Characterization of metabolite quantitative trait loci and metabolic networks that control glucosinolate concentration in the seeds and leaves of Brassica napus


Author for correspondence:
Jinling Meng
Tel: +86 27 87282457


  • Glucosinolates are a major class of secondary metabolites found in the Brassicaceae, whose degradation products are proving to be increasingly important for human health and in crop protection.
  • The genetic and metabolic basis of glucosinolate accumulation was dissected through analysis of total glucosinolate concentration and its individual components in both leaves and seeds of a doubled-haploid (DH) mapping population of oilseed rape/canola (Brassica napus).
  • The quantitative trait loci (QTL) that had an effect on glucosinolate concentration in either or both of the organs were integrated, resulting in 105 metabolite QTL (mQTL). Pairwise correlations between individual glucosinolates and prior knowledge of the metabolic pathways involved in the biosynthesis of different glucosinolates allowed us to predict the function of genes underlying the mQTL. Moreover, this information allowed us to construct an advanced metabolic network and associated epistatic interactions responsible for the glucosinolate composition in both leaves and seeds of B. napus.
  • A number of previously unknown potential regulatory relationships involved in glucosinolate synthesis were identified and this study illustrates how genetic variation can affect a biochemical pathway.


Glucosinolates are found in all members of the family Brassicaceae and are a class of secondary metabolites, of which > 120 different types have been found within plants. Glucosinolates are hydrolyzed by myrosinases following plant wounding (Husebye et al., 2002). Following hydrolysis the breakdown products can have beneficial effects such as preventing cancer in humans and enhancing plant protection (Fahey et al., 2001; Talalay & Fahey, 2001; Tierens et al., 2001). Deleterious effects have also been reported and the types of glucosinolates produced by the plant are important in determining the overall impact (Fenwick & Curtis, 1980; Griffiths et al., 1998; Vaughn et al., 2005). The discovery of a forage cultivar of Brassica napus, Bronowski, with reduced amounts of glucosinolates in its seeds was a major breakthrough in the breeding of modern oilseed crops (Kondra & Stefansson, 1970). Most modern varieties of B. napus are referred to as ‘double lows’ as they have seeds with low glucosinolate and erucic acid. The reduced concentrations of glucosinolates are often a consequence of crossing programs which have used the Bronowski germplasm. The glucosinolate concentration within their seeds has been reduced dramatically from > 100 to < 20 μmol g−1 (Toroser et al., 1995). Moreover, this reduction is thought to be associated with a concomitant decrease in glucosinolate production in leaves (Mithen, 1992; Li et al., 1999). A major goal of B. napus plant breeders is to further reduce the concentration of deleterious glucosinolate within seeds in which the cake is to be used for fodder and yet retain the protective effects of high glucosinolate concentrations in the vegetative organs. For this to be achieved, it is necessary for us to develop a better understanding of the processes underlying glucosinolate biosynthesis and accumulation in leaves and seeds.

The chemical structure of glucosinolates is formed around a common core together with a variable side chain. For each of the three different classes of glucosinolates, the aliphatic, aromatic and indolyl glucosinolates, the side chain is derived from a different type of amino acid precursor (Zukalová & Vašák, 2002; Grubb & Abel, 2006; Halkier & Gershenzon, 2006). The corresponding pathways, involved in glucosinolate metabolism, have been reported for the Brassicaceae (Fig. 1). It is thought that the glucosinolates are synthesized mainly in vegetative organs such as young leaves and silique walls, and then transported actively to embryos through the phloem by unknown transporters; however, synthesis in immature seeds has also been proposed (Toroser et al., 1995; Du & Halkier, 1998; Chen & Halkier, 2000; Chen et al., 2001b; Kliebenstein et al., 2001b). Blocking the transport of glucosinolates from vegetative organs to the seeds could provide a means of reducing glucosinolate concentrations in seeds without affecting other tissues.

Figure 1.

