Theor Appl Genet (2006) 113:931–942

DOI 10.1007/s00122-006-0352-9

ORIGINAL PAPER

Inheritance of inXorescence architecture in sorghum

P. J. Brown · P. E. Klein · E. Bortiri · C. B. Acharya · W. L. Rooney · S. Kresovich

Received: 6 February 2006 / Accepted: 21 June 2006 / Published online: 18 July 2006© Springer-Verlag 2006

Abstract The grass inXorescence is the primary foodsource for humanity, and has been repeatedly shapedby human selection during the domestication of diVer-ent cereal crops. Of all major cultivated cereals, sor-ghum [Sorghum bicolor (L.) Moench] shows the moststriking variation in inXorescence architecture traitssuch as branch number and branch length, but thegenetic basis of this variation is little understood. To

study the inheritance of inXorescence architecture insorghum, 119 recombinant inbred lines from an elite byexotic cross were grown in three environments andmeasured for 15 traits, including primary, secondary,and tertiary inXorescence branching. Eight character-ized genes that are known to control inXorescencearchitecture in maize (Zea mays L.) and other grasseswere mapped in sorghum. Two of these candidategenes, Dw3 and the sorghum ortholog of ramosa2, co-localized precisely with QTL of large eVect for relevanttraits. These results demonstrate the feasibility of usinggenomic and mutant resources from maize and rice(Oryza sativa L.) to investigate the inheritance of com-plex traits in related cereals.

Keywords InXorescence architecture · Sorghum · Quantitative trait locus · Candidate gene

Introduction

With 10,000 species, the grass family (Poaceae) is one ofthe most species-rich groups of Angiosperms. Poaceaeincludes major cereal crops such as rice, wheat (Triticumaestivum L.), and maize, which, together with other cere-als, provide over 70% of the food consumed by humans.Grasses bear grains in spikelets, a key reproductive inno-vation consisting of a short shoot with two bracts calledglumes subtending a variable number of reduced Xow-ers. Spikelets are borne on the main inXorescence rachisor on branches formed oV the main rachis, in a patternthat is highly variable among grass species. Thus, thegenetic control of the number and growth of branches,or inXorescence architecture, is very important in thisfamily because it directly aVects grain yield.

Communicated by R. Bernardo

Electronic Supplementary Material Supplementary material is available to authorised users in the online version of this article at http://dx.doi.org/10.1007/s00122-006-0352-9.

P. J. Brown (&) · C. B. Acharya · S. KresovichInstitute for Genomic Diversity, Cornell University, 158 Biotechnology Building, Ithaca, NY 14853, USAe-mail: pjb34@cornell.edu

P. J. Brown · S. KresovichDepartment of Plant Biology, Cornell University, Ithaca, NY 14853, USA

P. E. KleinInstitute for Plant Genomics and Biotechnology, Texas A&M University, College Station, TX 77843, USA

P. E. KleinDepartment of Horticulture, Texas A&M University, College Station, TX 77843, USA

E. BortiriPlant Gene Expression Center, Department of Plant and Microbial Biology, University of California-Berkeley, Albany, CA 94710, USA

W. L. RooneyDepartment of Soil and Crop Sciences, Texas A&M University, College Station, TX 77843, USA

123

932 Theor Appl Genet (2006) 113:931–942

Maize and rice are two of the leading model systemsfor grass research, and inXorescence mutants fromthese species have been used to characterize a numberof genes involved in the control of grass inXorescencearchitecture (Bommert et al. 2005). Some of thesegenes also appear to aVect quantitative variation ininXorescence traits (Upadyayula et al. 2005) and theirexploitation for cereal improvement is just beginning(Ashikari et al. 2005). Robertson (1989) was an earlyproponent of the idea that the genes underlying quanti-tative and mutant/qualitative variation in a speciesmight be the same. By extension, mutants from onespecies might also be used to identify candidate genesin related species. Today, this logic underlies anapproach which has met with considerable success inidentifying genes involved in agronomic traits in thegrasses, including maize Xowering time (Thornsberryet al. 2001), wheat vernalization (Trevaskis et al. 2003),sorghum photoperiod sensitivity (Childs et al. 1997),sorghum lignin composition (Bout and Vermerris2003), and inXorescence architecture in both maize andrice (Gallavotti et al. 2004; Komatsu et al. 2003a).There are many reasons why a given candidate genemight not aVect a quantitative trait. The expression offunctional variation at a candidate locus could bemasked by alleles at epistatic loci (Lauter and Doebley2002), or the functional variation itself might havebeen lost due to selection or drift (Clark et al. 2004).However, the co-localization of candidate genes andquantitative trait loci (QTL) for relevant traits pro-vides a means of prioritizing speciWc genomic regionsfor Wne-mapping and/or association mapping (Salviand Tuberosa 2005).

Sorghum is a tropical C4 grass with a relatively com-pact genome of 736 MB which will soon be sequenced(http://www.jgi.doe.gov), and is the world’s Wfth mostimportant cereal crop, with extensive subsistence use insome of the world’s poorest countries in sub-SaharanAfrica. Although sorghum itself has few characterizedgenes or mutant lines, it is only 11.9 million yearsdiverged from maize (Swigonova et al. 2004), whichboasts an exceptional collection of inXorescencemutants. Sorghum inXorescences, called panicles, aremore extensively branched than those of maize or rice,and there is tremendous variation within sorghum bothin the degree of inXorescence branching and in thedegree of rachis and branch elongation. Variation inpanicle morphology is a primary factor in the sub-division of domesticated sorghum into Wve major races:bicolor, caudatum, kaWr, durra, and guinea (Harlanand De Wet 1972).

The objective of this study was to investigate theinheritance of sorghum inXorescence architecture by

determining trait heritabilies, correlations betweentraits, and identifying QTL of large eVect. Weemployed an integrative approach similar to that usedpreviously with foxtail millet (Doust et al. 2005) toassess the possible contribution of known grass inXo-rescence genes to genetic variation in sorghum inXores-cence morphology. Eight candidate genes from maizeand other grasses were mapped, along with inXores-cence QTL, in a population of sorghum recombinantinbred lines (RILs) that shows extensive variation ininXorescence architecture. The co-localization of can-didates and QTL in this study provides a starting pointfor isolating genes with quantitative eVects on inXores-cence architecture in the grasses.

