Unraveling ancient hexaploidy through multiply aligned angiosperm gene maps

12 Pages·2008·0.83 MB·English

Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Methods
Unraveling ancient hexaploidy through multiply
aligned angiosperm gene maps
Haibao Tang,1,2 Xiyin Wang,1,3 John E. Bowers,1 Ray Ming,4 Maqsudul Alam,5
and Andrew H. Paterson1,2,6
1PlantGenomeMappingLaboratory,UniversityofGeorgia,Athens,Georgia30602,USA;2DepartmentofPlantBiology,
UniversityofGeorgia,Athens,Georgia30602,USA;3CollegeofScience,HebeiPolytechnicUniversity,Tangshan,
Hebei063000,China;4DepartmentofPlantBiology,UniversityofIllinoisatUrbana–Champaign,Champaign,
Illinois61801,USA;5AdvancedStudiesinGenomics,ProteomicsandBioinformaticsunit,UniversityofHawaii,
Honolulu,Hawaii96822,USA
Large-scale (segmental or whole) genome duplication has been recurring in angiosperm evolution. Subsequent gene
loss and rearrangements further affect gene copy numbers and fractionate ancestral gene linkages across multiple
chromosomes. The fragmented “multiple-to-multiple” correspondences resulting from this distinguishing feature of
angiosperm evolution complicates comparative genomic studies. Using a robust computational framework that
combines information from multiple orthologous and duplicated regions to construct local syntenic networks, we
show that a shared ancient hexaploidy event (or perhaps two roughly concurrent genome fusions) can be inferred
based on the sequences from several divergent plant genomes. This “paleo-hexaploidy” clearly preceded the
rosid–asterid split, but it remains equivocal whether it also affected monocots. The model resulting from our
multi-alignments lays the foundation for approximating the number and arrangement of genes in the last universal
common ancestor of angiosperms. Comparative analysis of inferred homologous genes derived from this model
shows patterns of preferential gene retention or loss after polyploidy and reveals large variability of nucleotide
substitutionratesamongplantnucleargenomes.
[Supplementalmaterialisavailableonlineatwww.genome.org.]
Ancient genome duplications are evident for many lineages of thatdidnotexperiencepolyploidyorundergomassivegeneloss
fungi(Kellisetal.2004),animals(Jaillonetal.2004),andplants (Van de Peer 2004). For example, “bridging” of ghost duplica-
(Bowersetal.2003),offeringopportunitiesfortheevolutionof tionsusingoutgroupshasclarifiedthehistoryofpolyploidyin
new (Spillane et al. 2007) or modified (Hittinger and Carroll bothSaccharomycesandTetraodon(Jaillonetal.2004;Kellisetal.
2007) gene functions, altering gene dosages, and creating new 2004;Scannelletal.2007).
gene arrangements. Traces from past whole-genome duplica- Continuous stretches of duplicate genes can be computa-
tion events can often be detected from pairwise syntenic seg- tionallydeducedthroughsynteny,usingsomevariantsofclus-
ments, including two sets of retained paralogs that have main- teringapproaches(Vandepoeleetal.2002;Hampsonetal.2005)
tainedrelativegenomiclocationsonsyntenicchromosomes.In ormorespecificallyusingdynamicprogrammingwithacustom-
angiosperms, genome duplications are recurring in many lin- izedscoringschemeifconservedgeneorder(collinearity)isalso
eages (Bowers et al. 2003), generating large numbers of paralo- considered(Haasetal.2004;Wangetal.2006).Traditionalmeth-
gousloci. odsfordeductionofsyntenybasedon“best-in-genome”criteria
Genelossatduplicatedlocieffectivelyfractionatesancestral (Milleretal.2007),uncoveringone-to-onebestmatchingregions
linkagepatternsandreducesthedensityofcontinuousstretches during pairwise genome comparisons, are relatively straightfor-
of“paleologous”genepairs,whicharetheremainingsignatures ward in vertebrates yet difficult in angiosperms because of
ofpaleo-polyploidy(Thomasetal.2006).Dependingonthelevel additional challenges that are more prominent in angiosperm
of gene loss, the remaining signatures of duplication are some- genomes (Tang et al. 2008). These challenges include frequent
timessoerodedthatthehomologoussegmentscannolongerbe genome duplications and convoluted genome shuffling (rear-
identifiedbasedonlyonsimilaritytooneanother.Theproblem rangements,chromosomalfusionsandfissions),suchastheex-
ismultipliedwhenthespeciesinquestionhasundergoneseveral tensive rearrangement that has occurred in Arabidopsis within
genome duplications, with recent duplications tending to ob- thepast5millionyears(Kuittinenetal.2004).
scuresyntenyfrommoreancienteventsasisfoundinmostan- Oneapproachforthecomputationalde-convolutionofpa-
giospermgenomes.Suchhighlydegenerateduplicatedsegments leopolyploidyfordeductionofancestralgeneordersisabottom-
have been referred to as “ghost duplications” and can often be up approach in which one attempts to resolve one duplication
resolved by comparison to an appropriate “outgroup” genome eventatatime,startingwiththemostrecentone.Thisisexem-
plifiedbystudiesinArabidopsisandParameciumwherethemost
recentlyduplicatedsegmentsaremergedtogeneratehypotheti-
6Correspondingauthor. cal intermediate profiles that are further recursively merged
[email protected];fax[phone].
(Bowersetal.2003;Auryetal.2006).
Articlepublishedonlinebeforeprint.Articleandpublicationdateareathttp://
www.genome.org/cgi/doi/10.1101/gr.080978.108. Herein,weelaborateonanalternativetop-downapproach
18:000–000©2008byColdSpringHarborLaboratoryPress;ISSN1088-9051/08;www.genome.org GenomeResearch 1
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Tang et al.
Table1. Summaryofsequencedplantgenomesbasedonrespectivegenomepublications
Species Assemblystatusa Assembled/estimatesize Annotationversion Annotatedgeneno.
Arabidopsis(Arabidopsisthaliana) BAC-by-BAC 115Mb/160Mb TAIRversion7 26784
Papaya(Caricapapaya) WGS,N50=11kb 278Mb/372Mb UniversityofHawaii 25536
Poplar(Populustrichocarpa) WGS,N50=125kb 410Mb/485Mb JGIversion1.1 45554
Grape(Vitisvinifera) WGS,N50=65kb 468Mb/487Mb Genoscoperelease 30434
Rice(Oryzasativassp.japonica) BAC-by-BAC 371Mb/389Mb RAPrelease2b 29389
a(BAC)bacterialartificialchromosome;(WGS)whole-genomeshotgun;(N50)maximumlengthLsuchthat50%ofallbasesareincontigsoflengthat
leastL.
bWeonlyusedmappedrepresentativelociforthericeannotationproject(RAP)release(Itohetal.2007).
