全 文 :书!"#$%&
,2016,36(6):1269-1277
犃犮狋犪犅狅狋.犅狅狉犲犪犾.犗犮犮犻犱犲狀狋.犛犻狀.
!"#$:10004025(2016)06126909 犱狅犻:10.7606/j.issn.10004025.2016.06.1269
%&(
:20151012;)*&%+(:20160428
,-./
:
()*+%,-
(31401457);./01234546789(SDAIT0204);:.%;<4=>?@
0123
:
A B(1978-),C,DE,FGHI,JKLMNOPQR
BCDEFGHI=JK
! "1,#$%1,&(2,) *1,#$+1,,-.1,/011
(1./034+%XNOGHY,Z[250100;2\]^<_`a,./\]265200)
L M:b,cdef67Sgb,cdhijklmnopqrst(SNP)uvwx8kyz{|}~ BLUP
z=stkR<
,
=R<
(GEBV),*k_st=28R<
ef
。
,cdefkJKc
———
v¡x8k¢z£i¤
、
zk}~¥
¦
、
stk¢z§¨{©
、
rªPQ«
,
¬§,cdefk®¥¯kGH°±²
,
³´µb,
cdef67¶NOR<·¸¹º¬§»k°¼
。
NOP
:
b,cdef
;
NO
;
=R<
QRST$
:Q789 !UVWX:A
犚犲狊犲犪狉犮犺犘狉狅犵狉犲狊狊狅狀犘犾犪狀狋犌犲狀狅犿犻犮犛犲犾犲犮狋犻狅狀(犌犛)
犪狀犱犐狋狊犃狆狆犾犻犮犪狋犻狅狀犻狀犕犪犻狕犲犅狉犲犲犱犻狀犵
SUNQi1,LIWenlan1,CHENLitao2,ZHAOMeng1,LIWencai1,
YUYanli1,MENGZhaodong1
(1MaizeInstitute,ShandongAcademyofAgriculturalSciences,Jinan250100,China;2LaiyangCitySeedCorporation,Laiy
ang,Shandong265200,China)
犃犫狊狋狉犪犮狋:Markerassistedselection(MAS)technologycouldrealizedirectgeneticselection,butitmust
baseonQTLmapping.Genomicselection(GS),asthenewestMASmethod,hasmuchadvantagecom
paredtotraditionalMAStechnology,especialyQTLmappingnotnecessary.Inthispaper,thefactorsaf
fectingpredictionaccuracyofGSwerereviewed,includingtrainingpopulationtype,predictionmodel,
markernumber,populationsize,populationstructure,hereditaryoftraitsandsoon.Theapplicationof
GSinmaizebreedingwasalsointroducedaswelashybridsperformanceprediction.Wethenpredicated
thefutureresearchandapplicationofGSinmaizebreeding.
犓犲狔狑狅狉犱狊:genomicselection(GS);maize;GEBV
Withrapiddevelopmentofthemolecularbiol
ogy and genomics, markerassisted selection
(MAS)emergedasthetimesrequire.MAStech
nologyisasakindofcropgeneticimprovement
methodcombingthephenotypicandgeneticvalue,
whichcanrealizegeneticdirectselectionandeffec
tivepolymerization[1].Whencomplextraitscon
troledby multiplegenesneedtobeimproved,
MAShastwoaspectsofflaws.First,selectionof
theprogenypopulationisestablishedonthequan
titytraitslocation(QTL)mapping.Buttheresult
ofQTLmappingbasingonthebiparentalpopula
tionshasnouniversalityandcouldn’tbeapplied
accuratelyinbreeding[2].Second,theimportant
traitswerecontroledbylotsofsmal effective
genes,lack ofappropriatestatistic methodand
breedingtechnology which wil apply quantity
genestocomplextraitsimprovement[3].NewMAS
technologygenomicselection(GS)emergedasthe
timesrequire.
1 Originationandadvantageofge
nomicselection(GS)
Meuwissenfirstputforwardgenomicselection
(GS)breedingstrategy.GSusesa“trainingpopu
lation”ofindividualsthathavebeengenotypedand
phenotyped. Best linear unbiased prediction
(BLUP)modelisestablishedonthebasisofthe
genotypedresultofanindividualanditsbreeding
value (Meanperformanceofcrosseswithsame
tester)forthetrainingpopulation.Thebreeding
valueof“Candidatepopulation”isestimatedby
BLUPmodelandgenotypicdata.withoutcrossto
testerand phenotypesrecord[4].BLUP model
takesgenotypicdataofuntestedindividualsand
producesgenomicestimatedbreedingvalues(GEB
Vs).TheseGEBVssaynothingofthefunctionof
theunderlyinggenesastheidealselectioncriteri
on[5].GenomicselectionbasisofGEBVsissuperi
ortotraditionalbreedingforincreasinggainsper
unittimeevenifbothmodelsshowthesameeffi
ciency.Inprinciple,phenotypesvalueofthecan
didateindividualsisnonessentialfortheselection,
henceshorteningthelengthofthebreedingcy
cle[6].
Genomicselectionhaveseveralmeritscom
paredtothetraditionalMAS.(1)QTLmappingis
notnecessaryforGS.Genomicselectiondiffers
frompreviousstrategiessuchaslinkageandassoci
ationmappinginthatitabandonstheobjectiveto
maptheeffectofsinglegeneandinsteadoffocu
singontheefficientestimationofbreedingvalues
onthebasisofalargenumberofmolecularmark
ers,idealycoveringthefulgenome[5].(2)Ge
nomicselectionismoreprecisionespecialyforear
lyselection.Genotypinguseshighdensitymolecu
larmarkerswhichcanestimateal oftheQTL
effectsandexplainthegeneticvarianceformostof
thetraits.ButMASonlyusesseveralmarkersin
traitsselection.Sogenomicselectionismoreaccu
ratethan MAS[7].(3)Genomicselectioncan
shortengenerationinterval,accelerategeneticpro
gressandreduceproductioncost.Geneticprogress
ofGSismorethanphenotypicselection4% -
25%.CostofGSislessthantraditionalbreeding
26%-56%[8].(4)Selectionefficiencyoflowheri
tabilitytraitsishigherforGSthanMAS.(5)The
criterionofGSisbreedingvalue,sumofalofthe
alelegeneticeffectsforeachindividual.Itis
judgedbythemeanperformanceofitscrossproge
ny,nottheperformanceofitself.SoGSismore
accurate[9].
