Genome-wide association study of ancestry-specific TB risk in the

14 Pages·2013·0.55 MB·English

HumanMolecularGenetics,2014,Vol.23,No.3 796–809
doi:10.1093/hmg/ddt462
AdvanceAccesspublishedonSeptember20,2013
Genome-wide association study of ancestry-specific
TB risk in the South African Coloured population
EmileR.Chimusa1,{,∗,NoahZaitlen7,{,MichelleDaya6,MarloMo¨ller6,PaulD.vanHelden6,
NicolaJ.Mulder1,AlkesL.Price2,3,5,4,{andEileenG.Hoal6,{
1DepartmentofClinicalLaboratorySciences,InstituteofInfectiousDiseaseandMolecularMedicine,UniversityofCape
Town,CapeTown,SouthAfrica,2DepartmentofEpidemiology,3DepartmentofBiostatistics,4PrograminMolecularand
D
GeneticEpidemiology,HarvardSchoolofPublicHealth,Boston,MA,USA,5PrograminMedicalandPopulationGenetics, o
w
BroadInstitute,Cambridge,MA,USA,6MolecularBiologyandHumanGenetics,MRCCentreforMolecularandCellular nlo
a
Biology,DST/NRFCentreofExcellenceforBiomedicalTBResearch,FacultyofHealthSciences,Stellenbosch d
e
University,Tygerberg,SouthAfricaand7DepartmentofMedicine,UniversityofCaliforniaSanFrancisco,SanFrancisco, d fro
CA,USA m
h
ttp
s
ReceivedJuly30,2013;RevisedJuly30,2013;AcceptedSeptember17,2013 ://a
c
a
d
e
Theworldwideburdenoftuberculosis(TB)remainsanenormousproblem,andisparticularlysevereinthe m
ic
admixedSouthAfricanColoured(SAC)populationresidingintheWesternCape.Despiteevidencefromtwin .o
u
p
studiessuggestingastronggeneticcomponenttoTBresistance,onlyafewlocihavebeenidentifiedtodate. .c
o
Inthiswork,weconductagenome-wideassociationstudy(GWAS),meta-analysisandtrans-ethnicfinemap- m
/h
pingtoattemptthereplicationofpreviouslyidentifiedTBsusceptibilityloci.OurGWASresultsconfirmthe m
WT1chr11susceptibilitylocus(rs2057178:oddsratio50.62,P52.71e206)previouslyidentifiedbyThye g/a
etal.,butfailtoreplicatepreviouslyidentifiedpolymorphismsintheTLR8geneandlocus18q11.2.Ourstudy rtic
le
demonstratesthatthegeneticcontributiontoTBriskvariesbetweencontinentalpopulations,andillustrates -a
b
thevalueofincludingadmixedpopulationsinstudiesofTBriskandothercomplexphenotypes.Ourevaluation s
tra
oflocalancestrybasedontherealandsimulateddatademonstratesthatcase-onlyadmixturemappingiscur- c
rentlyimpracticalinmulti-wayadmixedpopulations,suchastheSAC,duetospuriousdeviationsinaverage t/23
/3
localancestrygeneratedbycurrentlocalancestryinferencemethods.Thisstudyprovidesinsightsintoidenti- /7
9
fyingdiseasegenesandancestry-specificdiseaseriskinmulti-wayadmixedpopulations. 6
/5
9
4
0
6
6
b
y
INTRODUCTION butonly10%goontodevelopactiveTBduringtheirlifetime gu
e
(6,7). In addition, twin studies in humans and animal models s
Tuberculosis(TB)isasignificantsourceofmorbidityandmor- alsodemonstrateastronggeneticinfluenceonTBsusceptibility t on
tality worldwide, particularly in developing countries. It is a (5,8,9). The rate of concordance of TB among monozygotic 12
leadingcauseofhumanimmunodeficiencyvirus(HIV)-related twins(18of55,32.7%)wasmorethantwice(oddsratioofcon- Ap
deaths,asalmostoneinfourdeathsamongpeoplewithHIVin- cordance:2.4;95%CI:1.4–4.0)thatobservedamongdizygotic ril 2
fectionisduetoTB(1,2).In2010,therewere8.8millionnew twins (21 of 150, 14.0%) (8,9). These estimates suggest that 01
9
cases of TB, of which 1.1 million were among people living genetic factors play an important role in TB susceptibility in
with HIV (3,4). TB susceptibility is known to be a complex both understanding the host response and determining the
traitinfluencedbybothgeneticandenvironmentalfactors(5). outcomeofinfection(2,8).
The environmental factors that influence TB susceptibility GWASidentificationofgeneticassociationsofhostsuscepti-
include socio-economic conditions, smoking and acute infec- bilitytoinfectiousdisease,suchasTB,hasbeenslowandlimited
tion.One-thirdoftheworld’sindividualsareinfectedwithTB, bysmallsamplesizes(10,11).FewGWASpapersthatidentify
∗Towhomcorrespondenceshouldbeaddressed.Email:[email protected]
†Theauthorswishittobeknownthat,intheiropinion,thefirsttwoauthorsshouldberegardedasjointFirstAuthors.
‡Theauthorswishittobeknownthat,intheiropinion,thefinaltwoauthorsshouldberegardedasjointSeniorAuthors.
#TheAuthor2013.PublishedbyOxfordUniversityPress.Allrightsreserved.