(Tangetal.2008)thatisconceptuallymoreattractiveinthatit Results
onlyrequiresonecycleofdeduction—firstsearchingforpairwise
MCscan:Algorithmformultiplegeneorderalignments
syntenyinformationandthencombiningtheresultingpairsto
formamulti-waycorrespondenceamongallstructurallysimilar Whenseveralgenomesandsubgenomes(resultingfromancient
chromosomalsegments.Theefficacyofthetop-downapproach, duplication events) are compared simultaneously, synteny and
however, depends on the searching strategy because of the de- collinearitybetweenallpossiblepairsofgenomesaretediousto
generate synteny resulting from post-duplication gene loss. In enumerate because chromosomal homology is “transitive.” For
particular,atop-downsearchstrategycanincorporate“ghostdu- example, if there are corresponding chromosomal regions in
plications”(VandePeer2004),whicharenotdiscernibleusinga threegenomesA,B,andC,comparisonsbetweenthegenomes
bottom-up approach based on information from only one spe- would reveal three pairwise synteny blocks (A-B, B-C, A-C),
cies. whereasitcouldbebetterrepresentedasasinglemultiplesyn-
New angiosperm genome sequences (Table 1) promise to teny block (A-B-C). To solve this problem, we implemented a
qualitativelyimproveourdeductionsabouttheevolutionofan- novelalgorithm,MCscan,thatexploitsthistransitivityproperty
giosperm gene repertoire and arrangement. Arabidopsis (Arabi- ofcollinearitytoperformmultiplealignmentsbyincorporating
dopsisGenomeInitiative2000),rice(Oryzasativa)(International pairwisesyntenythatisderivedfromsharedevolutionaryevents.
Rice Genome Sequencing Project 2005), poplar (Populus tricho- The algorithm involves a four-stage pipeline illustrated in
carpa)(Tuskanetal.2006),grapevine(Vitisvinifera)(Jaillonetal. Figure1,witheachindividualstagedescribedinfurtherdetailin
2007),andpapaya(Caricapapaya)(Mingetal.2008)havebeen Methods.
sequenced, and more are in the pipeline. Indeed, Arabidopsis Wefirstuseasequencesimilaritysearchprogramtodetect
thaliana—a leading botanical model—is
now known to be a relatively difficult
system from which to deduce ancient
gene orders. For example, many Carica
segments show collinearity with three
or four Arabidopsis segments, showing
that two genome duplications have af-
fected the Arabidopsis lineage since its
divergence from Carica (Ming et al.
2008). Individual Arabidopsis genome
segmentscorrespondtoonlyoneCarica
segment, showing that Carica has not
duplicatedsinceitsdivergencefromAra-
bidopsis. Both Vitis and Carica have
only one duplication event, (cid:1), while (cid:2)
and (cid:3) occurred in the Arabidopsis line-
age after its divergence from the Carica
lineage (Ming et al. 2008; Tang et al.
2008).
Some newly sequenced genomes
havelesscomplicatedgenomestructure
and thus may represent better models
for comparative genomics than Arabi-
dopsis.Inthisstudy,weexploitfragmen-
tary conservation of plant gene orders
from multiple genomes along with a
new top-down algorithm MCscan, to
improvedeductionsaboutthecourseof
angiosperm genome structural evolu-
tion. Figure1. Flow-chartofMCscancorealgorithm.
2 Genome Research
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Paleopolyploidy and angiosperm genome evolution
matchingsamonggenesinallpossiblepairsofchromosomesand changeinthefuturewhenweincludeadditionalgenomes;how-
scaffoldsandinbothtranscriptionaldirections.Thisisfollowed ever, using Vitis as the current “reference” would produce the
by the “pairwise collinearity” stage, in which the neighboring bestsolutionsofar.
matches are chained along using dynamic programming. The ThetriplicationofgenelociisalsoevidentfromTable2.For
pairwise collinear blocks are combined in the “multi- example,wefoundthat88alignedlociinCaricahavemultiplic-
collinearity”stage,byfixingonegeneorderasreferenceandthen ity levels of three (triplication (cid:1)), with only one aligned locus
heuristically stacking the pairwise synteny tracks one after an- exceedingamultiplicityof3;54alignedlociinPopulushavethe
other.Inthisstep,weneedtousea“reference”geneorderasthe expectedmultiplicitylevelof6(triplication(cid:1)(cid:1)duplicationp),
basisforstackingthetracks;wethendescribethealignedsynteny butonlythreelociexceed6.Thelocithatexceedtheexpected
blocksas“threadedbythereferenceorder,”aprocedureinspired multiplicity level are likely produced by additional small-scale
byTBAaligner(Blanchetteetal.2004).Oncethemulti-syntenic (singlegeneorsegmental)duplicationsineachlineage.
blocks are identified, we can classify the segments and index
themtodifferentevolutionaryevents,mainlyduplicationsand Furthercircumscribingthe(cid:1)duplicationevent
divergence. The (cid:1) duplication event was previously dated to have occurred
As a result, MCscan condenses the combinatorial matches
afterthemonocot–dicotseparationbutbeforetheexpansionof
between multiple chromosomal segments resulting from diver-
therosids(Jaillonetal.2007).Weinvestigatedthelowerbound-
genceandrecursiveduplicationeventsandcreatesaviewofthe
ary of this claim by sampling genomic regions from other eu-
multiplyalignedsegments.
dicots outside the rosids for which long, contiguous sequences
(BACs) were available in GenBank, including tomato (Solanum
Patternsofsyntenyconservation
lycopersicum)andbanana(Musaacuminata).
Using the top-down algorithm MCscan, we have aligned large Wefirstmappedunigenesonto194sequencedtomato(So-
portions of the five sequenced genomes (Arabidopsis, Carica, lanum lycopersicum) BACs as preliminary gene annotation and
Populus,Vitis,andOryza)basedonsynteny.Atotalof61%ofthe inspected synteny to Vitis. Among the 78 Solanum BACs that
Arabidopsisgeneshavepreservedtheirancestrallocationsbased havemorethan10distinctivelymappedunigenes,72havemore
oncross-speciessynteny(Table2),versus44%,51%,and46%of than 50% of genes showing primary synteny to a single Vitis
Carica,Populus,andVitisgenes,respectively. chromosome (Supplemental Data 2). Each individual tomato
Thevariationinfrequenciesofalignedgenesmightbedue BAC corresponds closely to only one of the triplicate regions
to different levels of synteny conservation in different species. ratherthanshowingequalmatchestoeachofthethree(cid:1)paleo-
However,itisalsocorrelatedwiththedegreeofcontiguityofthe homeologous chromosomes in Vitis. Figure 2A shows one ex-
respectivesequences(Table1),withahigherpercentageofgenes ampleofaSolanumBACthatalignstotheVitisgeneorder.Al-
explainedbysyntenyinthegenomeswithhigherN50.Indeed,if though the Solanum BACs that we inspected only represent
mostgenesareinsmallorunanchoredscaffolds,itwouldbevery ∼2.5%ofthegenome,theevidencesofarstronglysupportsthe
difficultforMCscantodetectthemassyntenic,eveniftheydo hypothesisthat(cid:1)triplicationoccurredinacommonancestorof
remainintheirancestrallocations. asterids and rosids. Under this scenario, each Solanum segment
Alignments with gene order preserved across four eudicot wouldbeexpectedtohaveuptofourprimarysyntenicsegments
species show clear triplicated structure in many local regions. inArabidopsis,ashasbeensuggested(Kuetal.2000).
