The evolution of modern human populations continues to be a topic of controversy. Evidence from mtDNA, Y-chromosome polymorphisms, autosomal markers, and fossil material supports both the expansion of early modern human populations in Africa and the partial or complete replacement of other hominid groups (Cann et al. ¹⁹⁸⁷; Vigilant et al. ¹⁹⁹¹; Stoneking et al. ¹⁹⁹⁷; Harpending et al. ¹⁹⁹⁸; Jorde et al. ¹⁹⁹⁸; Krings et al. ¹⁹⁹⁹; Relethford and Jorde ¹⁹⁹⁹; Seielstad et al. ¹⁹⁹⁹; Ovchinnikov et al. ²⁰⁰⁰). Interpretation of other, primarily fossil, data suggests both an early expansion of hominid lines and multiregional development of modern humans (Hawks et al. ²⁰⁰⁰; Wolpoff et al. ²⁰⁰⁰). The increasing number and variety of genetic markers offer additional opportunities for more-detailed analysis of human evolution and of genetic diversity within and between human populations.

Numerous studies utilizing a variety of polymorphic loci suggest an overall pattern of higher gene diversity in African populations compared with that in non-African populations (Merriwether et al. ¹⁹⁹¹; Vigilant et al. ¹⁹⁹¹; Bowcock et al. ¹⁹⁹⁴; Deka et al. ¹⁹⁹⁵; Jorde et al. ¹⁹⁹⁷; Stoneking et al. ¹⁹⁹⁷; Kaessmann et al. ¹⁹⁹⁹; Seielstad et al. ¹⁹⁹⁹; Forster et al. ²⁰⁰⁰; Jorde et al. ²⁰⁰⁰). These studies have focused primarily on neutral restriction-site polymorphisms (RSPs), short tandem-repeat polymorphisms (STRPs), noncoding autosomal sequences, Y chromosomes, and mtDNA. Analyses of protein-coding regions, including the neurofibromatosis type 1 (NF1), angiotensin-converting enzyme (ACE), myotonic dystrophy (DM), dopamine D2 receptor (DRD2) and fragile X (FMR1) loci, have also shown higher levels of diversity in African populations compared with levels found in non-African populations (Purandare et al. ¹⁹⁹⁶; Kidd et al. ¹⁹⁹⁸; Tishkoff et al. ¹⁹⁹⁸; Rieder et al. ¹⁹⁹⁹; Crawford et al. ²⁰⁰⁰). Some loci, including the melanocortin 1 receptor (MCR1) and phenylalanine hydroxylase (PAH) loci, do not consistently show patterns of higher diversity in African populations, revealing the potential influence of natural selection on patterns of genetic diversity (Rana et al. ¹⁹⁹⁹; Harding et al. ²⁰⁰⁰; Kidd et al. ²⁰⁰⁰). Noncoding DNA sequences on chromosomes 22, 15, and 1 show higher nucleotide-diversity estimates for African populations than for non-African populations, consistent with a recent human population expansion (Zhao et al. ²⁰⁰⁰; Yu et al. ²⁰⁰¹; L. B. Jorde, W. S. Watkins, M. J. Bamshad, D. Dunn, and R. B. Weiss, unpublished data).

