Mitochondrial genome evidence reveals successful Late Paleolithic settlement on the Tibetan Plateau

Abstract

Due to its numerous environmental extremes, the Tibetan Plateau—the world's highest plateau—is one of the most challenging areas of modern human settlement. Archaeological evidence dates the earliest settlement on the plateau to the Late Paleolithic, while previous genetic studies have traced the colonization event(s) to no earlier than the Neolithic. To explore whether the genetic continuity on the plateau has an exclusively Neolithic time depth, we studied mitochondrial DNA (mtDNA) genome variation within 6 regional Tibetan populations sampled from Tibet and neighboring areas. Our results confirm that the vast majority of Tibetan matrilineal components can trace their ancestry to Epipaleolithic and Neolithic immigrants from northern China during the mid-Holocene. Significantly, we also identified an infrequent novel haplogroup, M16, that branched off directly from the Eurasian M founder type. Its nearly exclusive distribution in Tibetan populations and ancient age (>21 kya) suggest that M16 may represent the genetic relics of the Late Paleolithic inhabitants on the plateau. This partial genetic continuity between the Paleolithic inhabitants and the contemporary Tibetan populations bridges the results and inferences from archaeology, history, and genetics.

The Tibetan Plateau is characteristic of most extreme environmental conditions, with high absolute elevation, low temperature, extreme aridity, and hypoxia. Nonetheless, modern humans settled on this plateau by the Paleolithic Age. A number of Paleolithic sites excavated throughout the Tibetan Plateau have been dated to >20 thousand years ago (kya) [Fig. 1 and supporting information (SI) Table S1] (1–3), documenting the earliest human presence on the plateau well before the last glacial maximum (LGM, 22–18 kya). In contrast, evidence from classical genetic studies on the contemporary indigenous Tibetan population argues for a northern East Asian origin during the Neolithic (4), a scenario that seems compatible with the available historic records. According to the Xin Tang Shu (New Tang Annals; 11th century A.D.), proto-Tibetans (“Bo” people) can in fact trace their ancestry to the Di-Qiang, an ancient tribe that resided in northwest China about 3 kya (5). One possibility is that the Late Paleolithic settlers might have been eliminated due to exacerbated environmental conditions during the LGM or the Younger Dryas (12.8–11.6 kya), or were largely, if not completely, replaced by the Neolithic immigrants. This notion receives some support from archaeological observations; in particular, the main type of Neolithic tools excavated on the plateau, microliths, show typical features of the northern Chinese tool culture (6). However, these microliths also display some characteristics of the Tibetan paleoliths (7, 8). This mosaic feature raises another possibility that the Neolithic immigrants had received some contribution from the Paleolithic settlers through either cultural or demic contact.

Fig. 1.

Sampling locations of the Chinese populations under study and of the excavated archaeological sites summarized from the literature. Codes 1–72 label the populations (with solid pentacles labeling locations of Tibetan populations collected in this study; see Table S2 for details). Solid triangles refer to excavation sites dating to pre-LGM, open triangles represent sites with Paleolithic age based on lithic comparison, and inverted triangles indicate post-LGM sites (see Table S1 for more details).

Based on the genetic evidence obtained so far from Y chromosome (9, 10) and mitochondrial DNA (mtDNA) (11–13) data, the majority of Tibetan genetic components can trace their origins to the Neolithic immigrants from northern East Asia. No solid genetic evidence indicates the existence of any ancient genetic relics from Paleolithic settlers. Nearly all of the Y chromosome markers in Tibetans analyzed recently (14) are indeed suggestive of more recent genetic inflow, except for the paragroup O3a5*-M134 (comprising the O3a5-M134 Y chromosomes not belonging to O3a5a-M117) which has a more ancient age of 22 kya. The high frequency of haplogroup D-M174 (the Eurasian YAP+ founder haplogroup) in Tibetans had previously led some researchers to propose an additional genetic contribution from Central Asians (9) or to infer an ancient relationship between Tibetans and Japanese (15).