Glucosinolate metabolic pathways in Brassicaceae. Pathway A: precursors of aliphatic and aromatic glucosinolates are synthesized by a series of reactions (condensation, isomerization and oxidative decarboxylation mainly), which elongate the side chains of amino acids. Pathway B: formation of basic glucosinolates from amino acids and chain-elongated homoamino acids is catalyzed by a series of enzymes. Pathway C: glucosinolates are translocated from vegetative organs to seeds by the unknown glucosinolate transporter to satisfy the huge demand for glucosinolate in seeds. Pathway D: side chains of basic glucosinolates are modified by oxidation and elimination, etc., to form other modified glucosinolates in leaves and seeds. Pathway E: when plants are injured, glucosinolates are hydrolyzed by myrosinase to the breakdown products that are involved in stress defense. The figure is constructed following several publications (Iqbal et al., 1995; Chen & Andreasson, 2001a; Mikkelsen et al., 2002; Grubb & Abel, 2006; Halkier & Gershenzon, 2006; Padilla et al., 2007; Hansen et al., 2008; Gigolashvili et al., 2009; Mugford et al., 2009; Sawada et al., 2009; Dixon et al., 2010; Sonderby et al., 2010; Yatusevich et al., 2010).

Quantitative trait loci (QTL) have been determined to control different glucosinolates in the leaves and seeds of Arabidopsis. Kliebenstein et al. (2001a), found three tissue-specific QTL for aliphatic glucosinolates in both leaves and seeds and a similar number were found for the indolyl glucosinolates, with only one QTL identified as being active in both organs. These results indicate that genes underlying these QTL seem to be responsible for glucosinolate accumulation in the specific tissue rather than both tissues. Subsequently, the genes AOP2–AOP3 were cloned through fine mapping of these QTL and were demonstrated to be involved in glucosinolate modification (Kliebenstein et al., 2001c). Other gene families involved in the control of glucosinolate metabolism in Arabidopsis have been identified through the analysis of loss-of-function mutations such as the CYP79 gene (Hull et al., 2000; Reintanz et al., 2001; Chen et al., 2003; Mikkelsen et al., 2003). Such results together with comparative analysis between phenotypic QTL and expression QTL (eQTL) has resulted in the proposal of a complex pathway for glucosinolate biosynthesis and regulation in Arabidopsis (Kliebenstein et al., 2006; Wentzell et al., 2007). Numerous unknown metabolic regulatory relationships that are complementary to this network and are based on QTL co-location information have also been predicted using metabolic analysis (Keurentjes et al., 2006). A piece of software, MetaNetwork, has subsequently been developed which facilitates such approaches (Fu et al., 2007).

In addition to the work carried out in Arabidopsis, our current knowledge about the accumulation of glucosinolates has been enhanced by QTL mapping in Brassica. In Brassica oleracea (CC, 2n = 18), four QTL, together with the underlying candidate genes, have been identified as key players in the biosynthesis of multiple glucosinolates in the leaves of this species (Li & Quiros, 2001, 2002, 2003; Li et al., 2003). Recently, up to 22 QTL were identified for the accumulation of glucosinolates in the leaves of B. rapa (AA, 2n = 20) (Lou et al., 2008). In QTL studies of total glucosinolate accumulation in the seeds of B. napus, seven QTL have been identified on several linkage groups (Toroser et al., 1995; Uzunova et al., 1995; Howell et al., 2003; Zhao & Meng, 2003; Quijada et al., 2006). Some common QTL were detected between many of these studies which may be a result of the common ancestry of these populations. Most populations are derived from a shared ancestor with low seed glucosinolate concentration (Sharpe & Lydiate, 2003; Hasan et al., 2008). Interestingly, little work has been done on the genetic variation of glucosinolates in the leaves and other vegetative organs, or on the metabolic network of glucosinolate synthesis in this tetraploid species. It is likely that the regulation of glucosinolate synthesis will be very complex compared with that of Arabidopsis and diploid Brassica crops.

In this study, we have attempted to address this gap in our knowledge of the genetic control of glucosinolate concentration in leaves of B. napus and the metabolic network of glucosinolate synthesis. This has been achieved through the measurement of total glucosinolate and individual glucosinolates in seeds and leaves within a doubled-haploid (DH) mapping population of B. napus.

