co O co or N- cs) (0 ) C c cDm cD ,iHs *0302 The Myth of Eve: Molecular _ II II " IIIIIIil _lI , IHs*1103 Biology and Human Origins Hs *0302 I I I 11 1 Pt*0302 Francisco J. Ayala Hs*1103 I , , Il ,, , ,, XPt*0309 Fig. 1. Comparison of two human (Hs) and two It has been proposed that modern humans descended from a single woman, the "mi- chimpanzee (Pt) DNA sequences. The segments tochondrial Eve" who lived in Africa 100,000 to 200,000 years ago. The human immune compared are 270 nucleotides that make up exon system DRB1 genes are extremely polymorphic, with gene lineages that coalesce into an 2 of the DRB1 gene. The differences between ancestor who lived around 60 million years ago, a time before the divergence of the apes sequences are represented by vertical bars. from the Old World monkeys. The theory of gene coalescence suggests that, throughout the last 60 million years, human ancestral populations had an effective size of 100,000 other two genes, Hs*0302 and Pt*0302, are individuals or greater. Molecular evolution data favor the African origin of modern humans, also more similar to one another than to but the weight of the evidence is against a population bottleneck before their emergence. Hs-*l 103 or Pt*0309, and they are also The mitochondrial Eve hypothesis emanates from a confusion between gene genealogies paired in Fig. 2. The two doublets are then and individual genealogies. linked to indicate that all four genes derive from a common ancestral gene. Branch lengths can be made proportional to the number of nucleotide changes that occurred The DNA of organisms embodies virtually ancestor. Other mutations are favored by in each branch, as inferred from Table 1. unlimited information about their evolu- naturalselection because they benefit the One noteworthy property of these four tionary history. The information is encoded organism.Some of these mutationsspread genes is that the two human sequences are in the linear sequence of the four nucleo- among the individualsand accumulatein more different from one another than tide components of, DNA (adenine, cy- the specie, and this process is also time- each is from one chimpanzee gene. Hence, tosine, guanine, and thymine) in the same dependent.The regularityof the processby the lineages of the two human genes di- fashion as semantic information is encoded which nucleotide substitutions occur makes verged from each other more than 6 mil- in the sequence of letters of an English text. it possibleto reconstructthe historicalrela- lion years ago (Ma), the approximate time Molecular evolution investigations of DNA tions amongspecies and also to assigndates when the lineages of humans and chim- sequences have been largely directed to- to relevantevents. That is, there is a molec- panzees diverged. The ancient origin of ward reconstructing the phylogeny (evolu- ularclock of evolution,which is not a met- these and other' DRBl gene lineages is a tionary history) of species relations. But ronomicclock like ordinarytimepieces that property that makes them particularly many other issues can be explored. I have measuretime intervalsprecisely, but rathera suitable for ascertaining the history of an- used genes of a particular kind, those in- stochasticclock that is dependent,like ra- cient human populations. volved in the immune response, to investi- dioactivedecay, on events that occur with gate the size of human populations from the constantprobabilities. Genealogy of DRB1 Genes time of the origin of our species and earlier The evolutionaryinformation encoded in order to ascertain whether the existence in the DNA has the notoriousattribute of DRB 1 is one of -100 genes that make up of the mythical Eve has been confirmed by being effectively inexhaustible.The nucle- the human leukocyte antigen (HLA) com- science, as some reports claim (1). otide sequence of a single gene or small plex, whiclh extends over a DNA segment 4 Evolution is a time-dependent gradual DNA fragmentis often sufficientto solve a million nucleotides long located on chro- process. At the genetic level, evolution oc- particularevolutionary question, but organ- mosome 6. The HLA genes specify mole- curs by accumuLlatedsubstitutions of one nu- isms have many genes (-100,000 in pri- cules with a critical role in tissuLecompati- cleotide by another in the DNA of the or- matesand other mammals).The evolution- bility and in defense against pathogens and ganism. Nucleotide mutations arise with ary informationobtained by investigating parasites. These genes are arranged in two constant probabilities, but most are lost by one gene can be supplementedwith tlhe distinct groups, class I and class II, separated chance shortly after their origin. The fate of investigationof additionalgenes until the by several dozen genes that have functions the rest depends on their effects on the informationbecomes sufficient to settle a mostly unrelated to the immune response organism. Many mutations are injurioulsand particularissue. The practicallimits are re- (Fig. 3). are readily eliminated by natural selection. sources,financial costs, and time. The HLA complex is homologous to the Other mutations are adaptively neutral or major h-istocompatibility gene complex nearly so; that is, the replacement of one The Reconstruction of Genealogy (MHC) of mammals and other vertebrates nucleotide by another is of little or no con- (3). MHC molecules on the surfaces of cer- sequence to the organism's welfare. These The methodology used to reconstruct evo- tain cells bind protein fragments (antigens) mutations occasionally spread and become lutionarygenealogies can be illustratedwith and present them to lymphocytes called T fixed in the species at rates that are constant, a simple example (2). Figure 1 shows a so that the number of differences between schematic comparisonamong four DRB1 Table 1. Numbers of nucleotide differences two species is roughly proportional to the gene sequences,two fromhumans and two among four DRB1 genes, two from humans (Hs) time since their divergence from a common fromchimpanzees, each of 270 nucleotides; and two from chimpanzees (Pt). the datafor the six pairwisecomparisons are The author is the Donald Bren Professor of Biological given in Table 1. The most similarpair are Gene Hs*11O3 Pt*0309 Pt*O3O2 Sciences at the Universityof California,Irvine, and chair- the human gene Hs*1103 and the chim- man of the AAAS Board of Directors. This article is adapt- panzeegene Pt*0309,which have been ar- Hs*0302 18 18 12 ed from a lecture delivered at the 161 st annual meeting of Hs*11O3 9 20 the American Association for the Advancement of Sci- rangedin Fig. 2 to indicatethat they share Pt*03O9 20 ence in Atlanta on 18 February 1995. a recent ancestor with each other. The

