mtDNA

How powerful has genetic drift been in recent human evolution? That's the question I raised the other day with reference to the claim that a heart disease risk-inducing allele had become common by drift in India during the last 30,000 years.

Another paper released earlier this week in PLoS Genetics claims that mtDNA haplotypes have been recently lost from the Icelandic population by strong genetic drift. The evidence for such changes in haplotypes comes from sequencing the mtDNA of thousand-year-old skeletons unearthed in Iceland during the last 150 years. These ancient remains have haplotypes that are found elsewhere in Europe today, but not in Iceland. The conclusion is that the modern-day descendants of these early Iceland settlers have experienced genetic drift within the last 1000 years, relieving them of of a load of rare mtDNA haplotypes.

Could genetic drift have accomplished this loss of haplotypes? Although the paper does not present any analysis of this question, a quick consideration of some theory will show that genetic drift could easily have caused the observed results. It also shows a contrast between this case and others where genetic drift has been described as "strong". Even in this case, on an island with a limited human population, genetic drift is only "strong" in the sense of eliminating alleles that are already quite rare in the population.

Today, Iceland is a large population of more than 300,000 individuals. As in many countries, the current size is a result of twentieth-century population growth. From Wikipedia:

The population of the island is believed to have varied from 40,000 to 60,000 in the period from initial settlement until the mid-19th century. During that time, cold winters, ashfall from volcanic eruptions, and bubonic plagues adversely affected the population several times. The first census was carried out in 1703 and revealed that the population was then 50,358. After the destructive volcanic eruptions of the Laki volcano during 1783–1784 the population reached a low of about 40,000. Improving living conditions have triggered a rapid increase in population since the mid-19th century - from about 60,000 in 1850 to 320,000 in 2008.

Forty thousand is a large population, in evolutionary terms. But not all these 40,000 people would have counted toward the variance in reproduction of gene lineages (See my discussion of the Wright-Fisher model). If we took a census of the medieval Iceland population, we would find that a large fraction of individuals, maybe half, were children. A rather small fraction would be postreproductive adults. So at most half the population, and probably less, were of reproductive age at any given time. At a first approximation, this population of 40,000 individuals would reproduce like a Wright-Fisher population of less than 20,000. Around half that number will be females -- although probably a bit more than half, since reproductive variance is usually higher among men. So less than 10,000 mating females per generation. Other factors, such as within-family fitness correlations, may limit the effective number even further. I'll assume a value of 5000 effective women, which is certainly an underestimate for the last 300 years, but may not be too much of an underestimate for earlier times.

We should keep in the back of our minds that the population did not grow instantaneously to this value -- it would have taken a couple hundred years to reach its ultimate medieval size. So genetic drift might have been substantially stronger when these skeletons were being buried than it would have been 300 or 400 years later.

Helgason et al. (2009) applied an unusual test to demonstrate the genetic difference between early and current Icelanders. They don't consider the frequencies of haplotypes. Nor do they use a measure that would ultimately be influenced by frequencies, like F_ST. Instead, they consider the number of shared haplotypes between the two samples as a measure of similarity.

The statistical question is whether the modern and ancient samples share significantly fewer haplotypes than expected if the ancient and modern samples had been randomly drawn from a single population. The authors tested this hypothesis with a randomization test: they repeatedly drew random samples of the same sizes as the ancient and modern samples, from a distribution including both samples. It's a clever technique.

It doesn't yield an easy answer to the question of how strong genetic drift needs to be to explain the data. But we can address this question by considering the data and nature of the comparison.

Here, we're talking about the presence or absence of haplotypes that each have a very low frequency. For example, the modern Scandinavian sample in the paper includes 337 distinct haplotypes. The modern Icelandic sample, with a slightly larger same sample size (947 versus 898 individuals), has many fewer haplotypes -- only 172. That's a substantial deficit in the Icelandic sample.

Now, let's consider the ancient sample, with 73 individuals and 58 haplotypes. Obviously, nearly all the haplotypes were present in only a single sampled individual. This implies that many haplotypes existed in that population that were not sampled in these 73 individuals. It also implies that the average frequency of a unique mtDNA haplotype in that population was less than 1.4 percent (less, in other words, than 1/73).

We don't know that these alleles were lost in Iceland; what we know is that they weren't sampled today. So a real answer to this question would include some considerations of sampling theory.