Eachtriplicatedbranchcontainsorthologoussegmentsfromup Basedonasimilarnotion,Jaillonetal.(2007)calculatedthe
to four Arabidopsis regions, one Carica region, two Populus re- relativeabundanceofone-to-threecasesbetweenOryzaandVitis
gions,andoneVitisregion,supportingthehypothesisthatthis andsuggestedthatthetriplicationoccurredafterthemonocot–
genome triplication ((cid:1)) occurred in a common ancestor of all dicotsplit.Itistemptingtopushthedatingof(cid:1)further,yetwe
fourspecies;Populushasoneduplicationevent(p)initssalicoid consider such dating to have uncertainties in view of current
lineage, and Arabidopsis has two duplications ((cid:2) and (cid:3)) in its evidence.Contrarytothewell-conservedsyntenywithintheeu-
cruciferlineage.ThemultiplealignmentswerethreadedbyVitis dicotgroup,only14%ofOryzagenescouldbeplacedincross-
asthereferenceorder(SupplementalData1),sinceVitisappeared species gene clusters (Table 2). This proportion represents the
to have the most close-to-ancestral karyotype among the ge- actualextentofcollinearitybetweenOryzaandanyofthefour
nomesthatweinvestigated(Jaillonetal.2007).Thisislikelyto eudicots,asOryzaistheonlymonocotgenomeincludedinthis
Table2. Numberofclusteredgroupsofgenesatdifferentmultiplicitylevelsinfiveangiospermspecies
Multiplicitylevel
No.ofancestral WGDorsegmental
Species 1 2 3 4 5 6 7 8 9 10 loci No.ofgenes(%) expansion
Arabidopsis 6742 2642 868 282 80 32 6 5 1 1 10,659 16,451(61%) 54%
Carica 9118 942 88a 1 0 0 0 0 0 0 10,149 11,270(44%) 11%
Populus 5147 6362 763 618 96 54a 3 0 0 0 13,043 23,457(51%) 80%
Vitis 9926 1671 239a 15 2 0 0 0 0 0 11,853 14,055(46%) 18%
Oryza 2197 685 140 35 9 2 0 0 0 0 3068 4184(14%) 36%
Thestatisticsarebasedonlyongroupsthatcontaingenesfromatleasttwodifferentspecies,asconstructedfromsyntenicalignments.Thenumberof
inferredancestrallociiscalculatedby∑10 N ,andthenumberofgenesthatmaintaintheirancestralpositionsiscalculatedby∑10 m(cid:4)N ,wheremis
m=1 m m=1 m
themultiplicitylevelvaryingfrom1to10andN isthenumberofgroupsforeachmultiplicitylevel.
m
aExpectedmultiplicitiesforCarica,Populus,andVitis.ThemultiplicityforArabidopsisis12(yetnogenegroupsretainedall12copies),andequivocalfor
Oryza.
Genome Research 3
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Tang et al.
to-oneprimarysyntenypatternofSola-
numandVitis,MusaBACsshowroughly
equal matches to any of the three (cid:1)
homeologs in Vitis (Fig. 2B), a pattern
similar to Oryza–Vitis. However, fail-
uretodetectone-to-one(asopposedto
one-to-three) correspondence between
monocot regions and Vitis cannot be
viewed as strong evidence that (cid:1) oc-
curred after the eudicot–monocot split.
Analternativebutequallyplausiblesce-
narioisthatthemonocotsandeudicots
share (cid:1) but diverged soon after (cid:1) oc-
curred.Underthisscenario,thegenear-
rangements between two orthologous
chromosomes would share very lit-
tle synteny because of stochastic, inde-
pendent gene losses in both lineages—
leading to similarly low levels of corre-
spondenceofchromosomeinonetaxon
toeachofitsthree(cid:1)paralogsinanother
taxon.
While highly specific one-to-one
synteny is indicative that two lineages
sharethe(cid:1)triplication,frequentone-to-
three synteny is not necessarily indica-
tive that one lineage lacks the triplica-
tion. So far we can only confidently
place the (cid:1) triplication before the aste-
rid–rosidsplitandconsiderthestatusof
the paleo-hexaploidy in the monocot
lineagetobeunclear.
Itisdifficulttotestthehypothesis
that the (cid:1) triplication predated the di-
vergenceofmonocotsandeudicots.For
example, additional data from an out-
groupgenomesuchasAmborellawould
help,butdoesnotnecessarilysolvethe
placement of the triplication if (cid:1) is
foundabsentinthatoutgroup.Muchof
theuncertaintyisrootedinthefactthat
the(cid:1)triplicationisanancienteventthat
atleastpredatedtheasterids–rosids,and
comparisons across this evolutionary
distance are often less effective. There-
Figure 2. Collinearity between triplicate Vitis (cid:1)-homeologous regions with BAC sequences from fore, we need broader and more judi-
Solanum(A)andMusa(B).(Blackglyphs)Geneswiththetipshowingthetranscriptionaldirection; cious sampling of plant taxa. Indeed,
(grayshades)syntenymatchesbetweenaVitisgeneandSolanumorMusasequences. fortuitous discoveries of genomes like
grapevine that have close-to-ancestral
study.Therefore,itismoredifficulttomakeaccurateinferenceof karyotypes facilitate comparisons across major angiosperm
syntenypatternsbecauseofthegreaterevolutionarydistancein- clades. Similarly, additional karyotypically conserved monocot
volved and additional duplication in the cereal lineage. While orbasalangiospermgenomesthatarefreeofrecentpolyploidies
several studies hinted that additional monocot duplication(s) mightbetterelucidatethescenario.
predatedthecerealduplication(cid:5)(Zhangetal.2005;Jaillonetal.
2007), whether such additional duplication(s) found in Oryza Comparisonsof(cid:1)paleologsshowthattriplicatesubgenomes
correspondtothe(cid:1)triplicationwesawincoreeudicotsremains
aremostlyhomogeneous
tobedetermined.
WealsoexaminedsyntenytoVitisforchromosomalregions Wetestedwhetheranytwoofthethreesubgenomesaregeneti-
from a monocot species that is basal to the cereals—banana callyclosertooneanotherthanthethird.Weretrieved(cid:1)paleo-
(Musa acuminata). On average, the levels of synteny between loggroupsthathaveretainedgenesfromallthree(cid:1)subgenomes,
MusaBACsandVitischromosomesare50%lowerthansynteny ondifferentchromosomesorscaffoldsinCaricaorVitis,thetwo
betweenSolanumandVitis.Furthermore,incontrasttotheone- genomes that are unaffected by additional duplications other
4 Genome Research
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Paleopolyploidy and angiosperm genome evolution
2006).Byexcludingthesinglegenedu-
plications,wewereabletoanalyzetheK
s
distribution using mixtures of log-
normals(seeMethods).