1930 SCIENCE * VOL. 270 * 22 DECEMBER1995 Hs*0302 relative to those from other species. For populationwould be heterozygotes(that is, Pt*0302 example,the nine human genes at the top carryingtwo differentgenes), each for two of Fig. 4 are closely relatedto one another, genes differentfrom all others. The theory Hs*1103 but they are moredistantly related to other of gene coalescenceand other mathemati- Pt*0309 humangenes in the genealogythan they are cal theories formulatedin the last decade Fig. 2. Genealogical tree of the four DRB1 genes to the six genes immediatelybelow them, providemodels and analyticaltools for ex- in Fig. 1, representing their patterns of evolution- includingone drill gene and four macaque tractinginformation about human popula- ary divergence. Branch lengths are proportionalto genes. The evolutionarydivergence of the tions from the DRB1 genealogy. the nucleotide differences between the genes (Ta- hominoids(humans and apes)from the cer- The coalescence theory examines the ble 1). copithecoids (Old World monkeys) oc- genealogicalrelations between genes (11). curredaround 35 Ma, at the boundarybe- Accordingto this theory,all genes (alleles) cells. When antigen-presenting cells contact tween the Eocene and Oligocene epochs. presentin an extant populationmust have T cells that bear receptors matching a par- The relationsshown in Fig. 4 demonstrate descendedfrom a singlegene, to which they ticular combination of protein fragment and that several human gene lineages already coalesce. The theory was first formulated MHC molecule, the T cells are stimulated to existed at that time. for neutralor nearly neutralgenes that do proliferate and to initiate the specific arm of The ageof the humanDRB 1 lineagescan not modifythe welfareof the organism.In a the immune response, including the secre- be appraisedby calibratingthe DRBl molec- randommating populationat equilibrium, tion of specific antibodies. ularclock, which yields a rateof 1.06 x 1o- the mean coalescencetime of neutralgenes The recognition of protein fragments is nucleotide substitutionsper site per year is given by T = 4N[1 - (I/i)] generations, mediated by a specialized groove on the sur- (10). Figure5 displaysa genealogyof the 59 where T is the numberof generationsto face of the MHC molecule, called the pep- humanDRBl genes,constructed by a meth- coalescence, N is the effective size of the tide-binding region, that consists of some 50 od that usesthe averagerate of evolutionto population, i is the number of sampled amino acids (4). The composition of these determinethe length of branches.The time genes, and the variance is large (12). For amino acids varies from one MHC molecule scale at the bottom highlightsthree points any two genes (i = 2), the mean coales- to another, and this variation is responsible that approximatelycorrespond to important cence time reducesto T = 2N generations; for the tremendouspolymorphism character- events in human evolution:the divergence for a large numberof genes, the mean co- istic of the MHC molecules and their encod- of the orangutanlineage at 15 Ma, the di- alescencetime is T 4N. Thus, in a pop- ing genes. In people, as well as in other vergenceof humansand African apes (chim- ulation with N = 1 million individuals, mammals, scores of gene variants (alleles) panzeesand gorillas)at 6 Ma, and the emer- genes are expectedto convergeto their one may exist at any one of several MHC loci, gence of Homoerectus at 1.7 Ma. ancestor4 million generationsearlier. and some of the allelic pairs may differ at These points of referenceare useful for The coalescenceequation can be used in >100 nucleotide sites (5). The MHC poly- determiningthe numberof gene lineagesin the oppositedirection, so that we can esti- morphisms are ancient, with gene lineages existence at a given time, which is done by matepopulation size if the coalescencetime that can be traced back for millions of years, countingthe numberof lineagesintersected is known.The humanDRB 1 genes coalesce in primates (6-8) as well as in rodents (9). by a vertical line drawn from the time to their last common ancestorat -60 Ma Figure 4 represents the genealogy of 119 point. Forexample, 32 humanlineages were (Fig. 5), in the middlePaleocene, 10 to 20 DRB 1 genes, of which 59 are from humans, alreadyin existence around 6 Ma, which million years(My) beforethe divergenceof 40 are from apes, and 20 are from Old implies,correspondingly, that the other 27 the New Worldand Old Worldmonkeys. If World monkeys (Table 2). In this genealo- lineages arose after the hominid lineage we assumethat the averagegeneration time gy, the length of each branch is proportion- divergedfrom the African apes. The gene- over this long span is 15 years (which is al to the number of nucleotide substitutions alogy of all human genes coalescesat -60 surely too high), the coalescence occurs 4 that have occurred in that lineage. Thus, Ma, which is to say that the humanDRBl million generationsbefore the present. If we see at the bottom of the figure that of genes starteddiverging that long ago. the two macaque genes, Mm*0301 and Mm*0302, the first one has changed more Human Ancestral Populations Table 2. Species and number of DRB1 genes. than the second. If 32 DRB1 gene lineages have persisted Sym- Scientific name Age of the Human since 6 Ma, it followsthat no fewerthan 16 bol (common name) Genes DRB1 Lineages individualscould have lived at any given Apes time over that long span. The minimum Hs Homo sapiens (human) 59 The DRB1 human gene lineages are very number of individuals must have been Pp Pan paniscus (bonobo 4 ancient, as is apparent from the distant much larger,because the probabilityis ef- or pygmy associations between some human genes fectively zero that all 16 individualsin a chimpanzee) Pt Pan troglodytes 24 (chimpanzee) Gg Gorillagorilla (gorilla) 10 ClassII region ClassI reagion Or Pongo pygmaeus 2 DPB1DPA1 DQBIDQAI DRB1 DRB3 B C A (orangutan) Old Worldmonkeys Mm Macaca mulatta 16 38 8 1914 59 4 60 18 41 (rhesus macaque) Mn Macaca nemestrina 1 0 300 400 700 800 900 2000 2100 2200 3400 3500 3800 (pigtailmacaque) Ml Mandrillusleucophaeus 2 Fig. 3. Location of some polymorphicgenes withinthe HLAcomplex in human chromosome 6. There are (drill) two sets of genes, class I and class 11,separated by a region with unrelated genes. The number of variant Ph Papiohamadryas 1 genes (alleles) known at a locus appears under the box that indicates the location of the gene. The scale (hamadryasbaboon) below gives the number of nucleotides (x 1000) along the segment.

