AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 118:63–76 (2002)

Genetic Relationship of Chinese Ethnic Populations

Revealed by mtDNA Sequence Diversity
Yong-Gang Yao,1 Long Nie,1 Henry Harpending,2 Yun-Xin Fu,3 Zhi-Gang Yuan,4 and Ya-Ping Zhang1*
1
Laboratory of Molecular Evolution and Genome Diversity, Kunming Institute of Zoology, Chinese Academy of
Sciences, Kunming 650223, People’s Republic of China.
2
Department of Anthropology, University of Utah, Salt Lake City, Utah 84112.
3
Human Genetics Center, University of Texas-Houston, Houston, Texas 77030.
4
Guangxi Medical University, Nanning, Guangxi 530021, People’s Republic of China.

KEY WORDS mtDNA; Chinese ethnic populations; origin; genetic differentiation; language

ABSTRACT The origin and demographic history of
the ethnic populations of China have not been clearly
resolved. In this study, we examined the hypervariable
segment I sequences (HVSI) of the mitochondrial DNA
control region in 372 individuals from nine Chinese pop-
ulations and one northern Thai population. A relatively
high percentage of individuals was found to share se-
quences with those from other populations of the same
ethnogenesis. In general, the populations of southern or
Pai-Yuei tribal origin showed high haplotype diversity
and nucleotide diversity compared with the populations of
northern or Di-Qiang tribal origin. Mismatch distribu-
tions from these populations showed concordant features.
All except the northern groups Nu, Lisu, Tibetan, and
Mongolian showed typical signatures of ancient popula-
tion expansions in the mismatch distributions and neu-
trality tests. Episodes of extreme size reduction in the past
are one of the likely explanations for the absence of evi-
dence of expansion in northern populations. Small sample
sizes as well as samples from isolated subpopulations
contributed to the bumpy mismatch distributions ob-
served. Phylogenetic analysis and haplotype sharing
among populations suggest that current mtDNA variation
in these ethnic populations could reveal their ethnohistory
to some extent, but in general, linguistic and geographic
classifications of the populations did not agree well with
classification by mtDNA variation. Am J Phys Anthropol
118:63–76, 2002. © 2002 Wiley-Liss, Inc.

There are 56 officially recognized ethnic popula-
tions in China. Among them, Hans constitute the
vast majority and are dispersed nearly all over the
country. The other ethnic populations are found
mainly in peripheral regions (Du and Yip, 1993).
How these ethnic populations came into existence,
and how and when they spread in China are of
great interest, particularly to Chinese anthropol-
ogists. According to their current distributions,
languages, culture, habits, and historical docu-
ments, most of the minority ethnic populations in
southwest China can be traced back to two main
ancient tribes: Di-Qiang and Pai-Yuei. The Di-
Qiang tribe moved from the current Gansu,
Ningxia, and Qinghai Provinces in northwest
China to southwest China in two waves, one
around 4,000 –5,000 years ago and the other
around 2,000 –2,500 years ago (Du and Yip, 1993;
You, 1994). During their migration, the Di-Qiang
people intermarried with other groups and differ-
entiated into many ethnic groups, some of which
settled in highlands and mountainous regions.
The ancient Pai-Yuei tribe was widely distributed
along the southeast coast of China up to Yunnan
Province and the northern part of Southeast Asia
2,000 –3,000 years ago (Du and Yip, 1993; You,
1994). Some Pai-Yuei people later migrated to
north Thailand and contributed to the ancestral
gene pool of the Thai (Du and Yip, 1993; Ma, 1994;
You, 1994). Do current genetic structures of these
populations have the signatures of their earlier
ethnogenesis? Is there a correlation between the
genetic and the language differentiation? Given
that these ethnic populations mainly practiced
patrilocal endogamy and had different cultural
traditions, it is of interest to see at what extent
their cultural traditions may be reflected in the
current mtDNA pools of these populations.
Previous analyses of clustering of frequency dis-
tributions of immunoglobulin Gm haplotypes (Zhao
Grant sponsor: Natural Science Foundation of Yunnan Province;
Grant sponsor: National Natural Science Foundation of China; Grant
sponsor: Chinese Academy of Sciences.
*Correspondence to: Ya-Ping Zhang, Laboratory of Molecular Evo-
lution and Genome Diversity, Kunming Institute of Zoology, Chinese
Academy of Sciences, Kunming, Yunnan 650223, People’s Republic of
China. E-mail: zhangyp@public.km.yn.cn
Received 22 February 2000; accepted 26 November 2001.
Published online in Wiley InterScience (www.interscience.wiley.
com).
DOI 10.1002/ajpa.10052

© 2002 WILEY-LISS, INC.