Materials and Methods

Plant materials and growing environments

The TN (Tapidor and Ningyou7) DH population was derived from the European winter-type cv Tapidor, which has a low glucosinolate concentration (Sharpe & Lydiate, 2003), and Ningyou7, a Chinese semiwinter-type cultivar with high glucosinolate concentration (Qiu et al., 2006). The parental cultivars and 202 DH lines from the population were grown at Wuhan (W1) in China for 3 yr (September–May, 2003–2006) and at Daye (near Wuhan, W2) for 1 yr (September–May, 2005–2006), coded as 4W1, 5W1, 6W1, and 6W2, respectively. In each of the four field trials, all lines were planted according to a randomized complete-block design of three replicates, with each plot containing 30 plants as described by Shi et al. (2009).

Sampling and determination of glucosinolate concentration

At approx. 130 d after sowing when the plants had eight true leaves and before bolting, the fifth leaf from the bottom was collected from 12 plants selected from the middle of each plot, and stored at −80°C. At maturity, seeds were harvested from the same 12 plants and fully dried. The samples of leaves or mature seeds from each plot were combined before measurement of the glucosinolates. The glucosinolate profiles in seeds and leaves were analyzed as described by Agerbirk et al. (2001) and Wathelet et al. (2004), respectively, with minor modifications. After the petioles were removed, leaves were ground under liquid nitrogen, and 1.50 g of each fresh sample was extracted twice with 10 ml of boiling 70% methanol. The seed samples were crushed, and 200 mg of each sample was extracted twice with 2 ml of boiling 70% methanol. The concentration of glucosinolate in the leaves and seeds was determined by high-performance liquid chromatography (Waters 2487/600/717) using the ISO9167-1 (1992) standard method.

Genetic map and QTL analysis

The TN genetic map contains 786 markers, including simple sequence repeats (SSR), single nucleotide polymorphism (SNP), sequence-tagged site (STS), single-strand conformational polymorphism (SSCP), restriction fragment length polymorphism (RFLP), cleaved amplified polymorphic sequences (CAPS), amplified fragment length polymorphism (AFLP) and methylation sensitive AFLP (MS-AFLP) markers. It contains 19 linkage groups (A1–A10 and C1–C9), spanning 2117.2 cM with an average distance of 2.7 cM between markers (Shi et al., 2009). Fifty-three genes (Supporting Information, Table S1) known to be involved in glucosinolate metabolism in A. thaliana were obtained from the Arabidopsis Information Resource (TAIR) website ( and mapped on the genetic map by in silico mapping according to the description of Long et al. (2007) and Shi et al. (2009).

Quantitative trait locus mapping was carried out through composite interval mapping using WinQTL cartographer 2.5 (Zeng, 1994; The default genetic distance and walking speed were set to 5 and 2 cM, respectively, for defining individual QTL. A significance threshold for QTL at the level = 0.05 was determined through permutation analysis using 1000 repetitions (Churchill & Doerge, 1994), and logarithm of the odds ratio (LOD) values (2.7–3.5) were used to identify the QTL, which were designated as significant QTL. The confidence intervals were set to 95%.

Significant QTL were integrated using a ‘two-round’ strategy of QTL meta-analysis (Shi et al., 2009) with the BioMercator2.1 software (Goffinet & Gerber, 2000; Arcade et al., 2004). In the first round, significant QTL for individual traits that were detected repeatedly in multiple environments were integrated to give consensus QTL. In the second round, overlapping consensus QTL for the different traits in leaves and seeds were integrated into new QTL. For example, the process of QTL integration in a region of 12–41.7 cM in the A9 linkage group was shown in Fig. 2. A method is provided by this software to calculate 95% confidence intervals for the integrated QTL,

image(Eqn 1)

in which CI represents the confidence intervals of the integrated QTL, inline image is the variance of position of the QTLi, and k is the total number of the integrated QTL.

Figure 2.

Demonstration of the process of quantitative trait loci (QTL) integration in a region of 12–41.7cM in the A9 linkage group. Original QTL identified from different environments are shown by curves above the line of linkage group, and their confidence intervals are shown by dotted lines of the same color below the line of the linkage group. The solid colored lines show the confidence intervals of the integrated QTL by meta-analysis. LOD, logarithm of the odds ratio.

Given that the new QTL had pleiotropic effects on the metabolism of multiple glucosinolates in the key metabolic pathways, these QTL were named mcQTL (metabolite QTL for complex products). Nonoverlapping consensus QTL were detected that affected only one specific glucosinolate product in the metabolic process and were defined as msQTL (metabolite QTL for a specific product). Thereafter, a mQTL (metabolite QTL) could be regarded as a mcQTL or msQTL.