SCIENCE * VOL. 270 * 22 DECEMBER1995 1931 the DRB1 polymorphismwere neutral,the mate of the mean of the population over time, with the value of 4N = 60 My obtainedby coalescencedate wouldrequire a long-term the estimate refers to the hannonic mean, reconstructingthe genealogy. populationof 1 million individuals. which has the important property of being We need to qualifythis conclusionwith affected disproportionately more by smaller Experimental Simulated two observations.First, although N is an esti- than by largernumbers. The estimate of N is Populations therefore consistent with population sizes much greater than 1 million for many gener- The coalescence theory draws inferences H.-0406 H. .0407 ations, but it is not compatible with much about past events on the basis of observa- rHes40401 Hs040 0407 H". . 0402 smaller population sizes for very many gener- tions about currentpolymorphisms. These rHs-0405 Hs-0410 F. .80 4 124 Hs.04 1 1 ations. Second, because of the large variance Hs00402 ,m1-0401 of the coalescence equation, the estimate of N M0H 0403 t ES412 is readilycompatible with values between, say, H.-0407 M.-0404 Hs '0408 L ~~~r-r1l-0401 500,000 and 2 million individuals (13). Hs'0401 'M.-0402 Hs'0406 F-Hs-0403 Pst0101 The previous calculations are for neutral H.. __ Hs '0404 DRB1 are subject to Hs-0 05 genes, but the genes _H:::0409 rH*410 Hg0 101 overdominant natural selection, that is, het- s-041H 1 Gg-0101 Hs-0402 Hei0701 erozygotesfor HLA genes are better off than Hs04 12 Hs.0702 _.Hs 1304 Pt 0702 _ are homozygotes (3, 14); for instance, het- _ ~~Hs e1303 Pt .0701 __ lHs-0805 H..-'090 11 erozygotesdisplay enhanced resistanceto Plas- Hs'08031 Hs'09012 {&Hs-080 1 Mm.0101 Hs-08032 _ .0 102 modiumfalciparum, the parasite that causes M Hs' 1402 Hs.040e malignant malaria (15). Overdominant selec- Hs' 1409 -HS-0409 Hs-1406 nHO1501 Hs'1403 H:- 1503 tion, like other forms of balancing selection, .502 . L I ~~~Hs* 084 a~ Hs 80e021 1601 Hr. increases the probability that gene polymor- E,HsoeO822 Hs-0103 { H_ -1602 phisms will persist over time, and thus it _ ~~~~~Hs-0101 Gg.0201 . ~~~~~~~Hs-0102 PtP 0205 reducesthe numberof individualsrequired for Hs-1001 Pt- 0213 Hs' 1501 Pt '0212 their persistence. Hs' 1503 I Hs 1502 Pt90207 In the case of balancing selection, the Hs 1601 l Pt 0209 LHs- 1602 Pt .0202 coalescence process has the same structureas Hs- 1301 PtI 0204 rEPp.0202 in neutral gene genealogy, except for a scal- Pt '0210 1302 __Hs- Hs* 13106 Pt .0203 Pt 0201 ing factor, fs so that the time to coalescence 1 Pt .020e Hs0 101 in generations is estimated by T = Nfs (16- Hs- 1 103 H-. .1305 ~Hso11012 'Gg.0306 18). The scaling factor depends on the se- HHs- l 1041 Gg.0307 HsFl11042 H.. 1201 lected mutation rate as well as on the selec- Hs-0302 I H. 1202 _~~~~~~~~,,O G9030 - r' ~~Hs-0303 0,'0301 tion coefficient, s, which measures the ad- Hs-030 1 Lto* 0302 Hse 1405 H.. oe04 vantage accruing to the heterozygotes. Esti- . ~~~~~~Hs-1407 103031g. Gg'103ee mates of s range from 0.0007 to 0.019 (7, 14, Hs-1401 15). If we assume values of s = 0.01 to 0.03, Hs- 1 404 ___ _ r ~~~H.-09011l H s' 060241 the size for maintaining Hss 090 1 2 Hs.oe10 population required Hs-1201 Hsoe.0032 -125,000 to ! t 12 Pt .030e the DRB1 polymorphism is F4~~~~~~~Vs-e' _rHs 1P eO20309 -300,000 individuals (Table 3) (19, 20). Fig.~~~~~~~5.Geeloia tre of 591 5 hua6 1.7 DRB The time to coalescence has been esti- Hs- 1301 Hs. 1302 at 60 on the basis of the geneal- 60 50 40 30 20 10 0 | 5 mated My EPHs' aK11 1 :n102 3 Hs 1103 M ogy reconstructed in Fig. 5. We can reach My the same estimate by a somewhat different Fig. 5. Genealogical tree of 59 human DRB1 Hs' 1 1012 route. According to the theory of coales- Hs'. 1303 genes (exon 2), constructed according to the 'av- Ho 1304 cence, if we sample any two genes at ran- also known as PP10301 erage linkage" method [(39), p. 31; M--,.0303 for coales- the unweighted pair-group method using arith- Ml .0301 dom, the expected value their 0301 metic averages (UPGMA)(43)], on the basis of - L'.0304 cence is 2N. Figure 6 displays the distribu- Hs-1305 genetic distances (D) estimated by Kimura'stwo- He- 1403 tion of genetic distance between pairs of the Pt .0301 parameter method (42) for the 1711 pairwise Pt .0302 59 human DRBJ alleles. The mean distance Gg.0304 comparisons between the genes. The average Pt.0304 value is 0.067, which, on the basis of the t- 0305 linkage method assumes equal rates of evolution Pt .0303 rate of the DRB1 evolutionary clock, yields Pt .0306 along the branches. The time scale is based on a Pt .0310 an average coalescence time of 2N = 31.6 x PP.0301 rate of evolution of 1.06 10-9 substitutions per Pp.0302 My. This value is in reasonable agreement the mini- Ho. 1401 nucleotide site per year, estimated by H..1407 mum method (8) on the basis of genetic distances H.. 1404 H..1410 between species that diverged at differenttimes. Ho. 1400 i Ho.1405 Fig. 4. Genealogical tree of 1 19 DRB1 genes (exon The species compared for estimating the rate of L Ho-.1409 H..1402 2) from humans (Hs) and other primates (Table 2). evolution are Homo-Pan, Homo-Gorilla,and Pan- and sH. 1406 The tree was made with the use of the neighbor- Gorilla at 6 My; Homo-Pongo, Pan-Pongo, H.-0302 joining algorithm (41) on the basis of genetic dis- Gorilla-Pongo at 15 My;Pan-Macaca and Gorilla- IL r HS10303 Ho .0 301 tances (D) estimated by Kimura'stwo-parameter Macaca at 35 My; and Homo, Pan, and Gorilla, Mm.0308 Ggo0302 method (42). The matrixof genetic distances has each compared with species of New World mon- Gg.0303 I4- Mp'.0309 7021 entries that represent all pairwise compari- keys at 40 My. The regression equation is D = I 11 1 . .. ~~~~Mr.-030I Mm,f FtM030302 sons between the 1 19 genes. Branch lengths are 0.0099 + 0.00106 My; if the minimum-minimum _F GGg.0305 Mm.0307 proportionalto the estimated nucleotide substitu- method (8) is used instead, the regression equa- Mm.0305 Mm.0306 tions along the branch; the scale bar represents tion is D = 0.0027 + 0.00111 My. The fit in both 0.01 D. Sequence sources are listed in (5). cases is r > 0.85.

