic and later European Middle Neolithic samples, the last time I saw this checked by Mathieson, and it was about 25-35% for both.)
I don’t think just looking at global solar output is a good marker of how much Vitamin D people would be getting. You also need to consider the overall climate of an area, along with the presence of a winter season. After all, subsistence farmers living in cold climates will likely be indoors most of the winter, whereas in overcast areas where year-round agriculture is possible they’ll still be outside and getting a modicum of Vitamin D.
@Matt – Oh. That makes it look very complicated – too much for my simple mind to take in.
@John, to be fair, very speculative on my part based on that presentation at SMBE and a few other things!
(But there is some published backing to that signal of selection on pigmentation being more evident in Italy in the genetically south and east -shifted post-Roman Republic sample. Saupe et al 2021, which looked at Neolithic, Copper Age, Bronze Age and post-Bronze Age samples from Italy – https://www.sciencedirect.com/science/article/pii/S0960982221005352 – “For the variants that are significant in both tests (TLR1 [rs5743618], TNF [rs1800629], HLA [rs3135388], and SLC45A2 [rs16891982]), the signal is driven almost entirely by the post-Roman Republic Central Italy sample group (Cen_postRep), which includes Roman, Late Antiquity, and Medieval individuals. There is no detectable difference between the Italian BA samples presented here and the Italian N/Chalcolithic groups, despite the additional Steppe-related ancestry.”.
The Fig S29 from the paper than introduced the Imperial Roman samples also suggests that SLC45A2 continued to increase despite the Near Eastern shift of the Roman pool, or at least didn’t decrease by enough to shift back to the Copper Age level – https://imgur.com/a/XYyyDqJ – though it’s not very fine grained on that enough to be very clear.)
One other suggestive constraint on the evolution of skin pigmentation in Northern Europe might be that within Estonians, individuals with sections of the genome around alleles which influence hair and eye colour who bear more Siberian ancestry at those regions tend to have darker hair and eye colour – https://www.cell.com/current-biology/fulltext/S0960-9822%2822%2900108-7
That would somewhat constrain things to suggest that the European ancestral population(s) to at least Estonians would have had lighter hair and eye colour variants than the Siberian ancestors already at the time of admixture, which would have been at least prior to the Iron Age (500 BCE).
It’s not much of a constraint but it is one.
….
In longer term I’m interested in whether a comparison with the North China Plain, when the transect there gets better. One the one hand, we have a good reason to believe that there is greater autosomal similarity maintained here since the Early Neolithic, compared to Britain (though not an absolute one). On the other hand, if the population expanded quite heavily from this region, evolved in slightly varying environments, and then backmigrated fairly frequently, we’d have lots of trouble detecting this, because the overall genetic divergence wouldn’t be that high. So it might be even more difficult to be certain that evolution was happening in that specific environment, at that time?
It’s a common misconception that milk naturally has much vitamin D. Very little in fact, unless added. What is has is lots of calcium. Look at the list of foods that contain lots of V-D. Milk is not among them.
https://ods.od.nih.gov/factsheets/VitaminD-HealthProfessional/
My suspicion is that skin color started lightening after milk consumption increased. Increased calcium is of little use if vitamin D is not present. Milk consumption caused an imbalance in the diet that was solved when skin lightened, which seems to be what we see in the timeline.
One other suggestive constraint on the evolution of skin pigmentation in Northern Europe might be that within Estonians, individuals with sections of the genome around alleles which influence hair and eye colour who bear more Siberian ancestry at those regions tend to have darker hair and eye colour – https://www.cell.com/current-biology/fulltext/S0960-9822%2822%2900108-7
no way this explains the difference. look at the numbers
@Razib, what difference are you talking about?
look at the change in OCA-HERC2 haplotype. the difference is way too great to be explained by the 5% (at most) of siberian ancestry
@razib, I wasn’t arguing that it was, rather that the association with Siberian segments in Estonian genomes and darker pigmentation tells us something about *relative* pigmentation at the time of admixture.
