Genetic variation and population structure of Sudanese populations as indicated by 15 Identifiler sequence-tagged repeat (STR) loci

Informativeness for assignment

Rosenberg et al.[29] developed a statistic, the informativeness for assignment (Ι_n), which describes the information content of a particular genetic marker for ancestry inference (it ranges from zero (no information) to the natural logarithm of the number of populations (maximum information)). We computed Ι_n values for the 15 Identifiler STRs (all tetranucleotide microsatellites) from the present study. The mean, across markers, (and standard deviation) Ι_n was 0.167 (0.070) for the Sudanese populations, and 0.154 (0.066) for the Sudanese populations and the populations from Egypt, Somalia and Uganda (the Karamoja). To determine the relative informativeness of the 15 Identifier microsatellites, we computed I_n for 377 microsatellites (of which 274 were tetranucleotide microsatellites) for two groups of African populations from Rosenberg et al.[29]; six sub-Saharan African populations (Kenyan Bantu speakers, Mandenka, Yoruba, San, Mbuti Pygmy, Biaka Pygmy) from the Human Genome Diversity Panel (HGDP); the 'HGDP sub-Saharan group'; and a subgroup of three populations from the HGDP, who all speak Niger-Congo languages (Kenyan Bantu speakers, Mandenka, Yoruba), termed the 'HGDP Niger-Congo group'. Based on the 377 microsatellites from Rosenberg et al.[29], the mean I_n (across markers) was 0.234 (0.098) for the HGDP sub-Saharan group and 0.106 (0.054) for the HGDP Niger-Congo group, which was a significant difference (P < 0.001, Wilcoxon signed-rank test). The I_n statistic depends on both marker information about ancestry and the level of differentiation between the investigated populations. Because the same markers were used for both the HGDP sub-Saharan group and the HGDP Niger-Congo group, the difference in I_n between the two groups was due to the greater level of differentiation between populations in the HGDP sub-Saharan group. The I_n values for the Sudanese populations and the larger group of northeast African populations (from Sudan, Egypt, Somalia and Uganda) were found to lie between the values for the HGDP Niger-Congo group and the HGDP sub-Saharan group. This result is not surprising, considering that the sampled populations from Sudan belonged to different linguistic groups, although the groups are not as differentiated as the Pygmy, the San and the Niger-Congo-speaking populations in the HGDP sub-Saharan group. We concluded that the CODIS STRs contain information on population structure in the same range as many other microsatellites. However, it is possible to select more informative STR loci for population structure inference; for example, the top 15 most informative loci of the 377 microsatellites [29] (for the HGDP sub-Saharan group and for the HGDP Niger-Congo group) had mean I_n values of 0.489 (0.062) and 0.249 (0.041) respectively. This observation can be explained by the fact that the Identifiler STRs have been selected to have high levels of variation, because of the potentially high mutation rates of these STRs, in order to be powerful tools for identifying individuals, but the high level of variation also makes it challenging to separate alleles that are identical by state from those identical by descent [15].

Population diversity and sub-structure in Sudanese and neighboring populations

The greatest number of distinct alleles [30], considering a sample size of 58 chromosomes (29 individuals) from each of the Sudanese populations, the Somali population, the Egyptian population- and the Karamoja population from Uganda, was found in the Karamoja (mean ± SE 8.37 ± 0.63) followed by the Zagawa (8.27 ± 0.74), the Nuba (8.04 ± 0.67) and the Nilotic (8.01 ± 0.67) (Figure 2A) populations. The mean number of private alleles was greatest in the Somali (0.400 ± 0.171), followed by the Nilotic (0.292 ± 0.093), the Zagawa (0.273 ± 0.122) and the Karamoja (0.250 ± 0.066) groups (Figure 2B). This observation of greater levels of diversity for the Nilo-Saharan populations from southern and western Sudan and for the Karamoja population from Uganda indicates larger effective population sizes of Nilo-Saharan populations compared with Afro-Asiatic populations. The low number of private alleles found for the Beja (0.031 ± 0.012) and the Nubian (0.129 ± 0.040) (Figure 2B) groups could be the result of high migration rate and greater gene flow from Eurasian populations, because these populations traditionally occupy the entry ports to Sudan, that is, the Beja at Red Sea and the Nubian at the Nile River Valley. A similar result was found previously [4], based on Y-chromosome data. The low number of private alleles in the Coptic groups (0.075 ± 0.053) may be a result of the recent migration of this population from Egypt, where they may have been influenced by gene flow from Asia and Europe. The Nuba also contained few private alleles (0.123 ± 0.065), which may be a consequence of the population being an amalgamation of many groups with high levels of diversity (Hassan et al., unpublished data).

