Comparative Mapping of Quantitative Trait Loci Sculpting the Curd of Brassica oleracea

Corresponding author: Andrew H. Paterson, Plant Genome Mapping Laboratory, University of Georgia, Rm. 162, Riverbend Research Center, 110 Riverbend Rd., Athens, GA 30602., paterson{at}dogwood.botany.uga.edu (E-mail)

Communicating editor: J. A. BIRCHLER

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The enlarged inflorescence (curd) of cauliflower and broccoli provide not only a popular vegetable for human consumption, but also a unique opportunity for scientists who seek to understand the genetic basis of plant growth and development. By the comparison of quantitative trait loci (QTL) maps constructed from three different F₂ populations, we identified a total of 86 QTL that control eight curd-related traits in Brassica oleracea. The 86 QTL may reflect allelic variation in as few as 67 different genetic loci and 54 ancestral genes. Although the locations of QTL affecting a trait occasionally corresponded between different populations or between different homeologous Brassica chromosomes, our data supported other molecular and morphological data in suggesting that the Brassica genus is rapidly evolving. Comparative data enabled us to identify a number of candidate genes from Arabidopsis that warrant further investigation to determine if some of them might account for Brassica QTL. The Arabidopsis/Brassica system is an important example of both the challenges and opportunities associated with extrapolation of genomic information from facile models to large-genome taxa including major crops.

THE genus Brassica (tribe Brassicaceae), which includes many important crops, is in the taxonomic family Cruciferae, as is Arabidopsis thaliana, the first flowering plant whose genome will be completely sequenced. Such a close relationship suggests that crop plants of the genus Brassica will be among the earliest beneficiaries of the Arabidopsis sequence. Among the members of the genus Brassica, Brassica oleracea (n = 9) has long been of great interest to plant morphologists. B. oleracea includes many vegetables with distinct characteristics that are either not known or are less well developed in A. thaliana, including the enlarged inflorescence of cauliflower (B. oleracea subsp. botrytis) and broccoli (B. oleracea subsp. italica); the enlarged stem of kohlrabi (B. oleracea subsp. gongylodes) and marrow-stem kale (B. oleracea subsp. medullosa); the enlarged apical bud of cabbage (B. oleracea subsp. capitata); and the enlarged lateral buds of Brussels sprouts (B. oleracea subsp. gemmifera) (KALLOO and BERGH 1993 ) as well as morphologically simple "rapid-cycling" genotypes with a life cycle of 6 to 8 wk (WILLIAMS and HILL 1986 ). The fact that many members of the species B. oleracea can be intercrossed to generate fertile progenies provides an expedient means to investigate the genetic basis of enlargement of these organs. Classical quantitative studies show the inheritance of most of these novel morphological characteristics to be complex (KRISTOFFERSON 1924 ; PEASE 1926 ).

B. oleracea originated from the eastern Mediterranean area and appears to have been subjected to intensive selection pressure in the past century (PRAKASH and HINATA 1980 ). Formal genetic studies of B. oleracea started in the 1920s (KRISTOFFERSON 1924 ; DETJEN 1926 ; PEASE 1926 ). The heritability of the curd (inflorescence) size of cauliflower has been studied in particular detail (WATTS 1964 , WATTS 1965 ; CRISP 1977 ; FREEMAN and CRISP 1979 ). In these studies, curd weight is closely correlated with curd diameter, an easily measured, nondestructive trait that is particularly useful as a selection tool. The genetic basis of curd weight is complex and consequently has been described in terms of mathematical genetic parameters of multiple loci rather than specific locations and phenotypic effects of individual loci. Curd diameter showed additive genetic variation and heterosis (KALLOO and BERGH 1993 ). In related studies, the genetic factors determining the conformation of curd was analyzed (WATTS 1966 ). Curd shape was associated with acute angles between the curd branches. More recently, DNA markers were used to dissect the inheritance of curd size (KENNARD et al. 1994 ); however, the use of a very small population (95 individuals) constrained the number of quantitative trait loci (QTL) that could be resolved (LANDER and BOTSTEIN 1989 ).

This work, based on three B. oleracea F₂ populations and eight curd-related traits, identified a total of 86 quantitative trait loci, providing a detailed dissection of complex traits in B. oleracea and a comparative analysis of corresponding mutations in Arabidopsis. Its close relationship to Arabidopsis provides an efficient way for Brassica scientists to utilize the extensive toolboxes of yeast artificial chromosome (YAC)/bacterial artificial chromosome (BAC) contigs, expressed sequence tags (ESTs), genetic/physical maps, and genomic sequences available in Arabidopsis. Brassica/Arabidopsis comparative maps (KOWALSKI et al. 1994 ; LAGERCRANTZ et al. 1996 ; OSBORN et al. 1997 ; LAN et al. 2000 ) provide a means to evaluate the relationship between Brassica QTL and mutants, mapped ESTs, and inferred open reading frames in Arabidopsis. Further, the model of Brassica/Arabidopsis may possibly be extended to more distantly related plants (PATERSON et al. 1996 ).

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Plant materials:
Three B. oleracea F₂ populations were used for QTL analysis: rapid-cycling B. oleracea (RCB; self-compatible) x B. oleracea var. Cantanese (CAN; USDA accession no. PI462224), RCB x B. oleracea var. Pusa Katki (PK; USDA accession no. PI274783), and RCB x B. oleracea var. Bugh Kana (BK; USDA accession no. PI249556), composed of 247, 250, and 246 individuals, respectively (Fig 1). One B. oleracea F₂ population, RCB x B. oleracea var. Green Comet (GC; USDA accession no. G30771) containing 56 individuals, was used to facilitate the construction of a composite linkage map. Rapid-cycling Brassica was from the Crucifer Genetics Cooperative, Madison, Wisconsin. A single-plant selection derived by selfing a randomly chosen rapid-cycling plant was used. Seed and/or pollen of other B. oleracea varieties were generously provided by Dr. J. McFerson, and Dr. S. Kresovich, USDA/ARS, Geneva, New York. To minimize environmental effect among populations, all the plants used for QTL analysis were grown in the same experimental field of the Texas Agricultural Experiment Station in College Station, Texas, in the same period of time (from October 1994 to April 1995). A completely randomized design was used for field planting. Spacing of plants between rows and within rows is 1 m. Two parental self-compatible B. oleracea (SCO), one of the other parent and one F₁, were planted in every row at randomly determined positions within the row. Brassica seedlings were nursed in the growth chamber for 2 wk before transplanting to the field. Fertilization, irrigation, and insecticide application were consistent with commercial production conditions.