ForPermissions,pleaseemail:[email protected]
HumanMolecularGenetics,2014,Vol.23,No.3 797
the signal of association with pulmonary TB were published RESULTS
until recently. Davila et al. identified four polymorphisms in
theTLR8geneonchromosomeX,inanIndonesianpopulation GWASwithcorrectionforgenome-wideancestry
andreplicatedtheresultinaRussiancohort(12).Thyeetal.con-
Weanalyzeddatafrom642TBcasesand91controls(afterQC)
ductedacombinedGWASinAfricanpopulationsinGhanaand
genotyped on the Affymetrix 500K chip (see Materials and
GambiatoinvestigatethehostsusceptibilitytoTB,andidenti-
Methods).Usingimpute2(26),we imputedtheuntypedgeno-
fied the 18q11.2 locus (13). Recently, Thye et al. reported a
typesusingHapMap3release2(27)and1000Genomeproject
newTBsusceptibilitylocusonchromosome11p13inaGhan-
populations (28). There were 1453294 and 4467279 genetic
aianpopulationneartheWT1gene.Thisfindingwasreplicated
variants retained from each imputation panel respectively. As
in samples from Gambia, Indonesia and Russia (14). A case–
expected, we observed substantial population substructure in
control study in a Chinese population failed to replicate the
theSAC(seeMaterialsandMethods).Toaccountforbothpopu-
18q11.2 locus (15). However, it was replicated in a separate
HanChineseTBcohort(16). lationstratificationandhiddenrelatedness,weappliedthemixed D
o
ThesestudiesandworkshowingthatratesofTBvaryconsider- modelapproachEMMAX(29)onthesedatasets;theQ–Qplots w
n
ably between populations and regions suggest that association of genomic control factors effects are shown in Figure 1. The lo
a
genomic control lambda (30) was l ¼1.05 for the typed d
effects may vary across ethnic groups (4,17). Analysis of the GC e
cadomrreixlaetdiopnobpeutlwateioennsgceanneitmicparnocveesdtriyseaansdepprheednicottiyopneainndrpercoevnitdlye 1d.a0ta9sfeotr,tlhGeC10¼001G.05imfopruttehdedHataapsMetaapn3dlimGpCu¼ted1.d0a8tfaosrett,hleGcoCm¼- d from
cetruacl.iaslhoinwsiegdhtthsaitngtoenmeteidciacnaclegsetrnyeteixcser(t1s8a)m.Faojorreixnaflmuepnlec,eKonumima-r bTihneesde(gtyepneodm,iHcacpoMntarpo3l ilmampubtdeadsaanrde1a0c0c0epGtaibmlep,utaendd)dsuagtagseestt. https
provinglung-functionestimatesandcategorizingasthmaseverity little departure from the null expectation, except at the right ://a
c
end tail of the distribution. To control for cryptic relatedness, a
inanAfricanAmericanpopulation(18). d
The second highest incidence of TB in the world is in the weusedEMMAX(29),avariancecomponent-basedstatistical em
testshowntocontrolforbothstructureandrelatedness. ic
Western Cape in South Africa, particularly in the admixed .o
As shown in Figure 2, a SNP on chromosome 14 q24.2, u
South African Coloured (SAC) population (3,19). Investiga- p
rs17175227 (P¼8.99e209 and OR¼0.141) appears to be a .c
tions based on candidate gene studies and genome-wide o
m
genome-widesignificantassociation signal.TheSNPrs17175
linkage scans on the SAC population were previously con- /h
227hasalowminorallelefrequencyof0.01642.Weperformed m
ducted (3,10,11,19,20). Many of the candidate gene studies g
fwaeilreedetxoamobinseerdvaensdtawtiestriecaolftaesnsoincciaotniocnlusswivieth(1t1h,e20m).arAkletrhsotuhgaht awhweetlhle-cratlhibersapteedcitfiecstSfNorPrawroeuSldNPstsi,llFbisehgeer’nsoemxea-cwtitdeest,sitgonsiefie- /article
a few genetic association studies have identified candidate cant. The result suggested that rs17175227 was not genome- -a
genesforTBsusceptibilityusingdatafromtheSACpopulation wMiadteersiiagl,nFifiigc.aSnt1)(.PT¼his2h.7ig7he2lig06h,tsOaRn¼im0p.o1r4ta1n)t(cShuapllpelnegmeeinntaarsy- bstra
(11,20), neither ancestry-specific TB risk nor GWAS for TB c
has been considered in this population. In addition, no study sociation analysis of low-frequency (1–5%) variants, which t/23
onotherAfricanethnicgroupshasbeenattemptedtoreplicate may often attain genome-wide significance in standard tests /3/7
the susceptibility loci on chromosomes 18q11.2 and 11p13, such as mixed model association or logistic regression due to 96
reported in Thye et al. (13,14), respectively. It is not known theimperfectasymptoticdistributionofthosetestsinthecase /59
oflow-frequencyvariants(MaterialsandMethods). 4
whether the predisposition due to these SNPs applies to other 0
African populations. TheHapMap3imputedSNPrs12294076(P¼9.56e208)on 66
wiHdeeraen,cwesetryco,ncodmucpteudtedatGheWpAoSwewrittohrceoprlriceactteioenacfohrpguebnliosmheed- cghernoommoes-owmidee1s1igqn2i1fi–cqan2c2e.1, wnahrircohwwlyemdiesfisensetahsePth¼res6h.4oeld20o8f, by gue
1.7e208 and 5.5e209 based on 390887; 1453294 and 4467 s
odds ratio and specified whether the published odds ratio lies t o
withinthe 95%confidence interval (CI)estimated intheSAC 279 SNPs tested (Fig. 2) from typed genotype data and n 1
pdraetavsieotu.sOlyuirdsetnutdifiyerdeipnliTcahtyeedetthael.p(o1l4y)m.oBropthhisimmpsuitnatWioTn1-bgaesende iimveplyute(dsedeataMusaitnegriaHlsapManadp3Manedth1o0d0s0).GTenhoemgesendeattiac,revsaprieacnt-t 2 Ap
meta-analysis andtrans-ethnic finemappingofourstudywith rs12294076 has a minor allele frequency of 0.16 in the SAC, ril 2
recent African TB studies confirm that genetic variants in the 0.22 in Yoruba and 0.00 in other HapMap populations, and is 01
9
likely to be an African-specific SNP. The closest gene to this
WT1geneconferriskofTB.However,ourstudydidnotvalidate
SNP is DYNC2H1. This gene encodes a large cytoplasmic
previously identified polymorphisms in the TLR8 gene on
dyneinproteinknowntobeinvolvedinretrogradetransportin
chromosomeX(12)orthe18q11.2locus(13).
theciliumwithamajorroleinintraflagellartransport(31).Muta-
TheSAChaveamixedancestrytracedbackover350years
tionsinDYNC2H1causeaheterogeneousspectrumofconditions
from various populations including European Caucasoid,
relatedtoalteredprimaryciliumfunction.Thesub-cellulardis-
SouthandEastAsians,andSANandBantuAfricans(21–24).
tribution of dynein shows specific association with elements
Ourresultsshowthatancestry-specificTBriskisnotduetocon-
of the late endocytic pathway (31). Additional 36 typed and
foundingbysocio-economicstatus(SES)asisthecasefortype-2
62imputedgeneticmarkerswithsuggestiveP-values(10205–
diabetesinLatinopopulationsfromMexicoandColombia(25).
10206)thatdidnotpassourgenome-widesignificancethresh-
Furthermore,ourfindingsindicatedthattheSANancestralcom-
olds are listed in Supplementary Material, Tables S1 and S2,
ponent confers risk while non-African ancestral components
respectively.
conferprotectionagainstTBintheSAC.
