Ancestral Origins and Genetic History of Tibetan Highlanders

15 Pages·2016·6.56 MB·English

by  Dongsheng Lu

Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
ARTICLE
Ancestral Origins and Genetic History
of Tibetan Highlanders
Dongsheng Lu,1,2,9 Haiyi Lou,1,9 Kai Yuan,1,2,9 Xiaoji Wang,1,2,3,9 Yuchen Wang,1,2,9 Chao Zhang,1,2,9
Yan Lu,1 Xiong Yang,1,2 Lian Deng,1,2 Ying Zhou,1,2 Qidi Feng,1,2 Ya Hu,4 Qiliang Ding,4 Yajun Yang,4
Shilin Li,4 Li Jin,4 Yaqun Guan,5 Bing Su,6 Longli Kang,7 and Shuhua Xu1,2,3,8,*
TheoriginofTibetansremainsoneofthemostcontentiouspuzzlesinhistory,anthropology,andgenetics.Analysesofdeeplysequenced
(303–603)genomesof38Tibetanhighlandersand39HanChineselowlanders,togetherwithavailabledataonarchaicandmodern
humans,allowustocomprehensivelycharacterizetheancestralmakeupofTibetansanduncovertheirorigins.Non-modernhuman
sequencescompose~6%oftheTibetangenepoolandformuniquehaplotypesinsomegenomicregions,whereDenisovan-like,Nean-
derthal-like, ancient-Siberian-like, and unknown ancestries are entangled and elevated. The shared ancestry of Tibetan-enriched
sequencesdatesbackto~62,000–38,000yearsago,predatingtheLastGlacialMaximum(LGM)andrepresentingearlycolonization
of the plateau. Nonetheless, most of the Tibetan gene pool is of modern human origin and diverged from that of Han Chinese
~15,000to~9,000yearsago,whichcanbelargelyattributedtopost-LGMarrivals.Analysisof~200contemporarypopulationsshowed
thatTibetansshareancestrywithpopulationsfromEastAsia(~82%),CentralAsiaandSiberia(~11%),SouthAsia(~6%),andwestern
EurasiaandOceania(~1%).OurresultssupportthatTibetansarosefromamixtureofmultipleancestralgenepoolsbutthattheirorigins
aremuchmorecomplicatedandancientthanpreviouslysuspected.Weprovidecompellingevidenceoftheco-existenceofPaleolithic
andNeolithicancestriesintheTibetangenepool,indicatingageneticcontinuitybetweenpre-historicalhighland-foragersandpresent-
dayTibetans.Inparticular,highlydifferentiatedsequencesharboredinhighlanders’genomesweremostlikelyinheritedfrompre-LGM
settlersofmultipleancestralorigins(SUNDer)andmaintainedinhighfrequencybynaturalselection.
Introduction human (AMH) or non-AMH (archaic hominid) species.
Furtherinvestigationisrequiredtoconfirmwhetherthere
Current knowledge of the origin and population history is a genetic continuity, or merely a continuity of culture,
of Tibetan highlanders is still very much in its infancy between the pre-historical and present-day populations
andcontroversial.Inparticular,twokeyquestionsremain of Tibetans.
unsolved:(1)whodidTibetansdescendfrom,and(2)how A recent study based on sequencing a single gene in
long have human beings been living at the Tibetan Tibetans suggested that aparticular haplotype motif con-
Plateau? Both archaeological and genetic studies have sisting of five SNPs surrounding EPAS1 (MIM: 603349)
documented a human presence on the plateau as early wasintrogressedfromDenisovans,anextinctnon-modern
as 30,000 years before present (YBP).1,2 Linguistic studies human species.7 This indicated the existence of a non-
havesuggestedthattheTibetanandChinesepeopleshare AMH-derivedsequenceinTibetangenomes.Nonetheless,
a common root ancestor and that the Tibetan-Chinese the extent to which the non-AMH sequences contribute
splittookplace~6,000YBP.3Arecentgeneticstudyutiliz- to the genetic makeup is unknown. Neither is it clear
ingexomesequencingdataestimatedadivergencetimeof where the non-AMH sequences in Tibetans originated.
2,750yearsbetweenTibetansandHanChinese.4Existing Were these sequences introduced indirectly by recent
archaeological and genetic data are not sufficient to migrationsofpopulationswithbothAMHandnon-AMH
address the discrepancy between times estimated by ancestries? Or were the archaic sequences inherited
different studies. Analysis of Y chromosome and mito- directly from early Tibetan Plateau inhabitants and are
chondrial DNA (mtDNA) has demonstrated the co- thusindependentofthosefoundinotherEastAsianpop-
existence of both Paleolithic and Neolithic ancestries in ulations? Answering these questions is indispensable for
the paternal and maternal lineages of present-day Ti- understanding the demographic history of Tibetans and
betans5,6buthasnotexplicitlydeterminedwhetherthose elucidating the mechanism of their adaptation to high-
Paleolithiclineagesoriginatedfromanatomicallymodern altitudeenvironments.
1KeyLaboratoryofComputationalBiology,MaxPlanckIndependentResearchGrouponPopulationGenomics,CAS-MPGPartnerInstituteforComputa-
tionalBiology,ShanghaiInstitutesforBiologicalSciences,ChineseAcademyofSciences,Shanghai200031,China;2UniversityofChineseAcademyof
Sciences,Beijing100049,China;3SchoolofLifeScienceandTechnology,ShanghaiTechUniversity,Shanghai200031,China;4StateKeyLaboratoryof
GeneticEngineeringandMinistryofEducationKeyLaboratoryofContemporaryAnthropology,SchoolofLifeSciences,FudanUniversity,Shanghai
200433,China;5DepartmentofBiochemistryandMolecularBiology,PreclinicalMedicineCollege,XinjiangMedicalUniversity,Urumqi830011,China;
6State Key Laboratory of Genetic Resources and Evolution, Kunming Institute of Zoology, Chinese Academy of Sciences, Kunming 650223, China;
7KeyLaboratoryforMolecularGeneticMechanismsandInterventionResearchonHighAltitudeDiseaseofTibetAutonomousRegion,SchoolofMedicine,
XizangMinzuUniversity,XianyangShaanxi712082,China;8CollaborativeInnovationCenterofGeneticsandDevelopment,Shanghai200438,China
9Theseauthorscontributedequallytothiswork
*Correspondence:[email protected]
http://dx.doi.org/10.1016/j.ajhg.2016.07.002.
(cid:1)2016AmericanSocietyofHumanGenetics.