One must concede that most of those genetic studies were hampered by either limited resolution of the classification tree (9, 11, 13), relatively small sample sizes (9–12), or, most importantly, potentially biased sampling coverage, in that most of the Tibetan samples came from the peripheral regions of Tibet, including Yunnan and Qinghai Provinces (9, 10, 12, 13) or from an undifferentiated “general population” (14). Consequently, phylogeographic analyses performed on Tibetans were only rudimentary and proved largely inconclusive, as fine-scale founder types could not be identified.

Results and Discussion

To investigate at a finer scale whether any genetic relics from the Paleolithic inhabitants have survived in the modern Tibetan population, we analyzed 680 individuals, representing 6 populations sampled from all major residential regions of Tibetans across China (Fig. 1 and Table S2), for the (nearly) entire mtDNA control region sequence variation (Table S3). We then selected mtDNAs to sequence the entire genome.

With the exception of a few subjects (classified coarsely as M*, N*, or R*), most of the samples (653/680) were unambiguously allocated to the known Eurasian mtDNA classification scheme (Table S3), among which the vast majority (637/653) belong to (a fraction of) east Asian haplogroups, whereas the remaining ones belong to haplogroups prevalent in either west Eurasians (14/653) or south Asians (2/653). Haplogroups prevalent in northern East Asia, including A, M8 (encompassing M8a, C, and Z), M9, D, and G, were found at relatively high frequencies in Tibetans (64.0% on average; Table S4), an observation consistent with previous reports (4, 11–13). This is also well expressed by a principal components analysis (PCA): most regional Tibetan populations (except for Sichuan-Tibetans) show closer relationships with the northern East Asian populations (Fig. 2).

Fig. 2.

PCA of the populations under study. The map was constructed based on the basal haplogroup frequency matrix in Table S4. N_Tib, Nakchu Tibetan; R_Tib, Shigatse Tibetan; T_Tib, Tibet Tibetan; Q_Tib, Qinghai Tibetan; Y_Tib, Yunnan Tibetan; S_Tib, Sichuan Tibetan; G_Tib, Gansu Tibetan. See Table S3 for details of the group classifications.

Comparison with the published East Asian mtDNA datasets revealed that most of the East Asian lineages observed in Tibetans nearly match those in other northern East Asian populations (Fig. S1). This reflects rather recent genetic contributions, in agreement with the historically recorded assimilation of ancient Di-Qiang people into the proto-Tibetans (5). Complete sequencing of representative Tibetan lineages (as judged from the control region variation motifs; Table S3) disclosed 3 novel subhaplogroups: M9c, M13b, and A10 (Fig. 3). Phylogenetic analysis showed that several haplogroups, including M9a, M9c, M13a, M13b, G3a1, and A10, are prevalent in Tibetans (Tables S3 and S4). Most of these (M9c, M13a, M13b, and G3a1 in particular) share no terminal but only root types, with their counterparts in other populations from China (Figs. S2–S6). This strongly suggests that these specific lineages have de novo origins within Tibetans. Therefore, these haplogroups could serve as optimal molecular markers for dating the start of the major migration into the Plateau. The situation for haplogroups M9a and A10 seems somewhat different. Both of these haplogroups contain several major clusters that composed nearly exclusively of non-Tibetans or a mixture of Tibetans, Han Chinese, and individuals from the other Chinese ethnic populations, suggesting that haplogroups A10 and M9a had already differentiated before their arrival in Tibet. Fig. 4 summarizes the estimated ages of these haplogroups by adopting the most recent fine-tuning of the calibration rate of different segments of the mtDNA (see Table 1 for details). The ages of M9a, A10, and G3a1 fall into the period of post-LGM warming, whereas M9a2, M9c, M13a, and M13b are likely of early Holocene origin. It is noteworthy that the arrival time of these haplogroups at the Tibetan Plateau may have been somewhat more recent than their coalescent ages would indicate, because some of these haplogroups (A10 and M9a in particular) had already differentiated before their arrival on the plateau (Figs. S2 and S6). It is then conceivable that most, if not all, of these haplogroups may have actually arrived and spread on the plateau only after the 8.2 ka event (8.0–8.4 kya) at the beginning of the Holocene climatic optimum, with first Epipaleolithic and later Neolithic settlers from the upper and middle Yellow River. The distribution and frequencies of the geographically differentiated haplogroups M9a and M13 strikingly parallels that of the Y-chromosome haplogroup D-M174, which has relatively high frequencies (14.0%–72.3%) among most Tibeto-Burman populations and in Japanese (35.1%) (15). Given the limited resolution of the current set of Y-chromosome SNPs and the difficulty of identifying paternal founder types in a population, it is likely that the Y-chromosome haplogroup D-M174 is rather to be compared to the mtDNA macrohaplogroup M. Therefore, the question of whether the Tibetan-Japanese genetic link is of Pleistocene (15) or Holocene origin (as reflected by the sharing of mtDNA haplogroups M9a and M13) still awaits further investigation.

