The great bottleneck after the post-Eemian separation

I’ve been thinking about effective population size. Basically it’s the inferred breeding population you estimate in the present, or in many cases the past, based on the genetic variation you see within the population. Another way to say it is that it’s the population size that can explain the genetic drift that you see in the data.

To give a concrete example, the population of the New England states of America was ~1,000,000 during the 1790 Census. The vast majority of this was due to natural increase from a settler population of about ~50,000 in 1650 (total fertility rate of women in New England was seven children in the years between 1650 and 1700). Of these, ~23,000 were Puritans or the offspring of Puritans who migrated around between 1630 and 1643 (due to religious differences with the English government of the period). One might think that a population of ~1,000,000 would be genetically diverse, but the ~50,000 in 1650 matter a lot more than the ~1,000,000 in 1790. The rate of mutation accumulation is pretty slow, so a population bottleneck or subsample has a huge long-term effect.

In fact, as you probably know one of the biggest determinants of genetic variation in New England whites of 1790 is the bottleneck that they share with all other non-Africans that dates to 50,000 years or more before 1790!

And these are just the coarse demographic considerations on the broader population/historical scale. In any normal random-mating human population, there’s some reproductive variance by chance (usually it is modeled as a poisson distribution; mean and variance being the same, though from I have read the variance in mammals is usually greater than the mean).

Some people have more children, and some people have fewer children. That means that there is a census population, and a breeding population, and the breeding population is invariably smaller than the census population. Some individuals don’t reproduce to the next generation, obviously. But there are also cases where some individuals have large numbers of surviving offspring, while others have only a few.

To make it concrete I plotted the distribution of the number of children of women older than 50 years of age from the year 2000 and later in the General Social Survey (GSS). You can see that the most common number is two, but there are a fair number with three. Only about 10% of women 50 years and older have no children in the GSS.

But the curious thing is that if you weight the number by the proportion, you notice that women who have three children may not be as common as women who have two children, but they are contributing more children to the next generation than women who have the more typical two children. And, though the number of women who have five or more children is only 11% of the sample, as opposed to 14% who have one child, they contribute nearly five times as many children as those with one child to the next generation (women with six children alone contribute more than women with one child).

Basically, not all the genetic variation in a given generation is created equally. Some people will contribute more to the next generation, and that has a homogenizing effect (there are models of mutation/selection/drift which establish equilibria values of variation in a stationary state).

I’m revisiting all of this for two reasons. First, in Who We Are And How We Got Here David Reich talks about a long period of a shared population bottleneck for “Out of Africa” (all non-Africans) groups before the primary expansion ~60,000 years ago. Second, in my conversation with Matt Hahn, he was very skeptical of drawing any correspondence between effective population and some inferred census size. In hindsight I think part of it is that in most organisms census quotes are more an art than science. Not so with humans.

This made me look more into the literature for humans again. Recently Browning et al. published Ancestry-specific recent effective population size in the Americas. It’s a great paper. Basically, it uses identity by descent tracts of different ancestry to tease apart the distinctive pre-admixture effective population sizes. If you take an admixed population and assume that it was a single population random-mating indefinitely, and then work backward in time, you’re probably going to produce rather strange effective population sizes (if the two groups are about the same genetic diversity beforehand, they’ll probably show an inflated effective population, because you are assuming the two groups were a big random-mating population long before they were randomly mating!).

There are many ways to infer effective population, and the identity by descent method seems reasonable for recent time periods. And one thing about recent population size estimates for humans is that you have reasonable census estimates (you don’t just check with simulations):

Our simulations showed that biased sampling of a structured population results in underestimation of most recent effective population size. When we compare the estimated current effective sizes of HCHS/SOL country-of-origin populations to World Bank population sizes (accessed via Google Public Data Explorer) from 1995 (when the average age of the sampled individuals was around 25), we find that the ratio of current estimated effective size to 1995 population size ranges from approximately 1/60 (Ecuador) to approximately 1/4 (Cuba), with typical values around 1/10. Although estimates of effective size in the most recent generations are affected by these issues, our simulations also showed that less recent generations are not affected. Thus our estimates are useful for learning about the effective population sizes at and before admixture.

The structured part is important. For example, the paper On the importance of being structured: instantaneous coalescence rates and human evolution—lessons for ancestral population size inference? explores how structured models of gene-flow might be confused when genomic inferences assume a panmictic population. Last year a paper in PNAS, Early history of Neanderthals and Denisovans, suggested that Neanderthals were characterized by a high structured meta-population, and that low effective populations from sampled genomes in this group of humans reflects this, rather than a genuinely low census size.

Browning et al. focused on recent population size inferences. I was curious about these inferences because we can compare them to real census sizes. From this I think I can tune my intuition at least to the possibily that census size of a random mating population is not likely to be two orders of magnitude above the inferred effective population size. Conversely, the rough mammalian value of an effective population size of ~1/3 the census size seems to be a ceiling. Population structure and bottleneck aside, humans seem to have enough basal reproductive skew that effective population size is less than half of the census size.

