Genomic microsatellites identify shared Jewish ancestry intermediate between Middle Eastern and European populations
© Kopelman et al. 2009
Received: 23 October 2009
Accepted: 8 December 2009
Published: 8 December 2009
Skip to main content
© Kopelman et al. 2009
Received: 23 October 2009
Accepted: 8 December 2009
Published: 8 December 2009
Genetic studies have often produced conflicting results on the question of whether distant Jewish populations in different geographic locations share greater genetic similarity to each other or instead, to nearby non-Jewish populations. We perform a genome-wide population-genetic study of Jewish populations, analyzing 678 autosomal microsatellite loci in 78 individuals from four Jewish groups together with similar data on 321 individuals from 12 non-Jewish Middle Eastern and European populations.
We find that the Jewish populations show a high level of genetic similarity to each other, clustering together in several types of analysis of population structure. Further, Bayesian clustering, neighbor-joining trees, and multidimensional scaling place the Jewish populations as intermediate between the non-Jewish Middle Eastern and European populations.
These results support the view that the Jewish populations largely share a common Middle Eastern ancestry and that over their history they have undergone varying degrees of admixture with non-Jewish populations of European descent.
Large-scale genomic studies have contributed to a growing body of knowledge about the population structure of a wide variety of human populations [1–5]. Such studies have enabled precise inferences about the relationships of closely related groups, about the extent to which individuals in neighboring populations can be genetically distinguished, and about the potential of genetics for inference of ancestry at the intracontinental level. In general, Jewish populations, whose genetic origins and population relationships have long been of interest, have been excluded from such studies or examined only peripherally. Although some studies have included members of Jewish populations in the context of analyses of broader geographic regions [6–9], Jewish populations have only recently become a focus of investigation for genome-wide studies of population structure .
The population genetics of Jewish populations has been considered primarily from the perspective of the Y chromosome and mitochondrial DNA, and in smaller-scale studies using as many as 20-30 autosomal genetic markers. Although several studies have supported a genetic affinity among most Jewish populations, potentially due to shared ancestry [11–16], others have suggested similarity between Jewish and non-Jewish populations as a result of some level of gene flow among groups [12, 14, 17–19]. The discovery of shared Y chromosomes common in separate Jewish populations from different geographic regions has strengthened the evidence for shared Jewish genetic ancestry, but as evidenced in the considerable attention given in Israel to the 2008 scholarly book "When and how was the Jewish people invented" , debate continues regarding the issue of whether separate Jewish populations have any deep shared genetic ancestry beyond that shared with non-Jewish groups. The difficulty of fine-scale resolution of Jewish population relationships is highlighted by the different conclusions reached in two early genetic investigations that proceeded concurrently using similar data on classical markers, and that even today remain among the most comprehensive evaluations of Jewish population relationships [13, 17]. Whereas Karlin et al.  observed that most Jewish populations had lower genetic distance to other Jewish populations than to non-Jewish European and Middle Eastern populations included in their study, Carmelli & Cavalli-Sforza  found that a discriminant analysis scattered Jewish populations among clusters corresponding to various non-Jewish European and Middle Eastern groups.
Increasing the number of autosomal markers used in population-genetic studies has the potential to provide more detailed information that may help to resolve the population structure of Jewish populations and their historical neighbors. Here we extend the use of genome-wide markers to evaluate genetic relationships among Jewish populations and other Middle Eastern and European populations. To assess patterns of genetic structure among Jewish populations as well as the relationship of Jewish genetic variation to that of other populations, we examine 678 microsatellites in a collection of 78 individuals of Jewish descent representing four groups defined by community of origin, as well as genotypes of 321 Middle Eastern and European non-Jewish individuals at the same markers. We find that the Jewish populations cluster together in several analyses, separately from the remaining populations. In addition, we find that the genetic ancestry of the Jewish populations is intermediate such that in several types of analysis of population structure, the Jewish populations are placed centrally, between the Middle Eastern populations and the European populations. These results are compatible with an ancient Middle Eastern origin for Jewish populations, together with gene flow from European and other groups in the Jewish diaspora.
