Signatures of founder effects, admixture, and selection in the Ashkenazi Jewish population

Abstract

The Ashkenazi Jewish (AJ) population has long been viewed as a genetic isolate, yet it is still unclear how population bottlenecks, admixture, or positive selection contribute to its genetic structure. Here we analyzed a large AJ cohort and found higher linkage disequilibrium (LD) and identity-by-descent relative to Europeans, as expected for an isolate. However, paradoxically we also found higher genetic diversity, a sign of an older or more admixed population but not of a long-term isolate. Recent reports have reaffirmed that the AJ population has a common Middle Eastern origin with other Jewish Diaspora populations, but also suggest that the AJ population, compared with other Jews, has had the most European admixture. Our analysis indeed revealed higher European admixture than predicted from previous Y-chromosome analyses. Moreover, we also show that admixture directly correlates with high LD, suggesting that admixture has increased both genetic diversity and LD in the AJ population. Additionally, we applied extended haplotype tests to determine whether positive selection can account for the level of AJ-prevalent diseases. We identified genomic regions under selection that account for lactose and alcohol tolerance, and although we found evidence for positive selection at some AJ-prevalent disease loci, the higher incidence of the majority of these diseases is likely the result of genetic drift following a bottleneck. Thus, the AJ population shows evidence of past founding events; however, admixture and selection have also strongly influenced its current genetic makeup.

The Ashkenazi Jewish (AJ) population has long been viewed as a genetic isolate, kept separate from its European neighbors by religious and cultural practices of endogamy (1). Population isolates are frequently used in genetic research, as such groups are presumed to have reduced genetic diversity, along with increased frequencies of recessive disorders, identity-by-descent (IBD), and linkage disequilibrium (LD) as the result of founder events and population bottlenecks (2, 3). Accordingly, the AJ population is often the subject of Mendelian and complex disease studies, although evidence that the AJ population carries all of the hallmarks of a genetic isolate has not been fully established.

The most compelling genetic evidence of founder events in the AJ population is the elevated frequency of at least 20 rare recessive diseases attributed to genetic drift following bottlenecks. Coalescence times ascribed to founder mutations for some of these diseases correspond well with historical migrations or episodes of extreme persecution, supporting the argument for genetic drift (1, 4, 5). Strong evidence of founder events also comes from studies of mitochondrial DNA (mtDNA), which show significantly less diversity of mtDNA haplotypes among the AJ population (6, 7). Y-chromosome studies also indicate only a low amount of admixture with neighboring Europeans (8–10). Additionally, some reports have measured higher LD in the AJ genome compared with reference populations (11, 12). Furthermore, a recent study showed increased IBD in Jewish Diaspora populations, including the AJ population, in support of a bottleneck (13).

Despite the evidence for founder events, questions still remain whether population bottlenecks and genetic isolation can account for the current genetic structure of the AJ population. For example, some have posited that selection can explain the increase in rare recessive disorders, arguing for a heterozygous advantage for the mutant alleles (14–16). Moreover, Y-chromosome studies, in contrast to mtDNA results, reveal that Y-chromosome diversity in the AJ population is comparable to their non-Jewish European neighbors (8, 17). Admixture estimates using markers at a few autosomal loci or based on STRUCTURE clustering results have also shown much higher European admixture than reflected in the Y chromosome (13, 18, 19). Furthermore, recent studies have found no increase in LD at short distances between markers and suggest increased heterozygosity compared with Europeans, concluding that the AJ population is likely an older and larger population that is distinguished by its Middle Eastern origin, rather than the effect of population bottlenecks (20–22).

To better understand the genome-wide genetic structure of the AJ population and search for genetic signatures of founder events, we genotyped 471 unrelated AJ individuals at 732k autosomal SNPs. Our analysis of this large cohort clarifies many inconsistencies described above, establishing that the AJ population has had substantial admixture with European populations, increasing its genetic diversity and LD, yet maintaining a significant level of founder haplotypes identical-by-descent. In addition, we applied extended haplotype tests for detecting regions under positive selection and, although this does not account for most AJ-prevalent disorders, we did identify several regions of putative selection in the AJ genome.

Results and Discussion

Genetic Diversity and Linkage Disequilibrium.