64 Y.-G. YAO ET AL.
et al., 1991), HLA alleles (Chen et al., 1993), and 38
allele frequencies (Du et al., 1998) in Chinese ethnic
populations have identified a clear difference be-
tween northern and southern Chinese groups. Stud-
ies of archaeological assemblages revealed that dis-
tinction between the south and the north
populations might have existed since the Neolithic
period (Wu et al., 1989). Recently, data from Y chro-
mosome biallelic markers as well as nuclear micro-
satellites suggested a southern origin of the north-
ern groups (Chu et al., 1998; Su et al., 1999;
reviewed in Jin and Su, 2000). However, by combin-
ing the data from mtDNA RFLPs, short tandem
repeat loci, and published data of Y-chromosome as
well as of human-carried JC virus, an ordinarily
benign urinary tract virus that is primarily trans-
mitted from parent to child, Ding et al. (2000) em-
phasized that the regional difference in genetic
markers in southern and northern populations
might be more properly explained by simple isola-
tion by distance. Even though the ethnic populations
of northern Di-Qiang origin and of southern Pai-
Yuei origin have been included in previous studies,
none of these studies discussed the relationship be-
tween documented ethnohistory and current genetic
structure of these populations.
In the present study, we analyzed mtDNA control
region hypervariable segment I sequence (HVSI) in
nine Chinese ethnic populations and one northern
Thai population, together with the data from four
previously reported Chinese populations (Horai et
al., 1996; Betty et al., 1996; Yao et al., 2000a).
Among them, Bai, Sali (a branch of the Yi ethnic
group), Nu, Lisu, and Tibetan trace their origins to
the ancient Di-Qiang tribal group in northwest
China, while Dai and Zhuang are descendants of the
ancient Pai-Yuei tribe. There are two main language
families in our sample, with Uygur, Kazak, Tu, and
Mongolian belonging to the Altaic language family,
while Bai, Sali, Lisu, Nu, Tibetan, Dai, Zhuang, and
Han people speak Sino-Tibetan languages (Du and
Yip, 1993; Ma, 1994). Our focus here is not on the
distinction between southern and northern Chinese
groups; rather, we are interested in understanding
1) the possible relationship between population eth-
nohistory and its genetic architecture, and 2) the
correlation between language, culture, and genetic
differentiation.
MATERIALS AND METHODS
Sampling
Nine Chinese ethnic populations were sequenced
in the present study. Bai (N ⫽ 31), Dai (N ⫽ 38),
Lisu (N ⫽ 37), Sali (N ⫽ 31), and Nu (N ⫽ 30) were
collected in Yunnan Province. Tu (N ⫽ 35) and Mon-
golian (N ⫽ 15) were from Qinghai Province. An
indigenous ethnic population of Guangxi Province,
Zhuang (N ⫽ 83), was included. Except for the 32
Yunnan Tibetans collected in an isolated village in
the hilly region of DeQin County, Yunnan Province,
all other samples, including 8 Tibetans from Qing-
hai Province, were from various villages in concen-
trated communities of each ethnic population. The
blood donors of Zhuang were students of Guangxi
Medical University. Thirty-two Thais from north
Thailand were also sequenced in the current study.
The locations and language families of the studied
ethnic populations are shown in Table 1 and Figure
1. The geographic origin, nationality, and maternal
pedigree (unrelated through at least three genera-
tions) of each individual were ascertained before
sampling.
DNA extraction, amplification, and sequencing
Genomic DNA was extracted from the whole blood
by standard phenol/chloroform methods. The HVSI
sequence was amplified and sequenced as described
elsewhere (Yao et al., 2000a,b). For some individu-
als, internal primer H16401 (5⬘-TGATTTCACG-
GAGGATGGTG-3⬘) (Vigilant et al., 1991) and L16209
(5⬘-CCCCATGCTTACAAGCAAGT-3⬘) (Mountain et
al., 1995) were used in sequencing.
Data analysis
The sequences were edited and aligned using
Dnastar software (DNASTAR, Inc.), and sequence
haplotypes were identified. Other data used in the
current study included 20 Han individuals from
Hongkong (Cantonese) (Betty et al., 1996), 52 Han
individuals from Taiwan (Horai et al., 1996; the
other 14 individuals were discarded for short se-
quences compared with our data.), 45 Uygurs, and
30 Kazaks from Xinjiang Province (Yao et al.,
2000a). Because some of the published mtDNA data
included here for comparison were of only 360-bp
length (nucleotide positions 16024 –16383), we re-
stricted our analyses to that 360-bp segment when
the published data were used.
Three methods were used to compare the genetic
component of each ethnic population. Firstly, we
identify the shared haplotypes among ethnic popu-
lations, then computed the matching probability (m)
between population samples according to
m⫽ 冘 XY,i i
i
where Xi, Yi are the sample frequencies of the
shared haplotype i in population X and Y, respec-
tively. Secondly, haplotype diversity (h) and nucleo-
tide diversity (␲) were used. Haplotype diversity
was estimated by using the formula:
h ⫽ 共n/n ⫺ 1兲共1 ⫺ 冘 p 兲,
i
2
i
where pi is the sample frequency of the i-th haplo-
type and n is the number of individuals in the sam-
ple (Nei, 1987). Nucleotide diversity was estimated
by
冘共1 ⫺ 冘␹ 兲,
␲ ⫽ 共1/L兲共n/n ⫺ 1兲 2
ij
j i
 GENETIC RELATIONSHIP OF CHINESE POPULATIONS 65
TABLE 1. Sample information of populations analyzed in present study
Sample No. of No. of unique
Population Language1 Census size1 Location size haplotypes haplotypes2
1. Lisu Yi branch, Tibeto-Burman 574,856 Gongshan, 37 22 15 (68.2%)
group, Sino-Tibetan Yunnan
family
2. Nu Yi branch, Tibeto-Burman 27,123 Gongshan, 30 9 4 (44.4%)
group, Sino-Tibetan Yunnan
family
3. Sali Yi branch, Tibeto-Burman 6,572,173 Nuxi, Yunnan 31 26 17 (47.2%)
group, Sino-Tibetan
family
4. Tibetan Tibetan branch, Tibeto- 4,593,330 Deqin, Yunnan; 40 19 15 (78.9%)
Burman group, Sino- Qinghai
Tibetan family
5. Bai Yi branch, Tibeto-Burman 1,594,827 Dali, Yunnan 31 29 21 (72.4%)
group, Sino-Tibetan
family
6. Dai Zhuang-Dai branch, 1,025,128 Jinghong, 38 36 24 (66.7%)
Zhuang-Dong group, Yunnan
Sino-Tibetan family
7. Zhuang Zhuang-Dai branch, 15,489,630 Guangxi 83 66 50 (75.8%)
Zhuang-Dong group,
Sino-Tibetan family
8. Cantonese3 Chinese, Sino-Tibetan Hong Kong 20 20 14 (70.0%)
family
9. Taiwanese Han3 Chinese, Sino-Tibetan Taiwan 52 46 34 (88.5%)
family
10. Tu Mongolian group, Altaic 191,624 Huzu, Qinghai 35 29 25 (86.2%)
family
11. Mongolian Mongolian group, Altaic 4,806,849 Qinghai 15 10 9 (90.0%)
family
12. Kazak3 Turkic group, Altaic 1,111,718 Kashen, Xinjiang 30 27 21 (77.8%)
family
3
13. Uygur Turkic group, Altaic 7,214,431 Yili and Kashen, 45 41 32 (78.0%)
family Xinjiang
14. Thai North Thailand 32 29 22 (75.9%)
Total 2 519 409 303
1
The language and total population size of each ethnic population (1990 Census) are from Du and Yip (1993) and Ma (1994).
2
Number of haplotypes that are not shared between or among the 14 populations. Percentages of these haplotypes in each population
are in parentheses.
3
Data from Betty et al. (1996), Horai et al. (1996), and Yao et al. (2000a), respectively.