Materials and methods

Plant materials and growing conditions

A series of RILs were developed from the cross(BTx623 £ IS3620C) during the creation of a high-density molecular map of sorghum (Menz et al. 2002).In this study, we grew and phenotyped 119 of theselines at the F9–F10 stage. The female RIL parent,BTx623, is a subtropically-adapted parental line with astout rachis, numerous short primary branches, exten-sive secondary and tertiary branching, and a copiousnumber of large seeds. The male RIL parent, IS3620C,is an exotic accession with a thin rachis and long pri-mary branches that have reduced secondary and ter-tiary branching, and a reduced number of small seeds.The BTx623 panicle has a compact overall appearance,whereas the IS3620C panicle is often described as“open” or “loose”. Although most elite sorghum varie-ties have relatively compact panicles and are highyielding, the eVect of panicle architecture on yieldcomponents has not been adequately addressed.IS3620C is a “converted” line, meaning that the origi-nal line from which it was derived was crossed to atemperately-adapted line and then backcrossed to theoriginal exotic while selecting for short stature andphotoperiod-insensitivity (Stephens et al. 1967). RILsand parent lines were planted in three locations (Wes-laco, College Station, and Halfway, TX, USA) in 2004.A randomized complete block design was used, withtwo replicates per location and 2 rows of each parentper replicate, for a total of 246 rows per location. TheWeslaco location (26°N, 98 °W) was planted in Febru-ary, harvested in June, received 36.1 cm of rainfall andexperienced an average daily temperature of 23.6°C.The College Station location (30.5 °N, 96 °W) wasplanted in April, harvested in August, and received

123

Theor Appl Genet (2006) 113:931–942 933

72.6 cm of rainfall with an average daily temperature of25.6°C. The Halfway location (34 °N, 101.5 °W) wasplanted in May, harvested in September, and received39.1 cm of rainfall with an average daily temperature of22.9°C (http://www.ncdc.noaa.gov).

Phenotyping of panicle traits

Mature panicles were harvested and dried before mea-surement of panicle traits. Rachis diameter was mea-sured with calipers at the bottom-most branch of thepanicle. Primary branches were removed individuallyand counted, and rachis length was measured as the dis-tance from the bottom-most branch to the end of theprimary axis. Next, two branches were randomlyselected from the “long-branch zone” in the bottomthird of each panicle. On each of two long branches, wemeasured branch length, secondary branch number,and tertiary branch number. Tertiary branch numberwas measured as the number of secondary branchescontaining tertiary branches. The trait percent tertiarybranching was calculated as (number of secondarybranches with tertiary branching/total number of sec-ondary branches) £ 100. A “branch” is deWned as anaxis bearing two or more spikelet pairs; an axis bearinga single spikelet pair is considered a pedicel (Ikeda et al.2004). Thus, for the bottom-most branch in Fig. 1, sec-ondary branch number is 10, tertiary branch number is6, and percent of tertiary branching is 60. After under-going branch abortion in the shaded area, the samepanicle branch would yield a secondary branch numberof 5, a tertiary branch number of 1, and a percent of ter-tiary branching of 20. All panicle traits described weremeasured in three to Wve panicles per row, and rowmeans were used for quantitative analysis. For traitsmeasured in two branches per panicle, means were cal-culated for each panicle prior to calculating a row mean.Traits evaluated on a per row basis included plantheight, peduncle length (the distance from the Xag leafto the lowest inXorescence branch), Xowering time,panicles per row, seed set, 100-seed weight, seeds perpanicle, and total seed weight. To assess 100-seedweight, seeds per panicle, and total seed weight, 5–8panicles were threshed from each row. Seeds per pani-cle was calculated as (seed weight per panicle/100-seedweight) £ 100. Total seed weight was calculated as(panicles per row) £ (seed weight per panicle). Alltraits were measured in three environments, with theexception of Xowering time, which was not measured inHalfway, and the seed traits (100-seed weight, seeds perpanicle, seed set, and total seed weight) which were notmeasured in College Station because of damage fromsorghum midge (Contarinia sorghicola).

Scanning electron microscopy

InXorescence meristems were dissected from 4 to8 week old, greenhouse-grown plants, washed over-night in FAA (45% EtOH, 5% acetic acid, 10% forma-lin), and transferred to 95% EtOH. Fixed meristemswere critical-point dried, mounted, sputter-coated, andphotographed using a Hitachi S4700 Weld emissionSEM at the Plant Gene Expression Center in Albany,CA. USA

Quantitative analysis

Trait variances were partitioned using the randomeVects ANOVA model y = � + E + G + G £ E + error,where E represents environment, G represents geno-type, and G £ E represents the genotype by environ-ment interaction. The error term includes the variancebetween row means for the two replicates of eachgenotype at each location. Heritability estimates foreach trait were calculated as (�2

G)/(�2G + �2

G £ E/L + �2

error/LR), where L and R represent the numberof locations and the number of replicates per location,respectively. ANCOVA was used to partition the phe-notypic covariance between traits into genetic and

Fig. 1 InXorescence branching in sorghum. Spikelet pairs may beborne on primary, secondary, or tertiary branches, or directly onthe rachis, as indicated. The shaded triangle indicates the regionof branch abortion observed in IS3620C

on tertiary branches

on secondary branches

on primary brancheson the rachis

Spikelet pairs:

Region ofbranch abortion

123

934 Theor Appl Genet (2006) 113:931–942

environmental components. Genetic correlations werecalculated between traits as (covT1T2)/q(varT1varT2),where T1 and T2 represent the traits being compared,and the numerator and denominator represent thegenetic components from the ANCOVA and ANOVAanalyzes, respectively. All statistical analyzes were per-formed using R 2.0.1 (http://www.r-project.org).

QTL analysis

The RILs used in this study have been genotyped atthe F6 stage for » 3,000 AFLP, RFLP, and SSR mark-ers (Menz et al. 2002), of which we used 396 frame-work markers to perform QTL analysis. The complete(BTx623 £ IS3620C) linkage map, including theframework markers used in this study, is available on-line at http://www.sorgblast2.tamu.edu. Compositeinterval mapping (CIM) was performed using Win-QTLCart version 2.5 (Wang et al. 2005). GenotypiccoeYcients from the ANOVA models were used toperform CIM analysis on each trait across all threeenvironments. Seed set was included as a covariate inthe models for both seeds per panicle and total seedweight. Cofactors for all analyzes were selected bystepwise regression using the forward selection, back-ward elimination method with a cutoV of 0.05 forcofactor selection and elimination. A window size of10 cM and a walk speed of 0.5 cM were used for allanalyzes. SigniWcance was established by performing1,000 permutations and setting a highly stringent cut-oV of � = 0.001.