Genomic selection originated from animal
breedingduringlastcentury.Ithasbeenwidely
usedindairycattlebreedinginAmerica,Austral
ia,NewZealandandsoon[1011].Itwasalsoap
pliedinbroilerchickensandpigsbreeding[1213].
GS’applicationinplantbreedingwasdevelopedin
recentyears,whichfocusedonsimulationstudies.
Itisusedinmaize[14],wheat[15],tree[16],sugar
beet[17],Barley[18],triticale[19]andsoon.
Empiricalstudyisperformedinlargercompa
niessuchasMonsantoandPioneerDupond.Mark
SorrelsandJeanLucJanninkaretryingtouseGS
toincreasethespeedofvarietyimprovement3-4
times.TheworkiscarriedoutwithCYMMITand
performedfouraspectstoimprovetheyieldof
maizeandwheat[20].
Undertheabovecontext,theobjectiveofthis
studyistoreviewtheessentialfactorsaffectingthe
GSinplantbreeding.Maizeisessentialforglobal
foodsecurity.Moreresearchofgenomicselection
onmaizelauchedinrecentyears[2123].Thepaper
wilintroducetheadvanceontheapplicationofGS
0721 ! " # $ % & 36½
inmaizebreeding.Wethanputforwardthefuture
research whichshouldbecarriedoutin maize
breedinginChina.
2 Affectingfactorsofgenomicselec
tion
FactorsthataffectGSpredictionaccuracyof
includethenumberofmarkersusedforestimating
the GEBVs[10],trait heritability[7],calibration
populationsize[5],statisticalmodels[24],number
andtypeofmolecularmarkers[2526],linkagedise
quilibrium[27],effectivepopulationsize[28],rela
tionshipbetweencalibrationandtestset(TS)[2931]
andpopulationstructure[3234].
2.1 犜狉犪犻狀犻狀犵狆狅狆狌犾犪狋犻狅狀狅犳犵犲狀狅犿犻犮狊犲犾犲犮狋犻狅狀
Inanimalbreeding,weonlydiscussedGSin
thecontextofpopulationwidelinkagedisequilibri
um,wherethepopulationmightbedefinedasan
entirebreedofcattle,pig,orchicken.Theneed
forhighmarkerdensitiesinGSmaybereducedif
thecandidatepopulationconsistsofprogenyofthe
trainingpopulation.Inthatcase,anevenlyspaced
lowdensitysubsetofthemarkerstypedonthe
trainingpopulationcanbeusedonthecandidates,
andscoresforthefulcomplementofmarkerscan
beinferredbycosegregation[35].Becauseplantsof
tenproduceverylargefulsibships(anF2popula
tionderivedfromasingleF1byselfingisanexam
pleofsuchasibship),however,thereisalsoa
traditionofQTLdetection,MASandGSwithin
suchsibships[5].BernardocomparedF2,BC1,and
BC2populationsfromanadapted×exoticmaize
crossastrainingpopulationinthesimulationex
periment[14].Theresultindicatesthatgenomewide
selectionshouldstartatF2ratherthanbackcross
population,evenwhenthenumberoffavorableal
lelesissubstantialylargerintheadaptedparent
thanintheexoticparent.Comparedtonatural
populations,geneticbasisofF2populationsissim
plerbecauseF2populationsderivefromonlytwo
inbredlines.Sothebiparentalpopulationsize
mightbesmalerthanthatofnaturalpopulations.
Simulationstudieshavepreviouslyindicatedthat
forthreecyclesofgenomewideselectioninana
dapted×exoticcross,apopulationsizeof犖犆0=
144wasgeneralysufficient[21].Lowdensitymark
ersaresuitabletoF2populations[22].Buttwodis
advantagesofF2populationsexist.Biparentalpop
ulationrequiresseparatemodelfortrainingwithin
eachcross.TheBLUPmodelisonlysuitforthe
progeniesselectionfromthetwoparentallines.
TheprogenyofF2populationmustbeselectedby
thephenotypicvalueofF3testcrosses.Folowing
progenyselectionmaybeonlyaccordingtoBLUP
modelafterF3.
F2astrainingpopulationoftenbesuiltfor
crosspolinated plantsuch as maize.Yusheng
Zhaobasedonexperimentaldataofsixsegregating
populationsfromahalfdialelmatingdesignwith
788testcrossprogeniesfromanelitemaizebreed
ingprogram[23].InthestudyofVannesa犲狋犪犾.[36],
markereffectsestimatedin255diversemaizehy
bridswereusedtopredictgrainyield,anthesis
date,andanthesissilkingintervalwithinthediver
sitypanelandtestcrossprogeniesof30F2derived
linesfromeachoffivepopulations.
Wegenast犲狋犪犾.suggestedthatgenomicselec
tionwasappliedinplantbreeding,however,not
onlywithinaspecificbiparentalcrossorwithina
diversepanelofelitelinesbutalsoratherwithin
andamongcrosses[37].Selfpolinationplantoften
adoptnaturalpopulationsuchaswheatorsugar.
Würschumetalused924sugarbeetlinesastrain
ingpopulation.Theresultssuggestthatatraining
population derivedfrom intensively phenotyped
andgenotypeddiverselinesfromabreedingpro
gramdoesholdpotentialtobuilduprobustcalibra
tionmodelsforgenomicselection[17].Hans犲狋犪犾.
accessedtheaccuracyofGEBVsforrustresistance
in206hexaploidwheatlandraces[15].