798 HumanMolecularGenetics,2014,Vol.23,No.3
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/h
m
g
/a
rtic
le
-a
b
s
tra
c
t/2
3
/3
/7
9
6
/5
9
4
0
6
6
Figure1.Q–QPlotofpopulationstratificationeffectstocomparethedistributionofobservedP-valueswiththeexpecteddistribution:ThelGCvaluesindicatethe b
residualpopulationstratificationeffects(aftercorrection)whichareminimal.PlotsA–Darefromtypedgenotype,imputedgenotypefrombothHapMap3andthe y g
1000Genomesprojectandcombineddatasets,respectively. u
e
s
t o
n
1
ReplicationofSNPsreportedinpreviousstudies MAF¼0.31 and OR¼0.78), was not accurately imputed in 2 A
our study (CALL≈0.8), and could therefore not be examined p
Weattemptedtoreplicatethechr11,chr18andchrXassociations withconfidence(SupplementaryMaterial,TableS3).Thers205 ril 2
0
withTBpreviouslyreportedbyThyeetal.(13,14)andDavila 7178,rs11031728andrs110317311000GimputedSNPsinthe 1
9
etal.(12),respectively.Ineachcase,wecomputedourpower SACarenotgenotypedinGIHandSANdata.Nevertheless,we
to detect an association at the previously reported odds ratio, computedther2LDbetweenthesethreeSNPsandotherSNPs
and evaluated whether the previously reported odds ratio was intheWT1locususingtheSACdataandtheYorubainIbadan
within our 95% CI (see Materials and Methods). We found (YRI), CEU, Japanese in Toyko (JPT)+CHB data from the
thattheassociatedSNP,rs2057178(P¼2.63e209,OR¼0.77 1000 Genomes project. Previous results reported that the
and MAF¼0.33) on 11p13 reported by Thye et al. (14), is rs2057178,rs11031728andrs11031731SNPsareinstrongLD
close to genome-wide significance (P¼2.71e206, OR¼0.62 in the Ghanaian population (14). We obtained r2(rs2057178,
and MAF¼0.08) in the SAC-TB imputation GWAS rs11031728)¼0.90,0.90,1.00and0.8;r2(rs2057178,rs110317
(Table 1). A second reported significant SNP in the Ghanaian 31)¼0.70, 0.90, 1 and 1 and r2(rs11031728, rs11031731)¼
study group rs11031728 (P¼5.25e209, MAF¼0.32 and 0.70,1.00,1.00and0.90inSAC,CEU,YRIandJPT+CHB,re-
OR¼0.77) was also associated in our imputation GWAS spectively.ThesethreevariantsarealsofoundtobeinLDinthe
study (P¼2.86e206, MAF¼0.08 and OR¼0.61). The third SACdata.AdditionallociinLDwiththeWT1locusareprovided
mostsignificantSNPintheirstudy,rs11031731(P¼7.01e209, inSupplementaryMaterial,TableS3.Thers2057178,rs11031728
HumanMolecularGenetics,2014,Vol.23,No.3 799
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/h
m
g
/a
rtic
le
-a
b
s
tra
c
t/2
Figure2.Manhattanplotofgenome-wideassociationanalysesofTBintheSAC(A)fromtypeddatasetonly,and(BandC)bothimputeddatasetsbasedonHapMap3 3/3
and1000Genomesreferencepopulations,and(D)thecombineddataset. /7
9
6
/5
9
andrs11031731SNPsandotherslistedinSupplementaryMater- andhasbeenseeninsomecancersofblood-formingcells(leuke- 4
0
6
ial,TableS3areassociatedwithWT1. mias),suchasacutelymphoblasticleukemia,chronicmyeloid 6
b
LookingatTable1andSupplementaryMaterial,TableS3,the leukemiaandchildhoodacutemyeloidleukemia(32). y
g
effectofthesusceptibilitylociatWT1intheSACisinthesame Theidentifiedsusceptibilitylocusrs4331426onchromosome u
e
danirceecttoioTnBas.Hthoowseevpeurb,ltihsehepdubinliTshheydeoedtdasl.r(a1t4io),isnuTghgyeesteintaglr.e2s0is1t2- 1an8dq1O1.R2¼in1T.h1y8e, Gethaaln.a2:0P10¼(M0.A00F4¼an0d.4O8,RG¼am1b.1ia9:Pan¼d C0.o0m03- st on
(14)doesnotliewithinthe95%CIestimatedintheSACdataset bined data: P¼6.8e209 and OR¼1.19), for TB in the study 12
A
(Table1).ThismaybeduetodifferentLDpatternsbetweenthe of combined Gambia and Ghanaian populations (13), did not p
SACandWTCCC-TBdata,sincetheSACisadmixed.SNPscan yieldanassociationwithTBinourstudy(Table1).Weobtained ril 2
be associated with the same phenotype in different ethnicity P¼0.83, MAF¼0.19 and OR¼1.00, and no suggestive 01
9
groups, but the effect sizes may differ substantially, because signals in the SAC data located near the variant at SNP
thetagSNPmaytagdifferentsetsofcausalvariants. rs4331426(SupplementaryMaterial,TableS4).Asabove,we
WT1isasuppressorgenelocatedonchromosome11p13and computed r2 in the region of 18q11.2 in the data of the SAC,
providesinstructionsformakingaproteinthatisinvolvedinthe CEU, YRI, JPT+CHB, GIH and SAN. Four SNPs
developmentofthekidneysandgonads(ovariesinfemalesand rs4264496,rs4331426,rs4239431andrs4239432intheentire
testesinmales)beforebirth(32).Furthermore,itisalsoknownas regionof18q11.2haver2. ¼0.5,butallhaveweakP-values
atranscriptionfactorsinceitregulatestheactivityofothergenes from the association study with TB (Supplementary Material,
bybindingtospecificregionsofDNA.Queryingacomprehen- Table S4)inthe SACdata. Inaddition, thers4331426SNPis
sivehumanProtein–ProteinInteractionnetwork(http://cbg.ga notinLDwithanySNPsinWT1locusinthedataoftheSAC,
rvan.unsw.edu.au/pina/), WT1 has known direct interactions CEU, YRI and JPT+CHB. One possible explanation for our
(32) with UBE2I, AREG, WTAP, AREGB, U2AF2, TP73, failuretoreplicatethislocuscouldbeourlowpowertodetect
SDGF,PRKACAandtheP53gene.Inparticular,thisgeneisun- thereportedoddsratio(power¼0.12).However,thepublished
usuallyexpressedinlungandcertaintypesofprostatecancer, odds ratio of rs4331426 is not included in the 95% CI (CI)
800 HumanMolecularGenetics,2014,Vol.23,No.3
CI) 1–0.84)1–0.8) –1.3] –1.8)–1.8)–1.8) 95%CI erteesrptnilmiscbaateteitodwnienewntahtsehnSeoAStACdCudeaatsnaodsleeGtly(aTmtaobbliloae/wG1)hp,aosnuwage/gMre.sDatliianfwfgeitrhpeanottpouLulDartnipooannts--
% 77 1 766 e
R(95 77(0.78(0. 19[1. 4(1.04(1.04(1.0 hinth proTvoidceoamnpaalrteeronuartivsteuedxypwlainthatpiornev.ious findingsofassociation
O 0.0. 1. 1.1.1. wit withTBsusceptibilityatfourpolymorphismsintheTLR8gene
e
Thyeetal.(14)MAFP 2090.332.63e2090.327.01e Thyeetal.2010(13)2090.486.8e Davilaetal.2008(12)–0.014–0.016–0.01 hyeetal.(13,14)donotli oetTaraprentnnlLegadnaRtyiahdol8ePd.enddAt(Xgoi1iRt(ewn2incno2)hoaenT.)rrnaadoOooPlbmsnfAulieatmortRhshsi1eps)oemoumXccaXtpoaienuatmcdifctthoarihpoornttraoimwnoormGemnooDoWfooGsapusotvAsWohrmiemeSlruaAseedeeeosS,oitunnfaialosntulttusu.chhtgro2leewugs0pdSioen0toismAho8nltynsCg(at-m1hlpa.r2oosTsr)sneer3,ehupogw7edhnfi6eoroi-r4ossneca8mimssoug8untsn0o(lDditPs,ifisunoaArccvdmstatRii3henlsa17dae-tl Dow
T n