Candidate gene analysis

DNA was extracted from leaf tissue pooled from 8 to10 plants per RIL using a modiWcation of the CTABprotocol of Doyle and Doyle (1987). Genomic contigsfor inXorescence candidate genes from Zea, Oryza,and Lolium were assembled, in most cases frompublicly-available sequences, and BLAST analysis wasperformed against the SAMI database (SorghumAssembled genoMic Islands, http://www.magi.plantgenomics.iastate.edu), which is an assembly of methyla-tion-Wltered GSS reads from Orion Genomics (Bedellet al. 2005). Primers were designed manually in regionsof the resulting gene contigs that showed high similar-ity across all species and/or all copies of the gene withina given species. Multiple fragments from each genecontig were treated with exonuclease I and shrimpalkaline phosphatase and sequenced on an ABI3730 tosearch for polymorphism between the mapping par-ents. Wherever possible, RIL genotypes at candidateloci were scored using indel polymorphisms assayed on

either an ABI3730 or on agarose gels. If only SNPswere found between the mapping parents, RILs weregenotyped using the derived cleaved ampliWed poly-morphic sequences (dCAPS) technique (Michaels andAmasino 1998). Genotyped candidates were placed onthe (BTx623 £ IS3620C) map with MapPop (Brownand Vision 2000).

Results

Anatomy of the sorghum inXorescence

Sorghum inXorescences show primary, secondary, andtertiary branching, with primary branches becomingprogressively shorter and less branched towards thetop of the panicle (Fig. 1). In IS3620C, the lower pri-mary branches often bear spikelet pairs only at theirtips due to the abortion of basal secondary and tertiarybranches, a phenomenon known to sorghum breedersas “blasting”. The eVect of branch abortion on paniclearchitecture is clearly evident when comparing scan-ning electron microscopy photographs of immatureIS3620C inXorescences with mature IS3620C panicles(Fig. 2). Whereas the scanning electron microscopyphotos reveal extensive secondary and tertiary branch-ing in the lower primary branches, mature panicleshave empty nodes or short aborted branches withoutspikelets on the lower portion of the corresponding pri-mary branches. At Xowering time, many of thesebranches are still visible as white, stunted tissue thatfalls oV as the panicle dries and matures.

Quantitative genetic analysis of panicle traits

Genotype, environment, and genotype by environ-ment interactions contributed signiWcantly to the vari-ance of all 15 traits. Means, standard deviations, andranges of the traits measured are provided as supple-mental data (Table S1). BTx623 panicles consistentlydisplay more primary, secondary, and tertiary branch-ing than IS3620C panicles. Branch length is shorter inBTx623, although rachis length is similar between thetwo parents. BTx623 has heavier seeds, increased seednumber, and higher total seed weight, while IS3620Cbears many more panicles per row. BTx623 andIS3620C have similar values for plant height and Xow-ering time across all three environments. However,transgressive segregation is observed for all traits inthe RILs and is especially pronounced for height andXowering time.

With regard to traits involved in elongation, herit-abilities are much higher for plant height, rachis length,

123

Theor Appl Genet (2006) 113:931–942 935

and branch length than for peduncle length (Table 1).With regard to inXorescence branching, heritability ismuch higher for primary branch number than for sec-ondary and tertiary branching traits. Flowering timeand seed-related traits have much higher G £ E vari-ance components than the other traits.

A number of trait pairs show signiWcant genetic cor-relations (Table 2), indicating that their inheritance islinked either functionally (pleiotropy) or physically(linkage disequilibrium). Rachis length and branchlength are highly correlated with each other, but show

non-signiWcant or negative correlations with plant heightand peduncle length, Similarly, the genetic correlationbetween primary and secondary branch number is esti-mated at only 0.34, and the correlation between primaryand tertiary branch number is not signiWcant. Seeds perpanicle is strongly and positively correlated with allthree orders of branch number. Much of the variance fortotal seed weight in this population is due to diVerencesin seed set [ra(Total seed weight:Seed set) = 0.62]. Rachis diam-eter shows signiWcant positive correlations with all threeorders of branch number. When rachis diameter was

Fig. 2 Branch abortion in IS3620C. White arrowheads indicateanalogous secondary branches in each frame. i SEM image of animmature IS3620C panicle, with proliWc secondary and tertiarybranching on the lower primary branches; scale bar 0.1 mm. iiMature IS3620C panicle, with a “loose” overall appearance pro-

duced in part by sparse spikelet production on the lower primarybranches; scale bar 1 cm. iii Primary branches from the lower por-tion of a mature IS3620C panicle. The lower secondary brancheson each primary branch have aborted without producing spik-elets; scale bar 1 cm

i ii iii

Table 1 Quantitative genetic statistics for sorghum inXorescence traits

Trait type Trait name Variance components(% of total �2

P)Heritability(h2)

�2G �2

E �2G £ E �2

Error

Branching Primary branch number 71 9 9 11 0.94Secondary branch number 33 2 13 52 0.72Tertiary branch number 44 2 11 43 0.80% tertiary branching 45 8 10 37 0.83

Elongation Plant height 78 6 8 8 0.94Peduncle length 46 11 9 34 0.84Rachis length 70 6 8 16 0.93Branch length 62 12 4 22 0.93

Seed 100-seed weight 56 9 22 13 0.85Seeds per panicle 49 8 22 21 0.82Seed set 20 30 31 19 0.60Total seed weight 14 44 25 17 0.56

Miscellaneous Rachis diameter 55 11 4 30 0.90Flowering time 65 5 19 11 0.89Panicles per row 20 41 3 36 0.74

123

936 Theor Appl Genet (2006) 113:931–942

included as a covariate in the linear models forprimary, secondary, and tertiary branch number, itaccounted for 25, 15, and 6% of the phenotypic vari-ance, respectively.

Detection of QTL

Nineteen QTL, each accounting for 8.5–38.4% of thephenotypic variance for a given trait, were detected atP < 0.001 (Fig. 3; Table S2). Of the nineteen QTL,twelve are for traits that show clear diVerencesbetween the two parents, such as primary branch num-ber, branch length, and 100-seed weight. Of these 12,nine show eVects in the “expected” direction, with theBTx623 allele contributing to increased branching orseed weight, for example.