Although (cid:1) apparently occurred in
a common ancestor of Carica, Populus,
andVitis,themedianK betweenVitis(cid:1)
s
Figure3. Topologiesforfiveproximal(cid:1)ancestrallocithatcontainthreecollinearVitisgenes.Vitis paleologs (1.22) is much lower than
genenamesareabbreviatedas“[chromosome].[geneindex]”forgraphing.Eachtreewasrootedusing that of Carica (1.76) and Populus (1.54)
one best-matching moss gene, identified by JGI protein accession number. The numbers above (Table 3). The median values of K
s
branchesarebootstrapvaluesinthephylogeneticreconstruction.Thereareatotalof10localblocks among (cid:1) duplicates in these three ge-
thathavemorethanfivetripletsinCaricaandVitisthatarestudiedinthesameway.Phylogenetic
nomes show highly significant differ-
analysiswasperformedusingPHYLIPversion3.67(Retief2000).Theanalysiswascarriedoutusingthe
protdist program (default parameters) followed by neighbor-joining using neighbor. We used the ence (Kruskal-Wallis one-way ANOVA,
seqboot program to simulate 100 bootstrap replicates and the consense program to retrieve one P=2.25(cid:1)10(cid:2)142).
consensustree. The K distributions analyzed with
s
mixture models show the expected
than(cid:1).Wetheninferredgenetreesforthesetripletgroupsunder numberofcomponentsforeachspecies,exceptforArabidopsis,
theassumptionthatiftwosubgenomesare,indeed,moresimilar wherewecanfindonlytwoinsteadofthreedistinctcomponents
toeachotherthantothethird,weexpecttoseeonlyonepreva- (Table 3). This two-peak distribution (Fig. 4B) is similar to the
lenttreetopologyalongpaleologgroupswithinthesameances- resultsofapreviousstudy(Maereetal.2005)eventhoughMC-
tral duplicated (triplicated) segment. Only a limited data set is scanprovidesbetterdeductionsabouttheidentitiesofpaleologs.
suitable for this study since we need to have enough triplets Wepostulatethatmorerapidsubstitutionsoccuratsynonymous
alongthethreesubgenomesthatarederivedfromthesamean- sitesinArabidopsisthanintheotherthreeeudicotspecies,with
cestral segment. We picked 10 blocks with five or more Vitis Arabidopsis(cid:1)paleologsbeingsaturatedwithsynonymoussubsti-
triplets (this cutoff was chosen arbitrarily as we need enough tutions. Therefore, within Arabidopsis, K-based distances be-
s
tripletswithineachblockforinference,yetwedonothavemany tweenparalogscannotdifferentiate(cid:1)duplicatesfromeitherthe
blocksthathavemorethansixorseventriplets).Nonetheless,we tailofthedistributionof(cid:3)duplicates,orfromnoise,orboth.The
failedtofindonedominanttopologyforanyblock,withatypi- medianK valuesbetweenArabidopsis(cid:3)and(cid:1)duplicatesareclose
s
calexampleshowninFigure3.Thefactthatthe(cid:1)subgenomes tosaturation(2.00),muchlargerthanthoseofthe(cid:1)duplicatesin
areindistinguishablefromeachothermakesitunlikelythatone theotherthreespecies(Table3).Repeatingtheanalysisusinga
of the triplicated subgenomes may have originated from large- moreconservativegeneticdistance–transversionrateatfourfold
scalesegmentalduplicationsoraneuploidy.Instead,the(cid:1)tripli- degeneratesites(4DTV)(Fig.4C)showsalmostthesamepattern
cationmayhavebeenanancientauto-hexaploidyformedfrom asusingK,suggestingthatthesaturationeffectofDNAsubsti-
s
fusions of three identical genomes, or allo-hexaploidy formed tutionsmayhavealsoaffected4DTVdistance.
fromfusionsofthreesomewhatdivergedgenomes.Wearenot Differences in the median values of distances between the
able to determine whether the fusion(s) were a single event or paralogsthatarederivedfromthecommon(cid:1)eventcanbeex-
twoeventsarelativelyshorttimeapart(thelattercase,e.g.,char- plainedbydifferentsubstitutionratesamongthefourrosidlin-
acterizing the well-studied evolution of hexaploid wheat). A eages. We constructed a phylogenetic tree with per-branch K
s
more definitive test of allo-hexaploidy versus auto-hexaploidy estimates, based on orthologous gene groups that are strictly
wouldonlybepossibleifextantdiploidparentalspeciescanbe single copy in all five species (Fig. 4D). The same trend was
found,andthisisunlikelysincethegenomeduplicationsappear found,withincreasingevolutionaryratesinbranchesleadingto
to be pervasive throughout most angiosperm clades including Vitis, Populus, Carica, and Arabidopsis, respectively, suggesting
thebasallineages(Cuietal.2006). thatthevariationsofsubstitutionratesarenotconfinedtopopu-
lationsofduplicategenesbutareratherlineage-specific.Asimilar
rangeofnuclearratevariationinfloweringplantshasbeendocu-
Discussion mentedinpreviousstudies,andisoftenassociatedwithlifehis-
tory (Gaut et al. 1996; Koch et al. 2000). In general, the short
Byexploitingfragmentaryconservationofplantgeneorders,to-
generationtimeintheannualArabidopsismighthavecontribut-
getherwithanewtop-downmulti-alignmentapproach,limita-
edtothefastsubstitutionratescomparedwithPopulusorVitis,
tions of Arabidopsis for comparative genomics are mitigated by
using new angiosperm genome sequences to qualitatively im-
proveourdeductionsaboutthetempoandmodesofevolutionof Table3. MixturemodelestimatesfordistributionsofK between
s
angiospermgenesandgenomes. paleologsineachspecies
No.of
Ratevariationsbetweenpaleologswithinfoureudicotspecies
Sample mixture
Species size components Median Variance Proportion
Deductionofaconsensusgeneorderformultipletaxapermitsus
to directly compare estimates of the ages of gene duplications
Arabidopsis 7435 2 0.86 0.08 0.51
based on rates of nucleotide substitution per synonymous site 2.00 0.20 0.49
(K)betweenpaleologpairs(syntenicparalogs),filteringoutthe Carica 907 1 1.76 0.32 1
s
inevitable influence of background (i.e., single gene) duplica- Populus 13,113 2 0.27 0.01 0.62
1.54 0.24 0.38
tions, which superimpose an L-shaped curve on the relics of
Vitis 2288 1 1.22 0.16 1
whole-genome duplications (Blanc and Wolfe 2004; Cui et al.
Genome Research 5
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Tang et al.
Figure4. (A,B)DistributionofK distancesamongCarica,Populus,Vitis,andArabidopsispaleologs.K valuesaregroupedintobinsof0.1intervals.
s s
Certain K intervals are highlighted as they correspond to several presumed whole-genome duplication events. Dotted lines are fitted mixtures of
s
log-normaldistributionsforthepaleologK distributions(seeMethods).(C)Distributionof4DTVdistanceamongpaleologsinthesamefoureudicot
s
lineages.(D)Phylogenyofsingle-copyorthologsetusedinrelativerateestimates.Atotalof47orthologousgenesthataresinglecopyinallfivespecies
wereusedintheanalysis.ProteinalignmentsforeachorthologgroupwereconstructedandthenusedtoguideDNAalignments.Thealignmentsare
thenconcatenated,with53,856alignednucleotidepositions.Per-siteK valuesoneachbranchwereestimatedbycodemlinthePAMLpackage(Yang
s
1997)usingaconstrainedtopologythatreflectsorganismalrelationships.
which are perennials. However, because life history attributes Thestrikingdifferencesinevolutionaryratesamongthesetaxaat
tend to change over evolutionary time, the generation-time ef- theDNAsequencelevelmay,inpart,explainthecontroversial
fect is not sufficient to explain the rate heterogeneity among placementofVitisinsidetheeurosidsbysomeinvestigators(Jail-
differentorganisms(Gautetal.1996). lonetal.2007).Indeed,wefoundthatifweuseArabidopsisasthe
Because substitution rates vary among lineages, timing of referencepoint,theincreasingK distancesfromCarica,Populus,
s
duplication or speciation events is hard to determine using ge- andVitisappeartosupporttheviewthatVitisisanoutgroupto
netic distance measures alone. For the same reason, dating of therosids(Table4).
ancienteventsbasedonphylogenetictrees(Bowersetal.2003;
Tuskanetal.2006)couldproduceincongruousresultssincethe
Inferringthenumberandarrangementofgenes
drastic differences in rates may lead to incorrect trees that are
intheancestralangiosperm
artifactsbecauseoflong-branchattractions(Felsenstein2004).