E.g. intuitively, if population A and B mix, A is genetically shorter than B, at the time, then whatever degree of selection the admixed population C undergoes, we’ll still see that segments of A ancestry bear variants associated with shorter height and B segments with taller height (whether C is over time selected to be taller or shorter than the expectation of the immediate product of admixture).
@TD, that’s an interesting theory. Though I think it would be one that needs multiple transects to really test properly. For Britain, it would seem to fit with the simultaneous trajectories of LCT and SLC45A2.
Is this so for Spain as well? Another preprint by Mathieson (perhaps superceded by this one), estimated that selection on LCT was later than in Britain (see Fig 3 – https://www.biorxiv.org/content/10.1101/2020.11.17.387761v1.full). That preprint also seemed to find that LCT was not selected heavily in Italy, at some point rose to a frequency of 0.1 (which they couldn’t detect how/when it rose) but didn’t have any sustained selection over a long time.
Patterson and Isakov’s paper likewise puts selection on LCT in their Central European dataset about 1,000 years later than in Britain which is qualitatively similar to Iberia (Fig 4 – https://reich.hms.harvard.edu/sites/reich.hms.harvard.edu/files/inline-files/Patterson_BritainMigration_Nature_MainManuscript_2021_3.pdf).
Helpfully P&I’s paper gives a specific contrast of the SLC45A2 variant and LCT variant in their Extended Figure 4 – https://i.imgur.com/GpBLysJ.png . It seems like in Central Europe SLC45A2 is at modern frequencies by the Iron Age, while LCT is still hovering around 0.07-0.1. (That figure also shows a bit more heterogenity in the derived variant of SLC45A2 among Neolithic farmers than I thought.). That seems like derived SLC45A2 was selected well before derived LCT.
Off topic but a matter upon which this blog have frequently opined: MIT now using SAT and ACT scores again. https://mitadmissions.org/blogs/entry/we-are-reinstating-our-sat-act-requirement-for-future-admissions-cycles/
Kind of related to topic – https://www.frontiersin.org/articles/10.3389/fgene.2022.833190/full – “Evolutionary Trajectories of Complex Traits in European Populations of Modern Humans” (28/03/2022)
GWAS on UK Biobank on “height, body mass index, lipoprotein concentrations, cardiovascular disease, and intelligence”. Intelligence measured (in UKBB) by Education Years and short IQ tests.
Cognitive variables are higher in neolithic and post-neolithic (with steppe ancestry) than pre-neolithic (WHG) and higher in modern sample set than neolithic and post-neolithic. Neolithic and post-neolithic look roughly the same at population mean for the PRS. So, seems to mirror skin colour changes a bit. Usual caveats apply that particularly for education years, population structure in UKBiobank could affect the results, and that modern day people who are out-of-sample for UKBiobank but drawn from 1000 Genomes, might be enriched for particular outcomes (not a random sample necessarily). Also note in paper “It remains important, however, to realize that differences between post-Neolithic and modern samples could in part be explained by differences in sequencing quality which are lower for the fragmented aDNA samples.”
@ohwilleke, although off topic, that is probably good to hear. I am still a skeptic of the importance though. Saw the boss (per John Massey’s nickname) retweet a claim that the SAT explains one-third of the US’s social mobility. Seems a bit questionable!
US is pretty middle of the pack on various social mobility indices – https://en.wikipedia.org/wiki/Global_Social_Mobility_Index – comparable to Spain, compared to many countries that don’t even use SATs at all (many use subject content exams). If we assumed that there was a large decline in social mobility – perhaps a country shifts 30 points down such an index – then that would suggest that the tweeter would believe that the US would be as socially mobile as India or Pakistan without SATs. Surely that’s highly unlikely and the US’s free market system, and social norms, and Affirmative Action etc must be doing something? But this is pretty OT.
@Matt
So what are your thoughts regarding the evolution of lighter skin pigmentation in Europeans? When would you say it started occuring?
@James, probably since Out-of-Africa since there are some shared variants across Eurasia (https://www.biorxiv.org/content/10.1101/2020.05.08.085274v1.full – “At the KITLG locus, we find some support for a shared selective sweep in East Asians and Europeans”). Many of the European ancestors weren’t living in the sampled areas for the Upper Paleolithic, which poses a problem. It seems like it accelerated during the period post 20kya in West Eurasian populations though and there was a pulse of acceleration again particularly at this point in time after 2500 BCE.