Genomic organization of QTL

Several regions of the genome display clustering ofQTL with large eVects (Fig. 3). A cluster on the longarm of chromosome 7 includes QTL for plant height,rachis length, and branch length with peaks within afew cM of each other. The closest marker to these QTLpeaks is the Dw3 gene.

A cluster on chromosome 6 includes QTL for plantheight, rachis diameter, and primary branch number.The latter two QTL have large eVects in the oppositedirection of what would be expected based on parentalphenotypes: the BTx623 allele is associated with

reduced primary branching and reduced rachis diame-ter, even though BTx623 has more primary branchesand its rachis is almost twice as thick as that of IS3620C(Fig. 3).

A cluster on chromosome 3 includes QTL for plantheight, branch length, and primary branch number.The closest marker to the primary branch numberQTL is the Sb-ra2 gene.

Candidate gene mapping

Map locations for all eight candidate genes (Table 3;Fig. 3) agree well with current models of the sorghumgenome’s colinearity with maize and rice (Devos 2005).Candidate gene map locations could not have beenpredicted unambiguously using colinearity alone, dueto the complex history of duplications and deletionsbetween grass genomes (Gaut 2002). The primer pairsdesigned ampliWed single bands for every candidategene with the exception of Sb-ids1. The original prim-ers used to amplify Sb-ids1 ampliWed a single band thatyielded clean sequence data, but subsequent primersdesigned for mapping detected a second, Sb-ids1-likelocus on the short arm of chromosome 10 (data notshown). One of the candidate genes, Sb-ba1, showedno polymorphism between the mapping parents over a2.1 kb contig which included the entire coding regionand approximately 1.4 kb of upstream sequence. Sincethis was unusual given the genetic distance betweenthese lines, we sequenced the corresponding fragments

Table 2 Genetic correlations (rA) between traits

SigniWcant correlations are shown in bold

Prim

ary

bran

ch n

umbe

r

Primary branch number Seco

ndar

ybr

anch

num

ber

Secondary branch number .34 Ter

tiary

bra

nch

num

ber

Tertiary branch number -.13 .65 % te

rtia

ry b

ranc

hing

% tertiary branching -.33 .27 .90 Plan

t hei

ght

Plant height .08 .09 -.07 -.15 Pedu

ncle

leng

th

Peduncle length -.26 -.22 -.27 -.23 .22 Rac

his

leng

th

Rachis length .06 .06 .15 .17 -.16 .01 Bra

nch

leng

th

Branch length -.23 .06 .33 .39 -.23 -.19 .71 100-

seed

wei

ght

100-seed weight .14 .06 .09 .10 .11 .01 .07 -.01 Seed

s pe

r pa

nicl

e

Seeds per panicle .44 .41 .36 .24 .28 -.06 .22 .09 .42 Seed

set

Seed set -.04 .00 .08 .10 .32 .10 .03 -.10 .35 .57 Tot

al s

eed

wei

ght

Total seed weight .32 .20 .09 .01 .46 .11 .02 -.16 .52 .82 .62 Rac

his

diam

eter

Rachis diameter .58 .50 .24 .04 -.19 -.15 .23 .06 .30 .55 .05 .26 Flow

erin

g tim

e

Flowering time .28 .33 .21 .09 -.07 -.28 -.05 .16 -.29 -.35 -.59 -.50 .20Panicles per row -.14 -.35 -.49 -.44 .33 .29 -.24 -.37 -.33 -.29 .01 .16 -.49 -.27

123

Theor Appl Genet (2006) 113:931–942 937

in the exotic progenitor of IS3620C (NSL51039), whichis tall and photoperiod-sensitive, and in the “conver-sion parent” (BTx406), the elite line which was used todonate dwarWng and photoperiod-insensitive alleles.The sequencing results showed that IS3620C had been“converted” at the Sb-ba1 locus (data not shown).Therefore, Sb-ba1 was mapped using a diVerent sor-ghum population (BTx623 £ Sorghum propinquum, incollaboration with J. Bowers and A. Paterson), and itsposition on the present map is inferred based on thealignment of these two sorghum genetic maps (Feltuset al. 2006).

Co-localization of candidate genes and QTL

Two of the eight mapped candidate genes co-localizedwith QTL. Given that the 2-LOD intervals of thedetected QTL cover 235 cM, or 14% of the 1,677 cMgenome (Table S2), the proportion of candidates co-localizing with QTL is not signiWcantly greater thanwhat would be expected due to chance alone using achi-square test (P = 0.37). Even if 1-LOD QTL inter-vals are considered, the detected QTL collectivelycover 127 cM or 7.5% of the genome, and the propor-tion of candidates co-localizing with QTL is still not

Fig. 3 Candidate genes and QTL for sorghum inXores-cence traits. The ten sorghum chromosomes are shown in order and calibrated at 10 cM intervals. Black ellipses repre-sent centromeres and pericen-tromeric heterochromatin as reported by Kim et al. (2005). Mapped candidate genes and QTL are shown to the left and right, respectively, of each chromosome. The Sb-ba1 gene location is shown with a bracket to indicate that the precise location of this gene on this map is not known. Each QTL is indicated by a box and arrowhead that repre-sent the 2-LOD interval and LOD peak, respectively. The direction of the arrowhead indicates the direction of the QTL eVect (left-pointing arrow: BTx623 allele increas-es trait value; right-pointing arrow: IS3620C allele in-creases trait value), and the width of the QTL box is pro-portional to the LOD score

6 7 8 9 10

Sb-lfy

Dw 3

1 2 3 4 5

Sb-

bd1

100cM Sb-

ra 1

Sb-

ids 1

Sb-

tfl1

}Sb-

ba1

Sb-

ra 2

150cM

50cM

200cM

Branchlength

100-seedweight

Branchlength

Rachisdiameter

Primarybranchnumber

Plantheight

Branchlength

Primarybranchnumber

100-seedweight

Totalseed

weight

Floweringtime

Plantheight

100-seedweight

Flowering time

Plantheight

Plantheight

Rachislength

Branchlength

Secondarybranchnumber

123

938 Theor Appl Genet (2006) 113:931–942

123

Tab

le3

Can

dida

te g

enes

for

gra

ss inX

ores

cenc

e ar

chit

ectu

re

a Sb

-ba1

was

map

ped

in a

pop

ulat

ion

of 4

8 (B

Tx6

23£

Sorg

hum

pro

pinq

uum

) F

2s c

ourt

esy

of J

. Bow

ers

and

A. P

ater

son

b T

he S

b-ra

1 lo

cus

is t

ande

mly

dup

licat

ed (

Vol

lbre

cht e

tal.

