the plant genome july 2016 vol. 9, no. 2 1 of 13

original research

Development of a High-Density Linkage Map and Tagging Leaf Spot Resistance in Pearl Millet Using Genotyping-by-Sequencing Markers

Somashekhar M. Punnuri,* Jason G. Wallace, Joseph E. Knoll, Katie E. Hyma, Sharon E. Mitchell, Edward S. Buckler, Rajeev K. Varshney, and Bharat P. Singh

AbstractPearl millet [Pennisetum glaucum (L.) R. Br; also Cenchrus ameri-canus (L.) Morrone] is an important crop throughout the world but better genomic resources for this species are needed to facilitate crop improvement. Genome mapping studies are a prerequisite for tagging agronomically important traits. Genotyping-by-sequencing (GBS) markers can be used to build high-density linkage maps, even in species lacking a reference genome. A recombinant inbred line (RIL) mapping population was developed from a cross between the lines ‘Tift 99D2B1’ and ‘Tift 454’. DNA from 186 RILs, the parents, and the F1 was used for 96-plex ApeKI GBS library development, which was further used for se-quencing. The sequencing results showed that the average num-ber of good reads per individual was 2.2 million, the pass filter rate was 88%, and the CV was 43%. High-quality GBS markers were developed with stringent filtering on sequence data from 179 RILs. The reference genetic map developed using 150 RILs contained 16,650 single-nucleotide polymorphisms (SNPs) and 333,567 sequence tags spread across all seven chromosomes. The overall average density of SNP markers was 23.23 SNP/cM in the final map and 1.66 unique linkage bins per cM covering a total genetic distance of 716.7 cM. The linkage map was further validated for its utility by using it in mapping quantitative trait loci (QTLs) for flowering time and resistance to Pyricularia leaf spot [Pyricularia grisea (Cke.) Sacc.]. This map is the densest yet re-ported for this crop and will be a valuable resource for the pearl millet community.

Published in Plant Genome Volume 9. doi: 10.3835/plantgenome2015.10.0106 © Crop Science Society of America 5585 Guilford Rd., Madison, WI 53711 USA This is an open access article distributed under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

S.M. Punnuri and B.P. Singh, Agricultural Research Station, Fort Val-ley State Univ., 1005 State University Drive, Fort Valley, GA 31030; J.G. Wallace, Dep. Crop and Soil Sciences, the Univ. of Georgia, Athens, GA 30602 and Inst. for Genomic Diversity, Cornell Univ., Ithaca, NY 14853; J.E. Knoll, USDA-ARS, Crop Genetics and Breed-ing Research Unit, Tifton, GA 31793; K.E. Hyma and S.E. Mitchell, Genomic Diversity Facility, Inst. of Biotechnology, Cornell Univ., Itha-ca, NY 14853; E.S. Buckler, USDA-ARS, Inst. for Genomic Diversity, Dep. Plant Breeding & Genetics, Cornell Univ., Ithaca, NY 14853; R.K. Varshney, International Crops Research Institute for the Semi-Arid Tropics (ICRISAT), Hyderabad, 502324 Telangana, India. SMP and JGW contributed equally to this work. Received 14 July 2015. Ac-cepted 29 Jan. 2016. *Corresponding author (punnuris@fvsu.edu).

Abbreviations: DArT, Diversity Arrays Technology; GBS, genotyping-by-sequencing; H2, broad-sense heritability; LD, linkage disequi-librium; LG, linkage group; LOD, logarithm of odds; NGS, next-generation sequencing; QTL, quantitative trait locus; RFLP, restriction fragment length polymorphism; RIL, recombinant inbred line; SNP, single-nucleotide polymorphism; SSR, simple sequence repeat

Core Ideas

• Pearlmillet[Pennisetum glaucum(L.)R.Br;alsoCenchrus americanus(L.)Morrone]isanimportantforageandgraincropinmanypartsoftheworldbutgenomicresourcesforthisspeciesareneededtofacilitatecropimprovement.

• Thereferencegeneticmapdevelopedusing150recombinantinbredlinescontained16,650single-nucleotidepolymorphismsand333,567sequencetagsspreadacrossallsevenchromosomes.

• Thismapisthedensestyetreportedforthiscropandwillbeavaluableresourceforthepearlmilletcommunity.

• Genomemappingstudiesareaprerequisitefortaggingagronomicallyimportanttraits.

• Genotyping-by-sequencingmarkerscanbeusedtobuildhigh-densitylinkagemaps,eveninspecieslackingareferencegenome.

Published May 6, 2016

2 of 13 the plant genome july 2016 vol. 9, no. 2

Pearl millet,widelyknownforitstolerancetoheat,droughtandsoiltoxicity,isgrownforbothgrain

andforageinmanypartsoftheworld,particularlyinwarm,dryregions(BurtonandPowel,1968;Chemisquyetal.,2010).Pearlmillethashigherwater-useefficiencyandnitrogen-useefficiencythanmanyothercereals(Muchow,1988;Mamanetal.,2006;Vadezetal.,2012)andshowsusefulgeneticvariationfortolerancetohightemperaturesduringseedlingestablishment(Peacocketal.,1993;Howarthetal.,1994)andduringreproduc-tivegrowthstages(Guptaetal.,2015)andcanthriveonacidic,sandy,orinfertilesoilswherefewothercropscangrow(AndrewsandKumar,1992).Forthesereasons,pearlmilletisanessentialstaplefoodgrainand/orfod-dercropinmanydevelopingcountries.

ThemarketforpearlmilletgrainisalsoincreasingintheUnitedStatesbecauseofconsumerspreferringgluten-freefoodanddemandformilletflourbymanyethnicgroups(Dahlbergetal.,2004;Guliaetal.,2007).Inaddition,alternativesourcestomaize-(Zea maysL.)andsoybean[Glycine max(L.)Merr.]-basedlivestockfeedaresoughttolowerproductioncostsforthepoultryindustryinthesoutheasternUnitedStates(Durham,2003;Farrell,2005;CunninghamandFairchild,2012).Wholepearlmilletgrainhasbeenshowntobeasatis-factoryfeedingredientforbroilerchickensandforeggproductionwhilereducingfeedcosts(Collinsetal.,1997;Davisetal.,2003;GarciaandDale,2006).Comparedtosorghum[Sorghum bicolor(L.)Moench],pearlmilletgrainofferslowerstarch,superiorproteinqualityandcontent,ahigherproteinefficiencyratio,andgreatermetabolizableenergylevelsforpoultrydiets(Sullivanetal.,1990;Bramel-Coxetal.,1992;Andrewsetal.,1993;Nambiaretal.,2011).Over70%oftheapproximately10MhaofpearlmilletgrownannuallyinIndiaissowntoF1hybrids(YadavandRai,2011;Yadavetal.,2011a)andthedevelopmentofpearlmilletgrainhybridsintheUnitedStateshasshownsomeprogress.Forexample,theUSDA-ARSatTifton,GA,incollaborationwiththeUniversityofGeorgia,released‘TifGrain102’asacommercialgrainhybrid(Durham,2003;Leeetal.,2004).TifGrain102offersseveraladvantagescomparedtootherrowcrops,especiallyitsabilitytogrowonsandy,acidicsoilswithminimuminputsanditsresistancetorootknotnema-tode(Meloidogyne incognitaKofoid&White),rust(Puc-cinia substriataEllis&Barth.var.indicaRamachar&Cummins),andPyricularialeafspot[Pyricularia grisea(Cke.)Sacc.(teleomorph:Magnaporthe grisea(T.T.Her-bert)M.E.Barr](HannaandWells,1989;Wilsonetal.,1989;Timperetal.,2002;Guptaetal.,2012).Becauseofitshighforagequality,pearlmilletisalsogrownasanannualfoddercropinthesoutheasternUnitedStates(BurtonandPowel,1968;Chemisquyetal.,2010).

Pearlmilletisdiploidwithsevenpairsofhomologouschromosomesandanestimatedgenomesizeof2350Mb(or2C=4.71pgbasedonflowcytometry),muchofwhichconsistsofrepetitivesequences(BennettandSmith,1976;WimpeeandRawson,1979;Marteletal.,1997;Jauhar

andHanna,1998;Thomasetal.,2000).SomeDNAmarkershavebeendevelopedandusedoverthepasttwodecadesinpearlmilletforgeneticresearchorforappliedbreedingandselection(HashandBramel-Cox2000;BidingerandHash,2004;Galeetal.,2005).Nonetheless,pearlmilletcropimprovementsuffersfromarelativelackofgeneticandgenomicresourcescomparedtomostothercereals.Characterizationandutilizationofpearlmilletdiversitycanbeaidedbyexpandingthe(currentlyfew)genomicresourcesavailableinthiscrop.

Geneticmarkersarethebuildingblocksforcon-structinglinkagemaps.Linkagemapsfurthersupportnumerousapplicationsinplantbreeding.Geneticmapsofseveralpearlmilletpopulationshavebeenmadeusingdifferentmarkersetsoverthepast20y(Liuetal.,1994;Devosetal.,2000;Qietal.,2004;Pedraza-Garciaetal.,2010;Supriyaetal.,2011;Sehgaletal.,2012).Recently,asimple-sequencerepeat(SSR)consensusmapwith174lociwasdevelopedusingfourRILmappingpopulations(Rajarametal.,2013).Despitetheseefforts,pearlmil-letlinkagemapsfrequentlyhavelargegapsatthedistalends,whichisprobablycausedby(i)alackofeithersuf-ficientmarkersorpolymorphismsintheseregions,(ii)extremelyhighratesofgeneticrecombinationintheseregionsrequiringlargenumbersofphysicallycloselylinkedmarkerstopermitlinkagedetection,and/or(iii)thenatureofthemarkersandparentsusedinthesestudies(Devosetal.,2000;Vadezetal.,2012).Asugges-tionthatthesegapsarecausedbysomecombinationofthelattertwoexplanationsisprovidedbySupriyaetal.(2011),whodemonstratedgreatlyimprovedgenomecov-eragewithDiversityArraysTechnology(DArT)markerscomparedwiththatprovidedbyavailableSSRmarkers.MostoftheremainingpreviousmapshavegenerallyreliedonSSRs,restrictionfragmentlengthpolymor-phisms(RFLPs),orrelatedmarkers;however,inmanycrops,linkagemapsbasedonSNPsarenowbecomingcommonbecauseofthelowcostofhigh-throughputsequencingmethods(Ganaletal.,2009;Kumaretal.,2012).Becauseoftheirabundanceinthegenome,SNPscanbeusedtobuildmuchdenserlinkagemapsthanothertypesofmarkers.SuchSNP-basedgeneticmapsarehighlyinformativeastheynotonlyrevealthecomplex-ityofgenomearchitecture(structureandorganization)butalsotracethegeneticbasisofQTLsunderlyingatraitwithbetterresolution(Krawczak,1999;Mammadovetal.,2012).Next-generationsequencing(NGS)technolo-gieshavefacilitatedtherapiddetectionofgenome-wideSNPmarkers.Genotyping-by-sequencingisonesuchpowerfulapproachtodevelopgenome-wideSNPdata-sets(Elshireetal.,2011).Thistechniqueusesrestric-tionenzymestoselectivelydigestgenomicDNA;next,‘barcoded’DNAadaptersareligatedtothefragmentstomultiplexmanysamplesinasinglesequencinglane(Elshireetal.,2011).Thechoiceofrestrictionenzyme(s)andmultiplexingmakesGBSaversatilesystemandtheabilitytomultiplexenableslow-cost,high-throughput-markerdiscovery(PolandandRife,2012).Importantly,

punnuri et al.: development of a high-density linkage map in pearl millet 3 of 13

italsoworksinlessexploitedcrops,includingthoseforwhichnoreferencegenomesequenceisavailablepub-licly,suchaspearlmillet.ThepotentialutilityofGBSmarkersindevelopinghigh-densitymolecularmapsforseveralcerealcrops,includingmaize,barley(Hordeum vulgareL.)andoat(Avena sativaL.)hasbeenexten-sivelyreviewedandshowntobeuseful(Heetal.,2014).Recently,apearlmilletlinkagemapwasalsodevelopedwith2809high-qualitySNPmarkersusingamodifiedGBSprotocol(Moumounietal.,2015).