It is well established that populations that have been isolated for extended periods show two major patterns of diversity: (i) low genetic variation and (ii) high LD (2, 3). To compare the genetic diversity of the AJ population with their European neighbors, we merged our genotype data for 471 Ashkenazi Jews with two large European cohorts, a continental European (Euro) cohort of 1,705 individuals (242k overlapping SNPs), and a European American (EA) cohort of 1,251 individuals (732k overlapping SNPs). Consistent with recent reports (13, 20, 23–25), principal component analysis (PCA) using these combined datasets confirmed that the AJ individuals cluster distinctly from Europeans, aligning closest to Southern European populations along the first principal component, suggesting a more southern origin, and aligning with Central Europeans along the second, consistent with migration to this region (Fig. S1).

To explore the amount of genetic variation within the AJ and European populations, we first measured the mean heterozygosity. Surprisingly, we found a higher level of heterozygosity among AJ individuals compared with Europeans (P < 1e-40) (Table 1), confirming speculation made in one recent report and a trend seen in another (20, 22). Although this difference may appear small, it is highly statistically significant because of the large number of individuals and markers analyzed, even after pruning SNPs that are in high LD. The higher diversity in the AJ population was paralleled by a lower inbreeding coefficient, F, indicating the AJ population is more outbred than Europeans, not inbred, as has long been assumed (P < 1e-7) (Table 1). The greater genetic variation among the AJ population was further confirmed using a pairwise identity-by-state (IBS) permutation test, which showed that average pairs of AJ individuals have significantly less genome-wide IBS sharing than pairs of EA or Euro individuals (empirical P value < 0.05). Thus, our results show that the AJ population is more genetically diverse than Europeans.

Table 1.

Genetic diversity in Ashkenazi Jews and Europeans

Pop	LD pruned	SNPs	HET(Exp)	HET(Obs)	F
AJ	No	242k	0.2296	0.2298**	−0.0011*
Euro	No	242k	0.2262	0.2257	0.0022
AJ	Yes	107k	0.2390	0.2394**	−0.0014*
Euro	Yes	107k	0.2363	0.2358	0.0020
AJ	No	732k	0.2705	0.2714**	−0.0033*
EA	No	732k	0.2696	0.2697	−0.0004
AJ	Yes	207k	0.2518	0.2527**	−0.0038*
EA	Yes	207k	0.2510	0.2511	−0.0005

Expected heterozygosity, HET(Exp), observed heterozygosity, HET(Obs), and inbreeding coefficient, F, for the AJ, Euro, and EA populations. One SNP of a pair was pruned out of the dataset if the pair was in high LD, defined as pairwise r² > 0.5. The shading of rows highlights the AJ population with the corresponding SNP-matched European population.

**P < 1e-40 and *P < 1e-7 for AJ compared with its paired European population (t test).

We next compared LD between populations by calculating the average r² in sliding windows across the genome and found that LD is consistently higher in the AJ population, with only occasional regions of greater LD in Europeans (Fig. 1A). Of note, for this and all subsequent analyses, we limited the European cohorts to 471 individuals to avoid any sample size differences with the AJ population. Analysis of LD decay also demonstrates stronger LD at all distances between SNPs (Fig. 1B), contrary to a recent study that found less LD at short distances (21), illustrating the value of a larger sample size for LD calculations. The higher LD was confirmed by using a binomial test to show that significantly more SNP pairs had higher r² in the AJ population when directly compared with the same SNP pair in the European populations (P < 9.22e-15) (Table S1). Similar results were also seen when D′ was used to measure LD.

Fig. 1.

LD in Ashkenazi Jews compared with Europeans. (A) A 1.7-Mb sliding window shifting 100 kb along chromosome 22. (Upper) AJ and Euro populations; (Lower) AJ and EA populations. For each window, the average r² was calculated between all pairs of SNPs within 500 kb of each other. The Insets show the AJ r² minus the Euro or EA r², respectively. (B) The decay of LD across all of the autosomes was assessed by grouping pairs of SNPs according to the distance separating them (0–5 kb, 5–10 kb, …, 495–500 kb) and then plotting the average r² of each bin. (Left) Distances from 0 to 50 kb; (Right) distances from 50 to 500 kb. Note the scales on the y axis are different for the Left and Right.

We also compared the genome-wide haplotype structure between the AJ and European populations using a haplotype modeling algorithm (26), which models phased haplotypes as edges that pass through nodes at each SNP across the genome. The number of nodes in the model is correlated to the genetic variation, and the number of edges per node is inversely correlated to the haplotype length. Using this model, we found that the AJ population has a greater number of nodes (0.88–1.11% more) but fewer edges per node (3.82–4.76% fewer) compared with the Europeans (P < 1e-50) (Table S2), indicating both higher genetic variation and longer haplotypes in the AJ population, consistent with our previous results.