2005

)c W

hen

a si

ngle

ort

holo

gous

cop

y of

a c

andi

date

gen

e co

uld

not b

e un

ambi

guou

sly

iden

tiW

ed a

ll pu

tati

ve o

rtho

logs

from

tha

t gra

ss s

peci

es a

re li

sted

d Q

uest

ion

mar

ks a

re in

clud

ed w

here

the

full

amin

o ac

id s

eque

nce

of a

pro

tein

is n

ot k

now

n

Sorg

hum

can

dida

teP

utat

ive

orth

olog

s

Nam

eC

hr. (

cM)

Am

ino

acid

sSp

p.G

ene

Ref

eren

ceA

min

o ac

ids

% I

D w

/sor

ghum

pr

otei

nH

omol

ogy

Fun

ctio

n

Sb-b

a13

(174

–199

)a22

4Z

eaO

ryza

ba1

LA

XG

alla

vott

i et

al. (

2004

)K

omat

su e

tal.

(200

3b)

211

215

87 62bH

LH

TF

Pri

mar

y br

anch

In

itia

tion

Sb-r

a12

(99.

1)11

3/18

4bZ

eara

1V

ollb

rech

t et

al. (

2005

)17

471

/71

EP

F-c

lass

C2H

2 zi

nc W

nger

Bra

nch

dete

rmin

acy

Sb-r

a23

(30.

5)26

8Z

eaO

ryza

ra2

BA

B92

516.

1B

orti

ri e

tal.

(200

6)(h

ttp:

//ww

w.g

ram

ene.

org)

261

251

87 77L

OB

dom

ain

Bra

nch

dete

rmin

acy

Sb-b

d12

(196

.0)

326

Zea

Ory

zabd

1fz

pC

huck

eta

l. (2

002)

Kom

atsu

eta

l. (2

003a

)31

331

883 73

1 A

P2-

dom

ain

ER

FSp

ikel

et d

eter

min

acy

Sb-i

ds1

1 (1

3.2)

432

Zea

Ory

zaid

s1B

AC

2144

8.1c

AC

1044

33

Chu

ck e

tal.

(199

8)G

ram

ene

Gra

men

e

433

436

434

94 64 69

2 A

P2-

dom

ain

ER

FSp

ikel

et d

eter

min

acy

Sb-l

fy6

(82.

7)?d

Zea

Ory

zaL

oliu

m

zX1-

2rX L

tLF

Y

Bom

blie

s et

al. (

2003

)K

yozu

ka e

tal

. (19

98)

Goc

al e

tal.

(200

1)

391/

394

389

400

? ? ?

FL

OR

I-C

AU

LA

/LE

AF

YR

epro

duct

ive

tran

siti

on/

inX

o. b

ranc

hing

Sb-tX

15

(27.

5)17

3Z

eaO

ryza

Lol

ium

Zm

cen1

-3O

scen

2L

pTF

L1

this

pap

erZ

hang

eta

l. (2

005)

Jens

en e

tal

. (20

01)

177/

173/

?17

317

3

92/9

2/?

91 89

TF

L1/

CE

N (

PE

BP

)R

epro

duct

ive

tran

siti

on/

inX

o. b

ranc

hing

Dw

37

(116

.5)

1,40

2Z

eaO

ryza

br2

BA

D12

940.

1M

ulta

ni e

tal

. (20

03)

Gra

men

e1,

394

1,34

491 82

P-g

lyco

prot

ein

Aux

in t

rans

port

Theor Appl Genet (2006) 113:931–942 939

signiWcant (P = 0.06). Both co-localizing candidatesmapped within 2 cM of the respective QTL peaks,whereas the 2-LOD intervals averaged 15 cM.

Discussion

Genetic architecture of sorghum panicle traits

Variation in sorghum inXorescence architecture resultsfrom diVerences in the branching, elongation, andbranch abortion of the developing inXorescence.Although the genic basis of these diVerences isunknown, the high heritabilities of easily-measuredinXorescence traits like branch length and primarybranch number suggest that inXorescence architecturemay be quite amenable to quantitative analysis. Sincethe sorghum panicle is more highly branched than theinXorescences of rice or maize, and it shows great vari-ability among inbred lines, it is an excellent model withwhich to study quantitative variation in higher-orderinXorescence branching in grasses.

InXorescence branching

Primary branch number had signiWcantly higher herita-bility than secondary and tertiary branching traits inthis study. There are several probable reasons for this.First, only a subset of the primary branches were usedto estimate secondary and tertiary branching. Second,secondary and tertiary branching are aVected not onlyby meristem initiation but also by the abortion of basalbranches, which takes place later in development andis presumably controlled by a distinct set of genes. Sub-sequent studies could potentially disentangle thesephenomena by phenotyping secondary branch initia-tion separately from secondary branch abortion.

Analysis of both genetic correlations and QTL sup-ports the idea that primary, secondary, and tertiaryinXorescence branching in sorghum are largely underseparate genetic control. There is no overlap betweenQTL detected for primary and secondary branch num-ber, in keeping with the many reports of speciWc genesthat control speciWc stages of inXorescence branching(Bommert et al. 2005).

Three QTL for inXorescence branching weredetected, two for primary branch number and one forsecondary branch number. The primary branch num-ber QTL on chromosome 6 maps within a few cM of aQTL for rachis diameter. When rachis diameter wasincluded as a covariate in the linear model for primarybranch number, this QTL was no longer signiWcant atP < 0.001. Therefore, this primary branch number QTL

may be detected due to its proximity to the rachisdiameter QTL. Alternatively, these two QTL may rep-resent a single locus that exerts a concomitant increasein primary branch number and rachis diameter.

This genomic region is also known to harbor Ma1, amajor gene for photoperiod-sensitivity, as well as themajor height gene Dw2 (Lin et al. 1995). Since mostexotic grain sorghum has been selected for photope-riod-insensitivity through crossing to an insensitive line,this region of the genome is usually converted to anelite state in exotic germplasm, and is absent from somesorghum maps (e.g. see Sanchez et al. 2002). Some ofthe QTL in this genomic region may be conditioned byelite alleles. In particular, the plant height QTL is likelyDw2, which is dominant in BTx623 but recessive in theconversion parent BTx406. The QTL for rachis diame-ter and primary branch number could be conditionedby either elite or exotic alleles in this population, buttheir proximity to Ma1 and Dw2 suggests that at leastsome converted lines of sorghum may have been unin-tentionally converted at these inXorescence QTL.