One phylogenetic model placed Vitis within the eurosid I Top-downmultiplealignmentsmitigatethefragmentationand
clade(Jaillonetal.2007),incontrastwiththeprevailingviewof decayofancestralgeneorders,improvingourabilitytodeduce
theVitaceaeassistertobotheurosidIandeurosidII(Daviesetal. thenumberandarrangementofgenesinthelastcommonan-
2004;Soltisetal.2005).Indeed,PopulusandVitisdoshowsmall cestor of a group of genomes. When we align gene orders to
K or K values for substitutions between inferred orthologs produce multiple collinear segments, corresponding genes are
a s
(Table 4). However, the seemingly smaller distance between collectedandmergedintoadeduced“ancestrallocus.”Atotalof
PopulusandVitisgenesshouldbeinterpretedwithcautionsince 18,447 deduced ancestral loci (corresponding gene groups) col-
both species appear to have relatively slow evolutionary rates. lectivelyrepresent77,059genesinthefivespecieswestudied(see
6 Genome Research
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Paleopolyploidy and angiosperm genome evolution
Table4. K andK valuesforsyntenicorthologsoffivesequenced Implicationsforparticulareudicotgenefunctionalgroups
s a
plantgenomes
Arabidopsis Carica Populus Vitis Oryza By combining available positional information with sequence
homologies,ourmethodimprovesonotherorthology/paralogy
Arabidopsis — 0.24 0.23 0.25 0.37 mapping algorithms that depend mainly on similarity scores,
Carica 1.57(6913) — 0.17 0.19 0.35 such as OrthoMCL (Li et al. 2003), Inparanoid (O’Brien et al.
Populus 1.64(8366) 1.08(8504) — 0.16 0.31
2005), and the like. Since the clusters are inferred by syntenic
Vitis 1.72(7381) 1.12(7920) 0.98(10,143) — 0.32
alignments,anygenefamilyconstructedbyourmethodcontains
Foreachsyntenicgroup,thesmallestK orK valueamongallortholo- at least two genes. Genes duplicated by single-gene or tandem
s a
gouspairswasretrievedtorepresentthevalue.Thelowertriangleshows duplications do not fall on collinear chains, and thus are ex-
medianKsvalues,andtheuppertriangleshowsmedianKavalues.Num- cludedfromthesyntenicgenegroupsbyouralgorithm.Incon-
bersinbracketscorrespondtothenumberofsyntenicgroupsusedin
trast,sincesomeofthesesmall-scaleduplicationsarerecentand
eachcomparison.K valuesbetweenOryzaandfoureudicotsshowsatu-
s showhighersimilaritiesthanthepaleo-duplicates,theyaremore
ratedsubstitutionsandhighvariancesandthereforeshouldnotbecon-
sideredreliableestimates. easilyincludedbytraditionalhomology-basedclusteringmeth-
ods.
The exponential growth in gene numbers resulting from
Supplemental Data 3 for complete compilation). Among these recurring polyploidies is often tempered by a massive yet pro-
loci, 3680 (20%) are specific to only one species, and 14,767 gressiveamountofgenedeathinthesubsequentdiploidization
(80%)containgenesfromatleasttwodifferentspecies.Westud- process. However, the probability of gene loss is not uniformly
iedthecompositionsofthecross-speciesgroups(Table2).Ifall distributedamongallgenefunctionalgroups(Maereetal.2005).
duplicatesderivedfromeachgenomeduplicationeventhadbeen Convergent restoration of some genes to singleton status after
retained,eachclusteredancestrallocusideallywouldhavethree multipleroundsofduplicationinindependentlineagessuggests
Caricagenes((cid:1)only),threeVitisgenes((cid:1)only),sixPopulusgenes thattheremaybeselectiveadvantagesfortheorganismtohave
((cid:1),p),and12Arabidopsisgenes((cid:1),(cid:3),(cid:2)).Suchextremecaseswere onlysinglecopiesofthesegenes(Patersonetal.2006).However,
not observed. However, we still find cases in which the copy the most extreme cases of “duplication resistance,” gene func-
numbersareclosetosaturationinthesegenomes(Table2),and tional groups for which one and only one copy per nucleus is
twospecificcasesarefurtherdiscussedinthenextsection. adaptive,wouldprovidetoolittleinformationtobeinferredas
Conceptually, the number of cross-species syntenic gene duplication-resistant by previous (cid:6)2-based statistical methods
clusterswouldreflectthegenenumberpriorto(cid:1),byfarthemost (Paterson et al. 2006). Multi-alignment improves our ability to
ancientduplicationdetectedbyourcollinearityalgorithm.Ifwe identifycandidateduplication-resistantgenesthatfallintothis
assumethatmostgenesretaintheirancestralpositions,thenby mostextremecategory,inthatifasinglegeneisalwaysrestored
usingonlythesetofgenesthatshowcross-speciessyntenyand to singleton following a sufficient number of independent du-
correcting for the “inflation” induced by genome duplications, plications,thenduplicationresistanceofthatsinglegenemight
wecanhavearelativelyaccurateestimateofancestralgenenum- beinferred.Suchgeneshavecuriously“resisted”multipledupli-
ber.Thefourfullysequencedeudicotseachyieldslightlydiffer- cationcyclesinmultipleindependentlineages,specificallyone
entestimatesofthisnumber,varyingfrom10,149forCaricato roundofduplication((cid:1))inCarica andVitis,two(p,(cid:1))inPopulus,
13,043 for Populus (Table 2). This range coincides closely with three ((cid:2), (cid:3), (cid:1)) inArabidopsis, and one ((cid:5)) or more in Oryza. In-
previous estimate of ancestral angiosperm gene numbers of deed,somesyntenicgroupshavepreservedexactlyonecopyin
12,000–14,000basedanindependentgenebirthmodel(Stercket theancestrallocationforeachofthefivespecies.Somegenesin
al.2007).Ournumber,however,maybeanunderestimatecon- thesegroupsarenottrue“singletons,”withnon-synteniccopies
sidering that the alignment algorithm does not achieve perfect presentinthegenomebecauseofsinglegeneduplications.After
sensitivity. Moreover, lower contiguity and less progress in an- filteringoutsuchnon-synteniccopies,wefound47strictsingle-
notationmaytendtoreducetheCaricanumber,andappreciable ton groups for five angiosperm genomes preserved in collinear
heterozygosity in the sequenced genotype (resulting in alleles linkage groups, supporting their inferred orthology to one an-
sometimesbeingconsidereddifferentloci)maysomewhatinflate other.Ifweassumethatthediploidizationprocessiscompletely
thePopulusnumber. independent in each of the five species, we can estimate the
In contrast to estimation of ancestral gene number, infer- expectednumberofsingletongroupsbymultiplyingthepropor-
enceofancestralgeneorderisamuchharderproblem,andour tions of singleton genes in each genome by the average gene
computationallyreconstructedgeneordershouldnotbeconsid- number.Underthisestimate,the47singletongroupswefound
ered as truly “ancestral.” In our analysis, the inferred ancestral arenearly10timesmorethantheexpectedfivegroups.Wealso
geneorderwasdeducedbytakingtheconsensusofthealigned found 247 strict singleton groups for only the four eudicot ge-
gene orders among various chromosomes and scaffolds in the nomes(versus20explicablebychance).ThegeneIDsandfunc-
fivespeciesweinvestigated,similartosomepreviousapproaches tional annotations for the singleton groups are available in
(Blanc et al. 2003). Ideally such consensus orders would be re- Supplemental Data 5. Many of the singleton genes have only
quiredtoreflectallthegenearrangementsalignedinthesame putativeclassifications,andthoseofknownfunctionsaremostly
block.However,thesolutionisnotuniqueastheremaybesev- enzymes.