dd uted ddd from M64a8t7e9ri,arls,3F7ig6.1S622)4.aTnhdesres3fo7u8r89S3N5P(sTaarbelien1LaDndwSituhpopnleemanenottahreyr loade
werSNPtype 5641000GImpute4541000GImpute 185HapMap3Imp 5321000GImpute5211000GImpute5321000GImpute Allpublishedoddsratio (aw(r2drpsr0ae23eot0.a7wh8p8aaer(8¼ndor19dv2¼i30ni)td5.h)c50eoei).(dni5mtrih5ntipreh)netl,ehelraaSeetSetnpueAddoppdaCrpotttPalhw,ee-deComevfErapTetlUtuhnBuobete,alrsaSYiressyAfpRhsrlooCeIiMcdc,m,iaJaaaotPttdtehedTedreddtiih+staSieSloAr,NCpanCTtuaHPiblaosaBbSlsiiil,NssnnehGoPecDDSIsdiH5aaai.novtviadiioBLnlldanadDescseeSratwttauAutiaadsitNllohye.. d from https://academ
Po 0.10.1 0.1 0.50.50.5 er). (12) lies within the 95% CI estimated in the SAC dataset ic.o
w u
o (Table 1), the lack of a statistically significant association in p
OR(95%CI) 0.62(0.50–0.75)0.61(0.50–0.75) 1.00(0.95–1.04) 1.30(0.91–1.85)1.27(0.89–1.81)1.23(0.87–1.80) heSACdataset(P oMIdueerntdatai-ftayaninmaglaycysoibmsemduoenvtoariinacnotmsopflmeteodpeoswteefrf.ectisachallenge,and .com/hmg/article
nt largesamplesizeisawaytoincreasepower.Thesamplesizesof -a
i b
SACTBStudyMAFP 2060.082.71e2060.082.86e 0.190.83 0.3860.14650.3820.18440.3860.2854 chpublishedoddsratio TpncfiWimooinBfimTwepc-CucebamtarnCiensatcdCeoepesd-.oSTeTabNtBxnthoaidPesii(nstnGciccnaouiargsnnemrsrtaoresbsfiocineignilatastne/htGd-aiiemonlahsntpaaatsophnbsawepiaystt/ieMnasrpwgtsteautoirrltdaafihdnyonwegraetimdlee)xyonciisdnsttentalgciseotno.avtgimemMaplmndrpoeodauofrvtgtnsaiiantidatvncieitagnoliursnsitduotaihicfrnonfideatngselc,ritsettwwihhgtnooeee-t stract/23/3/796/5940
a 6
dies CallINFO 0.840.80.840.80 0.990.9 111111 epowertoreplicatee gTatteuhcieBtrernyeo)So,,sLAmswaDCteeht-oaewiantnastidwdsltheeGooseifasmseSmt1udpAbdu0Citit0ahea9/estaGi3(nodhS6dneaA4gGnpCraoaWe/-useMTstAoiBoabSsflloaeadmwhnadedtiaitasfleWsfareeSmotrTNgsepCen(PlSnetCsAesLCiwcCtD-ayeT-nrTpiBecnBaa)tm.uttaehSsnereeitdnnahsccW-eaeobtnemTeaartdCoblwymgiCnezeiCeeenxndd--- 6 by guest on 12 A
previousstu A1/A2 G/AC/G G/A A/CA/CA/C computedth s(tsofitaumxdheipedetlseionrirognrgeaoernrnrddoeeorirmtaytnoebdfcefahetwoctoote)ts.eaenlWaasnmetuaoodpbuiptnearstionop(efSrdihAae8tCet4ed-.r8Wom0g%TeetCnao-eCafintCvyaa)l(ryTrisaaaibtsuhil-meistrqyeutthhdaoaurndee pril 2019
n e
i w heterogeneity estimator of DerSimonian–Laird) was 0.0288
ofSNPsreported RPOS 3236418732363616 19676176 129226591292368112924697 hresholdof0.05,ntSACdataset. (cbMSooEemTthtoab¼Sriaano0cnefc.dtd0oo2supm2tnru6otd)agfi,noeriadsrnm.dtbhiicen(a3bat3ireny)twg-e(eafsefeehneci-gtssMthumdlaeytevetrhehileoatlodesfsrohaigmenetdepnrleoeMimgtyeee,ntnhwetoieetddyas)pii.nnplWoitehudeer
Table1.Replication SNPCH rs205717811rs1103172811 rs433142618 rs3788935Xrs3761624Xrs3764879X AssumingaP-valuetestimatedinthecurre Wftr(rereTaorpnTaimodabCroll,teCmeF2Cr-a)eeic-g.sfThf.WuBelcSties(tn3lad)(Glo.lisvCfGoIin¼Cdoruab¼aa1ntld.ad01dion4.si0mte9ti6ud5o2d)nr)ay,e,rn,aetdbossiSopnnAsbeatacirCanbtynia-l-vdTereeyafBilfr-nyedeflc(f(SfatlPetui(cG-olpvtCnpGal¼lCmreuam¼eet1este.a,s01n-9f.atw0ra4no5re)aym)lyMaaatslnnhseaoddes-
HumanMolecularGenetics,2014,Vol.23,No.3 801
Table2. Imputationgenome-widemeta-analysisoftwoTBcase–controlstudies,SAC-TBandWTCCC-TB
SACTB
SNP CHR P_RAN P_BE ST1 ST2 P-value M-value
rs2057178 11 3.26e23 9.83e213 53.05 2.91 2.75e206 1.0
rs11031728 11 4.73e207 4.08e210 41.19 0.0 2.98e206 0.988
SACTB
rs4331426 18 0.282554 1.8950e208 1.15 32.6 0.002 0.0
P_RANistheP-valueofrandomeffect,P_BEistheP-valueofbinaryeffect,ST1andST2arethestatisticmeaneffectandheterogeneity,respectively.M-valueisthe
posteriorprobabilitythattheeffectexistsineachstudy.TheP-valueofrs4331426differsfromthatinTable1duetoMetasoftfittingittogetherwiththeThyeetal.
studies.
D
o
w
mayinfluencetraitvariabilityatlociofWT1.Thebetween-study n
lo
heterogeneityduetodifferentpatternsofLDandimputationac- a
d
curacyaremajorlimitationsforfinemappingofGWASsignals ed
(26,34),neverthelessthiscurrentresult(SupplementaryMater- fro
m
ial,TablesS3andS6)demonstratedthatfinemappingisapower- h
ful approach to better characterize TB susceptible loci risk in ttp
s
diversepopulations. ://a
Additional meta-analysis of our TB study and four poly- c
a
morphismsintheTLR8geneonchromosomeXpreviouslyiden- de
m
tifiedinDavilaetal.(12)arefoundinSupplementaryMaterial, ic
Figure S2. Although Metasoft provides a slightly different .o
u
P-value (Table 2) for SNP rs4331426 than the one from the p.c
standardGWAS(Table1),whichmaybeduetohighheterogen- om
eity(ST2¼32.6,seeTable2),thissusceptibilitylocusreported /h
m
inThyeetal.2010(13)doesnotsurvivegenome-widesignifi- g
/a
canceintheTBmeta-analysisoftheSACandThyeetal.2010 rtic
(13).Moreover,thecombinedSACTBstudyandTBsuscepti- le
bilityatfourpolymorphismsintheTLR8geneonXchromosome -ab
s
intheIndonesian populationreportedinDavilaetal.2008by tra
Figure3.ForestplotofcombinedTBstudies,includingourTBstudy,Thyeetal. meta-analysisdonotyieldanyconvincingassociationevidence. ct/2
(13,14). Thefailuremaybemainlyduetolowpower(56.2and12.3%)to 3
/3
detect the four polymorphisms in the TLR8 gene on chrX and /7
9
locus at 18q11.2 associations using the meta-analysis sample 6
examined the posterior probability (m-value) that the effect /5
size, respectively. In addition, variability of patterns of LD 9
exists in each study (33). Using a threshold m.0.7, we across distinct ethnic groups (SAC, Gambia/Ghana/Malawi 406
observed two genetic variants, rs2057178 and rs11031728 6
andIndonesian)mightexplainourreplicationfailure. b
(Fig.3andTable2)withsimilarP-valuestothosefromthestand- y
g
ardGWAS(Table1),thatresultedinasignificantriskofTBand u
e
ShaNdPesfaferectbsoitnhoounrcshturodmyoasnodmtheereTghioyne e1t1pa1l.3staunddyre(1p4li)c.aTtehethsee RelationshipbetweenTBriskandgeneticancestry st on
1
recentfindingsofThyeetal.(14)inourimputationGWAS. Weestimatedtheproportionofgenome-wideancestryin733un- 2 A
Weevaluatedwhethertrans-ethnicmeta-analysiscanrefineor relatedSACindividuals(642casesand91controls)fromthese p
confirm the replicated association signals by narrowing the five proxy ancestral populations (Supplementary Material, ril 2
0
genomicregionsatlocusWT1wherefunctionalvariantsmight Table S7), including European (CEU), Bantu Africans (YRI), 1