InXorescence elongation

InXorescence elongation appears to be under separategenetic control from plant height, as has been notedpreviously (Doggett 1988). In this population, elitealleles of the QTL near Dw3 cause moderate elonga-tion of the rachis and primary branches and a drasticcompaction of the lower vegetative internodes, consis-tent with the reported eVects of Dw3 (Multani et al.2003). Aside from these putative Dw3 QTL, no closeco-localization of QTL for plant height, rachis length,and branch length was observed. Thus, not only is theelongation of the inXorescence distinct from that of thevegetative internodes, but rachis elongation and rachisbranch elongation also appear to be distinct processes.Four QTL were detected for branch length, more thanfor any other trait except plant height. Branch length isof relevance to grain mold resistance and bird preda-tion, and is a promising target for further quantitativegenetic study in sorghum.

InXorescence branch abortion

The overall appearance of panicle “compactness”,which is used as a racial indicator in sorghum, is inXu-enced not only by inXorescence branching and elonga-tion, but also by the abortion of higher-order branchesprior to spikelet maturity. In this study, branch abortionwas assessed by measuring secondary and tertiarybranching on two basal branches per panicle: since a“branch” was deWned as an axis bearing multiple spikelet

123

940 Theor Appl Genet (2006) 113:931–942

pairs, aborted branches were not counted. The expecta-tion for a QTL resulting from diVerences in branchabortion is that it should aVect both secondary and ter-tiary branching. However, only a single QTL wasdetected for secondary branch number, and no QTLwere detected for tertiary branching. Future studies thatmeasure branch abortion directly may have more suc-cess in uncovering the genetic basis of this complex trait.

Relationship between inXorescence and seed traits

Grass inXorescence architecture is of primary interest tobreeders because it contributes directly to yield, yieldstability, and grain quality. The strong positive correla-tion between the number of seeds per panicle and allthree orders of inXorescence branching suggests thatthis yield component can be selected for throughincreased inXorescence branching. None of the threeQTL detected for 100-seed weight were linked to inXo-rescence architecture QTL, suggesting that selection forboth increased seed weight and increased inXorescencebranching is potentially feasible. However, these resultsshould be interpreted with caution, and the relationshipbetween seed weight, seed number, and inXorescencearchitecture should be reassessed in a sorghum popula-tion that does not show such wide variation in seed set.

Mapping of grass inXorescence genes in sorghum

Eight candidate genes for grass inXorescence architec-ture were mapped in sorghum. Based on both genomiccolinearity and sequence similarity, it is likely that thesorghum genes are orthologous to their counterpartsfrom other grasses. While Southern blots are stillneeded for absolute veriWcation of copy number,results from BLAST analysis of the SAMI databasealso suggest that most of these genes are present in asingle copy in the sorghum genome (data not shown).Finally, given the high degree of similarity between thepredicted sorghum proteins and their putative ortho-logs in maize, rice, and ryegrass, it is likely that the pro-teins perform similar functions. If functional variationexists at these candidate gene loci in sorghum, thenthey have the potential to contribute to quantitativevariation in inXorescence architecture.

During the mapping of Sb-ids1, we detected an addi-tional Sb-ids1-like locus on the short arm of chromo-some 10 (data not shown). Intriguingly, this genomicregion in sorghum has several markers collinear withmaize 9S near the map location of the rgo1 mutant(Kaplinsky and Freeling 2003). Since ids1 and rgo1show non-allelic non-complementation in maize, it islikely that they encode similar proteins.

Sb-ba1 could not be mapped in our populationbecause of the introgression of an elite allele indistin-guishable from that of BTx623. For this reason, we hadlittle ability to assess the possible contribution of Sb-ba1 to phenotypic variation in sorghum inXorescencearchitecture. The conversion of Sb-ba1 in IS3620C islikely due to linkage with a height gene rather thandirect selection on Sb-ba1 itself, since the Sb-ba1region co-localizes with previously reported heightQTL and conversion events (Lin et al. 1995), and func-tional characterization of ba1 in maize does not suggestthat it has any eVect on plant height or Xowering time(Gallavotti et al. 2004).

Correspondence between sorghum QTL and candidate genes

Two of eight mapped candidate genes were found toco-localize with QTL, which is not signiWcantly diVer-ent from the expectation due to chance alone. Boththese cases of co-localization merit further examina-tion, however, since these two candidates map veryclose to the peaks of QTL for traits that closely matchtheir characterized gene function. The QTL reportedhere also suggest a short list of other candidate genesfor subsequent mapping in sorghum. For example, aXowering time QTL detected in this study maps to theshort arm of chromosome 9 between rz390 and cdo580,which both map to maize bin 8.05 within a few cM ofthe maize Xowering time QTL Vgt1 (Salvi et al. 2002).

The precise co-localization of the Dw3 gene withQTL for plant height, rachis length, and branch lengthstrongly suggests that IS3620C carries a wild-type Dw3allele. The BTx623 Dw3 allele, which has an » 800 bptandem duplication in exon 5, is similar in sequence tothe unstable, dwarWng allele characterized by Multaniet al. (2003), whereas the IS3620C Dw3 allele does notcontain this duplication (data not shown). Therefore,we hypothesize that the QTL at this locus are condi-tioned by the » 800 bp duplication in exon 5 of theDw3 gene.

The Sb-ra2 gene co-localizes closely with one of onlytwo QTL detected for primary branch number acrossthe genome. A » 1.9 Kb fragment of the Sb-ra2 locuswas sequenced from each parent, including the entireSb-ra2 protein-coding region, » 400 bp of upstreamsequence, and » 700 bp of downstream sequence. Onlya single polymorphism was found, a 4 bp indel 72 bpdownstream of the stop codon (data not shown). Thus,the predicted Sb-RA2 proteins are identical betweenmapping parents, although it is possible that theparents diVer in expression and/or translation of theSb-ra2 transcript.