TheAmericanJournalofHumanGenetics99,1–15,September1,2016 1
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
Existing archaeological and genetic data are not suffi- thelociwithanancestralallele(A,C,T,andGora,c,t,andg)
cient to address the fundamental questions mentioned matchingthe‘‘REF’’or‘‘ALT’’alleleintheVCFfiles.ASNP’sances-
above. This manuscript describes a whole-genome deep- tral state was not used if the SNP mapped to more than one
sequencing effort of Tibetan highlanders and samples of genomicregion.Finally,11,021,340SNVswithanunambiguously
derivedalleleremainedforcomparisonofderived-allelefrequency.
Han Chinese, a lowland-residing group that shares close
ancestry with the Tibetan people. A careful analysis of
PublicandPublishedData
these genomes, together with available data on archaic
AncientgenomesusedintheanalysisincludedthoseofanAltai
and modern humans, can provide further insights into
Neanderthal,14aDenisovan,15andUst’-Ishim,a45,000-year-old
thegeneticprehistoryandadaptationsofTibetans.
modern human from Siberia.16 The Affymetrix Human Origins
genotyping dataset17 and variants from phase 3 of the 1000
Material and Methods GenomesProject18werealsoincludedinthedataanalysisofthis
study(seeTablesS3andS4).
PopulationsandSamples
DeterminingYChromosomalandmtDNALineages
Peripheral-blood samples of 33 Tibetans (TBN) and 5 Sherpas
YchromosomalandmtDNAhaplogroupsweredeterminedonthe
(SHP)werecollectedfromsixprefectures(Lhasa,Chamdo,Nagqu,
basisofthekeymutationscommonlyusedfornomenclature.We
Nyingchi, Shannan, and Shigatse) in the Tibet Autonomous
developed an algorithm to search all possible combinations of
Region (Figure S1 and Table S1), and blood samples of 39 Han
the key mutations used for nomenclature from our sequence
Chinese (HAN) were collected from diverse regions in China
datatodeterminethefine-scalepaternalandmaternalhaplogroup
(Figure S1 and Table S1). Each individual was the offspring of a
affiliated with each sample (Tables S5 and S6). To compare the
non-consanguineousmarriageofmembersofthesamenational-
frequency distribution of worldwide populations (because fine-
ity within three generations. All samples were collected with
scalesub-lineagesofmostworldwidepopulationswerenotavail-
informed consent and approved by the Biomedical Research
able), we combined sub-lineages under each major paternal or
Ethics Committee of the Shanghai Institutes for Biological Sci-
maternalhaplogroup.
ences.Priortosequencingandanalysis,allsampleswerestripped
of personal identifiers (if any existed). All procedures were in
accordancewiththeethicalstandardsoftheResponsibleCommit- EstimatingGeneticDifferencesbetweenPopulations
tee on Human Experimentation (approved by the Biomedical GeneticdifferencesbetweenpopulationsweremeasuredwithFST
ResearchEthicsCommitteeoftheShanghaiInstitutesforBiolog- according to Weir and Cockerham;19 this metric accounts for
ical Sciences) and the Helsinki Declaration of 1975 (revised in differencesinthesamplesizeineachpopulation.TheSNP-specific
2000). FST between Tibetan and Han Chinese for each SNP was also
calculated with the same formula. Confidence intervals of the
F overlociwerecalculatedbybootstrapresamplingwith1,000
GenomeSequencingandDataProcessing ST
replications. To reduce the influence of large sample-size
Whole-genomesequencing,withhightargetcoverage(303–603)
differencesbetweenpopulations,wediscardedHuOriginpopula-
for150bppaired-endreads,wascarriedoutonanIlluminaHiSeqX
tions17withasamplesizesmallerthanfiveforpairwisecompari-
TenaccordingtoIllumina-providedprotocolswithstandardlibrary
son;theseincludedHimba(4),Kikuyu(4),Datog(3),Oromo(4),
preparationatWuXiNextCODE(Shanghai).Eachsamplewasrun
Italian_South(1),Canary_Islanders(2),Saami_WGA(1),Scottish
onauniquelanewithatleast90GBofdatathathadpassedfiltering,
(4),Dolgan(3),Tlingit(4),Australian(3),Chane(1),Chilote(4),
and read data were quality controlled so that 80% of the bases
Inga(2),Piapoco(4),Ticuna(1),andWayuu(1).Wealsodiscarded
achieved at least a base qualityscore of 30. Readswere merged,
SNPswithamissingratelargerthan20%inanysinglepopulation.
adaptor trimmed, and mapped to the human reference genome
Thisleftuswith563,816autosomalSNPs,whichweanalyzedto
(GRCh37) with the Burrows-Wheeler Aligner.8 Variant calling
estimatepairwiseF for188populations.
wascarriedoutwiththeHaplotypeCallermoduleintheGenome ST
AnalysisToolkit(GATK).9,10
Principal-ComponentAnalysis
WefilteredtheresultsofjointvariantcallingofHANandTIB
(TBNandSHP)samplesandretainedautosomalsingle-nucleotide Principal-componentanalysis(PCA)wasperformedattheindivid-
variants(SNVs)that(1)werebi-allelicinHANandTIBand(2)had uallevelwithEIGENSOFTv.3.0.20Toinvestigatefine-scalepopula-
a missing rate less than 20% in any one of the HAN, TBN, and tion structure and individual genetic affinities, we performed a
SHPpopulations.Finally,wewereleftwith11,452,436SNVsfor seriesofPCAsbygraduallyremoving‘‘outliers’’onthebasisofa
estimating minor allele frequency, heterozygosity, inbreeding plotof the first and secondprincipal components (PCs) and re-
coefficient, and runs of homozygosity, etc. Individual heter- analyzingtheremainingsamplesonthebasisofthesamesetof
ozygosity, identity by state, identity by descent, and inbreeding SNVmarkers.