To compare the genetic variability of Jewish populations with that of other Middle Eastern and European groups, we examined a sample of 399 individuals, representing four Jewish groups defined by their origin prior to 20th century migrations, as well as 12 other Middle Eastern and European populations from the HGDP-CEPH Human Genome Diversity Cell Line Panel . Our primary interest was in the relationship of Jewish populations to each other and to non-Jewish Middle Eastern and European populations. Previous analysis had demonstrated that the Middle Eastern and European HGDP-CEPH populations form genetic clusters separate from other populations such as those from Central and South Asia [4, 22]. Because inclusion in population structure analyses of distant populations has the potential to obscure genetic differences that might exist among closely related populations [22, 23], we did not include HGDP-CEPH populations from Central/South Asia or other geographic regions unlikely to be relevant for the genetic study of the Jewish populations analyzed.
The Middle Eastern populations included in the study were Bedouin (46), Druze (42), Mozabite (29), and Palestinian (46). The European populations were Adygei (17), Basque (24), French (28), Italian (13), Orcadian (15), Russian (25), Sardinian (28), and Tuscan (8). Middle Eastern and European non-Jewish individuals were taken from the H952 subset of the HGDP-CEPH panel . The Jewish samples included Ashkenazi Jews (20), Moroccan Jews (20), Tunisian Jews (20), and Turkish Jews (20). Two Tunisian Jewish individuals were omitted from the analysis following a procedure for detection of relatives (see below). Jewish individuals were sampled at the Barzilai Medical Center in Ashkelon, Israel, and included immigrants and second-generation immigrants from the source populations. Informed consent was obtained from all participants, and the project was approved by the ethics committee of the Barzilai Medical Center.
The Jewish individuals were genotyped by the Mammalian Genotyping Service for microsatellite loci in Marshfield Screening Sets 16 and 54 http://research.marshfieldclinic.org/genetics. The collection of markers genotyped in the Jewish populations overlaps to a large extent with a set of 783 markers previously reported for the HGDP-CEPH individuals [25, 26], but is not completely identical to the earlier marker set. Thus, to enable comparison of the 80 newly included Jewish individuals with commensurable genotypes previously reported for the HGDP-CEPH individuals, data analysis was restricted to 678 loci typed across all populations. Preparation of genotypes for the Jewish populations proceeded in the same manner as the preparation of genotypes in the study of Wang et al. , which used the same set of 678 markers; for the Middle Eastern and European non-Jewish populations, the data used here are the same as in that study, except that we considered only individuals from the H952 subset that excluded close relatives.
Considering all pairs among the 80 Jewish individuals, we examined identity-by-state sharing to detect relatives. In addition, separately for each Jewish population we screened pairs of individuals for close relatives by utilizing the RELPAIR program [28, 29]. Both approaches were applied in a similar manner to that used in a previous study . Two second-degree relative pairs were detected in the Tunisian sample, and for each pair, one individual was omitted from further analysis (individuals 2345 and 2348).
Expected heterozygosity was computed by using the sample-size-corrected estimator, averaging across loci to obtain an overall estimate . Paired values for individual loci were used in Wilcoxon signed-rank tests of heterozygosity across populations. For each locus, the number of distinct alleles and the number of private alleles, that is, alleles unique to one population, were measured as functions of the number of sampled chromosomes. This analysis used the rarefaction procedure, as implemented in ADZE , averaging the number of distinct alleles and the number of private alleles across possible subsets of sampled chromosomes while adjusting for differences in sample size across populations. We obtained the mean number of distinct alleles and the mean number of private alleles for each of three combined sets of samples (European, Jewish, Middle Eastern), averaging across loci. Our ADZE analysis used only 656 of the 678 loci, omitting loci with >15% missing data in any one of the three combined samples. This choice accords with that of Szpiech et al. , producing similar results to those obtained with all 678 loci while permitting higher numbers of sampled chromosomes to be considered.
The program Structure 2.2.3  was used to assess population structure for the full dataset used in this study, using the F model of correlation in allele frequencies. The program Structure is the most widely used in a family of programs that cluster individuals based on their diploid genotypes, in an unsupervised manner, without using prior knowledge of their populations of origin (additional programs in this collection include BAPS [33, 34], mStruct , and Structurama ). Using the admixture model of individual affiliations, for each individual Structure determines the fractions of genetic affiliation of the individual in each of a predetermined number of clusters (K). The admixture model is particularly suitable in complex populations for which mixed membership of individuals in multiple clusters is expected [32, 37]. We ran Structure for K ranging from 2 to 16, with 40 replicates for each K and a burn-in period of length 30,000 iterations followed by 30,000 additional iterations. For each K, and for each pair of replicates, we determined the similarity of the estimated affiliations using the symmetric similarity coefficient (SSC) scores based on the best alignment of the replicates. This alignment was obtained using the LargeKGreedy algorithm of the software CLUMPP , with 10,000 random input sequences. Using a threshold of 0.8 for the SSC scores, we separated different convergence modes among the 40 replicates with a given value of K, where a mode was defined as a clique such that all pairs of replicates within the clique had SSC≥0.8. For each mode and each K, CLUMPP was again used to obtain the average cluster memberships of the replicates placed into the mode. The program Distruct  was used to produce plots of these average memberships. Our combined application of Structure and CLUMPP to summarize clustering results follows the approach employed in previous studies [2, 27].