Theobjectivesofthisstudyweretoconstructahigh-densitylinkagemapusingGBS-derivedmarkerstopro-videaplatformfordownstreamstudiesandtodevelopgenomicresourcesforthegreaterpearlmilletresearchcommunity.Thetwoparentsusedinthisexperiment(Tift99D2B1andTift454)arealsotheparentsofthecommercialgrainhybridTifGrain102(Hannaetal.,2005a,2005b).Floweringtimewaschosenasapheno-typictraitforQTLanalysistodemonstratetheutilityofthismap.Also,Tift99D2B1carriesgenesforresistancetoPyricularialeafspot(HannaandWells,1989)andhencethismappingpopulationwasusedtoevaluateresistancetothisdiseaseaswell.

Materials and Methods

Pearl Millet Mapping PopulationTheparentallinesusedinthisstudyareTift99D2B1andTift454,whereTift99D2B1wasusedasthefemaleparent.BothTift99D2B1andTift454aredwarf,early-maturinggraintypesthatsharegenesfromTift23D2.Tift99D2B1hasrustandPyricularialeafspotresistanceallelesandTift454hasnematoderesistanceandpollenfertilityrestorercapability.ThispopulationwasdevelopedbyDr.JeffreyP.Wilson(USDA-ARS(retired),Tifton,GA)andwasprovidedtoFortValleyStateUniversityaspartofthecol-laborativepearlmilletprojectfundedbyUSDA-NationalInstituteofFoodandAgriculture,Grant#GEOX-2008–02595toDr.BharatSingh(retired).Thepopulationusedforsequencingwasasetof184RILsattheF7generation.

Plant DNA Preparation for SequencingPlantleaftissuewascollectedfrom1.5-mo-oldseedlingsraisedinthegreenhouse.Thetissuewaslyophilizedfor8handthengenomicDNAwasisolatedwithaDNeasy96PlantKit(6)(QiagenInc.,Valencia,CA).TheDNAwasquantifiedtocontain10ngL–1persampleand50Lofeachsamplefrom184lineswassentin96-deepwellplatestotheGenomicDiversityFacilityatCornellUniversityinIthaca,NY,forGBSmarkerdevelopment.EachplateincludedDNAsamplesfrombothparentsandTifGrain102inrandomwellsaswellasarandomblankwellcontainingonlywater.

Genotyping-by-SequencingLibrarypreparationandsequencingwereperformedbytheGenomicDiversityFacilityatCornellUniver-sity,Ithaca,NY.Genomiccomplexityreductionwas

performedwiththeApeKIrestrictionenzyme(recog-nitionsiteG/CWCG)andsamplesweresequencedin96-plexonanIlluminaHiSeq2000(IlluminaInc.,SanDiego,CA).Onehundredandeighty-fourRILsweresequenced;fivesamplesyieldedlessthan5000readseachandwereexcludedfromfurtheranalysis.Single-nucleo-tidepolymorphismswerecalledfromtheremaining179lines.

Single-Nucleotide Polymorphism CallsRawFASTQfileswereprocessedtoSNPcallsusingtheGBSpipelineinTASSEL(version4.3.6)(Glaubitzetal.,2014).Readswerealignedagainst~19,000contigsofpearlmilletgenomesequenceprovidedbythePearlMilletGenomeSequencingConsortium(Varshneyetal.,unpublisheddata,2015)usingBowtie2(LangmeadandSalzberg,2012).Toseetheeffectofusingareferencegenomeforsequencealignment,wealsogeneratedamapthatdidnotusethereferencegenometoaligntags.Thispipelinewasidenticaltothatusedforthereference-basedmap,exceptthattagswerealignedtoeachotherusingtheUNEAK(Luetal.,2013)filterinTASSELver-sion5.2.1.15(commands--UTagCountToTagPairPluginand--UTagPairToTOPMPlugin).

Initial Map Generation and OrderingMapcreationwasdoneinthreeiterativesteps.AllscriptsandparametersusedinthisprocessareincludedinSupplementalFileS1.First,high-qualitySNPcallswereselectedbyfilteringforthosewithatleast70%coverageacrossRILsandwithallelefrequenciesbetween0.25and0.75.Sitesshowing>12%heterozygositywereremovedasprobableparalogmisalignments.AllRILsshowing>50%missingdataor>10%heterozygositywerethenremoved.Potentialoutcrosseswerealsoidentifiedbyusingthefrac-tionofrarealleles(minorallelefrequency0.05)ineachRILtodefineanormaldistribution.AllRILswhosevaluehad<1%probabilityafterBenjamini–Hochbergcorrec-tion(BenjaminiandHochberg,1995)wereexcluded.Fil-teringresultedinadatasetof146RILsand17,400SNPs.

ToorderSNPs,heterozygouscallswerefirstsetto“missing”andthegenotypesweretransformedtonumericalequivalentsusingTASSEL(Bradburyetal.,2007).TheSNPswerethenclusteredusingthehclust()functioninR(RCoreTeam,2014).Theclustertreesweresplitatvariouslevelsandlinkagedisequilibrium(LD)amongclustersweremanuallyinspectedforthesmall-estlevelthatclearlyseparatedallsevenlinkagegroups(LGs).EachLGwasseparatedandmarkerswereimputedonthebasisofnearest-neighboranalysis;onlyperfectlycosegregatingSNPswereusedtoimputeeachother.Redundantmarkerswerethenremovedand100boot-strapsofeachLGweremadebyrandomlyresamplingtheRILs.EachbootstrapwasorderedindependentlyusingMSTmap(Wuetal.,2008)andtheresultsweremerged,keepingthe95%moststablemarkers.Markerposi-tionwasfine-tunedwiththeripple()functioninR/qtl

4 of 13 the plant genome july 2016 vol. 9, no. 2

(Bromanetal.,2003).MapdistanceswerealsoestimatedwithR/qtlusingtheKosambimappingfunction.

Usingthefirstiterationmapasabase,aseconditerationmapwasbuiltbytestingalloriginalSNPs’linkagetooneofthefirstiterationLGs;onlythosewithanR2value0.6weretakenasbeinganchored.TheseSNPswerethenfilteredforthosewithcallsinatleast60RILs,minorallelefrequencies0.25,andheterozygosity0.05.EachLGwasthenbootstrappedandreorderedwithMSTmap(Wuetal.,2008)asabove,thencleanedwithPLUMAGE(Spindeletal.,2014)andrippledwithR/qtl(Bromanetal.,2003).

ThemarkerorderfromthisseconditerationwasusedtoimputemarkergenotypesusingFSFHap(Swartsetal.,2014),whichusesahiddenMarkovmodeltoimputegen-otypesinbi-parentalpopulations.Theimputedgenotypeswereagainbootstrapped,ordered,andcleanedasabove.Togetthefinalmap,weputtheoriginalgenotypesintotheorderidentifiedbytheimputedmap,cleanedthemwithPLUMAGE(Spindeletal.,2014),andestimatedmapdistanceswithR/qtl(Bromanetal.,2003).

Comparison to the Consensus MapMapLGswerenumberedandorientedonthebasisoftheircorrelationtotheconsensusmapofRajarametal.(2013).ThreehundredandfiveSSRprimersequencesfromtheconsensusmapwerealignedtothecontigsusedinSNP-callingusingBowtie2(Langmead&Salzberg,2012).ThepositionoftheSSRwastakenatthecontig’slocationintheconsensusmap.Itscorrespondingloca-tioninthecurrentmapwascalculatedastheconsensuslocationoftheSNPsoriginatingfromeachcontig.Link-agegroupswerenumberedandorientedonthebasisoftheirbestcorrelationtotheconsensusmap.

Anchoring Sequencing TagsSequencetagswereanchoredbasedonthedominant-markermethodofElshireetal.(2011),whereeachtag’sdistributionacrossRILswascomparedtotheSNPsfromthefinalmapusingabinomialtestofsegregation.TheSNPswhosebestp-valuewasbelow0.0001wereconsideredtobeanchored;allotherswerediscarded.Inthisway,333,567tags(outof9.33millionintotal)wereanchoredtothegeneticmap.

Test for Segregation DistortionInarecombinantinbredpopulation,theexpectedseg-regationratioforanygivenmarkershouldbe50%fromeachparent.Eachofthe16,650markerswastestedforsegregationdistortionusinga2testwith1degreeoffreedomat=0.05usingMicrosoftExcel(MicrosoftCorp.,Redmond,WA).Thecriticalχ2valuewasadjustedformultipletestingusingthefalsediscoveryrateproce-dureofBenjaminiandHochberg(1995).

Field Layout for RILsOnehundredandseventy-nineRILs,twoparentallines,andTifGrain102weresowninsingle-rowplotsthatwere

1.5mlongand0.7mapart,witha1-malleybetweenplotsatFortValleyAgriculturalResearchStationfarm(32°31N,83°53W)on16July2013.Theexperimentaldesignwasarandomizedcompleteblockwiththreerep-lications.Grainsorghumwasplantedasaborderaroundtheexperimentplot.

