We investigated how the genetic variance, based on mtDNA haplotype frequencies (HVS-I sequences) was distributed in a hierarchical mode using an AMOVA analysis 6. The overall differentiation was low but statistically significant (Fst = 0.013; P < 0.000). Differences among ethnic groups explain about 0.6% (P < 0.001) of the overall variance. The comparison of Turkic populations versus Tajik Indo-Iranian populations showed that differences between these two groups constitutes 0.55% (P < 0.0283) of the total genetic variance. Intra-ethnic group genetic differentiation was significant for the Tajik group (Fst = 0.0197; P < 0.001) but not for the Turkic groups (0.3% P = 0.10). Differences among Turkic ethnic group was low but globally significant (0.66% P < 0.001). [See additional file 1].

TABLE 1

When taking into account all populations, no correlation between genetic and geographical distances was detected at the global level (Mantel test, r = -0.00682, P = 0.502). This lack of correlation remains if we test separately for each language family.

With respect to the Y chromosome, the AMOVA analysis performed using the 20 populations showed that about 5.6% of genetic differentiation is due to differences among ethnic groups (P < 0.02) and that the overall differentiation between populations is RST = 0.186 (P < 0.001). When populations were grouped by language affiliation/mode of subsistence -- Turkic versus Tajik -- ~9.1% of the genetic variability was due to differences between these two groups. In addition, the analysis at the intra-group level revealed a high degree of differentiation both for Tajik and Turkic populations except for the two Uzbek populations. [See additional file 1].

TABLE 2

A Mantel test of correlation between geographical and genetic distance was non-significant (r = -0.0145 p = 0.4755). Note: this test was based only on 19 populations since KRI-TY could not be assigned a precise geographical location (individuals were sampled in a military camp and come from several places in Kyrgyzstan). This lack of correlation remains if we test inside language family or even among a sub-region for one ethnic group (Kyrgyz).

TABLE 3

Kyrgyz, Kazak, Turkmen and Karakalpak have significantly lower effective population sizes than Tajik and Uzbek populations. Conversely, Uzbek and Tajik populations show higher growth rates but confidence intervals overlap the growth rates of other populations, except for the Kyrgyz when compared with the Uzbek. The date of the first split event is older than 1000 years except in the case of the Karakalpak, but confidence intervals are large.

We found evidence that overall, the Y chromosome has a significantly higher level of differentiation between populations than does mtDNA, in agreement with previous studies. The present study also shows that the level at which differentiation occurs is different between the two markers. There are more differences between populations of the same ethnic group than between ethnic groups for the Y chromosome, whereas the opposite is observed for mtDNA in the Turkic group. Indeed, no differences are observed in the Turkic group between populations belonging to the same ethnic group but there is a significant (although low) genetic differentiation between ethnic groups. This is not the case for Tajik populations where the mtDNA differentiation, like Y-chromosomal differentiation, is also significant between populations within this ethnic group.

Ethnologists describe the social organisation of Turkic populations as exogamous at the clan level or the lineage level (depending on the population) but endogamous at the tribe level - a man chooses his spouse outside the clan or his lineage but inside the tribe and inside his ethnic group. The geographical spread of a given tribe is wide 1, and this could explain the lack of mtDNA genetic differentiation between populations that are defined on geographical criteria. However, we would have expected stronger differences between ethnic groups. One explanation for our observations of low levels of maternal differentiation could be that ethnic groups are not actually highly endogamous. An ethnological study during our field expedition in Karakalpakia measured the level of endogamous mating at the tribe level among Karakalpak. Of 506 matings considered, 443 (87.5%) were among members of the Karakalpak ethnic group, and 78.5% among members of the same tribe 9. Thus, even if the ethnic group's endogamy is high (87.5%) from an ethnological perspective, it is low from a genetic point of view and insufficient to create high levels of genetic differentiation for mtDNA between these ethnic groups. An alternative explanation is that ethnic groups are a recent aggregation of tribes of different origins. This low level of endogamy, combined with an aggregation of unrelated tribes to form an ethnic group, leads to a low level of matrilineal genetic differentiation among ethnic groups. By contrast, Tajik populations are endogamous - a male tends to choose his spouse in the same village, and within the same family. This is shown by the significant Fst between Tajik populations for mtDNA. Further, the strong sex-specific difference in the pattern of genetic differentiation in Turkic populations (i.e. no mtDNA genetic differences between populations but strong Y-chromosomal differences within them) is explained by their strongly patrilineal social organisation. This type of organisation is absent in Tajik and explains the less sex-specificity in the genetic differentiation observed in this ethnic group (see Figure 1), consistent with no sex-specificity in the effective population size that has been demonstrated recently 5.