2.2 犘狉犲犱犻犮狋犻狅狀犿狅犱犲犾狅犳犵犲狀狅犿犻犮狊犲犾犲犮狋犻狅狀
Genomicselectionmodelingtakesadvantageof
theincreasingabundanceof molecular markers
throughmodelingofmanygeneticlociwithsmal
effects[26,35,38].Overthelastdecade,simulation
andempiricalcrossvalidationstudiesin plants
haveshownGSismoreeffectivethantraditional
MASstrategiesthatuseonlyasubsetofmarkers
17216¾ A B,®:#$b,cdef67kGH°§¨¶NOR<·k¸(¿)
withsignificanteffects[57,39].
Estimationmethodsofaleliceffectsinclude
leastsquaresregression[40],ridgeregressionBLUP
(RRBLUP),principlecomponentanalysis[4142]
and Bayesregression[43].In essenceforleast
squares,chromosomefragmentsormarkersarese
lectedassociatedtothetraitsbygenomewideas
sociationstudies(GWAS)atthesametimeand
thentheeffectofthefragmentsisestimated[44].
RRBLUPmethodregardsthefragmenteffectsas
randomeffects.Themarkereffectwasestimated
bylinearmixed models.Thesum offragments
effectisbreedingvalueforanindividual[43].Bayes
methodscombinesthepriordistributionofmarker
effectvarianceanddatacolection.Frenquently
usedBayesmethodsconcludeBayesAandBayes
B.MaindifferencebetweenBayesAandBayesBis
thatBayesApermitsdifferentvariancefordiffer
entmarkersandBayesBpermitsthatthevariance
ofsomemarkersiszero[45].
Simulationstudiesshowthattheprediction
accuracyofBayesmethodisbestandleastsquares
isweakest.TheaccuracyrateofRRBLUPis
slightlysmalerthanBayesA.Evenso,RRBLUP
hasfouraspectssuperiorto Bayesian method.
First,Bayesianmethodiscomplexandneedsuper
computer.Butcomputerrequirementislowerand
calculationspeedishigherforRRBLUP.Marker
effectsareestimatedbyRRBLUPinSASPROC
IML[46].Second,prediction withinfamilieswas
moreaccurateinBLUPthanBayesB.Regression
coefficientbofRRBLUPisnearerto1thanBayes
A[47].Habier犲狋犪犾.showedthatRRBLUPismore
effectiveatcapturinggeneticrelationshipsbecause
itfitsmoremarkersintothepredictionModel[27].
Incontrast,BayesBismoreeffectiveatcapturing
LDbetweenmarkersandQTL.Third,RRBLUP
ismoreaccuratethanothermethodwhenthenum
berofQTLsincreasesortheheredityishigher[18].
Fourth,BLUPledtolowerinbreedingandasmal
lerreductionofgeneticvariancecomparedtoBayes
andPLS[48].From above,wecanconcludthat
BLUPmethodsisbetterthanBayesianregression
forplantmodels.
Inaddition,machinelearning methodsalso
canbeusedtopredictthemarkereffect,including
supportvectormachine(SVM),bootingandran
domforest(RF).Ogutu犲狋犪犾.comparedthese
methodsforgenomicselection.Theresultshows
thatthecorrelationbetweenthepredictedandtrue
breedingvaluesis0.547forboosting,0.497for
SVMs,and0.483forRF,indicatingbetterper
formanceforboostingthanforSVMsandRF[49].
2.3 犗狋犺犲狉犳犪犮狋狅狉狊犪犳犳犲犮狋犻狀犵狆狉犲犱犻犮狋犻狅狀犪犮犮狌狉犪犮狔
Ingenomewideselectionmethods,prediction
accuracyisaffectedbypopulationsize(犖),aver
agehereditaryoftraits(犺2)andmarkernumbers
(犖犕)[50].Simulationstudiesshowedthatthepop
ulationstructureisalsocrucialfortheprediction
accuracyingenomicselection[27].
Predictionaccuracyincreases with markers
density.Markersnumberonacertainlengthge
nomealsodirectlyaffectstotalinformationofge
neticmarkers.IfSSR markersdensityincreases
from0.25犖犲/犿狅狉犵犪狀 (犖犲,effectivepopulation
size)to2犖犲/犿狅狉犵犪狀,predictionaccuracywilbe
improvedfrom0.63to0.83.IfSNPmarkersden
sityincreasesfrom1犖犲/犿狅狉犵犪狀to8犖犲/犿狅狉犵犪狀,
predictionaccuracywilbeimprovedfrom0.69to
0.86.Evenatthehighesttesteddensitiesof2犖犲
SSRmarkersperMorganor8犖犲SNPmarkersper
Morgan,accuracyhadnotreachedaplateau[5].
Meanwhile,moremarkersnumber,moreeasyto
getthe Linkagedisequilibrium (LD)markers.
Emilyfoundthatinthebiparentalpopulations,
therewasnoconsistentgainingenomewidepre
diction(狉犿狆)fromincreasingmarkerdensitya
boveonemarkerper12.5cM[22].Zhao犲狋犪犾.re
vealedthattheaccuracywasnearlyreachingaplat
eauat800SNPswhenthenumberofmarkersvar
iedfrom100to800[23].Thereasonisthatgenome
issufficientlysaturated with markerswhenthe
predictionaccuracyarrivesataplateau[28,50].The
numberofmarkersneededforaccuratepredictions
ofgenotypicvaluesdependsontheextentoflink
agedisequilibrium (LD)between markersand
QTL[4]andalsoonthegermplasmunderconsidera
tion[18].
