ARTICLE

Top Abstract Article References

When and where did modern humans evolve? This question remains the focus of much scientific controversy. Traditionally, answers were sought in the human fossil record, which tells us that upright bipedal hominids who made stone tools have occupied much of the temperate Old World for 0.5-1.5 million years. The earlier forms of these hominids are usually called Homo erectus, whereas the later forms, with larger brains and more sophisticated tool kits, are called Archaic Homo sapiens. The Neandertals of Europe are the most familiar of these archaics. In Europe they were replaced by modern humans over several millennia about 40,000 years ago. In Indonesia archaics may have persisted until as recently as 25,000 years ago (1).

Although fossils provide unique and invaluable information, they are very limited in quantity and quality. Huge gaps remain in the human fossil record, and it is difficult to assess, for example, whether there was continuity between archaic and modern humans. Genetic data, in contrast, are easy to collect, and they are accumulating rapidly. Ancient demographic events have left imprints that can be detected in present-day gene differences. Our purpose is to write an accessible summary of current genetic research about human population history.

Genetic methods and data are providing fresh perspectives on a long-standing debate about the origins of our species, which, in its simple form, can be summarized as two competing hypotheses. The multiregional hypothesis suggests that modern humans evolved directly from archaic forms in several different locations in the Old World. Gene flow among these populations, combined with natural selection for advantageous genes, maintained genetic homogeneity of the species. Under this hypothesis, our species had hundreds of thousands, perhaps millions, of ancestors for most of the last million years. Without a large population, gene flow among populations distributed widely over the temperate and tropical Old World would have been impossible.

The other hypothesis is called variously the Garden of Eden, the Noah's Ark, or the single origin model. According to this hypothesis, a specific population ancestral to modern humans underwent demographic expansion and populated those parts of the world occupied by archaics and then beyond into northern parts of Eurasia and eventually the New World. The contribution of archaic populations to the modern gene pool was negligible. The number of our ancestors just before the expansion ("origin") of modern humans was small, only several thousand breeding adults.

A clear difference between these two hypotheses is the implied size of the past human population. If the size of the human population had been large throughout much of its history, extant genetic variation should be substantial. Conversely, a small human population would result in relatively little genetic variation. Many genetic systems provide reassuringly congruent estimates: all indicate that human genetic variation is relatively low and that the approximate "effective" size (i.e., the number of breeding adults) of humans is on the order of 10,000 (2, 3). Because several thousands or even tens of thousands of humans could not have occupied the whole temperate Old World, genetic data provide strong support for the single origin hypothesis. Archaeological and skeletal evidence also generally support some version of the single origin hypothesis (4, 5).

The genetic relationship between modern humanity and the world population of archaics at, say, a half million years ago is still unspecified. If the small effective size of humans reflects a transient but drastic reduction in size of a large population, and subsequent recovery, then before the reduction, the number of our ancestors was large, and a graph of our population history looks like an hourglass. By contrast, if we are descended from a subpopulation of archaic humans that was effectively a separate species for the last million years or so, then the graph of our population history is a bottleneck, a short bottle and a very long neck.

The hourglass hypothesis posits that there was a contraction in the number of our ancestors at some time during the Pleistocene, perhaps before the last interglacial, but specifies that the small ancestral population that later expanded had been part of a network of gene flow over the whole temperate Old World occupied by archaic humans before the contraction. In other words, if the genes in the small founding population of modern humans were traced backward in time, they would be dispersed over a large part of the Old World in a population of hundreds of thousands to millions, as in the multiregional hypothesis. Humanity's small apparent effective size is the result of loss of genetic diversity during the contraction.

The long-neck hypothesis, in contrast, posits that the small ancestral population was small during most of the Pleistocene, for the last million years or so, and that genes in this population traced backward in time were restricted to the range of the particular species of archaic humans that were our ancestors. The essential difference in the two hypotheses is the effective size before the constriction in the middle or upper Pleistocene. They have different consequences for the shape of trees of descent of nuclear genes. Below we show that there is no support in current genetic data for the hourglass hypothesis, whereas the long-neck hypothesis finds strong support.

Estimating Human Effective Size

Effective Size and Census Size. Effective size is the breeding size of an abstract population, and relating effective to census size of human populations is complicated.