To focus on ancient population growth (or lack thereof), I reread Inferring human population size and separation history from multiple genome sequences (Schiffels et al. 2014), Exploring Population Size Changes Using SNP Frequency Spectra (Liu et al. 2015) and Neutral genomic regions refine models of recent rapid human population growth (Gazavea et al. 2014). The first two papers seem to suggest an “Out of Africa” population bottleneck that’s pretty long, with an effective population that’s somewhat lower than 5,000 individuals. In contrast, the last paper seems to have a sharp bottleneck of 200 individuals.

Remember, different models can produce the same empirical patterns in the genome. You can reduce genetic diversity by a modest, but long, bottleneck. Or, through a very sharp short bottleneck.

In Who We Are and How We Got Here David Reich definitely leans toward a long, but more modest, bottleneck. For anthropological and archaeological reasons this seems more plausible now than it did ten years ago.

But perhaps it makes more sense now that we have more ancient DNA and a more elaborated model of human history seen through the lens of population genetics. In Schlebusch and Jakkbonson’s Tales of Human Migration, Admixture, and Selection in Africa the authors come out say “For our species’ deep history in Africa, both paleoanthropological and genetic evidence increasingly point to a multiregional origin of AMHs [anatomically modern humans] in Africa.”

They’re only saying what I hear other people talking about.

Instead of the “Out of Africa bottleneck” being defining for our species, it’s only a phenomenon which is important for peoples outside of Sub-Saharan Africa. Arguably for the majority of the existence of our species something closer to multi-regionalism was operative within modern humans.

If fact, isn’t that what the new ancient DNA shows? Pulses of admixture and gene-flow between distinct groups? Arguably multiregionalism might be the answer to our origins, but also characterize many of the dynamics after the “Out of Africa” event.

In any case, the best evidence now points to the likelihood that modern human lineages began to diversify and diverge before 200,000 years ago. Conversely, most of the ancestry of modern humans outside of Africa dates to an expansion around ~60,000 years before the present (ancient DNA and archaeology seem to agree here).

This is probably right before the Neanderthal admixture event with non-African humans, at least the modern lineages we have around today. But, it turns out it does not define the point when non-African humans diverged from the ancestral African population. Another group, “Basal Eurasians” (who may not have been Eurasian at all), diverged before the expansion of all eastern non-Africans, Oceanians, as well as the ancestors of Pleistocene Europeans and Siberians. It does not seem that Basal Eurasians had any Neanderthal admixture. Basal Eurasian ancestry is substantial in the Middle East today (although lower than 50%), and non-trivial across broad swaths of Europe and South Asia, due to the expansion of farming. They seem to have been well mixed in places like North Africa with other Eurasian groups ~15,000 years ago. Presumably that was a “back to Africa” migration, since these people had Neanderthal ancestry.

All of this leads to the conclusion that the ancestors of Basal Eurasians/non-Africans must have gone through their shared bottleneck well before ~60,000 years before the present. And, it may have happened on the African continent. So with that, I’ll quote Schiffels et al.:

This comparison reveals that no clean split can explain the inferred progressive decline of relative cross coalescence rate. In particular, the early beginning of the drop would be consistent with an initial formation of distinct populations prior to 150kya, while the late end of the decline would be consistent with a final split around 50kya. This suggests a long period of partial divergence with ongoing genetic exchange between Yoruban and Non-African ancestors that began beyond 150kya, with population structure within Africa, and lasted for over 100,000 years, with a median point around 60-80kya at which time there was still substantial genetic exchange, with half the coalescences between populations and half within (see Discussion). We also observe that the rate of genetic divergence is not uniform but can be roughly divided into two phases. First, up until about 100kya, the two populations separated more slowly, while after 100kya genetic exchange dropped faster.

David Reich’s group, and others, now posit the existence of “Basal Human” population that mixed into West Africans, who can be modeled as primarily proto-East African (without Eurasian admixture), as well as this ancient outgroup. This means that estimates of divergences with non-Africans from something like MSMC may generate a composite if proto-East Africans are closer to the ancestors of non-Africans, which seems likely. One likely model is that the “Out of Africa” population emerged out of the northern edge of this proto-East African distribution of modern humans over 100,000 years ago (but after groups like the Khoisan and Basal Humans had already diverged).

Looking at Schiffel et al., they seem to posit lower in divergence times than seems likely to me. Is that perhaps due to unaccounted for admixture in lineages which fuse together groups which were earlier distinct?

In any case, with details about the divergence dates set aside, the MSMC results are actually in line with a new congealing consensus. Deep structure within Africa, but gene-flow between distinct populations, for at least ~100,000 years (possibly more). This is the period when population structure was quite fluid and indistinct along the East Africa continuum out of with non-Africans emerged.

Also, the archaeological evidence is now strongly suggestive of modern humans in places like Southeast Asia over 10,000 years before the wave which led to the ancestry of most extant populations. In fact, we know that this sort of early migration with no descendants isn’t abnormal. The first modern humans in Europe left no descendants (at least in any appreciable quantity). And the Altai Neanderthal seems to have modern-like admixture that dates to ~100,000 years before the present.