View larger version (325K): In this window In a new window Download PPT slide   Figure 1. The parents used in this study. (A) The profound phenotypic difference between B. oleracea var. Cantanese (left) and rapid-cycling B. oleracea (right), 6 wk after planting in the field. The rapid-cycling parent has flowered and set seeds, while the flower bud of Cantanese has not even emerged. (B) B. oleracea var. Bugh Kana has small and loose curd. (C) B. oleracea var. Pusa Katki has a large and compact curd. At maturity, the curd of Cantanese will be similar to Bugh Kana.

Phenotyping:
The following curd-related traits (Fig 2) were measured:

1. Days to budding (dfb): number of days from planting to appearance of the first floral bud.

   
   

   View larger version (103K): In this window In a new window Download PPT slide   Figure 2. Illustration of curd-related traits studied. Abbreviations for the traits are as listed in the MATERIALS AND METHODS section.

2. Days from budding to flowering (dofdbf): number of days from budding to the first flower blossom.

3. First-rank branching (rk1): the number of branches within the curd that originated from the main stem (measured on the first flowering day).

4. Side-branches (sidb): number of lateral branches from the main stem that were outside the curd (measured on the first flowering day).

5. Cluster width (cluw): the width of the largest floral cluster measured on the first flowering day.

6. Curd width (headw): the width of the curd measured on the first flowering day.

7. Apical shoot length (rrl1st): stem length from the apical meristem to where the closest first-rank branch originated from the main stem (measured 180 days after planting).

8. First branch length (rrl2nd): length of the branch that is the nearest to the apical meristem.

Genotyping:
DNA extraction, electrophoresis, Southern blotting, and autoradiography were as previously described (KOWALSKI et al. 1994 ). The DNA markers and assembly of the Brassica composite linkage map were described elsewhere (LAN et al. 2000 ).

Data analysis:
Trait means, correlations, and histograms were calculated by using SAS and Microsoft Excel. Broad sense heritabilities were estimated from the variances of F₂ and F₁ (FALCONER 1989 ). Restriction fragment length polymorphism (RFLP) linkage maps were constructed by using MapMaker (LANDER et al. 1987 ) and the Kosambi centimorgan function. QTL likelihood maps, percentage of variance explained, and gene actions were determined by using interval mapping in individual and multiple QTL models by MapMaker/QTL (LANDER and BOTSTEIN 1989 ). The threshold of LOD (logarithm of odds) = 2.5 was used, appropriate for the marker density and number and length of B. oleracea chromosomes (LANDER and BOTSTEIN 1989 ). For each phenotype, the "free genetics mode" was used to "scan" the whole genome and locate QTL. Then the QTL was "fixed," and the genome rescanned to seek QTL that were masked by the largest QTL. If a new QTL was found, then the second largest QTL was also fixed, and the genome was scanned again until no further QTL were found. For each QTL, LOD 1 and 2 support intervals were plotted.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Phenotypic variation:
Fig 3 illustrates the phenotypic distribution, parental means, and population means for each trait in original units. Log transformation was used for dfb, dofdbf, rk1, sidb, and headw in all populations to improve normality. The distribution of dfb and dofdbf in the CAN population clearly indicates a bimodel distribution, which suggests segregation for both qualitative and quantitative factors influencing flowering. Further, substantial transgression of parental phenotypes was observed for sidb (CAN populations), cluw (all three populations), rrl1st (BK population), and rrl2nd (BK population).

View larger version (45K): In this window In a new window Download PPT slide   Figure 3. Histograms of curd-related phenotypes measured. The arrows indicate the average values of parents and F2 grown in the same environment. Too few F1 plants were available to obtain meaningful data.

Table 1 summarizes the broad sense heritabilities for each trait investigated. Generally, for the same trait, broad sense heritability varied greatly between populations. Such variations suggest complex inheritance and genetic differences among populations. Phenotypes that showed substantial transgression (list above) had broad sense heritabilities averaging 0.6, markedly lower than the average of 0.788 for the other phenotypes.

Correlations among traits:
From Table 2, dfb and dofdbf were highly correlated (r 0.73) across the three populations, suggesting that genes that accelerate or delay floral meristem initiation have similar effects on floral meristem maturation. As expected, traits that were related to curd size such as rk1 and headw were also correlated with one another (r 0.53), with the exception of cluw (-0.27 r 0.21). Late flowering time (dfb and dofdbf) was correlated with larger curd size (rk1 and headw), consistent with previous data (WATTS 1964 ). The strength of correlation between curd size and curd conformation varied in different populations. Specifically, rrl1st was negatively correlated with rk1 and headw in CAN and PK populations.

QTL detected for each trait:
Fig 4 illustrates and Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 Table 10 summarize the QTL detected for each trait.

View larger version (314K): In this window In a new window Download PPT slide   Figure 4. QTL maps of curd-related traits, plotted on RFLP linkage maps of B. oleracea RCB x CAN, RCB x PK, and RCB x BK F2 populations. Linkage maps of these three crosses used for QTL mapping are also aligned with a more detailed map of RCB x GC, with homeologous locations in Brassica and with homologous locations in Arabidopsis (as described in more detail elsewhere; LAN et al. 2000 ). The solid circles next to the loci indicate homeologous Brassica loci (chromosomes 1–9, near right) or homologous Arabidopsis loci (chromosomes 1–5, far right) detected by the same probe. All open circles indicate that no polymorphism was detected for homeologous (Brassica) or homologous (Arabidopsis) loci. The letter "R" next to the probe name indicates that the probe hybridizes to a repetitive DNA sequence in Arabidopsis. Chromosome segments that appear to be homeologous (Brassica) or homologous (Arabidopsis) were connected by shaded columns. Open columns indicate possible triplicated (Brassica) or duplicated regions (Arabidopsis). The likelihood intervals of QTL associated with different phenotypes (as indicated by fill; see legend) are plotted as bars (1 LOD) and whiskers (2 LOD).

dfb: In the BK population, four QTL were found on chromosomes 1, 4, and 7 (2). A full model containing these QTL explained 64.9% of phenotypic variance, with individual QTL models explaining 20.9–49.7% of variance. Allele effects were generally consistent with the difference between parents, as three of the BK alleles delayed budding, while one of the SCO alleles delayed budding. BK alleles were dominant or additive to SCO alleles.