We can do a quick calculation to figure out the chances that an allele of 1.5 percent would be eliminated by drift in less than 1000 years (around 40 generations). Here's a small piece of Mathematica code that begins with an allele at 0.015 frequency, a population of 5000 effective individuals, runs it under drift for 40 generations:

The variable count returns the number of trials out of 10,000 that the allele disappeared. For an initial frequency of 1.5 percent, roughly 3 percent of trials (308 out of 10000) end in loss of the allele. That's not a very high proportion of loss by drift.

Still, we're not interested in loss from the population; we're interested in why haplotypes might not show up in the present-day sample. That might happen for two reasons:

1. The alleles weren't so common to begin with -- many haplotypes were present in the early population that weren't in the skeletal sample. We can use the simulation to see what happens to rarer haplotypes. For example, if the initial frequency was only half a percent (0.005), then they are lost from the population after 40 generations nearly a third of the time (3142 out of 10000).

2. The alleles are still in the Icelandic population, but rare enough they didn't get sampled. A sample of 900 people will miss roughly half of haplotypes with frequencies of 0.005 or less. Altering the simulation again, we find that drift achieves this result in roughly 5 percent (526 out of 10000) of cases, beginning at an initial frequency of 0.015.

So altogether, the observed result is not at all unlikely. Genetic drift in Iceland had the power to eliminate rare alleles, even after the population was founded and grew to its medieval size.

Still, this "strength" of genetic drift is manifested in its ability to eliminate initially rare alleles. If we ask what effect it would have on a common allele, we see a somewhat different dynamic. For mtDNA, the variance of allele frequency after one generation of drift is p(1-p)/(N), where N is the female effective population size (for autosomal genes with two copies, this expression would include 2N instead of N). After 40 generations, we expect a variance of 40p(1-p)/(N). What that means is that in 95 percent of cases, an allele that begins at 20 percent frequency will be between 13 and 27 percent after 40 generations.

In other words, even in the small population of medieval Iceland, genetic drift is not strong enough to cause large increases or losses of common mtDNA haplotypes. And the effect of drift is four times greater for mtDNA than for autosomal genes.

As a result of the loss of rare haplotypes, the medieval Iceland sample shares more alleles with Eastern Europe, Russia, and Spain and Portugal than with today's Icelanders. But if we look at a measure like F_ST, which is dominated by common alleles, the differences between these populations are very slight.

I've gone through the example to point out that some kinds of observations are explained well by drift, and others aren't. Genetic drift has been an important cause of some evolutionary changes, even in recent human populations. But in any given case it is a hypothesis. We test this hypothesis by applying our knowledge of demographic history, mathematical models, and other genes. The hypothesis of drift might well be useful to explain one kind of observation (loss of rare alleles), while it is useless to explain others (elevation of one rare allele to a high frequency).

References:

Helgason A, Nicholson G, Stefánsson K, Donnelly P. 2003. A reassessment of genetic diversity in Icelanders: Strong evidence from multiple loci for relative homozygosity caused by genetic drift. Ann Hum Genet 67:281-297. doi:10.1046/j.1469-1809.2003.00046.x

Helgason A, Lalueza-Fox C, Ghosh S, Sigurðardóttir S, Sampietro ML, Gigli E, Baker A, Bertranpetit J, Arnadottir L, Þorsteinsdotter U, Stefánsson K. 2009. Sequences From First Settlers Reveal Rapid Evolution in Icelandic mtDNA Pool. PLoS Genet 5(1): e1000343. doi:10.1371/journal.pgen.1000343

Roni Caryn Rabin reports on a study linking blood glucose spikes to age-related memory decline:

Researchers said the effects can be seen even when levels of blood sugar, or glucose, are only moderately elevated, a finding that may help explain normal age-related cognitive decline, since glucose regulation worsens with age.

I wonder if this physiological link may also underlie the association between some mtDNA haplogroups and Alzheimer's. In this case, the paper is able to analyze functional consequences of glucose levels because the authors were able to manipulate glucose levels in experimental animals. Remember this:

Previous observational studies have shown that physical activity reduces the risk of cognitive decline, and studies have also found that diabetes increases the risk of dementia. Earlier studies had also found a link between Type 2 diabetes and dysfunction in the dentate gyrus.

Here the causality is not necessarily clear. Maybe people who have healthy metabolic profiles are more likely to be active and less likely to exhibit cognitive declines. In that scenario, you wouldn't necessarily benefit from changing your activity pattern.