Alu-insertion polymorphisms are ideal markers for human evolutionary studies because retroposition produces infrequent, irreversible, widely distributed insertion events, each with a known ancestral state (Batzer et al. ^{1994, 1996}; Stoneking et al. ¹⁹⁹⁷; Melton et al. ¹⁹⁹⁸; Novick et al. ¹⁹⁹⁸). These insertions are short interspersed repetitive elements (SINEs), and they account for 10% of the human genome (Deininger and Batzer ¹⁹⁹³; Smit ¹⁹⁹⁶; Dunham et al. ¹⁹⁹⁹). Each full-length Alu element is a dimeric, ~300-bp retroposon that is homologous to the 7SL RNA component of the signal-recognition particle. The Y, Ya, and Yb subfamilies of Alu elements are still active and produce new Alu insertions that are polymorphic in human populations (Arcot et al. ^1995a; Batzer et al. ^{1995, 1996}). Since the Y, Ya, and Yb subfamilies did not become active until after the human lineage diverged from the last common ancestor with the nonhuman apes, insertion events of these Alus are restricted to humans, and the ancestral allele is the absence of an insertion. Assignment of the ancestral allele for polymorphic Alu loci does not require comparisons with other species such as chimpanzees and provides a unique opportunity for the rooting of phylogenetic networks. At each polymorphic locus, the chromosome containing the Alu insert is the derived allele, and different individuals with an Alu insert at a given locus share a chromosomal region that is identical by descent. Previous studies using a small number of polymorphic Alu insertions have suggested an African origin for modern humans and have argued for an earlier, more extensive expansion of modern humans in tropical regions (Batzer et al. ^{1994, 1996}; Stoneking et al. ¹⁹⁹⁷).

In the present study, we examine 35 Alu loci and 30 gene-related RSPs in 31 world populations, to characterize diversity and genetic structure in modern human populations. The ancestral state for each Alu and RSP locus has been assigned. This analysis of the largest set of Alu-insertion polymorphisms examined to date provides new data on the distribution of ancestral and derived alleles in human populations and is consistent with an African origin of anatomically modern humans. Genetic-distance estimates based on Alu-insertion polymorphisms, RSPs, STRPs, and mtDNA, using the same individuals in the same populations, are highly correlated. These data build upon a growing body of evidence characterizing human genetic diversity and population structure.

For Alu loci beginning with the prefix NBC, screening of nonredundant and high-throughput genomic sequence (HTGS) databases was performed through use of the Basic Local Alignment Search Tool (BLAST [^GenBank]). The database was searched for exact complements to the oligonucleotide 5′-CCATCCCGGCTAAAACGGTG-3′, which is an exact match of that portion of the Alu Ya5-subfamily consensus sequence that contains unique diagnostic mutations. Sequences that were exact complements of the oligonucleotide were then subjected to more-detailed analysis. A region of 1,000 bases, which was directly adjacent to the sequences identified from the databases and matched the initial ^GenBank BLAST query, was analyzed through use of either the REPEATMASKER2 or the CENSOR program. PCR primers were designed from unique flanking DNA sequences adjacent to individual Alu elements, through use of PRIMER3 software (Whitehead Institute for Biomedical Research). The PCR primers were screened, against the ^GenBank nonredundant, Alu, and HTGS databases, for the presence of repetitive elements or duplications, through use of BLAST. The sequences of the oligonucleotide primers and PCR conditions are available from the Eccles Institute of Human Genetics Web site. The chromosomal location of Alu repeats identified from clones that had not been mapped previously was determined by PCR amplification of National Institute of General Medical Sciences human/rodent somatic cell hybrid mapping panel 2 (Coriell Institute for Medical Research).

Polymorphic Alu loci were genotyped by amplification of 25 ng of genomic DNA in a standard 30-cycle, three-step PCR. Appropriate annealing temperatures and additives were optimized for each system. For most systems, samples were amplified with 5 μl of cresol red loading buffer (34% sucrose, 0.02% cresol red) to eliminate the need to add dye to the samples before gel loading. After PCR, the samples were loaded onto multiple-combed 3% Nusieve agarose (3:1) gels and were electrophoresed at 175 V for 2 h. Ethidium bromide–stained gels were visualized by UV and were documented. The methodology and genotypes for the STRP, RSP, and hypervariable segment 1 (HVS1) data have been described elsewhere (Jorde et al. ^{1995, 1997}).

The ancestral allele of each RSP polymorphism was determined by comparison with that of a common chimpanzee and that of a pygmy chimpanzee. Human and chimpanzee DNA was amplified with primers for each of the 30 RSP loci. The locations, primer sequences, and PCR conditions for each RSP locus are provided by The Genome Database and are listed at the Eccles Institute of Human Genetics Web site. PCR products were digested with 4 units of the appropriate restriction endonuclease in PCR buffer for 2 h and then were visualized by agarose-gel electrophoresis and ethidium bromide staining. The ancestral status was assigned to the allele that was shared by the chimpanzees and humans. No polymorphic variation was seen in the chimpanzees. Of the 30 RSPs used here 11 are documented single-nucleotide polymorphisms (SNPs), whereas 19 have not been characterized at the sequence level.