Estimates from Mean Pairwise Difference and Segregating Sites. The simplest genetic evidence about our demographic history is from estimates of overall human effective size. There are two standard approaches to estimating effective size. Each does not estimate size per se but the product of size and mutation rate: these two parameters are almost always confounded in population genetic models. An exception is the estimate from human-specific Alu insertions described below.

[ 1 ]

New Effective Size Estimates. While the above methods require knowledge of the mutation rate to estimate N, a different approach is to use the time of separation between the ancestral chimpanzee and human species to calibrate a genetic estimate of human effective size using Alu insertions. Most Alu elements are short (300 bp) pseudogenes (10). They are stable, transcriptionally inactive copies of a few active Alu elements, scattered randomly throughout the entire nuclear genome. Collectively, there are about 500,000 copies per haploid genome or 5% of the genome by mass. Some Alu elements are shared with prosimians, monkeys, and apes, whereas others are so recent that they are polymorphic in humans. We have studied insertions of the Ya5 and Yb8 subfamilies whose active elements have been present since before the separation of gorillas from the chimp-human lineage. There are several thousand Ya5 and Yb8 elements in the human nuclear genome, and they are still being inserted. The unique value for evolutionary inference of these loci is that inserted elements are never precisely deleted, so the ancestral state is always known.

View larger version (63K): [in this window] [in a new window]   Fig. 1.   Schematic history of a nuclear locus in humans and chimps.

Changing Effective Size

There are almost 6 billion members of our species on earth today, but the genetic indications of our low effective size suggest that we have not been so numerous for long. Effective size estimates from genetic data refer to a kind of average of population size from the present back to the coalescent of the sample. To understand how genetic data are used to read population size history, it is necessary to look at trees of descent of genes and how population size change affects their characteristics. Because common practice in the literature about human evolution is to reconstruct a gene tree from differences among sequences and then infer history from the reconstructed tree, we include some comments and caveats about this practice.

Properties of Gene Trees. In this section we show simulated gene trees from populations that have been stationary, that have undergone expansion, and that have undergone contraction. Each of these histories can generate characteristic signatures in the structure of gene trees. In these simulations there are two populations that have always exchanged members at the rate of 0.5 gene per generation; the two populations are approximately as different from each other as two human populations from different continents. The sample is 10 sequences or genes from each population, the red and the green populations.

Constant Population Size. The trees shown in Fig. 2 are generated by assuming that the two populations have never changed size. The four trees in the figure are derived from exactly the same demographic scenario, yet they vary widely from one to another. Typically we would have only a single tree to analyze, for example, a tree of mtDNA sequences, so it is important to look at variability among trees in these simulated populations to assess how reliable inference from a single reconstructed intraspecific tree can be.

View larger version (15K): [in this window] [in a new window]   Fig. 2.   Simulated gene trees from a pair of populations of constant size that exchange an average of one-half a gene every generation. These are four simulated loci from the same populations with the vertical axes drawn to the same time scale.

Population Expansion. As we follow a sample going backwards in time to the coalescent, the rate at which lineages coalesce is proportional to the inverse of the effective size. If a population is large now, but expanded rapidly from a small population in the past, then coalescent events will be relatively few since the expansion. They will be concentrated just before the expansion when the population was small. The result is a characteristic star-like gene genealogy as shown in Fig. 3. Prolonged exponential population growth generates similar trees (16). Trees like these are also produced by "selective sweeps," replacements by an advantageous new allele that is fixed by selection. In the case of population expansion and of a selective sweep, today's genes have a smaller-than-expected number of ancestors at some time in the past.

View larger version (52K): [in this window] [in a new window]   Fig. 3.   Simulated gene trees from a pair of populations that expanded by a factor of 1,000. The populations exchange an average of one-half a gene every generation.

Population Contraction. Reduction in population size can lead to rapid loss of genetic diversity. Because the expected depth of a coalescent tree is 2N generations, a contraction in population size lasting this long would erase preexisting diversity in many loci, whereas others would retain two or a few variants that would differ according to the coalescence time, hence the population size, before the contraction. Fig. 4 shows trees from a population that has undergone an instantaneous hundredfold size reduction. The loci in the two right panels retain several variants from before the contraction, whereas those in the left panels lost all precontraction diversity. The visible result of a contraction should be many loci with very little diversity, like those on the left, along with others with a few very divergent gene lineages, like those on the right.

View larger version (9K): [in this window] [in a new window]   Fig. 4.   Simulated gene trees from a pair of populations that have contracted by a factor of 100. The populations exchange an average of one-half a gene every generation. Because the vertical axes are drawn to the same scale, the trees in the left two panels are so shallow that they are invisible.