1932 SCIENCE * VOL. 270 * 22 DECEMBER1995 ?EESSC<+ S ; E S 4S S>ASSOCIATION AFFAIRS

inferencescan be testedby computerexper- persistat the end of the experiment. heritance of the mitochondrial DNA iments, in which the time direction is re- Mutation is ignored in the simulated (mtDNA) is distinctive and follows a pat- versed and the process is examined as it populations.If the mutationrate to alleles tern of maternal inheritance. Sons and proceeds from past to present conditions. favored by natural selection is 5 X 10' daughtersinherit mtDNA fromtheir moth- The experimentssimulate populations con- mutations per gene per generation (8), a ers, but only the daughterstransmit it to sisting of a certain numberof individuals new mutation appearsevery 100 genera- their progeny. that reproduceeach generationby random tions in a populationof N = 10,000 indi- Analysesof the mtDNA from>100 eth- reassortmentof genes. Figure7 presentsthe viduals.The mutationprocess increases the nically diverseindividuals have shown that resultsof threesets of experiments.The fate numberof genes in the populations,but it the mtDNA sequencesof modernhumans of 60 differentgenes, initially in identical would not be nearlysufficient to maintain coalesceto one ancestralsequence, the "mi- frequencies,is examinedover 100,000 gen- 60 genes if s 0.01 to 0.02. With N = tochondrial Eve" that existed in Africa erations, a time approximatelyequivalent 10,000 and s 0.01 or 0.02, the processof -200,000 yearsago (21). This Eve, howev- to the 1.7 My fromthe emergenceof Homo eliminationof the first40 genes is veryfast. er, is not the one mother from whom all erectusto the- present. The probabilityof Forpopulations of N = 100,000 individ- humans descend, but rather a mtDNA mol- reproductionis s = 0 to 0.2 higher for uals (Fig. 7, bottompanel), with s = 0.01 to ecule (or the woman carrierof that mole- heterozygotesthan for homozygotes. 0.03, 30 to 50 genes remainat the end of cule) fromwhich all modernmtDNA mol- Forpopulations of N = 1000 individuals the experiment.If we assume,as above, a ecules descend.The inferencethat all hu- (Fig. 7, top panel),without selection, all 60 selected mutationrate of 5 x 10-7 muta- mans descend from only one or very few genesbut one arerapidly eliminated, consis- tions per gene per generation,a new over- women (1) is basedon a confusionbetween tent with the theoreticalexpectation that dominantmutation will ariseevery 10 gen- gene genealogies and individual genealo- fixation for one gene will occur in 4N = erations.Most newly arisenmutations have gies. Gene genealogies graduallycoalesce 4000 generations.If the heterozygoteshave s a high probabilityof being quickly lost by toward a single DNA ancestralsequence = 0.01 to 0.02 advantage,which is the chance, even if they are favorablyselected, (such as in Figs.4 and 5), whereasindivid- magnitudeoperating in real populations,all but mutationwould increasethe numberof ual genealogies increase by a factor of 2 genes but two or threeare eliminated.Even persistinggenes. Dependingon the muta- each generation: An individual has two with unrealisticallylarge s advantagesof 0.1 tion rate, the numberof genes might ap- parents,four grandparents, and so on. (The to 0.2, only 8 to 11 genes persist after proach or reach 60, the numberof DRB1 100,000 generations.Populations of 10,000 genes. In conclusion,th-ese computer-simu- 201 n= 60 individuals(Fig. 7, middle panel) are like- lated populations yield results consistent N= 1000 wise incapableof maintaining60 genes over with the theoretical calculation that to 10 s =0.2 many generations.Without natural selec- maintain -60 DRB1 genes, human ances- 0=0.1

tion, all genes but one are soon eliminated. tral populations must have consisted of s = 0.02 s 0.01 With s = 0.01 to 0.02, only 9 to 11 genes 100,000 or more individualsover their long 0 s= 0.00 history. 0 10 20 30 40 50 60 70 80 90 100 60 - n= 60 Table 3. Effectivepopulation size, N, as a function The Mitochondrial Eve NN= 10,000 of various parameter values. T, time in genera- 50 tions; s, selection advantage of the heterozy- 40 gotes; u, mutationrate per gene per generation for Most of the genetic informationis storedin s =0.2 selected alleles. Ifthe long-term generation time is the chromosomesinside the cell nucleus. 15 years, then T = 4.0 x 106 corresponds to 60 The total DNA in a human cell nucleus My. Ifthe mutation rate per nucleotide site per year amounts to 6 X 109 nucleotides or base *l 30 s=0.1 is m = 10-9, half the sites yield selected alleles (n pairs (bp), half in each set of 23 chromo- E~10iL,s o= 0.0 = 135), and the generation time is g = 15 years, somes that are inheritedfrom each parent. 2 0 s= 0.00 then u = 2.0 x 10-6; the same value of u results 0 10 20 30 40 50 60 70 80 90 100 if m = 5 x 10-9, n = 40, and g = 10. A relatively small amount of DNA (-16,569 bp) exists in the mitochondria, 60 s=0.1

T u s N cell organellesoutside the nucleus.The in- 50 , ss=0.03 s = 0.02 40: 4.0 x 106 0 1,000,000 s =0,01 4.5 x 10-6 0.01 297,789 140 30 0.02 257,430 20 n=60 0.03 230,213 120- N= 100,000 2.0 x 10-6 0.01 194,240 10 0.02 149,072 s = 0.00 100 0O 0.03 125,343 0 10 20 30 40 50 60 70 80 90 100 3.0 x 106 0 750,000 80 4.5 x 10-6 0.01 211,793 Time(103 generations) 0.02 178,691 60- Fig. 7. Loss of genetic polymorphism in comput- 0.03 157,795 er-simulated populations initiated with 60 genes 2.0 x 10-6 0.01 40 131,196 -ll (alleles)at identicalfrequencies. The 100,000 gen- 0.02 98,992 erations correspond to -1 .7 My, the time elapsed 0.03 82,505 20 ~ - since the emergence of Homo erectus. The effec- 2.5 x 106 0 625,000 tive size of populations is N individuals;the selec- x 4.5 10-6 0.01 169,720 tion advantage of the heterozygotes is s. The per- 0.02 141,111 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 sisting genes are censused at 100-generation in- 0.03 123,668 D by random 2.0 x 10-6 0.01 101,920 tervals.The populations"reproduce" 0.02 76,11 8 Fig. 6. Distributionof 1711 genetic distancesbe- sampling,at each generation,of N pairsof genes 0.03 63,1 05 tween pairsof 59 humanDRB1 genes (exon 2); froman infinitepooi that has the gene frequencies the mean is D = 0.0675 + 0.0276. observedin the previousgeneration.