Measurement of Phenotypic TraitsMultiplephenotypictraitswerescoredamongthe179RILsforoneseason;twotraitsarereportedhereastestcasesforthelinkagemapandtheotherswillbereportedinasepa-ratepublication.Thenumberofdaysfromsowingto50%floweringwasrecordedforeachplot.The50%floweringdatewasdecidedwhenatleasthalfoftheplantsineachplothadstartedfloweringandhalfofthepaniclesonindividualplantshadexsertedstigmas.Pyricularialeafspotinfestationhadoccurredundernaturalconditionsbecauseoftherainy,humidweatherduringourexperiment.Tenplantsineachplotwerevisuallyscoredandgivenanaverageratingforthatplot.Thediseasemanifestationwasveryclearandcon-spicuousonallRILsandontheirparents.Thediseasescor-ingwasperICRISATusinga1–9scale(Thakuretal.,2011),where1indicatesnodiseaseand9indicatescompletedeathoftheplantfromdisease.Asdiseaseprogressdependsongrowthstage,someofthelate-maturinglinesshowedadif-ferentdiseaseresponsefromotherlines.Therefore,diseasescoresforplantsthatvarywidelyinmaturationrateswereadjustedbasedontheirmaturityanddiseaseprogresscurve(WilsonandHanna,1992).

Broad-senseheritability(H2)forthesetwotraitswascalculatedusingtheTypeIIImeanssquaresfromPROCGLMinSASversion9.3(SASInstitute,Cary,NC)usingtheformula:

H2=MSG(MSG+MSR+MSE)–1,

whereMSGisthemeansquareforgenotype,MSRisthemeansquareforreplication,andMSEisthemeansquareerror.Thedenominatoristhusthetotalphenotypicvariance.

QTL AnalysisBeforeperformingQTLmapping,therawfloweringtimescoresweretransformedusingBox–Coxtransforma-tionascodedintheMASSpackageforR(VenablesandRipley,2002)becausethedatawerenotnormallydistrib-uted.Theoptimalvalueforwasdeterminedbytestingallvaluesbetween2.0and+2.0instepsof0.008(500stepsintotal);=–1.335hadthehighestlog-likelihoodvalueandsowasusedfortransformation.

MappingofQTLswasthenperformedusingsingle-markerregressionascodedintheR/qtlpackageforR(Bromanetal.,2003).ThephenotypesusedweretherawdiseasescoresandBox–Coxtransformedfloweringtimescoresandthegenotypeswerethefinallinkagemap.Wealsosmoothedthelogarithmofodds(LOD)scoresin5-cMslidingwindows,takingthemaximumLODwithineachwindowtoidentifypeaksofassociationmoreclearly.

punnuri et al.: development of a high-density linkage map in pearl millet 5 of 13

Results

Genotyping-by-Sequencing Analysis and SNP CallingOnehundredandeighty-fourRILsweresequenced,whichgeneratedatotalof438.6millionreadsthat,withtheexceptionoffivefailedsamples,arespreadmostlyevenlyacrossthesamples(SupplementalFig.S1).Asthetwoparentallinesandtheircommercialhybridweresequencedtwice,wehadhigh-depthcoverageof5,964,312readsforTift99D2B1,3,077,835readsforTift454,and4,704,803readsforTifGrain102.Themeanreaddepthacrossallsuccessfulsampleswas2.2±0.95million(CV=0.43),thepassfilterratewas88%,andthemedianwas2.16millionreadspersample.Thetotalnumberofgoodreadsamong179RILswas387,339,046;theindividualwiththefewestreadshad390,047andtheindividualwiththehighestreadshad4,854,147.RawreadswerethenconvertedtoSNPcallsusingtheTASSEL-GBSpipeline(Glaubitzetal.,2014;seetheMethodssectionfortheparametersused).Sincepearlmilletdoesnotyethaveapublishedreferencegenome,wealignedthereadsagainstacollectionof~19,000scaffoldsandcontigskindlypro-videdbythePearlMilletGenomeSequencingConsor-tium(Varshneyetal.,unpublisheddata2015).DuringSNPcalling,88.8%ofthetotalreadsweremappedtoscaf-foldsandcontigsfromthepearlmilletgenomicsequence.

Relationship between Founder Lines and the RILsThesequencingdatafromtheRILsshowsacloserela-tionshiptotheparentallinesusedinthisstudy(Supple-mentalFig.S2).TheRILsclusteraroundthetheoreticalvalueof50%relatednesstoeachparent(0.5,0.5).Asexpected,afewindividualsshowupto80%orhigherrelatednesstooneparentortheother,asaresultofsto-chasticgametesamplingduringmeiosis.

Identification of Polymorphic MarkersCallingofSNPsresultedin>500,000rawSNPs,manyofwhichwerefalsepositivescausedbysequencingerrors.FilteringforSNPswithcallsinatleast60%oflinesandwithminorallelefrequenciesabove0.25resultedin~24,000high-qualitypolymorphicSNPs.TofilteroutfalseSNPsfromparalogoussequencesaligningtogether,wealsoremovedsitesthatshowed>12%heterozygosity(the12%cutoffwasdeterminedempiricallybylookingatthedistributionofheterozygoussites).WethenalsoremovedanyRILswith>50%missingdataor>10%het-erozygosity.Thisresultedinusing17,400sitesacross146RILindividuals,wheremissingdatawasintherangeof0.5to43.4%perindividual(median8.7%)and12.7%missingacrosstheentiredataset.

Construction of the Genetic MapWebuiltthelinkagemapinaseriesofiterativesteps.Firstasubsetofveryhigh-quality“core”SNPswastakenandusedtodefineLGsandaninitialordering.Weobtainedacoresetof1192uniquemarkerswith

stringentfiltering(seeMaterialsandMethods)coveringsevenLGs.Oncethecoremapwasassembled,lower-qualitySNPswereanchoredtoLGsandtheorderingwasrepeated.ThissecondorderingwasthenusedtoimputeallthemarkersusingFSFHap(Swartsetal.,2014),whichusesahiddenMarkovmodeltoimputeindividualsinbiparentalpopulations.Theseiterativestepsaddedanother15,458markerstothemapacross150RILs.AheatmapofLDshowsclearclustersbetweensevendif-ferentLGscorrespondingtothesevenpearlmilletchro-mosomes(SupplementalFig.S3).Thesethreeiterativestepsresultedinthefinalgeneticmapof16,650SNPsin1191uniquerecombinationbins(Fig.1).OurLGswerethenrenumberedandreorientedtomatchthoseofRaja-rametal.(2013),whichwerebasedonmappingtotheampliconsusedtogeneratetheirmap.Forcomparison,wealsocreatedamapwithoutusingthegenomiccontigstoanchorthesequencingreads.Instead,GBSreadswerealignedagainsteachotherwiththeUNEAKfilter(Luetal.,2013;seeMaterialsandMethods);allotherstepswereidentical.Thefinalgenome-freemapincluded4900markers.Thisisstillasignificantnumberofmarkers,andifnogenomicdatawereavailable,theywouldstillformausefulmap.However,the>3×highernumberofmarkersfromtheoriginalmapdemonstratethevalueofhavinggenomicsequencestoalignagainst,evenifthesesequencesarenotassembledintoareferencegenome.

Expanding the Genetic Map with Sequencing TagsAfterobtainingthefinalmap,wethenanchoredsequencingtags(thesame64-bpreadsusedintheGBSpipeline)toit.Weusedthedominant-markermethodofElshireetal.(2011),whichanchored333,567(outof9.33million)tagsontothegeneticmap.

Togaugetheaccuracyofmapping,welookedattheoverlapbetweensequencingreadsandtheSNPstheygenerated.ThereisonlypartialoverlapbetweenthesetoftagsthatgiverisetoSNPsinthemapandthetagsthatweremappedontheirown(Fig.2).Thisismostlybecause(i)atagcanstillbeanchoredevenifanySNPsitgivesrisetoarefilteredout,(ii)sometagsarecausedbypresence–absencevariationandsowillnotgiverisetoSNPsthemselvesbutcanstillbeanchoredtonearbySNPs,and(iii)sequencingerrorscanmakeatagappearunique,soevenifagoodSNPcanbecalledinonepartofatag,anerrorelsewhereinitmakesthetagtoorareforthebinomialsegregationtesttowork.

Ofthetagsthatdooverlap,~87%ofthemanchortowithin10cMoftheirassociatedSNP,manyofthemtotheexactsamerecombinationbin.Thisimpliesthatourmappingaccuracyishighandthepositionsofthereadsshouldbeveryclosetotheirtrueposition.

Intotal,16,650SNPsand333,567additionaltagsweredistributedonallsevenLGs(Table1).Overall,20.10%ofdataweremissingfor16,650lociacross150individuals.Themissingvaluesacrosstheselocirangedfrom4to55perlocus(2.6–36.6%).Inthefinalmap

6 of 13 the plant genome july 2016 vol. 9, no. 2

containing16,650SNPSand333,567tags,theaveragedensitiesofSNPmarkersandofadditionaltagsacrossallchromosomeswere23.23and465.42tagspercM,respectively,coveringthegenomelengthof716.7cM.ThemarkerdensitiesperLGwerespreadfromaminimumof9.24cM–1onLG4toamaximumof35.13cM–1onLG7.Whenonlyuniquelinkagebinswerecounted,markerdensitiesareintherangeof0.81binscM–1(LG4)to1.90binscM–1(LG2)withanaveragedensityof1.66binscM–1acrossthegenome.

Amongallthechromosomes,LG4hadthefewestmarkersandtags,whereasLG5hadthehighestnumberof

SNPsandLG2hadthehighestnumberoftags.ThehighestnumbersofSNPmarkerswereanchoredandorderedonLG5,whichhad3085markersinthefinalmap.

Comparison to an Existing Pearl Millet Consensus MapRajarametal.(2013)recentlyproducedaconsensuspearlmilletmapbycombiningSSRdatafromfourdifferentlinkagepopulations.Thecurrentmapwasmatchedtotheconsensususing305SSRprimerpairsfromRajarametal.(2013).Ofthese,191aligneduniquelywhilebeinginthecorrectrelativeorientationsanddistancesapart;16alignedconcordantlybutatmultiplelocations,onealigneddiscordantly(incorrectorientation),and97eitherdidnotalignatallorhadonlypartialalignments(meaningoneprimerwasalignedbutnotboth)(SupplementalFig.S4).

Fig. 1. Linkage map of pearl millet developed using genotyping-by-sequencing (GBS) markers. Gray bars represent each linkage group, with black bands showing the unique map locations on each linkage group. (Linkage groups were extended past the first and last mark-ers for visual clarity.) Blue bars to the left of each group are proportional to the number of single-nucleotide polymorphisms (SNPs) at each location; red bars to the right show the number of sequencing tags mapped to each location.

Table 1. Linkage group (LG) statistics with marker and tag density in pearl millet.

