Most of the extant mtDNA boundaries in South and Southwest Asia were likely shaped during the initial settlement of Eurasia by anatomically modern humans

Background Recent advances in the understanding of the maternal and paternal heritage of south and southwest Asian populations have highlighted their role in the colonization of Eurasia by anatomically modern humans. Further understanding requires a deeper insight into the topology of the branches of the Indian mtDNA phylogenetic tree, which should be contextualized within the phylogeography of the neighboring regional mtDNA variation. Accordingly, we have analyzed mtDNA control and coding region variation in 796 Indian (including both tribal and caste populations from different parts of India) and 436 Iranian mtDNAs. The results were integrated and analyzed together with published data from South, Southeast Asia and West Eurasia. Results Four new Indian-specific haplogroup M sub-clades were defined. These, in combination with two previously described haplogroups, encompass approximately one third of the haplogroup M mtDNAs in India. Their phylogeography and spread among different linguistic phyla and social strata was investigated in detail. Furthermore, the analysis of the Iranian mtDNA pool revealed patterns of limited reciprocal gene flow between Iran and the Indian sub-continent and allowed the identification of different assemblies of shared mtDNA sub-clades. Conclusions Since the initial peopling of South and West Asia by anatomically modern humans, when this region may well have provided the initial settlers who colonized much of the rest of Eurasia, the gene flow in and out of India of the maternally transmitted mtDNA has been surprisingly limited. Specifically, our analysis of the mtDNA haplogroups, which are shared between Indian and Iranian populations and exhibit coalescence ages corresponding to around the early Upper Paleolithic, indicates that they are present in India largely as Indian-specific sub-lineages. In contrast, other ancient Indian-specific variants of M and R are very rare outside the sub-continent.

surprisingly limited. Specifically, our analysis of the mtDNA haplogroups, which are shared between Indian and Iranian populations and exhibit coalescence ages corresponding to around the early Upper Paleolithic, indicates that they are present in India largely as Indian-specific sub-lineages. In contrast, other ancient Indian-specific variants of M and R are very rare outside the sub-continent.
The overwhelming majority of the Iranian mtDNAs have been shown to lie in the West Eurasian domain of the global human mtDNA pool [27,28]. Here we focus on the analysis of mtDNA lineages that are shared between Indians and Iranians and bear signals of pre-Holocene expansion in the region.
India congregates four linguistic domains (Indo-European, Dravidic, Austro-Asiatic and Tibeto-Burman) that occupy non-random spheres of the geographic distribution of its populations. The majority of the recent studies based on mtDNA variation have, in contrast to some [21], provided evidence that linguistic groups of India do not represent genetically homogeneous units and are not, therefore, traceable to different immigration waves from distinct sources [8,13,19]. The complexity that arises in defining populations and groups of populations in India based on genetic and cultural criteria has been recently demonstrated in South Indian tribal and caste populations. The combined data from mtDNA, Y-chromosome and autosomal genes indicated that the tribes and castes derive largely from the same genetic heritage of Late Pleistocene southern and southwestern Asians, and have received limited gene flow from external sources since the Holocene [15]. Similar results were obtained by Cordaux et al. [29], who demonstrated that caste and tribal groups exhibit similar levels of molecular variance. However, genetic distances indicated that the Tibeto-Burman speaking tribal populations (from eastern India) were more closely related to East Asians than to other Indians [29]. This is consistent with an earlier suggestion placing the origin of these tribal groups east of India -in Tibet and Myanmar [30].
In this study, we have analyzed the mtDNA variation in a sample of 796 Indians and 436 Iranians (Table 1), and combined the results with previously published data from the same geographic area. We also compared the mtDNA variation in India and Iran with that of Europe, China, and Thailand. The overall aim was to improve our understanding of the origins and composition of the Indian and Iranian gene pools and to determine the nature and the extent of gene flow between these regions. Through the analyses of the genetic variation of extant Southwest and South Asian populations we took an endeavor to envisage the exodus of anatomically modern humans from Africa. the southern states, and peaked at 86% in West Bengal (Table 8, see Additional file 3).
With the exception of the diverse set of largely Indian-specific R lineages, the most frequent mtDNA haplogroup in India that derives from the phylogenetic node N is haplogroup W [13]. The frequency peak of haplogroup W is 5% in the northwestern states -Gujarat, Punjab and Kashmir. Elsewhere in India its frequency is very low (from 0 to 0.9%) ( Table 2) forming a significant spatial cline ( Figure  4).