2721 ! " # $ % & 36½
Differentmarkertypehasdifferentpolymor
phisminformationcontent(PIC).ComparingSSR
andSNPmarkers,theyfoundthatforsimilarac
curacies,theSNPmarkersrequiredadensityof2
to3timesthatoftheSSR[5].
Simulationstudiesshowedthatthepopulation
sizeiscrucialforthepredictionaccuracyingenom
icselection[27].TheresultofEmily犲狋犪犾.indicated
thatpredictionaccuracy狉犿狆increasedaspopula
tionsizeNincreased.Inthebiparentalmaizepop
ulationandwiththehighestmarkersnumber犖犕,
(1213markers)andhereditary犺2= 0.30,the
predictionaccuracyforgrainyieldwas狉犿狆=0.19
with犖=48,狉犿狆=0.26with犖 =96,and狉犿狆
=0.33with犖 =192[22].ZhaoYushengobserved
amonotonicincreaseinthepredictionaccuracyfor
grainyieldwithincreasingpopulationsizewithout
anysubstantialdecreaseintheslope[23].Thestudy
ofBernardoalsoindicatedthatlagerpoluationsize
wouldgethigherpredictionprecision[14].ButF2
populationsizeof犖犆0=144wasgeneralysuffi
cient[21].
Trainingpopulationstructureisalsoanim
portantfactoraffectingpredictionaccuracyofge
nomicselectionfor multiparentalpopulations.
Trainingpopulationstructuresetmethodsconclude
randomsampling,unidirectionalsampling(selec
tingindividualswithhighestgenotypicvalues),bi
directionalsampling (selectingindividuals with
highestorlowestgenotypicvalues)[5051].Thisbi
directionalselectionshowedtobemuchmorepow
erfulthanrandomsampling[52].YushengZhaoob
servedasubstantiallossintheaccuracytopredict
genomicbreedingvaluesinunidirectionalselected
populations.Bidirectionalselectionisavaluable
approachtoefficientlyimplementgenomicselection
inappliedplantbreedingprograms[53].
Forthesametraitwithinthesamepopulation,
predictionaccuracy(狉犿狆)wilremainunchanged
fordifferentcombinationsofpopulationsize(犖)
andtraithereditary(犺2).Decreaseonh2canbe
compensatedbyaproportionalincreasein犖 (and
viceversa)sothat狉犿狆ismaintained.Ontheoth
erhand,traitswithinitialylowh2canbeevalua
tedwithlarger犖orthe犺2forasubsetoftraits
canbeincreasedbytheuseofadditionaltestingre
sources.Differenttraits,however,varyintheir
predictionaccuracyeven when 犖,犺2,and 犖犕
(markersnumber)areconstant.Yieldtraitshad
lowerpredictionaccuracythanothertraitsdespite
theconstant犖,犺2,and犖犕.Simulationresults
indicatedthat狉犿狆isalsolowestforyieldtraitse
venwhenits犺2isashighasothertraits.Plant
heightandlodgingarealwayspredictedmostaccu
ratelyfolowedbyfloweringtime[22].Empiricalev
idenceandexperienceonthepredictabilityofdif
ferenttraitsarenecessaryindesigningtraining
populations.
3 Genomicselectioninmaizebreed
ing
3.1 犗狉犻犵犻狀犪狋犻狅狀狅犳犌犛犻狀犿犪犻狕犲
ThekeytechnologyofGSisthemaizehybrid
predictionbyBLUPmodelwithmarkerseffectsor
coefficientofparentage.Itwasusedtopredictthe
singlecrossperformanceinmaizehybridbreeding
atfirst.TheBLUPmodelisestablishedbasedon
thetestedhybridsdataandthemarkersinforma
tionoftheirparents.Theperformanceofuntested
hybridsispredictedbytheBLUPmodelandthe
markersdataoftheparents[54].
Bernardodevotedhimselftohybridsprediction
byBLUPmodelinmaize[5558].Thecoefficientof
relativebetweentheoryandactualobservationwas
0.688~0.800byRFLPmarkers[54].BLUPissuit
ableforhybridperformancepredictionsincethe
traitonlyhasmoderateheritability.Predictionac
curacyofmolecularmarkereffectsishigherthan
phylogeneticrelationship[58].Withthedevelop
mentofmolecularmarkers,newmolecularmarker
typeemerged.Simplesequencerepeats(SSR)and
singlenucleotidepolymorphism(SNP)werewide
lyused.ManjeGowda犲狋犪犾.foundthatprediction
accuracyofflowertimeandplantheightwasabove
0.8withSSR markersinmaize[19].Researchof
Massman犲狋犪犾.indicatedthatpredictionaccuracy
ofgrainyieldwas0.8,androotloggingratiowas
0.87usingSSRmarkers[59].Buttheprediction
37216¾ A B,®:#$b,cdef67kGH°§¨¶NOR<·k¸(¿)
effectofgrainyieldwasonly0.50~0.66,androot
loggingratiowasonly0.31~0.45withcoefficient
ofparentage[55].Thenitindicatedthatmolecular
markerswasmoresuitableforhybridperformance
predictionthancoefficientofparentage.
ThenscientistsfoundthatBLUPwasnotonly
usedtohybridperformanceprediction,butalsothe
breedingvalueofindividualsamongthemaizepop
ulation.SoBLUPwasusedtoindividualsselection
ofF2populationinselectionandbreedingofinbred
lines.Hybridperformancepredictionlaythefoun
dationforthegenomewideselectioninmaize.
