Genetic data and the African origin of humans.

S. A. Tishkoff et al. (1) provide an intriguing analysis of human genetic variation at the CD4 locus. We are concerned, however, that their data do not provide significant support for the estimate that modern humans first emerged from Africa in the last 100,000 years. It appears that a robust estimate of this migration time will require the use of numerous loci. The estimate, which is predicated on a set of assumptions listed in their paper, depends in part on estimating the relative ages of the Alu deletion [Alu(-)] allele in African and non-African populations. Tishkoff et al. argue that among Africans, the frequency of Alu(H-) chromosomes linked to the progenitor [90 base pair (bp)] specific short tandem repeat polymorphism (STRP) allele is given by eNA 1, where NA is the age of the Alu(-) allele and ,. is the STRP mutation rate. (They give an equivalent expression for non-Africans, in which NB represents the time of migration out of Africa.) Under the assumption of no back mutations, this expression does give the expected frequency of the 90-bp allele on Alu( -) chromosomes. Because many of the individuals in the sample will have a shared ancestry, the alleles found in different individuals are highly correlated, and so an estimate based on this procedure may have an extremely high variance. In estimating the age of the Alu( -) mutation, it is convenient to consider the problem in a coalescent framework (2). In this view, the individuals in a sample are related to one another by some ancestral tree (strictly speaking, this is ancestry at a specified locus). When a mutation occurs at some point on the tree, all the individuals who trace their ancestry through that point on the tree will carry that mutation (recall the assumption of no back mutation). This means that a mutation that occurs near the root of the tree will often be carried out by a large proportion of the sample. We have investigated the problem of establishing a lower bound on NARp, given the authors' observation that 34 out of 85 non-recombinant African Alu(-) chromosomes carry the progenitor allele [some of the Alu( -) chromosomes seemed to be descended from a single recombinant and were excluded from the original analysis]. Their estimate of the migration time out of Africa is crucially dependent on this lower bound. No detailed theory exists for finding such a bound analytically. We can, however, approximate confidence intervals with the use of simulations of the coalescent process. If the entire African sample consisted of non-recombinant Alu(-) chromosomes, it would be reasonable to set bounds on NAP. by using standard coalescent assumptions to generate random trees with 85 tips (3). In this case, we also know the Alu(-) frequency in the total sample, and it is possible to use the coalescent approach to get the distribution of mutations in the Alu( -) chromosomes conditioned on that frequency. In order to do this, we have generated trees of 806 chromosomes [the sample size in the article (1)] and selected only those that contain a clade of 132 chromosomes [the total number of Alu( -) chromosomes]. The relative times of the nodes in the tree of 806 are drawn from an exponential distribution [the parameter is (n) between nodes n and n 11 (4). Taking the clade of 132 to correspond to the 132 Alu(-) chromosomes, we have now specified the relationships within a simulated data set in which all the relative branch lengths have been drawn from the appropriate conditional distribution. In the original data set, 47 of the 132 Alu( -) chromosomes were recombinants and were excluded from the analysis. In order to further condition our own analysis on this information, we have selected only those trees in which a clade of 47 lies within the clade of 132 Alu(-) chromosomes. This procedure has allowed us to generate trees of 85 individuals whose relationships to the larger sample closely mimic those in the original data set. Each simulation specified the relative lengths of all the branches, and so picking a trial value of NAU. for the top of the Alu(-) clade determined the expected number of mutations along each branch. For each simulated tree, and trial value of NAP, the number of mutations on each branch was drawn from a Poisson distribution with that expected value. Our results, based on 10,000 random trees that meet the above criteria, are rather striking. The value of NAp., estimated using the method of Tishkoff et al., is 0.916; however, we have found the lower bound to be 0.12 at the 5% level of statistical significance (5). This indicates that their estimate of NAPL could be seriously in error, and even without taking into account the variance in the other estimates in the original calculation, the technique used cannot reject a migration time out of Africa as much as sevenfold greater than the original estimate. It is also possible to analyze the estimate of NBpL in a similar manner. In this case, however, the coalescent assumptioni of single origin might break down because there may have been numerous Alu( -) chromosomes in the proposed migration event. A further problem with the argument presented by Tishkoff et al. (1 ) is that, in light of other studies of population coalescent times for mitochondrial DNA (6), the Y chromosome (7), and autosomal microsatellite markers (8), it seems unlikely that the Alu polymorphism is as old as 5 million years, as implicitly suggested. However, if it were this old, the absence of the Alu(-) allele in chimpanzees would not rule out an age of more than 5 million years. That is, the Alu(-) allele could have existed in the ancestral population of both humans and chimpanzees and subsequently been lost from the chimpanzee lineage. This analysis shows that as a result of the shared ancestry of individuals in a population, estimates of mutation times-or population divergences based on a single mutating locus (STRP here)-can be highly unreliable, even when large samples of individuals are used. Jonathan K. Pritchard Marcus W. Feldman Department of Biological Sciences, Stanford University, Stanford, CA 94305, USA E-mail: [email protected] marc@charles. stanford.edu