At 15% among the caste and 8% among the tribal populations haplogroup U is the most frequent sub-clade of R in India (Table 12, see Additional file 7). Approximately one half of the U mtDNAs in India belong to the Indianspecific branches of haplogroup U2 (U2i: U2a, U2b and U2c) [13,27] (Table 2). They are present throughout India without a clear geographical cline (Figure 2, panel U2i, SAA p > 0.05). However, the spread of another subset of U, haplogroup U7 [13], is similar to that of haplogroup W, peaking at 12% and 9% in Gujarat and Punjab, respectively (Table 11, see Additional file 6). The frequency of U7 is also high in neighboring Pakistan (6%) and particularly in Iran (9%) ( Table 9, see Additional file 4).

MtDNA haplogroups in Iran
Over 90% of the mtDNAs found in Iran belong to haplogroups HV, TJ, U, N1, N2 and X, commonly found in West Eurasia (Table 2). In contrast to Europe, where H is predominant among the mtDNA haplogroups, in Iran the frequency of haplogroup U (29%) is higher than that of haplogroup H (17%) ( Table 9, see Additional file 4). This difference accounts, at least partly for the presence in Iran of U sub-groups, such as U7 (9.4%), that are virtually absent in Europe.
Compared to India, haplogroup M frequency in Iran is marginally low (5.3%) and there are no distinguished Iranian-specific sub-clades of haplogroup M. All Iranian haplogroup M lineages can be seen as derived from other regional variants of the haplogroup: eleven show affiliation to haplogroup M lineages found in India, twelve in East and Central Asia (D, G, and M8) and one in northeast Africa (M1).  Table 1 in the Supplementary Material for the complete set of studied populations including those whose mtDNA variation was previously published) The spatial distribution of Indian-specific mtDNA haplogroups (M2, M6 and M3) and their sub-haplogroups Figure 1 The spatial distribution of Indian-specific mtDNA haplogroups (M2, M6 and M3) and their sub-haplogroups.
Maps of South and Southwest Asia depicting the spatial frequency distribution of various Indian-specific mtDNA haplogroups. For India, the tribal populations were excluded (see 'Methods' for explanation). The published (full reference in Table 6, see Additional file 1) and new data were averaged to the resolution of states in India, geographic regions in Iran and provinces in China and Thailand. Numbers adjacent to the data points indicate sample sizes. In the case of haplogroup M3a some datasets had to be excluded because the discrimination between M* and the respective subgroup was not possible on the basis of the HVS-I data alone (see footnote of Table 3).
The spatial distribution of Indian-specific mtDNA haplogroups (U2, M4a, M18 and M25) and their sub-haplogroups Figure 2 The spatial distribution of Indian-specific mtDNA haplogroups (U2, M4a, M18 and M25) and their sub-haplogroups. In the case of haplogroup M25 some datasets had to be excluded because the discrimination between M* and the respective subgroup was not possible on the basis of the HVS-I data alone (see footnote of Table 3). For other details see legend to Figure 1.
Indian-specific (R5 and Indian-specific M and U2 variants) and East Asian-specific (A, B and East Asian-specific M subgroups) mtDNAs, both, make up less than 4% of the Iranian mtDNA pool. We used Turkey (88.8 ± 4.0%) as the third parental population for evaluating the relative proportions of admixture from India (2.2 ± 1.7%) and China (9.1 ± 4.1%) into Iran. Therefore we can conclude that historic gene flow from India to Iran has been very limited.

The package of the most ancient mtDNA haplogroups in India
Approximately one tenth of the Indian haplogroup M mtDNAs fall into its major sub-clade M2, which is defined by the motif 477G-1780-8502-16319 [15]. M2 can be further subdivided into haplogroups M2a (transitions at nps 5252 and 8369) and M2b [15]. Haplogroup M2 and its two major sub-clades reveal coalescence times of 50 to 70 thousand years (Table 3). Due to the increased frequency towards the southern part of India ( Figure 1, panel M2, SAA p < 0.05 Figure 4), M2 is significantly (p < 0.05) more frequent among the Dravidic speakers than among the Indo-European speakers who are spread mostly in the northern regions of India (Table 2). It is more plausible that geography rather than linguistics is behind this pattern, because the frequency of M2 amongst the Indo-European speaking populations in southern India is significantly higher than that in the north, while there is no significant difference between Dravidic and Indo-European speaking populations from the same geo-graphic region ( We found that R5, which is defined by transitions at nps 8594 [27], 16266 and 16304, is the second most frequent sub-clade of R in India after haplogroup U. The coalescence age estimate for R5 was similar to that of M2 (Table  3), whereas individual boughs within the R5 limb showed expansions from ca. 20,000 ybp to ca. 50,000 ybp ( Figure  12, see Additional file 8). Our data indicate that this diverse and ancient haplogroup is present over most of India ( Figure 3, panel R5). Though absent among the Austro-Asiatic tribal groups, R5 is spread across the other language boundaries. In contrast to M2, R5 is more frequent among the caste (CR: 2.5-4.5%) than among the tribal populations (CR: 0.7-1.9%) ( Table 2).