Of the 35 polymorphic Alu insertions, 33 are present in all continental groups. Ya5NBC54 and Ya5NBC135 insertions were not found in African populations. No Alu insertion was fixed in African populations, but two, five, and six insertions have achieved fixation in Indian, Asian, and European populations, respectively. Most Alu loci are in HWE. For the four major population groups, five, four, five, and nine loci produced significant values (P<.05) in African, Asian, European, and Indian populations, respectively, but these values fell to two, four, two, and three when a Bonferroni correction was applied. The loci that deviated from HWE differed from population to population. Ten Alu polymorphisms were examined for Mendelian segregation in two three-generation CEPH families. Normal codominant segregation patterns were observed (data not shown).

Figure 1a shows the Alu allele-frequency profile for African and non-African populations. This plot displays the frequencies of the Alu inserts at each locus, sorted according to their frequency in the African population. Two patterns are discernible: (1) African populations show an overall trend toward lower Alu-insertion frequencies, and (2) loci that have a low (<.3) frequency in non-Africans also have low frequencies in Africans, whereas loci with higher frequencies in non-Africans tend to have lower frequencies in Africans. These patterns were also observed when the continental groups were divided into subpopulations. African subpopulations showed trends toward lower Alu-insertion frequencies, whereas the highest insertion frequencies were found in southeastern Asian populations. The mean Alu-insertion frequencies for the major population groups are as follows: Africans, .4226; Asians, .5690; Europeans, .5604; and Indians, .5453. Allele-frequency distributions in the four major population groups are significantly different between Africans and Asians (P<.001, by a Wilcoxon signed-rank test), Africans and Europeans (P<.003), and Africans and Indians (P<.002) but are not significant different between non-African groups.

Figure 1 Allele frequencies of the derived alleles at 35 polymorphic Alu (a) and 30 RSP (b) loci, in African and non-African populations, sorted by the insertion frequency in the African population. Note the trend toward higher Alu-insertion frequencies in populations (more ...)

We also evaluated the frequency of Alu insertions in a data set, published elsewhere (Stoneking et al. ¹⁹⁹⁷), consisting of eight Alu loci in 34 independently ascertained world populations. Average Alu-insertion frequencies over all loci in this data set show a similar trend toward an increasing frequency of Alu insertions in populations outside Africa (Africa, .374; western Asia, .435; Europe, .464; southeastern Asia, .529; and Americas, .543). Dividing the polymorphic Alu insertions into groups based on the method of ascertainment (directed HeLa library screening [Arcot et al. ^{1995b, 1998}], random library screening, or in silico database mining) showed lower Alu-insertion frequencies in Africans, for all ascertainment methods.

To compare ancestral-allele and derived-allele patterns based on Alu elements versus those seen for other polymorphic markers, the derived-allele–frequency profiles for 30 randomly chosen, gene-related, unlinked RSPs were examined in a subset of 274 individuals (Jorde et al. ¹⁹⁹⁵). The ancestral allele for each of these previously published loci was determined by typing the system in chimpanzees, and new analyses were performed with the “rooted” RSP data. The allele-frequency profiles for the 30 RSPs in African, Asian, and European populations show trends toward higher frequencies of derived alleles in Asia and Europe and lower frequencies of derived alleles in Africa (fig. 1b). Thus, through use of either widely distributed Alu markers or RSP markers from the same individuals, an increase in the overall frequencies of derived alleles can be observed in continental populations outside Africa.