The glucosinolate traits were named as follows: the nomenclature began with the code for the individual glucosinolate (Table 1), and was followed by the letter ‘S’ (seeds) or ‘L’ (leaves), e.g. Ali-4C-PRO-S and Aro-GST-L. The nomenclature for the mQTL began with ‘q’ (abbreviation of QTL), followed by ‘mcG’ (mcQTL for the glucosinolate concentration) or ‘msG’ (msQTL for the glucosinolate concentration), and then the linkage group number (A1–A10 and C1–C9) and the serial number of the QTL (a, b, c…) in the linkage group, such as q.mcG-A1a and q.msG-C1c.

Table 1.   Denomination of different glucosinolates
Serial numberCommon nameSystematic nameAbbreviation
 1Total glucosinolateGlucosinolateTGS
 2The sum of Ali-4C and Ali-5CAliphatic glucosinolateAli
 3The sum of Ali-4C-GRA, Ali-4C-GNA and Ali-4C-PRO4C-aliphatic glucosinolateAli-4C
 4Glucoraphanin4-methylsulfinylbutyl glucosinolateAli-4C-GRA
 5Gluconapin3-butenyl glucosinolateAli-4C-GNA
 6Progoitrin2-hydroxy-3-butenyl glucosinolateAli-4C-PRO
 7The sum of Ali-5C-GAL, Ali-5C-GBN and Ali-5C-GNL5C-aliphatic glucosinolateAli-5C
 8Glucoalyssin5-methylsulfinylamyl glucosinolateAli-5C-GAL
 9Glucobrassicanapin4-pentenyl glucosinolateAli-5C-GBN
10Gluconapoleiferin2-hydroxy-4-pentenyl glucosinolateAli-5C-GNL
11Gluconasturtiin2-phenylethyl glucosinolateAro-GST
12The sum of Ind-GBS, Ind-4OH,Ind-4ME and Ind-NEOIndolyl glucosinolateInd
13Glucobrassicin3-indolyl-methyl glucosinolateInd-GBS
144-hydroxyglucobrassicin4-hydroxy-3-indolylmethyl glucosinolateInd-4OH
154-methoxyglucobrassicin4-methoxy-3-indolylmethyl glucosinolateInd-4ME
16Neoglucobrassicin1-methoxy-3-indolylmethyl glucosinolateInd-NEO

The analysis of genetic correlations and partial correlation

A metabolic network of the glucosinolates was constructed using MetaNetwork software (Fu et al., 2007), which explored potential associations between metabolites by computing correlations among QTL for different traits in two steps. Firstly, the genetic correlations between two glucosinolates were computed on the basis of colocation of the QTL, and then the partial correlation between pairs of glucosinolates was calculated to eliminate spurious correlation, which could reveal potential regulatory relationships among different glucosinolates. The threshold of partial correlation at an α level of 0.05 after Bonferroni correction was estimated by doing 10 000 permutation tests, and a correlation coefficient at 0.1 was regarded as significant (Fu et al., 2007).

The analysis of epistatic interaction

The maximum-likelihood estimation method in the software QTLmapper V2.0 (Wang et al., 1999; was used to identify the whole-genome epistatic interactions, based on a mixed linear model. The walking speed and likelihood ratio value were set to 1 cM and 0.005, respectively, to determine the presence of putative epistatic interactions. The significance of the epistatic effect was further tested by running the submenu of the Bayesian test (P < 0.001).

The nomenclature for the epistatic interactions began with ‘EI’ (epistatic interaction), followed by the position of two interacting loci in parentheses. The position of the interacting locus represented the linkage group number (A1–A10 and C1–C9) and the peak position of the interacting locus in the linkage group, and then connected the positions of two interacting loci with ‘/’, such as EI(A3-135.7/C2-41.1).