3.2 犃狆狆犾犻犮犪狋犻狅狀狅犳犵犲狀狅犿犻犮狊犲犾犲犮狋犻狅狀犻狀犿犪犻狕犲
Bernardo’slaboratorybegantostudyapplying
GStomaizebreedinginMinnesotaUniversityof
America[21].Theydidplentyofsimulationandem
piricalexperiments.PiephoinGermanandRobert
inBrazilalsotriedtostudyusingGSin maize
breeding[6061].GSutilityinmaizebreedingconsist
oftwosides,hybridsperformancepredictionand
improvementofinbredlines.Hedevotedtoinbred
linesimprovementusingGS.TheBLUPmodelof
biparentalpopulationsfromtwoinbredlinesison
lysuitfortheprogenyoftheparents.Genomewide
selection as proposed in maize involves two
steps[21].First,asegregatingmaizepopulationis
genotypedandevaluatedfortestcrossperformance
ofF3family.Basedonthegenotypicandphenotyp
icdata,breedingvaluesassociatedwithalargeset
ofmarkers(e.g.,256to512markers)arecalcu
latedforthetraitsofinterest.Significancetests
formarkersarenotused,andtheeffectsofal
markersarefittedasrandomeffectsinalinear
modelbybestlinearunbiasedprediction(BLUP).
Second,twoorthreegenerationsofselectionbased
onalmarkersareconductedinayearroundnurs
ery(e.g.,HawaiorPuertoRico)orgreenhouse.
Traitvaluesarepredictedasthesumofanindivid
ualplant’smarkervaluesacrossal markers,and
selectionissubsequentlybasedonthesegenome
wideprediction.Accordingtothesteps,Emily
(2013b)introgressedsemidwarfgermplasmtoU.
S.Cornbeltinbredandfoundthatgenomewidese
lectionfromCycle1untilCycle5eithermaintained
orimprovedonthegainsfromphenotypicselection
achievedinCycle1[62].
TheresultsofBernardoindicatedthatauseful
strategyfortherapidimprovementofanadapted×
exoticcrossinvolves7to8cyclesofgenomewide
selectionstartingintheF2[14].Benjamin犲狋犪犾.
demonstratedthatprogressiveselfinghadasignifi
cantand positiveimpacton genomicselection
gains.Inparticular,selfingtotheF8produceda
72%increaseoverF2gains[63].However,mostof
thegainsarerealizedbytheF5generation(95%of
theF8gains).AlsonotethattheF8andDHper
formedsimilarly,consistentwithpreviousobser
vations[64].
IntheresearchofBernardo,thetrainingpop
ulationisthespecificbiparentalpopulationsfrom
thetwoparentallines,sotheBLUPmodelissuit
fortheprogenyofthetwoinbredlines.Otherex
perimentsofGSinmaizeareaboutmultiparental
populationsastrainingpopulation.StudyofYush
engZhaowasbasedonexperimentaldataofsix
segregatingpopulationsfromahalfdialelmating
design.Asformaizeuptothreegenerationsare
feasibleperyear,selectiongainperunittimeis
highand,consequently,genomicselectionholds
greatpromisefor maizebreeding programs[23].
Theseresultofthestudymightbeasgenomicpre
dictionmodelforfurtherbreedingelitemaizelines
betweenthesixpopulations.InthestudyofVan
essa犲狋犪犾.,markereffectsestimatedin255diverse
maizehybridswereusedtopredictgrainyield,an
thesisdate,andanthesissilkingintervalwithinthe
diversitypanelandtestcrossprogeniesof30F2de
rivedlinesfromeachoffivepopulations[36].Poten
tialusesforgenomicpredictionin maizehybrid
breedingarediscussedemphasizingtheneedof(1)
acleardefinitionofthebreedingscenarioinwhich
genomicpredictionshouldbeapplied(i.e.,predic
tionamongorwithinpopulations),(2)adetailed
analysisofthepopulationstructurebeforeper
formingcrossvalidation,and(3)largertraining
setswithstronggeneticrelationshiptothevalida
tionset.
4721 ! " # $ % & 36½
4 Futureresearchinmaizebreeding
GSisjustbeginningtobeimplemented,butit
wiltakelongtimetobeusedinmaizebreeding.In
previousstudy,trainingpopulationwasonlyfrom
severalinbredlines,eveniftwoinbredlines.It
couldntbeimplementedbyotherbreedingpro
gram.Futureresearchshouldfocusontwosidesof
work.First,weshouldcommittobuildageneral
izedpredictionmodelforsomekindsofinbredlines
suchasyield,qualityandsoon.Butthesetraits
werecomplexcomposedofagreatdealofgenes.
TraditionalMAStechnologycouldntrealizethe
traitsselectioninmaizebreeding.973Plan“Basic
studyonbreedingofgenomewideselectionofyield
andqualitytraitsinmaize”hasbeencarriedoutin
2014.Theplanwilsystematiclyanalyzethege
neticbasisofmaizeyieldandquality,andthen
buildgenomewideselectionbreeding model.It
wil affordnewtechnologyfor maizebreeding.
Seond,inChina,abioticstresstolerancealsore
ducesthe yield seriously in maize especialy
droughttolerance.Droughtistheforemostfactor
restrictingmaizeproduction,oftenresultingin20
-50% maizeyieldreductioneveryyearinChi
na[65].Ifweestablishpredictionmodelofdrought
tolerance,itwilaffordthetheoryandtechnology
supportofmaizebreeding.Consequently,ourre
searchteamwilcarriedoutstudyonthegenomic
selectionprogramofdroughttolerance.
犚犲犳犲狉犲狀犮犲狊:
[1] STUBERCW,POLACCOM,SENIORML.Synergyofem
piricalbreeding,markerassistedselection,andgenomicsto
increasecropyieldpotential[J].犆狉狅狆犛犮犻犲狀犮犲,1999,39:
15711583.
[2] MOOSESP,MUMMRH.Molecularplantbreedingasthe
foundationfor21stcenturycropimprovement[J].犘犾犪狀狋
犘犺狔狊犻狅犾狅犵狔,2008,147:969977.
[3] BERNARDOR.Molecularmarkersandselectionforcomplex
traitsinplants:learningfromthelast20years[J].犆狉狅狆犛犮犻
犲狀犮犲,2008,48:16491664.