eral possible consensus arrangements under these constraints. The multiplicities in ancestral loci constructed by MCscan
For example, different permutations of interleaving genes be- also revealed extreme cases in which ancestral loci were “dele-
tweenthesyntenicanchorswouldhaveequallikelihoodofbeing tion-resistant,” with a tendency to be preserved in consistently
“ancestral.” In general, the gene groups that have fewer copies high copy numbers in multiple species (Table 5). Since both
mayhavefewerconstraintsintheconsensusarrangementsand CaricaandVitishaveonlyoneroundofduplicationwithmulti-
thereforecannotbepreciselyorderedcomputationally. plicity of 3, while Populus has two rounds of duplications and
Genome Research 7
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Tang et al.
Table5. Thirtyancestrallociselectedbasedonsaturatedpaleologcopynumbersin ofpolyploidyhaveimportantimplications
Carica(≥3copies),Populus(≥5copies),andVitis(≥3copies) for studying plant gene family evolution.
Ancestral Because of less gene loss, such gene fami-
locusID Carica Populus Vitis Arabidopsis Genefamilyb lies show improved power to resolve par-
ticular evolutionary events. Using two an-
N00011 3 6 3 4 cestral loci that are close to each other
N00123 3 6 3 3
in the local ancestral order and highly
N00137 3 6 3 6
N00535 3 6 3 5 GRAStranscriptionfactor saturated with paleo-duplicates, N01482
N00715 3 5 3 3 (C2H2 transcription factor family) and
N01470 3 6 3 8 N01483 (auxin-response protein), we con-
N01482a 3 6 3 7 C2H2transcriptionfactor
structed phylogenetic trees for the gene
N01483a 3 5 3 10
N01501 4 5 3 6 members. Both phylogenetic trees (Fig. 5)
N01504 3 5 3 6 support the coarse partitioning of three
N01732 3 6 3 8 subclades,witheachcladecontainingupto
N01831 3 5 3 3 four Arabidopsis genes, two Populus genes,
N02420 3 5 3 6 C3Htranscriptionfactor
oneCaricagene,andoneVitisgene.These
N02938 3 6 3 3
N03063 3 6 3 6 two examples also support the inference
N03148 3 5 3 3 Phosphatidylinositol-4-kinaseg thatArabidopsisgenesevolvemorequickly
N03158 3 5 3 4 than Vitis genes. This is reflected by the
N03159 3 6 3 5 C2C2-Doftranscriptionfactor
longer branches, that is, more nucleo-
N03285 3 5 3 3 Lateralorganboundariesgene,classII
N03326 3 5 3 3 tide substitutions for Arabidopsis genes
N03658 3 5 3 4 withinindividualsubclades.Indeed,differ-
N03794 3 5 3 4 ential evolutionary rates have some im-
N04406 3 5 3 6 Kinesin-likeproteins
pact on the N01482 tree topology, as one
N05304 3 6 3 4 C3Htranscriptionfactor
N05519 3 6 3 3 EF-handcontainingproteins:GroupIV Vitis gene (Vv4g1235) appears to be even
N05829 3 5 3 5 MADStranscriptionfactor closertooneofits(cid:1)paleologs(Vv18g1188)
N05866 3 6 4 6 AP2-EREBPtranscriptionfactor than to its orthologs in the three other
N06369 3 5 3 5 C2C2-Gatatranscriptionfactor
species. One possible alternative explana-
N07685 3 6 3 3 Lateralorganboundariesgene,classII
N07692 3 6 3 3 Corecellcyclegenes tion is that these two Vitis genes have
undergone homogenization, as has been
TheancestrallocireferenceIDsareavailableinSupplementalData3. shown to occur in some paleo-dupli-
aUsedinthephylogeneticreconstructioninFigure5. cated genes in Oryza genome (Wang et al.
bBasedoncuratedArabidopsisgenefamiliesfromTAIR(Hualaetal.2001). 2007).
Methods
multiplicityof6,wespecificallyselectedtheancestrallocithat
aresaturatedwithpaleologsforthesethreespecies,requiringthat
the groups that we chose have Carica multiplicity (cid:7)3, Populus Genesetandsequencehomologysearch
multiplicity(cid:7)5,andVitismultiplicity(cid:7)3atthesametime(Table Protein sequences from Arabidopsis, Carica, Populus, Vitis, and
5).WesetthesecopynumbercutoffsbecauseCarica,Populus,and Oryzagenomeannotationswereused(Table1).Afewannotated
Vitishaveexpectedcopynumbersof3,6,and3respectively,and moss (Physcomitrella patens) genes (JGI annotation version 1.1)
we slightly loosened the Populus cutoff to look at more groups werealsousedastheoutgroupingenetreeanalysis.Carica,Popu-
thatareclosetosaturation.Atotalof30suchgroupswerefound lus,andVitisgenenameswererenamedaccordingtotheirincre-
(Table 5). Considering that very few groups have exceeded the mental position on the chromosomes or scaffolds (see Supple-
thresholdforeachspecies(Table2),thechancethat30random mentalData4foraconversiontabletooriginalgeneidentifiers).
groupssatisfyallthreethresholdsisalmostnon-existent((cid:6)2-test, Incasetheoriginalgeneidentifiersaresubjecttofuturechanges,
P=2.2(cid:1)10(cid:2)16). theconversiontablewillbeupdatedaccordinglytoensureeasy
translation.Ifagenehadmorethanonetranscript,onlythefirst
In contrast to “duplication-resistant” genes, many “dele-
transcript in the annotation was considered. Each genome was
tion-resistant”lociofknownfunctionaretranscriptionfactors,
compared against itself and other genomes using BLASTP
consistentwithpreviousfindingsthattranscriptionalregulators
(Altschul et al. 1990), retrieving the best five hits meeting an
are significantly over-retained in WGD duplicates (Seoighe and E-valuethresholdof1(cid:1)10(cid:2)5.