Together with the Indian-specific sub-clades of haplogroup U2 [13,27], haplogroups M2 and R5 can be discriminated as a package of Indian-specific mtDNA clades harboring extremely deep coalescence times (around 50,000 -70,000 ybp). Together they constitute nearly 15% of the Indian mtDNAs. Importantly, these haplogroups are virtually absent elsewhere in Eurasia [13,15], this study]. Because most of Indian varieties of haplogroup M are still unclassified (M*), this package is likely to be extended when more mtDNA coding region information will become available for the M* lineages in India.
The spatial distribution of Indian-specific mtDNA haplogroups (R5 and R6) and West Eurasian-specific U7, W and R2 in South and southwest Asia Figure 3 The spatial distribution of Indian-specific mtDNA haplogroups (R5 and R6) and West Eurasian-specific U7, W and R2 in South and southwest Asia. For other details see legend to Figure 1. Figure 4 Spatial Autocorrelation Analyses Correlograms of different haplogroups and haplogroup groupings frequencies in South Asia. In the case of haplogroup M3a some datasets had to be excluded because lack of resolution (see footenote for Table 3. for detailes). There was no significant cline in the frequencies of haplogroup R2. In the case of haplogroups M6a and M6b only the potential cline along the Bay of Bengal is investigated. Therefor the number of distance classes is reduced to three. In contrast to haplogroup M, ancient sub-clades of haplogroup N are spread both east and west of India as well as within India itself. Several migration scenarios involving multiple "out of Africa" events punctuated by space, time or both, could be invoked to explain the phylogeography of these mtDNA haplogroups. Yet, using the parsimony criterion it can be argued that only a single early migration that brought ancestral lineages, M and N (with the latter having already given rise to R), to South Asia could account for the extant mtDNA phylogeography in Eurasia [15]. The finding of several largely South and West Asianspecific sub-clades (H, L, R2, and F*) of the major Eurasian Y chromosomal haplogroups F and K also supports this scenario [15]. From South and West Asia the colonization would have sprung both east and west as region-specific mtDNA and Y chromosomal sub-clades appeared both in West-and East-Eurasia as well as in India itself ( Figure 5). However, not all the West-Eurasian Y-chromosomal founder haplogroups are present in India. Haplogroup E, for example, was possibly carried to Europe and Western Asia later via the Levantine corridor [33]. Similarly, a Late Upper Palaeolithic origin and spread of mtDNA haplogroup X from Northeast Africa and Middle East has been suggested lately [34].

The improved structure of autochthonous Indian mtDNA clades
Nearly a third of Indian mtDNAs belonging to haplogroup M could be assigned to its existing boughs and limbs with the current knowledge of the mtDNA coding region polymorphisms (Table 9, see Additional file 4). It is likely that the unclassified Indian M* and R* mtDNAs are also to a large extent autochthonous because neither the East nor West Eurasian mtDNA pools include such lineages at notable frequencies.
Haplogroup M6 ( Figure 6) is primarily found in the Indus Valley and on the western shores of the Bay of Bengal where its sub-clades M6a and M6b are concentrated towards the southwest and the northeast, respectively ( Figure 1, panel M6, M6b cline is significant SAA p < 0.05, Figure 4). The highest frequencies of M6a and M6b were found amongst the Mukri scheduled caste from Karnataka (17%) and in Kashmir (10%), respectively. The Mukri form an endogamous group of no more than 10,000 individuals, who dwell on an area less than 2000 km 2 [35]. That, together with the observation that all the sixteen M6 sequences found among the Mukri belong to a single haplotype, suggests that genetic drift has played a major role in the demographic history of the Mukri. The statistical significance of the high M6 frequency in Kashmir is undermined by the small sample size (19 individuals), which results in the very wide error margins for the frequency estimate (CR: 3.2-31.7%).