coefficientwereestimatedwithPLINKv.1.07.11
Todeterminetheancestralandderivedalleleofeachvariantob- UsingOutgroupf3StatisticstoTestAncientAncestry
tained from our sequenced genomes and the public data inte- WeassessedthegeneticrelatednessbetweenancientandTIBge-
grated in our analysis, we downloaded the ancestral sequences nomesbycomputingoutgroupf statistics;thismethodisinsen-
3
fromthe1000GenomesProject;thesewereinferredfromsixpri- sitivetogeneticdriftandlessaffectedbysample-sizebiaswhen
matesintheEnredo-Pecan-Ortheus(EPO)pipeline.Weemployed ancientgenomesareinvolved.21
BEDToolsv.2.24.012,13togettheancestralalleleforeachlocusin Wecomputedf (X;ancient,African)accordingtoReichetal.21
3
thevariant callformat(VCF) files fromtheancestralsequences. to find the admixture signal from ancient genomes in modern
Intheanalysesrelatedtothederivedallelestates,weonlychose human populations. Ancient genomes used in the analysis
2 TheAmericanJournalofHumanGenetics99,1–15,September1,2016
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
included those of an Altai Neanderthal,14 a Denisovan,15 and sample size (or haplotype number) during removal of non-
Ust’-Ishim.16BAMfilesoftheancientgenomesweredownloaded modern human sequences, we directly masked the genomic
fromtherespectivereferencesandmanagedwithPicardv.1.112, regionsonthebasisofS*resultswithasignatureofnon-mod-
and the genotypes were called with GATK v.3.3.9 Data on DNA ern human ancestry identified in any of the genomes for
polymorphisms were stored in VCF files and managed with analysis. Therefore, we removed genomic regions with non-
VCFtools.22 X represents worldwide non-African modern hu- modern human ancestry rather than partially removing
man populations (23 Native American, 23 Central Asian and haplotypes.
Siberian, 25 East Asian, 4 Oceanian, 22 South Asian, and 58
West Eurasian). A significantly negative f (X; ancient, YRI
3 AncestryAnalysiswithADMIXTURE
[Yoruba in Ibadan, Nigeria]) (jZj > 2) suggests ancient gene
The model-based ancestry-estimation method ADMIXTURE32
introgression. There is no significant admixture signal from
was applied on the merged HuOrigin, TIB, and HAN dataset
any of the ancient genomes (Altai Neanderthal, Denisovan, or
consisting of 2,422 individuals from 205 worldwide popula-
Ust’-Ishim), given that all f values are significantly positive
3 tions. ADMIXTURE, which infers individual genetic-ancestry
(jZj>2).Therefore,theseanalysesdonotprovideclearinsight
coefficientsbyconditioningonaspecificnumber(K)of‘‘ances-
into the ancient admixture in modern humans, possibly
tral populations,’’ is a useful tool for analyzing population
because the analysis design of f (X; ancient, African) is not
3 structure. We used it mainly to identify the structure of TIB
powerful enough or the data are insufficient, for example, the
andHANinthecontextofworldwideandregionalpopulations.
admixtureproportionfromancientgenomesissmalloradmix-
Because the model in ADMIXTURE does not consider linkage
tureoccurredtoolongago.
disequilibrium (LD), we used PLINK v.1.0711 to prune the
Further,weusedan‘‘outgroup’’f statistic,f (African;ancient,
3 3 original dataset with dense SNPs. After we assigned an r2
X), according to Raghavan et al.23 to examine the relationship
threshold of 0.4 in every continuous window of 200 SNPs
between modern human populations and ancient genomes
advanced by 25 SNPs (--indep-pairwise 200 25 0.4), 288,979
(Altai Neanderthal, Denisovan, and Ust’-Ishim). Without
SNPs remained for the ADMIXTURE analysis.32 We ran
gene flow from an ancient population, f (YRI; ancient, X) has
3 ADMIXTURE with a random seed for the merged dataset
the same value for a different X. A larger f value suggests
3 from K ¼ 2 to K ¼ 20 with default parameters (--cv ¼ 5) in
more allele sharing with ancient individuals and ancient gene
tenreplicatesforeachK.
introgression.
Becausetheindividualancestryproportionvariedsignificantly
whenKwaslarge,weusedCLUMPP33toexaminetheconvergence
of different runs in the ten replicates for each K; the optimal
EstimationofN andDivergenceTime
e ancestry proportion was defined by the single replicate that
We applied pairwise sequentially Markovian coalescent (PSMC)
showedtheclosestvaluetotheancestryproportionthatappeared
analysis24 and multiple sequentially Markovian coalescent
most commonly among the ten replicates. We also assessedthe
(MSMC) analysis25 to infer the long-term effective population cross-validation error in the ten replicates to find the best K of
sizes (N) and time of divergence (T ) from the high-
e divergence theancestralpopulations.
coverage genomes. PSMC analysis was implemented according
to the original procedure.25 For MSMC analysis, rather than
phasingthedatafrombamCaller.pydirectly,weusedonlythose EstimatingAncientGeneFlowontheBasisof
SNVswith allelestatesidenticaltothoseof genotypescalledby f4Statistics
bamCaller.py fromthe VCFresultsfrom GATK analysis,and we Weusedanf test(Dtest),D(African,X,Neanderthal,Denisovan),
4
phasedthedatawithSHAPEIT2.26 accordingtoPattersonetal.34,35todeterminethemainsourceof
Ourestimationofeffectivepopulationsizeovertimeandthat gene flow (either Neanderthal or Denisovan) to the target pop-
of divergence times and time to the most recent common ulations (particularly TIB). A significantly negative D value
ancestor (TMRCA) in this study depends on estimates of the indicates more Neanderthal than Denisovan allele sharing in
germlinemutationrate(m).Toovercomeuncertaindateestimates population X, a non-African modern human population. In
duetoalackofconfidenceinthetruevalueofthemutationrate, contrast, a significantly positive D value (Z > 2) suggests more
we scaled time in units of 2mTand effective population size in gene introgression from Denisovans than from Neanderthals.
unitsof2mN 3103(wheremisthemutationrate,N istheeffec- WiththeexceptionofOceanianpopulations,non-Africanpopula-
e e
tivepopulationsize,andTistimeingenerations).Wealsocalcu- tions always share significantly more alleles with Neanderthals
lated absolute estimation by assuming a fast mutation rate of thanwithDenisovans.
1.0 3 10(cid:2)9 per site per year (the commonly used phylogenetic UsingD(African,ancient,TIB,X)accordingtoDurandetal.,17,34
mutation rate, i.e., ~2.5 3 10(cid:2)8 per base per human generation wefurtherdeterminedwhetheracertainancientgenomecontrib-
for a generation time of 25 years)27 or a slow mutation rate of utesmoreancestrytoTIBthantoanyothernon-Africanmodern
0.5 3 10(cid:2)9 per site per year (~1.25 3 10(cid:2)8 per base per human humanpopulation.AsignificantlypositiveDvalueindicatesthat
generationforagenerationtimeof25years).28–30Inmostocca- ancientgenomesshareahigherproportionofalleleswithpopula-
sions,we adoptedaslow mutation rate(0.5 3 10(cid:2)9 persite per tionXthanwithTIB.