Although the elevated LD and longer haplotypes could be the result of severe bottlenecks and founder events, such events would not account for the increased genetic diversity. However, increased LD can also arise from the admixture of genetically distinct populations (27–30), which could also explain the elevated diversity. Therefore, our data suggest that the higher diversity and LD observed in the AJ population may be the result of admixture rather than founder effects. An alternative explanation is that the AJ population simply arose from a more genetically diverse Middle Eastern founder population and, therefore, its diversity is reduced relative to the founder but not relative to the host Europeans. To address this possibility, we merged our autosomal AJ genotype data with data from the Human Genome Diversity Panel (HGDP) (31), with 168k overlapping SNPs. We removed SNPs in high LD and measured the mean heterozygosity per locus across the combined Middle Eastern populations (Bedouin, Palestinian, and Druze) and found that the AJ population had higher heterozygosity (0.3121 vs. 0.3053, P < 1e-23). Other reports showing no increased heterozygosity in the AJ relative to Middle Eastern populations (13, 22) were probably limited by lower AJ sample sizes, which our dataset overcomes. Thus, the increased genetic diversity and LD appear consistent with admixture rather than founding effects.

Admixture.

To evaluate admixture in the AJ population, we investigated the similarity between AJ and HGDP populations using PCA as well as a population clustering algorithm (32). Both analyses show that AJ individuals cluster between Middle Eastern and European populations (Fig. 2 A and B and Fig. S2A), corroborating other recent reports (13, 20, 22, 23, 25). Interestingly, our population clustering reveals that the AJ population shows an admixture pattern subtly more similar to Europeans than Middle Easterners (Fig. 2 A and C, Lower), while also verifying that the Ashkenazi Jews possess a unique genetic signature clearly distinguishing them from the other two regions (Fig. 2C, Upper). The fixation index, F_ST, calculated concurrently to the PCA, confirms that there is a closer relationship between the AJ and several European populations (Tuscans, Italians, and French) than between the AJ and Middle Eastern populations (Fig. S2B). This finding can be visualized with a phylogenetic tree built using the F_ST data (Fig. S2C), showing that the AJ population branches with the Europeans and not Middle Easterners. Two recent studies performing PCA and population clustering with high-density SNP genotyping from many Jewish Diaspora populations, both showed that of the Jewish populations, the Ashkenazi consistently cluster closest to Europeans (13, 25). Genetic distances calculated by both groups also show that the Ashkenazi are more closely related to some host Europeans than to the ancestral Levant (13, 25). Although the proximity of the AJ and Italian populations could be explained by their admixture prior to the Ashkenazi settlement in Central Europe (13), it should be noted that different demographic models may potentially yield similar principal component projections (33); thus, it is also consistent that the projection of the AJ populations is primarily the outcome of admixture with Central and Eastern European hosts that coincidentally shift them closer to Italians along principle component axes relative to Middle Easterners. Taken as a whole, our results, along with those from previous studies, support the model of a Middle Eastern origin of the AJ population followed by subsequent admixture with host Europeans or populations more similar to Europeans. Our data further imply that modern Ashkenazi Jews are perhaps even more similar with Europeans than Middle Easterners.

Fig. 2.

Admixture and its relationship with LD. (A–C) Ancestral population clustering and PCA were performed using the combined AJ and HGDP populations. The AJ population was divided into three random subgroups of 157 individuals to better match the population size of the Middle Eastern and European populations in the HGDP dataset. The data shown represent one subgroup, and all three had similar results. (A) Population clustering analysis of AJ and all HGDP individuals for seven, K = 7, theoretical ancestral populations. Each color represents a different ancestral population and each vertical line represents a single individual. The proportion of each color within an individual signifies the fraction of ancestry derived from the color's ancestral population. The red arrow highlights the AJ population. (B) PCA performed using only the AJ, Middle Eastern, and European HGDP populations. (C) Population clustering analysis with K = 3 (Upper) or K = 2 (Lower) ancestral populations, performed using only the AJ, Middle Eastern, and European HGDP populations. (D) The correlation between admixture and LD was determined by plotting the average r² for all SNP pairs within the given intervals of δ₁δ₂, where δ₁ is the allele frequency difference between the founding populations at locus 1, and δ₂ is the frequency difference at locus 2. The legend indicates the populations representing the founding populations: Middle Eastern (ME), European American (EA), or Yoruba in Ibadan, Nigeria (YRI). Error bars show SEM.