Gehring2004;FreelingandThomas2006).Forexample,N05829
containsfiveArabidopsisMADS-boxgenes(AGL14,AGL19,SOC1, Pairwisegeneorderalignments
AGL42, AGL72), all descended from a single ancestral pre-(cid:1)
Thesyntenicregionsweregroupedtoformmultiplealignments
MADS-box gene. N03285 (contains Arabidopsis genes LBD40,
usinganovelalgorithmMCscan(multiplecollinearityscan).We
LBD41,LBD42)andN07685(containsArabidopsisgenesLBD37, first took whole-genome BLASTP results and computed strictly
LBD38,LBD39)collectivelycompriseallsixclassIIlateralorgan collinear segments for all possible pairs of chromosomes and
boundaries (LOB) gene family members characterized to date scaffolds.Apairwisealignmentprocedurewasimplementedus-
(Shuaietal.2002),whichweinfertotracetotwoancestral(pre-(cid:1)) ing an empirical scoring scheme similar to that of Haas et al.
LOBclassIIgenes. (2004).Thedefaultscoringscheme(configurable)ismin(log E,
10
Comparative analysis for genes derived from “deletion- 50)matchscoreforonegenepair,and(cid:2)1gappenaltyforeach
resistant”locithathavelargelyexpandedfollowingeachround 10-kb distance between any two consecutive gene pairs. The
8 Genome Research
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Paleopolyploidy and angiosperm genome evolution
Figure5. PhylogeneticanalysisofancestrallociN01482(A)andN01483(B).Codingsequencesofallmembersinfoureudicotspeciesforeach
ancestrallocus(19genesinN01482,21inN01483)werealignedbyCLUSTALW(Thompsonetal.1994)usingparameterssuggestedbyHall(2007).
Phylogenetic relationships among the members and sequences were grouped into clades using MrBAYES (Ronquist and Huelsenbeck 2003). The
Bayesiananalysiswascarriedoutfor500,000generationsusingtheGeneralTimeReversibleplusGamma(GTR+G)substitutionmodelselectedbased
onMODELTEST(PosadaandCrandall1998).Allbrancheswithsupport<50%arecollapsedintoapolytomy.Amajoritytreewaspresentedinbothcases.
ThegenenamesforCarica,Populus,andVitisarerecodedtoreflectrelativeordersonchromosomeorscaffold(seeMethods).Theconversionsfromthe
originallocusidentifierstothere-indexedgenenamesareavailableasaconversiontableinSupplementaryData4.Incasetheoriginalgeneidentifiers
aresubjecttofuturechanges,theconversiontablewillbeupdatedaccordingly.ArabidopsisgenenamesfollowtheirstandardTAIRlocusIDs.Scalebars
representthenumberofsubstitutionspersitefollowingtheGTR+Gmodel.
scoreforeachpairwisecollinearchainisthencalculatedviady- multiplies by two since there are two possible orientation con-
namic programming through the following recurrence condi- figurationsbetweentwocollinearsegments.Thisisonlyanap-
tion, assuming that two gene pairs, u and v, are on the path proximation to a more rigorous yet computationally expensive
whereuprecedesv, permutationtest(VandePeer2004)andMonteCarlomethods
(Hampsonetal.2005);however,computationalexperimentsand
ChainScore(cid:1)v(cid:2)=MatchScore(cid:1)v(cid:2)+max{ChainScore(cid:1)u(cid:2)
u analytical results (Wang et al. 2006) suggest that this gives a
+GapPenalty(cid:1)u,v(cid:2),0} reasonable estimate for the significance of the syntenic blocks.
All the pairwise alignments that we reported are significant at
Tandem matches <50 kb apart are collapsed using a repre- E<1(cid:1) 10(cid:2)10.
sentativepairthathasthesmallestBLASTPE-value.Thisthresh-
old,indeed,didnotpurgealltandems—westillfoundaveryfew Multiplegeneorderalignments
long-distancetandemsinourclusteredancestralloci—however,
Pairwisesyntenicmatcheswereclusteredintomulti-wayanchors
thisisreasonabletrade-offsinceincreasingthethresholdwould
through a Markov clustering algorithm MCL (Enright et al.
remove some of the intra-chromosomal WGD duplicates. All
2002),inordertosimplifythecorrespondencesamongmultiple
pairwisesegmentswithscoresabove300arereported.Eachpair-
loci.Multiplechromosomalregionsthreadedbyconsecutivean-
wise segment consists of two distinct genomic locations with
cestrallociarerecoveredandalignedusingaheuristicthatcon-
aligned,collineargenesasanchors.
structs the multiple alignments progressively by aligning one
Theexpectednumberofoccurrencesofapairwisecollinear-
closest-related region at a time by dynamic programming. We
itypatterncouldbeestimatedwiththefollowing,similartothe
then use a reference genome to report all the multiple blocks.
oneusedinWangetal.(2006),
(cid:1) (cid:2) Noticethatwhenweusea“reference”asthebasis,welosesym-
m−1 l l metry.Forexample,letusassumeA-B-Casamultiplealignment,
E=2Pm (cid:3) 1i(cid:4) 2i ,
N i=1 L1 L2 formedbysyntenicregionsA,B,andC.Ifweallowtheblocksto
be threaded by A, B, or C, we can find this block three times;
where N is the number of matching gene pairs (by BLASTP or however, the resulting multiple alignment may be slightly dif-
BLAT, etc.) between two chromosomal regions defined by the ferent because of the order in which we stack A, B, and C. We
syntenic block; m is the number of collinear gene pairs in the found that the “once a gap, always a gap” rule applies to the
identifiedblock;L andL arerespectivelengthsofthetwochro- multiplealignmentofgeneorders,inthattheorderofprogres-
1 2
mosomalregions;andl andl aredistancesbetweentwoadja- sive stacking does affect the resulting alignment. Therefore, we
1i 2i
centcollineargenepairsinthesyntenicblock.Theexpectation implementarefinementproceduretoamelioratesucheffectby
Genome Research 9
www.genome.org
Downloaded from genome.cshlp.org on March 16, 2023 - Published by Cold Spring Harbor Laboratory Press
Tang et al.
iteratively realigning each segment, allowing the falsely placed lated4DTVvaluesbetweengenepairsusingin-housePerlscripts.
gapstobecorrectedandfurtheroptimizethegapplacement. 4DTV values are calculated for gene pairs having (cid:7)10 fourfold
degenerate sites. Fourfold degenerate sites are codons of amino
Clusteringthemultiplyalignedgenomicregions acidresiduesG,A,T,P,V,andR,S,L.Raw4DTVvaluesarethen
Ifweconsider“generetentionattheancestrallocus”asthean- correctedforpossiblemultipletransversionsatthesamesiteus-
cestralstateand“geneloss”asderived,theneachalignedchro- ingthisformula:
mosomalsegmentcanbedescribedasavectorofbinarycharac- 4DTV =−1(cid:4)2×ln(cid:1)1−2×4DTV (cid:2).
ters. We could then search for hierarchical clustering based on corrected uncorrected
“Camin-Sokal parsimony” since genes that had been lost are FinitemixturemodelsofgenomeduplicationsbasedonK
s
highly unlikely to re-emerge at original paleologous locations, distribution
that is, reversal to the ancestral state is prohibited (Camin and
Sokal1965).Usingthissimplisticparsimonyprinciple,syntenic TheactualdistributionofKsbetweenpaleologscanbemodeled
genomicregionsinmultiplealignmentblockscanbeclustered, asmixturesoflog-transformedexponentialsandnormals,repre-
using the “mix” program in the PHYLIP package (Retief 2000) senting single gene duplications and whole genome duplica-
with0/1-codedchromosomalregionswithineachblockasinput. tions, respectively. Since we have identified the paralogs that
show segmental correspondence with most of the single gene
MCscanimplementationandavailability duplicationsexcluded,theactualdistributionscanbedescribed
as mixtures of log-normal components that represent multiple
Themulti-alignedplantgeneordersandimplementedalgorithm
rounds of genome duplications, using the EMMIX software
and C++ source codes are publicly available (http://chibba.