LG Length (cM) MarkersAnchored

tagsMarker density

per cMTag density

per cM

LG 1 96.9 2509 49,855 25.89 514.50LG 2 98.1 2986 62,754 30.44 639.69LG 3 175.3 3000 61,367 17.11 350.07LG 4 55.5 513 15,789 9.24 284.49LG 5 118.3 3085 58,902 26.08 497.90LG 6 112.6 2449 46,685 21.75 414.61LG 7 60.0 2108 38,215 35.13 636.92Total 716.7 16650 333,567 23.23 465.42

Fig. 2. Single-nucleotide polymorphism–tag concordance in pearl millet. The overlap between sequencing reads and the SNPs they generated is shown. Of the tags that do overlap, ~87% of them anchor to within 10 cM of their associated SNP, many of them to the exact same recombination bin.

punnuri et al.: development of a high-density linkage map in pearl millet 7 of 13

Thelengthsofeachchromosomeinthecurrentmaprangedfrom55.5cM(LG4)to175.3cM(LG3),withanaveragelengthof102.3cMperchromosome.Inthecoremapmadefrom1192sites,theaverageintermarkerdistancesbetweentwoadjacentmarkersrangedfrom0.52cM(LG2)to1.23cM(LG4).Theinter-markerdistanceof0.01cMwasleastonLG2andLG3,andthemaximumdistanceof11.71cMwasobservedonLG4,withanoverallaveragemarkerdistanceof0.67cMacrosstheentirecoregeneticmap.Therewerethreeintervals[5.52cM(LG4),6.14cM(LG2),and11.71cM(LG4)]thatweremorethan5cMbetweenneighboringmarkers.Therestoftheintervalswerebelow5cMdistances,whichreflectsthatmorethan99%ofthemaphadsmallspacingsbetweenneighboringmarkers.LinkageGroup3hereappearedtobeextendedlongerthanLG3oftheconsensusmap,whereasLG7fairlyrepresenteditscounterpart.Therestofthechromosomeswereshorterthantheconsensusmap.WealsocomparedourLGlengthswithfourLGs(LGA,LGB,LGC,andLGG)intheGBS-basedSNPmapbyMoumounietal.(2015),whichrevealedthatourmapwasextendedinLG1andLG6,butitwasshorterinLG2,LG4,andLG7.Theseextensionsareverycommonintelomericregions,whichalsohavebeenobservedinDArT-basedmapsofpearlmillet(Supriyaetal.,2011).ThemapsreportedinallthepreviousstudiesusedHaldanemappingfunctions,whereasourmapusedtheKosambimappingfunctiondistances,whichcouldbeonereasonfordiscrepanciesinmaplengths.

Thecurrentmapappearstohaveroughlyequalcoveragetotheconsensusmapbutwithsomecaveats.Manyindividualmarkersandsomegroupsofmarkerswerelocalizedtodifferentlocationsinthetwomaps.Someofthismaybearesultoftechnicalerror,suchasmisalignmentoftheprimersequencesormisassemblycausedbysequencingerrors.Someofthediscrepanciesareprobablybiological,however,andrepresentsmall-andlarge-scalestructuralvariationsbetweenthepopulationsusedtomakethetwomaps.Pearlmillethassignificantgeneticdiversity(Oumaretal.,2008),tothepointthatonlyasingleSSRfromtheconsensusmapwasmappableinallfourofitsinputpopulations(Rajarametal.,2013).Inthatcontext,findingsignificantvariationwithafifthpopulation(theoneusedinthisstudy)istobeexpectedhereaswell.

Segregation DistortionOfthe16,650mappedSNPmarkers,6652(39.41%)showedsignificantsegregationdistortionafteradjust-mentformultiplecomparisons(BenjaminiandHoch-berg,1995).Mostofthesedistortedmarkersoccurredinlargelinkageblocks.LinkageGroup3showedthegreatestamountofsegregationdistortion,withnearlytheentireLG(98.67%ofmappedmarkers)significantlybiasedinfavorofTift99D2B1.Incontrast,LG1wasalsohighlydistorted(80.71%ofmappedmarkers)butwasbiasedinfavoroftheotherparent,Tift454.Linkage

Group2alsohadseveralhighlydistortedblocks,biasedtowardtheTift99D2B1parent,andLG6hadonemajorlinkageblockbiasedtowardTift99D2B1.LinkageGroup4showedtheleastsegregationdistortion,withonlytwomarkers(0.39%)distorted(SupplementalFig.S5).

Mapping Leaf Spot Resistance and Days to 50% Flowering TraitsThelinkagemapdevelopedinthisexperimentwasusedinregressionanalysistoidentifyQTLsfortwopheno-typictraits:leafspotresistanceanddaysto50%flower-ing.TheH2wasquitehighforthesetwotraits.Fordaystoflowering,H2=0.7578.ForBox–Coxtransformeddaystoflowering,H2dropsto0.5110.Forrawdiseasescore,H2=0.7978;fortheadjusteddiseasescore,itis0.9163.

Thetwoparentsshowedsignificantdifferencesforthesetwotraitsinthefield,whereastheirF1hybrid,TifGrain102,showedgoodleafspotresistance,similartoTift99D2B1,butfloweredlater,similartoTift454(SupplementalTableS1).R/qtlresultsidentifiedleafspotresistancelocionLG5andLG7withsignificantthresholdLODvaluesabove3.0(Fig.3,Table2).TheseQTLswerefoundtobeminor,withphenotypicvarianceof4.83to5.05%andafavorablealleliceffect(lowerdiseasescore)fromTift454.TwomoreQTLsforleafspotresistancewithafavorablealleliceffectfromTift99D2B1werelocatedonLG2andLG3,havingLODvaluesjustabove2.0.AsignificantQTLforfloweringtimewithaLODvalueabove3.0waslocatedontheupperarmofLG2,whichexplained6.0%ofthephenotypicvariance,withthepositivealleliceffect(laterflowering)comingfromtheparentTift454(Fig.3,Table2).TherestoftheQTLsforfloweringtimeweredetectedbelowLOD3.0onLG1,LG5,andLG7with0.49to4.75%phenotypicvarianceandpositiveadditiveeffectscomingfromtheotherparent,Tift99D2B1.

Discussion

Importance of a High-Density Genetic Map and Its Comparison to Existing MapsNext-generationsequencingtechnologieshaverevolution-izedmarkerdiscoveryandenabledhigh-throughputplantgenotypingthroughseveralnewmarkerplatformslikeGBS(PolandandRife,2012).Genotyping-by-sequencingisacost-effectiveandefficientsystemfordevelopinghigh-densitymarkers,whichareconcurrentlydiscoveredandgenotypedinlargermappingpopulations(Heetal.,2014).Theseabundantmarkers,coupledwithwell-developedbioinformatics,facilitatethedevelopmentofdensemolec-ularlinkagemaps.Inthisexperiment,wehadhigh-depthcoverageandabundanthigh-qualitySNPs.

EversincethefirstpearlmilletgeneticmapwasmadefromRFLPsin1994(Liuetal.,1994),therehasbeenacontinuousefforttoimprovesuchmapswithgreatermarkerdensityanduniformity.Manyofthesemapshadlargegapsinthedistalregionsofchromo-somes,probablycausedbyveryhighrecombinationrates,somostimprovementeffortstargetedthese

8 of 13 the plant genome july 2016 vol. 9, no. 2

regions.Forexample,expressedsequencetagandgenomicSSRswereaddedbySenthilveletal.(2008),DArTmarkersbySupriyaetal.(2011),andgene-basedSNPandconservedintronspanningprimersmarkersbySehgaletal.(2012).Despitetheseefforts,largegapsofmorethan30cMwerestillpresentinmostofthedistalregionsofchromosomes.Themostrecentconsensusmap(Rajarametal.,2013)usedexpressedsequencetagSSRsandalsocontainedlargegapsintherangeof18to27cMoneverychromosome.UsingNGS,Moumounietal.(2015)madeaGBSmapfrom314nonredundantSNPs.AlthoughthemapdevelopedbyMoumounietal.(2015)wasuniformincoveragewithnointervalgreaterthan20cMinlengthandonly10intervalslargerthan10cM,itstillhadamaximumgapof19.7cMonLG2thatcorre-spondsto3.0%ofthetotalmaplength.Thelinkagemapinthecurrentstudyhasamaximumgapof11.71cMonLG4,equatingto1.6%oftotalmaplengthandrepresent-ingasignificantimprovementinreducedgapsize.

Toourknowledge,thismaprepresentsthedensestgeneticmapinpearlmilletsofar.Itcontains16,650SNPsand333,567sequencetagscoveringallsevenLGs.Here,wereportanaveragedensityof1.66linkagebinscM–1and23.23SNPcM–1inthefinalmap,whichsignificantlysurpassesthe0.51SNPcM–1ofthenext-densestmap(Moumounietal.,2015).Thelinkagemapconstructedinthisstudyismoredense,uniform,andhighlysaturated,whichisreflectedthroughsmallermarkerspacing(<5cM)thananypreviouslypublishedpearlmilletgeneticmap.Themeandistancebetweentwoneighboringmarkersistheleast:0.6cMcomparedto2.1cM(Moumounietal.,2015)and3.7cM(Supriyaetal.,2011)publishedsofar.ThesmallmarkerspacingsoneverychromosomewithseveralcosegregatingredundantmarkersshowsthatwiththeexceptionofLG4,thismapisextensiveandreasonablyuniformingenomecoverage.Therefore,ourmapcomplementstherecentpearlmilletlinkagemapdevelopedbyMoumounietal.(2015),whichcontains2809GBSmarkersfrom85F2progenies.At716.7cM

Fig. 3. Quantitative trait locus (QTL) mapping in pearl millet. Quantitative trait loci were identified for Pyricularia leaf spot (top) and flowering time (bottom). The distribution of phenotype scores is shown on the left and logarithm of odds (LOD) values from single-marker regression (using R/qtl; Broman et al., 2003) are shown on the right. The light gray traces show the raw LOD scores, which vary depending on the different levels of missing data for each marker. The solid black line shows a smoothed trace, taking the maximum value in 5-cM sliding windows.

punnuri et al.: development of a high-density linkage map in pearl millet 9 of 13

intotallength,ourmapisslightlylongerthanthatofMoumounietal.(2015)(640.6cM),whichusedanF2populationandthusisexpectedtobeshorter.Thehighqualityandquantityofmarkersfoundinthisexperimentwerepossiblebecauseofhigh-depthcoveragefortwoparentsincallingSNPsandthelargenumberofRILs(150individualprogenies)availableafterstringentfiltering.

