The report of the Neanderthal genome sequence included the fascinating conclusion of a small amount of Neanderthal admixture in anatomically modern humans (AMH) (Green et al. 2010). Consistent with the fossil and archaeological records that show Neanderthals and AMH lived about the same time in parts of Europe and Asia, the signal of introgression was observed in genomes from non-African and North African populations (Green et al. 2010; Sanchez-Quinto et al. 2012) but was not apparent in sub-Saharan African populations (Hammer et al. 2011; Lachance et al. 2012; Wall et al. 2013). Given the signal of introgression occurs in diverse populations outside of sub-Saharan Africa, it may have been caused by gene exchange soon after the Out-of-Africa (OOA) migration of modern humans, perhaps in the Middle East region (Green et al. 2010; Stewart and Stringer 2012) in the range of 47–65 thousand year ago (kya) during the Late Pleistocene period (Tishkoff et al. 2007; Sankararaman et al. 2012). This model of relatively recent introgression outside of Africa is depicted in figure 1A (model i). Fig. 1.—

(A) Two models of Neanderthal gene flow into AMH. Model i: A single episode of admixture occurred between AMH and Neanderthal after OOA but before AMH subsequently spread throughout Eurasia. Model ii: Neanderthal admixture happened before OOA. A subset of ancient AMH African populations who were admixed with Neanderthal subsequently moved out of Africa. The red arrows represent introgression of Neanderthals with modern humans. Time period is not on scale. Figure modified from Campbell and Tishkoff (2010). (B) Eight sub-Saharan African populations sampled in this study. The populations are marked in black filled squares, along with their names and sample locations. ASW individuals have recent West African ancestry (denoted with a circle, as this ancestry does not trace back to a single location [Bryc et al. 2010]).

Our whole-genome single-nucleotide polymorphism (SNP) data set contains a total of 63 individuals, including 38 individuals who belong to eight sub-Saharan populations (fig. 1B and table 1) and 25 individuals who belong to five non-Africans populations (table 1). Four out of the eight sub-Saharan African populations have a hunter-gatherer life style (San, Hadza, Sandawe, and Pygmy). The other four populations include Yoruba in Ibadan Nigeria (YRI), Luhya in Webuye Kenya (LWK), Maasai in Kinyawa Kenya (MKK), and African Americans with African ancestry in the Southwest USA (ASW). The overall data set contains genotypes of 4.6 million (M) high-quality nongenic SNPs in 63 modern humans (see Materials and Methods for SNP selection detail). This data set was then combined with the Neanderthal genome from Green et al. (2010) to obtain loci where the Neanderthal sequence is available. A final data set of ∼2.9 M SNPs was used in subsequent introgression and ancestry analysis. Table 1

Populations Examined in This Study

We first estimated whether the Neanderthal genome shared more bases with the sub-Saharan African genomes or with the non-African genomes. This is achieved by calculating the pairwise D statistic, sometimes called the ABBA/BABA statistic, that was first introduced by Green et al. (2010) as a formal test for Neanderthal ancestry in modern human (see Materials and Methods). The theory and application of the D statistic was extended for SNP data in which the values were computed using either population or sample allele frequencies (Durand et al. 2011; Patterson et al. 2012), and the D statistic always considers a group of four populations. The pairwise D statistics are calculated by assigning H₁ to be one of the eight sub-Saharan African populations and H₂ to be one of the five non-African populations, as compared with Neanderthals. If H₁ and H₂ share the same amount of Neanderthal genomes, the D value is expected to be zero (Green et al. 2010; Durand et al. 2011; Patterson et al. 2012). A negative D statistic implies that H₂ is genetically closer to Neanderthals than H₁, whereas a positive D statistic implies H₁ is closer. The D statistics of the pairwise comparison between the sub-Saharan African population (H₁) and the non-African population (H₂) is summarized in figure 2A. A Bonferroni-corrected P value for 40 hypothesis tests is 0.0013 (0.05/40), corresponding to a standard normal Z score of ∼3 standard deviation (SD). All comparisons of sub-Saharan Africans (H₁) and non-Africans (H₂) give significant negative D statistics (D from −4.7% to −6.6%, |Z| ≫ 3 SD, for details see supplementary table S1, Supplementary Material online), indicating that Neanderthal genomes are closer to non-Africans than to sub-Saharan Africans, consistent with results from Green et al. (2010). Fig. 2.—

Neanderthal admixture with modern human populations. (A) D statistic for pairwise comparisons of non-Africans with sub-Saharan Africans. Each bar represents a D statistic which is calculated for an H₁ (sub-Saharan African) population and an H₂ (non-African) population. All D statistic values are significant. (B) D statistics among sub-Saharan African populations. Asterisks indicate significance (|Z| > 2.9 SD corrected for 28 hypotheses).