Alu-heterozygosity and -homozygosity estimates for the continental groups and their subpopulations are consistent with the Alu-frequency distributions (table 1a). African populations have higher overall Alu gene diversity (heterozygosity) than do other populations. This finding is concordant with results from other marker systems, including the mitochondrial genome, autosomal STRPs, and the Y chromosome (Deka et al. ¹⁹⁹⁵; Jorde et al. ²⁰⁰⁰). The observed homozygosity for the Alu insert (the derived state) is significantly higher (P<.05) in all non-African groups than it is in sub-Saharan Africans. All subpopulations outside Africa have higher observed levels of Alu homozygosity, as averaged over the 35 Alu loci, than do all populations within Africa. Comparisons of the major population groups and their subpopulations, through use of rooted RSP loci, show a similar trend, outside Africa, toward increasing frequency and homozygosity of derived alleles (table 1b). Despite the high levels of heterozygosity in Europe that are caused by ascertainment bias (Bowcock et al. ¹⁹⁹¹; Rogers and Jorde ¹⁹⁹⁶), observed levels of RSP homozygosity of derived alleles, over all loci, is lowest in Africa and its subpopulations, higher in Europe, and highest in populations of southeastern Asia.

Table 1 Population Statistics for 35 Alu-Insertion Polymorphisms and 30 “Rooted” RSPs

Using hierarchical F statistics, we examined the effect of population subdivision of individuals with respect to the total population (F_IT) and of subdivision of the subpopulations with respect to the total population (F_ST), at the worldwide and continental levels, for the 35 Alu loci (table 2). Dividing the total population into the four major population groups and into 31 subpopulations yields statistically significant F_ST estimates of 12.2% and 11.2%, respectively. Omitting Africans from the continental comparison reduces the F_ST to 4.8%. The F_ST for populations within Africa is 5.7%, more than twofold greater than those for populations within Asia (1.8%), Europe (2.0%), and India (2.4%). The F_ST for the five populations within Asia is not significantly different from 0. For most subpopulations with larger sample sizes, the fixation index of the individual with respect to its own subdivision (F_IS) is relatively low, for both Alu and RSP markers (shown in table 1).

Table 2 Hierarchical F Statistics for 35 Alu Loci

A neighbor-joining tree of Alu genetic distances illustrates the relationships among the 31 world populations (fig. 2a). A hypothetical population with an Alu allele frequency of 0 for each locus is included as an “ancestral” population (Batzer et al. ¹⁹⁹⁴). The ancestral population is located closest to the Mbuti and Zaire Pygmy populations. African populations are closer to the ancestral population than are non-African populations. The ancestral population and all African populations are separated from all non-African populations, with very high bootstrap support. The populations of Asia and Europe cluster by continent, and Indian populations are dispersed among non-African groups. Tribal Indians of close geographic proximity to one another (Maria Gond, Khonda Dora, and Santal) cluster together. The network shows greater separation among the 10 African populations than among Asian, European, and Indian populations.

Figure 2 Neighbor-joining networks of genetic distances, based on 35 polymorphic Alu loci (a) and 30 RSP loci (b). African Mbuti Pygmy populations are nearest to the ancestral outgroup. African populations are clustered together, with long branches. All African (more ...)

A neighbor-joining network based on the 30 RSPs shows similar relationships among a subset of the populations shown in the Alu network (fig. 2b). A derived–RSP-allele frequency (0) at each locus was used to create the ancestral population. The ancestral population for the RSP network falls very close to the African cluster. African populations branch off first, and bootstrap support for the non-African cluster is very high. Populations from Asia and Europe cluster into their respective continents. The populations of southeastern Asia have the greatest distance from the ancestral population, which reflects a higher overall frequency of derived RSP alleles in Vietnamese, Cambodian, and aboriginal Malay populations.

To compare genetic distances based on Alu loci versus those based on other marker systems, we created four matrices of genetic distances for 15 populations from Africa, Asia, and Europe, based on 35 Alu markers, 30 STRPs, 30 RSPs, and the mtDNA control-region HVS1. This subset of 15 populations was selected for the analysis because it contains data for all populations, loci, and marker types. Each matrix represents the between-population genetic distances derived from the same 245 individuals from 15 populations, for a given genetic system. Correlation coefficients between Alu distances and distances based on other marker systems are high and significant (P<.0001), with the highest correlations seen between the Alu and RSP systems (table 3).