9
beexpectedtoreside.Toaddressthis,were-imputedSNPsat SAN Africans, Gujarati Indian (GIH) and Chinese/Japanese
chromosomeregion11p13(32409321–32457176bp)fromboth (CHB+JPT). Our choice of the proxy ancestral populations
WTCCC-TBandSAC-TB.SupplementaryMaterial,TableS3 of the SAC is based on previous findings (21–23) (Materials
displaysGWASresults,thepatternsoflinkagedisequilibrium andMethods).Inparticular,Pattersonetal.(22)demonstrated
in region in both ethnic groups (WTCCC-TB and SAC-TB). that the admixed SAC population includes an east Asian
Although the imputations were moderately accurate (CALL related component. They inferred European, south Asian and
(cid:4)0.8,SupplementaryMaterial,TableS3),alltheimputedvar- Indonesiancomponentsofnon-Africanancestry.Itisappropri-
iantsatlocusWT1wereinLDwiththeleadSNPrs2057178in ate to model the Indonesian component using an east Asian
theSAC. population,becausePattersonetal.(22)reportedanF value
ST
Our trans-ethnic fine mapping (Supplementary Material, of 0.02 between Indonesians and east Asians, which is much
TableS6)basedontheimputationmeta-analysisatchromosome lower than the F value of 0.07 between Indonesians and
ST
region11p13confirmedtheassociationsignalsatlocusWT1and southAsiansortheF valueof0.10betweenIndonesiansand
ST
facilitatedthelocalizationofneighbouringvariantsinLDthat Europeans.Similarly,Pattersonetal.(22)andDeWitetal.(21)
802 HumanMolecularGenetics,2014,Vol.23,No.3
Table3. AssociationofgeneticancestrywithTBintheSAC,withnominalP-valuesbeforecorrectingforhypothesestested
AssociationofgeneticancestrywithTB Genome-wideancestry
POP Correlation95%CI,P-value(TB-ancestry) Cases Controls Allsamples
SAN 0.16[0.09,0.23],1.58e205 0.342+0.135 0.279+0.114 0.334+0.134
YRI 0.06[20.01,0.13],0.109 0.279+0.163 0.250+0.143 0.275+0.161
CEU 20.12[20.19,20.05],0.0007 0.184+0.116 0.228+0.124 0.190+0.118
GIH 20.11[20.18,20.04],0.002 0.127+0.086 0.156+0.079 0.130+0.086
CHB+JPT 20.13[20.20,20.06],0.0005 0.069+0.047 0.087+0.054 0.071+0.048
MeanandstandarderrorofancestryproportionfromeachoffivepopulationscontributingtotheadmixtureintheSAC(fromcases,controlsandallsamplesofthe
SAC.).
D
o
w
n
demonstratedthattheSACpopulationincludesbothSANAfrican asbotheffectswouldneedtobepresentinorderfortheancestry lo
a
andBantuAfricancomponents. analysistobeconfoundedbySES.Whentestingforcorrelations d
e
staTtuoseixnamthiinseSthAeCredlaattiaosnesth,iwpbeertewgereenssgeednectaiscea–nccoesnttrryolansdtaTtuBs btreietws,eneonneeaocfhthoeftrheesutwltsowSeErSesvtaartiiasbtilceasllaynsdigenacifihcoafntthaeftfievrecoarnrceecst-- d fro
m
againsttheestimatedfractionofYRI,SAN,GIH,CHB+JPT ingfor10hypothesestested(SupplementaryMaterial,TableS8). h
andCEUancestry,respectively, in733 unrelated SAC indivi- Inparticular,SANancestryhadanon-significanttrendtowards ttp
s
duals(Table3).Weobservedastatisticallysignificantpositive positive correlation (r¼0.05, 95% CI¼[20.17, 0.27], P¼ ://a
correlation(r95%CI¼0.16[0.09,0.23],P¼1.58e205)between 0.65) and (r¼0.09, 95% CI¼[20.13, 0.31], P¼0.45) with ca
d
SANancestryandTBstatus.TheCEU(r95%¼20.12[20.19, SES self and household incomes, respectively. Although TB em
20.05], P¼0.0007), CHB+JPT (r 95%CI¼20.13 [20.20, status is known to be associated with low SES (35–38), even ic
.o
20.06],P¼0.0005)andGIH(r95%¼20.11[20.18,20.04], consideringthelowendofthe95%CIof 20.17(forSESself u
p
P¼0.002)ancestryintheSACwerenegativelycorrelatedwith income)and 20.13(forSEShouseholdincome)thisstillcould .c
o
TBstatus.YRIancestryproportionwasnotsignificantlycorrelated not explain the correlation (r¼0.16, 95% CI¼[0.09, 0.23], m
(r95%CI¼0.06[20.01,0.13],P¼0.109)withTB.Further- P¼1.58e205)betweenSANancestryandTBstatusunlessthere /hm
g
more,weobservedastatisticallysignificantcorrelationofage is a virtually perfect negative correlation between SES and TB /a
(P¼1.01e205, mean age 37 in cases and 31 in controls) with status, which is highly unlikely. Thus, the observed association rtic
riskofTB,butnoevidenceofcorrelationbetweensexandTB betweenSANancestryandTBstatus(Table3)isnotentirelyacon- le-a
(P¼0.597). sequenceofSES. bs
We observed a correlation between ancestry proportions of tra
c
theancestralpopulations(SupplementaryMaterial,TableS8). t/2
Accuracyofmulti-waylocalancestryinferenceintheSAC 3
We checked to see if the tests above could be confounded by /3
this correlation by performing conditional risk tests between Recentlyintroducedlocus-specificancestrymethodsformulti- /79
6
pairsoftheancestralpopulations(seeMaterialsandMethods). way admixed populations, including ALLOY (39), PCAdmix /5
9
OurresultsdemonstratethatAfricanancestry(SAN,YRI)TB (40), MULTIMIX (41) and ChromoPainter (42) achieved 4
0
riskintheSACisnotsignificantconditionedonnon-Africanan- equivalentaccuracytoWinPOP(43)orLamp-LD(44).There- 66
cestry(CEUandJPT–CHB)risk.WiththeexceptionofIndian fore, we assessed the accuracy of two methods for inferring by
(GIH) ancestry, non-African ancestry (CEU and JPT–CHB) local ancestry in multi-way admixed populations, Lamp-LD gu
e
risk is significant conditioned on African ancestry risk. This (44)andWinPOP.Wesimulated750individualsofmixedEuro- s
shows that YRI and SAN are differently correlated with risk pean(CEU),ChineseandJapanese(CHB+JPT),Bantu(YRI) t on
than CEU, JPT–CHB and GIH (Supplementary Material, andSANancestryonchromosome1(SupplementaryMethods 12
TableS7).JPT–CHB,GIHandCEUarenotsignificantlycondi- Text1).Weassessedtheaccuracyfromtheinferredlocalances- Ap
tionedoneachotherandallarecorrelatedwithTBrisk(Supple- tryfrombothLamp-LDandWinPOPbycomparinganestimate ril 2
mentary Material, Table S8). We see that SAN confers risk, ofcorrelationbetweentheinferredlocalancestryfromLamp- 01
CEU, JPT–CHB and GIH confer protection, and YRI shows LD(CEU:86%,YRI:78%,GIH:84%,CHB+JPT:90%and 9
noevidenceofcorrelation(SupplementaryMaterial,TableS8). SAN: 80%) and WinPOP (CEU: 58%, YRI: 69%, GIH: 51%,
Anotherpotentialconcernwasthattheobservedrelationship CHB+JPT:48%andSAN:47%)andthetrueancestryinforma-
betweengeneticancestryandTBstatuscouldbeaconsequence tionbycomputingtheaccuracy(SupplementaryMethodsText1).