[4] MEUWISSENTH,HAYESBJ,GODDARD ME.Predic
tionoftotalgeneticvalueusinggenomewidedensemarker
maps[J].犌犲狀犲狋犻犮狊,2001,157:18191829.
[5] JANNINKJL,LORENZAJ,IWATAH.Genomicselection
inplantbreeding:fromtheorytopractice[J].犅狉犻犲犳犻狀犵狊犻狀
犉狌狀犮狋犻狅狀犪犾犌犲狀狅犿犻犮狊,2010,9(2):166177.
[6] HEFFNEREL,JANNINKJL,IWATAH,犲狋犪犾.Genomic
selectionaccuracyforgrainqualitytraitsinbiparentalwheat
populations[J].犆狉狅狆犛犮犻犲狀犮犲,2011,51:25972606.
[7] HEFFNEREL,SORRELLSME,JANNINKJL.Genomic
selectionforcropimprovement[J].犆狉狅狆犛犮犻犲狀犮犲,2009,49:
112.
[8] MAYORPJ,BERNARDOR.Genomewideselectionand
markerassistedrecurrentselectionindoubledhaploidversus
F2populations[J].犆狉狅狆犛犮犻犲狀犮犲,2009,49:17191725.
[9] MASSMANJM,JUNGHJG,BERNARDOR.Genome
wideselectionversusmarkerassistedrecurrentselectiontoim
provegrainyieldandstoverqualitytraitsforcelulosicethanol
inmaize[J].犆狉狅狆犛犮犻犲狀犮犲,2012,53(1):5866.
[10] SCHAEFFERLR.Strategyforapplyinggenomewideselec
tionindairycattle[J].犑狅狌狉狀犪犾狅犳犃狀犻犿犪犾犅狉犲犲犱犻狀犵犌犲狀犲狋犻犮,
2006,123:218223.
[11] GODDARDME,HAYESBJ.Genomicselection[J].犑狅狌狉
狀犪犾狅犳犪狀犻犿犪犾犅狉犲犲犱犻狀犵犌犲狀犲狋犻犮狊,2007,124:323330.
[12] DAETWYLER H D,VILLANUEVA B,BIJMA P.In
breedingingenomewideselection[J].犑狅狌狉狀犪犾狅犳犃狀犻犿犪犾
犅狉犲犲犱犻狀犵犌犲狀犲狋犻犮,2007,124:369376.
[13] TUL,WOOLLIAMSJA,SIGBJORNL.Theaccuracyof
genomicselectioninnorwegianredcattleassessedbycross
validation[J].犌犲狀犲狋犻犮狊,2009,183:11191126.
[14] BERNARDOR.Genomewideselectionforrapidintrogres
sionofexoticgermplasminmaize[J].犆狉狅狆犛犮犻犲狀犮犲,2009,
49:419425.
[15] HANSDD,BANSALUK,BARIANAHS,犲狋犪犾.Genom
icpredictionforrustresistanceindiversewheatlandraces[J].
犜犺犲狅狉狔犪狀犱犃狆狆犾犻犲犱犌犲狀犲狋犻犮狊,2014,127:17951803.
[16] MARIED,BOUVETJM.Genomicselectionintreebreed
ing:testingaccuracyofprediction modelsincludingdomi
nanceeffect[J].犅犕犆犘狉狅犮犲犲犱犻狀犵狊,2011,5(Supply7):12.
[17] WRSCHUMT,REIFJC,KRAFTT,犲狋犪犾.Genomicse
lectioninsugarbeetbreedingpopulations[J].犅犕犆犌犲狀犲狋犻犮狊,
2013,14:8592.
[18] ZHONGSQ,DEKKERSJCM,FERNANDORL,犲狋犪犾.
Factorsaffectingaccuracyfromgenomicselectioninpopula
tionsderivedfrommultipleinbredlines:abarleycasestudy
[J].犌犲狀犲狋犻犮狊,2009,182(1):355364.
[19] GOWDAM,ZHAOYS,MAURERHP,犲狋犪犾.Bestlinear
57216¾ A B,®:#$b,cdef67kGH°§¨¶NOR<·k¸(¿)
unbiasedpredictionoftriticalehybridperformance[J].犈狌
狆犺狔狋犻犮犪,2013,191:223230.
[20] ÀÁÂ,ÃÄÅ,ÆÇ],®.#${jrªb,cdef
GH°
[J].![34%&,2012,25(4):15101514.
WUYS,SHAOJM,ZHOURY,犲狋犪犾.Reviewsofge
nomewideselectionforquantitativetraitsinplants[J].
犛狅狌狋犺狑犲狊狋犆犺犻狀犪犑狅狌狉狀犪犾狅犳犃犵狉犻犮狌犾狋狌狉犪犾犛犮犻犲狀犮犲狊,2012,25
(4):15101514.
[21] BERNARDOR,YUJ.Prospectsforgenomewideselection
forquantitativetraitsinmaize[J].犆狉狅狆犛犮犻犲狀犮犲,2007,47:
10821090.
[22] EMILYC,BERNARDOR.Accuracyofgenomewideselec
tionfordifferenttraitswithconstantpopulationsize,herita
bility,andnumberofmarkers[J].犘犾犪狀狋犌犲狀狅犿犲,2013a,6
(1):17.
[23] ZHAOYS,GOWDAM,LIU WX,犲狋犪犾.Accuracyofge
nomicselectioninEuropeanmaizeelitebreedingpopulations
[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱 犃狆狆犾犾犻犲犱 犌犲狀犲狋犻犮狊,2012a,124:
769776.
[24] HESLOTN,YANGHP,SORRELLSME,犲狋犪犾.Genomic
selectioninplantbreeding:acomparisonofmodels[J].犆狉狅狆
犛犮犻犲狀犮犲,2012,52:146160.