Different geographic distributions characterize the subclades of M3 and M4 that we define by mtDNA coding region markers (Table 3 Awaiting further information from complete mtDNA sequences, we defined haplogroup M18 by using the transversion at np 16318. This star-like cluster (Figure 9) is spread at low frequencies across India, with the exceptions of the very north and the coast of the Arabian Sea ( Figure 2, panel M18). The high incidence (33%) of the M18 nodal haplotype among the Austro-Asiatic speaking Peopling of Eurasia Figure 5 Peopling of Eurasia. Map of Eurasia and northeastern Africa depicting the peopling of Eurasia as inferred from the extant mtDNA phylogeny. The bold black arrow indicates the possible "coastal" route of colonization of Eurasia by anatomically modern humans (ca. 60,000 -80,000 ybp.). This "Southern Coastal Route" is suggested by the phylogeography of mtDNA haplogroup M, the virtual absence of which in the Near East and Southwest Asia undermines the likelihood of the initial colonization of Eurasia taking a route north around the Red Sea. Therefore, the initial split between West and East Eurasian mtDNAs is postulated between the Indus Valley and Southwest Asia. Spheres depict expansion zones where, after the initial (coastal) peopling of the continent, local branches of the mtDNA tree (haplogroups given in the spheres) arose (ca. 40,000 -60,000 ybp), and from where they where further carried into the interior of the continent (thinner black arrows). Admixture between the expansion zones has been surprisingly limited ever since. We note that while there is no obvious need to introduce the "northern route" -from northeast Africa over Sinai to the Near East -to explain the initial colonization of Eurasia, the spread of some mtDNA and Y-chromosomal haplogroups implies that the "northern" passage might have been used in a later period [33,34].
Network of HVS-I haplotypes belonging to haplogroup M6  Table 6 (see Additional file 1). Coalescence estimates marked with an asterisk are calculated excluding tribal populations (see Materials and Methods for explanation). The coalescence estimate marked with two asterisks is calculated without the data on tribal and scheduled caste (the Mukri) populations (see text for details).
Lodha of West Bengal (Table 10, see Additional file 5) suggests a possible founder effect in this population. This explains the nearly two-fold difference between the coalescence estimates for this cluster calculated with and without the tribal data ( Table 3).
The G to A transition at np 15928 has been spotted on different branches (e. g. haplogroups T and M) of the mtDNA phylogeny [3,36]. Quintana-Murci and colleagues observed this transition within haplogroup M in combination with the HVS-I motif 16048-16129-16223-16390 [27]. None of the mtDNAs in our study which har-  Figure 10). In this case, monophylecity is the most parsimonious assumption. This haplogroup is moderately frequent in Kerala and Maharashtra but rather infrequent elsewhere in India (Figure 2, panel M25).
Several Indian-specific mtDNA clades demonstrate a similar spread-pattern in southern India. We found haplogroups M4a, M6a and M18 in southeastern Tamil Nadu and Andhra Pradesh while they were absent from neighboring Karnataka and Kerala (Figure 1 panel M6a and Figure 2 panels M4a and M18). One possible explanation is that admixture has been facilitated along the coastlines of the Arabian Sea and the Bay of Bengal. On the other hand, because the absolute frequencies of these haplogroups are rather low, it cannot be ruled out that an increase of sample sizes would disrupt the observed spread-pattern.

Were the Austro-Asiatic speaking tribal people the earliest inhabitants of India?
By calculating nucleotide diversities and expansion times (using the method from [37]) for different linguistic groups of India, some previous studies on mtDNA variation have distinguished the Austro-Asiatic speaking tribal groups as the carriers of the genetic legacy of the earliest Network of HVS-I haplotypes belonging to haplogroup M4a Figure 8 Network of HVS-I haplotypes belonging to haplogroup M4a. Underlined samples are those in which the diagnostic coding region marker (6620) of M4a was assayed by either RFLP analysis (+6618 MboI) or direct sequencing. For other details, see the legend to Figure 6.
settlers of the subcontinent [17,38]. However, because the linguistic groups of India do not cluster into distinct branches of the Indian mtDNA tree [13,15,19], this study], calculating the beginning of expansion for those groupings is problematic and likely controversial as well.