To quantify the level of admixture within the AJ genome given the model of a Middle Eastern origin and European admixture, we applied a likelihood method (34) to differentiate the relative ancestry of each locus across the genome. We used the combined Palestinian and Druze populations to represent the Middle Eastern ancestor and tested three different European groups as the European ancestral population (SI Materials and Methods). Using these proxy ancestral populations, we calculated the amount of European admixture in the AJ population to be 35 to 55%. Previous estimates of admixture levels have varied widely depending on the chromosome or specific locus being considered (18), with studies of Y-chromosome haplogroups estimating from 5 to 23% European admixture (8, 9). Our higher estimate is in part a result of the use of different proxies for the Jewish ancestral population. Our analysis used the Middle Eastern population frequencies as the putative Jewish ancestor, similar to a previous approach (18), whereas the studies of Y-chromosome admixture used a combination of several Jewish Diaspora populations. Our calculations will have overestimated the level of admixture if the true Jewish ancestor is genetically closer to Europeans than Middle Easterners; however, using the Jewish Diaspora populations as the reference Jewish ancestor will naturally underestimate the true level of admixture, as the modern Jewish Diaspora has also undergone admixture since their dispersion. Furthermore, because our analysis incorporates data from considerably more markers across all of the autosomes, thus including both male and female contributions to admixture, we believe our estimate is closer to the true level of admixture. Recent STRUCTURE analysis of the entire Jewish Diaspora estimated that the Ashkenazi and Syrian Jewish populations have between 20 and 40% European admixture (13). Our estimate overlaps this assessment, although we avoided a similar approach because STRUCTURE-like algorithms model hypothetical ancestral populations that likely never existed in reality (35).

Next, we wanted to consider if admixture could account for the increased LD in the AJ population. Admixture between genetically differentiated populations gives rise to an increase in LD proportional to δ₁δ₂, where δ₁ is the allele frequency difference between the founding populations at locus 1, and δ₂ is the frequency difference at locus 2 (27, 29, 30). Admixture LD decays within a few generation at long distances (>20 cM) but decays slowly at short distances (< 10 cM) (27, 29), allowing us to detect admixture LD in the AJ genome even if admixture occurred early in its history. We, therefore, tested if allele frequency differences between the Middle Eastern and European populations correlated with elevated AJ LD. Indeed, we see that the increase in δ₁δ₂ coincides dramatically with increased LD in the AJ population (Fig. 2D). Using an alternative ancestral population, the Yoruba (YRI), which has even greater allele frequency differences, does not show a similar trend, confirming that admixture between Middle Eastern and European populations contributes to the high LD seen in the AJ population. Models proposed by Kruglyak (36) imply that a population isolate would have to undergo an extreme bottleneck to see a significant increase in LD. Admixture models, on the other hand, reveal that LD can be elevated even when admixture rates are low (27–29). Therefore, our data suggest that the elevated LD in the AJ population has arisen primarily as the result of admixture rather than founder effects.

Identity-by-Descent.

Another genetic hallmark of population isolates is elevated IBD. We searched for regions of IBD using a hashing and extension algorithm (37) and discovered significantly more segments of IBD in the AJ population compared with Europeans or an out-group population, YRI (P < 1e-10) (Table S3). Plotting the frequency of IBD across the genome reveals that most loci are usually shared by less than 5% of AJ pairs (Fig. 3A and Fig. S3A). The exceptions to this are the pericentromeric regions that often show a high frequency of sharing in all populations, possibly because of low SNP density or other factors. A recent study by Atzmon et al. (13) also reported higher IBD in a small AJ cohort; however, we found considerably more IBD in our AJ population (4.5 segments per pair vs. 1.6 segments per pair), even when the IBD segments were similarly filtered (Table S3, Bottom). Given our much larger AJ cohort, as well as the availability of trios to help in haplotype phasing, our results are likely more precise. When we plotted the IBD length decay and maximum IBD length distribution we did see similar trends to Atzmon et al., consistent with a bottleneck in the AJ history (Fig. S3 B and C). Furthermore, the increased IBD in the AJ population was uniform across the genome, indicating it was not the byproduct of a few loci under selection, but likely the outcome of founder events and the persistence of long founding haplotypes.