[25] CHENX,SULLIVANPF.Singlenucleotidepolymorphism
genotyping:biochemistry,protocol,costandthroughput
[J].犘犺犪狉犿犪犮狅犌犲狀犲狋犻犮狊,2003,3:7796.
[26] POLANDJ,RIFETW.Genotypingbysequencingforplant
breedingandgenetics[J].犘犾犪狀狋犌犲狀犲狋犻犮狊,2012,5:92102.
[27] HABIERD,FERNANDORL,DEKKERSJCM.Theim
pactofgeneticrelationshipinformationongenomeassisted
breedingvalues[J].犌犲狀犲狋犻犮狊,2007,177:23892397.
[28] DAETWYLER H D,VILLANUEVAB,WOOLLIAMSJ
A.Accuracyofpredictingthegeneticriskofdiseaseusinga
genomewideapproach[J].犘犔狅犛犗狀犲,2008,3:3395.
[29] ALBRECHTT,WIMMERV,AUINGERHJ,犲狋犪犾.Ge
nomebasedpredictionoftestcrossvaluesinmaize[J].犜犺犲狅
狉犲狋犻犮犪犾犪狀犱犃狆狆犾犾犻犲犱犌犲狀犲狋犻犮狊,2011,123:339350
[30] CLARKS,HICKEYJ,WERFJ.Differentmodelsofgenetic
variationandtheireffectongenomicevaluation[J].犌犲狀犲狋犻犮
犛犲犾犲犮狋犻狅狀犈狏狅犾狌狋犻狅狀,2011,43:18.
[31] PSZCZOLAM,STRABELT,MULDERHA,犲狋犪犾.Relia
bilityofdirectgenomicvaluesforanimalswithdifferentrela
tionshipswithinandtothereferencepopulation[J].犑狅狌狉狀犪犾
狅犳犇犪犻狉狔犛犮犻犲狀犮犲,2012,95z:389400.
[32] SAATCHIM,MCCLUREMC,MCKAYSD,犲狋犪犾.Accu
raciesofgenomicbreedingvaluesinAmericanAngusbeef
cattleusingkmeansclusteringforcrossvalidation[J].犌犲
狀犲狋犻犮犛犲犾犲犮狋犻狅狀犈狏狅犾狌狋犻狅狀,2011,43:40.
[33] WINDHAUSENVS,ATLINGN,CROSSAJ,犲狋犪犾.Ef
fectivenessofgenomicpredictionofmaizehybridperformance
indifferentbreedingpopulationsandenvironments[J].犌犲狀犲狊
犌犲狀狅犿犲狊犌犲狀犲狋犻犮,2012,2:14271436.
[34] GUOZ,TUCKERDM,BASTENCJ,犲狋犪犾.Theimpactof
populationstructureongenomicpredictioninstratifiedpopu
lations[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱犃狆狆犾犾犻犲犱犌犲狀犲狋犻犮狊,2014,127:
749762
[35] HABIERD,FERNANDORL,DEKKERSJCM.Genomic
selectionusinglowdensity markerpanels[J].犌犲狀犲狋犻犮狊,
2009,182:343353.
[36] VANESSASW,ATLINGN,HICKEYJM,犲狋犪犾.Effec
tivenessofgenomicpredictionofmaizehybridperformancein
differentbreedingpopulationsandenvironments[J].犌犲狀狅犿犻犮
犛犲犾犲犮狋犻狅狀,2012,2(14):14271436.
[37] WEGENASTT,LONGINCFH,UTZHF,犲狋犪犾.Hybrid
maizebreedingwithdoubledhaploidsIV.Numberversussize
ofcrossesandimportanceofparentalselectionintwostage
selectionfortestcrossperformance[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱犃狆
狆犾犾犻犲犱犌犲狀犲狋犻犮狊,2008,117:251260.
[38] SOLBERGTR,SONESSONAK,WOOLLIAMSJA,犲狋
犪犾.Genomicselectionusingdifferentmarkertypesanddensi
ties[J].犑狅狌狉狀犪犾狅犳犃狀犻犿犪犾犅狉犲犲犱犻狀犵犌犲狀犲狋犻犮狊,2008,86
(10):24472454.
[39] LORENZANARE,BERNARDOR.Accuracyofgenotypic
valuepredictionsfor markerbasedselectioninbiparental
plantpopulations[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱 犃狆狆犾犻犲犱 犌犲狀犲狋犻犮狊,
2009,120:151161.
[40] WOLDH,JOHNSONNL,KOTZS.Partialleastsquares
[C].EncyclopediaofStatisticalScience.NewYork:Wiley,
1985:58191.
[41] SOLBERGTR,SONESSONAK,WOOLLIAMSJA,犲狋
犪犾.Reducingdimensionalityforpredictionofgenomewide
breedingvalues[J].犌犲狀犲狋犻犮犛犲犾犲犮狋犻狅狀犈狏狅犾狌狋犻狅狀,2009,41
(1):29.
[42] CROSSAJ,CAMPOSG,P?REZP,犲狋犪犾.Predictionofge
neticvaluesofquantitativetraitsinplantbreedingusingpedi
greeandmolecularmarkers[J].犌犲狀犲狋犻犮狊,2010,186(2):
713724.
[43] MEUWISSENTHE,SOLBERGTR,SHEPHERDR,犲狋
犪犾.AfastalgorithmforBayesBtypeofpredictionofgenome
wideestimatesofgeneticvalue[J].犌犲狀犲狋犻犮犛犲犾犲犮狋犻狅狀犈狏狅犾狌
狋犻狅狀,2009,41:2.
[44] WANG W YS,BARRATTBJ,CLAYTONDG,犲狋犪犾.
Genomewideassociationstudies:theoreticalandpractical
concerns[J].犖犪狋狌狉犲犚犲狏犻犲狑犌犲狀犲狋犻犮狊,2005,6(2):109118.