Phenotypic variation and QTL identification for glucosinolate concentration

Tapidor and Ningyou7 were found to differ significantly in the total glucosinolate and the aliphatic glucosinolate concentrations in both seeds and leaves; the total glucosinolate concentrations were 15 and 20 times higher in seeds compared with leaves, respectively. The total glucosinolate concentration in the leaves and seeds of Ningyou7 were more than 2.5 and 4 times higher, respectively, than that in Tapidor (Fig. 3a). A continuous skewed distribution was observed for total glucosinolate concentration and the majority of the individual glucosinolates in seeds and leaves in the TN DH population (Fig. 3b).

Figure 3.

Distribution of glucosinolate concentration in the two parents (a) and the TN population (b). (a) Concentration of glucosinolates in the seeds and leaves of Ningyou7 and Tapidor; the standard errors for the total seed glucosinolate concentration in four environments and the total leaf glucosinolate concentration in two environments are shown at the top of each column. (b) Frequency distributions of total glucosinolate in seeds and leaves in the TN population.

Four hundred and thirty-six significant QTL were detected for the 16 traits which included total glucosinolate and the individual glucosinolates in leaves and seeds from the four field trials (Table S2). Among them, 59% of the significant QTL (257) were found in two or more of the field trials and these were integrated, resulting in 85 nonredundant consensus QTL; the remaining 41% significant QTL (179) were only identified in a single trial and were also considered as consensus QTL. A combined total of 264 consensus QTL were identified in the genome. Of these 193 were seed-specific, 59 leaf-specific, and 12 were found in both organs. These QTL were distributed throughout the genome, being mapped to 18 of the 19 linkage groups (Fig. 4, Table 2). The confidence intervals of 217 of the consensus QTL overlapped with other QTL and as a result these were merged, giving 58 mcQTL. These might act early in the glucosinolate metabolic pathways or play an important role in the late steps of aliphatic glucosinolate biosynthesis with a pleiotropic effect on several glucosinolates. The remaining 47 consensus QTL were defined as msQTL in which the underlying genes are thought to have a specific effect on a single glucosinolate product. In total, 105 mQTL (including mcQTL and msQTL) were identified (Fig. S1, Table S3).

Figure 4.

Overview of the distribution of quantitative trait loci (QTL) on the linkage groups. The 19 sectors of the disc represent different linkage groups of Brassica napus. The black lines perpendicular to the linkage group bars at the outer edge represent molecular markers. The bars on the 16 dotted arc lines (from the outside to the inside) indicate the consensus QTL in turn for the total glucosinolate concentration and the 15 glucosinolate profiles. Bars on the inner edge represent the mQTL. The confidence interval of each QTL is shown by the width of the bar. QTL identified in different organs are represented by different colors. The catalog of QTL is shown in Table 2. The four mQTL inside the blue trapezia are considered as major QTL which were mentioned in the ‘Results’ section (see Table 3). In the center, the presumed relationships between mcQTL and metabolic pathways was estimated by genetic correlations among different glucosinolates as described by Barker et al. (2007) and Zhao et al. (2008). The capitals (A, B and D) in the small cycle indicate the glucosinolate metabolic pathways, which correspond to those in Fig. 1. The connecting dashed lines represent the presumed relationships between mcQTL and metabolic pathways.

Table 2.   Catalog of the quantitative trait loci (QTL) and epistatic interactions
OrganQTLEpistatic interactions
Consensus QTLmcQTLmsQTLWithin the A genomeWithin the C genomeBetween the A/C genomesQTL/QTL interactionsSingle interacting locus overlapping with QTLNonoverlapping with QTL
  1. mcQTL, metabolite QTL for complex products; msQTL, metabolite QTL for a specific product.


In order to determine if any of the genes underlying the mQTL had been previously identified, 53 Arabidopsis genes known to have roles in glucosinolate biosynthesis and regulation were aligned using comparative genomics to collinear regions identified between the Arabidopsis genome and the TN linkage groups. In all, 41 mQTL (39% of the total mQTL identified) coincided with regions that contained 27% of the orthologs (Table S3, Fig. S1), which suggested that most of these genes were not identifiable using this population.