In the CAN population, four QTL were found on chromosomes 3, 4, 5, and 7. A full model containing these QTL explained 90.0% of phenotypic variance, with individual QTL models explaining 6.5–87.6% of variance. Allele effects were generally consistent with the difference between parents, as three of the CAN alleles delayed budding, while one of the SCO alleles delayed budding. CAN alleles were dominant or additive to SCO alleles.

In the PK population, six QTL were found on chromosomes 3, 4 (2), 7, 8, and 9. A full model containing these QTL explained 81.9% of phenotypic variance, with individual QTL models explaining 5.9–45.7% of variance. Allele effects were all consistent with the difference between parents, as all of the PK alleles delayed budding. PK alleles were dominant or additive to SCO alleles.

Across the three Brassica populations, the total of 14 significant marker-trait associations may reflect variation at as few as 11 different genetic loci. BK QTL on chromosome 4 correspond closely to CAN and PK QTL, and a BK QTL at the top of chromosome 7 corresponds to a CAN QTL.

Further, the 11 different genetic loci may be derived from as few as seven ancestral genes. This is based on the observation that some pairs of QTL were located in apparently duplicated chromosomal regions, based on a detailed map of B. oleracea (LAN et al. 2000 ). To further illustrate the point, we have presented (Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 Table 10) the LOD scores for each QTL and also for putatively homeologous (duplicated) locations in the genomes of the relevant B. oleracea crosses in our study. For example, the chromosome 1 QTL (BK: EST429a+7) falls in a region that corresponds to the chromosome 4 QTL (PK: EST114a+1) and chromosome 7 QTL (BK: EW8C11a+2, CAN: WG3D11+14); a chromosome 3 QTL (CAN: EW8F03a+5) corresponds to the region of a chromosome 8 QTL (PK: EW8 D10b+0); and a chromosome 7 QTL (PK: EW6B07a+3) corresponds to the region of a chromosome 9 QTL (PK: EST131a+17).

Finally, these seven ancestral genes correspond approximately to the locations of at least 13 known mutations in Arabidopsis that are implicated in flowering (Table 3).

dofdbf: In the BK population, three QTL were found on chromosomes 4, 7, and 8. A full model containing these QTL explained 48.4% of phenotypic variance, with individual QTL models explaining 7.8–39.7% of variance. Allele effects were all consistent with the difference between parents, as all of the BK alleles delayed flowering. BK alleles were mostly additive to SCO alleles, the one exception being dominant.

In the CAN population, six QTL were found on chromosomes 3, 4 (2), 5, 7, and 8. A full model containing these QTL explained 72.3% of phenotypic variance, with individual QTL models explaining 7.3–59.7% of variance. Allele effects were generally consistent with the difference between parents, as five of the CAN alleles delayed flowering, while one of the SCO alleles delayed flowering. CAN alleles were mostly dominant or additive to SCO alleles, the two exceptions being recessive.

In the PK population, six QTL were found on chromosomes 3, 4, 5, 7, 8, and 9. A full model containing these QTL explained 91.1% of phenotypic variance, with individual QTL models explaining 7.4–76.6% of variance. Allele effects were all consistent with the difference between parents, as all of the PK alleles delayed flowering. PK alleles were mostly dominant or additive to SCO alleles, the one exception being recessive.

Across the three Brassica populations, the total of 15 significant marker-trait associations may reflect variation at as few as 12 different genetic loci. The BK QTL on chromosome 4 corresponds closely to CAN and PK QTL, all showing strictly additive gene action. The BK QTL on chromosome 7 corresponds closely to a CAN QTL, though the two have somewhat different gene action (A and RA, respectively).

Further, the 12 different genetic loci may be derived from as few as nine ancestral genes. The chromosome 3 QTL (CAN: EW8F03a+1) corresponded closely to chromosome 8 QTL (BK: EW8D10b+1, CAN: EW8E09c+2); the chromosome 4 QTL (CAN: EST453g+3) corresponded to a chromosome 7 QTL (BK: EW8C11a+1, CAN: EW8C11a+4); and a second chromosome 7 QTL (PK: EW7E01a+1) corresponded to a chromosome 9 QTL (PK: EST131a+4). QTL in putatively homoeologous locations often showed different gene action.

rk1: In the BK population, one QTL was found on chromosome 7. The model containing this QTL explained 15.8% of phenotypic variance. Its allele effect was consistent with the difference between parents, where the BK allele increased the first-rank branching. The BK allele was dominant to the SCO allele.

In the CAN population, six QTL were found on chromosomes 1, 3, 4, 5, 7, and 8. A full model containing these QTL explained 62.9% of phenotypic variance, with individual QTL models explaining 5.9–27.2% of variance. Allele effects were generally consistent with the difference between parents, as five of the CAN alleles increased branching, while one of the SCO alleles increased branching. CAN alleles were mostly dominant or additive to SCO alleles, the one exception being recessive or additive.

In the PK population, six QTL were found on chromosomes 1, 4, 5, 6, 7, and 9. A full model containing these QTL explained 55.2% of phenotypic variance, with individual QTL models explaining 7.6–41.7% of variance. Allele effects were generally consistent with the difference between parents, as five of the PK alleles increased branching, while one of the SCO alleles increased branching. Two PK alleles were mostly dominant to SCO alleles, one PK allele was additive, and three were largely recessive.

Across the three Brassica populations, the total of 13 significant marker-trait associations may reflect variation at as few as 11 different genetic loci. BK QTL on chromosome 7 corresponds closely to a CAN QTL, and CAN QTL on chromosome 4 corresponds to a PK QTL.