Under model ii of ancient AMH substructure and within-Africa Neanderthal-like gene flow, the Neanderthal similarity could vary among sub-Saharan Africans (fig. 1A, model ii). Therefore, we compared the relative similarities of the genomes from eight sub-Saharan African populations to the Neanderthal genome (see Materials and Methods). As shown in figure 2B, although none of the D values approached the range found between African and non-African genomes, we did observe statistically significant D values for several sub-Saharan African pairs (|Z| > 2.9 SD, corrected for 28 hypothesis tests at P = 0.05). For the test of D(San, H₂; Neanderthal, Chimpanzee), where H₁ is San and H₂ is any other sub-Saharan African populations (fig. 2B), the D statistics are always negative (D from −0.69% to −1.68%), suggesting that Neanderthal genome shares more alleles with these populations than with San (for details see supplementary table S2, Supplementary Material online). Specifically, the comparison of San with Sandawe, MKK, and ASW gives statistically significant D (|Z| > 2.9 SD, for details see supplementary table S2, Supplementary Material online, significant Z scores were highlighted in red). These observations are consistent with the results between pairs of African and non-African populations, where the D values were the lowest for comparisons involving Sandawe, MKK, and ASW (fig. 2A). Taken together, these results suggest that Sandawe, MKK, and ASW share higher number of Neanderthal alleles than other African populations. In addition to the comparisons between San and Sandawe, MKK, or ASW, ASW is more similar to Neanderthal (|Z| > 2.9 SD) in comparisons with Pygmy, LWK, and YRI (fig. 2B). The variation in D statistics among sub-Saharan African comparisons is consistent with the model ii of ancient AMH substructure and within-Africa Neanderthal-like gene flow, as well as some recent findings (Hammer et al. 2011; Wall et al. 2013).

The differences in D statistics among sub-Saharan African comparisons are considerably lower than those found in comparisons involving pairs of sub-Saharan African and non-African genomes. Three scenarios could explain this difference. First, if model ii is true, a lower D value between African populations is expected if there has been gene exchange among African populations since the time of introgression from Neanderthal-like genomes and if this African gene exchange did not affect the population that gave rise to the OOA migration. This would leave the non-African populations showing higher contrasts with African populations (i.e., more extreme D values) and modest D values among some pairs of African populations. Second, if the model ii is true, then a low value of D between sub-Saharan African populations may be expected because the timing of Neanderthal-like introgression would have been longer ago, when the Neanderthal ancestors were more closely related to the sub-Saharan AMH populations present at that time. However, this explanation would not explain why Green et al. (2010) observed larger values of D between non-African and sub-Saharan African populations. Third and finally, the model ii may be false, and the cause of low but varying D values between African populations is the result of varying amounts of recent introgression into sub-Saharan African populations from non-African populations. Under this view, the original model of Neanderthal introgression into non-African genomes (fig. 1A, model i) is correct, and varying amounts of recent introgression into Africa would have introduced these Neanderthal alleles into some African populations more than others.

To examine this last scenario, we determined the relationship between the D statistics and the mixed ancestry of sub-Saharan Africans. We used the program ADMIXTURE to determine genetic ancestry of all 63 individuals in this study. ADMIXTURE assumes that each individual is drawn from an admixed population with contributions from a set number of ancestral populations (Alexander et al. 2009). Because we are estimating the amount of European or Asian ancestry in sub-Saharan individuals, the program was run with k = 3 populations, corresponding to the three major continents, Africa, Europe, and Asia, where individuals in our data set were collected. On average, ASW, Sandawe, and MKK individuals show more than 16% non-African ancestry in the analysis (fig. 3). Overall, the estimated amount of non-African admixture in African populations is highly correlated with their D statistics (r = 0.93, P = 0.002, table 2). This high correlation supports the explanation that recent admixture with non-Africans had introduced Neanderthal alleles into some African populations more than others. It should be noted that although ADMIXTURE predicts the contribution of three putative ancestral populations in each individual genome, the contribution could be indirect. For example, as noted in Tishkoff et al. (2009), MKK and Sandawe might not have large amounts of Eurasian ancestry. They have large amounts of Cushitic ancestry which originated in Ethiopia, and those populations are more genetically similar to the non-Africans and may have experienced gene flow (Tishkoff et al. 2009). So the non-African ancestry predicted in these populations could be through indirect contribution from their ancestral populations. Fig. 3.—

Individual ancestry inferred by the ADMIXTURE program. Results from k = 3 are shown. Each individual’s genome is represented by a vertical bar composed of colored sections, where each section represents the proportion of an individual’s ancestry derived from one of the three ancestral populations. Individuals are arrayed horizontally and grouped by population as indicated.

In summary, although we observed varying D values in an extensive comparison among sub-Saharan African genomes, our results are consistent overall with the original report of Neanderthal admixture into non-African modern humans (Green et al. 2010) and several recent studies (Sankararaman et al. 2012; Yang et al. 2012). Our data do not provide support for the alternative model (model ii) of Neanderthal-like admixture with some African populations prior to the OOA migration (Eriksson and Manica 2012).

It is worth noting that our test does not have a lot of power to address the related and more general question of how modern humans evolved from archaic humans in Africa and about the role of population structure in that process. In this light it is useful to recall that for neutral variants, the expected age of a variant is correlated with the effective population size and the allele frequency (Kimura and Ohta 1973). Assuming an effective population size of 10,000 and a generation time of 20 years, the average age of a SNP with an allele frequency of 20% is approximately 322 thousand years, far older than the origin of modern humans. Therefore, we all harbor a large amount of variation that arose before the origin of modern humans, and a substantial fraction of this arose before the origin of Neanderthals. The comparison of modern human and Neanderthal genomes offers an exceptional but limited glimpse into the history of modern human population structure. With the advancement in technology and analytical methods in the genomic field, we will soon have more data (e.g., sequence data from more, older archaic hominid) and better methods to understand the role of population structure in human evolution.

Supplementary tables S1 and S2 are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/).