With all the evidence that modern humans were present in Africa, and expansively so, for hundreds of thousands of years, it seems unlikely that they never mixed with “archaic” Eurasian  lineages (and vice versa). In fact, as we obtain more and more Neanderthal and Denisovan genomes perhaps we’ll find that a rapid expansion like the one that occurred ~60,000 years ago across Eurasia and Oceania happened before, out of and/or into Africa.

Looping back to the effective population issue, the effective population of modern non-Africans seems to have been below ~5,000 for a while. There was minimal gene-flow with other populations for many generations. Reich has a schematic of 40,000 years between 90,000 and 50,000 BP in Who We Are and How We Got Here. But that’s obviously just a ballpark figure. I have a hard time believing that the census size was around 500,000. The world population 10,000 years ago is usually estimated to be 1 to 10 million. Human populations were probably much larger at the end of the Pleistocene than 100,000 years ago. But a figure of 10% effective would give 50,000, which seems a reasonable number, especially with the likelihood that we’re talking about many tribes over a wide ecological zone. Meta-population dynamics of extinction and resettlement in inclement periods probably drove down the effective population.

The separation seems to be distinct from the older multiregional phase. What could explain it? The existence of the Sahara, and periods of extreme desertification seems the most likely candidate. I can’t say much with any credibility because I don’t know the archaeology and paleoclimate literature, but before domesticated animals, it was probably difficult for hunter-gatherers to make a go of it in the deep Sahara during the driest phases.

If I had to bet, the Eemian interglacial, 130 to 115 thousand years ago, is when I would assume there was:

1. Lots of gene flow across the Sahara, perhaps in both directions
2. A major population expansion of humans, of all sorts

This gives plenty of time for a wave of modern humans to push east, probably going through milder climates, rather than expanding north into Neanderthal or Denisovan territory. Eventually, some group must have mixed with the ancestors of the Altai Neanderthals. It seems likely that a cold and dry spell after the Eemian would have been optimized more to the well adapted Eurasian groups, and modern populations would have withdrawn into refugia. The brutally expanding Sahara would have divided the majority of modern humans, who existed in the meta-populations to the south that dated back hundreds of the thousands of years, from the groups on the northern fringe.

One can imagine that large numbers of modern humans were either absorbed or went extinct with the expansion of Neanderthals and other archaics. Though Neanderthals and Denisovans were interfertile with moderns, the lineages were still distinct enough that it looks like there was some hybrid breakdown. Just as modern humans seem to have purged many Neanderthal alleles from our genome, the opposite dynamic was probably at work.

There was clearly some structure in the relict modern human group that was separated from the African populations. Basal Eurasians did not mix with Neanderthals, but the ancestors of all other non-African humans did. Though one has to be careful about such geographical inferences, that suggests to me that the range of modern humans in the period between 60,000 to 80,000 years ago extended further back into pockets of northeast Africa, where no contact with Neanderthals would have occurred. Perhaps, in the end, we’ll end up thinking that the Basal Eurasians in some ways were a lot more like Africans south of the Sahara, as they didn’t undergo the massive range expansion of other populations during the Upper Paleolithic.

I’ll end with some predictions.

• Ancient DNA of proto-moderns and archaics in eastern Eurasia dated to between 50,000 to 100,000 years BP will be analyzed at some point and will exhibit a fair amount of admixture. That is, the Altai Neanderthal was not exceptional, and probably relatively attenuated. I’m moderately confident of this.
• The pre-60,000 year eastern Eurasians will be found to have left some of their genes in modern eastern Eurasians. Especially in Southeast Asia and Oceanian. Probably in the 1-10% range. I’m moderately confident of this.
• The Denisovan ancestry in Oceanians is mediated by a “first wave” group “Out of Africa.” I have low confidence in this, but I really wouldn’t be surprised either way. My confidence in my confidence is low!
• At some point we’ll obtain sequence from a 1 million year old hominin somewhere in the colder/drier climes of Eurasia (we have a 900,000 year old horse genome). This will predate Neanderthal/Denisovans. We will see from this that some of these super-archaic populations left their heritage in later archaics, and therefore our own lineage. I’m rather confident of this.
• By hook or crook we’ll get more ancient genomes out of African samples, and confirm a lot of ancient population structure, as well as some gene-flow from archaic non-modern lineages. Probably around the same range you see in non-Africans (though some of the gene-flow may also apply to non-Africans, since they didn’t separate from eastern Africans until 100,000 to 150,000 years ago). I’m rather confident of this.
• H. naledi will return sequence at some point. I’m very confident of this. I don’t have inside knowledge, but I know they’re going to keep trying. They are getting more samples.
• H. naledi will be found to have contributed ancestry to modern southern African populations. I’m moderately confident of this.
• At some point ancient genomes from the Americas will confirm the existence of an earlier group which was only distantly related to modern New World populations descended mostly from Siberians. There is indirect evidence of this group from South American populations, but we’ll get individuals who are much more distinct at some point in the future. I’m moderately confident of this.
• Basal Eurasians will be found to have inhabited Southern Arabia/Persian Gulf region. But “pure” population will have been found to have disappeared around the Last Glacial Maximum ~20,000 years ago, as the human populations to the north moved south, and the Near East’s southern fringe became drier. I’m moderately confident of this.