Recently, Basu et al. (2003) supported the conclusions of [17,38] by reporting that the frequency of the ancient haplogroup M2 among the Austro-Asiatic tribal populations is as high as 19%, and that they lack the slightly younger haplogroup M4 (the likely paraphyletic mother-clade of M4a). The authors have regarded the HVS-I transition at np 16319 as sufficient in defining haplogroup M2. This assumption, however, might lead to an overestimation of M2 frequency and age. Indeed, the 16319 transition has arisen several times on the background of other Indian haplogroup M lineages (Table 7, see Additional file 2), more specifically, in mtDNAs lacking the coding region markers that define M2 [15].  Language families present today in India, such as Indo-European, Dravidic and Austro-Asiatic, are all much younger than the majority of indigenous mtDNA lineages found among their present-day speakers at high frequencies (see Additional file 9). It would make it highly speculative to infer, from the extant mtDNA pools of their speakers, whether one of the listed above linguistically defined group in India should be considered more "autochthonous" than any other in respect of its presence in the subcontinent.
Additionally, we note that some recent linguistic and archaeological evidence place the spread of the Austro-Asiatic languages in the Neolithic, in conjunction with the dispersal of rice cultivation from the Yangtze River basin [39]. If this were the case, it would imply that the arrival of this linguistic phylum in India was not associated with female gene flow.

Gene flow from West Eurasia
Broadly, the average proportion of mtDNAs from West Eurasia among Indian caste populations is 17% (Table 2).
In the western States of India and in Pakistan their share is greater, reaching over 30% in Kashmir and Gujarat, nearly 40% in Indian Punjab, and peaking, expectedly, at approximately 50% in Pakistan (Table 11, see Additional file 6, Figure 11, panel A). These frequencies demonstrate a general decline (SAA p < 0.05 Figure 4) towards the south (23%, 11% and 15% in Maharashtra, Kerala and Sri Lanka, respectively) and even more so towards the east of India (13% in Uttar Pradesh and around 7% in West Bengal and Bangladesh). The low (<3%) frequency of the western Eurasian mtDNAs in Rajasthan may be in part a statistical artifact due to the limited sample size of 35 Rajputs.
In comparison to an overall frequency of 17% (CR 15.1-19.3%) among the caste populations, only 7% (CR 6.2 -9.0%) of the mtDNAs from the tribal groups show affiliation to the West Eurasian haplogroups. The observed difference could be caused by differences in the extent of gene flow from the west to different social layers of the Indian society [19], and/or a more pronounced genetic drift among the tribal groups.
This western Eurasian contribution into the Indian maternal gene pool can be broadly divided into two different components. Over two thirds of the West Eurasian-specific mtDNAs found in India are made up by haplogroups HV, TJ, N1 and West Eurasian-specific branches of haplogroup U. It is likely that these mtDNA haplogroups have been carried to western India both by relatively low-intensity long-lasting admixture at the border regions as well as a consequence of numerous but probably limited migrations during the last 10, 000 ybp [13]. The remaining one third of the West Eurasian-specific mtDNAs found in India is comprised of haplogroups U7, R2 and W showing much deeper time depths in India -approximately forty thousand years before present ( Similarly to HV, TJ, N1, the spread of U7, W and R2 in India is geographically uneven -the three haplogroups are much more frequent in the northwestern states (SAA p < 0.05, Figure 4). Together, they constitute nearly 14% of the maternal gene pool in Indian Punjab, Gujarat and Rajasthan. Their frequency is also high in Iran (13%) and Pakistan (10%) but declines in Central Asia (5%).
Although haplogroup W is not highly frequent in European populations, it is nevertheless quite common [9], reaching the highest frequencies among the central-northern Finns (9% [40]). Yet, it is virtually absent from the Finno-Ugric speaking populations of the Volga basin [41]. In Central Asia the frequency of haplogroup W stays below 2%.
Many European populations are lacking haplogroup U7 [9], but its frequency climbs over 4% in the Near East and up to 5% in Pakistan, reaching nearly 10% level in Iranians (Table 9, see Additional file 4). In India, haplogroup U7 frequency peaks at over 12% in Gujarat, in the westernmost state of India, while for the whole of India its frequency stays around 2%. Expansion times and haplotype diversities for the Indian and Near and Middle Eastern U7 mtDNAs are strikingly similar ( Table 4). The possible homeland of this haplogroup spans likely in Gujarat and Iran because from there its frequency declines steeply both to the east and to the west. If the origin were in Iran rather than in India, then its equally high frequency as well as diversity in Gujarat favors a scenario whereby U7 has been introduced to the coastal western India either very early, or by multiple founders. Notably, the overlap of the Indian and Iranian lineages is largely restricted to the ancestral nodes while the coalescence age estimates for the nearly exclusively Indian (16207-16309-16318T) and West Eurasian (16126-16309-16318T) founder HVS-I motifs both yield time depths of about 20,000 to 30,000 ybp.