Geneticmapdistancesarerelativedistancesbasedonrecombinationfrequencies,unlikephysicalmaps,whichestimateactualdistancesinbasepairs.Themapdistancesandpositionsofindividualmarkerscanvaryfromonemappingpopulationtotheotherdependingontheparentsusedintheinitialcrossandtypeofmappingpopulationused.OurmapdistancesarerepresentedthroughtheKosambimappingfunctionalthoughpreviousstudiesusedHaldanemappingfunction,whichmayexplainsomeofthedifferencesinmaplength.Thecomparisonbetweenourmapandthepreviousconsensusmaphasshownsomeagreementbutalsosomediscrepancies.Forexample,somemarkersareatdifferentlocationsinthetwomaps(SupplementalFig.S4).OurtotalmaplengthisshorterthanthetotalmaplengthsreportedbySupriyaetal.(2011),Sehgaletal.(2012),andRajarametal.(2013).Althoughsomeofthesedisagreementsareprobablycausedbytechnicaldifferencesinthewayseachmapwasprepared,manyofthedisagreementsareprobablyaresultofbiologicaldifferences,includingafewlargelinkageblocksthatmayrepresentactualtranslocationsinonepopulationrelativetotheother.GiventhequalityofLDwithinthecurrentmap(SupplementalFig.S3),anymajordiscrepancies

areprobablycausedbystructuralvariationsoriginatingfromthegermplasmusedinthecurrentstudy.

High-densitymapsdevelopedthroughGBSnotonlysupportfunctionalgenomicsthroughconnectingphenotypetogenotypebuttheyalsoserveasreferencemapsinfundamentalstudieslikegenomesequencingtorefine,order,andassemblescaffoldsandcontigsofpseudochromosomes(PolandandRife,2012;Wardetal.,2013).ThismaphasbeenpartlyusedincontigassemblyofthepearlmilletgenomesequencingprojectledbyICRISAT.Furthermore,awell-ordereddensemapallowsacomparativegenomestructureanalysisandinformsaboutimportantevolutionarychanges(GaleandDevos,1998).Thislinkagemapwillalsohelpotherresearchersworkingonmappingtraitsinpearlmillet.Forexample,otherscandirectlyusethe64-bptagsusedtodevelopSNPsinthisstudyforthesamepurpose.Theresultingdatasetscanbeusedtomakegeneticmaps,minealleles,andcharacterizediversepearlmilletaccessions.

Imputation of SNP DataThemajordrawbackofsequencing-basedgenotypingtechnologyisthelargeamountofmissingdata;GBSisnoexception.Severalapproachescanbeusedtoreducethesemissingdata,suchassequencingtohighdepth,filteringtosaveonlyhigh-qualitydata,orperformingimputationofhaplotypes(PolandandRife,2012).Weusedcarefulfilteringtoachieveamissingrateof20.1%inourfinal(unimputed)geneticmap,althoughoneofthestepsusedtogenerateitincludedimputinganotherversiondowntoonly~3%missingdata.Wefocusouranalysesontheunimputedmapbecauseimputationcanintroducebiases.BoththeimputedmapandunimputedmapareavailableinSupplementalFileS2.

The Parents and Their AncestryTheparentsofthismappingpopulation,Tift99D2B1andTift454,aredwarf,early-maturinggraintypes.Bothpar-entscarrytherecessivedwarfinggened2,whichliesonLG4(Parvathanenietal.,2013).WediscoveredveryfewmarkersonLG4comparedtootherLGs.SincethetwoparentsinheritedgenomicregionsfromTift23D2B1,itispossiblethatthisLGhasfewSNPsbecauseofaregionofcommondescentaroundthedwarfinggened2.Themale-sterileA-lineTift99D2A1andTift454aretheparentsofthecommercialhybridknownasTifGrain102(Hannaetal.,2005a,2005b).Tift99D2B1wasselectedforresistancetorustandisderivedfromTift89D2andalsosharessomegenomicregionswithTift23D2(HannaandWells,1993;Hannaetal.,2005b).ItalsoappearstohaveresistancetoPyricularialeafspot.Tift454wasderivedfromaninterspecificcrossbetweenpearlmilletTift23D2A1andanapiergrass[Cenchrus purpureus(Schum.)Morrone]–pearlmillethybridandcarriesatleastoneAchromosomefromthenapiergrassparent(Hannaetal.,2005a).Tift454isresistanttonematodes[Meloidogyne areniaria(Neal)ChitwoodandMeloidogyne incognitaKofoid&White]andhasmale-fertilityrestorercapabilityinA1cytoplasm.

Table 2. Quantitative trait loci for flowering time and Pyricularia leaf spot disease identified in a pearl millet recombinant inbred line population.

Flowering time

LG§ Location SNP interval Peak SNP LOD VarianceAdditive effect†

cM % d1 32.3 S1_1423–S1_3590 S1_2196 2.61 3.03 1.82 23.3 S2_1896–S2_2803 S2_2223 4.86 6.00 2.05 0.0 S5_0012–S5_1669 S5_0451 2.38 4.75 1.57 14.4 S7_0244–S7_2067 S7_0774 2.48 0.49 1.3

Leaf spot disease

LG Location SNP interval Peak SNP LOD Variance Effect‡

cM %2 85.0 S2_7773–S2_8331 S2_7983 2.18 1.78 0.63 114.2 S3_0019–S3_4763 S3_4544 2.25 1.82 0.55 30.5 S5_2145–S5_4145 S5_3817 4.56 4.83 0.97 30.5 S7_0738–S7_3864 S7_2251 3.01 5.05 0.9

† A negative sign indicates that the later flowering allele was derived from the Tift 454 parent, whereas a positive sign indicates that the allele from parent Tift 99D2B1 delayed flowering.

‡ 1 indicates no disease symptoms; 9 indicates complete susceptibility. A negative sign indicates that the Tift 99D2B1 allele increased resistance (lower score), whereas a positive sign indicates that the Tift 454 allele increased resistance.

§ LG, linkage group; SNP, single-nucleotide polymorphism; LOD, logarithm of odds.

10 of 13 the plant genome july 2016 vol. 9, no. 2

Regionsofsignificantsegregationdistortionhavebeenreportedinpreviousgeneticmappingstudiesinpearlmillet(Qietal.,2004;Rajarametal.,2013;Moumounietal.,2015),soitisnotsurprisingthattheyweredetectedinthispopulationaswell.However,wefoundtworegionsofsegregationdistortioninthispopulationthateachspansnearlyanentireLG(LG1andLG3)(SupplementalFig.S5).Suchlargeregionsofsegregationdistortionhavenotbeenreportedinpreviousstudiesinpearlmillet.LinkageGroup1and3alsohadthehighestnumberofdiscrepanciesincomparisontothemapofRajarametal.(2013)(SupplementalFig.S4).AccordingtoHannaetal.(2005a),theparentallineTift454(2n=2x=14)carriesatleastonepairofchromosomesfromtheAgenomeofnapiergrassinplaceofahomologouschromosomepairfromtheAgenomeofpearlmillet.Theevidencehere,namelynearlycompletesegregationdistortionoftwoentireLGsalongwithalargenumberofmapdiscrepancies,suggeststhatTift454mayinfactcarrytwonapiergrasschromosomes.LinkageGroups1and3appeartorepresenttwoA–Achromosomepairs.ThoughtheAandAgenomesarereportedtobehomologous(Hanna,1990),itispossiblethattherateofrecombinationbetweenthenapiergrassandpearlmilletchromosomesislowerthantherateofrecombinationbetweenchromosomesoriginatingfromthesamespecies.EvidencereportedbyTechioetal.(2006)suggeststhattheAandAchromosomesarelikelytobehomeologousratherthanhomologous.Inaddition,meioticirregularitieshavealsobeenreportedintriploid(Techioetal.,2006)andhexaploid(Paivaetal.,2012)pearlmillet–napiergrasshybrids.Interestingly,mostofLG1isbiasedinfavoroftheTift454parent,suggestingthattheAchromosometransmitsmorefrequently,whereasLG3isbiasedinfavorofTift99D2B1,suggestingreducedfrequencyoftransmittingthisAchromosome.ThoughtheRILswereselectedrandomly,thebiastowardoneparentortheothermayalsobeanartifactofunintentionalselectionbasedoncharacteristicssuchaspollenviabilityorseedsetundertheselfingbag.

Utility of the Map in Tagging Disease Resistance Loci and Flowering TraitsThehigh-densityGBS-basedlinkagemapwasvalidatedbymappingQTLsforfloweringtimeandPyricularialeafspotresistance.TheleafspotresistancelociidentifiedinthisstudyindicatethatthistraitiscontrolledbyseverallocifromdifferentLGs.Inapreviousstudy,arandomamplifiedpolymorphicDNAmarkerwasidentifiedasbeingassociatedwithPyricularialeafspotresistancebutwasnotassignedtoanyLG(Morganetal.,1998).ResearchfromICRISAT,India,hasmappedaleafspotresistanceQTLtoLG4inaRILpopulationbasedon‘ICMB841-P3’ב863B-P2’(Dr.RKSrivastava,personalcommunication,2015),whichwasalsoassociatedwithstoverqualitytraitsandwasintrogressedintothehybridseedparent‘ICMA/B95222’(Nepoleanetal.,2006).ICMA/B95222istheseedparentofhybrid‘HHB146’releasedfromChaudhary

CharanSinghHaryanaAgriculturalUniversity,Hisar(Dwivedietal.,2012).ThepresentstudyalsoidentifiedasignificantfloweringtimeQTLonLG2,thesameLGwherethePHYCgenewassignificantlyassociatedwithfloweringtime(Saïdouetal.,2009)andseveralotherflow-eringanddroughttoleranceQTLswerereported(Yadavetal.,2002,2004,2011b;Bidingeretal.,2007;Sehgaletal.,2012).Primersequencesfromthesestudieswereusedtocomparetheirlocationonourmap(SupplementalTableS2).Basedontheirmarkerpositioninourmap,ourfloweringtimeQTLlocationsdonotcorrespondtothelocationsreportedinthesepreviousstudies.However,itwillbeinterestingtoexplorethepotentialcandidategenesoncethecompletepearlmilletgenomesequenceisavailable.TheamountofphenotypicvariationexplainedbytheseQTLswaslowforthesetwotraitsdespitethefactthatH2wasquitehigh[H2=0.511fordaystoflower(transformed)andH2=0.916foradjusteddiseasescore]andwerecomparabletootherstudies(Yadavetal.,2002,2004;Nepoleanetal.,2006;Dwivedietal.,2012;Sehgaletal.,2015).Theheritabilitiesforfloweringtraitwerereportedtobeintherangeof47to94%inthepreviousstudies(Yadavetal.,2004;Sathyaetal.,2014;Sehgaletal.,2015).OneexplanationisthatnumerousQTLs,eachwithaverysmalleffect,contributetothesetraits(Yadavetal.,2003).Additionally,theQTLdetectionmethodusedhere(single-markerregressioninR/qtlsoftware)mayunderes-timateindividualQTLeffects(LanderandBotstein,1989;Zeng,1994).TherelativelackofmarkersonLG4[becauseoftheapparentdescentofmuchofLG4inbothparentsoftheRILpopulationsfromacommonancestor(Pyricularialeafspot-susceptibleTift23D2B1)],wherealeafspotresis-tanceQTLwaspreviouslyidentified,couldalsoexplainwhywedidnotidentifythisQTL.Whenthesetraitsweremappedusingageneticmapmadewithoutgenomicsequences,manyoftheQTLswerestillidentifiablebutappearedtohavelostsomesignificance,probablybecausetheylackedtheSNPsthatwereintightestlinkagewiththecausallocus(SupplementalFig.S6).ThisalsoreflectsthathavinggenomesequenceinformationwillenhanceQTLmapping.TheQTLresultsreportedherearebasedonasingleseasonofdata,sotheywillneedtobevalidatedbyadditionalstudiesinmoreenvironments.Evenso,theexamplespresentedheredemonstratetheutilityofthisgeneticmapforidentifyingQTLs.