The combination of mtDNA and Y-chromosomal data from these large collections of populations and ethnic groups of Central Asia can shed light on the history of these groups. In addition, the Y-chromosomal analysis of genetic differentiation between populations belonging to the same ethnic group can give some estimation of the minimal age of these ethnic groups. The median estimate of the age of first split is always older than 1000 years (except for Karakalpak, for which it is 880 years). Actually, this estimation does not represent the age of the group sensu stricto, but the lower bound at which the group originated. In any case, this estimate is older than what is known from historical records for most of the Turkic ethnic groups, further, even if the confidence intervals are large, they do not overlap with historical estimates in two of the ethnic groups (and marginally three). Historical sources state that the Kazakh, Kyrgyz and Uzbek living in Central Asia arose in the sixteenth century. Genetic data show that populations belonging to one of these ethnic groups have an older common ancestor (more than one thousand years ago). Although these estimates are based on only one genetic system (linked Y chromosome microsatellites), we can propose that these ethnic groups are a heterogeneous conglomerate of tribes or populations. This hypothesis has been previously formulated in the case of Brahmin caste in India, whose subcastes seem to result from a fusion rather than a fission process 10. Such heterogeneous conglomerate of populations could have its origins at the foundation of the ethnic group or later during its history, as a result of the agglomeration of new unrelated tribes. The second hypothesis is compatible with historical records regarding the Uzbek and the Kyrgyz. Soucek 11 records that what is now called 'Uzbek' encompasses the seventeenth century Uzbek and former Chagatai Turk groups who were already settled in Uzbekistan. Therefore the name refers to a tribal union of different tribes including Chagatai Turks who were strongly mixed with Iranian dwellers of Central Asia. The same type of scenario is proposed by historians regarding the Kyrgyz living in Kyrgyzstan: this group is made up of Kyrgyz who arrived in the country in the fourteenth century and of Turkic groups who were already leaving in TienShan. The minimum age of the origin of the group is compatible with a common ancestry for the Turkmen group. This does not prove the common ancestry hypothesis, but does not refute it formally as for the other ethnic groups. In any case, additional sampling would certainly help to test these hypotheses, especially because our Turkmen group is composed of only two populations. Similar analyses based on mtDNA information are not feasible because of the high uncertainty in mtDNA mutation rate calibration and the near absence of genetic differentiation among populations belonging to the same ethnic group. Recent common ancestry or older common ancestry with high levels of gene flow are both possible explanations for this absence of mtDNA genetic differentiation. Despite the limitations associated with mtDNA data, our study shows that for the Turkic, there is a slight but significant mtDNA genetic differentiation between ethnic groups. This is consistent with the results on the Y chromosome revealing genetic differentiation between ethnic groups. The refutation of the common ancestry hypothesis for several of these ethnic groups, together with the observation of inter-group genetic differentiation, suggest that genetic boundaries separate them.

We combined our results with previous data published on the populations of Central Asia - [see additional file 2] for the list of populations and Figure 2 for their locations 47131415. For each village, our sampling strategy was to sample individuals who were more distantly related than the first and second degree, and belonged to the same ethnic group. Such samples are considered as "populations" in our study. Regarding data from the literature, the sampling strategy is not always precisely described and when the information was not published, we contacted authors to obtain more detailed information. Except for one Kyrgyz sample (not included in the geographical analyses), all samples followed the strategy adopted by us.

DNA was extracted from blood samples using standard protocols. Informed consent was obtained from all participants.

In order to determine how overall genetic diversity is distributed within and between populations, an analysis of molecular variance (AMOVA) was performed using Arlequin v 2.0 software 6. For mtDNA, the mutation model assumed was the Kimura 2-parameter model with a transition/tranversion ratio of 10 and an alpha (Gamma shape parameter) of 0.26. For the Y-linked microsatellites, we used the RST genetic distance 17, which takes into account the probability of recurrent mutation. We performed a global AMOVA analysis including all populations and also considering several groupings corresponding to the ethnic affiliation of populations. For the ethnic grouping, we divided populations into six ethnic groups: Karakalpak, Kazakh, Uzbek, Kyrgyz, Turkmen and Tajik. Correlations between genetic and geographical distances were performed using a Mantel test implemented in the R package 18.

Based on the generally high levels of population differentiation observed with the Y-chromosomal microsatellites we decided to perform a BATWING 19 analysis to estimate different population history parameters: (a) the population parameter θ for the populations altogether of the same ethnic group (2Mu, where u is the mutation rate 2021and M is equal to Ne - the effective size - for a uniparentally inherited gene and to 2Ne for a biparentally inherited gene); (b) the total growth rate; (c) the parameters of the population 'supertree', namely the dates of the splitting events, the identity of the populations that split and the proportional size taken up by each population.

The program assumes that the populations under study have diverged from an ancestral population at different points in time, have the same growth rate (growth or stationarity can be assumed) and have not exchanged migrants after the splits. The date of the first split represents the minimum age of the ethnic group. A generational interval of 30 years was assumed 22.