Haplogroup R2 appears at low frequencies in Near and Middle East and India and is virtually absent elsewhere. The spread of haplogroup R2 in Europe is restricted to a few populations in the Volga basin where it is represented by nodal haplotypes and by a region-specific subclade characterized by the HVS-I motif 16037-16172 [41]. The coalescence estimate of this sub-clade is 11,400 ± 9,000 ybp. However, its wide error range prevents us from drawing any firm conclusions.
To summarize, the West and South Asian phylogeography of haplogroups W, U7 and R2 can be viewed as a genetic continuum that spans from the Near East into India, The segregation of West Eurasian, East Eurasian and South Asian mtDNA pools extending north into Central Asia. The coalescence times of these haplogroups suggest that this continuum took shape somewhere between 30,000 to 50,000 ybp (Table  4), thus falling within the climatically favorable interglacial period. We notice that the extant U7 and W frequencies along the proposed continuum are not uniform. U7 is more predominant in Iran, Pakistan, northwestern India and the Arabian peninsula, while W is more frequent in the western Near-East, Anatolia and the Caucasus. The coalescence ages of the Indian-and Iranianspecific U7 clades suggest that the time-window of this continuum was closed by ca. 20,000 ybp. The inferred extreme aridity of eastern Iran and western India during the last glacial maximum, which is well documented in paleovegetation reconstructions [42] may explain the observed segregation.
It has been suggested that the Jews settled in southwest India on the coast of the Arabian Sea sometime during the early Middle Ages. However, the mtDNA pool of the extant Cochin Jews is overwhelmingly Indian-specific (Table 10, see Additional file 5). We found exact or close matches to the fourteen HVS-I haplotypes observed among the Cochin Jews in other Indian populations. It is not clear whether the Near Eastern mtDNA lineages have been lost or the initial Jewish settlers did not include women.

Gene flow from East Eurasia
The East Eurasian-specific mtDNA haplogroups are less common in India and more sharply geographically segregated than the haplogroups of western Eurasian ancestry (Table 2; Figure 11, panel C). Indian caste populations harbor only about 4% of such mtDNAs, compared to 17% of the West Eurasian ones (Table 2). Elevated frequencies of haplogroups common in eastern Eurasia are observed in Bangladesh (17%) and Indian Kashmir (21%) and may be explained by admixture with the adjacent populations of Tibet and Myanmar (and possibly further east: from China and perhaps Thailand). On the other hand, the high frequencies of East Eurasian-specific mtDNAs found in the southern Indian state of Tamil Nadu (21%) are unexpected when considering their relatively low frequen-cies (~1%) in West Bengal and Andhra Pradesh. We notice, however, that the haplogroup assignments used here for the Tamil Nadu sample (A4, B4, F1a and M7) ( Table 7, see Additional file 2) are based on HVS-I sequences alone [29]. As shown and discussed elsewhere [7], such type of assignment is prone to mistakes.
Two varieties of haplogroup M, D4c and G2a, were recently identified as largely specific to Central Asia [43].
In spite of geographical proximity we did not find these haplogroups in northern or northwestern India. Haplogroup G2a did, though in marginally low frequency, turn up in Iran (CR: 0.1 -1.6%) and in southern Indian states Andhra Pradesh (CR: 0.3 -2.9%) and Sri Lanka (CR: 0.2 -4.1%).
Tibeto-Burman speaking tribal populations of eastern and northern India exhibit the highest frequencies of East Eurasian-specific mtDNA haplogroups. As inferred from the published HVS-I sequences [29], their share sums up to approximately two thirds of mtDNAs among the tribal groups in Assam, Nagaland, Arunachal Pradesh and Tripura, (Table 8, see Additional file 3; Figure 11, panel D). MtDNA haplogroups native to East Eurasia are also highly frequent in the northern states of India, reaching a peak of nearly 50% among the Kanet of Himachal Pradesh. Papiha and colleagues have previously demonstrated through the typing of immunoglobin allotypes that the Tibetan admixture among the regional Kanet groups decreases as the distance from the Tibetan border increases [44]. Thus, mtDNA data are consistent with an ancestral origin of the Tibeto-Burman speaking tribal populations outside (east of) India in the neighboring Tibet and Myanmar [30,45].