Fig. 3.

Reconstructed mtDNA tree of the completely sequenced representatives of the major Tibetan mtDNA lineages. Suffixes “A,” “C,” “G,” and “T” refer to transversions, “d” denotes deletion, and “+” indicates an insertion event (without specifying the number of inserted nucleotides). Suffixes “Y” and “R” denote heteroplasmic mutations (C/T and A/G, respectively); recurrent mutations are underlined; “@” denotes a reverse mutation; “s” means synonymous and “ns” means nonsynonymous mutation; “−nc” refers to mutations at the intergenic noncoding regions in segments 577–16023; and “∼r” and “∼t” denote mutations in rRNA genes and tRNA genes, respectively. The C stretch length polymorphism in region 303–315 was disregarded for the tree reconstruction. Suffixes “MT#,” “MI#,” “QPK#,” and “PS#” next to the sample names refer to the sources Tanaka et al. (38), Ingman et al. (39), Kong et al. (19, 24), and Soares et al. (40), respectively. Codes “N,” “R,” “Q,” “Y,” and “G” refer to sampling locations (Nakchu, Shigatse, Qinghai, Yunnan, and Gansu, respectively) of different regional Tibetan populations.

Fig. 4.

Estimated ages of the Tibetan-prevalent mtDNA lineages. Open cycles indicate the estimated ages of a haplogroup, with red, brown, green, blue, and black cycles referring to the calibration rates for segment 16051-16400 substitutions (35), coding region synonymous substitutions (35), modified coding region synonymous transitions (34), modified coding region substitutions (34), and complete genome substitutions (35), respectively.

Table 1.

Estimated ages of specific Tibetan mtDNA haplogroups based on different calibration rates

Haplogroup	n	Control region substitutions		n	Coding region substitutions		Coding region synonymous transitions		Coding region synonymous substitutions		Complete genome substitution
		Soares rate for segment 16051–16400*			Modified Mishmar rate†		Modified Kivisild rate†		Soares synonymous rate*		Soares rate*
		ρ ± σ	T (ky)		ρ ± σ	T (ky)	ρ ± σ	T (ky)	ρ ± σ	T (ky)	ρ ± σ	T (ky)
M16	23	1.61 ± 0.53	26.83	13	4.77 ± 1.29	21.99	3.08 ± 1.00	23.54	3.08 ± 1.00	24.26	8.23 ± 1.61	22.34
			[18.04, 35.62]			[16.05, 27.92]		[15.87, 31.21]		[16.35, 32.17]		[17.72, 27.05]
A10	60	1.07 ± 0.54	17.79	2	1.00 ± 0.71	4.61	0.50 ± 0.50	3.83	0.50 ± 0.50	3.94	1.50 ± 0.87	3.90
			[8.75, 26.83]			[1.35, 7.87]		[0.00, 7.65]		[0.00, 7.88]		[1.63, 6.19]
G3a1	25	1.04 ± 0.39	17.34	3	1.00 ± 0.75	4.61	0.33 ± 0.33	2.55	0.33 ± 0.33	2.63	2.00 ± 0.94	5.22
			[10.81, 23.88]			[1.17, 8.05]		[0.00, 5.10]		[0.00, 5.26]		[2.73, 7.74]
M9a	143	1.02 ± 0.27	17.03	15	3.67 ± 1.14	16.90	2.47 ± 0.97	18.87	2.47 ± 0.97	19.45	5.67 ± 1.32	15.23
			[12.58, 21.48]			[11.64, 22.16]		[11.46, 26.28]		[11.81, 27.08]		[11.56, 18.98]
M9a2	20	1.05 ± 0.42	17.51	3	1.33 ± 0.67	6.15	0.67 ± 0.47	5.10	0.67 ± 0.47	5.26	3.33 ± 1.15	7.89
			[10.48, 24.54]			[3.07, 9.22]		[1.49, 8.71]		[1.54, 8.97]		[4.94, 10.90]
M9c	60	0.58 ± 0.19	9.73	3	1.34 ± 0.67	6.15	—	—	—	—	3.33 ± 1.05	8.79
			[6.62, 12.84]			[3.07, 9.22]						[5.96, 11.67]
M13a	24	0.33 ± 0.19	5.56	5	2.60 ± 1.22	11.99	1.00 ± 0.82	7.65	1.00 ± 0.82	7.88	4.20 ± 1.51	11.16
			[2.45, 8.67]			[6.38, 17.59]		[1.34, 13.96]		[1.38, 14.39]		[7.06, 15.36]
M13b	25	0.40 ± 0.16	6.67	3	2.33 ± 0.88	10.76	1.33 ± 0.67	10.20	1.33 ± 0.67	10.51	3.00 ± 1.00	7.89
			[4.00, 9.34]			[6.69, 14.82]		[5.10, 15.30]		[5.26, 15.77]		[5.22, 10.61]