structed from net genetic distances (dA), defined as

Fig. 1. Locations of samples in current study. Numbers cor-
respond to population names in Table 1.
where ␹ij is the frequency of the i-th nucleotide at
site j, L is the sequence length, and n is the sample
size (Nei, 1987). Thirdly, a neighbor-joining tree
(Saitou and Nei, 1987) of the populations was con-
dA ⫽ dXY ⫺ (dX ⫹ dY)/2, where dXY is the mean
pairwise difference between individuals from popu-
lation X and Y, and dX (dY) is the mean pairwise
difference between individuals within population X
(or Y) (Nei, 1987). Since a tree presentation of the
distance matrix might be misread as a succession of
population splits, we also performed principal com-
ponent analysis (PCA) of the populations based on
the distance matrix.
In order to examine whether there are genetic
differences among different geographic groups and
language groups, we grouped the ethnic populations
according to their geographic locations and lan-
guages, respectively, and performed analyses of mo-
lecular variance (AMOVA) (Excoffier et al., 1992),
using the Arlequin package (Schneider et al., 2000).
The demographic history of each population was
examined by two different approaches. First, the D
test of Tajima (1989) and the Fs test of Fu (1997)
were used to test if neutrality holds (i.e., the popu-
lation under study evolves with a constant effective
population size, all mutations being selectively neu-
tral). A population that has experienced population
66 Y.-G. YAO ET AL.
TABLE 2. Genetic diversities and population demographic parameters in the populations
Population No. Haplotype diversity Nucleotide diversity Fs1 P2 D3 Tau Expansion time (YBP)4
Lisu 37 0.957 ⫾ 0.017 0.020 ⫾ 0.011 ⫺4.848 0.043 ⫺0.919 6.152 51,800
Nu 30 0.858 ⫾ 0.038 0.017 ⫾ 0.009 2.295 0.816 ⫺0.024 3.927 33,100
Sali 31 0.989 ⫾ 0.011 0.019 ⫾ 0.010 ⫺16.567 0.000 ⫺1.390 7.223 60,800
Tibetan 40 0.933 ⫾ 0.020 0.013 ⫾ 0.007 ⫺5.834 0.021 ⫺1.565 4.058 34,200
Bai 31 0.996 ⫾ 0.010 0.018 ⫾ 0.010 ⫺24.460 0.000 ⫺1.599 7.188 60,500
Dai 38 0.996 ⫾ 0.007 0.020 ⫾ 0.011 ⫺24.971 0.000 ⫺1.389 7.847 66,100
Zhuang 83 0.992 ⫾ 0.004 0.020 ⫾ 0.011 ⫺24.868 0.000 ⫺1.482 8.116 68,300
Cantonese 20 1.000 ⫾ 0.016 0.019 ⫾ 0.010 ⫺15.576 0.000 ⫺1.464 7.438 62,600
Taiwanese Han 52 0.993 ⫾ 0.006 0.019 ⫾ 0.010 ⫺25.030 0.000 ⫺1.600 7.515 63,300
Tu 35 0.987 ⫾ 0.011 0.019 ⫾ 0.010 ⫺19.557 0.000 ⫺1.417 6.002 50,500
Mongolian 15 0.943 ⫾ 0.045 0.016 ⫾ 0.009 ⫺1.459 0.258 ⫺0.561 6.839 57,600
Kazak 30 0.993 ⫾ 0.011 0.018 ⫾ 0.010 ⫺20.781 0.000 ⫺1.553 7.063 59,500
Uygur 45 0.995 ⫾ 0.006 0.017 ⫾ 0.010 ⫺25.170 0.000 ⫺1.895 6.184 52,100
Thai 32 0.994 ⫾ 0.010 0.020 ⫾ 0.011 ⫺22.002 0.000 ⫺1.627 7.833 65,900
Chinese5 519 0.996 ⫾ 0.001 0.020 ⫾ 0.001 ⫺24.239 0.000 ⫺1.953 7.471 62,900
1
Fu’s Fs test.
2
P value of Fu’s Fs statistic.
3
Tajima’s D test.
4
Expansion time of the population was estimated from the tau value, assuming the divergent rate to be 33% site/million years (Ward
et al., 1991). Only 360-bp HVSI sequences (16024 –16383) were included.
5
Total number of individuals in 13 Chinese ethnic populations and the Thai population.