Multimodality in clustering solutions was observed for some values of K. The mode containing the largest number of replicates (the "major mode") for K = 2 contained 39 of 40 Structure runs. For K = 3 and K = 5, only one mode was found, containing all 40 runs. For K = 4, the major mode contained 15 of 40 runs. The second-largest mode contained 11 runs, and was very similar to the major mode of K = 5, except that it did not separate the Mozabites and the Bedouins (results not shown). For K = 6, the major mode contained 20 of 40 runs. The second-largest mode, containing 14 runs, was very similar to the major mode, except that it showed greater similarity of the Bedouins to the Palestinians (results not shown). For K = 7 and K = 8, the number of replicates in the major mode was well below half of the total number of replicates examined, equaling 13 for K = 7 and 12 for K = 8. Two new clusters were identified in the major mode for K = 8 compared to the analysis for K = 6; one of these clusters largely corresponded to the Tunisian Jews and the other largely corresponded to the Sardinians (results not shown). The major mode for larger values of K contained fewer replicates, at most 7 for values of K>8. For the larger values of K (K>8), the second-largest mode contained nearly as many replicates as the major mode - for example, for K = 7, K = 8, K = 9, and K = 10, the second-largest mode possessed 8, 7, 6, and 4 runs, respectively, compared to 13, 12, 7, and 5 replicates for the major mode. Because inferences based on K>6 were less replicable than those based on smaller values of K, we chose for display the major mode for each K from 2 to 6.
Neighbor-joining population trees were produced using the neighbor program in the software package Phylip 3.65 , considering each of three genetic distance measures. Distance measures were chosen among those found to produce relatively high bootstrap support in comparisons of multiple trees in past microsatellite studies [41–43]. The distance matrices for the allele-sharing distance (computed as one minus the proportion of shared alleles under Hardy-Weinberg proportions ), chord distance  and Nei's standard distance (computed as one minus Nei's identity ) were obtained with the software Microsat , bootstrapping across loci 10,000 times. For each collection of 10,000 bootstrap replicates, we constructed a majority-rule consensus tree, resolving multifurcations by sequentially incorporating the groupings that had the highest frequencies in the set of bootstraps and that were compatible with groupings already incorporated.
For each Jewish population we examined the genetic distances between the allele frequency vector of that population and linear combinations of allele frequency vectors for pairs of other populations. For each Jewish population and each linear combination of two other populations, we obtained a mean allele-sharing genetic distance across loci. For each pair of populations considered in obtaining linear combinations, we examined combinations in which the fraction from the first population ranged from 0 to 1, with a step size of 0.01.
Pairwise distances between individuals were calculated using allele-sharing distance . We then performed multidimensional scaling (MDS) for the individual distance matrices using the cmdscale function in R. This function performs classical MDS based on the approach of Cailliez . MDS analysis was also performed for several subsets of the full collection of individuals: Jewish individuals alone, Jewish and European individuals, Jewish and Middle Eastern individuals, and Jewish and Palestinian individuals.
In the two-dimensional MDS plots, we evaluated distances between groups of individuals by using the average linkage distance [49, 50]. For a pair of groups in an MDS plot, this quantity, denoted here by L 0, is the mean Euclidean distance between the location in the plot of a randomly chosen member of the first group and a randomly chosen member of the second group. The significance of the separation of two groups was evaluated by permutation of labels within groups, as specified in the contexts of the various plots. The probability that a random permutation of the labels gives rise to a smaller average linkage distance for two groups than that seen using the actual labels was obtained from a distribution of the average linkage distance across 1000 permutations. While the magnitude of a value of L 0 is not itself meaningful, the relative size of L 0 values for multiple pairs of groups in the same MDS plot carries information about the relative levels of separation of the various pairs.
We also performed Structure analysis for the Jewish individuals alone. Using the same Structure model and the same lengths for runs as in the analysis of the full data, we considered values of K ranging from 2 to 6, performing 40 replicates for each value. CLUMPP and Distruct were used to process the Structure results in the same manner as in the analysis with the full dataset. We found that for K = 2, the major mode contained all 40 replicates, and that for K>2, the additional subdivision observed beyond that seen for K = 2 was negligible (results not shown).