year)becauseresultsbasedontheslowmutationrateagreebetter Finally,webasedfourdesigns(fourequations)onf statisticsto
4
with the paleoanthropological record and with estimates from quantitatively estimate the ancient ancestry in modern human
mtDNA.31 populationstofindoutwhetherthereisexcessarchaicancestry
Tostudytheinfluenceofthenon-modernhumansequences inTIB.Wefirstrandomlyselectedapopulationasthereference
on the estimation of N and divergence time, we removed population(ref)andobtainedaseriesofratiovalues.Thepopula-
e
non-AMH sequences from the genomes used for analysis. To tionshowingthesmallestratiovalue,whichisalwaysnegative,
facilitate the analysis and avoid any bias of the fluctuation of mighthavetherelativelysmallestgeneintrogressionfromancient
TheAmericanJournalofHumanGenetics99,1–15,September1,2016 3
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
populations. Then, we used these reference populations to esti- astatewithahigherE-allelerateascandidatesforarchaicintro-
mateancientgeneflowintheothernon-Africanpopulations: gressionsandimplementedathree-steppedapproachtofilterfalse
positives.42–44Foreachcandidatesegment,wefirstreconstructeda
ratio ¼ Sðref; X; ancient; AfricanÞ 36 (Equation1) phylogenetictreeforthecandidatesegment,alongwithNeander-
Sðref; ancient; ancient; AfricanÞ thal, Denisovan, African, and chimpanzee sequences. To detect
Neanderthal-likeandDenisovan-likesequences,werequiredthe
AsinEquation1,weusedchimpanzee(chimp)asanoutgroup
segment to coalesce first with Neanderthals and Denisovans,
and used Africans (YRI) as a reference population, which was
respectively. Second, we estimated the divergence time between
assumedtohavenoancientgeneflow:
thecandidateandNeanderthalsegments,denotedast (orbe-
Nean
ratio ¼SðASfðrAicfarnic;aann;cXie;natn;cainencite;ncth;icmhpimÞpÞ36 (Equation2) tdweteeecnt cNaenadniddearttehaanl-dlikDeesneiqsouveanncesse,gwmeenretsq,udireendotteNdeanastotDebnei).leTsos
than550,000yearsago14andt tobelessthant .Todetect
Nean Deni
InviewofthesimilaritybetweenNeanderthalsandDenisovans, Denisovan-like sequences, we required t to be less than
Deni
thepreviousformulasmightoverestimatethegeneflowofeach 550,000yearsagoandt tobelessthant .Third,werequired
Deni Nean
ancientpopulation.So,wealsousedanotherancientpopulation thegeneticlengthofthesegmentstobeconsistentwithanintro-
asanoutgrouptocontrolsuchaneffect: gressiontimelaterthan550,000yearsagotorejecttheancestral
polymorphismmodel.
SðAfrican; X; Neanderthal; DenisovanÞ 37
ratio ¼ OnthebasisofE-allelesandthealgorithmdescribedabove,we
SðAfrican; Neanderthal; Neanderthal; DenisovanÞ
further extended the method to the haplotype level and imple-
(Equation3)
mented the algorithm in a computer program, ArchaicSeeker,
In addition, we can also estimate the non-ancient admixture which is a more heuristic method of detecting archaic DNA se-
proportion.Here,NeanderthalsandDenisovansarereferencepop- quences in present-day human genomes. For those detected
ulationsforeachother: archaicgenomicsegments,weclassifiedthemintothreegroups:
Neanderthal-like,Denisovan-like,andunknownarchaic.Thefirst
Sðref; X; African; chimpÞ 37 twogroupswereresultsfromthethreemethodsbasedoncompar-
ratio¼1(cid:2) (Equation4)
Sðref; African; African; chimpÞ ativeanalysesofthearchaicgenomes(DenisovanorNeanderthal).
However, the unknown archaic segments, which could be de-
tectedonlybytheS*method,weredefinedasarchaic-likesignals
DetectingArchaicSequencesinModernHuman that cannot be attributed to either modern human genomes or
Genomes sequencedarchaicgenomes.
We further locally identified genomic segments of non-modern
human origins by using the S* statistic38 and ArchaicSeeker, a
Local-AncestryInference
methoddevelopedinthisstudy.Theprincipleofthismethodis
Tounderstandthegeneticmakeupofsomeinterestingregionson
based on the high divergence between modern humans and
afinerscale,wedevelopedamethodonthebasisoftheresultsof
archaichominins38–40andthefindingthatarchaichomininintro- ChromoPainter(cp)45toobtainthefine-scaleancestrymakeupof
gressions are absent from sub-Saharan Africans.41 We defined
agivengenomicregion.ToinferthelocalancestriesofTIBwith
‘‘E-allele’’ as an allele absent from Africans but observed in Eur-
cp,weselectedHanChinesegenomesandsomeavailablearchaic
asians.38–40Insimulations,wefoundthattheE-allelerate(defined genomes,includinganAltaiNeanderthalgenome14andaDeni-
asthenumberofE-allelesonasegmentdividedbythenumberof sovan genome,15 as reference data. Moreover, because we
polymorphicsitesintheregionspanningthesegment)ismuch
observed that Tibetans shared a considerable proportion of
higheronarchaicintrogressivesegmentsthanonmodernhuman
ancestry with Siberian populations, we also included the Ust’-
segments. We further implemented a hidden-Markov-model- Ishim as a reference.16 In addition, to avoid the false-positive
based method to partition each Eurasian chromosome into seg-
inference of archaic ancestral segments, we included full
ments with different E-allele rates and labeled each segment
sequencedataofYRIfromphase3ofthe1000GenomesProject
with one of the two hidden states. One hidden state represents
inourreferencepopulationpanel.Withthisreferencepanel,we
segments that are of modern human origin and have a lower
classified the local genomes into six categories regarding
E-allele rate, and the other one represents segments that are of
their ancestry: Denisovan-like, Neanderthal-like, Ust’-Ishim-like,
archaichomininoriginandhaveahigherE-allelerate.Theprocess
Han-Chinese-like,African-like,anduncertainancestry.
ofthealgorithmwasasfollows.Wefirstsetapenaltyforstatetran-
sitions(l)ofthealgorithmandinitialE-alleleratesfortwostates.