ThisstudyusedaRILpopulation,whichallowedforareplicatedfieldscreenfordiseaseresponseandfloweringtime.Suchreplicationincreasestheaccuracyofphenotyping,despitehavingonlyoneseasonofdata,andisnotpossiblewithF2populations.Additionally,seedsoftheRILpopulationcanbedistributedtootherresearcherstomapothertraitsofinterestwithouttheneedtoreconstructthegeneticmap.

ConclusionsPearlmilletisconsideredaminorcropintheUnitedStatesandEurope,sodevelopmentofgeneticandgenomicresourcesinthiscrophaslaggedbehindother

punnuri et al.: development of a high-density linkage map in pearl millet 11 of 13

cereals.Itis,however,anessentialstaplecropinmanypartsoftheworld,particularlydevelopingcountriesinhotsemiaridandaridregionswherelittleelsewillgrow.Thusimprovementofthiscropiscriticallyimportantforfoodsecurityintheseareasandmaybecomecriticaltocurrentlymorefavorableareasifglobalclimatechangecontinuesunabated.ToolslikemolecularmarkerscanfacilitaterapidadvancesincropimprovementbutthedevelopmentofsuchresourceswasaformidabletaskinpearlmilletuntiltheadventofNGS-basedmarkerslikeGBS.Inthisexperiment,GBSmarkersweresuccessfullyusedtomakeahigh-densitymapcontaining16,650SNPsand333,567additionalsequencetags,whichisthedens-estmapyetcreatedinpearlmillet.High-densitylink-agemapsprovidebettermapresolutionandabundantgenomicresources.Arecombinantinbredmappingpopu-lationcreatedfromanelitegermplasmwasusedtocon-structthismapsothatusefulandrepeatablevariationcanbestudiedusingthisresource.Thesegenome-widemark-erscanbeusedforapplicationssuchasmarker-assistedselection,genomicselection,diversitystudies,andcom-parativegenomicanalyses.Theresultswillalsohelptoidentifyandtagseveraltraitsrelatedtodiseaseandnema-toderesistanceinpearlmillet.Inaddition,understandingthegenesunderlyingimportanttraitsinpearlmillet,suchasdroughttoleranceandnitrogenuseefficiency,couldhelptoimprovethesetraitsinothercrops.

Supplemental Information AvailableSupplementalmaterialisavailablewiththisarticle.

Supplemental Table S1:Leafspotscoresanddaysto50%floweringforparentallines,theirF1hybrid(Tif-Grain102),andtheRILpopulation.

Supplemental Table S2:Markerpositionsonthecurrentmapbasedonbasiclocalalignmentsearchtool(BLAST)hits.

Supplemental Figure S1: Read depth per sample.Thenumberofsequencingreadsmatchedtoeachindi-vidualisshowninorderofincreasingreaddepth.GraybarsrepresentRILsthatweresequencedonceeach;blackbarsarethetwoparents(Tift99D2B1andTift454)andtheirF1hybrid(TifGrain102),whichweresequencedtwice(onceoneachplate).Fivesampleswereremovedbecausetheyhad<5000mappedreadseach.

Supplemental Figure S2: Relatedness of RILs to parents.RILs(palebluecircles)areplottedaccordingtotheirdegreeofrelatednessrelativetobothparents.Darkercolorsindicatewherepointshavestackedontopofeachother.

Supplemental Figure S3: Linkage disequilibrium heatmap.Linkagedisequilibrium(r2)heatmapshownacrossthefinalgeneticmapforallpairwiseSNPcom-parisons.Single-nucleotidepolymorphismsarearrayedinmaporderonboththexandyaxesandeachpointshowsthepairwiselinkagedisequilibriumbetweenasetofSNPs.ThesizeofeachblockisproportionaltothenumberofSNPsineachLG;thesmallnumberofSNPsin

LG4isprobablycausedbyalargechromosomalsegmentthatisidenticalinbothparentsthatislikelytohavebeeninheritedfromtheircommonancestor,Tift23D2B1.

Supplemental Figure S4: Comparison to existing pearl millet consensus map.SimplesequencerepeatprimersequencesfromanexistingSSRconsensuspearlmilletmap(Rajarametal.,2013)werealignedagainstthecontigsusedtocallSNPsinthecurrentmap.Thelinkagemapfromthisstudy(left-handside,darkgray)iscomparedwiththeSSRconsensusmap(right-handside,lightgray).Blackbarsindicatemarkersthatcouldbeidentifiedinbothmaps,withcoloredlinesconnect-ingeachmarkerpositiontoitscorrespondingpositionintheothermap.Solidlinesindicatemarkersthatmaptomatchinglinkagegroups(LGs);dashedlinesindicatemarkersthatmaptodifferentLGs;andlinecolorindi-catestheLGinthecurrentSNP-basedmap.Althoughmanymarkersshowgoodcorrelation,manyalsoshowinconsistentordering.Largeblocksofinconsistentmark-ersmayrepresentlargetranslocations,suchasbetweentheconsensusLG1andourLG4andbetweenthecon-sensusLG6andourLG1.

Supplemental Figure S5: Map of segregation distor-tion in the pearl millet RIL population.MarkersshadedinredarebiasedinfavorofTift99D2B1;markersshadedbluearebiasedinfavorofTift454.Markerswithaχ2valuegreaterthanthecriticalvaluearesignificantlydistorted.

Supplemental Figure S6: Effect of genomic sequence on mapping quality.Quantitativetraitlocusmapsforfloweringtimeandleafspotdiseasecomparedbetweenthefulllinkagemapandthemapmadewithoutaligningsequencestothepearlmilletgenomicdata.

Supplemental File S1 (Textfiles):Allscriptsandparametersusedinthecurrentexperiment.

Supplemental File S2(Excelfiles):Genotypicdatafor16,550lociusedforfinalmapcreationandphenotypicdataforleafspotdiseaseandfloweringtraitsin179RILs.

AcknowledgmentsWethankEliRodgers-MelnickforpartoftheRcodeforripplingLGs,thePearlMilletGenomeSequencingConsortiumforuseofprepublica-tioncontigsandtheGenomicDiversityFacility(CornellUniversity)forhelpfuladviceonGBSanalysis.WethankMs.ChrisdonB.Bonnerforhelpingustoimprovethequalityofthemanuscript.WearealsogratefulforfundingsupportreceivedfromthecapacitybuildingprojectbyUSDA-NationalInstituteofFoodandAgricultureGrant#GEOX-2008-02595,NSFGrantIOS-1238014,theUniversityofGeorgia,andtheUSDA-ARS.Theauthorsdeclarethattheyhavenocompetinginterestsrelatedtothecontentsofthismanuscript.Mentionoftradenamesorcommercialprod-uctsinthisarticleissolelyforthepurposeofprovidingspecificinforma-tionanddoesnotimplyrecommendationorendorsementbytheU.S.DepartmentofAgriculture.

ReferencesAndrews,D.J.,andK.A.Kumar.1992.Pearlmilletforfood,feed,andfor-

age.Adv.Agron.48:89–139.doi:10.1016/S0065-2113(08)60936-0Andrews,D.J.,J.F.Rajewski,andK.A.Kumar.1993.Pearlmillet:Newfeed

graincrop.In:J.JanickandJ.E.Simon,editors,Newcrops.JohnWiley&Sons,NewYork.p.198–208.

Benjamini,Y.,andY.Hochberg.1995.Controllingthefalsediscoveryrate:Apracticalandpowerfulapproachtomultipletesting.J.R.Stat.Soc.,B57(1):289–300.

12 of 13 the plant genome july 2016 vol. 9, no. 2

Bennett,M.D.,andJ.B.Smith.1976.NuclearDNAamountsinangio-sperms.Phil.Trans.Roy.Soc.Lond.Ser.B.274:227–274.doi:10.1098/rstb.1976.0044

Bidinger,F.R.,andC.T.Hash.2004.Pearlmillet.In:H.T.NguyenandA.Blum,editors,Physiologyandbiotechnologyintegrationforplantbreeding.MarcelDekker,NewYork.p.221–233.

Bidinger,F.R.,T.Nepolean,C.T.Hash,R.S.Yadav,andC.J.Howarth.2007.Quantitativetraitlociforgrainyieldinpearlmilletundervariablepostfloweringmoistureconditions.CropSci.47:969–980.doi:10.2135/crop-sci2006.07.0465

Bradbury,P.J.,Z.Zhang,D.E.Kroon,T.M.Casstevens,Y.Ramdoss,andE.S.Buckler.2007.TASSEL:Softwareforassociationmappingofcomplextraitsindiversesamples.Bioinformatics23:2633–2635.doi:10.1093/bioinformatics/btm308

Bramel-Cox,P.J.,K.AnandKumar,J.H.Hancock,andD.J.Andrews.1992.Sorghumandmilletsforforageandfeed.In:D.A.V.Dendy,editor,Sor-ghumandmillets,chemistryandtechnology.AmericanAssociationofCerealChemists,St.Paul,MN.p.325–364.

Broman,K.W.,H.Wu,S.Sen,andG.A.Churchill.2003.R/qtl:QTLmap-pinginexperimentalcrosses.Bioinformatics19:889–890.doi:10.1093/bioinformatics/btg112

Burton,G.W.,andJ.B.Powel.1968.Pearlmilletbreedingandcytogenetics.Adv.Agron.20:49–89.doi:10.1016/S0065-2113(08)60854-8

Chemisquy,M.A.,L.M.Giussani,M.A.Scataglini,E.A.Kellogg,andO.Morrone.2010.PhylogeneticstudiesfavourtheunificationofPen-nisetum, CenchrusandOdontelytrum(Poaceae):Acombinednuclear,plastidandmorphologicalanalysis,andnomenclaturalcombinationsinCenchrus.Ann.Bot.(Lond.)106:107–130.doi:10.1093/aob/mcq090

Collins,V.P.,A.H.Cantor,A.J.Pescatore,M.L.Straw,andM.J.Ford.1997.Pearlmilletinlayerdietsenhanceseggyolkn-3fattyacids.Poult.Sci.76:326–330.doi:10.1093/ps/76.2.326

Cunningham,D.L.,andB.D.Fairchild.2012.BroilerproductionsystemsinGeorgiacostsandreturnsanalysis.UniversityofGeorgiaCooperativeExtensionB1240.http://www.caes.uga.edu/departments/agecon/exten-sion/pubs/documents/B1240_3.PDF(accessed11Mar.2016).