I must have seen a dozen stories today that started this way:

Reuters:

"Otzi," Italy's prehistoric iceman, probably does not have any modern day descendants, according to a study published Thursday.

Washington Post:

Sparking a new mystery about early man, Italian scientists have unraveled the DNA of the 5,300-year-old "Iceman" mummy, only to discover that he doesn't appear to have modern descendants anywhere near where he was found in Europe.

I didn't really think about how funny that line is, until I was talking to someone about it this evening -- there's no way that the mtDNA can be informative about Ötzi's descendants, because, well, it's maternally inherited. D'oh!

Meanwhile, we can ask what it means that a randomly picked Neolithic man would have a now-extinct mtDNA lineage. According to ScienceNews, Antonio Torroni thinks that it is a case of marker loss:

It’s possible but unlikely that Ötzi belonged to a fourth branch of K1 that is now extinct or rare, Torroni says. He considers it more probable that a random mutation in the Iceman’s mitochondrial DNA erased the only genetic marker currently used to identify members of the most common K1 branch.

Could be.

Whether the sequence was a unique branch of K or a slight variation on a well-represented subtype, there's a natural hypothesis for why it no longer exists, that somehow is mentioned in none of the reports. The Iceman is hardly singular: Remember that the mtDNA pool of Central Europe in the Neolithic was dominated by lineages that are now rare. And Medieval Danes had several mtDNA sequences that are now rare or absent in Scandinavia. And the Cambridge sequence has been increasing in frequency in Britain since medieval times. And so on.

There's no mystery here. These are large populations, and mtDNA haplogroups have been changing in frequencies between ancient DNA samples and the present. MtDNA has functions that plausibly were subject to changing environments after the Neolithic. This seems like a good candidate for recent selection.

UPDATE (2008-10-31): A reader writes:

Hi John,

You raised the hypothesis that recent selection might explain the apparent dearth of modern examples of his haplotype, with private mutations at 3513T and 8137T. Those mutations are synonymous, which would seem to rule that out, leaving drift as the better alternative.

That's a good point, and one often raised as a criticism of the selection hypothesis. If a sequence differs from some extant (still-existing) variant by only synonymous mutations, then selection can't explain why one is gone and one is still here. Only genetic drift can explain the extinction of one and the survival of the other.

But this is not the entire story. In this instance, we have synonymy with one major haplogroup (K) which has coding differences from other haplogroups. Those haplogroups have been changing in relative frequencies in Europe over the last few thousand years. A decrease in the frequency of K would naturally cause rare K variants to become rarer or extinct, even though they are neutral with respect to each other. In this instance, lineage extinction would be the result of selection, even though the extinct lineages have no disadvantage relative to some that still survive.

Now, is that the case with Ötzi's haplotype? I would say it's a good hypothesis, but not yet testable. We really would like to know the frequency of the haplogroup in the Neolithic. For that matter, further ancient DNA sampling will test many hypotheses of genetic drift and selection, because direct observation of ancient frequencies gives us a source of information that does not depend on sampling models from living populations.

Is there anything surprising about finding the Cambridge Reference Sequence in Paglicci 23?

UPDATE follows at the bottom.

Original post:

I don't think it's surprising in the least. No European who has yet been sampled carries a mitochondrial DNA sequence that looks like any known Neandertal mtDNA sequence. The Neandertal sequences had to disappear from the European population sometime. The sequence diversity of modern mtDNA haplogroups in Europe suggests that several of them entered Europe during Upper Paleolithic times, although some entered later. Some Upper Paleolithic Europeans must have had sequences like some living Europeans, and the CRS is one of the most common.

Does that mean that all Upper Paleolithic Europeans carried sequences that are present in living Europeans? Now, this question cannot really be answered without an exhaustive sampling of Upper Paleolithic remains. The sampling of living Europeans has been pretty exhaustive, but not so complete as to rule out the presence of Neandertal mtDNA entirely.

But David Serre and colleagues (2004) did a fairly thorough sampling of early Upper Paleolithic European remains, finding that five preserved recoverable mtDNA sequences (out of 40 sampled), none of which had a Neandertal-like sequence. If we add in the two Paglicci specimens, that makes seven. That is enough to show that it's unlikely that more than a quarter of Upper Paleolithic Europeans had Neandertal-like mtDNA sequences. Still, the mtDNA distribution may have changed under the influence of selection, so it's not possible to say much more about the ancestry of Upper Paleolithic Europeans for
other genetic loci. To know that, we'll want to sample more genes.