123

Theor Appl Genet (2006) 113:931–942 941

Since the evidence from QTL and candidate geneco-localization is purely coincidental, the possible con-tributions of Dw3 and Sb-ra2 to variation in sorghuminXorescence architecture merit closer examinationthrough Wne-mapping and/or association mapping. Thelatter approach oVers several advantages in that it ismuch quicker and can evaluate a much wider range ofalleles, which is particularly important for a trait poten-tially subjected to diversifying selection. However, sor-ghum is a predominantly-selWng organism withmoderate linkage disequilibrium (Hamblin et al. 2005),so association mapping may not be able to isolate sin-gle polymorphisms conditioning the phenotypes ofinterest. Association mapping is also potentially disad-vantaged, relative to Wne-mapping, when the target is anon-coding polymorphism not in close physical prox-imity to the gene whose expression it aVects. At the Blocus in maize, for example, expression is aVected bytandem repeats over 100 KB upstream of the start codon(Stam et al. 2002). Used together, the Wne-mapping andassociation mapping approaches can provide a signiW-cant increase in power (Meuwissen et al. 2002).

It is worth noting that the two candidate genes ofgreatest promise, Dw3 and Sb-ra2, are both genes thatwere identiWed as maize mutants. In contrast toLEAFY and TFL1, which were discovered as Arabid-opsis and Antirrhinum mutants, and to ba1/LAX andbd1/fzp, which were identiWed in both maize and rice,br2/Dw3 and ra2 (as well as ra1 and ids1) are geneswith mutant phenotypes described only in maize.Although Wrm inference cannot be drawn from such asmall sample, this study nonetheless suggests that phy-logenetic proximity is a critical consideration whenselecting candidate genes from other species.

Conclusions

In this study, we have mapped a number of candidategenes for grass inXorescence development, several ofwhich co-localize closely with QTL for relevant inXo-rescence traits in sorghum across multiple environ-ments. These initial results suggest that the approachemployed here, which uses candidate genes fromrelated model organisms to investigate quantitativevariation in a target species with high phenotypic vari-ance, may be highly eVective. Since inXorescence archi-tecture in sorghum is a complex trait with dramaticintraspeciWc variation, the possible contribution of can-didate genes from other grasses, not limited to thesmall number described here, needs to be evaluatedacross a broader range of sorghum germplasm. Ulti-mately, the genes identiWed in sorghum may helpexplain how human selection, acting upon the common

genetic and physical architecture of the grass inXores-cence, produced a range of viable, alternative out-comes in diVerent domesticated cereals.

Acknowledgments We thank Toby Kellogg, Sarah Hake, andGael Pressoir for critical advice, Robert Klein for providing plantmaterials, Delilah Wood for assistance with scanning electronmicroscopy, and Mark Sorrells and David Benscher for providingthreshing equipment. The MORPH Research Coordination Net-work provided Wnancial assistance to PJB.

References

Ashikari M, Sakakibara H, Lin S, Yamamoto T, Takashi T, Ni-shimura A, Angeles W, Qian Q, Kitano H, Matsuoka M(2005) Cytokinin oxidase regulates rice grain production.Science 309:741–745

Bedell JA, Budiman MA, Nunberg A, Citek RW, Robbins D,Jones J, Flick E, RohlWng T, Fries J, Bradford K, MeMenamyJ, Smith M, Holeman H, Roe BA, Wiley G, Korf IF, Rabi-nowicz PD, Lakey N, McCombie WR, Jeddeloh JA, Mart-ienssen RA (2005) Sorghum genome sequencing bymethylation Wltration. PLoS Biol 3:e13

Bomblies K, Wang R-L, Ambrose BA, Schmidt RJ, Meeley RB,Doebley JF (2003) Duplicate FLORICAULA/LEAFYhomologs zX1 and zX2 control inXorescence architecture andXower patterning in maize. Development 130:2385–2395

Bommert P, Satoh-Nagasawa N, Jackson D, Hirano H-Y (2005)Genetics and evolution of inXorescence and Xower develop-ment in grasses. Plant Cell Physiol 46:69–78

Bortiri E, Chuck G, Vollbrecht E, Rocheford T, Martienssen R,Hake S (2006) ramosa2 encodes a LATERAL ORGANBOUNDARY domain protein that determines the fate ofstem cells in branch meristems of maize. Plant Cell 18:574–585

Bout S, Vermerris W (2003) A candidate-gene approach to clonethe sorghum Brown Midrib gene encoding caVeic acid O-methyltransferase. Mol Genet Genom 269:205–214

Brown DG, Vision T (2000) MapPop 1.0: software for selectivemapping and bin mapping. http://www.bio.unc.edu/faculty/vision/lab/mappop/

Childs KL, Miller FR, Cordonnier-Pratt MM, Pratt LH, MorganPW, Mullet JE (1997) The sorghum photoperiod sensitivitygene, Ma3, encodes a phytochrome B. Plant Physiol 113:611–619

Chuck G, Meeley RB, Hake S (1998) The control of maize spike-let meristem fate by the APETALA2-like gene indetermi-nate spikelet1. Genes Dev 12:1145–1154

Chuck G, Muszynski M, Kellogg E, Hake S, Schmidt RJ (2002)The control of spikelet meristem identity by the branchedsilkless1 gene in maize. Science 298:1238–1241

Clark RM, Linton E, Messing J, Doebley JF (2004) Pattern ofdiversity in the genomic region near the maize domesticationgene tb1. Proc Natl Acad Sci 101:700–707

Devos KM (2005) Updating the ‘crop circle’. Curr Opin PlantBiol 8:155–162

Doggett H (1988) Sorghum, 2nd edn. Longman ScientiWc & Tech-nical, Essex

Doust AN, Devos KM, Gadberry MD, Gale MD, Kellogg EA(2005) The genetic basis for inXorescence variation betweenfoxtail and green millet (Poaceae). Genetics 169:1659–1672

Doyle J, Doyle JL (1987) Genomic plant DNA preparation fromfresh tissue: CTAB method. Phytochem Bull 19:11

123

942 Theor Appl Genet (2006) 113:931–942

Feltus FA, Hart GE, Schertz KF, Casa AM, Kresovich S, KleinPE, Paterson AH (2006) Alignment of genetic maps andQTLs between inter- and intra-speciWc sorghum popula-tions. Theor Appl Genet (in press)

Gallavotti A, Zhao Q, Kyozuka J, Meeley RB, Ritter MK, Doeb-ley JF, Pe ME, Schmidt RJ (2004) The role of barren-stalk1in the architecture of maize. Nature 432:630–635