*See ref. 35 for details.

^†See ref. 34.

Significantly, one unclassified mtDNA lineage (designated as M* in Table S3), recognized by the control region motif 150-204-16223-16260, has a distribution restricted to Tibetans, with only 2 descendant lineages observed sporadically in Uygur and Uzbek (16); both populations were in contact with Tibetans during the Yuan Dynasty (Fig. 5). The whole mtDNA genome analysis revealed 10 diagnostic coding region mutations (Fig. 3). This lineage, named M16 here, branched off directly from the M founder type and is absent in >5,000 (nearly) complete mtDNA genomes from the worldwide mtDNA database. With hindsight, this lineage was detected in previous studies on Tibetans (11, 13, 16, 17), but has remained unrecognized due to limited information. For instance, among 54 Tibetans, there was one sample (type AS155) with salient RFLP status (−2734AluI, +8148HaeIII, and + 15520HaeIII) which clearly points to M16 status (11). The low frequency of M16 (1.9%; 1/54) in Tibetans sampled from 3 regions of Tibet (Nakchu, Tsedang, and Linchi) (11) is in line with our results (2.1%; 14/680).

Fig. 5.

Constructed median network displaying the control region information of M16 lineages. This network was constructed manually according to Bandelt et al. (31). The data used here were collected from the literature (Table S2) and the present study (Table S3). The sequence information used for network construction was confined to segment 16047-16497. Time estimation was carried out based on segment 16051-16400 as described previously (35). The asterisk denotes the ancestral node of the haplogroup defined by motif 16223-16260-16295-73-150-204-263. See the legend of Fig. 3 for more information.

Comparison of nonsynonymous and synonymous substitutions on the mtDNA protein-encoding genes, as well as on internal and terminal branches of the East Asian mtDNA tree, revealed no significant difference between the variations seen within haplogroup M16 and within other East Asian haplogroups of similar ages (Tables 2 and 3). This indicates that M16 might have been under similar natural selection pressure as the other haplogroups, at least since the beginning of the LGM. This finding is supported by the observation that different timing methods, including the method that considers only synonymous transitions (18), have led to coincident ages of M16 (>21 kya; Table 1 and Fig. 4). The rather long stem of M16 in the mtDNA phylogeny shows a somewhat different picture, with 5 nonsynonymous changes and 1 rRNA mutation, opposed to just 4 synonymous mutations (Fig. 3). A more extreme pattern among the basal East Asian haplogroups is seen only in the stem for haplogroup R11 (19). This could imply that the evolutionary pathway leading to the root of haplogroup M16 might have been under strong selection pressure during the millennia before the LGM, which eventually left M16 as the sole surviving mtDNA lineage among the earliest Paleolithic inhabitants of the Tibetan Plateau.

Table 2.