Therein lies the problem. We have enough trouble telling whether Neandertal genomic sequence has been affected by contamination (see my earlier entry on that topic). Neandertals must share a substantial sequence similarity with humans, and most randomly-chosen 100-bp fragments will be identical to any randomly-chosen living person. The problem is even worse when we consider Upper Paleolithic and more recent people. It is inevitable that they will be identical to some living people, and the most common alleles for most loci then will often be the same as the most common alleles today.

The real debate is about how far we should be skeptical when sequences from ancient specimens look like sequences found in living people. This debate is important not because Upper Paleolithic people shouldn't look like living people, but because Neandertals sometimes should. Sequence identity between Neandertal specimens and living people is expected for almost all the nuclear genome sequence. That means that it is impossible to authenticate Neandertal genomic fragments based on sequence alone, and we must resort to other characteristics of the fragments, such as length, proportion of base misincorporations or deamination, or intrasample polymorphism (more than two alleles is usually bad).

In this regard, I think that the current study is a helpful example. When we deal with any kind of evidence, we must establish its provenience. Forensic scientists call this "chain of custody" -- that is, how do we know that the evidence is really the same thing that was originally found, instead of something caused by events along the way during analysis? If you're doing a radiocarbon sample, you want to make sure that stray radioactivity hasn't affected your sample. If you're extracting DNA, you want to find the sequence of everyone who could be a possible source of contamination. So the fact that Caramelli and colleagues have done this is a great thing.

In fact, a 2006 paper authored by Maria Lourdes Sampietro and colleagues, including Caramelli, covered a much larger sample of Neolithic teeth and attempted to track down all cases of contamination by the excavation and laboratory teams. Here's the abstract:

DNA contamination arising from the manipulation of ancient calcified tissue samples is a poorly understood, yet fundamental, problem that affects the reliability of ancient DNA (aDNA) studies. We have typed the mitochondrial DNA hypervariable region I of the only 6 people involved in the excavation, washing, and subsequent anthropological and genetic study of 23 Neolithic remains excavated from Granollers (Barcelona, Spain) and searched for their presence among the 572 clones generated during the aDNA analyses of teeth from these samples. Of the cloned sequences, 17.13% could be unambiguously identified as contaminants, with those derived from the people involved in the retrieval and washing of the remains present in higher frequencies than those of the anthropologist and genetic researchers. This finding confirms, for the first time, previous hypotheses that teeth samples are most susceptible to contamination at their initial excavation. More worrying, the cloned contaminant sequences exhibit substitutions that can be attributed to DNA damage after the contamination event, and we demonstrate that the level of such damage increases with time: contaminants that are >10 years old have approximately 5 times more damage than those that are recent. Furthermore, we demonstrate that in this data set, the damage rate of the old contaminant sequences is indistinguishable from that of the endogenous DNA sequences. As such, the commonly used argument that miscoding lesions observed among cloned aDNA sequences can be used to support data authenticity is misleading in scenarios where the presence of old contaminant sequences is possible. We argue therefore that the typing of those involved in the manipulation of the ancient human specimens is critical in order to ensure that generated results are accurate.

The second half of that is kind of scary. Post-excavation contamination rapidly converges on the damage pattern of genuine endogenous DNA. That means that patterns of damaged DNA will not authenticate a sequence from an ancient skeleton.

Yuck.

We have to find criteria that will let us assign confidence to ancient sequences. To some extent, a lab that has relatively recent skeleton or mummy can drown out contamination by grinding up more material. But some preservation contexts won't allow that kind of sampling, and remains that are more and more ancient preserve less and less DNA.

Obviously for some samples it will be impossible to provide equivalent information. Bones from important sites like Krapina have been touched by hundreds of people over the years, many of whom are no longer available for DNA extraction. For newer sites, it might be possible to provide a list of all people who have contacted the bones, but it still might not be desirable.

Imagine a policy requiring credible researchers to put their DNA sequences on file, before they are allowed to touch a specimen. That would greatly restrict access to new finds, at least until DNA testing becomes much more routine than today. More access restrictions are the last thing we need.