Further, the 11 different genetic loci may be derived from as few as six ancestral genes. The chromosome 1 QTL (PK: EST217a+3) falls in a region that corresponds to the chromosome 5 QTL (PK: EW8A11a+0); another chromosome 1 QTL (CAN: EST566a+2) falls in a region that corresponds to a chromosome 6 QTL (PK: WR1D12a+0) and a chromosome 8 QTL (CAN:EW8E09c+0); a chromosome 4 QTL (CAN: EST453g+3, PK: EST122b+6) falls in a region that corresponds to a chromosome 7 QTL (BK: WG3D11+11, CAN: WG3D11+8); and another chromosome 7 QTL (PK: EW6B07a+3) falls in a region that corresponds to a chromosome 9 QTL (PK: EST131a+14).

sidb: In the BK population, one QTL was found on chromosome 7, which explains 45.9% of phenotypic variance. The allele effect is consistent with the difference between parents, as the BK allele increases branching. The BK allele is additive to the SCO allele.

In the CAN population, three QTL were found on chromosomes 4, 7, and 8. A full model containing these QTL explained 53.6% of phenotypic variance, with individual QTL models explaining 8.1–30.4% of variance. Allele effects were generally consistent with the difference between parents, as all of the CAN alleles increase branching. All of the CAN alleles were recessive or additive to SCO alleles.

In the PK population, four QTL were found on chromosomes 4, 5, 7, and 9. A full model containing these QTL explained 53.1% of phenotypic variance, with individual QTL models explaining 11.4–42.2% of variance. Allele effects were generally consistent with the difference between parents, as all of the PK alleles increase branching. PK alleles were mostly dominant or additive to SCO alleles, the one exception being recessive.

Across the three Brassica populations, the total of eight significant marker-trait associations may reflect variation at as few as six different genetic loci. The BK QTL on chromosome 7 corresponds closely to a CAN QTL, and a CAN QTL on chromosome 4 corresponds to a PK QTL.

Further, the six different genetic loci may be derived from as few as four ancestral genes. The chromosome 4 QTL (CAN: EST55b+16, PK: EST429e+1) fall in a region that corresponds to the chromosome 7 QTL (BK: EW8C11a+1, CAN: EW8C11a+5), and a different chromosome 7 QTL (PK: EW8F06+0) falls in a region that corresponds to the chromosome 9 QTL (PK: EW1G03+0).

cluw: No QTL were detected in the BK population. In the CAN population, three QTL were found on chromosomes 1, 2, and 4. A full model containing these QTL explained 18.2% of phenotypic variance, with individual QTL models explaining 5.9–8.0% of variance. Allele effects were generally consistent with the difference between parents, as two of the CAN alleles increase cluster width, while one of the SCO alleles increase cluster width. CAN alleles were recessive or additive to SCO alleles, the one exception being dominant or additive.

In the PK population, three QTL were found on chromosomes 1, 3, and 4. A full model containing these QTL explained 42.6% of phenotypic variance, with individual QTL models explaining 5.6–33.4% of variance. Allele effects were generally consistent with the difference between parents, as one of the PK alleles increase cluster width, while two of the SCO alleles increase cluster width. PK alleles were mostly dominant to SCO alleles, the one exception being largely recessive.

Across the three Brassica populations, the total of six significant marker-trait associations may reflect variation at as few as five different genetic loci. The CAN QTL on chromosome 4 corresponds closely to the PK QTL.

Further, the five different genetic loci may be derived from as few as four ancestral genes. The chromosome 2 QTL (CAN: EW9F06a+0) falls in a region that corresponds to the chromosome 4 QTL (CAN: EW9E10+0, PK: EW9B02x+0).

headw: In the BK population, 2 QTL were found on chromosomes 4 and 7. A full model containing these QTL explained 21.4% of phenotypic variance, with individual QTL models explaining 10.5–14.9% of variance. Allele effects were generally consistent with the difference between parents, as both of the BK alleles increase curd width. BK alleles were largely dominant to SCO alleles.

In the CAN population, seven QTL were found on chromosomes 1, 3, 4, 5, 7, 8, and 9. A full model containing these QTL explained 65.5% of phenotypic variance, with individual QTL models explaining 5.1–28.5% of variance. Allele effects were generally consistent with the difference between parents, as six of the CAN alleles increase curd width, while one of the SCO alleles increases curd width. CAN alleles were mostly dominant or additive to SCO alleles, the one exception being largely recessive.

In the PK population, five QTL were found on chromosomes 3, 4, 5, 7, and 9. A full model containing these QTL explained 66.1% of phenotypic variance, with individual QTL models explaining 6.4-59.9% of variance. Allele effects were generally consistent with the difference between parents, as all of the PK alleles increase curd width. PK alleles were mostly dominant or additive to SCO alleles, the one exception being recessive.

Across the three Brassica populations, the total of 14 significant marker-trait associations may reflect variation at as few as 12 different genetic loci. A BK QTL on chromosome 7 corresponds to a CAN QTL, and a CAN QTL on chromosome 4 corresponds closely to a PK QTL.

Further, the 12 different genetic loci may be derived from as few as 10 ancestral genes. Several chromosome 7 QTL (BK: EW8C11a+4, CAN: WG3D11+6) fall in a region that corresponds to the chromosome 4 QTL (CAN: EST453g+4, PK: EST122b+4), and another chromosome 7 QTL (PK: EW5A12a+0) falls in a region that corresponds to the chromosome 9 QTL (PK: EW8E09b+2).

rrl1st: In the BK population, 4 QTL were found on chromosomes 1, 4, and 7 (2). A full model containing these QTL explained 48.9% of phenotypic variance, with individual QTL models explaining 7.5–22.7% of variance. Allele effects were generally consistent with the difference between parents, as three of the BK alleles increase length, while one of the SCO alleles increases length. BK alleles were mostly recessive or additive to SCO alleles, the one exception being not recessive.

In the CAN population, five QTL were found on chromosomes 3, 4, 5, 7, and 9. A full model containing these QTL explained 39.1% of phenotypic variance, with individual QTL model explaining 6.8–12.8% of variance. Allele effects were consistent with the difference between parents, as one of the CAN alleles increases length, while four of the SCO alleles increase length. CAN alleles were mostly dominant or additive to SCO alleles, the two exceptions being largely recessive.