Fig. 3.

IBD and frequency of shared haplotypes. (A) The percentage of IBD for all pairs of individuals within the AJ, EA, and YRI populations is plotted for chromosome 1. Pairs consisting of one Ashkenazi Jew with one European American (AJ-EA) were also analyzed. The symbol ‡ mark the three loci whose haplotype frequencies are analyzed in B. (B) The 100-SNP haplotype blocks centered at three loci marked in A were analyzed for their frequency in the AJ and EA population. Red circles represent the haplotypes identified with IBD in the AJ population and the size of the circle indicates the frequency that the haplotype is shared (i.e., 50 indicates that the given haplotype is shared by half of the AJ pairs with IBD at that locus). The scale of the circles is given on the far right. (Center) Blue circles represent the haplotypes with IBD in the EA population.

Because less than 5% of the AJ pairs share IBD at any given locus, this implies that the long founding haplotypes are present at low frequency in the AJ population. To explore this, we measured the relative IBD-sharing frequencies of long haplotypes at three loci on chromosome 1 (Fig. 3B). This analysis confirms that many long haplotypes sharing IBD are found at low frequency in the AJ population and are not present in the EA population, presumably representing the existence of older founder haplotypes. The presence of multiple haplotypes contributing to the overall level of IBD at a given locus also confirms that the IBD is not the result of selection of a single haplotype. Additionally, some haplotypes sharing IBD in the AJ population are also present in lower frequency in the EA population, which may be evidence of early admixture. Taken together, our IBD results support the existence of founder effects in the history of the AJ population, as well as admixture with European neighbors.

Positive Selection.

Positive selection at disease loci is an alternative explanation to genetic drift that could account for the prevalence of AJ diseases (14–16). Our study is unique in applying extended haplotype tests, the integrated haplotype score (iHS) and cross-population extended haplotype homozygosity test (XP-EHH) (38, 39), to the AJ population. Both tests are designed to uncover selected alleles with higher frequency than expected relative to their haplotype length. The iHS method has greater power to detect selected alleles at intermediate frequency, and the XP-EHH test has greater power when selected alleles approach fixation in one population relative to another (38). Importantly, the XP-EHH test normalizes for differences in genome-wide haplotype length to account for demographic differences between populations, allowing us to directly compare signals of selection in the AJ and European populations (38). Our IBD analysis suggests that the AJ population shares long haplotypes as the result of founder events or bottlenecks. However, because of the low frequency of these long shared haplotypes, they should not interfere with identifying true regions under strong selective pressure. Furthermore, these long-range haplotype tests are designed to test the hypothesis of neutrality, so rejection suggests that selection has likely occurred, but failure to reject does not mean that selection is absent, implying that undetected loci under selection exist.

We first performed the iHS test separately for the AJ and EA populations and we observed similar magnitudes of integrated haplotype scores between the populations, indicating that the frequency of founding haplotypes in the AJ population did not substantially alter the extended haplotype scores. We calculated the fraction of SNPs with standardized |iHS| > 2 for nonoverlapping, 40-SNP windows across the genome, an approach shown to be more powerful for detecting true regions of selection rather than relying on iHS scores for single SNPs (39). Comparing the top iHS windows in both populations reveals many regions of putative selection in common, including several regions previously implicated as being under selective pressure (Fig. 4 and Table S4). Approximately half of the top 1% of iHS windows in each population are shared by both. This finding is consistent with a recent report that found large overlap between selected regions in Middle Eastern and European populations (40).

Fig. 4.

The top 1% of iHS-selected regions. Each chromosomal ideogram shows the position of the top 1% of iHS windows in the AJ population, red, and the EA population, blue.

To explore whether regions of selection in the AJ population included any loci of known Ashkenazi diseases, we examined 21 disease- and cancer-susceptibility loci with known mutations found at higher frequency in the Ashkenazi population. Only 6 of the 21 genes fell in or near (within 500 kb) the top 5% of the AJ iHS windows (Table 2). Among these is the Tay-Sachs disease gene, HEXA, whose selection has been widely debated (4, 5, 14–16) and was found ~400 kb downstream of a window on chromosome 15 identified in the top 1% of the AJ iHS hits. Although none of the SNPs interrogated immediately adjacent to the HEXA locus showed elevated iHS signals, it is possible that the nearby region may contain regulatory elements under selection that affect HEXA expression. Cochran et al. (14) speculated that selection of many of the AJ-prevalent disease loci, especially the lysosomal diseases, conferred an increase in intelligence that was necessary historically for the AJ economic survival. Our data shows evidence of strong selection at or near only six disease loci, including only one out of the four AJ-prevalent lysosomal storage diseases, thus arguing that most AJ disease loci are not under strong positive selection, but rather rose to their current frequency through genetic drift after a bottleneck. However, we cannot exclude the possibility that selection of some AJ disease loci are outside the limits of detection by the extended haplotype tests, which are known to have less power to detect selection of lower frequency alleles (38, 41).