expansion may result in a rejection of the null hy-
pothesis. Second, mismatch distribution analyses
were used to evaluate 1) whether there was signa-
ture of population expansion that is found in most
samples from human populations in the mismatch
(Excoffier and Schneider, 1999), and 2) the timing of
demographic expansion measured in units of muta-
tional time. Typically, a population with a constant
size in the past has a multimodal mismatch distri-
bution, while a population that has undergone ex-
pansion usually shows a unimodal or Poisson-like
distribution (Rogers and Harpending, 1992; Rogers,
1995). These computations were also performed by
using the Arlequin package (Schneider et al., 2000).
The divergence rate from Ward et al. (1991; 33%
site/million years) was used to converting muta-
tional time (␶) into real time according to equation:
␶ ⫽ 2t␮, where ␮ is the mutation rate per sequence
per generation and t is the time in years.
RESULTS
Sequence diversity
Hypervariable segment I sequences of the mtDNA
control region (nucleotide positions 16001–16497)
were sequenced in 340 individuals from 9 Chinese
ethnic populations and 32 individuals from a north-
ern Thai population. One hundred thirty-five poly-
morphic sites (excluding the insertion sites) were
identified compared to the reference sequence
(Anderson et al., 1981; see Appendix). When the
published data of Chinese were considered, the Han
from Hongkong (Cantonese; Betty et al., 1996)
showed the highest haplotype diversity (h ⫽ 1.000 ⫾
0.016), followed by those of Bai, Dai, Zhuang, Tai-
wanese Han, Uygur, and Kazak (h ⬎ 0.99), while
that of Nu was the lowest, with a value of 0.858 ⫾
0.038 (Table 2). The haplotype diversity of the Thai
population showed a similar value to those of Dai
and Zhuang. In general, the ethnic populations of
ancient Di-Qiang origin showed lower haplotype di-
versity than those of southern Pai-Yuei origin pop-
ulations.
The two ethnic populations from Xinjiang Prov-
ince (Uygur and Kazak; Yao et al., 2000a) showed
similar nucleotide diversity (0.017– 0.018). However,
nucleotide diversities varied in different ethnic pop-
ulations from Yunnan Province, ranging from
0.013 ⫾ 0.007 (Tibetan) to 0.020 ⫾ 0.011 (Dai). The
two populations from Qinghai Province also differed
in their nucleotide diversities (Tu, 0.019 ⫾ 0.010;
Mongolian, 0.016 ⫾ 0.009), although with no statis-
tical significance (P ⬎ 0.05). With the exception of
the Lisu population, which had relatively high nu-
cleotide diversity (0.020 ⫾ 0.011), it seemed that the
higher nucleotide diversity was only present in Pai-
Yuei origin populations, such as Dai and Zhuang
(Table 2).
Haplotype sharing between or among the
ethnic populations
A total of 237 haplotypes was identified in the 340
Chinese and 32 Thais who were sequenced in this
study; among them, 23 haplotypes were shared by
two or more populations (see Appendix). The per-
centages of population-specific haplotypes were rel-
atively low in Nu (44.4%) and Sali (47.2%), while in
other samples, the percentages were higher, ranging
from 66 –90% (Table 1). As shown by the number of
sequences shared among the populations listed in
the Appendix and the matching probabilities in Ta-
ble 3, two main features could be discerned: 1) pop-
ulations of the same ethnic origin generally shared a
relatively large number of sequences with each
other and presented high matching probabilities be-
tween populations. The highest matching probabili-
ties were found between Nu and Lisu, between Nu
and Sali. 2) Populations of Pai-Yuei origin shared
relatively few sequences with the populations of Di-
 GENETIC RELATIONSHIP OF CHINESE POPULATIONS 67
Qiang origin. The matching probabilities even

All values have been multiplied by 100. Below the diagonal, net genetic distances (dA) between populations; above the diagonal, matching probability (m) between population
0.0877

0.0641
0.2151
0.3810
0.3226
0.1802
0.5556
0.1667

0.2963
Kazak
reached zero between some populations. However,

0.0

0.0
0.0

0.0
in the Bai population, which is of northern Di-Qiang
origin but has intermixed extensively with Han and
0.1170
0.2083
0.0535
0.0427
0.2151
0.4444
0.2151

0.2222

0.0278
Uygur
0.0 other local populations (Du and Yip, 1993; You,

0.0
0.0

0.0
1994), relatively high matching probabilities were
found with the populations of Pai-Yuei origin. The
Han populations (Hongkong and Taiwanese; Betty
Mongolian

0.0803

0.0472
0.1177
et al., 1996; Horai et al., 1996) presented relatively
0.0
0.0
0.0

0.0
0.0
0.0
0.0
0.0
0.0
0.0
higher matching probabilities with the populations
of Pai-Yuei origin than with those of Di-Qiang origin
and other populations that currently in northwest
TABLE 3. Matching probability between population samples and the net genetic distances between populations1
Tibetan

China.
0.3226
0.4286
0.5645

0.0903
0.1125
0.1959
0.0
0.0
0.0
0.0
0.0

0.0
0.0

Mismatch analyses
Among the 14 populations considered here, only
1.3333
2.0430
3.7838

0.1255
0.1249
0.1335
0.1613

Nu, Tibetan, Lisu, and Mongolian showed multimo-

Nu
0.0
0.0
0.0
0.0
0.0
0.0

dal mismatch distributions, characteristic of popu-

lations in equilibrium; the remaining populations all
have unimodal distributions, characteristic of popu-
0.0711

0.6950
1.0462

0.0477
0.1254
0.1057
0.0742
0.1201
Lisu

lations that have undergone large-scale expansion

0.0

0.0
0.0
0.0
0.0

(Fig. 2) (Rogers and Harpending, 1992; Rogers,

1995). The Fs test (Fu, 1997) and D test (Tajima,
0.3226
0.1698
0.4032
0.3109
0.4963
0.0922
1.1982

0.0326
0.0970
0.0677
0.0423
0.0188
0.0884

1989) agreed well with the mismatch analysis, with

Sali

insignificant test results for Nu (P ⬎ 0.05) and Mon-

golian (P ⬎ 0.05). For Tibetan and Lisu, the Fs tests
were marginally significant (Lisu, ⫺4.848, P ⫽
0.2857
0.2256
0.3571
0.3098
0.4945
0.3687

0.0124
0.0314
0.1006
0.0760
0.0704
0.0248
0.0781
Tu

0.043; Tibetan, ⫺5.834, P ⫽ 0.021), while the test for

the remaining 10 populations was highly significant
(P ⬍ 0.05; Table 2). Note that the mismatch analysis
0.7640
0.2016
0.5052
0.1861