Comparisons of nonsynonymous and synonymous substitutions between M16 and the other East Asian M lineages

Gene	M16		East Asian M lineages*
	NS†	S‡	NS†	S‡	P§
ND1	1	3	9	15	1.000
ND2	2	1	8	15	0.538
COX1	0	4	3	21	1.000
COX2	2	1	5	17	0.180
ATP8	0	1	2	5	1.000
ATP6	2	0	6	6	0.473
COX3	0	0	6	8	1.000
ND3	0	2	1	3	1.000
ND4L	0	0	0	8	1.000
ND4	0	4	7	15	0.546
ND5	1	3	9	24	1.000
ND6	0	1	4	11	1.000
CytB	2	4	14	13	0.656
Totally	10	24	74	161	1.000

*Data from refs. 19 and 24. Haplogroups M9a, M13, and G3a1, which are prevalent in Tibetans, were not considered.

^†NS refers to the number of nonsynonymous substitutions.

^‡S refers to the number of synonymous substitutions.

^§P values determined by the 2-tailed Fisher's exact test.

Table 3.

Comparison of internal and terminal NS/S on the East Asian mtDNA tree

	n*	NSi/SI†	NSt/St†	P
M16	13	0.78 (7/9)	0.20 (3/15)	0.13
East Asian M lineages‡	38	0.56 (24/43)	0.42 (50/118)	0.44
P		0.58	0.29

*Sample size.

^†Indices “i” and “t” refer to the corresponding fractions of NS/S (see Table 2) for the internal branches and terminal branches, respectively, of the East Asian mtDNA tree. P values were obtained by the 2-tailed Fisher's exact test.

^‡Data from refs. 19 and 24. Haplogroups M9a, M13, and G3a1, which are prevalent in Tibetans, were not considered.

Several characteristics of M16 observed so far, including its basal status on the Eurasian mtDNA tree, restricted distribution in Tibetans, and ancient age (>21 kya), strongly support that this lineage has a different origin from the other Tibetan-prevalent haplogroups of northern East Asian ancestry. The most reasonable explanation for these is that M16 represents the genetic relics of the initial Late Paleolithic settlers on the Tibetan Plateau. This inference is supported by the fact that our dating for the autochthonous haplogroup M16 is very close to the optical dating (20.6–21.7 kya) of the human handprints and footprints found at the center of the Tibetan Plateau (in the vicinity of Lhasa, 4,200 m above sea level) (1).

Our findings have significant implications for the seemingly conflicting inferences drawn from archaeology, genetics, and historic records. Essentially, the previous debate on the peopling of the Tibetan Plateau concerned the issue of whether or not the initial Late Paleolithic inhabitants on the plateau were completely replaced by the later Neolithic immigrants. In this study, the observed genetic continuity between the initial Paleolithic inhabitants and the modern populations on the Tibetan Plateau strongly suggests that modern humans did exist on the plateau before the LGM, and it is these Paleolithic people who have successfully overcome the extremely harsh climate and environments and made some genetic contribution (albeit limited) to the contemporary inhabitants. This also helps to explain why the excavated microliths on the Tibetan Plateau display mosaic features of both northern Chinese tool culture (6) and the Tibetan Paleoliths (7, 8).

In summary, although the vast majority of identified mtDNA lineages found in Tibetans can trace their origins to northern East Asia and may have entered the Tibetan Plateau in the Holocene, our study provides support for the existence of genetic relics of the Late Paleolithic settlers in Tibetans, indicating some genetic continuity between the initial Paleolithic inhabitants and the modern populations on the Tibetan Plateau. Our findings may contribute to resolving the long-standing debates among the fields of archaeology, history, and genetics.

Subjects and Methods

Sampling.

Blood samples from 680 unrelated individuals of 6 Tibetan populations were collected with informed consent. Total DNA was extracted by the standard phenol/chloroform method. The populations were labeled as follows: Nakchu-Tibetans, 168 Tibetans from Nakchu Prefecture of Tibet; Shigatse-Tibetans, 220 Tibetans from Shigatse Prefecture of Tibet; Yunnan-Tibetans, 71 Tibetans from Diqing Tibetan Autonomous Prefecture of Yunnan Province; Qinghai-Tibetans, 76 Tibetans from Qinghai Province; Sichuan-Tibetans, 62 Tibetans from Liangshan Yi Autonomous Prefecture of Sichuan Province; and Gansu-Tibetans, 83 Tibetans from Gannan Tibetan Autonomous Prefecture of Gansu Province.