TheinitialE-alleleratesaretheobservedE-alleleratesonmodern AnalyzingtheCorrelationbetweenAncestry
Eurasianandarchaichominingenomes.Second,weimplemented andAltitude
the classic Viterbi algorithm to partition the Eurasian chromo- Toinvestigatewhethertherecouldbeanyrelationshipbetweenthe
somes into segments and labeled each segment with a hidden ancestralcompositionofindividualsandthealtitudewherethey
state. Third, with segments from step two, we re-estimated the reside,wegroupedTibetanindividualsaccordingtogeographical
E-allele rate for both hidden states. Fourth, we repeated steps regions and calculated the correlation between percentages of
two and three until convergence. In the Viterbi algorithm, we non-modernhumansequencesandaltitude.Inthisanalysis,Han
usedthesameinitialprobabilityforeachstate.Wesetthetransi- ChineseandSherpasampleswerenotincluded,althoughamuch
tion probability between two states as e(cid:2)l and set the emission strongercorrelationwouldbeexpectediftheywere.TheTibetan
probabilityofbothstatesastheirE-allelerates. samplesweregroupedintosevenregionalpopulationsaccording
After we partitioned each chromosome into segments and to the information provided in Table S1; Shigatse samples were
labeledahiddenstateforeachsegment,weselectedsegmentsin grouped into two populations because three individuals were
4 TheAmericanJournalofHumanGenetics99,1–15,September1,2016
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
Figure 1. Genetic Affinities of TIB in the
A ContextofWorldwidePopulations
Africa (A)Afan-likechartshowinggeneticdiffer-
WestEurasia ences (F ) between TIB and worldwide
ST
SouthAsia populations. Each branch represents a
0.12 Oceania
comparisonbetweenTIBand 1ofthe256
EastAsia
0.09 populations; lengths are proportional to
CentralAsiaSiberia the F values, indicated by gray circles.
0.06 America ThepSoTpulationsareclassifiedbygeograph-
ical regions and indicated with the colors
0.03
shown in the legend. The populations in
each region are presented in a clockwise
TIB order according to great-circle distance to
Tibet.
(B)Afan-likechartshowingF betweenTIB
ST
andEastAsianpopulations.
(C) A fan-like chart showing F between
ST
TIB and Central Asian and Siberian
populations.
p
u
k
B Atayal C Sel _Pamir
0.08 Tajiks p
Han_KNHoXorAeirNbTtaNuhaojDinYxaaiuizruHaTnuHezhen0000T....MiaozuI0000B1234JapaneOroqeKninShheLCADaTUmaihahumyiagbuiordian TKuAyrvlgUtianyYlziicaKaaihkaanEniluYnvmtuMekyaongknhigrolUzbek000T...TubaIla000rB246 TurkmenMansiChukNcEghsKiIakAitonemlrlaeymosuaaetkninan mtnwgwieuohhanncteeolesrremdeeoatwimticdmdoaiHbaterTbhubeMmi.g¼dstiTihRoflth(cid:2)feaCnhebidesrAfwHoe3Amun¼lrniommtccuThHc2e(cid:2)tuaHua3llAmlubteamnieCnmmcobtHh(cid:2)ganwueuibtmHmltehrou3apCeamwthntliCiaemo:dlbhatnpeicbmhfn;arpenroo;amthbteeuespmoeotsfahstniiae--
s
e
reference genome and that of the most
recent common ancestor of humans and
fromaratherdifferentaltitude(Dingri,4,300m).Therefore,seven chimpanzees of region ab. THum (cid:2) HumChimp is the divergence
regional populations—Lhasa (3,650 m), Nyingchi (3,000 m), time of humans and chimpanzees, and 13 million years was
Chamdo(3,240m),Shannan(3,573m),Shigatse(3,853m),Nagqu usedinthisstudy.
(4,522 m), and Dingri (4,300 m)—were used for this analysis.
The altitude for each region was averaged over the number of
individuals residing in a given region. Accordingly, the Results
proportion of sequences from each category (non-modern
human sequences, Neanderthal-like sequences, Denisovan-like GeneticAffinitiesofTIBintheContextofGlobal
sequences, Neanderthal-specific sequences, Denisovan-specific Populations
sequences, and unknown archaic sequences) was averaged Acomparisonofgeneticdifferencesmeasuredbyunbiased
over the number of individuals in the same region. Pearson’s F 19betweenTIBand188worldwidepopulations17(Tables
correlation coefficient (R) and coefficient of determination (R2) ST
S3andS4)showedthatTIBismostcloselyrelatedtoEast
werecalculatedonthebasisoftheestimatedancestryproportions
Asianpopulations(Figures1Aand1B),followedbyCentral
andaltitudedata.
AsianandSiberianpopulations(Figure1C).Theserelation-
shipsremainedconsistentinseparateanalysesofTBNand
EstimatingTMRCA
SHPsamples(FiguresS2andS3).Inaddition,theseresults
ToestimatetheTMRCAofagroupofsequences(orhaplotypes)ina
werealsoconfirmed by PCA (Figure 2and Figures S4 and
certainregion,wefirstcalculatedtheaveragepairwisenucleotide
differences of these sequences ðpÞ according to the following S5) and outgroup f3 statistics21 (Figure S6). Interestingly,
formula: TBNandSHParedistinguishableinatwo-dimensionalPC
P P plot(Figure2D),indicatingthattheyaretwodistinctgroups
p¼2 3 ni¼(cid:2)11 nj¼iþ1pij; withgeneticdifferences(FST¼0.010),slightlysmallerthan
ðn þ 1Þ 3 n thosebetweenTBNandHAN(F ¼0.011).Theoverallge-
ST
wherenisthenumberofsequencesanalyzedinagivengenomic netic makeup of TIB is closer to surrounding populations
region ab, and pij is the nucleotide difference between two se- livingintheplateau,suchasTu(FST¼0.006),Yizu(FST¼
quences i and j (i s j). The TMRCA can be estimated with the 0.007),andNaxi(FST¼0.009),thantoanyotherpopula-
followingformula: tionworldwide(Figure1BandTablesS7andS8),mostlikely
TheAmericanJournalofHumanGenetics99,1–15,September1,2016 5
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
A B Figure 2. Analysis of the First Two PCs
of TIB Individuals and Other Population
Samples
Geographicalregionswheretheindividuals
are located are indicated with the colors
showninthelegend.Numbersinbrackets
denotethevarianceexplainedbyeachPC.
The HAN samples are classified into East
Asianpopulationsandarenothighlighted
inthisillustration.
(A) In this two-dimensional PC plot, TIB
samplesarelocatedcloselyintheEastAsian
cluster and share similar PC coordinates
with some populations in South Asia and
CentralAsiaandSiberia.
(B) PCA of TIB individuals and Eurasian
samples (western Eurasian samples were
C D excluded).
(C) PCA of TIB individuals and some East
Asiansamples.