0.0145
0.0048
0.0518
0.1133
0.1219
0.0610
⫺0.0089
0.0508

of the pooled Chinese sample presented a smooth

Bai
0.0

unimodal distribution (Fig. 2). Thus, to understand

the details of Chinese migration history, it is impor-
tant to identify the demographic events of different
Taiwanese Han

ethnic populations.
0.4808
0.4555
0.6486
0.5792

0.0353
0.0539
0.0692
0.0832
0.1602
0.1296
0.1016
0.0365
0.0854

In order to see whether small sample size could

mimic multimodal mismatch distribution, we ran-
domly chose 15 individuals from the Zhuang popu-
lation, which presented a unimodal mismatch dis-
tribution in a total of 83 individuals, to estimate the
Zhuang
0.3614
0.7609
0.6777

0.0593
0.0379
0.0731
0.0835
0.1300
0.2457
0.2440
0.1488
0.0564
0.0919

mismatch distributions. The resulting mismatch

distributions were found to be ragged (Fig. 2). The
smaller the sample size, the more ragged the mis-
match distribution (data not shown). The P values
0.4688
0.4934

⫺0.0025 ⫺0.0040
0.0468
0.0433
0.0819
0.1008
0.1273
0.2546
0.2737
0.1773
0.0655
0.1081

(P ⬎ 0.05) of the Fs test in these samples were all

Thai

statistically nonsignificant.
The tau value (␶), which reflects the location of the
0.2632

0.0011

0.0915
0.0393
0.0743
0.0777
0.1025
0.2410
0.2808
0.1666
0.0692
0.1248

mismatch distribution crest, provides a rough esti-

Dai

mate of the time when rapid population expansion

started (Rogers and Harpending, 1992; Rogers,
1995). The tau values of Zhuang, Dai, and Thai were
Cantonese

0.0086
⫺0.0067
⫺0.0142
0.0134
0.0051
0.0229
0.0483
0.0769
0.1641
0.1652
0.0908
⫺0.0014
0.0258

all larger than 7.8, while those of Nu and Tibetan

were smaller than 4.0. This figure should be re-
garded with caution, since the latter two popula-
tions (Nu and Tibetan) did not show clear evidence
of any expansion at all. Sali, Bai, and Kazak had
Taiwanese Han

similar tau values to those of the Han populations,

corresponding to estimated expansion times of more
Mongolian
Cantonese

samples.

than 66,000 years before present (YBP) for Dai,

Tibetan
Zhuang

Kazak
Uygur

Thai, and Zhuang; about 60,000 YBP for Bai, Sali,

Thai

Lisu
Sali
Dai

Bai

Nu
Tu

and Han from Hongkong and Taiwan; and about

1
68 Y.-G. YAO ET AL.

Fig. 2. Mismatch distributions of HVSI sequences within (a) Tibetan, (b) Lisu, (c) Nu, (d) Zhuang, (e) Mongolian, (f) 15 randomly
selected Zhuang samples, (g) pooled samples of Tibetan, Nu, Lisu, and Mongolian, and (h) total of 519 samples. Numbers of nucleotide
differences between all pairs of sequences are indicated along x-axis, and frequency of pairs is indicated by y-axis.