For pairs of Sudanese populations, the greatest mean number of alleles that are private to pairs of populations was found for the Zagawa-Nuba pair (0.131 ± 0.064), followed by the Zagawa-Nilotic pair (0.080 ± 0.051) and the Zagawa-Copt pair (0.076 ± 0.063; Figure 3). The lowest number of private alleles for pairs of populations were typically found when we compared a western group (for example, the Zagawa) with a northern (for example, the Nubian) or a central (for example, the Arab) group. When the Sudanese populations were compared with their neighboring populations (sample size of 58 chromosomes), three of the four highest numbers of private alleles for population pairs were seen between the Karamoja population (from Uganda) and either the Zagawa (0.055 ± 0.026), the Nilotic (0.034 ± 0.012) or the Nubian (0.029 ± 0.021) populations (Figure 4), indicating gene flow and/or shared ancestry between the Karamoja population and Nilo-Saharan populations. For smaller sample sizes (10-44 chromosomes), the Zagawa-Nuba pair also has a high number of private alleles. A similar result was found in a previous Y-chromosome study [6], in which all Nilo-Saharan populations (which included the Zagawa, the Nilotic and the Nubian) had little evidence of gene flow with other Sudanese populations. The second largest value for the number of private alleles for population pairs in our study was for the Arab-Somali pair (0.036 ± 0.029; Figure 4), which may be a result of the influence of Arab groups in east Africa as the product of continuous migrations from the Arabian Peninsula across the Gate of Tears over the past three millennia [31]. Among pairs of populations that included the Egyptian population, the Egyptian-Copt pair had the greatest number of private alleles (0.012 ± 0.008) indicating a connection between the Coptic and the Egyptian population (Figure 4).

We estimated population structure for the Sudanese populations together with the previously published data from the Somali, Egyptian and Karamoja populations using the software program Structure (version 2.3.3; http://pritch.bsd.uchicago.edu/software/structure2_2.html) [32]. Although most individuals had substantial parts of their genomes assigned to more than one cluster, three main patterns could be distinguished (Figure 5). Individuals from northern Sudan and Egypt were (generally) more likely to fall within the 'green' cluster (P < 0.001, Mann-Whitney U-test), these from southern Sudan and Uganda into the 'red' cluster (P < 0.001, Mann-Whitney U-test), and those from Somalia into a (partly) separate 'yellow' cluster (P < 0.001, Mann-Whitney U-test). The mixed ancestry of basically all individuals is probably a consequence of the limited number of markers (and of the set of markers not being particularly informative about ancestry) rather than an indication of recent admixture, a behavior of admixture analyses that has been reported previously (see Figure 4 in Rosenberg et al.[29]). Using principal component analysis (PCA), based on pairwise genetic distance between populations, the first principal component explained 25.5% of the genetic variation, and distinguished the Coptic, Egyptian, Somali and Nubian populations from the others (Figure 6A). The second component (10.6%) distinguished the Egyptian, and to some extent, the Somali and the Nubian population from most of the others (Figure 6A), and the third component (9.9%) distinguished the Somali population from all others (Figure 6B). PCA indicated that the Egyptian, Coptic, Somali, and to some extent the Nubian groups form genetically distinct populations. The Nuba, Zagawa, Nilotic and Gemar groups clustered with the Karamoja group from Uganda, as was also indicated by the clustering using Structure (Figure 5).

The number of unique alleles (Figure 2B) was greatest in the Somali population, and and in the population structure analyses (Figure 5), the Somali population grouped separately from other populations. Because the Somali population is separated both geographically and linguistically from the other populations included in our study, it is not surprising that it is also genetically distinct. It is possible that the Bantu expansion from West Africa had a stronger effect on the region of the Horn of Africa, where Somalia is located, compared with the region where Sudan is located. For example, the languages in Somalia belong to two major linguistic families, the Afro-Asiatic and Niger-Congo, whereas Nilo-Saharan is absent and the Bantu Swahili language is one of the major languages in Somalia (Ethnologue [1]). Another explanation could be that the Somali population is of both Eurasian and sub-Saharan origin, as suggested by a recent study [33], potentially explaining the differentiation of this population from some east African groups, although many of the Sudanese populations, such as Arabs and the Beja, may also have mixed Eurasian and sub-Saharan origin.

The patterns of population structure we found in northeast Africa, in particular the similarity of Nubian (a northern Sudanese group that speak Nilo-Saharan languages) and the Egyptian population. is consistent with the historical evidence for long-term interactions between Egypt and Nubia, probably resulting in genetic flow between the two regions. However, the Nubian group and the Karamoja group from Uganda share a relatively large number of private alleles (Figure 4), potentially reflecting the shared ancestry of the Nubians with populations from southern Sudan and Uganda. Our results, in addition to mtDNA [7] and Y-chromosome [6, 34, 35] data, suggest that migration, potentially bidirectional, occurred along the Nile between Egypt and Nubia.

Samples

In total, 498 unrelated subjects (366 men, 132 women) were recruited from different geographic regions of Sudan. For each individual, we collected blood and (self-reported) information about the gender and ethnic background of the participant and their parents and grandparents. For the calculation of forensic summary statistics, we excluded seven people because of a potential first-degree relationship, which was inferred using Relpair v2.0.1 [36, 37], leaving 491 subjects. For the population-genetics analyses, we excluded (i) the same 7 subjects because of a potential first-degree relationship, (ii) 29 for whom the sample size of the particular population was <6, and (iii) eight who failed genotyping for at least one marker, resulting in a final subject group size of 454 from 18 Sudanese populations. The geographic locations of the sampled populations are indicated on a map of Sudan (Figure 1), and sample sizes for the populations are given in Table 1. For the population-genetics analyses, we also combined the genotype data from our Sudanese sample (n = 454) with previously published genotype data from Uganda [25], Egypt [26] and Somalia [27].