Seventy-one of the 105 mQTL were seed-specific, with 50% of these being detected in different trials. Twenty-one of these are responsible for the total glucosinolate accumulation (Table S3). Two of these mQTL, q.mcG-A9a and q.mcG-A9b, explained 7.9–21.4 and 8.4–38.5% of the phenotypic variation for the total glucosinolate in seeds from the different field trials, respectively (Table 3), and also had a pleiotropic effect on several glucosinolates within the seed. Another mQTL, q.mcG-C2c, was found to affect seven traits, including the total seed glucosinolate concentration. 4-Methylsulfinylbutyl and 5-methylsulfinylamyl glucosinolates were principally affected, both of which are precursor of other aliphatic glucosinolates, which suggests that q.mcG-C2c affected the total glucosinolate accumulation in seeds through its action early in aliphatic glucosinolate synthesis (Table 3). These results strongly suggest that the genes underlying these QTL are likely to be involved in the biosynthesis of glucosinolates within the seeds (pathway A, B or D in Fig. 1) or in the transportation of glucosinolates from vegetative organs to embryos (pathway C in Fig. 1).

Table 3.   The list of major quantitative trait loci (QTL) for glucosinolates
mcQTLConsensus QTLTraitLinkage groupLODQuantitative variancePeak positionConfidence intervalAdditiveExperiment
  1. Ali, aliphatic glucosinolate; Ali-4C, 4C-aliphatic glucosinolate; Ali-4C-GNA, 3-butenyl glucosinolate; Ali-4C-GRA, 4-methylsulfinylbutyl glucosinolate; Ali-4C-PRO, 2-hydroxy-3-butenyl glucosinolate; Ali-5C, 5C-aliphatic glucosinolate; Ali-5C-GAL, 5-methylsulfinylamyl glucosinolate; Ali-5C-GBN, 4-pentenyl glucosinolate; Ali-5C-GNL, 2-hydroxy-4-pentenyl glucosinolate; Aro-GST, 2-phenylethyl glucosinolate; Ind, Indolyl glucosinolate; Ind-4ME, 4-methoxy-3-indolylmethyl glucosinolate; Ind-4OH, 4-hydroxy-3-indolylmethyl glucosinolate; Ind-GBS, 3-indolyl-methyl glucosinolate; Ind-NEO, 1-methoxy-3-indolylmethyl glucosinolate; TGS, total glucosinolate.

q.mcG-A4c  A4  94.493.9–95  
q.mcG-A9a  A9  14.113.5–14.6  
q.mcG-A9b  A9  25.224.6–25.8  
q.mcG-A9c  A9  35.734.7–36.7  
  C2  94.393.8–94.8  

Seventeen mQTL were only detected in leaves, which implied their specific role in the biosynthesis of glucosinolates (pathway A, B or D in Fig. 1). One of these mQTL, q.msG-C8b, was responsible for 8.1% of the phenotypic variation in the total glucosinolate concentrations in the leaves.

Seventeen mQTL were identified to be common between both leaves and seeds, which suggests that the genes that underlie these mQTL are likely to be involved in the core pathway, the side-chain elongation pathway, or the modification pathway of glucosinolate biosynthesis in different organs (pathway A, B or D in Fig. 1). One mQTL, q.mcG-A4c, was found to be responsible for 26.7–49.7 and 48.6–63.3% of the phenotypic variation in 4-pentenyl glucosinolate concentration in leaves and seeds from the different field trials, respectively (Table 3). It showed that the gene or genes underlying this locus has a similar effect within both organs.

Construction of a glucosinolate metabolic network and presumed function of the genes that underlie mQTL in the metabolic network

The complicated relationships among the different glucosinolates and in the different organs were analyzed on the basis of partial correlations computed by MetaNetwork software. Twelve pairs of significant partial correlations were identified that affect either the aliphatic glucosinolates or the indolic glucosinolates (Fig. 5). An advanced metabolic network that expands on the known pathways for glucosinolate metabolism was constructed using all of the partial correlations between different aliphatic or different indolyl glucosinolates. Five potential novel regulatory relationships were revealed (Fig. 6), which may lead to identification of genes involved in the biosynthesis of aliphatic and indolyl glucosinolates in both organs. For the indolyl glucosinolates, only two regulatory relationships of the biosynthetic metabolism were detected (Fig. 6b). There was no significant difference between the two parents in the concentration of 4-hydroxy-3-indolyl-methyl and 1-methoxy-3-indolylmethyl glucosinolate in leaves and seeds, which meant that other known regulatory relationships for the indolyl glucosinolates could not be verified in this population.