Gaut BS (2002) Evolutionary dynamics of grass genomes. NewPhytol 154:15–28

Gocal GF, King RW, Blundell CA, Schwartz OM, Andersen CH,Weigel D (2001) Evolution of Xoral meristem identity genes.Analysis of Lolium temulentum genes related to APETALA1and LEAFY of Arabidopsis. Plant Physiol 125:1788–1801

Hamblin MT, Salas Fernandez MG, Casa AM, Mitchell SE, Pat-erson AH, Kresovich S (2005) Equilibrium processes cannotexplain high levels of short- and medium-range linkage dis-equilibrium in the domesticated grass Sorghum bicolor.Genetics 171:1247–1256

Harlan JR, De Wet JMJ (1972) A simpliWed classiWcation ofcultivated sorghum. Crop Sci 12:172–176

Ikeda K, Sunohara H, Nagato Y (2004) Developmental course ofinXorescence and spikelet in rice. Breeding Sci 54:147–156

Jensen CS, Salchert K, Nielsen KK (2001) A TERMINALFLOWER1-like gene from perennial ryegrass involved inXoral transition and axillary meristem identity. Plant Physiol125:1517–1528

Kaplinsky NJ, Freeling M (2003) Combinatorial control of meri-stem identity in maize inXorescences. Development130:1149–1158

Kim J-S, Islam-Faridi MN, Klein PE, Stelly DM, Price HJ, KleinRR, Mullet JE (2005) Comprehensive molecular cytogeneticanalysis of sorghum genome architecture; distribution ofeuchromatin, heterochromatin, genes and recombination incomparison to rice. Genetics 171:1963–1976

Komatsu M, Chujo A, Nagato Y, Shimamoto K, Kyozuka J(2003a) FRIZZY PANICLE is required to prevent the for-mation of axillary meristems and to establish Xoral meristemidentity in rice spikelets. Development 130:3841–3850

Komatsu K, Maekawa M, Ujiie S, Satake Y, Furutani I, OkamotoH, Shimamoto K, Kyozuka J (2003b) LAX and SPA: majorregulators of shoot branching in rice. Proc Natl Acad Sci100:11765–11770

Kyozuka J, Konishi S, Nemoto K, Izawa T, Shimamoto K (1998)Down-regulation of RFL, the FLO/LFY homolog of rice,accompanied with panicle branch initiation. Proc Natl AcadSci 95:1979–1982

Lauter N, Doebley JF (2002) Genetic variation for phenotypi-cally invariant traits detected in teosinte: implications for theevolution of novel forms. Genetics 160:333–341

Lin Y-R, Schertz KF, Paterson AH (1995) Comparative analysisof QTLs aVecting plant height and maturity across the Poa-ceae, in reference to an interspeciWc sorghum population.Genetics 141:391–411

Menz MA, Klein RR, Mullet JE, Obert JA, Unruh NC, Klein PE(2002) A high-density genetic map of Sorghum bicolor (L.)

Moench based on 2926 AFLP, RFLP, and SSR markers.Plant Mol Biol 48:483–499

Meuwissen THE, Karlsen A, Lien S, Olsaker I, Goddard ME(2002) Fine mapping of a quantitative trait locus for twinningrate using combined linkage and linkage disequilibriummapping. Genetics 161:373–379

Michaels SD, Amasino RM (1998) A robust method for detectingsingle-nucleotide changes as polymorphic markers by PCR.Plant J 14:381–385

Multani DS, Briggs SP, Chamberlin MA, Blakeslee JJ, MurphyAS, Johal GS (2003) Loss of an MDR transporter in compactstalks of maize br2 and sorghum dw3 mutants. Science302:81–84

Robertson DS (1989) Understanding the relationship betweenqualitative and quantitative genetics. In: Helentjaris T, BurrB (eds) Development and application of molecular markersto problems in plant genetics. Cold Spring Harbor Lab Press,New York, pp 81–87

Salvi S, Tuberosa R (2005) To clone or not to clone plant QTLs:present and future challenges. Trends Plant Sci 10:297–304

Salvi S, Tuberosa R, Chiapparino E, Maccaferri M, Veillet S, vanBeuningen L, Isaac P, Edwards K, Phillips RL (2002) To-ward positional cloning of Vgt1, a QTL controlling the tran-sition from the vegetative to the reproductive phase in maize.Plant Mol Biol 48:601–613

Sanchez AC, Subudhi PK, Rosenow DT, Nguyen HT (2002)Mapping QTLs associated with drought resistance insorghum (Sorghum bicolor L. Moench). Plant Mol Biol48:713–726

Stam M, Belele C, Ramakrishna W, Dorweiler JE, Bennetzen JL,Chandler VL (2002) The regulatory regions required for B’paramutation and expression are located far upstream of themaize b1 transcribed sequences. Genetics 162:917–930

Stephens JC, Miller FR, Rosenow DT (1967) Conversion of aliensorghums to early combine genotypes. Crop Sci 7:396

Swigonova Z, Lai J, Ma J, Ramakrishna W, Llaca V, BennetzenJL, Messing J (2004) Close split of sorghum and maizegenome progenitors. Genome Res 14:1916–1923

Thornsberry JM, Goodman MM, Doebley JF, Kresovich S, Niel-sen D, Buckler ES (2001) Dwarf8 polymorphisms associatewith variation in Xowering time. Nat Genet 28:286–289

Trevaskis B, Bagnall DJ, Ellis MH, Peacock WJ, Dennis ES(2003) MADS box genes control vernalization-induced Xow-ering in cereals. Proc Natl Acad Sci 100:13099–13104

Upadyayula N, da Silva HS, Bohn MO, Rocheford TR (2005)Genetic and QTL analysis of maize tassel and ear inXores-cence architecture. Theor Appl Genet (in press)

Vollbrecht E, Springer PS, Goh L, Buckler ES, Martienssen RA(2005) Architecture of Xoral branch systems in maize and re-lated grasses. Nature 436:1119–1126

Wang S, Basten CJ, Zeng Z-B (2005) Windows QTL Cartogra-pher 2.5. Statistical Genetics, NCSU

Zhang S, Hu W, Wang L, Lin C, Cong B, Sun C, Luo D (2005)TFL1/CEN-like genes control intercalary meristem activityand phase transition in rice. Plant Sci 168:1393–1408

123