(D) PCA of TBN and SHP individuals and
samplesofsomecloselyrelatedsurrounding
populations(Tu,Yizu,andNaixi).
example, the frequency of B5 in TBN
(0%) was quite different from that in
HAN (10.26%) (Table S6). Some hap-
logroups, such as A and M5, also
showed substantial differences be-
tweenSHPandTBN;however,because
asaresultofdirectancestrysharingorreciprocalgeneflow samplesizes,especiallythatofSHP,inourdataweresmall,
betweenthesepopulations. estimationofhaplotypefrequencyisnotreliable.
Wecautionthattherelationshiprevealedbyanoverall TIBshowedanoveralllowergeneticdiversitythanHAN,
analysis of the genome-wide data only reflects a limited includingalowerlevelofheterozygosity,ahigherlevelof
aspectoftheoriginsofthepopulationsstudied.Theresults runsofhomozygosity(FigureS8),andaconsistentlysmaller
should not be directly interpreted as the genetic origins effectivepopulationsizeoverthelast~15,000years(Figures
andhistoryoftheTibetans;rather,theyreflectthepopula- 3and4A),mostlikelybecauseTIBisisolatedontheTibetan
tion’s present-day genetic makeup, which could be sub- Plateau,whereas HAN has experienced recent population
stantiallyinfluencedbyrecentgeneflowfromsurrounding expansion. The average time of divergence between TBN
populations. This highlights the need to account for andHANwasestimatedtobe~15,000–9,000YBPontheba-
admixturewheninferringpopulationhistory. sisof a sequentially Markovian coalescent analysis24,25 of
high-coverage genomes. This divergence time between
GeneticRelationshipandDivergencebetweenTIB TibetansandHanChineseisthusmuchearlierthanthees-
andHAN timateof2,750yearsagobyapreviousstudy.4Incontrast,
Among the lowland populations, the data indicated that theestimateddivergencetimebetweenSHPandHANwas
HanChinesearemostcloselyrelatedtoTIB(F ¼0.011) ~16,000–11,000YBP,andthatbetweenTBNandSHPwas
ST
(Figure1BandTableS7);thetwopopulationshadanover- ~11,000–7,000 YBP (Figure 4B). The divergence between
all strong correlation in the allele-frequency spectrum TIB and HAN most likely resulted from recent migration
(Figure S7), but Y chromosome and mtDNA data showed to the Tibetan Plateau after the Last Glacial Maximum
substantial differences between them (Tables S5 and S6). (LGM), a period of intense cold from ~26,500 to 19,000
Y chromosome haplogroup D-M174 is an East-Asian-spe- YBP.49 Subsequent gene flow between TBN and HAN or
cific male lineage and is rare in populations from regions continuous migrations to the plateau most likely caused
borderingEastAsia(CentralAsia,NorthAsia,andtheMid- thedivergencetimebetweenTBNandHANtobeslightly
dleEast),usuallylessthan5%.6,46–48Inourdata,D-M174 shorterthanthatbetweenSHPandHAN.
lineages were not observed in the 39 HAN samples (fre-
quency was 0) but had a high frequency in TIB (66.6%; AncestralMakeupofTIB
Table S5), indicating a very large difference in Y DNA AncestryanalysiswithADMIXTURE32suggestedthatpre-
lineages between the two groups. Such a large difference sent-day Tibetans share the majority of their ancestry
was not observed in mtDNA, but some haplogroups also makeup with populations from East Asia (~82%), Central
showed considerable differences between groups. For Asia and Siberia (~11%), and South Asia (~6%) and have
6 TheAmericanJournalofHumanGenetics99,1–15,September1,2016
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
5−10 kyr 50−100 kyr 0.5−1 Myr 5−10 Myr A
310) 4
× HAN
Ne3.5 TBN
µ4 SHP
of 3
s
unit 2.5
n
d i 2
e
al
sc 1.5
e (
siz 1
n
atio 0.5
ul
op 0
P
10−5 10−4 10−3 10−2
Time (scaled in units of 2µT)
Figure 3. Estimated Changes in Effective Population Size B
overTime
The estimation was based on PSMC analysis of single genomes
from three modern human groups with deeply sequenced ge-
nomesinthisstudy(33TBN,5SHP,and39HAN).Toovercome
uncertaindateestimatesdue toa lack ofconfidencein thetrue
valueofthemutationrate,wescaledtimeinunitsof2mT(where
misthemutationrateandTistimeingenerations).Wealsopro-
videanabsolute estimationoftime(top)undertheassumption
ofafastmutationrateof1.0310(cid:2)9oraslowmutationrateof
0.5310(cid:2)9persiteperyear.
minorancestralrelationshipswithwesternEurasian(<1%)
and Oceanian (<0.5%) populations (Figure 5 and Figures
S9–S11). In contrast, HAN share much less ancestry with
Siberian (~7%), South Asian (<0.5%), and Oceanian
(~0%) populations but higher ancestry with East Asian
populations(>90%). Figure4. EstimatedChangesinHistoricalEffectivePopulation
SizeandDivergenceTimebetweenPopulations
Weappliedoutgroupf statistics21andf statistics17,34to
3 4 (A)TheestimationwasbasedonMSMCanalysisoffourgenomes
assessancientancestralcontributionstoTIBbyanalyzing (eighthaploidgenomes)randomlyselectedfromeachofthethree
available ancient genomes, including those of an Altai modern human groups with deeply sequenced genomes in this
Neanderthal,14 a Denisovan,15 and Ust’-Ishim, a 45,000- study (TBN, SHP, and HAN). The analysis was repeated twice
with different combinations of different individuals, and two
year-old anatomically modern human from Siberia16 (see
curves are displayed for each group in the plot with different
Material and Methods). TIB was found to share slightly colors.Curveswithdashedlinesdenotetheestimationbasedon
morealleleswithancientgenomesthanmanyworldwide thesamesetofgenomesbutwithnon-modernhumansequences
removed(AMHonly).
populations, except for Oceanian populations, which
(B)TheestimationwasbasedontheMSMCalgorithm.Twoindi-
showed significantly higher shared archaic (both Deniso-
vidualgenomeswererandomlyselectedfromeachgroupinthe
van- and Neanderthal-like) ancestry than TIB, and East group pairs (TBN versus HAN, SHP versus HAN, or SHP versus
Asianpopulations(especiallythosesurroundingTIB:Yizu, TBN), and in total four genomes (eight haploid genomes) were
used for each MSMC analysis. The analysis was repeated eight
Tu, andNaxi),whichhadsimilarorevenhigher levelsof
times,andeightcurvesaredisplayedforeachgroupintheplot
sharedarchaicancestrythanTIB(TablesS9,S10,S11,S12,
withdifferentcolors.Here,weshowtheresultsbasedonabsolute
S13,S14,andS15andFiguresS12–S26).Inparticular,no estimationoftimeundertheassumptionofaslowmutationrate
modern human populations shared more alleles with of0.5310(cid:2)9persiteperyear.