And I still worry about other sources of contamination. When I was familiar with laboratories, the biggest contamination problems were not the people working on the DNA, but instead were the other samples being studied. Yes, yes, I know. Clean rooms, positive air pressure, the whole lot, all those things keep this kind of laboratory contamination. Well, except for one reason or another sometimes they don't. Sometimes the contaminant is from the next room; sometimes it's from another floor of the same building. It's very hard to detect the source of this kind of contamination unless it's some kind of exotic sequence. And when your result is the CRS, it's hard to get a less exotic than that.

I am especially interested in this question because of my work on recent evolutionary changes. Our findings show that people have many new alleles that should have been very rare or absent even 5000 years ago. Sampling ancient skeletons is an obvious way to test this hypothesis of recent evolutionary change. Already, results from ancient skeletons have confirmed some cases of recent selection -- for example, the absence of lactase persistence in Neolithic Germans (Burger et al. 2007).

With Neandertals we have the problems that stem from the use of difference as a means of authentication. If you throw out sequences that look similar to modern humans, then you have a biased estimate of population variation. With more recent people, difference may not exist, and changes in the pattern of variation may be subtle. Positively selected alleles are a rare case where we can bend the biases in our favor, since they are prominent cases where the common alleles today may have been rare or absent in the past.

UPDATE (2008/07/19): A reader writes:

In the post you commend the authors of the study for genotyping all people who have come into contact with the bone and its DNA extract. Their main line of evidence against contamination is that the sequence they recover is not any of the people who contacted the sample. However, in our experience this is a control that is nearly useless.

First, it is nearly impossible to know for sure that one has DNA samples for all people who really have come into contact with the samples. Similarly, it's not possible to know how much or how close contact must be in order to introduce a risk for contamination.

Second, in our experience, contamination introduced from laboratory reagents is also a great danger. Every enzyme, oligo, etc. can potentially introduce contamination. Extraction and PCR blanks, properly done, can control for contamination by lab reagents. However, the details of these controls are quite important to known (at least as important as knowing the sequences of the bone handlers) to evaluate the data presented.

In any case, I worry that the field of ancient DNA is sliding towards a criteria of authenticity that is not helpful. To imagine that one can round up, genotype, and then rule out all possible contamination suspects seems unreasonable.

This jibes with my experience, and the point about laboratory reagents is very well taken. Here's what I wrote back:

You know, I almost wrote that post more critically, because my contact with labs has always included stories about these weird contamination sources. With one case, it was DNA in the next building that had come in due to a faulty ventilation system. In another lab, an early UP specimen looked like Polynesians. Well, they were working on Polynesians down the hall, but they were pretty sure that all their precautions would keep the ancient stuff completely separate.

There's no chance you can rule out everything, so why try to get all the people who touched it? Especially since that's going to be useless for most of what you can sample, which has been touched by dozens of people.

On the other hand, it may be worthwhile to get people who excavate and handle things to be more careful, and taking their sequences adds some seriousness. It may be more for show than genuine contamination control, no doubt. I suppose that's the main purpose of all these proposals about protocols -- mainly ways to prevent arguments between labs, about "you can trust my sequence, but who knows about those guys?"

Looking from the outside, I think the contamination problem may be no worse for ancient DNA than for other kinds of sequencing, and errors are creeping into databases all the time. The metagenomics scheme is a tremendous improvement, since the computers actually can keep track of this contamination and correct for it. So working on Neandertals is really more a question of getting the algorithm right. Probably, we should try to get people to stop doing extractions for single-gene probing, it's a waste of material, and the question of contamination is never going away. It's really a matter of time before we have all these sequences from specimens that preserve them, and we want to keep as much as possible against the chance of future improvements.

References:

Burger J, Kirchner M, Bramanti B, Haak W, Thomas MG. 2007. Absence of the lactase-persistence-associated allele in early Neolithic Europeans.

Caramelli D and 13 others. 2008. A 28,000 years old Cro-Magnon mtDNA sequence differs from all potentially contaminating modern sequences. PLoS ONE 3:e2700. doi:10.1371/journal.pone.0002700

Sampietro ML, Gilbert MTP, Lao O, Caramelli D, Lari M, Bertranpetit J, Lalueza-Fox C. 2006. Tracking down human contamination in ancient human teeth. Mol Biol Evol 23:1801-1807. doi:10.1093/molbev/msl047

Serre D and 8 others. 2004. No evidence of Neandertal mtDNA contribution to early modern humans. PLoS Biol 2:313-317. doi:10.1371/journal.pbio.0020057