Heterozygosity and sample size for European, Jewish, and Middle Eastern populations.
Number of individuals
Mean heterozygosity across loci
Standard deviation of heterozygosity across loci
To investigate the possibility of further separation of the European and Jewish individuals, we also examined the MDS representation of distances for these individuals alone (Figure 5B). The Tunisian Jews are located further from the pooled European populations than are any of the other Jewish populations, with L 0 = 0.1180 and P > 0.999 based on permutation of the Jewish population labels, compared with L 0 = 0.0859 (P = 0.005), L 0 = 0.0899 (P = 0.065), and L 0 = 0.0946 (P = 0.327) for the Turkish, Ashkenazi, and Moroccan Jewish populations, respectively. The European populations that cluster closest to the pooled Jewish populations are the Tuscan, Italian, Sardinian, and Adygei populations, each with P < 0.001 based on permutations of the European population labels.
In a similar way, we also examined the MDS representation for the Middle Eastern and Jewish populations alone, excluding the relatively distinctive Mozabites. As can be seen in Figure 5C, the Palestinians are relatively close to the pooled Jewish populations (L 0 = 0.0488, P < 0.001 in permutations of the labels among the Bedouin, Druze, and Palestinian populations), whereas the Bedouin and Druze populations are more separated from the pooled Jewish populations (L 0 = 0.1102, P > 0.999, and L 0 = 0.1023, P = 0.990, respectively) and largely produce distinctive clusters of their own.
Because the closest population in Figure 5C to the four Jewish populations was the Palestinian population, we also considered the MDS representation of only the Jewish populations together with the Palestinians (Figure 5D). The plot places the Palestinians closer to the Moroccan and Turkish Jews than to the other Jewish populations (L 0 = 0.1153, P = 0.024, and L 0 = 0.1009, P < 0.001 for Moroccan and Turkish Jews, respectively, in permutations of the labels among the Jewish populations, in contrast with L 0 = 0.1356, P = 0.847, and L 0 = 0.1632, P > 0.999 for Ashkenazi and Tunisian Jews, respectively). It further suggests that the Tunisian Jews are the most distinctive Jewish population, whereas the Ashkenazi, Turkish, and Moroccan Jewish populations are genetically more similar to each other.
To examine the affinities of Jewish populations and their relationships to Middle Eastern and European populations, we have analyzed a sample of 78 individuals from four Jewish populations at 678 autosomal microsatellite loci together with corresponding genotypes of 321 Middle Eastern and European non-Jewish individuals from 12 populations. In various statistical analyses of population structure, the Jewish populations had a high level of genetic similarity to each other, grouping together in Bayesian clustering (Figure 2), neighbor-joining population trees based on three population-level genetic distances (Figure 3), and MDS analysis based on individual-level genetic distances (Figure 5). Moreover, in multiple analyses the Jewish populations were placed in an intermediate position in relation to the European and Middle Eastern populations, both in analyses of genetic variability (Table 1 and Figure 1) and in analyses of population structure (Figures 2, 3, and 5). When we searched for linear combinations of population pairs that produced minimal genetic distance to Jewish populations, the minima were often obtained from pairs that included one European population and one Middle Eastern population in similar proportions (Figure 4).
Whereas recent Y-chromosomal studies have identified a trend of genetic affinity among Jewish populations [12, 15, 18], most notably a shared group of haplotypes common in Jewish priests from different Jewish populations [16, 52, 53], past autosomal studies of multiple Jewish populations have been somewhat more equivocal regarding the clustering of Jewish populations separate from non-Jewish populations [13, 14, 17, 19, 54–56]. Recent genomic studies that have identified a component of distinctive ancestry for Jewish individuals have largely focused on Ashkenazi Jews sampled in the United States in relation to the broader European-American population [7–10], finding most recently that individuals with even partial Ashkenazi ancestry can be detected on the basis of principal components analysis . Our study furthers the results of these studies by showing that a distinctive component of genomic ancestry extends to Jewish populations more broadly.