Ust’-IshimthanTIB(TableS13andFigureS14).Although
theexactestimatesvariedandtheabsolutevaluesarenot ofnon-AMHsequencesintheTIBgenepool(6.17%)was
comparable, these genome-wide analyses indicate that significantly higher than that in the HAN gene
TIBsharesconsistentlyhigherancestrywithancientnon- pool(5.86%)(p<10(cid:2)5)(Figures6AandFigureS27A).Re-
AMH genomes than with HAN, the lowland population stricting the comparison to Neanderthal-like and Deniso-
thathastheclosestmodernrelationshipwithTIB. van-like sequences, we did not observe any significant
differencesbetweenTIBandHANasapercentageofNean-
IndividualArchaicAncestryIsHigherinTIBThan derthal-likesequencesperindividual(Figure6B),although
inHAN we did observe some marginally significant differences
Genomic segments of non-AMH origins were identified whenwerestrictedthecomparisontoNeanderthal-specific
with the S* statistic,38 as well as two methods developed sequences (p ¼ 0.01; Figure S27B). Significant differences
inthisstudy(seeMaterialandMethods).Thetotalamount werealsoobservedbetween TIB andHAN onthe basisof
TheAmericanJournalofHumanGenetics99,1–15,September1,2016 7
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
A a
LBNiethUolauIrkOcrwaruearnesclHnaSigaiaCauinciEaFdannnzdoSCrninngeeatipgcrtnacnaioCslFcrhinaaBshihriantseha_ianhanSsr_nqcyoNS_huuIpoeIsttharBlaatnluIhintaialsgdnlihaea_rnrNia_oSnArotlhbuatThnuisacnsahGnkyreeSneiackzilyia_SnJaerwdiniManaltesCeypriot 11.63.Karitian1%%SuruiWayuuTicunaChanePiapocoCabecarMixeGuaranKiaqchAiykelmaPriamaMixteZcaIpnotgeacQueBcolhiMuviaaayCnahilCohtiAelpgOejoiCwnbryTqelawiueniaAnlnEgeistCutkKiihItoaumMrenklYyoacNmEuaDhnYikkgvSeTsaoaeianuelkgnnglbuhkaaiatusrnlpaanr
Estonian 81.6% Altaian
Finnish Tuvinian
Mordovian Ulchi
Russians_North TIB Oroqen
Chuvashe Hezhen
Armenian Daur
Georgian Xibo
GeorAgNiboaCkrnthhYAh_eeaOdJLmcsKeyeshieBuagwszenmanheigKntlNtYeiikyeanealoak_BmmngrJIeSreaeydiaankwouqPeiuadisl_iIneeJrBDsareBetinudiwnzioaJeaunoirn_nLdJeAaebniwTauanrnekiSssyrehi_aJInreawniTuarnkiMsakhraBnirahBuailochKialashPathanSi7ndhiG.uja0ratiA%BurushoCochin_JewTiwari9GujaratiB2.GujaratiC3%MalaGujaratiDLodhiPunjabiVishwHabrahBenAgaliKNhariaOngeKusunHdaazarBaougAauinsvtPirlalaelipaunTaanjiTkus_rUkPmzabmeKeniyrkUrgyyAgzutiaAarnymSailhDeaMiHiaTaounKzjuiLianaKhhoHHurTaeAYInNaBNi_zJnaTCTNuaxMuhpaoiaarmotinhnbegosodelian
m
in
Figure5. SummaryPlotofGeneticAdmixture
Theresultsofindividualadmixtureproportionsestimatedfrom592,799autosomalSNPswithgenotypedataavailablefor38TIB,39
HAN,and2,345HuOriginsamples(Africansampleswerenotincluded).Eachindividualisrepresentedbyasinglelinebrokeninto
K¼7coloredsegmentswithlengthsproportionaltotheK¼7inferredclusters.ThepopulationIDsarepresentedoutsideofthecircle
oftheplot.Theresultsofpopulation-leveladmixtureofTIBandHANaresummarizedanddisplayedinthetwopiechartsinthecenterof
thecircleplot;admixtureproportionsaredenotedaspercentagesandwithdifferentcolors.
Denisovan-like sequences per individual (p ¼ 0.001; tion(see Material and Methods), a strong correlation was
Figure 6C), and this difference was more pronounced observed between the average proportion of non-AMH
whenthecomparisonwasrestrictedtoDenisovan-specific sequences present in Tibetan individuals and altitude
sequences(p<10(cid:2)5;FigureS27C).Together,thisanalysis (PearsonR2¼0.855;p¼0.0083;Figure6D),butnosignif-
suggeststhatthemajordifferencesbetweenTIBandHAN icantcorrelationwasobservedinrelationtoNeanderthal-
resultedfromthecontributionofDenisovan-likeandun- likeorDenisovan-likesequences(Figures6Eand6F).These
knownnon-AMHsequences(Figure6andFigureS27). results indicate that non-AMH sequences have some
association with altitude, possibly contributing to the
AltitudinalCorrelationofArchaicAncestry adaptation of Tibetans, but the main contribution seems
Interestingly,whentheTibetansamplesweregroupedinto to originate from unknown non-AMH ancestry
sevenregionalpopulationsaccordingtogeographicalloca- (Figure S27D) rather than from Neanderthal-like or
8 TheAmericanJournalofHumanGenetics99,1–15,September1,2016
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
A D Figure 6. Distribution of Non-modern
Human Sequences among Individuals and
CorrelationswithAltitude
(A)Acomparisonofthepercentageofnon-
modernhumansequencesperindividualbe-
tween TIB (6.17 5 0.10) and HAN (5.86 5
0.07) demonstrates significant differences
(p < 10(cid:2)5). On the horizontal axis, T, S,
andHdenoteTibetan,Sherpa,andHanChi-
nese,respectively.
(B)AcomparisonofthepercentageofNean-
B E
derthal-like sequences per individual be-
tween TIB (1.04 5 0.04) and HAN (1.02 5
0.04) shows significant differences (p ¼
0.379). On the horizontal axis, T, S, and H
denote Tibetan, Sherpa, and Han Chinese,
respectively.
(C)AcomparisonofthepercentageofDeni-
sovan-likesequencesperindividualbetween
TIB (0.42 5 0.02) and HAN (0.40 5 0.02)
shows slightly significant differences (p ¼
C F
0.001); the statistical significance (p value)
was obtained by permutationtests repeated
100,000 times.Onthehorizontal axis,T,S,
andHdenoteTibetan,Sherpa,andHanChi-
nese,respectively.