33,000 YBP for Nu and Tibetan, assuming a diver-
gence rate of 33% site/million years (Ward et al.,
1991) (Table 2). Again, the recent tau estimates of
Nu and Tibetan should be regarded with skepticism.
Genetic structure of the ethnic groups
The genetic structure of the ethnic populations
(Thai included) was investigated by AMOVA (Ex-
coffier et al., 1992). Considering haplotypes, 95.26%
of the genetic variation was found within popula-
tions, whereas 4.74% of the variation (P ⬍ 0.05,
1,000 iterations) was among populations when all 14
populations were grouped together. When the ethnic
populations from Xinjiang and Qinghai Province
were grouped together as a northwest geographic
group (45 Uygurs, 30 Kazaks, 35 Tus, 15 Mongo-
 GENETIC RELATIONSHIP OF CHINESE POPULATIONS 69
lians, and 8 Tibetans), and populations from Yun-
nan and Guangxi Province (31 Bais, 37 Lisus, 30
Nus, 32 Tibetans, 31 Salis, 38 Dais, and 83 Zhuangs)
were collected as a south geographic group, the ge-
netic variation among groups and within groups
among populations was ⫺0.39% (P ⫽ 0.94 ⬎ 0.05)
and 3.24% (P ⬍ 0.05), respectively. The genetic vari-
ation within the geographic groups was by far larger
than the between-group variation. The same results
were seen when the populations (Thai was not in-
cluded) were grouped into two major groups accord-
ing to their languages (Sino-Tibetan and Altaic, Ta-
ble 1). The proportion of genetic variation attributed
to the differences among language groups amounted
to ⫺0.17% (P ⫽ 0.52 ⬎ 0.05), while 4.86% (P ⬍ 0.05)
and 95.31% (P ⬍ 0.05) of the variation were ob-
served among populations within language groups
and within populations, respectively. We also
grouped the populations according to their historical
origin (Di-Qiang group: Sali, Tibetan, Nu, Lisu, and
Bai; Pai-Yuei group: Dai, Zhuang, and Thai); the
resulting AMOVA test showed a significant differ-
ence between these two groups. About 5.17% of the
genetic variation was observed between the Di-
Qiang and Pai-Yuei groups (P ⫽ 0.02 ⬍ 0.05).
Phylogenetic analysis and principal component
analysis of populations
An unrooted NJ tree of the 14 populations (Thai
included) was constructed using the net genetic dis-
tances (dA) shown in Table 3 (see also Fig. 3a). Those
with the same language family or from the same
geographic region (except for the two Xinjiang pop-
ulations) did not group together, which was consis-
tent with the results of AMOVA analyses. The Dai,
Zhuang, and Thai showed close affinity. The genetic
distances among Dai, Zhuang, and Thai even
showed negative estimated values. Tu and Sali oc-
cupied intermediate node positions in the tree, as
did Taiwanese Han and Bai. The two Xinjiang pop-
ulations (Uygur and Kazak) clustered together; how-
ever, the genetic distance between Uygur and Kazak
was not the smallest: the genetic distances between
Uygur and Han from Hongkong and between Uygur Fig. 3. a: Unrooted NJ tree of the 14 populations based on the
and Bai had negative estimates (Table 3). The Nu, net genetic distances (dA) given in Table 3. b: Principal compo-
Tibetan, and Lisu, with multimodal mismatch dis- nent (PC) map of the 14 populations based on the distance matrix.
tributions, also showed a long branch length to other First principal component accounts for 17% of the original vari-
populations. ation, while the second accounts for 15%. The PC map axes bear
no scales because the relationship shown among populations is
Figure 3b displays distances among the 14 popu- purely topological.
lations represented as the first two principal compo-
nents. The first two PCs incorporate about 32% of
the total squared distances among populations, intense drift, as also demonstrated by the mismatch
while the remaining dimensions mostly portray analyses.
unique characteristics of single populations. The DISCUSSION
cluster pattern of populations in the PC map, with
high-diversity southern-origin populations on the The vicissitudes of time, war, seizure of lands, and
left against low-diversity northern-origin popula- plague throughout history have produced numerous
tions on the right, was in good agreement with the large changes to the populations in China (Ge et al.,
results of the NJ tree. The genetic distances among 1997). Movement and admixture have thus obscured
the northerners are greater, probably reflecting much of the population history of Chinese groups.
their history of episodes of small population size and According to historical documents, Bai, Sali, Nu,
70 Y.-G. YAO ET AL.
Tibetan, and Lisu have their origins in the ancient
Di-Qiang group in northwest China, and some of
them (Lisu, Nu, and Sali) were separated from each
other about 1,000 –1,400 years ago (Du and Yip,
1993; Ma, 1994; You, 1994). Thus, it is not surpris-
ing that Sali, Nu, and Lisu share similar linguistic
backgrounds, geographic distributions, ethnic
names, customs, and habits. Bais have a different
demographic history. Around the 3rd and 4th cen-
turies and again in the 14th century, thousands of
Han moved to the region where Bais lived and were
integrated gradually into the Bai population. The
origin of Tu is still under debate, although the prev-
alent opinion is that their ancestors migrated from
eastern Liaoning Province to southeastern Qinghai
and southern Gansu Provinces around the early 4th
century. Subsequently there was some gene ex-
change with Mongolian, Tibetan, and Han people.
Another view is that Tus were the descendants of
Mongolian soldiers who intermarried with the indig-
enous nomadic herdswomen during the Yuan Dy-
nasty (1271–1368 AD). Both these historical ac-
counts suggest that the Tu had extensive genetic
admixture in their history (Du and Yip, 1993; Ma,
1994; You, 1994; Ge et al., 1997). Our results are
essentially in agreement with these historical docu-
ments. First, a large number of individuals from
these populations of the same ethnogenesis shared
sequences with each other. The populations with
documented gene flow with other populations, like
Bai and Tu, showed signatures of admixture on the
basis of sequence sharing and the matching proba-
bility, but the conclusion is without strong statisti-
cal support. Second, the genetic distances among
these populations of Di-Qiang origin, such as Lisu,
Sali, and Nu, were relatively small. The estimated
distance was even negative between the two Pai-
Yuei populations, Zhuang and Dai. However, larger
distances were observed between populations of Di-
Qiang origin and populations of Pai-Yuei origin (Ta-