In the PK population, two QTL were found on chromosome 1. A full model containing these QTL explained 18.5% of phenotypic variance, with individual QTL models explaining 7.9–11.0% of variance. Allele effects were mixed, as one of the PK alleles increases length, while one of the SCO alleles increases length. One PK allele is recessive to the SCO allele and the other shows indications of additive effect, but neither dominant nor recessive effects can be ruled out.

Across the three Brassica populations, the total of 11 significant marker-trait associations may reflect variation at as few as 10 different genetic loci. BK QTL on chromosome 4 correspond to CAN QTL.

Further, the 10 different genetic loci may be derived from as few as nine ancestral genes. The chromosome 4 QTL (BK: EW2B12s+10, CAN: EST453g+0) falls in a region that corresponds to the chromosome 7 QTL (BK: EW8C11a+1).

rrl2nd: In the BK population, five QTL were found on chromosomes 1, 4, 7(2), and 8. A full model containing these QTL explained 47.5% of phenotypic variance, with individual QTL models explaining 15.4–25.5% of variance. Allele effects were mixed, as three of the BK alleles increase the branch length, while two of the SCO alleles increase the length. BK alleles were mostly recessive or additive to SCO alleles, the one exception being largely dominant. No QTL was found to affect the first branch length in CAN and PK populations. The five different genetic loci may be derived from five ancestral genes.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The architecture of the curd (cluster of unopened flower buds) is an excellent example of the "morphological excursions" that appear to be typical of Brassica evolution. Enlargements of particular organs such as the inflorescence of cauliflower (B. oleracea subsp. botrytis) and broccoli (B. oleracea subsp. italica), the stem of kohlrabi (B. oleracea subsp. gongylodes), the apical bud of cabbage (B. oleracea subsp. capitata), and lateral buds of Brussels sprouts (B. oleracea subsp. gemmifera) (KALLOO and BERGH 1993 ) have led to the widespread use of Brassica as a major crop. The fact that such morphological divergence has occurred among taxa that can be intercrossed to generate fertile progenies makes Brassica a fascinating subject for genetic analysis.

The enlarged curd of cauliflower (B. oleracea subsp. botrytis) is under complex genetic control. By the comparison of QTL maps constructed from three different F₂ populations, we have identified a total of 86 QTL that control eight curd-related traits in B. oleracea. The three crosses shared one common parent, a rapid-cycling Brassica strain that does not generate a curd but instead quickly flowers. Among the three different parents, B. oleracea var. BK, which generates small curd compared to varieties CAN and PK, is believed to be the progenitor of broccoli and cauliflower (SONG and OSBORN 1992 ). CAN and PK, from Italy and India, respectively, represent curd-forming types of diverse morphology and geographical origin. The possibility that BK may represent a progenitor of types such as CAN and PK was supported by the observation that the BK population showed simpler genetic control of most traits; across the eight curd-related traits studied, only 20 QTL were found in BK, vs. 34 in CAN and 32 in PK.

Among the 86 QTL detected, 18 displayed dominance/additivity (d/a) absolute values larger than 1 (up to 8.95), suggesting overdominance. Except traits rrl2nd and cluw all the traits detected one to seven loci showing overdominance across three populations. Closer examination of the 18 QTL reveals that two genetic loci account for many of the cases of overdominance in different traits. One, on chromosome 3 near marker EST411c, has positive d/a values; the other, on chromosome 5 near marker EW8A11a, has negative d/a values. Both loci primarily appeared in the PK population. It has been observed that crosses made between parents of different origins generally exhibit greater heterosis than crosses between parents of similar origins in Brassica (GRAND and BEVERSDORF 1985 ; LEFORT-BUSON et al. 1987 ; BRANDLE and MCVETTY 1990 ; ALI et al. 1995 ). For most traits, we found more overdominant QTL in CAN and PK populations than in BK populations, further supporting the notion that BK may represent a progenitor of types such as CAN and PK (SONG and OSBORN 1992 ).

By the comparison of three diverse B. oleracea genotypes, we sought to investigate the extent to which curd-related traits were under common genetic control. Of particular interest was the notion that we might be able to discern particular genes that might have been fundamental to the evolution of the curd-like inflorescence, based on finding QTL at corresponding locations in all three populations. The discovery of a mutation that results in a "curd-like" phenotype in Arabidopsis (KEMPIN et al. 1995 ) has often been considered to support this notion. In a few cases, the locations of QTL did appear to correspond in all three populations. For example, the chromosome 4 QTL (BK: EW4D04w+6, CAN: EW9E10+12, PK: EW9B02x+1) appear to correspond to each other for trait dfb. However, the 86 QTL we found appeared to represent 67 different genetic loci, illustrating that most QTL differed between populations. Further, the comparison of ostensibly homeologous locations of duplicated Brassica chromosomal regions also showed mixed results. The 67 putatively different genetic loci accounting for the QTL found could be accounted for by 54 ancestral genes—i.e., about one-third of the QTL had possible homoeologs. On the basis of a genome of nine chromosomes with average length of 100 cM and QTL likelihood intervals averaging 25 cM in length, these levels of correspondence are somewhat higher than would be expected to occur by chance, but are not nearly so striking as the levels of correspondence found between much more distantly related members of the grass family (LIN et al. 1995 ; PATERSON et al. 1995 ). The level of diversity of QTL among these three populations was also reflected in comparisons to another population in which similar traits have been mapped. KENNARD et al. 1994 found three QTL associated with days to first bud that mapped to the same general regions of three of our QTL, but a fourth was on a linkage group where we found no QTL for this trait. A single QTL for bud cluster width did not correspond to any of the five QTL that we report for this trait. Each of two QTL for leaf lamina length did correspond to two of the five QTL that we found for this trait. One QTL each for leaf lamina width and petiole length corresponded to one of the QTL we found (for each trait), while a second QTL for each trait did not correspond to any of the three additional QTL we found for each trait. Two QTL associated with internode length corresponded in the two different studies, but a third QTL for this trait found by KENNARD et al. 1994 did not correspond to any of the four additional QTL that we found for this trait.