SCIENCE * VOL. 270 * 22 DECEMBER1995 1933 theoreticalnumber of ancestorsfor any one mean populationsize that rangesbetween history.We want,nevertheless, to explorethe individual becomes enormous after some 10,000 and >50,000 individualsthrough- likelihoodthat a bottleneckdid occur.When tens of generations,but "inbreeding"oc- out the Pleistocene. This finding is con- the numberof individualsin a population curs;that is, aftersome generations,ances- sistent with the estimate based on the oscillatesfrom generation to generation,small tors appearmore than once in the geneal- DRBI polymorphismof a population size numbershave a disproportionatelylarge effect ogy.) Coalescence to one ancestral gene on the order of 100,000 individualsover on the valueof N, becauseN is the harmonic originallypresent in one individual does the last 60 My. meanof the populationsize over time. There- not disallowthe contemporaryexistence of fore, very narrowor long-lastingbottlenecks many other ancestorsfrom whom we have The ZFY Adam areinconsistent with the DRB1results. How- inheritedthe other genes. ever, the questionremains whether an occa- The conclusion warranted by the A genetic gender counterpart of the sional populationbottleneck may have oc- mtDNA analysisis that the mitochondrial mtDNA is the Y chromosome,which is curred,and if so, how smallit couldbe. It has Eve is the ancestorof modernhumans in the transmittedfrom fathers to sons. Regionsof been suggestedthat a populationbottleneck maternalline. Any personhas a single an- the Y chromosomedo not have homologs occurredat the transitionfrom archaic to cestor in the maternalline in any given in the X chromosomeand thus are trans- modernH. sapiens,some 100,000to 200,000 generation. Thus, a person inherits the mittedthrough the paternalline, just as the yearsago (1, 21, 24). mtDNA from the great-grandmotherin the mtDNA is inheritedthrough the maternal The consequencesof a populationbot- maternalline, but also inheritsother genes line. A 729-bp fragmentof the Y-chromo- tleneck depend not only on the size of the fromthe threeother great-grandmothers and some ZFYgene (thought to be involved in bottleneck,Nb, but also on the numberof the four great-grandfathers(about one- the maturationof testes or sperm)has been bottleneckgenerations, tb. A usefulmeasure eighth of the total DNA from each great- sequencedin 38 men, of diversegeographic for evaluatingthe effects of a bottleneckis grandparent).The mtDNA that we have origins and representativeof majorethnic the ratio Nb/tb, which if <10 leads to a inheritedfrom the mitochondrialEve repre- groups,and no variationwas observed(24). drasticreduction in genetic polymorphism sents a four-hundred-thousandthpart of the Comparisonwith the homologousregion of (18). Thus, a bottleneckof 100 individuals DNA present in any modernhuman. The the greatapes yields a rateof substitutionof wouldsubstantially reduce genetic variation restof the DNA, 400,000 timesthe amount 1.35 X 10-9 nucleotides per site per year. if it lasted 10 or more generations, but of mtDNA, was inheritedfrom other con- Assuming that substitutions in this DNA balancing selection facilitates the persis- temporariesof the mitochondrialEve. segment are neutral, coalescence theory tence of polymorphismsthrough a bottle- The theoryof gene coalescencemakes it yields 270,000 years as the expected date for neck (25). The persistenceof HLA poly- possibleto estimatethe numberof ancestors the last common ancestor of the ZFY gene morphismsover millions of years requires who were contemporariesof the mitochon- of modern humans, with 95% confidence that the sizeof humanancestral populations drial Eve. The mtDNA is inherited as a limits from 0 to 800,000 years (24). The be at least Ns = 10 at all times (7, 17). If s single copy, from only one parent,and the carrier of the ancestral ZFY gene predicted = 0.01, the minimumpopulation size pos- mtDNA polymorphismis largely neutral. by these calculations is the ancestor of all sible at any time would thereforebe Nb = According to the theory, the coalescence modern humans in the paternal line. As was 1000. The minimum number must have into a single ancestralmolecule is expected the case for the mitochondrial Eve, this been in fact much larger,because human to be T 2Nfgenerations ago, whereNf is "ZFY Adam" is the individual from which population bottlenecks cannot last just a the numberof mothersper generation. This all humans have inherited the ZFY gene, few generations.Many generationsare re- inferenceassumes constant populationsize but he is not our only ancestor in his gen- quired for a human population to grow and other conditions that are unlikely in eration. We have inherited the other thou- from 1000 to its long-term mean, which reality,so that the conclusionsreached are sands of genes from many other contempo- we have estimated to be -100,000 indi- only roughapproximations. If we assume20 raries of this Adam. viduals.The rate of growthof human pop- yearsper generation,the 200,000 years of The expected coalescence of a DNA ulations throughout the Pleistocene has the mtDNA coalescenceyield T 10,000 polymorphism that is transmitted in a single been estimated to be -0.02% per genera- generations;therefore, Nf 5000 mothers copy and paternally inherited is T = 2NTm, tion (26). and N = 10,000 individuals,which is al- where Nm is the number of males. If we Computer-simulatedpopulations are use- most certainlyan underestimate(22). assume 20 years per generation, the coales- ful for exploringthe minimumbottleneck Otherestimates place the coalescenceof cence to the ancestral ZFY gene yields an size that would allow the persistenceof the the mtDNA at 143,000 and 298,000 years effective population size of 6750 fathers, or DRBI polymorphism.The resultsin Fig. 8 ago (23). One recent analysis,which takes 13,500 individuals, with a 95% confidence are basedon 200 separatecomputer runs. If into account variablesubstitution rates for upper limit of N 40,000 individuals. If we there is no selection (s = 0) and we ignore differentsites of the mtDNA genomeand is account for the standard deviation of the the time requiredfor a populationto grow supportedby computersimulations, yields mean coalescence, the 95% upper limit for back to its long-termpopulation size, and if an estimateof 622,000 to 889,000 yearsago N would