Graphical Methods, Trees, and Human History

Expansion in the Pleistocene. An important paper by Felsenstein (17) showed that estimates of population size could be dramatically improved if the branching order of the underlying tree were known. Because unambiguous reconstruction of this branching order is often impossible, computer-intensive methods have been developed that examine large numbers of trees, evaluating the likelihood of each tree and the likelihood of a demographic hypothesis given the tree (18, 19). As these methods become faster and more widely available, methods like those as we describe here will be relegated to screening roles. However, the simple approaches we describe are fast, simple, and easy to understand. Computer-intensive methods have the important disadvantage that one never really knows what they do; it is difficult to prove that they are working correctly.

View larger version (24K): [in this window] [in a new window]   Fig. 5.   Gene tree (Top), frequency spectrum (Middle) and mismatch distribution (Bottom) from a population that has undergone a population expansion. The circles on tree represent mutations. The simulation parameters match approximately those estimated from human mtDNA.

View larger version (20K): [in this window] [in a new window]   Fig. 6.   Gene tree (Top), frequency spectrum (Middle) and mismatch distribution (Bottom) from a population that has always been constant in size.

View larger version (25K): [in this window] [in a new window]   Fig. 7.   Frequency spectrum and mismatch distribution from a world sample of 636 mtDNA sequences at 411 positions of the first hypervariable segment (HVS-I). Compare this with Fig. 5. The diamonds show the expected number of segregating sites in each frequency interval expected in a constant size population.

View larger version (10K): [in this window] [in a new window]   Fig. 8.   Frequency spectrum and mismatch distribution from a world sample of 718 Y chromosome sequences. There are 20 segregating sites ascertained at approximately 20,000 positions. Ascertainment was done in two samples, one with 21 chromosomes and one with 53. If a site has population frequency x, then the probability that it will be detected in a sample of size n, the ascertainment function, is 1  (xn + (1  x)n). Diamonds show the expected number of sites in each frequency class under the hypothesis of constant population size, computed by multiplying the ascertainment function by the distribution in Eq. 1. The excess of low frequency sites is consistent with a Pleistocene population expansion. Because the whole nonrecombining portion of the Y chromosome is a single locus, these sites are not independent, and there is no simple statistical test of the constant size hypothesis.

Population Size Before the Bottleneck. Both mtDNA and Y chromosomes coalesce several hundred thousand years ago, so they provide no information about population size before then. The coalescent of nuclear genes should be four times as old as that of mtDNA and the Y in the absence of population size change. Unfortunately, nuclear genes undergo recombination, and recombination rapidly destroys evidence of population size in DNA sequences. The mutation rate at nuclear loci is also much lower than that of mtDNA so long sequences are necessary to achieve resolution comparable to that of mtDNA sequences, but the longer the sequence the more likely recombination has occurred.

View larger version (13K): [in this window] [in a new window]   Fig. 9.   Frequency spectrum of 23 Alu insertions in humans. The diamonds show the expected numbers of loci under constant population size as specified by the long-neck model. Circles show expected numbers of loci under the hourglass model of a population contraction. The hourglass hypothesis can be rejected by a statistical test, whereas the long-neck model cannot. Each of these was ascertained in a diploid. The probability of detecting at least one copy of an insertion in a diploid whose population frequency is x is x2 + 2x(1  x), the ascertainment function for this system. Expected values for the long-neck model were computed by multiplying the distribution in Eq. 2 by this function. Expected values for the hourglass model were computed by multiplying the ascertainment function by the uniform distribution because the distribution of the number of copies of a mutation in the top interval of the tree is uniform.

Conclusions

We have avoided hedging and qualifying our findings in the interest of making this paper simple and accessible. Nevertheless, the broad picture that we paint continues to gain empirical support. Most of the familiar specimens of Homo erectus and of archaic humans known from the Pleistocene were not members of populations ancestral to us, instead "the fate of most such populations appears to be tragic" (13). We are descended from a population that was effectively a separate species for at least the last 1 or 2 million years. Although the size of this population must have fluctuated over time, it was often reduced to the level of several thousands of adults. Such a population would have occupied an area the size of Swaziland or Rhode Island rather than a whole continent, although episodic expansions would have covered a much larger area. Archaeologists should find and identify this population.