Table 3 Correlations between Genetic Distance Matrices[Note]

Both the higher frequency of ancestral Alu loci and the higher genetic diversity in Africans are consistent with (1) the emergence of modern Homo sapiens within Africa, (2) the exodus and rapid expansion of a limited subset of humans into Eurasia, and (3) limited migration back into Africa. Higher African diversity has been observed for many neutral marker systems, including those for autosomal RSPs and STRPs (Bowcock et al. ^{1991, 1994}; Deka et al. ¹⁹⁹⁵; Jorde et al. ^{1995, 1997}), mtDNA (Merriwether et al. ¹⁹⁹¹; Vigilant et al. ¹⁹⁹¹; Penny et al. ¹⁹⁹⁵), and the Y chromosome (Seielstad et al. ¹⁹⁹⁹; Forster et al. ²⁰⁰⁰). The diversity patterns at several genes, including ACE (Rieder et al. ¹⁹⁹⁹), DM (Tishkoff et al. ¹⁹⁹⁸), and the plasma alpha (1,3) fucosyltransferase gene (FUT6) (Pang et al. ¹⁹⁹⁹), among others, also show high diversity in African populations. In addition to high African diversity, our analysis shows a directional increase in the overall frequency and homozygosity of Alu and derived RSP alleles. This finding is consistent with a rapid expansion and/or a bottleneck in early populations leaving Africa. A continued expansion of populations into Asia and southeastern Asia may account for the higher frequency of derived alleles in these populations. Additional work, using tightly linked markers flanking each Alu insert, will provide relative ages of the Alu insertions, which will allow better resolution of the potential mechanisms generating these patterns.

The F_ST estimates based on the 35 Alu-insertion polymorphisms for the four major population groups are also consistent with an African origin and rapid expansion. The continental-level Alu F_ST of 12.2% is similar to previous estimates made on the basis of autosomal RSPs and STRPs (Bowcock et al. ¹⁹⁹¹; Deka et al. ¹⁹⁹⁵; Barbujani et al. ¹⁹⁹⁷; Jorde et al. ²⁰⁰⁰), but removal of the African populations reduces the F_ST to 4.8%. Thus, substantially more Alu variation occurs between African and non-African continental populations than occurs between all non-African populations. Low F_ST estimates for European and Asian populations are consistent with rapid and recent expansion with relatively little time for population differentiation.

The Alu-insertion polymorphism and the RSP consensus neighbor-joining networks place the ancestral population among and near Africans, respectively. This result is consistent with the allele frequencies for both Alu and RSP markers and reflects a higher frequency of ancestral alleles in African populations. Recent studies of SNPs indicate that the level of sequence-verified polymorphism sharing between chimpanzees or gorillas and humans is low (<1%) (Hacia et al. ¹⁹⁹⁹), which suggests that rooted trees based on RSP/SNP are subject to only limited error due to convergent mutations. Using data sets for RFLPs, microsatellite data, protein polymorphisms, and a limited number of Alu loci, Nei and Takezaki (¹⁹⁹⁶) consistently place a root, for human phylogenetic trees, between African and non-African populations. Despite initial methodological concerns, a substantial body of evidence indicates a root, in African populations, for the maternally inherited mtDNA tree (Cann et al. ¹⁹⁸⁷; Vigilant et al. ¹⁹⁹¹; Stoneking and Soodyall ¹⁹⁹⁶; Ingman et al. ²⁰⁰⁰). Additionally, a human mtDNA “molecular fossil” inserted into chromosome 11 has been used as an independent outgroup for the rooting of mtDNA trees (Zischler et al. ¹⁹⁹⁵). The topology of the tree, although statistically weak, suggests an African origin. Like Alu-insertion frequencies, the insertion frequency of the mtDNA “fossil” is low in Africans and higher in non-Africans. Recent studies using independent sets of STRP and biallelic markers on the Y chromosome suggest that the earliest branches for male founder(s) of our species can be traced to ancestors of the African San population (Forster et al. ²⁰⁰⁰). In the Alu and RSP networks, the San individuals examined in this analysis fall near the ancestral populations.