Genetic markers

Genomic DNA was extracted from DNA storage cards (FTA Classic Cards; Whatman, Maidstone, Kent, UK), following the manufacturer's instructions. Fifteen STR loci (D3S1358, vWA, FGA, D8S1179, D21S11, D18S51, D5S818, D13S317, D16S539, TH01, TPOX, CSF1PO, D7S820, D2S1338 and D19S433) were co-amplified for each of the 498 subjects (AmpFl STR^® Identifiler™ PCR Amplification Kit; Applied Biosystems) following the manufacturer's recommendations [38]. The amplified PCR products were genotyped using an automatic analyzer (ABI PRISM 3730 XL Genetic Analyzer; Applied Biosystems). For each run on the analyzer, an allelic ladder, positive and negative controls, and an internal lane standard (GeneScan-600 LIZ; Applied Biosystems) were included. Allele calling was performed (GeneMapper ID software, version 4.0; Applied Biosystems). We followed the International Society for Forensic Genetics (ISFG) recommendations on the analysis of DNA polymorphisms. The recommended nomenclature was used, and we followed the guidelines on quality control [39].

Analysis of genotype data

Standard summary statistics for forensic applications, including allele frequencies, PE, power of discrimination, and polymorphism information content, were computed (PowerStats, version 12.0; Promega Corp., Madison, WI, USA; http://www.promega.com/geneticidtools/powerstats/). Testing for deviations from Hardy-Weinberg equilibrium was performed using Arlequin, version 3.11; (http://cmpg.unibe.ch/software/arlequin3/) [28]. To assess the power of the 15 Identifiler microsatellite loci for population structure inference, we calculated the informativeness for assignment, I_n, [29] for each locus. The I_n statistic is based on information theory, and evaluates the efficiency of a marker for assigning individuals to one of K populations. Following Rosenberg et al.[29] for a particular locus,

where N is the number of alleles for the locus, K is the number of populations, p _ij is the (parametric) frequency of allele j in population i, p _j is the mean frequency of allele j across populations and 'log' denotes the natural logarithm. The minimal I_n value is 0 (when all alleles have equal frequencies in all populations), and the maximal value is log K (which occurs when no allele is found in more than one population). I_n was computed for the combined dataset of northeast African populations from Sudan, Egypt, Somalia and Uganda. For comparison of the level of informativeness for the 15 Identifiler microsatellites, we computed I_n values for 377 microsatellites genotyped in the HGDP [29]. Average I_n values were computed for two groups of populations from the HGDP: six African populations (Kenyan Bantu speakers, Mandenka, Yoruba, San, Mbuti Pygmy, Biaka Pygmy), referred to as the 'HGDP sub-Saharan group', and a subgroup of three populations from the HGDP who all speak Niger-Congo languages (Kenyan Bantu speakers, Mandenka, Yoruba), referred to as the 'HGDP Niger-Congo group'.

Based on the genotype data from the 15 microsatellites, we inferred population structure for northeast African populations using the clustering software Structure [32, 40]. We used the admixture model, using the F model of correlated allele frequencies across clusters. Each replicate Structure run used a burn-in period of 100,000 iterations, followed by 10,000 iterations from which estimates were obtained. We replicated the Structure analysis 10 times for each choice, based on the number of assumed clusters (K), from K = 2 to K = 10. The 10 replicates for each choice of K was summarized (CLUMPP software; http://rosenberglab.bioinformatics.med.umich.edu/clumpp.html[41]), with the Large K Greedy algorithm (10,000 random permutations) to identify common modes of replicates, and to combine the clustering-results across replicates. The combined clustering result was visualized (distruct package; http://rosenberglab.bioinformatics.med.umich.edu/distruct.html[42]).Population structure was also assessed through calculating the mean squared distance for microsatellite loci [43] between pairs of populations (Populations version 1.2.30;http://bioinformatics.org/project/?group_id=84). The distance matrix was visualized using PCA.

We computed the mean number of distinct alleles, the mean number of private alleles and the mean number of private alleles for pairs of populations (ADZE; http://rosenberglab.bioinformatics.med.umich.edu/adze.html[44]) for the Sudanese populations (Table 1). For this analysis, we merged populations into larger 'ethnic groups' to avoid small sample sizes, and we also excluded the Gemar because of the small sample size (n = 6), and because they could not be grouped with any of the other populations. This analysis was repeated for the Sudanese ethnic groups, including the populations from Egypt, Somalia and Uganda (the Karamoja). The sample from Egypt was collected from five populations, but because no structure could be detected in these five populations [26], we treated them as one population. Analysis of molecular variance (AMOVA) [45] was used to partition genetic variation for predefined linguistic and geographic groups (Arlequin version 3.11 [28]) and the Nuba population was excluded because the group encompassed individuals speaking different languages belonging to the Niger-Congo and Nilo-Saharan linguistic divisions.