Dahlberg,J.A.,J.P.Wilson,andT.Snyder.2004.Sorghumandpearlmil-let:Healthfoodsandindustrialproductsindevelopedcountries.In:AlternativeUsesofSorghumandPearlMilletinAsia.CFCTechnicalPaperNo.34.ProceedingsofanExpertMeeting,Patancheru,AndhraPradesh,India.1–4July2003.ICRISAT,India,p.42–49.

Davis,A.J.,N.M.Dale,andF.J.Ferreira.2003.Pearlmilletasanalterna-tivefeedingredientinbroilerdiets.J.Appl.Poult.Res.12:137–144.doi:10.1093/japr/12.2.137

Devos,K.M.,T.S.Pittaway,A.Reynolds,andM.D.Gale.2000.Compara-tivemappingrevealsacomplexrelationshipbetweenthepearlmil-letgenomeandthoseoffoxtailmilletandrice.Theor.Appl.Genet.100:190–198.doi:10.1007/s001220050026

Dwivedi,S.L.,H.Upadhyaya,S.Senthilvel,C.Hash,K.Fukunaga,X.Diao,etal.2012.Millets:Geneticandgenomicresources.PlantBreed.Rev.35:247–375.doi:10.1002/9781118100509.ch5.

Durham,S.2003.Newstrainofpearlmillet.Agric.Res.Mag.51:19.Elshire,R.J.,J.C.Glaubitz,Q.Sun,J.A.Poland,K.Kawamoto,E.S.Buckler,

etal.2011.Arobust,simplegenotyping-by-sequencing(GBS)approachforhighdiversityspecies.PLoSONE6(5):e19379.doi:10.1371/journal.pone.0019379

Farrell,D.J.2005.Matchingpoultryproductionwithavailablefeedresources:Issuesandconstraints.WorldPoultrySci.J.61:298–307.doi:10.1079/WPS200456

Gale,M.D.,andK.M.Devos.1998.Plantcomparativegeneticsafter10years.Science282:656–659.doi:10.1126/science.282.5389.656

Gale,M.D.,K.M.Devos,J.H.Zhu,S.Allouis,M.S.Couchman,H.Liu,etal.2005.Newmolecularmarkertechnologiesforpearlmilletimprove-ment.Int.SorghumMilletsNewslett.42:16–22.

Ganal,M.W.,T.Altmann,andM.S.Röder.2009.SNPidentificationincropplants.Curr.Opin.PlantBiol.12:211–217.doi:10.1016/j.pbi.2008.12.009

Garcia,A.R.,andN.M.Dale.2006.Feedingofungroundpearlmillettolay-inghens.J.Appl.Poult.Res.15:574–578.doi:10.1093/japr/15.4.574

Glaubitz,J.C.,T.M.Casstevens,F.Lu,J.Harriman,R.J.Elshire,Q.Sun,etal.2014.TASSEL-GBS:Ahighcapacitygenotypingbysequencinganal-ysispipeline.PLoSONE9(2):e90346.doi:10.1371/journal.pone.0090346

Gulia,S.K.,J.P.Wilson,J.Carter,andB.P.Singh.2007.Progressingrainpearlmilletresearchandmarketdevelopment.In:J.JanikandA.Whip-key,editors,Issuesinnewcropsandnewuses.ASHSPress,Alexandria,VA.p.196–203.

Gupta,A.K.,K.N.Rai,P.Singh,V.L.Ameta,K.Suresh,A.K.Jayalekha,etal.2015.Seedsetunderhightemperaturesduringfloweringperiod

inpearlmillet(Pennisetum glaucumL.).FieldCropsRes.171:41–53.doi:10.1016/j.fcr.2014.11.005

Gupta,S.K.,R.Sharma,K.N.Rai,andR.P.Thakur.2012.Inheritanceoffoliarblastresistanceinpearlmillet(Pennisetum glaucumL.(R.)Br.).PlantBreed.131:217–219.doi:10.1111/j.1439-0523.2011.01929.x

Hanna,W.W.1990.TransferofgermplasmfromthesecondarytotheprimarygenepoolinPennisetum.Theor.Appl.Genet.80:200–204.doi:10.1007/BF00224387

Hanna,W.W.,andH.D.Wells.1989.InheritanceofPyricularialeafspotresistanceinpearlmillet.J.Hered.80:145–147.

Hanna,W.W.,andH.D.Wells.1993.RegistrationofparentallineTift89D2,rustresistantpearlmillet.CropSci.33:361–362.doi:10.2135/cropsci1993.0011183X003300020048x

Hanna,W.,J.Wilson,andP.Timper.2005a.RegistrationofpearlmilletparentallineTift454.CropSci.45:2670.doi:10.2135/cropsci2005.0171

Hanna,W.,J.Wilson,andP.Timper.2005b.RegistrationofpearlmilletparentallinesTift99D2A1/B1.CropSci.45:2671.doi:10.2135/crop-sci2005.0172

Hash,C.T.,andP.J.Bramel-Cox.2000.Markerapplicationsinpearlmil-let.In:B.I.G.Haussmann,H.H.Geiger,D.E.Hess,C.T.Hash,andP.Bramel-Cox(eds.)TrainingmanualforaseminarheldatIITA,Ibadan,Nigeria,16–17Aug.1999.ICRISAT,Patancheru,India.p.112–127

He,J.,X.Zhao,A.Laroche,Z.X.Lu,H.Liu,andZ.Li.2014.Genotyping-by-sequencing(GBS),anultimatemarkerassistedselection(MAS)tooltoaccelerateplantbreeding.Front.PlantSci.5:484.doi:10.3389/fpls.2014.00484

Howarth,C.J.,G.P.Cavan,K.P.Skot,R.W.H.Layton,C.T.Hash,andJ.R.Witcombe.1994.MappingQTLsforheattoleranceinpearlmillet.In:J.R.WitcombeandR.R.Duncan,editors,Useofmolecularmarkersinsorghumandpearlmilletbreedingfordevelopingcountries.OverseasDevelopmentAdministration,London,UK.p.80–85.

Jauhar,P.P.,andW.W.Hanna.1998.Cytogeneticsandgeneticsofpearlmil-let.Adv.Agron.64:1–26.doi:10.1016/S0065-2113(08)60501-5

Krawczak,M.1999.Informativityassessmentforbiallelicsinglenucleotidepolymorphisms.Electrophoresis.20:1676–1681.

Kumar,S.,T.W.Banks,andS.Cloutier.2012.SNPdiscoverythroughnext-generationsequencinganditsapplications.Int.J.PlantGenomics2012:831460.doi:10.1155/2012/831460.

Lander,E.S.,andD.Botstein.1989.MappingMendelianfactorsunderlyingquantitativetraitsusingRFLPlinkagemaps.Genetics121:185–199.

Langmead,B.,andS.Salzberg.2012.Fastgapped-readalignmentwithBowtie2.Nat.Methods9:357–359.doi:10.1038/nmeth.1923

Lee,D.,W.Hanna,G.D.Buntin,W.Dozier,P.Timper,andJ.P.Wilson.2004.Pearlmilletforgrain.UniversityofGeorgia.http://extension.uga.edu/publications/files/pdf/B%201216_3.PDF(accessed11Mar.2016).

Liu,C.J.,J.R.Witcombe,T.S.Pittaway,M.Nash,C.T.Hash,C.S.Busso,andM.D.Gale.1994.AnRFLP-basedgenetic-mapofpearlmillet(Pennisetum glaucum).Theor.Appl.Genet.89:481–487.doi:10.1007/BF00225384.

Lu,F.,A.E.Lipka,R.J.Elshire,J.C.Glaubitz,J.H.Cherney,M.D.Casler,etal.2013.Switchgrassgenomicdiversity,ploidyandevolution:Novelinsightsfromanetwork-basedSNPdiscoveryprotocol.PLoSGenet.9:E1003215.doi:10.1371/journal.pgen.1003215

Maman,N.,S.C.Mason,andD.J.Lyon.2006.Nitrogenrateinfluenceonpearlmilletyield,nitrogenuptake,andnitrogenuseefficiencyinNebraska.Commun.SoilSci.PlantAnal.37(1–2):127–141.doi:10.1080/00103620500406112

Mammadov,J.,R.Aggarwal,R.Buyyarapu,andS.Kumpatla.2012.SNPmarkersandtheirimpactonplantbreeding.Int.J.PlantGenomics2012:728398:doi:10.1155/2012/728398.

Martel,E.,N.D.De,S.Siljak-Yakovlev,S.Brown,andA.Sarr.1997.Genomesizevariationandbasicchromosomenumberinpearlmilletandfour-teenrelatedPennisetumspecies.Heredity88:139–143.doi:10.1093/oxfordjournals.jhered.a023072

Morgan,R.N.,J.P.Wilson,W.W.Hanna,andP.Ozais-Akins.1998.Molecu-larmarkersforrustandpyricularialeafspotdiseaseresistanceinpearlmillet.Theor.Appl.Genet.96:413–420.doi:10.1007/s001220050757

Moumouni,K.H.,B.A.Kountche,M.Jean,C.T.Hash,Y.Vigouroux,B.I.G.Haussmann,etal.2015.Constructionofageneticmapforpearlmil-let,Pennisetum glaucum(L.)R.Br.,usingagenotyping-by-sequencing(GBS)approach.Mol.Breed.35:5.doi:10.1007/s11032-015-0212-x

Muchow,R.C.1988.Effectofnitrogensupplyonthecomparativepro-ductivityofmaizeandsorghuminasemi-aridtropicalenviron-mentleafgrowthandleafnitrogen.FieldCropsRes.18(1):131–143.doi:10.1016/0378-4290(88)90056-1.

punnuri et al.: development of a high-density linkage map in pearl millet 13 of 13

Nambiar,V.S.,J.J.Dhaduk,N.Sareen,T.Shahu,H.Shah,andR.Desai.2011.Potentialfunctionalimplicationsofpearlmillet(Pennisetum glau-cum)inhealthanddisease.J.Appl.Pharm.Sci.01(10):62–67.

Nepolean,T.,M.Blümmel,A.G.BhaskerRaj,V.Rajaram,S.Senthilvel,andC.T.Hash.2006.QTLscontrollingyieldandstoverqualitytraitsinpearlmillet.Int.SorghumMilletsNewslett.47:149–152.