A simple explanation for the clustering of the Jewish populations is that this pattern is the consequence of shared ancestry with an ancestral Middle Eastern group. Under this scenario, the intermediate placement of the Jewish populations with respect to European and Middle Eastern populations would then result from early shared ancestry of the Jewish and Middle Eastern populations, followed by subsequent admixture of the Jewish populations that took place with European groups or other groups more similar to the Europeans than to the Middle Eastern populations in the study. Although it is difficult to assess the specific nature of the admixture on the basis of our analysis, this explanation is supported by other genetic studies that find a combination of shared ancestry and admixture among Jewish populations [56–59] and by historical records of conversions to Judaism [20, 60–64]. Further sampling of matched Jewish and neighboring non-Jewish populations will be informative for investigating the evidence for this scenario.
One frequently discussed conversion that likely occurred in the 8th century at the far eastern edge of Europe, north of the Caucasus and Black Sea regions, is that of the Khazarian kingdom [60, 62, 64]. The demographic effect of this conversion is debated, so that only a small minority of the Khazars may have adopted Judaism. While the ultimate fate of the Khazar population remains unknown, the theory has been advanced that a large fraction of the ancestry of eastern European Jews derives from the Khazars [60, 62–64]. This theory would predict ancestry for the eastern European Ashkenazi Jewish population to be distinct from that of the other Jewish populations in the study. Although we did not observe such a distinct ancestry, it is noteworthy that in some analyses (Figures 2 and 3), as was observed in the recent study of Need et al. , we did detect similarity of the Adygei, a north Caucasian group from the area once occupied by the Khazars, to the Jewish populations.
In several analyses, the population in the study that is most similar to the Jewish populations is the Palestinian population. This result is reflected by the fact that for K = 5, Bayesian clustering with Structure assigns the Jewish populations and the Palestinians to the same cluster (Figure 2), and by the relatively close placement of the Palestinians and the Jewish populations in MDS plots of individual distances (Figure 5). This genetic similarity, which is supported by several previous studies [12, 65, 66], is compatible with a similar Middle Eastern origin of the Jewish populations and the Palestinians. Admixture of the Palestinians with groups with European origins might have maintained or augmented this shared ancestry, especially if it was paralleled with similar admixture of these groups with Jewish populations.
Among the Jewish populations, the Tunisians were found to be the least variable and most distinctive, and their genotypes could be most easily distinguished from those of the three other Jewish populations. This result suggests a smaller population size and greater degree of genetic isolation for this population compared to the other Jewish groups, or a significant level of admixture with local populations. These explanations are not incompatible, as it is possible that early admixture was followed by a long period of isolation. Some Berber admixture of Tunisian Jews may very well have taken place [61, 63], and documentation of rare Mendelian disorders in Tunisian Jews [67–69] supports a view of isolation with relatively few founding individuals. A smaller-scale autosomal study that did not include Tunisian Jews found the neighboring Libyan Jewish population to be distinctive with respect to other Jewish populations , and our results concerning the Tunisian Jewish population might reflect a similar phenomenon.
We note that caution is warranted in interpreting some of our results. For example, in the population trees produced from three distance measures (Figure 3) there is disagreement on the branching order of three of the European populations closest to the Jewish populations (Adygei, Sardinian, and Tuscan). Thus, from these data, it is difficult to make strong inferences regarding the most similar European populations to Jewish groups. However, consistent with studies that have incorporated a single Jewish population in a broader European context [6–9], southern groups from Europe are placed closer to the Jewish populations than more northerly groups. An additional disagreement among the trees lies in the branching pattern of the Jewish populations themselves. However, this within-group disagreement does not affect the basic pattern visible in all three trees, in which the Middle Eastern and European populations cluster separately with the Jewish populations in the center. A possible additional concern is ascertainment bias on the loci favoring high levels of European polymorphism. However, no strong evidence for ascertainment bias has been detected for the loci considered here , and in general, ascertainment effects in humans are only significant in studies of populations from distant geographic regions. Two recent genomic studies of Ashkenazi Jews sampled in the United States [10, 71] have demonstrated the potential of the use of haplotypes and extremely densely placed markers for detailed investigation of genetic variation in a Jewish population, and it is possible that with the resolution provided by higher densities and haplotypic analysis, some of the discrepancies in our analyses might be overcome. Irrespective of the limitations of our study, however, our main results, namely the clustering of the Jewish populations and the intermediate placement of the Jewish populations compared to European and Middle Eastern populations, were robust across diverse types of analysis.
We thank M. DeGiorgio, J. Degnan, L. Maltz, T. Pemberton, and Z. Szpiech for assistance. Grant support was provided by the University of Michigan/Israeli Universities Partnership in Research, the Burroughs Wellcome Fund, and the Alfred P. Sloan Foundation.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.