Haplotype sharing between populations
The majority (70%) of the 1136 mtDNA haplotypes found among continental Indians (including Pakistan and Bangladesh) are singletons and 41% of those that occur more than once are restricted to a single population. Only a few haplotypes are shared among five or more populations. The number of shared haplotypes between pairs of social, linguistic and geographic groups of Indian populations is slightly (but in most cases insignificantly) lower than that between random groups of Indian populations taken for reference (see Materials and Methods). Where the decline of shared haplotypes is significant relative to the reference, it is most probably caused by large differences in the sample sizes of the groups under comparison (Table 5).
An alternative method that assesses the degree of haplotype sharing between populations is to investigate the combined frequency of the shared haplotypes in two population groups. Thus, amongst the northern and the southern population groups the combined frequency of the haplotypes present also in the other group is significantly lower than that which we observed in the case of random groups. This is not surprising because West Eurasian-specific mtDNA haplogroups are rather frequent in northwest India. Because the Indo-European and the Dravidic speakers of India are largely concentrated to the northern and southern parts of the subcontinent, respectively, the differences arising from geographic division of the Indian populations also correspond to these linguistic groupings (Table 5).

Conclusions
Three Indian-specific haplogroups, M2, U2i and R5, which encompass about 15% of the Indian mtDNA pool, exhibit equally deep coalescence ages of about 50,000 -70,000 years. Thus, their spread can be associated with the initial peopling of South Asia.
Haplogroups U7, W and R2 harbor a number of similar traits. Their overlapping geographic distributions and coalescence times suggest some degree of genetic continuum in the area spanning from the Near and Middle East through northwest India and reaching north into Central Asia. The coalescence estimates for these haplogroups are equally deep (around 30,000 -50,000 years) in these different regions. That may be a result of either relatively more recent albeit large in scale migrations that brought along most of the diversity or may indeed reflect the region-specific expansions of these haplogroups. The former explanation could be ruled out since it is impossible to envisage a substantial movement of mtDNAs from South Asia that would not include haplogroup M. The same is true for the opposite -the share of U7, W and R2 within the West Eurasian-specific mtDNA haplogroups is two-fold higher in India than it is in Iran. Moreover, the South-and West Asian-specific sub-branches of haplogroup U7 predate the last glacial maximum. Therefore, deep autochthonous history of these haplogroups in the region remains to be the most parsimonious explanation.
Through the use of mtDNA coding region markers, we were able to classify altogether a quarter of the Indian M and R mtDNAs into a number of Indian-specific mtDNA haplogroups, four of which we newly identified. Several of these are characterized by clear patterns in their geographic distribution and/or different frequencies among different socio-cultural groups of India. Additional efforts should be undertaken to identify new coding region markers in order to further improve the classification of Indian mtDNAs. Indeed, much of the information is still hidden by the poor resolution of the Indian mtDNA tree and further piling of HVS-I datasets would add little to deepen our understanding of the demographic history of South and Southwest Asia.
We found that haplogroup M frequency drops abruptly from about 60% in India to about 5% in Iran, marking the western border of the haplogroup M distribution. A similarly sharp border cuts the distribution of Indian-specific mtDNA haplogroups to the east and to the north of the subcontinent. We therefore propose that the initial mtDNA pool established upon the peopling of South Asia has not been replaced but has rather been reshaped in situ by major demographic episodes in the past and garnished by relatively minor events of gene flow both from the West and the East during more recent chapters of the demographic history in the region.

Subjects
MtDNA sequence variation in a total of 796 Indian samples most of which are held in a collection at Newcastle University [46] was analyzed. The samples cover a wide geographical range that spans from Himachal Pradesh in the north, Sri Lanka in the south, West Bengal in the east and Gujarat in the west (Table 1).
Tribal populations constitute 15% of the total sample size. The Lodha (n = 56) live mostly in the western part of Midnapore district of West Bengal where they are also known as Kheria or Kharia. Their total population size was ~59000 according to the 1981 census. Their language belongs to the Mundari branch of the Austro-Asiatic language family [47]. The Kanet (n = 37) make up two thirds of the ~50000 inhabitants of the Kinnaur district in Himachal Pradesh [48]. Their language belongs to the Himalayan group of the Tibeto-Burman language family [47]. Five Bhoksa and twenty-six Tharu individuals were included in the present study in addition to those from the same populations that we have previously reported [13].  Table 1).