Groups are resident in Western Europe are pretty heavily sampled before the Neolithic, though they are not exactly the majority of European ancestors today, which makes things tricky.
Looking at the one Near Eastern Anatolian HG at 15kya (https://www.nature.com/articles/s41467-019-09209-7#MOESM10) against the Neolithic Anatolians – https://imgur.com/a/kb0sGyv . That shows that the Anatolian pre-Neolithic HG at 13000 BCE was still ancestral for SLC24A5, while all the Neolithic Anatolians from around 8000 BCE were derived… which would seem to support the idea that agriculture made a big difference! Likewise I think we know that the Upper Paleolithic Russian ANE hunter gatherers (AG3) didn’t have a derived variant. But! Even that’s complicated by the fact that there are loads of hunter-gatherers (typically from further east than WHG) now from the Mesolithic and Pottery Neolithic period who have derived pigmentation variants. These are Caucasus Hunter Gathers, the Iron Gates Hunter Gatherers from SE Europe, Scandinavian HG, the EHG from Russia and Ukraine, etc. (I can’t remember if this was the case for the late Ancient North Eurasian HG populations sampled after 10000 BCE or not).
So it looks like a long term pressure (and even the WHGs have some derived light pigmentation variants at high frequency, just not the ones of large effect in people today), but there are pulses and fits and starts that I don’t have any model for why and when they happened.
@Matt
Always appreciate your replies. Something I’ve always found hard to make sense is the apparently intense pigmentation differences between Corded Ware and Yamnaya. Given the inferred Yamnaya phenotypes one would think that rapid selection occured within a very short time for lighter pigmentation and associated traits.
Not sure if you agree with this or not, or whether I am not remembering CW pigmentation accurately. But it at least points out that agriculture (given CW are basically Yamnaya+EEF) must have played a role in the rise in frequency of lighter pigmentation.
PS. I seem to recall that for Iran Neo-related samples you have ancestral on SLC24A5, was it really derived for CHG?
@James, according to the paper by Jones from 2015, apparently so! https://www.nature.com/articles/ncomms9912#MOESM1131 – “Like EF, but in contrast to WHG, CHG carry a variant of the SLC24A5 gene” (Satsurblia at 11700 BCE and Kotias at 7700 BCE”. Ganj Dareh 13a (Iran), dated to 7800 BCE (as I1290) in the Reich Lab data is also – “likely had at least one copy of the derived SLC24A5 allele (rs1426654) associated with (light skin pigmentation)” (https://www.nature.com/articles/srep31326).
Yeah, like you say, I think from what I remember the CWC and BBC seem to have at least slightly higher frequencies of the SLC45A2 and OCA2 derived variants than would be expected from a mix Yamnaya+Globular Amphora Culture. There could be various reasons for this. That could include rapid selection at this time, and maybe due to more agriculture.
(Another might be that the Yamnaya samples we have – who seem to have been largely from one patrilineal clan, though curiously there aren’t many close relatives at sites, more like over 3rd degree relatives I think, but most from the same y-dna ancestor – were just a little darker than was representative for CWC’s ancestor.
Another one might be that the rapidly expanding culture simply selected lighter skinned females from the GAC culture they interacted with – it looks to me like they’re taking something like 1 in 30 marriages/reproductive events per generation that way (to rise to 50% of ancestory over 500 years). There’s a finding from that paper I linked above that, present day individuals in Estonia who have more EEF ancestry on parts of their genome that are linked to hair colour tend to have lighter hair. This seemed odd and unexpected, but one possible explanation could be something like this, where even if the GAC and early steppe populations were similar and GAC had darker hair/skin, then selecting particular partners could change the associations. Or maybe it would be that the frontier populations in CWC/BBC who mixed more with EEF had more natural selection on light traits due to adopting more agricultural lifestyles. Effectively that’s still rapid natural selection, the process is just slightly different than selection within a single population?)