The discovery that the joint effects of two mutations in Arabidopsis, CAULIFLOWER and APETALA1, could produce a plant with a curd-like inflorescence (KEMPIN et al. 1995 ) is in contrast to the complex genetic control we report for Brassica. Other investigations of possible interactions between these genes in Brassica (CARR and IRISH 1997 ) show that each of the two duplicated copies of APETALA1 in B. oleracea contain insertions that lead to premature translation termination (LOWMAN and PURUGGANAN 1999 ). One CAULIFLOWER homolog that has been studied in Brassica does not contain a loss-of-function mutation such as would be predicted (KEMPIN et al. 1995 ), but does contain other changes that may have phenotypic consequences. Both the CAULIFLOWER and APET ALA1 mutations map to Arabidopsis chromosome 1. Due to the lack of polymorphism, we were unable to map the Arabidopsis genes in Brassica directly. However, from our comparative map, we predict they will be mapped to Brassica chromosomes 1 and 5, in locations that approximately coincide with the location of QTL that influence curd size, including rk1 QTL, EST566a+2 (CAN), and EW8F11c+0 (CAN). It should be noted that the "curd-like" mutation that occurred in Arabidopsis primarily referred to the phenotype of "semisterile" flowers and does not appear to be accompanied by the "enlarged inflorescence" observed in cauliflower or broccoli.

Our data (both herein and previously published; KOWALSKI et al. 1994 ) are consistent with the notion that the Brassica genome may be relatively rapidly evolving. The diversity of QTL we found among genotypes and generally low correspondence of QTL in ostensibly homeologous locations may reflect relatively rapid formation of new alleles in Brassica. The high level of duplication, perhaps triplication, in the Brassica genome would of course facilitate this.

Comparative data help to highlight how tools from Arabidopsis might be used to quickly explore for genes that may directly account for Brassica QTL. In these studies we have identified 14 dfb and 15 dofdbf QTL that associate with earliness in B. oleracea. Ten of these QTL fall in a homologous region involving the tops of B. oleracea chromosomes 1, 4, and 7, which correspond to a segment of Arabidopsis chromosome 5 that contains seven flowering mutations (tfl1, flc, tfl2, co, fy, art1, emf1); a homologous region in the middle of chromosome 7 and the lower half of chromosome 9, which corresponds to a region of Arabidopsis chromosome 1 containing the mutation efs; a homologous region involving the middle of chromosomes 3 and 8, which corresponds to a region of Arabidopsis chromosome 3 containing mutations hy2 and vrn1; a region of Brassica chromosome 3, which corresponds to a region of Arabidopsis chromosome 1 containing the mutation fha; a region in the middle of chromosome 5, which corresponds to a region of Arabidopsis chromosome 1 containing the mutation gi; and a region on the bottom of chromosome 7 in which comparative mapping data were insufficient to determine if there were corresponding mutations in Arabidopsis. As the Arabidopsis genome sequence unfolds, these and other candidates will provide a foundation for the use of "resequencing" methods to explore levels and patterns of allelic variation in Brassica that may help to implicate some of these in the genetic control of complex traits in well-defined Brassica gene pools (WANG et al. 1999 ).

Compared to two previous reports (OSBORN et al. 1997 ; BOHUON et al. 1998 ), we have identified at least six additional Brassica QTL corresponding to known Arabidopsis flower-timing mutations. The two dfb QTL EW8C11a+2 (BK) and WG3D11+14 (CAN) located on the top of Brassica chromosome 7, although very close to each other and both with very high LOD score (>20), appear to correspond to different loci. Evidence supporting this finding derives from the nonoverlapping LOD-support interval and the different gene action of two QTL (EW8C11a+2 is additive and WG3D11+14 is dominant). In addition, homologous QTL on chromosomes 7 (PK: EW6B07a+3) and 9 (PK: EST131a+17) possess different gene action, too, which might also suggest two different loci. The allelic variation of a homologous locus may reflect the different allelic effects among different crosses or may be the result of mutation (either gain-of-function mutation or loss-of-function mutation) at a particular locus. The latter, including homology-dependent gene silencing, i.e., cosuppression, caused by DNA methylation has been a common feature in many plant species including Arabidopsis (NAPOLI et al. 1990 ; FURNER et al. 1998 ).

Naturally occurring variants at QTL loci in Brassica may prove useful for new investigations of plant development. For example, the curd is essentially a cluster of branches near and surrounding the apical meristem. What makes it interesting is that the occurrence of those branches near the apical meristem violates the rule of apical dominance, which states that the germination of axillary buds should start from those buds that are distant from the apical meristem, i.e., near the ground. In contrast, there also exists a "spreading type" Brassica (vs. heading type, which makes curd), which generates many axillary branches, but does not make a curd. Thus, the trait of sidb was a measurement of Brassica conformation (spreading vs. heading) as well as a test of the apical dominance theory. QTL controlling sidb all fall in the same location as QTL controlling trait rk1, suggesting that QTL controlling the size of a curd in a heading-type Brassica may have the same effect in spreading-type Brassica. Variations in apical dominance may largely explain the difference between these phenotypes. Finally, Brussels sprouts (B. oleracea subsp. gemmifera), which produce many axillary buds that are not transformed to axillary branches, appear to have a phenotype of "neutral" apical dominance, lending further support to our theory.

On the basis of the progeny of crosses with B. oleracea var. Romanesco, WATTS 1966 suggested that acute angle of a curd was genetically controlled. Similar phenotypes were found in our Brassica populations. As illustrated in Fig 5C–E, the relative length of apical shoot and first branch will influence the acute angle of a curd, and therefore is a good indication of curd conformation. Noticeably, from the histogram, we can see that CAN and PK, both with large curds, have short rrl1st and rrl2nd with respect to SCO, while BK has roughly equal rrl1st and rrl2nd to SCO, which implies that there might be an interaction between curd conformation and curd size. Further crosses of SCO to other Brassica varieties should reveal more details.

View larger version (13K): In this window In a new window Download PPT slide   Figure 5. The bars indicate the first-rank branches (rk1) of a curd. As illustrated, A and B show that a curd with the same amount of first-rank branches could have different curd size (headw); C, D, and E show three different conformations of a curd, which can be documented by the different length between apical shoot and first branch. The angle between the thin lines is the acute angle of a curd, where curd C has an angle >180°, D = 180° and E <180°.