(http://www.maths.uq.edu.au/∼gjm/emmix/emmix.html). K
agtec.uga.edu/duplication/mcscan/).Theprogramusesonlytwo s
valuesthatare<0.005werediscardedtoavoidfittingacompo-
inputfiles—afilecontainingBLASTPresultsandafiledescribing
nenttoinfinity(Cuietal.2006),andthemixedpopulationswere
gene coordinates—and outputs both pairwise syntenic blocks
modeledwithonetofivecomponents.Weselectedonebestmix-
and the multi-aligned gene orders threaded by a reference ge-
turemodelforeachpaleologdistributiononthebasisofBayesian
nome.Thereareseveralparameterstoconfigureaccordingtothe
informationcriterion(BIC)andanadditionalrestrictiononthe
user’s need. For example, the significance cutoff would reduce
mean/variancestructureforK (Cuietal.2006).
sensitivity but increase specificity for the uncovered syntenic s
blocks.
Acknowledgments
ComparisonbetweenVitisandSolanum,Musa
WeappreciatefinancialsupportfromtheU.S.NationalScience
ForSolanum,wedownloaded195-ntsequencesfortomato(Sola-
Foundation(MCB-0450260toA.H.P.andJ.E.B.,DBI-0421803to
numlycopersicum)fromNCBI(September2007)thatwere(cid:7)100
R.M.andA.H.P.),theUniversityofHawaiitoM.A.,andtheU.S.
kb,discardingonechloroplastsequencefromanalysis,foratotal
Department of Defense W81XWH0520013 to M.A. We thank
of 25 Mb (representing ∼2.5% of the tomato genome). We re-
GuojunLiforhelpfuldiscussionsonthesyntenydeductional-
trieved53,792TIGRSolanumunigenes(S.lycopersicumTIGRtran-
gorithm.
scriptassemblyversion5),mappingthemtothecollectedBACs
(BLASTN E-value<1(cid:1)10(cid:2)6) and took the best hit that had
(cid:7)200-bpalignmentlengthand97%identity.Thisshouldaccom- References
modateminorsequencingerrorsorcultivardifferencesbetween
theESTsandBACs,ifany.Ifmultipleunigeneswentwithin300 Altschul,S.F.,Gish,W.,Miller,W.,Myers,E.W.,andLipman,D.J.1990.
bp on the tomato sequence, only the longest hit was retained. Basiclocalalignmentsearchtool.J.Mol.Biol.215:403–410.
ArabidopsisGenomeInitiative.2000.Analysisofthegenomesequence
This was to resolve cases in which the unigenes were not as-
ofthefloweringplantArabidopsisthaliana.Nature408:796–815.
sembledcompletelyorcorrectlyforageneandtherealgenewas Aury,J.M.,Jaillon,O.,Duret,L.,Noel,B.,Jubin,C.,Porcel,B.M.,
representedbymorethanoneunigene.Atotalof2243Solanum Segurens,B.,Daubin,V.,Anthouard,V.,Aiach,N.,etal.2006.
unigenes, 4.2% of the total, were anchored to BACs. Solanum Globaltrendsofwhole-genomeduplicationsrevealedbytheciliate
Parameciumtetraurelia.Nature444:171–178.
unigeneswereassignedtheirbase-pairlocationswithintheBACs,
Blanc,G.andWolfe,K.H.2004.Widespreadpaleopolyploidyinmodel
andweusedthesemappedunigenesastentativegenemodelson plantspeciesinferredfromagedistributionsofduplicategenes.Plant
theseSolanumBACs.Themappedunigeneswerethensearched Cell16:1667–1678.
for homology against the Vitis proteins using BLASTX Blanc,G.,Hokamp,K.,andWolfe,K.H.2003.Arecentpolyploidy
(E<1(cid:1) 10(cid:2)5). We analyzed synteny of Vitis chromosomal re- superimposedonolderlarge-scaleduplicationsintheArabidopsis
genome.GenomeRes.13:137–144.
gionsand17banana(Musaacuminata)BACsinasimilarproce- Blanchette,M.,Kent,W.J.,Riemer,C.,Elnitski,L.,Smit,A.F.,Roskin,
dure. K.M.,Baertsch,R.,Rosenbloom,K.,Clawson,H.,Green,E.D.,etal.
2004.Aligningmultiplegenomicsequenceswiththethreaded
Synonymoussubstitution(K)andfourfolddegeneratesite blocksetaligner.GenomeRes.14:708–715.
s Bowers,J.E.,Chapman,B.A.,Rong,J.,andPaterson,A.H.2003.
transversion(4DTV)calculation Unravellingangiospermgenomeevolutionbyphylogeneticanalysis
ofchromosomalduplicationevents.Nature422:433–438.
For each pair of homologs, we aligned their protein sequences
Camin,J.H.andSokal,R.R.1965.Amethodfordeducingbranching
using CLUSTALW (Thompson et al. 1994) and converted the sequencesinphylogeny.EvolutionInt.J.Org.Evolution19:311–326.
proteinalignmenttoDNAalignmentusingPAL2NAL(Suyamaet Cui,L.,Wall,P.K.,Leebens-Mack,J.H.,Lindsay,B.G.,Soltis,D.E.,Doyle,
al. 2006). Some homologous genes could not produce reliable J.J.,Soltis,P.S.,Carlson,J.E.,Arumuganathan,K.,Barakat,A.,etal.
2006.Widespreadgenomeduplicationsthroughoutthehistoryof
CLUSTALW alignment for various reasons and were discarded
floweringplants.GenomeRes.16:738–749.
from further analysis. Ks values were calculated using the Nei- Davies,T.J.,Barraclough,T.G.,Chase,M.W.,Soltis,P.S.,Soltis,D.E.,and
Gojoborialgorithm(NeiandGojobori1986)implementedinthe Savolainen,V.2004.Darwin’sabominablemystery:Insightsfroma
PAMLpackage(Yang1997).WerepeatedtheK calculationusing supertreeoftheangiosperms.Proc.Natl.Acad.Sci.101:1904–1909.
s Enright,A.J.,VanDongen,S.,andOuzounis,C.A.2002.Anefficient
other algorithms and found that the differences are small, sys-
algorithmforlarge-scaledetectionofproteinfamilies.NucleicAcids
tematic biases that do not affect major conclusions. We calcu- Res.30:1575–1584.
10 Genome Research
www.genome.org