Table 2.

Overlap of AJ disease loci and regions of selection

Disease	Carrier freq.	Gene	Selected region
AJ-prevalent disease
Bloom syndrome	1/100	BLM	No
Canavan	1/41	ASPA	No
Congenital adrenal hyperplasia	1/10	CYP21A2	Yes
Factor XI deficiency	1/19	F11	No
Familial dysautonomia	1/30	IKBKAP	No
Familial nonsyndromic deafness	1/25	GJB2	No
Fanconi anemia C	1/90	FANCC	Yes
Gaucher Type 1	1/18	GBA	No
Glycogen storage disease type 1a	3/200	G6PC	No
HMPS1 (colorectal cancer)	?	CRAC1	No
Mucolipidosis Type IV	1/50	MCOLN1	No
Niemann-Pick Type A	1/80	SMPD1	No
Tay-Sachs	1/30	HEXA	Yes
Torsion dystonia	1/2,000	DYT1	No
AJ-Prevalent alleles
Breast/Ovarian cancer	1/100	BRCA1	No
Breast/Ovarian cancer	1/100	BRCA2	Yes
Colorectal cancer	1/17	APC	Yes
Familial hypercholesterolemia	1/56	LDLR	No
Familial hyperinsulinism	1/89	ABCC8	Yes
HNPCC1 (colorectal cancer)	1/100	MSH2	No
Maple syrup urine disease	1/113	BCKDHB	No

Table compiled from Risch et al. (4), Kedar-Barnes and Rozen (56), Ostrer, H (1), and Charrow, J (57), showing disease and cancer susceptibility loci at increased frequency in the Ashkenazi Jews. The bottom portion of the table lists diseases for which specific mutant alleles are at higher frequency in the AJ population, although the overall disease incidence is similar in other populations. Being in a selected region was defined as falling in or near (within 500 kb of) the top 5% of iHS windows.

To uncover other loci that show evidence of stronger selection in either the AJ or EA populations, we identified regions found in the top 1% of iHS hits in one population but not the other (Fig. 4 and Table S5), as well as directly comparing populations using the XP-EHH test (Table 3). Both tests indicated that the strongest signal of selective pressure unique to Europeans was at the lactase locus, LCT, which we found to be under strong selective pressure in the EA population, but showed no signs of selection in our data for the AJ population. Multiple studies have found that the “lactase-persistence” allele at the LCT locus was selected for in Northern Europeans, with the selective sweep presumably occurring at the time of the domestication of cattle 2,000 to 20,000 y ago (42, 43). The absence of this allele in our data would suggest that the selective sweep was complete before the Ashkenazi establishment in Europe. Moreover, the prevalence of lactase deficiency in Ashkenazi Jews has been estimated at 60 to 80% (44), further corroborating the lack of selection for the LCT locus in the AJ population.

Table 3.

Top XP-EHH regions of selection

Chr: region (Mb hg18)	Max |XP-EHH|	Pop selected	Genes (number)
1: 160.36–160.54	4.53	EA	NOSIAP (1)
2: 134.63–137.29	13.36	EA	CCNT2, RAB3GAP1, LCT (13)
2: 237.19–237.34	4.62	EA	None (0)
3: 10.59–10.86	5.07	EA	SLC6A11, LOC285370 (2)
3: 11.59–11.71	5.07	EA	VGLL4 (1)
4: 30.66–31.01	4.86	EA	PCDH7 (1)
5: 56.95–57.29	4.65	AJ	None (0)
5: 112.64–113.18	4.73	AJ	MCC, YTHDC2 (2)
6: 31.35–32.47	5.95	AJ	TNF, CYP21A2, SLC44A4 (60)
6: 156.46–156.76	4.77	AJ	None (0)
8: 49.2–49.31	5.04	EA	None (0)
8: 52.12–52.36	4.74	EA	None (0)
8: 55.54–55.97	4.55	AJ	RP1 (1)
8: 80.25–80.73	4.69	EA	STMN2 (1)
11: 12.84–12.93	4.78	AJ	TEAD1 (1)
11: 79.65–79.79	5.05	AJ	None (0)
12: 109.84–111.85	5.95	AJ	ATXN2, ALDH2, TRAFD1 (18)
13: 98.06–98.21	4.54	EA	SLC15A1 (1)
15: 46.73–47.17	5.18	AJ	SHC4, EID1, SECISBP2L (4)
18: 73.91–74.11	4.48	AJ	None (0)