(D)Correlationbetweentheaveragepropor-
tion of non-modern human sequences in
regional Tibetan populations and altitude
(PearsonR2¼0.855;p¼0.0083).
(E)Correlationbetweentheaveragepropor-
tionofNeanderthal-likesequencesperindi-
vidual and altitude (Pearson R2 ¼ 0.216;
p¼0.354).
(F)Correlationbetweentheaveragepropor-
tionofDenisovan-likesequencesperindivid-
ual and altitude (Pearson R2 ¼ 0.412;
p¼0.170).
Denisovan-like sequences. A similar pattern and conclu- some2,encompasseseightgenes(EPAS1,LOC101805491,
sion was apparent when the analysis was restricted to TMEM247, ATP6V1E2, RHOQ [MIM: 605857],
Neanderthal-specific and Denisovan-specific sequences LOC100506142, PIGF [MIM: 600153], and CRIPT [MIM:
(FiguresS27EandS27F). 604594]), and has extreme differences between TIB and
HAN (F > 0.65) for SNVs with a high frequency of
ST
AHighlyDifferentiatedGenomicRegionwith derived alleles. These differences are in sharp contrast to
ElevatedArchaicAncestryinTIB thegenome-wideaverage(F ¼0.011).
ST
Theaboveresultsrevealthatnon-AMH-derivedsequences
not only exist in the Tibetan genomes but are also very EntangledAncestriesandTheirAncientOriginsin
common. However, it is not clear whether these non- the~300kbRegion
AMH sequences derived from recent migrations to Tibet The ancestral pattern in the ~300 kb region is extremely
withbothAMHandnon-AMHancestriesorwereinherited complicated, such that it contains a mix of Denisovan,
fromtheancientcolonizationoftheTibetanPlateau.The Neanderthal, ancient Siberian, and unknown archaic an-
overall spatial distributions of non-AMH-derived se- cestries,whichareelevatedinTIB(TableS16andFigure7).
quences are similar between TIB and HAN (Figure S28), The unusually high frequency of archaic sequences and
andtheTMRCAfornon-AMHsequencesinbothTIBand substantial differences between TIB and HAN—as well as
HANwasestimatedtobe>40,000years(seeMaterialand other worldwide populations—cannot be explained by
Methods). These results might be not surprising, given recent gene flows or incomplete linage sorting (Figure 7
that TIB and HAN share recent genetic drift (Figures 1 and Table S17). These highly differentiated sequences
and2)andthemajorityoftheirgeneticmakeup(Figure5). harbored in the genomes of present-day Tibetan high-
Nonetheless, when frequency was taken into account, a landers were most likely ‘‘inherited’’ from earlier settlers
~300kbregionwasfoundtobeconsiderablydifferentbe- rather than ‘‘introgressed’’ by later arrivers. The compli-
tween TIB and HAN, such that both Denisovan-like and cated ancestral architectures in these regions have many
Neanderthal-like ancestries were significantly elevated in implications for Tibetan origins and their pre-history.
TIB(FigureS29).This~300kbregionislocatedonchromo- The surviving archaic sequences in the ~300 kb region
TheAmericanJournalofHumanGenetics99,1–15,September1,2016 9
Pleasecitethisarticleinpressas:Luetal.,AncestralOriginsandGeneticHistoryofTibetanHighlanders,TheAmericanJournalofHuman
Genetics(2016),http://dx.doi.org/10.1016/j.ajhg.2016.07.002
Figure 7. Local-Ancestry Inference of
a~300kbRegioninTIBandEightOther
Populations
The local ancestry was jointly inferred
with ArchaiSeeker and ChromoPainter45
(see Material and Methods). The upper
panelshowsgeneslocatedinthe~300kb
region (chr2: 46,577,796–46,870,806).
The two vertical arrows indicate the two
previously identified Tibetan-specific
varianttags,thefive-SNPmotif,7andthe
Tibetan-enriched deletion (TED).50 The
positionsofancestry-informativemarkers
(AIMs)fortheDenisovan(DENAIM),the
Neanderthal (NEA AIM), and Ust’-Ishim
(UST AIM) are also displayed in the
following rows. For the AIMs of ancient
samples identified here (DEN, NEA, and
UST), we required the frequency of the
ancient samples’ alternative allele (with
respect to reference genome GRCh37) to
be >0.5 in TIB and <0.1 in other world-
wide populations. The lower panels
exhibittheinferredancestryofhaplotypes
inTIB,HAN,CHB(HanChineseinBejing,
China; 1000 Genomes), CHS (Southern
Han Chinese; 1000 Genomes), JPT (Japa-
nese in Tokyo, Japan; 1000 Genomes),
PAP (Papuan),51 SIB (Siberians),51–53 SSIP
(Singapore Indians),54 and SSMP
(Singapore Malay).55 Each row represents
a haplotype with ancestry derived
from Neanderthals14 (red), Ust’-Ishim16
(orange), Denisovans15 (blue), or Han
Chinese (green) or uncertain ancestry
(including unknown archaic and uncer-
tainmodernhuman;gray).
groups before the post-LGM arrivals.
We suggest that SUNDer, an un-
known ancient group of multiple
ancestral origins, introduced the
ancient haplotypes that are unique
to present-day Tibetan highlanders
(Figure8).
ATwo-Wave‘‘Admixtureof
Admixture’’Model
Our data support that the Tibetan
genomeappearstohavearisenfroma
mixture of multiple ancestral gene
pools, but the ancestral composition
ismuchmorecomplicated,anditshis-
tory can be traced back considerably
earlierthan previously suspected.For
trace their ancestries back to ~62,000–38,000 years ago instance,arecentstudyhassuggestedthat‘‘Tibetansarea
(Table S17), pre-dating the LGM49 and indicating that mixtureofancestralpopulationsrelatedtotheSherpaand
thecolonizationofhumansinTibetismuchmoreancient Han Chinese.’’56 We propose a two-wave ‘‘admixture of
thanpreviouslythought. Themixedarchaicancestriesin admixture’’ (AoA) model, despite its simplicity, to help
this ~300 kb region and simulation studies indicate that explain the ancestral makeup and pre-history of Tibetans
TibethasbeenahumanmeltingpotsincethePaleolithic and Sherpas (Figure 9). An ancient wave of admixture
age, despite its inhospitable environment and the fact occurredinapre-LGMera>40,000YBP,whichcouldhave
that interbreeding occurred among different hominin resulted in the unique mosaic pattern of Paleolithic
10 TheAmericanJournalofHumanGenetics99,1–15,September1,2016