ofconfoundingduetoSES,asdescribedinarecentstudyoftype The results suggest that Lamp-LD provides greater accuracy
2 diabetes in Latinos (25). We investigated this possibility by thanWinPOP(SupplementaryMaterial,FigsS4andS5).Given
studyingtwoSESvariables(seeMaterialsandMethod),house- this,wethenassessedtheabilityofLamp-LDtocorrectlyinfer
hold and individual incomes (Supplementary Material, thelocalancestryinamulti-wayadmixedpopulationbycomput-
Table S8). Although these variables were available in only a ingtherateofcallingtrueancestralandtheerrorofcallingother
subset of 82 SAC cases, it was possible to draw a conclusion ancestral populations instead of the true ones (Supplementary
aboutwhetherornotthereexistsbothahypotheticaldifference MethodsText1).SupplementaryMaterial,TableS9demonstrates
inSESincasesversuscontrols(whichwecannotquantify)anda thatevenLamp-LDdoesnotcorrectlyestimatethetruelocalan-
correlationbetweenancestryandSES(whichwecanquantify), cestry in a multi-way admixed population. Supplementary
HumanMolecularGenetics,2014,Vol.23,No.3 803
Material, Table S9 shows that the true CEU ancestry in the DISCUSSION
admixedpopulation(simulationdata)ismiscalledasGIHancestry
(17%)moreoftenthantrueGIHancestryismiscalledasCEUan- Weconductedagenome-wideassociationanalysisofTBcase–
cestry(9.4%).ThetrueSANancestryintheadmixedpopulation controlsfromtheadmixedSACpopulation,resultingintheiden-
(simulation data) is miscalled as YRI ancestry (15.8%) slightly tification of a low-frequency variant at the rs17175227 SNP.
moreoftenthantrueYRIancestryismiscalledasSANancestry After imputation we also identified a rare variant at rs122940
(14.2%).Duetothelimitationinaccuracyofthesemethodsinin- 76SNPattheborderlineofgenome-widesignificanceandwe
ferringlocalancestryinmulti-wayadmixedpopulations,wecon- moderately replicated a recently reported susceptibility locus,
clude that case-only admixture mapping is impractical in the rs2057178. Similar results were obtained when including age
multi-wayadmixedpopulationssuchtheSAC. andgenderascovariatesintheanalysis(SupplementaryMater-
ial,TableS1).Powertodetectassociationisafunctionofallele
frequencyandrarevariantsareunderpoweredwhensamplesizes
DeviationinlocalancestryintheSAC
arelimited.Becauseoftheimperfectasymptoticdistributionof D
o
Amajorlimitationofadmixturemappingandadmixtureassoci- mixed model association or logistic regression in the specific w
n
ationisthattheinferenceoflocus-specificancestryincomplex caseoflow-frequencyvariants,whichmayoftenreachgenome- lo
a
multi-way admixed populations such as the SAC may suffer wide significance; we computed Fisher’s exact test values for de
fcrhormomsopsuorimoualsrdeegvioiantsiofrnosmincaavseesraagnedlcoocnatlroalnsc,erestsruyltaintgpainrtsicpuulrair- vatairoinanPts-vthaaluteasc.hTiehviesdrtehseulmteodstinsigrns1ifi7c1a7n5t2m27ixnedotmroeadcehliansgsotchie- d from
onuostmcaesaes-uornelysyasdtemmixattiucreinaascscoucriaactiioesnsar(i4si5n,4g6f)r.oSmimthuelavtiioolnatmioanys guesntoomdeem-woindsetrcautet-tohfaft.aImraproervtaanritalyn,tFisisnhoetrg’seneoxmacet-wteisdteaslliogwniefid- https
ofmodelassumptionssuchasrandommating,numberofgenera- cant although it achieved significant mixed model association ://a
c
tionssinceadmixture,andrecombinationrate(45,46)inestimat- P-values.Ingeneral,theanalysisofrareorlow-frequencyvariants ad
ing local ancestry. Here, we conducted a real data-based poseschallengesatanydatasetofsmallertolargersamplesize. em
evaluationoflocalancestryusingtheSACpopulation.Weeval- Fisher’sexacttest,asitsnameimplies,maybeawell-calibrated ic.o
uated the deviations in local ancestry inferred from Lamp-LD test for rare SNPs and works fine with small sample sizes. up
(MaterialsandMethodsText2).SeveralSNPsinatotalof69 However,Fisher’sexacttestlacksflexibilitytohandlemultiple .co
m
chromosomal regions (Supplementary Material, Table S10) predictors, does not account for the contribution of the sample /h
showedstrongdeviationinlocalancestry(4standarddeviations structuretothephenotypeandmayinflated.Ourstudyaddressed m
g
aboveorbelowthegenome-wideaverage)intheSAC.Thesein- theseissues.UntilsuchtimeasGWASisabletoaccountforrare /a
ferredlocalancestriesregionswithlargedeviationareconsistent variants,ourcurrentrecommendationistoapplyFisher’sexact rtic
le
between cases and controls (Fig. 4), indicating that the devia- testtothefewlow-frequencyvariantsthatareshowntobesignifi- -a
b
tions in cases are not true disease associations and could lead cantfromthemixedmodelinGWAS.Forcombinedtypedand s
to spurious admixture associations if case-only admixture imputation GWAS, Table 4 displays the top 24 genetic trac
mappingisemployed(45,46).Weconcludethatcase-onlyad- markerswithsuggestiveP-values(,orequalto10206)obtained t/2
3
mixture mapping is currently impractical in multi-way fromtheassociationanalysiswiththeTBphenotypeintheSAC, /3
/7
admixedpopulationssuchtheSAC. includingsomepreviouslyreportedTBSNPsthatwediscussin 9
6
thenextsection.Anadditional36typedand62imputedSNPs /5
9
attained suggestive levels of significance (10205–10206), and 40
6
can be found in Supplementary Material, Tables S1 and S2, 6
b
respectively. y
g
Imputation-based meta-analysis, combining data from the u
e
GWAS conducted at different locations, was successful in a st o
recent study of severe malaria in three African populations n
1
(47).Toachieve sufficientpowerandidentifysharedriskloci 2
A
with a previously reported African TB case–control studies p
(13,14),aGWASmeta-analysiswasperformedunderrandom- ril 2
0
effect and binary-effecs models. In combining GWAS data 1
9
acrossthesestudies,twoloci(rs2057178andrs11031728)had
anassociationresultwithgenome-widesignificance,andshowed
strong effects in both our study and the previous African TB
case–controlstudy(14).