ble 3). Third, a statistically significant difference
was found between the Di-Qiang and Pai-Yuei
groups in the AMOVA test. Finally, in the popula-
tion tree and the PC map (Fig. 3), these populations
presented a cluster pattern that was consistent with
their ethnohistory or historical documents about ge-
netic mixture. Thus, mtDNA variation in these pop-
ulations does reflect ethnohistory to some extent.
There are preserved genetic traces of ancient shared
ethnicity among those populations of the same eth-
nogenesis, even in the face of numerous subsequent
movements and substantial gene flow.
Among the 14 populations, Nu, Tibetan, Lisu, and
Mongolian showed no conspicuous population ex-
pansion in the past. These four populations also
exhibited reduced genetic diversity compared with
the others in our sample. Both of these may suggest
possible episodes of severe size reductions in these
regional populations in the past. The previous signal
of Pleistocene expansion might have been erased in
these populations due to episodes of small sample
size (Excoffier and Schneider, 1999). The relatively
large genetic distances between Nu and other pop-
ulations, and between Tibetan and other popula-
tions, could also result from the impact of genetic
drift in these samples (Relethford, 1996; Excoffier
and Schneider, 1999). Given the large census sizes of
the current Tibetans, Mongolians, and Lisus (Table
1), and the fact that samplings of the populations
might be restricted to small subsections of these
ethnic populations (i.e., villages or regional popula-
tions), as actually performed in the 32 Tibetan sam-
ples from Yunnan Province, the ragged mismatch
distributions observed here might be restricted to
these samples analyzed, but not to entire ethnic
groups labeled as Lisu, Tibetan, or Mongolian.
Moreover, the mismatch distributions could be eas-
ily affected by the sample sizes considered. As
shown in Figure 2, small sample sizes could mimic
bumpy mismatch distributions, while pooled sam-
ples from populations showing ragged mismatch dis-
tributions could present roughly unimodal mis-
match distributions. Thus, the multimodal
mismatch distribution in the 15 Mongolian samples
could also be attributed to small sample size. The
mismatch distribution analyses and the neutrality
tests should be taken with caution when dealing
with samples from restricted regions or with a small
sample size. Nevertheless, the different mtDNA
pools of these regional samples from various ethnic
populations could be reflected by the different mis-
match distributions observed.
The expansion times of the populations that show
evidence of expansion in the past were estimated to
be 58,000 – 65,000 YBP. This result reflects the an-
cient imprint of the initial Paleolithic expansion and
agrees with the previous estimate, based on Y-chro-
mosome biallelic markers, that the settlement of
modern humans in East Asia occurred about
60,000YBP (Su et al., 1999), earlier than the later
recorded population migrations (Du and Yip, 1993;
Ma, 1994; You, 1994; Ge et al., 1997). The estimated
expansion times of the Zhuang and Dai population
were earlier than 63,000 YBP and earlier than other
populations considered. This is consistent with the
southern origin hypothesis (cf. Jin and Su, 2000) and
the earliest archaeological evidence of modern hu-
mans in China, i.e., the ancient Liujiang people who
lived in Guangxi Province, dated back to 67 Kya
(U-series dating; Wu et al., 1989).
Our previous work on the frequency pattern of the
9-bp deletion in the mtDNA COII/tRNALys inter-
genic region, in which the frequency of the 9-bp
deletion was present at higher values in populations
of southern Pai-Yuei origin than in populations of
northern Di-Qiang origin (Yao et al., 2000b, 2001),
agrees well with the genetic pattern observed in the
present study. As discussed above, populations of
northern Di-Qiang origin seem to have undergone
recent genetic drift during their spread to southwest
China. In addition, ongoing migration from north to
south in the past 2,000 years (Ge et al., 1997) and
 GENETIC RELATIONSHIP OF CHINESE POPULATIONS
before might account for some of the higher genetic
diversity observed in the southern populations.
The languages of our sample fall into two families,
Sino-Tibetan and Altaic. We might expect that an-
cient population divergences would have led to well-
defined genetic clusters that were consistent with
linguistic boundaries. This has been found in major
ethnic groups (Barbujani and Sokal, 1990; Cavalli-
Sforza et al., 1992). In our present study, however,
the correlation between female genetic lineages and
linguistic differentiation of the populations was not
evident, as shown by AMOVA analyses and the clus-
tering pattern in the phylogenetic tree as well as the
PC map. One reason for the discordance may be due
to the likely asymmetry of maternal and paternal
gene flow among the populations, as reported by
Seielstad et al. (1998). Exogamous marriages hap-
pen occasionally in some populations nowadays. The
tendency for women to move to their husbands’ lo-
cality (patrilocality) could have obscured the corre-
lation between language and female genetic lin-
eages. This process has been observed in Indian
castes, in which there is limited upward female gene
flow among the social ranks of stratified Hindu
castes (Bamshad et al., 1998). Moreover, when we
considered large-scale migrations in recorded his-
tory (Ge et al., 1997), which can be responsible for
more than “occasional” female exogamy, the absence
of concordance between language classification of
populations and mtDNA variation could also be ex-
plained, for the discordance may arise in a relatively
short amount of time under such destabilizing forces
as warfare and plague. Our finding here is interest-
ing because linguistic and cultural differences
among Chinese ethnic populations seem to have per-
sisted for a lengthy time, despite possibly significant
female gene flow among them.
ELECTRONIC-DATABASE INFORMATION
Accession numbers and URLs for the sequences in
this article are as follows: GenBank, http://www.
ncbi.nlm.nih.gov/web/Genbank (accession numbers:
AF 392063–AF 392434).
ACKNOWLEDGMENTS
We thank Hai-Peng Li, Dr. Francesc Calafell, Pro-
fessor Hans-Jürgen Bandelt, Dr. Yuan-Chun Ding,
and the three anonymous reviewers for their critical
comments on the manuscript and interesting discus-
sions on the subject. We also thank Professor Ai-
Hua Liu, Pai-Li Geng, and Dr. Wen Wang for their