increase to 80,000 individuals. we assume that 70 gene lineages were (22) and correspondsto N = 31,100 to The ZFYresults are thus consistent with the presentbefore the bottleneck,the smallest 44,450 individuals.As noted earlier, esti- mtDNA-derived and DRB I -derived esti- bottleneckthat wouldallow the persistence mates of mean coalescencetime as a func- mates that human ancestral populations of 60 of these 70 gene lineagesis 510 to 550 tion of N, and vice versa, have largevari- have been -100,000 individuals for mil- individuals(Fig. 8, top panel). When we ances. When the sample of genes is large, lions of years. accountfor the time requiredfor the popu- the standarddeviation of the mean for mi- lation to recover to its average size, the tochondrialDNA is >N/2 (13). The 95% Population Bottlenecks minimumpopulation size at the bottleneck confidence interval coalescence corre- becomes substantiallylarger. Assuming a spondingly extends at the upper end to Neither the mtDNA results nor the ZFY re- rate of population increase of R =1% per >88,900 generations and an equal number sults lead to the conclusion that narrow pop- generation (which is 50 times the average of individuals. Thus, despite considerable ulation bottlenecks consisting of one or very growth rate of humnanpopulations through- uncertainty, the mtDNA results yield a few couples have occurredin human ancestral out the Pleistocene) (26) and population

1934 SCIENCE * VOL.270 * 22 DECEMBER1995 ...... ~ ....;...Ef .E...... S! ..;.:;::::E ....iS...... SS ...:; ...Ei ..Q.EE ; .S....S ~~ ~ ~ ~ ~ ~ ~ ~ ~~ ~ ~ ~~~~~~~~~FS E i: E : i E EE iEE E::;: :::E: :::: iE ::;: : : ...: ...... ; s E ; growthto 100,000individuals, 4490 to 4590 Census Populations anderthalshave been thought to be ances- individualsare the minimumbottleneck for tral to anatomicallymodern humans, but passing60 alleles (Fig. 8, top panel). The parameterN that I have used for estimat- now we know that modern humans ap- Overdominant selection reduces only ing the size of human populations is a theo- peared at least 100,000 years ago, much slightlythe minimumnumber of individuals retical construct that corresponds approxi- before the disappearanceof Neanderthal requiredat the bottleneck.When s = 0.01, mately to the numberof synchronouslyrepro- fossils. It is puzzlingthat in caves in the the minimumbottleneck size for passing60 ducing individuals.The census numberis like- MiddleEast, fossils of anatomicallymodern of the 70 alleles is 500 to 540 individualsif ly to be about four or five times greater. In humansprecede as well as follow Neander- we ignorethe requiredgrowth back to long- humans and in other primates, a number of thal fossils.Some modernhumans are dated term numbers,but 4310 to 4380 when we individuals,possibly one-third of the total, do at 120,000 to 100,000 years ago, whereas take populationgrowth to 100,000 individ- not reproduce at all. Of the females who Neanderthals are dated at 60,000 and uals into account (Fig. 8, middle panel). reproduce, only about one-third are actively 70,000 years,followed by modernhumans Even with the unrealisticallyhigh s = 0.1, reproducingat any given time; the others are dated at 40,000 years (34). It is unclear the correspondingvalues become 490 to juveniles or are beyond reproductive age. whether the two formsrepeatedly replaced 530 and 3410 to 3510 (Fig. 8, bottom pan- With the use of these rough approximations, one another by migrationfrom other re- el). It maybe concludedthat, to accountfor we can conclude that N is about two-ninths of gions, or whetherthey coexisted,or indeed the DRB1 polymorphism,the minimum the census population. A long-term effective whether interbreedingmay have occurred. possiblenumber of individualsat a bottle- population size of N = 100,000 corresponds, There is considerablecontroversy about neck is at least 4000; this numberis consis- therefore,to a census population of 400,000 to the origin of modernhumans (22, 33-36). tent with the lower estimatesderived from 500,000 individuals.A bottleneck population Some anthropologistsargue that the transi- the mtDNA and the ZFYgene. of N = 4000 similarlycorresponds to 15,000 tion from H. erectus to archaic H. sapiens, to 20,000 census individuals. Population sub- and later to anatomicallymodern humans, division would affect the value of N estimated occurredconsonantly in variousparts of the 1.0 A B C from the DRB1 genealogy. If human ancestral Old World. Proponentsof this "multire- 0.8 populations consisted of several synchronous gional model" emphasize fossil regional but relatively isolated populations,fewer indi- continuityin the transitionfrom H. erectus 0.6 viduals than were estimated above would be to archaic and then modern H. sapiens. required to account for the DRB 1 polymor- However, they postulate that genetic ex- 0.4 phisms, although the order of magnitude changeoccurred from time to time between 0.2 -s=0.00 would not change (18). populations,so that the speciesevolved as a R 1.01 single gene pool, even though geographic o0. The Origin of Modern Humans differentiationoccurred and persisted,just 0 1000 2000 3000 4000 as geographicallydifferentiated populations The hominid lineage diverged from the exist in other animalspecies (29, 35). This B 1.0 A chimpanzee lineage at -6 Ma, and it explanationis dependenton the postulate 0.8 evolved exclusively in the African continent of persistentmigrations and interbreeding until the emergence of Homo erectus some- between populationsfrom differentconti- C=0.6 what before 1.7 Ma. The first known homi- nents, of which no direct evidence exists. of 0.4 nid, Ardipithecusramidus, lived around 4.4 Moreover, it is difficult to reconcile the Ma, but it is uncertain whether it was bipe- multiregionalmodel with the contemporary 0.2 s= 0.1 dal or in the direct line of descent to modern existence of different species or forms in R 1.01 humans (27). The recently described Austra- differentregions. 0.c lopithecusanamensis, dated at 3.9 to 4.2 Ma, Otherscientists argue instead that modern 0 1000 2000 3000 4000 was bipedal and has been placed in the line humansfirst arose in Africaor in the Middle of descent to A. afarensis, H. habilis, H. Eastsomewhat earlier than 100,000years ago 1.0 A B erectus, and H. sapiens (28). and spreadfrom there throughout the world, 0.8 Shortly after its emergence in Africa, H. replacingthe preexistingpopulations of H. erectusspread to other continents. Fossil re- erectusor archaic H. sapiens(21, 23, 37-39). 