Sequencing and RFLP Typing.

With the exception of 37 Qinghai-Tibetans, for which only the segment spanning from position 16001 to position 16497 (relative to the revised Cambridge reference sequence, rCRS) (20, 21) was amplified and sequenced as described elsewhere (22), the entire mtDNA control region for the other 643 samples was amplified, sequenced, and dealt with as described previously (23), with minor modifications in the reverse primers [i.e., replacement of the previous reverse primer H408 by H902 (5′-GACTTGGGTTAATCGTGTGAC-3′) or H575 (5′-TGAGGAGGTAAGCTACATAAACTG-3′), to cover more informative sites, such as position 489]. To confirm the haplogroup status inferred from the control region motifs, the following coding region sites were selected for typing by either RFLP or DNA sequencing [according to the reconstructed East Asian mtDNA tree (19, 24)]: 10397AluI (for macrohaplogroup M), 5176AluI/4883 (D), 3008TaqI/3010 (D4), 4831HhaI/4833 (G), 9820HinfI/6455 (M7), 6680 (M7b), 4715 (M8), 14465AccI (M8a), 13262AluI (C), 3391HaeIII/3394 (M9a), 12549 (M10), 10644RsaI (M10a), 7641AluI (M11), 12030 (M12), 6023/6253 (M13), 663HaeIII/663 (A), 5417 (N9), 10310 (F), 12406HincI (F1), 14766 (HV), 7025AluI (H), and 8281–8289del (B). For some Tibetan samples [i.e, 32 Yunnan-Tibetans and 8 Qinghai-Tibetans, for which segments from 16001 to 16497 have been reported by Yao et al. (12)], the corresponding HVS-II segments and some further coding region sites were sequenced and/or screened as well.

Data Analyses.

Based on the combined control region and coding region information, the majority of the samples were unambiguously assigned to haplogroups under the guidance of the reconstructed mtDNA trees of East Asian (19, 24) and South Asian (25, 26) mtDNA lineages. For those Tibetan mtDNAs that remained unassignable, complete mtDNA genome sequencing was performed, as described previously (27), to fully determine their exact phylogenetic status. Specifically, the whole mtDNA genome was amplified in 4 overlapping fragments by using 4 pairs of primers (L13894/H2187, L1677/H6505, L5868/H10718, and L9877/H14676), then each fragment was sequenced using a set of inner primers. (See table 1 in ref. 27 for detailed information on the primers.) With this approach, the previously reported problems in mtDNA genome datasets, such as artificial recombination (28) and amplification of pseudomitochondrial gene (29), could be minimized. Two samples (Uzb57 and Uyg5) reported with M16 control region variation motif (16) (GenBank accession numbers AY678062 and AY678009) were also selected for complete sequencing.

The haplogroup allocations of the reported mtDNA data from the literature (Table S2) were reevaluated by the near-matching strategy (30). The reduced median network for each haplogroup was constructed manually [as described by Bandelt et al. (31)] and then confirmed using Network 4.510 (http://www.fluxus-engineering.com/sharenet.htm). The time to the most recent common ancestor of a haplogroup was estimated as described previously (32–35). PCA was conducted as described previously (30).

The Chinese M sequences used in our comparative analysis of nonsynonymous and synonymous substitution were obtained from the literature (19, 24). Three sequences (GenBank accession numbers AY255153, DQ272115, and DQ272108) belonging to M9a, M13, and G3a1, respectively, were disregarded, because these haplogroups are very frequent in Tibetans and thus might have suffered similar high-altitude selection pressure as M16 did. Mutations were classified into nonsynonymous and synonymous substitutions for each gene using mtDNA-GeneSyn software (36). Each mutation was classified as internal or terminal on the mtDNA tree, as described previously (37). Fisher's exact test was used to examine the difference in each gene between M16 and the other East Asian M lineages.