invaluable help in collecting the samples.
LITERATURE CITED
Anderson S, Bankier AT, Barrell BG, de Bruijn MH, Coulson AR,
Drouin J, Eperon IC, Nierlich DP, Roe BA, Sanger F, Schreier
PH, Smith AJ, Staden R, Young IG. 1981. Sequence and orga-
nization of the human mitochondrial genome. Nature 290:457–
465.
Bamshad MJ, Watkins WS, Dixon ME, Jorde LB, Rao BB, Naidu
JM, Prasad BV, Rasanayagam A, Hammer MF. 1998. Female
gene flow stratifies Hindu castes. Nature 395:651– 652.
71
Barbujani G, Sokal RR. 1990. Zones of sharp genetic change in
Europe are also linguistic boundaries. Proc Natl Acad Sci USA
87:1816 –1819.
Betty DJ, Chin-Atkins AN, Croft L, Sraml M, Easteal S. 1996.
Multiple independent origins of the COII/tRNALys intergenic
9-bp mtDNA deletion in Aboriginal Australians. Am J Hum
Genet 58:428 – 433.
Cavalli-Sforza LL, Minch E, Mountain J. 1992. Coevolution of
genes and languages revisited. Proc Natl Acad Sci USA 89:
5620 –5624.
Chen R, Ye G, Geng Z, Wang Z, Kong F, Tian D, Bao P, Liu R, Liu
J, Song F, Fan L, Zhang G, Guo S, Xu L, Xu X, Cheng D, Zhao
X. 1993. Revelations of the origin of Chinese nation from clus-
tering analysis and frequency distribution of HLA polymor-
phism in major minority nationalities in mainland China. Acta
Genet Sin 205:389 –398 [in Chinese].
Chu JY, Huang W, Kuang SQ, Wang JM, Xu JJ, Chu ZT, Yang
ZQ, Lin KQ, Li P, Wu M, Geng ZC, Tan CC, Du RF, Jin L. 1998.
Genetic relationship of populations in China. Proc Natl Acad
Sci USA 95:11763–11768.
Ding Y-C, Wooding S, Harpending H, Chi H-C, Li H-P, Fu Y-X,
Pang J-F, Yao Y-G, Xiang YJG, Moyzis R, Zhang Y-P. 2000.
Population structure and history in East Asia. Proc Natl Acad
Sci USA 97:14003–14006.
Du R, Yip VF. 1993. Ethnic groups in China. Beijing: Science
Press.
Du R, Xiao CJ, Cavalli-Sforza LL. 1998. Genetic distances be-
tween Chinese groups calculated on gene frequencies of 38 loci.
Sci China [C] 28:83– 89.
Excoffier L, Schneider S. 1999. Why hunter-gather populations do
not show sign of Pleistocene demographic expansions. Proc
Natl Acad Sci USA 96:10597–10602.
Excoffier L, Smouse PE, Quattro JM. 1992. Analysis of molecular
variance inferred from metric distances among DNA haplo-
types: application to human mitochondrial DNA restriction
data. Genetics 131:479 – 491.
Fu Y-X. 1997. Statistical tests of neutrality of mutations against
population growth, hitchhiking and background selection. Ge-
netics 147:915–925.
Ge JX, Wu SD, Chao SJ. 1997. Zhongguo yimin Shi [the mi-
gration history of China]. Fuzhou: Fujian People’s Press [in
Chinese].
Horai S, Murayama K, Hayasaka K, Matsubayashi S, Hattori Y,
Fucharoen G, Harihara S, Park KS, Omoto K, Pan IH. 1996.
mtDNA polymorphism in East Asian populations, with special
reference to the peopling of Japan. Am J Hum Genet 59:579 –
590.
Jin L, Su B. 2000. Natives or immigrants: modern human origin
in East Asia. Nat Rev Genet 1:126 –132.
Ma Y. 1994. China’s minority nationalities. Beijing: Foreign Lan-
guages Press.
Mountain JL, Hebert JM, Bhattacharyya S, Underhill PA, Otto-
lenghi C, Gadgil M, Cavalli-Sforza LL. 1995. Demographic his-
tory of India and mtDNA-sequence diversity. Am J Hum Genet
56:979 –992.
Nei M. 1987. Molecular evolutionary genetics. New York: Colum-
bia University Press.
Relethford JH. 1996. Genetic drift can obscure population his-
tory: problem and solution. Hum Biol 681:29 – 44.
Rogers AR. 1995. Genetic evidence for a Pleistocene population
expansion. Evolution 494:608 – 615.
Rogers AR, Harpending H. 1992. Population growth makes waves
in the distribution of pairwise genetic differences. Mol Biol Evol
9:552–569.
Saitou N, Nei M. 1987. The neighbour-joining method: a new
method for reconstructing phylogenetic trees. Mol Biol Evol
4:406 – 425.
Schneider S, Roessli D, Excoffier L. 2000. ARLEQUIN, version
2.0: a software for population genetic data analysis. Geneva:
University of Geneva.
Seielstad MT, Minch E, Cavalli-Sforza LL. 1998. Genetic evi-
dence for a higher female migration rate in humans. Nat Genet
20:219 –220.
72 Y.-G. YAO ET AL.
Su B, Xiao C, Deka R, Seielstad MT, Kangwanpong D, Xiao J, Lu
D, Underhill P, Cavalli-Sforza L, Chakraborty R, Jin L. 1999.
Y-chromosome evidence for a northward migration of modern
humans into eastern Asia during the last Ice Age. Am J Hum
Genet 656:1718 –1724.
Tajima F. 1989. Statistical method for testing the neutral
mutation hypothesis by DNA polymorphism. Genetics 123:
585–595.
Vigilant L, Stoneking M, Harpending H, Hawkes K, Wilson AC.
1991. African populations and the evolution of human mito-
chondrial DNA. Science 253:1503–1507.
Ward RH, Frazier BL, Dew-Jager K, Pä äbo S. 1991. Extensive
mitochondrial diversity within a single Amerindian tribe. Proc
Natl Acad Sci USA 88:8720 – 8724.
Wu R, Wu X, Zhang S. 1989. Early humankind in China. Beijing:
Science Press. In Chinese.
Yao Y-G, Lü X-M, Luo H-R, Li W-H, Zhang Y-P. 2000a. Gene admix-
ture in the silk road of China—evidence from mtDNA and melano-
cortin 1 receptor polymorphism. Genes Genet Syst 75:173–178.
Yao Y-G, Watkins WS, Zhang Y-P. 2000b. Evolutionary history of
the mtDNA 9-bp deletion in Chinese populations and its rele-
vance to the Peopling of East and Southeast Asia. Hum Genet
107:504 –512.
Yao Y-G, Yuan Z-G, Zhou Z-D, Geng P-L, Li Q-W, Zhang Y-P.
2001. Frequency of the mtDNA 9-bp deletion among Chinese
ethnic groups. Prog Nat Sci 11:358 –364.
You Z. 1994. History of Yunnan nationalities. Kunming: Yunnan
University Press [in Chinese].
Zhao TM, Zhang GL, Zhu YM, Zheng SQ, Gu WJ, Chen Q, Zhang
X, Liu DY. 1991. Study on Immunoglobulin allotypes in the
Chinese: a hypothesis of the origin of the Chinese nation. Acta
Genet Sin 182:97–108 [in Chinese].
APPENDIX. Variable sites of mtDNA hypervariable segment I sequences (HVSI) in 340 individuals from 9 Chinese ethnic populations and 32 individuals from a northern Thai
population, with respect to nucleotide position 16001–16497 of the reference sequence (Anderson et al., 1981). Only the last three digits are given, e.g., site 051 means 16051. Two
additional insertions of a C and a G in the C homopolymer in region 16258 –16263 in an individual from Zhuang population (Zh68), and in the G homopolymer in region
16470 –16474 in an individual from Sali population (sali21), respectively, are not included. The numbers of individuals sharing a haplotype are listed at right. Zhuang, Tibetan,
and Mongolian are abbreviated as Zhua, Tibe, and Mong, respectively

(Continued)
APPENDIX. (Continued)

(Continued)
APPENDIX. (Continued)

(Continued)
APPENDIX. (Continued)