model of human evolution depends on the demonstration of evolutionarily recent time depths for alleles found in non-African pop- ulations. Multiregional model enthusiasts also argue for an African origin, but place this origin at 1 million years before present (Y.B.P.), the approximate time at which Homo erectus remains can be identified out- side Africa. Thus, for the "Out-of-Africa" model to be accepted, it is critical that allelic time depths be more recent than 1 million Y.B.P. In support of the recent "Out-of-Africa" model, we (1) attempted to show that a single chromosomal segment, a CD4 locus haplotype composed of an Alu(-) allele and an STRP allele of 90 bp separated by 10 kb, had a recent time depth in non-Afri- cans. As we emphasized in the article, in the absence of known recombination be- tween the sites or mutation rates at the STRP marker, it is impossible to estimate an exact time of origin of this haplotype in non-Africans. However, by making certain conservative assumptions, it is possible to place likely upper bounds for this date. We used several methods of analysis to derive an upper bound for the coalescent date for non-Africans. One was based on the vari- ance observed at the STRP on Alu(-) chromosomes outside versus inside Africa; this led to a date of 167,000 Y.B.P. Another analysis was based on the proportion of Alu(-) chromosomes with STRP alleles less than 110 bp outside versus inside Africa that carry the progenitor (90 bp) STRP allele. As an upper bound on this propor- tion, we examined its variability across five geographically diffuse sub-Saharan African populations that had more than 10 Alu( - ) chromosomes. The proportion carrying the 90-bp repeat ranged from 0.25 in the Woloff to 0.53 in the Herero. We used 0.53 as an upper bound for this value across sub-Saharan Africa. For non-Africans, be- cause of the small number of Alu( -) chro- mosomes not carrying the 90-bp allele, we assumed a Poisson distribution to obtain a lower 95% confidence bound for this num- ber. With these two bounds, we obtained a maximum age of 313,000 Y.B.P. We also performed other conservative analyses [notes 40 and 41 in (1)], which gave addi- tional estimates of maximal dates ranging to 450,000 Y.B.P. All of these estimates of maximum age depend on the assumption that the Alu(-) allele has a maximum age of 5 million years and originated in Africa. This upper-bound estimate was used because the allele was not observed in chimpanzees or gorillas. Prit- chard and Feldman state that the mutation could technically be even older, but they also agree that it is far more likely that this polymorphism is less than 5 million years old. A younger age seems likely because of the lifetime survival distribution for neutral mutations (2). In fact, our data argue for a more recent origin, albeit still ancient [note 42 in (1)]. Comparing variation in STRP allele size (calculated by any of several methods) shows that Alu(-) chromosomes have less variation than do Alu(+) chro- mosomes and are therefore likely to have a more recent coalescent. Pritchard and Feldman use coalescent theory and a simulation to calculate a lower 95%/o confidence bound for NA[t. The sam- ple of chromosomes on which their analysis is based derived from 10 extremely disparate African populations, spanning the entire continent, for which there must have been considerable relative endogamy. Such pop- ulation structure would make more recent ages for the Alu(-) allele far less likely than would appear in Pritchard's and Feld- man's simulation (3). Also, it is implausible that the population has been constant in size since the Alu deletion first occurred. Its rather high frequency in Africa suggests a rapid increase in the numbers of this chro- mosome soon after its introduction. Such growth would lead to a smaller estimate of variance for NAIL than that calculated by Pritchard and Feldman. Still, even under assumptions implausi- bly more conservative than ours, the upper bound for the estimate of the coalescent date of the Alu(-) chromosome in non- Africans is about 700,000 Y.B.P. (using Pritchard's and Feldman's estimate), still short of the 1 million years speculated by the "Multiregional" model. Their analysis thus supports our conclusion that a more in their article "Comparative Earth history and Late Permian mass extinction" (1), A. H. Knoll et al. suggest that Late Permian extinctions were caused by the release to the atmosphere of massive quantities of carbon dioxide (CO2) from the deep ocean; that the CO, buildup in the ocean resulted from pri- mary production in the surface layer; and that, despite sluggish ocean circulation rates, the release of phosphorus from decaying or- ganic matter in deep anoxic waters would have been sufficient to further stimulate photosynthesis (2), which would in turn have led to further organic decay (that is, positive feedback) before oceanic overturn and release of CO2. Knoll et al. otherwise deemphasize the role of nutrients in the Permian extinctions, but if ocean circulation had been sufficient- ly slow in the Late Permian, phytoplankton could have largely stripped the surface recent date for an exodus of modern hu- mans from Africa is more likely and that the CD4 data argue for the "Out-of-Africa" model rather than for the "Multiregional" model. We originally stated (1) that the data we have obtained for the CD4 locus rep- resent only a single realization of evolu- tionary history for Africans and non-Afri- cans. As Pritchard and Feldman point out, it is tenuous to derive statistical distribu- tions for coalescent times based simply on theory because of the arbitrary demo- graphic assumptions required. The best way to derive such a distribution is empir- ically, combining the results of numerous different loci. Examination of linkage dis- equilibrium patterns for other systems in a fashion similar to what we have presented for CD4 should provide more definitive conclusions regarding the coalescence time for non-Africans. Neil Risch Department of Genetics, Stanford University School of Medicine, Stanford, CA 94305, USA Kenneth K. Kidd Sarah A. Tishkoff Department of Genetics, Yale University School of Medicine, Post Office Box 208005, New Haven, CT 06520, USA REFERENCES 1. S. A. Tishkoff et al., Science 271, 1380 (1996). 2. N. Takahata and M. Nei, Genetics 124, 967 (1990); N. Takahata, Mol. Biol. Evol. 10, 2 (1993) 3. N. Takahata, Genetics 129, 585 (1991). 3 October 1996; accepted 10 October 1996 mixed layer of nutrients (3) so that a "nu- trient collapse" could have occurred. Also, the expansion of gymnosperms during this time (4) and the greatly increased interior drainage associated with the formation of the Pangean supercontinent (5) could have sequestered large amounts of nutrients on land (4, 6). Greatly decreased nutrient availability during the Late Permian is con- sistent with the loss of many suspension- feeding invertebrates and nekton and the differential survival of infaunal taxa that fed on organic-rich sediment (6, 7, 8), as de- scribed by Knoll et al. Moreover, before Late Permian extinctions, the Permo-Car- boniferous was a time of increasing nutrient and food availability in the water column (6, 7). Thus, just as global marine ecosys- tems were becoming increasingly depen- dent on greater food availability in the Late Paleozoic, the rug, so to speak, could have SCIENCE * VOL. 274 * 29 NOVEMBER 1996 Late Permian Extinctions 1 549 Downloaded from https://www.science.org on January 16, 2022