[45] ÈÉÊ,ËÌÍ,A.,cdef§¨¸[J].PQ,
2011,33(12):13081316.
LIHD,BAOZM,SUNXW.Genomicselectionanditsap
plication[J].犎犲狉犲犱犻狋犪狊,2011,33(12):13081316.
[46] SASInstitute.TheSASsystemforWindows.Release9.2.
6721 ! " # $ % & 36½
SASInst.,Cary,NC,2009.
[47] LUNDMS,SAHANAG,KONINGDJ,犲狋犪犾.Comparison
ofanalysesoftheQTLMASXIIcommondataset.I:Genomic
selection[J].犅犕犆犘狉狅犮犲犲犱犻狀犵狊,2009,3(Suppl.1):S1.
[48] BASTIAANSENJW,COSTER A,CALUSM P,犲狋犪犾.
Longtermresponsetogenomicselection:effectsofestima
tionmethodandreferencepopulationstructurefordifferent
geneticarchitectures[J].犌犲狀犲狋犻犮狊犛犲犾犲犮狋犻狅狀犈狏狅犾狌狋犻狅狀,2012,
4:316.
[49] OGUTUJO,PIEPHOHP,TORBENSS.Acomparison
ofrandomforests,boostingandsupportvectormachinesfor
genomicselection[J].犅犕犆犘狉狅犮犲犲犱犻狀犵狊,2011,5(Suppl3):
S11.
[50] DAETWYLERHD,WONGRP,VILLANUEVAB,犲狋犪犾.
Theimpactofgeneticarchitectureongenomewideevaluation
methods[J].犌犲狀犲狋犻犮狊,2010,185:10211031.
[51] JULIOI,JANNINKJL,AKDEMIRD,犲狋犪犾.Trainingset
optimizationunderpopulationstructureingenomicselection
[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱 犃狆狆犾犾犻犲犱 犌犲狀犲狋犻犮狊,2014,128(1):
145158.
[52] NAVABIA,MATHERDE,BERNIERJ,犲狋犪犾.QTLde
tectionwithbidirectionalandunidirectionalselectivegenoty
ping:markerbasedandtraitbasedanalyses[J].犜犺犲狅狉犲狋犻犮犪犾
犪狀犱犃狆狆犾犾犻犲犱犌犲狀犲狋犻犮狊,2009,118:347358.
[53] ZHAOYS,GOWDA M,LONGINFH,犲狋犪犾.Impactof
selectivegenotypinginthetrainingpopulationonaccuracy
andbiasofgenomicselection[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱犃狆狆犾犾犻犲犱
犌犲狀犲狋犻犮狊,2012b,125:707713.
[54] ÎÏÐ,Ñ}Ò.¸{ÓPQ%[M]."Ô:h34+%
67ÕÖ
,2007:185204.
ZHAIH Q,WANGJK.Applied QuantitativeGenetics
[M].Beijing:ChinaAgriculturalScienceandTechnology
PublishingHouse,2007:185204.
[55] BERNARDOR.Predictionofmaizesinglecrossperformance
usingRFLPsandinformationfromrelatedhybrids[J].犆狉狅狆
犛犮犻犲狀犮犲,1994,34:2025.
[56] BERNARDOR.Geneticmodelsforpredictingmaizesingle
crossperformanceinunbalancedyieldtrialdata[J].犆狉狅狆
犛犮犻犲狀犮犲,1995,35:141147.
[57] BERNARDOR.Bestlinearunbiasedpredictionofmaizesin
glecrossperformance[J].犆狉狅狆犛犮犻犲狀犮犲,1996,36:5056.
[58] BERNARDOR.Markerassistedbestlinearunbiasedpredic
tionofsinglecrossperformance[J].犆狉狅狆犛犮犻犲狀犮犲,1999,39:
12771282.
[59] MASSMANJM,GORDILLO A,LORENZANARE,犲狋
犪犾.Genomewidepredictionsfrommaizesinglecrossdata[J].
犜犺犲狅狉犲狋犻犮犪犾犪狀犱犃狆狆犾犾犻犲犱犌犲狀犲狋犻犮狊,2013,126:1322.
[60] PIEPHOHP.Ridgeregressionandextensionsforgenome
wideselectioninmaize[J].犆狉狅狆犛犮犻犲狀犮犲,2009,49:1165
1176.
[61] ROBERTOFN,JULIOCD,?DERCML,犲狋犪犾.Genome
wideselectioninfortropicalmaizeroottraitsunderconditions
ofnitrogenandphosphorusstress[J].犃犮狋犪犛犮犻犲狀狋犻犪狉狌犿,
2012,34(4):389395.
[62] EMILYC,BERNARDOR.Genomewideselectiontointro
gresssemidwarfmaizegermplasmintoU.S.CornBeltin
breds[J].犆狉狅狆犛犮犻犲狀犮犲,2013b,53:14271436.
[63] BENJAMINM,COMBEJL,TANKSLEYSD.Selfingfor
thedesignofgenomicselectionexperimentsinbiparental
plantpopulations[J].犜犺犲狅狉犲狋犻犮犪犾犪狀犱 犃狆狆犾犾犻犲犱 犌犲狀犲狋犻犮狊,
2013,126:29072920.
[64] BORDESJ,CHARMETG,VAULXRD,犲狋犪犾.Doubled
haploidversussingleseeddescentandS1familyvariationfor
testcrossperformanceinamaizepopulation[J].犈狌狆犺狔狋犻犮犪,
2007,154:4151.
[65] HU,RF,MENGEC,ZHANGSH,犲狋犪犾.Prioritization
formaizeresearchanddevelopmentinChina[J].犛犮犻犲狀狋犻犪
犃犵狉犻犮狌犾狋狌狉犪犛犻狀犻犮犪,2004,37:781787.
(
!"
:
#$%
)
77216¾ A B,®:#$b,cdef67kGH°§¨¶NOR<·k¸(¿)