The Iranian sample of 436 individuals was collected in different locations mainly from southwestern and northwestern Iran ( Table 1).
An even wider range of data was used for the production of the isofrequency maps: the published data from Armenia (n = 192) [32] and Kyrgyz (n = 95) [57], and the unpublished data from Yemen (n = 118), Oman (n = 79), Saudi Arabia (n = 205), Ethiopia (n = 270), Uzbekistan (n = 114) and Tajikistan (n = 41). Further details about the mtDNA variation of these populations will be published elsewhere.

MtDNA molecular analyses
DNA was extracted using standard phenol-chloroform methods [58]. The hypervariable segment (HVS)-I (between nucleotide positions (np) 16024-16400) of the control region was sequenced in all the 796 Indian and 436 Iranian samples. Preparation of sequencing templates was carried out following Kaessmann et al. [59]. Purified products were sequenced with the DYEnamic™ ET terminator cycle sequencing kit (Amersham Pharmacia Biotech) and analyzed on ABI 377 DNA or Megabace2000 Sequencers. Sequences were aligned and analyzed with the Wisconsin Package (GCG). In addition, informative mtDNA coding region positions [3,11,27] were assayed ( Table 7, see Additional file 2) in selected individuals from different HVS-I haplotypes to determine haplogroup affiliations.

Data analysis
Median networks [60,61] were constructed using Network 3.111 and Network 2.10B programs [62] with default settings (r = 2; ε = 0). MtDNA coding region markers were given five times the weight of the HVS-I positions.
Coalescence of mtDNA haplogroups and sub-haplogroups was calculated using ρ (the averaged distance to a specified founder haplotype) and a mutation rate of one transition per 20,180 years between nps 16090-16365 [63]. Standard errors for coalescence estimates and efficiency of a sample for coalescence time calculation (ρ/ nδ 2 ) were calculated following [64]. Coalescence times were further calculated with the exclusion of tribal populations sequences, yet preserving the cluster topology (coalescence times marked with an asterisk). A more intense genetic drift (particularly founder effects) could introduce a bias into the coalescence time calculation (for example, see below the coalescence time of haplogroup M18 with and without the Lodha sample).
The software kindly provided by Vincent Macaulay was adopted in order to calculate the 95% credible regions (CR) from the posterior distribution of the proportion of a haplogroup/sub-haplogroup in the population.
Haplotype diversity was estimated as where n is the number of sequences, k the number of distinct haplotypes, and n i number of sequences with one distinct haplotype.
Haplogroup isofrequency maps were generated using the Kriging method in Surfer 7 program of Golden Software. Haplogroup frequencies were averaged over populations from the same state in India, provinces in China and Thailand and geographic divisions in Iran. The data points for Kriging are shown as black dots, while the sample size applicable to the data point is given adjacent to the dot. In relatively small and isolated groups (e.g. tribal groups) random genetic drift might seriously affect the haplogroup frequencies, which may become uninformative when a whole region (e.g. state) is considered (e.g. M18 among the Lodha, see below). Therefore, the tribal data were excluded from the haplogroup isofrequency maps calculation. When illustrating the spread of mtDNA haplogroups native to West Eurasia, East Eurasia and India ( Figure 11, panel D) we present these data as pie diagrams. The respective sample size and origin are indicated adjacent to the diagrams.
Spatial autocorrelation analysis was done using the PAS-SAGE software packet [65,66]. The correlograms of Moran's I autocorrelation coefficient were calculated using binary weight matrix with five distance classes. Data on tribal populations was not used (see previous paragraph for explanation).
Haplogroup frequencies based admixture proportions were calculated using the ADMIX 2.0 software [67].
Haplotypes were defined as the HVS-I motif combined with the haplogroup label for haplotype analysis. In this way it was possible to discriminate between mtDNAs with identical HVS-I motifs but otherwise known to belong to different haplogroups (e.g. sequences with the CRS motif in HVS-I belonging to haplogroups H or R). In order to assess the results of haplotype sharing analysis between population groups (e.g. tribal and caste populations) we divided the Indian populations ten times randomly into two sets and analyzed the level of haplotype sharing between these sets.