We used a combination of two complementary methods to
examinewhetherthegeneticancestrycontributioncanincrease
TBrisk,andevaluatedthecontributionofSEStotheancestry-
TBrelationshipintheSAC.Ourresultsdemonstratedsignificant
Figure4.Deviationsintheaveragelocalancestry(DLA)intheadmixedSouth
AfricanColouredpopulation(4standarddeviationsaboveorbelowthegenome- evidenceofanassociationbetweenSANancestryandTBstatus
wideaverage).Weplotregionsofstrongdeviationinaveragelocalancestryin thatisnotconfoundedbySES.Thisisanimportantepidemio-
case(x-axis)versuscontrol(y-axis)samplesoftheSAC.Theplotshowsspurious logicalresultandillustratesthevalueoftheinclusionofadmix-
deviationsinaveragelocalancestry(e.g.regionsinwhichthemodelledancestral
tureassociationmethodsinthesetofmethodsusedtoconduct
populationisunusuallydifferentfromthetrueancestralpopulationduetothehis-
TBassociationstudiesinthispopulation.Whentheextremely
toricalactionofnaturalselection).
804 HumanMolecularGenetics,2014,Vol.23,No.3
Table4. Top24geneticmarkersobtainedfromtheassociationanalysiswiththeTBphenotype(typedandimputeddatasets)
SNP CHR POS A1/A2 Call INFO MAF P OR ClosestGene Type
rs10916338 1 228715923 A/G 1 1 0.14 4.74e206 0.40 RNF187 Typed
rs1925714 1 226769912 A/G 0.99 0.97 0.14 1.80e207 0.35 RNF187 Imputed
rs6676375 1 240941508 C/T 0.96 0.87 0.13 7.58e207 0.33 PLD5 Imputed
rs1075309 2 5253598 C/T 1 1 0.09 7.07e207 0.35 SOX11 Typed
rs2202157 3 26518424 C/T 0.88 0.65 0.16 7.80e207 0.25 Unknown Imputed
rs958617 4 78824393 A/G 0.74 0.50 0.30 2.90e207 0.15 CNOT6L Imputed
rs2505675 6 2300674 C/T 0.87 0.61 0.15 3.87e206 0.22 LOC100508120 Imputed
rs17217757 8 106613321 C/G 1 1 0.18 3.09e206 0.43 ZFPM2 Typed
rs1934954 10 96792202 C/T 1 1 0.02 2.81e206 0.14 CYP2C8 Typed
rs12283022 11 102485804 A/G 0.76 0.48 0.24 1.88e206 0.14 DYNC2H1 Imputed
rs12294076 11 102632275 C/T 0.85 0.60 0.17 9.57e208 0.16 DYNC2H1 Imputed D
rs2057178 11 32364187 G/A 0.84 0.80 0.08 2.71e206 0.62 WT1 Imputed ow
rs11031728 11 32363616 C/G 0.843 0.80 0.08 2.86e206 0.61 WT1 Imputed nlo
rs7105967 11 102434653 C/T 0.75 0.45 0.25 3.51e206 0.15 DCUN1D5 Imputed a
d
rs7947821 11 102452675 C/T 0.75 0.47 0.25 1.95e206 0.14 DCUN1D5 Imputed e
d
rrss61593080144402 1123 7461246023163764 AC//TG 00..8917 00..6931 00..2165 44..4762ee220066 00..2337 EV2WFA78 IImmppuutteedd from
rs17175227 14 70502050 A/G 1 1 0.02 8.99e209 0.14 SMOC1 Typed h
rs40363 16 3449057 A/G 0.75 0.51 0.28 3.13e206 0.09 NAA60 Imputed ttp
rrss2485317389507 2211 4412123186852759 CC//TG 00..8808 00..6665 00..1350 21..4508ee220067 00..3202 DC2SCCDA2M IImmppuutteedd s://a
c
rs3218255 22 35874432 A/G 0.80 0.66 0.31 2.32e207 0.25 IL2RB Imputed ad
rs5928363 X 33784063 A/C 1 1 0.02 3.72e206 0.12 Unknown Imputed em
rs142513793 X 47906480 A/C 1 1 0.03 1.84e207 0.20 Unknown Imputed ic
.o
u
p
A1/A2arereference/derivedalleles. .c
o
m
/h
m
high incidence of TB in the SAC population is considered, We detected an association at the previously identified WT1 g
taongceetshtreyriwsidtehriovuerdfifrnodmintghethSaAt aNsaingdniofitchaenrtApferirccaenntpaogpeuolaftitohnesir, ltohceulsimthiattedissaalmmopsletgseizneo.mSee-wcoindde,siigmnpifiuctianngtimniosusirndgatgaedneostpyiptee /article
itappearspossiblethatthereisanelementofpopulationlevel dataofacomplexadmixedpopulationisanimportantchallenge -a
b
genetic susceptibility to this disease. Although, our study had withseveralissuesrelatedtothechoiceandsizeofexistingref- stra
only82caseswithSESinformation,itwaspossibletodrawa erencepanels.Inparticular,theimputationofmissinggenotype c
conclusionabouttheextenttowhichbothahypotheticaldiffer- dataoftheadmixedSACpopulationwassuboptimal.Nonethe- t/23
enceinSESincasesversuscontrols(whichwecouldnotquan- less,theincreasednumberofSNPsgeneratedbyimputationana- /3/7
tify) and a correlation between ancestry and SES (which we lyses was useful in this study, yielding the replication of TB 96
quantified)couldleadtotheancestryanalysisbeingconfounded susceptibility loci (14). Third, despite applying Fisher’s exact /59
4
bySES.Specifically,ifancestryandSESwerecorrelated,this testtoaddressthecaseofrarevariants,theimplementationof 0
6
correlation would exist irrespective of TB status and would newer sequencing technologies is still required to search for 6 b
exist in both cases and controls, regardless of whether or not rareriskvariants.Thismaypotentiallyprovidecrucialinsights y g
thereisameanSESdifferencebetweencasesandcontrols.We in identifying TB susceptibility genes and, therefore, inform ue
s
justified our conclusion by the results we presented in Florez the development of novel interventions. Fourth, the between- t o
etal.(25),inwhichtheyobservedasimilarcorrelationbetween studyheterogeneityduetodivergentevolutionaryandmigratory n 1
ancestryandSESinbothcasesandcontrols. histories,variabilityofpatternsoflinkagedisequilibriumacross 2 A
Ourstudyisthefirstancestry-specificGWASofTBriskinthe distinctethnicgroups,andthedifferingimputationaccuracyare p
complexadmixedSACpopulation.Oursubsequentimputation- majorlimitationsforfinemappingofGWASsignalsinmulti-way ril 2
0
based meta-analysis and trans-ethnic fine mapping including admixedpopulationssuchastheSAC.Lastly,amajorlimitation 1
9
otherGWAS onAfrican populations confirmedlociidentified ofadmixturemappingandcase-onlyadmixtureassociationana-
previously. The results (Table 2 and Supplementary Material, lysis is currently due to errors in inference of local ancestry in
TablesS3andS6)demonstratedthattrans-ethnicfinemapping complexmulti-wayadmixedpopulationssuchastheSAC.Our
is a powerful approach to better characterize TB susceptible simulationresultdemonstratedthatexistingmethodsmayattain
lociindiversepopulations. highaccuracyonaveragebutmaysufferfromspuriousdeviations
Overall,somelimitationsshouldbenotedinourcurrentstudy. inaveragelocalancestryatparticularregions.Thesedeviations
First,thepresentstudyisunderpoweredtodetectriskvariantsof wouldbepresentinbothcasesandcontrols,andwouldleadto
smalleffectsize,becauseofourmodestsamplesize.However, spuriouscase-onlyadmixtureassociations.