0.6 mains of H. erectushave been found in Africa, Someproponents of this Africanreplacement 0.4 Indonesia (Java), China, the Middle East, and modelargue further that the transitionfrom Europe (29-33). Homo erectus fossils from archaicto modernH. sapienswas associated 0.2 s=0.1 Java have been dated at 1.81 ? 0.04 and 1.66 with a verynarrow bottleneck, and that this R=1.01 0.0 ? 0.04 Ma (30), and from Georgia between bottleneckconsisted of a small numberof 1.6 and 1.8 Ma (31 ). Anatomically distinctive individualswho arethe ancestorsof all mod- 0 1000 2000 3000 4000 H. erectus fossils deposited before 780,000 ern humans.The postulateof a narrowpop- Effectivepopulation size years ago have been found in Spain (32). ulationbottleneck derives from a misunder- Fig. 8. Minimum size of bottleneck populations. The transition from H. erectusto H. sapi- standingand need not be entertained.The The ordinate displays the average probability ens occurred around 400,000 years ago, but most serious difficulty with the replacement P60170 (based on 200 computer-simulated pop- there is uncertaintyas to whether some fossils modelis that it leavesunexplained the appar- ulations) that 60 of the initial 70 genes will per- are H. erectusor "archaic"forms of H. sapiens ent morphologicalcontinuity observed in The selective of the sist. advantage heterozy- (33). Moreover,H. erectuspersisted further in some regions, most notably in Australasia gotes is s, and the rate of population growth per Asia, until 250,000 years ago in China and generation is R 1.01. The bottleneck lasts 10 (29, 35). generations. A gives the probabilityof survival at perhaps until 100,000 years ago in Java (33). The reconstruction of the rntDNA gene- the end of the bottleneck; B and 0 take into The subspeciesH. sapiensneaniderthalensis ap- alogical tree places its root in Africa (21, account the growth of the populationto N= peared in Europe around 200,000 years ago 23), consistent with the African origin fa- 10,000 and 100,000 individuals,respectively. and persisteduntil 30,000 years ago. The Ne- vored by proponents of the replacement

SCIENCE * VOL. 270 * 22 DECEMBER1995 1935 model (37-39). This evidence is, however, 5. R. E. Bontrop, Immunogenetics 39, 81 (1994); R. E. a different Y chromosome fragment from five men, Bontrop et al., Immunol.Rev. 143, 33 (1995); J. Klein have estimated the coalescence at 37,000 to 49,000 far fromconclusive. The mtDNA root pin- and F. Figueroa, Crit.Rev. Immunol.6, 295 (1986); S. years ago. The discrepancy between these two es- points only the ancestor in the maternal G. E. Marsh and J. G. Bodmer, Immunogenetics 33, timates and Dorit et al.'s estimate underscores the line, but the much largernuclear DNA may 321 (1991); ibid. 37, 79 (1993); H. McDevitt, tmmu- need for more extensive and accurate data. These nol. Rev. 143, 113 (1995); C. O'hUiginet al., Immu- new estimates are consistent with the conclusions have genealogicalroots in otherparts of the nogenetics 38, 165 (1993); World Health Organiza- reached here concerning the size of ancestral hu- world. tion, Bull. WHO 70, 801 (1992). man populations. Additionalmolecular evidence favoring 6. T. Bergstrom and U. Gyllenstein, Immunol.Rev. 143, 25. Because alleles behave as neutralwhenever Ns < 1, fromthe 13 (1995); W. Fan et al., Hum. Immunol. 26, 107 if the selection is weak (for example, s = 0.01), N Africanancestry derives, however, (1989); U. B. Gyllenstein and H. A. Erlich,Proc. Natl. must be correspondingly large (at least 100) for se- analysisof 30 DNA polymorphismsfrom 14 Acad. Sci. U.S.A. 86, 9986 (1989); U. B. Gyllenstein lection to play a role. worldwidepopulations (38). The genealogical et al., ibid. 87,1835 (1990); D. A. Lawloret al., Nature 26. J. N. Spuhler, in Genetics of Cellular,Individual, Fam- tree derivedfrom averagegenetic distances 335, 268 (1988); W. E. Mayeretal., EMBOJ. 7, 2765 ily, and Population Variability,C. F. Sing and C. L. (1988). Hanis, Eds. (Oxford Univ. Press, New York, 1993), separatesancestral African from derived non- 7. F. J. Ayala et al., Proc. Natl. Acad. Sci. U.S.A. 91, pp. 262-297. Africanpopulations. The deepestsplit in the 6787 (1994). 27. T. D. White, G. Suwa, B. Asfaw, Nature 371, 306 genealogyis dated at -156,000 years ago, 8. F. J. Ayala and A. A. Escalante, Mol. Phylogenet. (1994); ibid. 375, 88 (1995); G. WoldeGabriel et al., Evol., in press. ibid. 371, 330 (1994). which thus estimatesthe time when modern 9. B. Arden and J. Klein,Proc. Natl. Acad. Sci. U.S.A. 28. M. G. Leakey et al., ibid. 376, 565 (1995). humansspread from Africa throughoutthe 79, 2342 (1982); F. Figueroa et al., Nature 335, 265 29. G. A. Clark, in The Middle Paleolithic: Adaptation, world.These results are particularly notewor- (1988); T. J. McConnell et al., ibid. 332, 651 (1988). Behavior and Variability,H. Dibble and P. Mellars, 10. This rate is obtained by the "minimum''method de- Eds. (Univ. of Pennsylvania Museum, Philadelphia, thy becausethe analysissought to determine scribed in (8). See also Y. Satta et at., in Molecular 1992), pp. 183-205; Recherche (Paris) 263, 316 the historyof humanpopulations rather than Evolutionof the MajorHistocompatibility Complex, J. (March 1994); and J. M. Lindly, Am. An- the ancestryof individualgenes [see also Kleinand D. Klein,Eds. (Springer,Heidelberg, 1991), thropol. 9, 962 (1989). pp. 51-62. 30. C. C. Swisher Illet al., Science 263, 1118 (1994). (39)].Even so, thereseems to be no definitive 11. R. C. Griffiths,Theor. Popul. Biol. 17, 37 (1980); R. R. 31. L. Gabunia and A. Vekua, Nature 373, 509 (1995). reasonto excludethe possibilitythat different Hudson, Oxf. Surv. Evol. Biol. 7, 1 (1990). 32. E. Carbonell et al., Science 269, 826 (1995); J. M. genesmay have different populational origins. 12. J. F. C. Kingman,Stochast. Processes Appl. 13, 235 Pares and A. Perez-Gonzalez, ibid., p. 830. wouldthen reflectthe (1982); J. Appl. Probab. 19, 27 (1982); F. Tajima, 33. S. Jones, R. Martin,D. Pilbeam, Eds., The Cambridge The averagedistances Genetics 105, 437 (1983); N. Takahata and M. Nei, Encyclopedia of Human Evolution (Cambridge Univ. relativegenomic contribution of variousan- ibid. 110, 325 (1985); S. Tavare, Theor. Popul. Biol. Press, Cambridge, 1992), pp. 246-257. cestralpopulations. The resultswould thus be 26, 119 (1984). 