Head width (= curd diameter), generally considered to be the best nondestructive proxy for the mass of the curd (WATTS 1964 , WATTS 1965 ; CRISP 1977 ; FREEMAN and CRISP 1979 ), was closely related to both branching pattern and flowering time. In the BK and CAN populations, the closest correlate of head width was the number of first-rank branches. In the PK population, the closest correlate of head width was the number of days between the first bud and first flower. Earliness has long been a priority for crop breeders because reducing the time in the field also reduces the cost and risk of crop cultivation. Several QTL controlling earliness (dfb, dofdbf) fall into the same chromosomal regions as QTL that control the size of the curd (rk1, headw). Such an association could be caused by pleiotropic effects or by tightly linked genes. From an applied standpoint, the breeding of early varieties that also produce high yield should utilize varieties carrying QTL for curd size that are not associated with late-flowering QTL. For example, PK carries rk1 QTL (EST217+3) on chromosome 1, where no late-flowering QTL were detected.

While the number of first-rank branches, rk1, appears to be a major component of curd width, curd density also needs to be taken into account. As illustrated in Fig 5A and Fig B, both curds possess five first-rank branches; however, they have different width, due to the growth angle of the branch relative to the stem. Thus, measuring only the curd width to represent the mass of a curd could be misleading. Therefore we measured rk1, cluw, and headw at the same time to better represent curd morphology. By comparing the QTL associated with rk1, cluw, and headw, we might be able to identify QTL that influence the density of the curd. For example, in BK population, there was only one QTL (C7: WG3D11+11) mapped for trait rk1, and two QTL (C4: EW4D04w+7 and C7: EW8C11a+4) were identified for trait headw. Comparing two traits, both QTL that mapped to C7 were in the same chromosomal region. The chromosome 4 QTL for headw (EW4D04w+7) has a log likelihood of 3.05 but only 1.40 for trait rk1, which is indicative but not statistically significant. Thus the QTL EW4D04w+7 is likely a component of curd density. From the above, one can see the power of QTL mapping. By comparing associated phenotypes, one can either confirm an identified QTL or detect new alleles that differentiate the phenotype, so that a complex trait can be analyzed in detail.

The Brassica genus, and B. oleracea in particular, illustrates the "specialized" features that distinguish major crops from A. thaliana, a botanical model for molecular genetic studies. The prospect of a complete sequence for Arabidopsis in the near future will open new doors into many aspects of plant biology by using comparative approaches, with special importance for its close relatives such as Brassica. With the morphological diversity of Brassica and the abundance of Arabidopsis resources, we believe that establishing the Cruciferae as a model for dissecting crop development will bear fruit soon.

	ACKNOWLEDGMENTS

We thank Scott Davis, Mark Hussey, and Terry Thomas for critical suggestion, Jim McFerson, Stephen Kresovich, Kenneth Feldmann, and JoVan Currie for technical help, and the Pioneer HiBred Production Ltd. for providing a subset of the DNA probes used. We thank the Texas Higher Education Coordinating Board, United States Department of Agriculture Plant Genome Program, and Texas Agricultural Experimental Station for funding.

Manuscript received August 18, 1999; Accepted for publication April 25, 2000.

	LITERATURE CITED

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

ALI, M., L. O. COPELAND, and S. G. ELIAS, 1995  Relationship between genetic distance and heterosis for yield and morphological traits in winter canola (Brassica napus L.). Theor. Appl. Genet. 91:118-121.

BOHUON, E. J. R., L. D. RAMSAY, J. A. CRAFT, A. E. ARTHUR, and D. F. MARSHALL et al., 1998  The association of flowering time quantitative trait loci with duplicated regions and candidate loci in Brassica oleracea.. Genetics 150:393-401[Abstract/Free Full Text].

BRANDLE, J. E. and P. E. B. MCVETTY, 1990  Geographical diversity, parental selection and heterosis in oilseed rape. Can. J. Plant Sci. 70:935-940.

CARR, S. M. and V. F. IRISH, 1997  Floral homoeotic gene expression defines developmental arrest stages in Brassica oleracea L. vars. botrytis and italica. Planta 201:179-188[Medline].

CRISP, P., 1977  Genotype x environment interactions in early winter cauliflowers in south-west Britain. J. Hort. Sci. 52:357-366.

DETJEN, L. R., 1926  A preliminary report on cabbage breeding. Proc. Am. Soc. Hort. Sci. 23:325-332.

FALCONER, D. S., 1989 Introduction to Quantitative Genetics. Longman Scientific & Technical Press, New York.

FREEMAN, G. H. and P. CRISP, 1979  The use of related variables in explaining genotype-environment interactions. Heredity 42:1-11.

FURNER, I. J., A. SHEIKH, and C. E. COLLETT, 1998  Gene silencing and homology-dependent gene silencing in Arabidopsis: genetic modifers and DNA methylation. Genetics 149:651-662[Abstract/Free Full Text].

GRAND, I. and W. D. BEVERSDORF, 1985  Heterosis and combining ability estimates in spring oilseed rape (Brassica napus L.). Can. J. Genet. Cytol. 27:472-478.

KALLOO, G., and B. O. BERGH, 1993 Genetic Improvement of Vegetable Crops. Pergamon Press, Oxford.

KEMPIN, S. A., B. SAVIDGE, and M. F. YANOFSKY, 1995  Molecular basis of the cauliflower phenotype in Arabidopsis. Science 267:522-525[Abstract/Free Full Text].

KENNARD, W. C., M. K. SLOCUM, S. S. FIGDORE, and T. C. OSBORN, 1994  Genetic analysis of morphological variation in Brassica oleracea using molecular markers. Theor. Appl. Genet. 87:721-732.

KOWALSKI, S. P., T.-H. LAN, K. A. FELDMANN, and A. H. PATERSON, 1994  Comparative mapping of Arabidopsis thaliana and Brassica oleracea chromosomes reveals islands of conserved organization. Genetics 138:499-510[Abstract].

KRISTOFFERSON, K. B., 1924  Contributions to the genetics of Brassica oleracea.. Hereditas 5:297-364.

LAGERCRANTZ, U., J. PUTTERILL, G. COUPLAND, and D. LYDIATE, 1996  Comparative mapping in Arabidopsis and Brassica, fine scale genome collinearity and congruence of genes controlling flowering time. Plant J. 9:13-20[Medline].