weaddressedtheseissuesbycomputingthepowertoreplicate Overall,ourGWASprovidespromisingresultsandsuggests
TBpublishedoddsratioandforeachpublishedassociationwe that more power can be obtained by designing an extension-
specified whether or not the published odds ratio lies within association model that can combine both SNP case–control
the95%CIestimatedinourcurrentdataset.Thisallowsustoin- andadmixtureintheSAC.Prioritiesforfutureworkwillbeto
terpretthefailuretoreplicatepreviousTBfindingsinourstudy. examine an accurate, unbiased estimation of the ancestry at
HumanMolecularGenetics,2014,Vol.23,No.3 805
everySNPinthismulti-wayadmixedpopulationtopotentially Imputationofmissinggenotypes
provide crucial insights into identifying disease genes. This
ToaccountforthepopulationstructureintheadmixedSACin
willprovideamethodtoaccountforacombinedgenome-wide
imputingtheuntypedgenotypes,weconsideredtheimputation
SNP case–control and admixture analysis in a multi-way
modelbasedonpopulationgeneticparametersinthecoalescent
admixedpopulationsuchastheSAC.
framework implemented in IMPUTE2 (26). Exploring the
advantage of the model in IMPUTE2, we combined all avail-
able reference-phased haplotype data from both release 2
MATERIALSANDMETHODS
oftheHapMap3dataset(NCBIBuild36)(includesUtahresi-
Samplesandgenotypequalitycontrol dents (CEPH) with Northern and Western European ancestry
(CEU),JapaneseinToyko(JPT),ChineseinDenver,Maasaiin
TheSACpopulationunderstudyislocatedinthemetropolitan
Kinyawa,ToscaniinItalia(TSI),GIH,AfricanAncestryinSouth-
area of Cape Town in the Western Cape Province in South
west(ASW),LuhyainWebuye(LWK),MexicanAncestryinLos
Africa (12). Since the ethnicity, SES and HIV infection may D
Angeles(MEX),HanChineseinBeijing(CHB)andYorubain o
be confounders inTB associationstudies (2–4),this areawas w
selectedduetothehighincidenceofTBaswellastheuniform IYbRadI,anBr(iYtiRshI)f;roamndE1n0g0l0anGdeannodmSescoPtlraonjedct(G(2B8R))i,nFclinundiesshCfrEoUm, nload
ethnicity, SES and low prevalence of HIV (1). This is due to e
Finland(FIN),HanChineseSouth(CHS),PuertoRican(PUR), d
thefollowingreasons: ChineseinDenver(CHB),JPT,LWK,MXL,ASW,TSI,Colom- fro
m
(1) UniformethnicityandSESstatusareimportantindisease bianinMedellin(CLM)andIberianpopulationsinSpain(IBS). h
associationstudies,asitremovessomeoftheconfounding WedecidedtoimputeSNPsbysplittingeachchromosomeinto ttp
s
variables. 5Mb regions for analysis by IMPUTE2. On both resulting ://a
(2) LowprevalenceofHIVisimportantbecauseinthepresence imputeddatasets,post-imputationqualitycontrolsweresimilarly ca
d
of HIV infection, an individual has a greatly increased conductedinordertoaccountforimputationuncertainty. e
m
chanceofprogressingtoTBdiseaseonceinfected,simply ic
.o
becauseoftheirimpairedimmunesystem,andnotnecessar- u
p
ilybecauseoftheirgeneticsusceptibility. Admixtureestimationandprincipalcomponentanalysis .c
o
m
ThedefinitionforTBdiagnosisandrecruitmentofappropriate ForadmixtureanalysisintheSAC,weincludedatotalof25un- /h
controlsfor infectious diseases suchas TB has been shownto related SAN (5 samples of Ju|’hoan from HGDP) and 20 mg
fboeriem,tphoertTaBntpinattiheentisntienrpthreistasttiuodnyowfGerWeiAdSenrteifisueldtsth(4ro8u).gThhbearec-- Jgue|n’ohmoaen-wsiadmepSlNesPodbattaaionfefdoufrropmopuSlcahtiloenbsufsrcohmetthealI.nt(e5r0n)a,tiaonnd- /article
teriological confirmation (smear positive and/or culture alHapMap3project(27),includingEuropean(CEU),Yoruba -a
b
pliovsinitgivuen).deCrothnetrsoalms ewceornedsietiloencsteidncflruodmingthSeESsasmtaetucsoamndmauvnaiitly- ((YCHRIB),+GJuPjTar)at(iSuIpnpdlieamnsen(tGarIyH)MaantedriaCl,hiTnaebslee aSn7d). JWapeanpeesre- strac
abilityofhealthfacilities.Thesehealthyindividualshadnopre- formedthequality-controlfiltersoneachpopulationseparately. t/23
vious history of TB disease or treatment. Approval from the After the quality-control filter checks, the resulting merged /3/7
Ethics Committee of the Faculty of Health Sciences, Stellen- datasetoftheSACwithinitsfiveputativeancestralpopulations 96
boschUniversity(projectnumber95/072)wasobtainedbefore had272798autosomalSNPs.Weadditionallyremoved37556 /59
4
blood samples were collected with informed consent, and A/TandC/Gand6176sexSNPsfromthemergeddatasetthat 0
6
knownHIV-positiveindividualswereexcludedfromthestudy. contained 272798 autosomal SNPs, resulting in a dataset of 6 b
ThestudysamplesweregenotypedontheAffymetrix500K 229076 SNPs. We additionally remove each SNP that has an y g
chip and SNP calling was done as described by De Wit et al. LD r2 value of .0.1 with any other SNP within a 50-SNP ue
s
(21). Quality-control filters were applied to 797 cases and 91 sliding window (advanced by 10 SNPs at a time). We finally t o
controls.Atotalof6450SNPsfailedtheminorallelefrequency obtained the subset of unlinked SNPs (n¼49930). We then n 1
(MAF,1%)andmissingnesstest(GENO.0.05),aswellas appliedthealgorithmimplementedinADMIXTURE(51)tode- 2 A
theHWEtestincontrols(alphalevel0.0001).Weretained390 terminetheindividualadmixtureproportionintheSACdataand p
887 SNPs for 888 individuals (381558 autosomal SNPs; 797 inbothcase–controlonlydataoftheSAC.SupplementaryMa- ril 2
0
casesand91controls;489malesofwhich444arecasesand45 terial,FigureS6(AandB)andTableS3showthatgenome-wide 1
9
controls). Further relatedness analysis using PLINK (49) was ancestryvarieswidelyacrossSACindividuals,SAN(33+1%),
conductedandresultedintheremovalof155relatedindividuals, YRI(28+2%),CEU(19+1%),GIH(13+0.9%)andCHB+
producingadatasetsuitableformethodsthatassumeindepend- JPT (7+0.5%). To evaluate the extent of substructure in the
entsamples.Ithas390887SNPsfor733individuals(381558 SAC (separating cases and controls as distinct groups) and
autosomal SNPs, 642 cases and 91 controls; 406 males of examine whether stratification can be accounted for in the
which361 are cases and45 controls). The resulting cases and GWAS,thesmartpcaprogrammeintheEIGENSOFTpackage
controls can provide the same amount of power in detecting (52,53) was applied to the merged data of the 888 SAC
theassociationsignalas159casesand159controls.Tofurther samples(alsocase-onlyandcontrol-only)anditsfiveancestral
check the homogeneity of the samples, we additionally per- populations(SupplementaryMaterial,Fig.S6C)onautosomal
formed the identity-by-state (IBS) permutation test, where SNPsandSNPsontheXchromosome(SupplementaryMaterial,
case–control labels were permuted, and then recalculated Fig.S6D).Weadditionallyregressedcase–controlstatusfrom
betweengroupmetricsbasedontheaverageIBS(fixed10000 autosomal SNPs against the two first principal components in
permutations). order to quantify the difference in genetic ancestry between