Oumar,I.,C.Mariac,J.-L.Pham,andY.Vigouroux.2008.Phylogenyandoriginofpearlmillet(Pennisetum glaucum[L.]R.Br.)asrevealedbymicrosatelliteloci.Theor.Appl.Genet.117:489–497.doi:10.1007/s00122-008-0793-4

Paiva,E.A.A.,F.O.Bustamante,S.Barbosa,A.V.Pereira,andL.C.Davide.2012.Meioticbehaviorinearlyandrecentduplicatedhexaploidhybridsofnapiergrass(Pennisetum purpureum)andpearlmillet(Pennisetum glaucum).Caryologia65(2):114–120.doi:10.1080/00087114.2012.709805

Parvathaneni,R.K.,V.Jakkula,F.K.Padi,S.Faure,N.Nagarajappa,A.C.Pontaroli,etal.2013.Fine-mappingandidentificationofacandidategeneunderlyingthed2dwarfingphenotypeinpearlmillet,Cenchrus americanus(L.)Morrone.G3.3:563–572.doi:10.1534/g3.113.005587.

Peacock,J.M.,P.Soman,R.Jayachandran,A.V.Rani,C.J.Howarth,andA.Thomas.1993.Effectofhighsoilsurfacetemperatureonseed-lingsurvivalinpearlmillet.Exp.Agric.29:215–225.doi:10.1017/S0014479700020664

Pedraza-Garcia,F.,J.E.Specht,andI.Dweikat.2010.AnewPCR-basedlinkagemapinpearlmillet.CropSci.50:1754–1760.doi:10.2135/crop-sci2009.10.0560

Poland,J.A.,andT.W.Rife.2012.Genotyping-by-sequencingforplantbreedingandgenetics.PlantGen.5:92.doi:10.3835/plantgen-ome2012.05.0005

Qi,X.,T.S.Pittaway,S.Lindup,H.Liu,E.Waterman,F.K.Padi,etal.2004.Anintegratedgeneticmapandanewsetofsimplesequencerepeatmarkersforpearlmillet,Pennisetum glaucum.Theor.Appl.Genet.109:1485–1493.doi:10.1007/s00122-004-1765-y

RCoreTeam.2014.R:Alanguageandenvironmentforstatisticalcomput-ing.RFoundationforStatisticalComputing,Vienna,Austria.http://www.R-project.org(accessed11Mar.2016).

Rajaram,V.,T.Nepolean,S.Senthilvel,R.K.Varshney,V.Vadez,R.K.Sriv-astava,etal.2013.Pearlmillet[Pennisetum glaucum(L.)R.Br.]consen-suslinkagemapconstructedusingfourRILmappingpopulationsandnewlydevelopedEST-SSRs.BMCGenomics14:159.doi:10.1186/1471-2164-14-159

Saïdou,A.A.,C.Mariac,V.Luong,J.L.Pham,G.Bezançon,andY.Vigouroux.2009.AssociationstudiesidentifynaturalvariationatPHYClinkedtofloweringtimeandmorphologicalvariationinpearlmillet.Genetics182:899–910.doi:10.1534/genetics.109.102756

Sathya,M.,P.Sumathi,N.Senthil,S.Vellaikumar,andA.JohnJoel.2014.GeneticstudiesforyieldanditscomponenttraitsinRILpopulationofpearlmillet(Pennisetum glaucum[L.]R.Br.).ElectronicJ.PlantBreed-ing5(2):322–326.

Sehgal,D.,L.Skot,R.Singh,R.K.Srivastava,S.P.Das,J.Taunk,etal.2015.Exploringpotentialofpearlmilletgermplasmassociationpanelforassociationmappingofdroughttolerancetraits.PLoSONE10:e0122165.doi:10.1371/journal.pone.0122165

Sehgal,D.,V.Rajaram,V.Vadez,C.T.Hash,andR.S.Yadav.2012.Integra-tionofgene-basedmarkersinpearlmilletgeneticmapforidentificationofcandidategenesunderlyingdroughttolerancequantitativetraitloci.BMCPlantBiol.12:9.doi:10.1186/1471-2229-12-9

Senthilvel,S.,B.Jayashree,V.Mahalakshmi,P.S.Kumar,S.Nakka,T.Nepolean,andC.T.Hash.2008.Developmentandmappingofsimplesequencerepeatmarkersforpearlmilletfromdataminingofexpressedsequencetags.BMCPlantBiol.8:119.doi:10.1186/1471-2229-8-119

Spindel,J.,M.Wright,C.Chen,J.Cobb,J.Gage,S.Harrington,etal.2014.Bridgingthegenotypinggap:Usinggenotypingbysequencing(GBS)toaddhigh-densitySNPmarkersandnewvaluetotraditionalbi-parentalmappingandbreedingpopulations.Theor.Appl.Genet.126(11):2699–2716.doi:10.1007/s00122-013-2166-x

Sullivan,T.W.,J.H.Douglas,D.J.Andrews,P.L.Bond,J.D.Hancock,P.J.Bramel-Cox,etal.1990.Nutritionalvalueofpearlmilletforfoodandfeed.In:G.Ejeta,E.T.Mertz,L.Rooney,R.Schaffert,andJ.Yohe(eds)ProceedingsoftheInternationalConferenceonSorghumNutritionalQuality,26Feb.–1March1990,PurdueUniversity,WestLafayette,IN.PurdueUniversity,WestLafayette,IN.p.83–94

Supriya,A.,S.Senthilvel,T.Nepolean,K.Eshwar,V.Rajaram,R.Shaw,etal.2011.DevelopmentofamolecularlinkagemapofpearlmilletintegratingDArTandSSRmarkers.Theor.Appl.Genet.123:239–250.doi:10.1007/s00122-011-1580-1

Swarts,K.,L.Huihui,J.A.RomeroNavarro,A.Dong,M.C.Romay,S.Hearne,etal.2014.Novelmethodstooptimizegenotypicimputationforlow-coverage,next-generationsequencedataincropplants.PlantGen.7.doi:10.3835/plantgenome2014.05.0023

Techio,V.H.,L.C.Davide,andA.V.Pereira.2006.Meiosisinelephantgrass(Pennisetum purpureum),pearlmillet(Pennisetum glaucum)(Poaceae,Poales)andtheirinterspecifichybrids.Genet.Mol.Biol.29(2):353–362.doi:10.1590/S1415-47572006000200025

Thakur,R.P.,R.Sharma,andV.P.Rao.2011.Screeningtechniquesforpearlmilletdiseases.InformationBulletinNo.89.ICRISAT,Patancheru,AndhraPradesh,India.

Thomas,G.,T.Bhavna,andN.C.Subrahmanyam.2000.HighlyrepetitiveDNAsequencesofpearlmillet:ModulationamongPennisetumspe-ciesandcereals.J.PlantBiochem.Biotechnol.9:17–22.doi:10.1007/BF03263077

Timper,P.,J.P.Wilson,A.W.Johnson,andW.W.Hanna.2002.EvaluationofpearlmilletgrainhybridsforresistancetoMeloidogynespp.andleafblightcausedbyPyricularia grisea.PlantDis.86:909–914.doi:10.1094/PDIS.2002.86.8.909

Vadez,V.,T.Hash,F.R.Bidinger,andJ.Kholova.2012.Phenotypingpearlmilletforadaptationtodrought.Front.Physiol.3:1–12.doi:10.3389/fphys.2012.00386

Venables,W.N.,andB.D.Ripley.2002.ModernappliedstatisticswithS.4thed.Springer,NewYork.

Ward,J.A.,J.Bhangoo,F.Fernández-Fernández,P.Moore,J.D.Swanson,R.Viola,etal.2013.SaturatedlinkagemapconstructioninRubus idaeususinggenotypingbysequencingandgenome-independentimputation.BMCGenomics14:2.doi:10.1186/1471-2164-14-2

Wilson,J.P.,andW.W.Hanna.1992.Effectsofgeneandcytoplasmsubsti-tutionsinpearlmilletonleafblightepidemicsandinfectionbyPyricu-laria grisea.Phytopathology82:839–842.doi:10.1094/Phyto-82-839

Wilson,J.P.,G.W.Burton,H.D.Wells,J.D.Zongo,andI.O.Dicko.1989.Leafspot,rustandsmutresistanceinpearlmilletlandracesfromcen-tralBurkinaFaso.PlantDis.73:345–349.doi:10.1094/PD-73-0345

Wimpee,C.F.,andJ.R.Y.Rawson.1979.Characterisationofthenucleargenomeofpearlmillet.Biochim.Biophys.Acta562:192–206.doi:10.1016/0005-2787(79)90165-5

Wu,Y.,P.R.Bhat,T.J.Close,andS.Lonardi.2008.Efficientandaccurateconstructionofgeneticlinkagemapsfromtheminimumspanningtreeofagraph.PLoSGenet.4(10):E1000212.doi:10.1371/journal.pgen.1000212

Yadav,R.S.,C.T.Hash,F.R.Bidinger,G.P.Cavan,andC.J.Howarth.2002.Quantitativetraitlociassociatedwithtraitsdetermininggrainandstoveryieldinpearlmilletunderterminaldroughtstressconditions.Theor.Appl.Genet.104:67–83.doi:10.1007/s001220200008

Yadav,O.P.,andK.N.Rai.2011.HybridizationofIndianlandracesandAfricanelitecompositesofpearlmilletresultsinbiomassandstoveryieldimprovementunderaridzoneconditions.CropSci.51:1980–1987.doi:10.2135/cropsci2010.12.0731

Yadav,O.P.,K.N.Rai,I.S.Khairwal,B.S.Rajpurohit,andR.S.Mahala.2011a.Breedingpearlmilletforaridzoneofnorth-westernIndia:Con-straints,opportunitiesandapproaches.AllIndiaCoordinatedPearlMilletImprovementProject,Jodhpur,India.

Yadav,R.S.,F.R.Bidinger,C.T.Hash,Y.P.Yadav,O.P.Yadav,S.K.Bhatnagar,etal.2003.MappingandcharacterizationofQTL×Einteractionsfortraitsdetermininggrainandstoveryieldinpearlmillet.Theor.Appl.Genet.106:512–520.doi:10.1007/s00122-002-1081-3.

Yadav,R.S.,C.T.Hash,F.R.Bidinger,K.M.Devos,andC.J.Howarth.2004.Genomicregionsassociatedwithgrainyieldandaspectsofpost-floweringdroughttoleranceinpearlmilletacrossstressenviron-mentsandtesterbackground.Euphytica136:265–277.doi:10.1023/B:EUPH.0000032711.34599.3a

Yadav,R.S.,D.Sehgal,andV.Vadez.2011b.Usinggeneticmappingandgenomicsapproachesinunderstandingandimprovingdroughttoler-anceinpearlmillet.J.Exp.Bot.62:397–408.doi:10.1093/jxb/erq265

Zeng,Z.-B.1994.Precisionmappingofquantitativetraitloci.Genetics136:1457–1468.