34. C. B. Stringer, Sci. Am. 263, 98 (December 1990). compatiblewith a modelin which a modem 13. In the case of nuclear genes, the standard deviation 35. M. H. Wolpoff et al., Science 241, 772 (1988); A. G. for the mean is >2N [see M. Nei, Molecular Evolu- Thorne and M. H. Wolpoff, Sci. Am. 266, 76 (April Africanreplacement was concomitantwith tionary Genetics (Columbia Univ. Press, New York, 1992); D. M. Waddle, Nature 368, 452 (1994); G. someregional continuity (40). 1987), p. 395, eq. 13.74]. Brauer, in Continuityor Replacement? Controversies In conclusion,the weightof the molecu- 14. P. W. Hedrick, Am. Nat. 143, 945 (1994); A. L. in Homo sapiens Evolution, G. Brauer and F. H. Hughes and M. Nei, Nature 335, 167 (1988); ibid. Smith, Eds. (Balkema, Rotterdam, Netherlands, lar evidence favorsa recent Africanorigin 355, 402 (1992); J. Klein and C. O'hUigin, Philos. 1992), pp. 83-98. for modernhumans. Ethnic differentiation Trans.R. Soc. London Ser. B 346, 351 (1994); W. D. 36. The controversy about the origin of modern humans betweenmodern human populations would Potts and P. R. Slev, Immunol.Rev. 143,181 (1995); has at times been heated and prone to overstate- thereforebe evolutionarilyrecent, a resultof Y. Satta et al., Proc. Natl. Acad. Sci. U.S.A. 90, 7480 ment. Recently, a science journalistquoted with im- (1993). plicit approval a distinguished scientist saying that divergentevolution between geographically 15. A. V. S. Hillet al., Nature 352, 595 (1991); A. V. S. Hill "there is no genetic data that is consistent with the separatedpopulations during the last 50,000 et al., ibid. 355, 403 (1992); A. V. S. Hillet aL, ibid. multiregionalhypothesis' [A. Gibbons, Science 267, 360, 434 (1992); L. H. Miller,Proc. Natl. Acad. Sci. 1272 (1995); my italics]. Apparently, neither the sci- to 100,000years. However, the replacement U.S.A. 91, 2415 (1994). entist nor the journalistreflected that Mendel's laws, of archaicH. sapiensby anatomicallymod- 16. If u is the selected mutation rate per gene per gen- the double helix, and the immense majorityof genet- ern humansmay not have been complete eration and s is the selection coefficient, then ic data are consistent with the multiregionalhypoth- esis, as well as with the African replacement or any interbreedingbetween (__2 N-s[ iNs -1 -3/2 everywhere.Some other model of human origins. fs ( IL n J the colonizing modern humans and local 2Nu 881N2u2 37. A. R. Rogers and L. B. Jorde, Hum. Biol. 67, 1 populationswould account for the apparent [see (7, 8, 17, 18)]. (1995); C. B. Stringer, Philos. Trans. R. Soc. London morphologicalcontinuity in some regions, 17. N. Takahata, Proc. Natl. Acad. Sci. U.S.A. 87, 2419 Ser. B 337, 217 (1992); and P. Andrews, (1990); and M. Nei, Genetics 124, 967 Science 239, 1263 (1988). particularlyAustralasia (29, 35, 39). The (1 990). 38. D. B. Goldstein et al., Proc. Natl. Acad. Sci. U.S.A. currentblossoming of molecularevolution- 18. N. Takahata, Mol. Biol. Evol. 10, 2 (1993). 92, 6723 (1995). The polymorphisms are for auto- ary anthropologysurely will soon provide 19. Among the values in Table 3, more nearly appropri- somal microsatellite loci. answers. ate for DRB1 are T = 4.0 x 1 06 (which corresponds 39. L. L. Cavalli-Sforza, P. Menozzi, A. Piazza, The His- moredefinitive and precise to 60 My at 15 years per generation), u = 2.0 x 10 6 tory and Geography of Human Genes (Princeton or 4.5 x 10-6, and s = 0.01 or 0.02. The range Univ. Press, Princeton, NJ, 1994). REFERENCESAND NOTES corresponding to these values is N = 149,072 to 40. Regional continuity is supported by some molecular 297,789 (Table 3). evidence [W.-H. Li et al., in Genetics of Cellular,In- 1. The claimthat evolutionistshave shown that all modern 20. Estimates on the order of N = 100,000 for ancestral dividual,Family, and Population Variability,C. F. Sing humans descend from one or few women has repeat- human populations over the last 60 My are also ob- and C. L. Hanis, Eds. (OxfordUniv. Press, New York, edly been made in the media, as well as in popular tained by analyzingthe HLADQB1 polymorphisms(8). 1993), pp. 253-261; S. S. Deeb et al., Proc. Natl. scientificpublications [J. M. Lowenstein,Pac. Discovery 21. R. L. Cann et al., Nature 325, 31 (1987); M. Stone- Acad. Sci. U.S.A. 91, 7262 (1994); see (7)]. 39, 38 (1986)] and by scientists [W. M. Brown, Proc. king et al., Genetics 124, 717 (1990); L. Vigilantetal., 41. N. Saitou and M. Nei, Mol. Biol. Evol. 4, 406 (1987). NatI.Acad. Sci. U.S.A. 77, 3605 (1980)]. Science 253, 1503 (1991). 42. M. Kimura, The Neutral Theory of Molecular Evolu- 2. A good introductionto various methods used in the 22. C. Wills, Evolution49, 593 (1995). tion (Cambridge Univ. Press, Cambridge, 1983). molecular reconstruction of phylogeny is D. L. Swof- 23. S. Horaiet al., Proc. Natl. Acad. Sci. U.S.A. 92, 532 43. P. H. A. Sneath and R. A. Sokal, Numerical Taxon- ford and G. J. Olsen, in MolecularSystematics, D. M. (1995); M. Ruvolo et al., Mol. Biol. Evol. 10, 1 1 15 omy (Freeman, San Francisco, 1973). Hillisand C. Moritz, Eds. (Sinauer Associates, Sun- (1993). 44. I am grateful to A. Escalante for help with the com- derland, MA, 1990), pp. 411-501. 24. R. L. Dorit,H. Akashi, W. Gilbert,Science 268,1183 puter simulations and data analysis, and to him and 3. J. Klein, Natural History of the Major Histocompati- (1995). M. F. Hammer [Nature 378, 376 (1995)] has R. R. Hudson for helpful discussions. I am particu- bilityComplex (Wiley, New York, 1986); J. Kaufman, analyzed a polymorphic fragment of the Y chromo- larlyindebted to J. Klein,who stimulated my interest H. Vo5lk,H.-J. Wallny,Immunol. Rev. 143, 63 (1995); some in 16 men of diverse geographic origin, and in HLApolymorphisms; many ideas herein present- F.-W. Schwaiger and J. T. Epplen, ibid., p. 199. estimated the age of the ancestral sequence at ed derive from work done in collaboration with him, 4. P. J. Bjorkmanetal., Nature 329, 506 (1987); ibid., p. 188,000 years ago. L. S. Whitfield,J. E. Sulston, and although he should not be held responsible for what 512; J. H. Brown et al., ibid. 364, 33 (1993). P. N. Goodfellow (ibid.,p. 379), based on analysis of I now say. Supportedby NIHgrant GM 42397.

1936 SCIENCE * VOL. 270 * 22 DECEMBER1995