LAN, T.-H., T. A. DELMONTE, K. P. REISCHMANN, J. HYMAN, and S. KOWALSKI et al., 2000  An EST-enriched comparative map of Brassica oleracea and Arabidopsis thaliana.. Genome Res. in press.

LANDER, E. and D. BOTSTEIN, 1989  Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics 121:185-199[Abstract/Free Full Text].

LANDER, E., P. GREEN, J. ABRAHAMSON, A. BARLOW, and M. DALY et al., 1987  MAPMAKER: an interactive computer package for constructing primary genetic linkage maps of experimental and natural populations. Genomics 1:174-181[Medline].

LEFORT-BUSON, M., B. GUILLOT-LEMOINE, and Y. DATTEE, 1987  Heterosis and genetic distance in rapeseed (Brassica napus L.): crosses between European and Asiatic selfed lines. Genome 29:413-418.

LIN, Y., K. SCHERTZ, and A. PATERSON, 1995  Comparative analysis of QTL affecting plant height and maturity across the Poaceae, in reference to an interspecific sorghum population. Genetics 141:391-411[Abstract].

LOWMAN, A. C. and M. D. PURUGGANAN, 1999  Duplication of the Brassica oleracea APETALA1 floral homoeotic gene and the evolution of domesticated cauliflower. J. Hered. 90:514-520[Abstract/Free Full Text].

NAPOLI, C. C., C. LEMIEUX, and R. JORGENSEN, 1990  Introduction of a chimeric chalcone synthase gene into petunia results in reversible co-suppression of homologous genes in trans.. Plant Cell 2:279-289[Abstract/Free Full Text].

OSBORN, T. C., C. KOLE, I. A. P. PARKIN, A. G. SHARPE, and M. KUIPER et al., 1997  Comparison of flowering time genes in Brassica rapa, B. napus and Arabidopsis thaliana.. Genetics 146:1123-1129[Abstract].

PATERSON, A. H., S. DAMON, J. D. HEWITT, D. ZAMIR, and S. E. LINCOLN et al., 1991  Mendelian factors underlying quantitative traits in tomato: comparison over species, generations, and environments. Genetics 127:181-197[Abstract].

PATERSON, A. H., Y. LIN, Z. LI, K. SCHERTZ, and J. DOEBLEY et al., 1995  Convergent domestication of cereal crops by independent mutations at corresponding genetic loci. Science 269:1714-1718[Abstract/Free Full Text].

PATERSON, A. H., T.-H. LAN, K. P. REISCHMANN, C. CHANG, and Y.-R. LIN et al., 1996  Toward a unified map of higher plant chromosomes, transcending the monocot-dicot divergence. Nat. Genet. 14:380-382[Medline].

PEASE, M. S., 1926  Genetic studies in Brassica oleracea.. J. Genet. 16:363-387.

PRAKASH, S. and K. HINATA, 1980  Taxonomy, cytogenetics, and origin of crop Brassica, a review. Opera. Bot. 55:1-59.

SONG, K. and T. C. OSBORN, 1992  Polyphyletic origins of Brassica napus: new evidence based on organelle and nuclear RFLP analyses. Genome 35:992-1001.

WANG, R., A. STEC, J. HEY, L. LUKENS, and J. DOEBLEY, 1999  The limits of selection during maize domestication. Nature 398:236-239[Medline].

WATTS, L. E., 1964  Studies of maturity in F1 and F2 generations of cauliflowers from crosses between summer, autumn and winter types. J. Hort. Sci. 39:84-91.

WATTS, L. E., 1965  The inheritance of curding periods in early summer and autumn cauliflower. Euphytica 14:83-90.

WATTS, L. E., 1966  Conformation of cauliflower curds. Euphytica 15:111-115.

WILLIAMS, P. H. and C. B. HILL, 1986  Rapid-cycling populations of Brassica.. Science 232:1385-1389[Abstract/Free Full Text].

This article has been cited by other articles:

				D. V. Duclos and T. Bjorkman Meristem identity gene expression during curd proliferation and flower initiation in Brassica oleracea J. Exp. Bot., February 1, 2008; 59(2): 421 - 433. [Abstract] [Full Text] [PDF]

				Y. Long, J. Shi, D. Qiu, R. Li, C. Zhang, J. Wang, J. Hou, J. Zhao, L. Shi, B.-S. Park, et al. Flowering Time Quantitative Trait Loci Analysis of Oilseed Brassica in Multiple Environments and Genomewide Alignment with Arabidopsis Genetics, December 1, 2007; 177(4): 2433 - 2444. [Abstract] [Full Text] [PDF]

				P. Lou, J. Zhao, J. S. Kim, S. Shen, D. P. Del Carpio, X. Song, M. Jin, D. Vreugdenhil, X. Wang, M. Koornneef, et al. Quantitative trait loci for flowering time and morphological traits in multiple populations of Brassica rapa J. Exp. Bot., November 28, 2007; (2007) erm255v1. [Abstract] [Full Text] [PDF]

				T.-J. Yang, J. S. Kim, S.-J. Kwon, K.-B. Lim, B.-S. Choi, J.-A Kim, M. Jin, J. Y. Park, M.-H. Lim, H.-I. Kim, et al. Sequence-Level Analysis of the Diploidization Process in the Triplicated FLOWERING LOCUS C Region of Brassica rapa PLANT CELL, June 1, 2006; 18(6): 1339 - 1347. [Abstract] [Full Text] [PDF]

				C. J. Murren and M. Pigliucci Morphological responses to simulated wind in the genus Brassica (Brassicaceae): allopolyploids and their parental species Am. J. Botany, May 1, 2005; 92(5): 810 - 818. [Abstract] [Full Text] [PDF]

				I. M. L. Somorjai, R. G. Danzmann, and M. M. Ferguson Distribution of Temperature Tolerance Quantitative Trait Loci in Arctic Charr (Salvelinus alpinus) and Inferred Homologies in Rainbow Trout (Oncorhynchus mykiss) Genetics, November 1, 2003; 165(3): 1443 - 1456. [Abstract] [Full Text] [PDF]

				L. Lukens, F. Zou, D. Lydiate, I. Parkin, and T. Osborn Comparison of a Brassica oleracea Genetic Map With the Genome of Arabidopsis thaliana Genetics, May 1, 2003; 164(1): 359 - 372. [Abstract] [Full Text] [PDF]