The eight Alu loci reported by Stoneking et. al. (¹⁹⁹⁷) show a pattern of increasing Alu-insertion frequencies in populations outside Africa, with the highest insertion frequencies being found in American Indians. An exception to this trend was a low frequency of inserts in the aboriginal populations of Papua New Guinea and Australia. Using a principal-components analysis, those authors suggest an early tropical expansion of modern humans, because these populations’ distance to the root is similar to that of Africans. The tribal populations of the Indian subcontinent may also represent remnants of an early Paleolithic expansion out of Africa (Cavalli-Sforza et al. ¹⁹⁹⁴). However, the Irula, Khonda Dora, Maria Gond, and Santal tribes of (sub)tropical India that are examined here do not show an excess of ancestral alleles. The analysis of more-isolated “Negrito” populations of the tropics, including the Kadar (India), Jarawa (Andaman Islands), and Onge (Andaman Islands), may reveal signals of early hominid expansion in tropical regions between Africa and Australia.

Patterns of higher Alu frequencies in non-African populations could be produced solely by in silico ascertainment of Alu markers from mostly non-African DNA sequences (Europeans or Asians). Several results argue against this possibility. First, Alu markers either ascertained in African American–derived genomic HeLa libraries or ascertained by other screening methods produce frequency patterns similar to those ascertained in silico. Second, gene-diversity estimates for the Alu loci are highest in Africa, as is the case with unbiased mtDNA and STRP markers (Jorde et al. ²⁰⁰⁰). Third, the highest reported Alu frequencies occur in populations from the Americas rather than in populations from Europe or Asia. Fourth, the large continental sample sizes used here diminish the effects of potential ascertainment bias. If ascertainment bias for Alu loci exists, the bias is a function of 1/n, where n is the haploid sample size, under the assumption of rapid population growth (A.R.R., unpublished data). Thus, it is unlikely that the results seen here are simply due to marker-ascertainment methods.

A more likely explanation for the observed pattern stems from the fact that allele frequencies are driven toward fixation (frequencies of either 0 or 1) in populations that have undergone bottlenecks. If a bottleneck accompanied the exodus of human populations out of Africa, then non-African populations should exhibit a more strongly U-shaped allele-frequency distribution than should African populations (Sherry et al. ¹⁹⁹⁷). Because Alu polymorphisms are ascertained on the basis of a limited number of individuals, low-frequency polymorphisms are more likely to be missed and high-frequency inserts are likely to be detected. Thus, the observed pattern of insertion frequencies is influenced strongly by the right half of the allele-frequency distribution, in which non-African populations should exhibit an excess of insertion frequencies that are closer to 1. Indeed, this pattern is evident in figure 1a. The higher Alu-insertion frequencies in European and Asian populations are consistent with a pronounced bottleneck having occurred in these populations.

The similarity between the Alu, RSP, STRP, and mtDNA genetic-distance patterns is notable, especially in light of the differences in mutation rates and in mechanisms producing each type of variation. The Alu-insertion polymorphism and RSP genetic distances analyzed in the present study are based on the frequencies of loci with known polarity, and both show a gradient of derived alleles that increases from Africa into Europe, eastern Asia, and southeastern Asia. These results suggest a distribution pattern for ancestral alleles and for genetic diversity, in world populations. Further characterization of many loci should provide greater insight into genetic variation and allele-distribution patterns in the human species.

The authors thank J. Hawks and B. V. R. Prasad for their helpful comments and suggestions for the manuscript. We are appreciative for the DNA samples provided through collaborations with P. Fischer, T. Jenkins, K. Kidd, J. Kidd, J. Kere, J. M. Naidu, and B. B. Rao. This research was supported by National Institutes of Health grants GM-59290 and RR-00064 and National Science Foundation grants SBR-9514733, SBR-9512178, and SBR 9818215.