The XP-EHH test directly compared the AJ and EA populations for regions that show stronger selection in one population relative to the other. The top 10 selected regions for each population are reported. A subset of genes in each interval is listed, with the total number in that region in parentheses.

The strongest signal of selection unique to the AJ population was on chromosome 12 (110.6–111.72 Mb hg18). This locus was not found to be under selection in either the EA population or among reported selected regions of the HGDP Middle Eastern populations (40), making it an apparent Jewish-specific selected locus. Interestingly, this locus is also within a region that was recently identified as having high IBD across many Jewish Diaspora populations (13). This region contains 18 genes, including the mitochondrial aldehyde dehydrogenase gene, ALDH2, which is part of the major oxidative pathway of alcohol metabolism. Genetic variation within ALDH2 has been shown repeatedly to affect alcohol dependence (45). Intriguingly, the AJ population has long been known to have lower levels of alcoholism than other groups (16, 46), with one study showing that Jewish males have a 2.5-fold lower lifetime rate of alcohol abuse/dependence compared with non-Jews (47). In his analysis of the historical record, Keller (46) concluded that drunkenness was common among Jews before their Babylonian captivity but that drunkenness vanished around the time they returned to Israel in 537 B.C.E. This dramatic shift and persistence of low alcoholism rates in modern Jews has largely been attributed to social and religious practices (16, 46), with little support for a biological explanation. Our results, together with a recent study showing that variation in the ALDH2 promoter affects alcohol absorption in Jews (48), now suggest that genetic factors and selective pressure at the ALDH2 locus may have contributed to the low levels of alcoholism. The mechanism driving selection of the ALDH2 locus is unknown, but a plausible target of selection also within this selected region is the TRAFD1/FLN29 gene, which is a negative regulator of the innate immune system, important for controlling the response to bacterial and viral infection (49). TRAFD1/FLN29 may have conferred a selective advantage in the immune response to a pathogen, perhaps near the time that the Jews returned to Israel from their Babylonian captivity. Despite the unclear selective mechanism, this remains a remarkable example of a putatively selected region accounting for a known population phenotype.

Materials and Methods

More detailed methods are described in SI Materials and Methods and Dataset S1.

Genotype Data.

Unrelated Ashkenazi Jews were genotyped using the Affymetrix 6.0 array, with 471 individuals and 732K SNPs passing quality control filters. Genotype data for all other populations were obtained from previously published studies or publicly available sources.

Principal Component Analysis, F_ST, and Phylogenetic Tree Building.

PCA was performed using smartpca in the EIGENSOFT software package (v 3.0) (50, 51). An F_ST matrix was calculated using smartpca concurrently with the PCA analysis. The unrooted phylogenetic tree built using the F_ST matrix was created using the FITCH program in the PHYLIP package (v 3.69) (52).

Ancestral Clustering and Locus-Specific Admixture.

The frappe algorithm (v 1.0) (32) was used to determine the ancestral population clustering. The LAMPANC algorithm was used from LAMP (v2.3) (34) to calculate the locus-specific admixture given two ancestral populations.

Genetic Diversity.

Heterozygosity (HET) and inbreeding coefficients (F) were calculated using all nonmissing genotype calls. The pairwise IBS test was run in PLINK (53).

Linkage Disequilibrium.

The r² and D′ were calculated for all pairs of SNPs within 500 kb of each other in Haploview (v 4.1) (54).

Haplotype Phasing and Frequency Modeling.

BEAGLE (v 3.04) was used to phase haplotypes (55) and build a graphical model of haplotype frequency (26).

Identity-by-Descent.

Phased haplotypes from BEAGLE were analyzed with the GERMLINE program (v 1.4.0) (37) to identify segments of IBD (> 5 Mb, 150 SNPs).

Positive Selection.

The iHS and XP-EHH were implemented according to previous methods (38–40).