Figure 5.

Demonstration of genetic correlations and partial correlations among different glucosinolates based on the colocation of quantitative trait loci (QTL).

Figure 6.

The advanced metabolic network of glucosinolate in Brassica napus. (a) The advanced metabolic network of aliphatic glucosinolate. (b) The advanced metabolic network of indolyl glucosinolate. Five potential novel regulatory relationships are shown with red lines, and the numbers beside the lines show partial correlation coefficients. The information of the known metabolic networks were obtained from Windsor et al. (2005) and Padilla et al. (2007) for aliphatic glucosinolate, and from Iqbal et al. (1995) and Grubb & Abel (2006) for indolyl glucosinolate.

Results showed that the amounts of aliphatic and indolyl glucosinolates were correlated positively with each other, with a total of 106 individual correlations detected (Fig. 5). On the basis of correlations between different glucosinolates, the relationships between the mcQTL and the metabolic pathways were assumed to reveal how genetic variation could affect a given biochemical pathway (Fig. 4). For example, q.mcG-A9b was found to affect the total glucosinolate and seven other traits, including both aliphatic and indolyl glucosinolates in seeds (Table 3). This suggests that it might play a key role in the core glucosinolate biosynthesis pathway (pathway B) or the modification pathway (pathway D). Ultimately, the functions of the genes that underlay each mcQTL were assigned to their corresponding pathway in the glucosinolate metabolic network (Fig. 4).

Epistatic network for glucosinolate concentration in leaves and seeds

An average of five significant epistatic interactions per trait was found, although the number varied from one to 10 for each trait in a single organ per field trial. A very high proportion (97%) of these epistatic interactions was only identified in a single trait and one trial. No epistatic interaction was found to be common between leaves and seeds. This means that a large number of epistatic interactions could be identified within this population. A total of 471 significant (P < 0.001) epistatic interactions were detected for the 16 traits and in both organs from the four field trials; 177 (38%) and 294 (62%) epistatic interactions were detected in leaves and seeds, respectively (Table S4). The general phenotypic contribution for total glucosinolate in the seeds attributable to epistatic interactions and QTL were 17.8% and 57.2%, respectively. However, in leaves epistatic interactions and QTL accounted for 55.5% and 8.2% of the variation, respectively. Epistatic interactions obviously play a significant role in regulating the glucosinolate concentration in leaves but only a minor role in the seeds.

Networks were constructed based on the epistatic interactions identified in both leaves (Fig. 7, Table S4) and seeds (Fig. S2, Table S4). Approx. 40% of the epistatic interactions detected in leaves or seeds corresponded to one or more QTL at one interacting locus or QTL/QTL interactions (Table 2), which indicated that genes underlying mQTL controlled the concentration of glucosinolate not only directly, but also possibly through interactions with other locus. For example, two epistatic interactions, EI(A9-72.9/C2-53.5) and EI(A9-72.9/C7-64.6), involved the same mcQTL (q.mcG-A9g, confidence interval: 72.1–73.3 cM) and the same locus (A9-72.9) (Fig. S2). This implied that the other two interacting loci (C2-53.5 and C7-64.6) that interacted with q.mcG-A9g might also play a role in regulating the biosynthesis of aliphatic glucosinolate in seeds.

Figure 7.

Demonstration of a complex epistatic network for glucosinolate in leaves. The 19 sectors of the disc indicated the 19 linkage groups of TN doubled-haploid (DH) linkage map. The boxes of different colors perpendicular to the linkage group bars at the outer edge represent the different pseudochromosome fragments of Arabidopsis that are aligned with chromosome fragments of Brassica napus, and the green, black and red lines perpendicular to the linkage group bars at the inner edge indicate the confidence intervals of mQTL for glucosinolate in leaves, seeds and both, respectively. The long, thick, green lines indicate epistatic interactions for glucosinolate in leaves. The position of two interacting loci involved in EI(A2-12.4/C1-39.4) are marked with a pink triangle, in which one interacting loci (A2-12.4) involved MYB29 and MYB76, and another interacting loci (C1-39.4) involved AOP2, BCAT3, CYP79F1 and CYP79F2.

Using in silico mapping, 242 orthologous genes that corresponded to 53 genes with recognized functions for glucosinolate metabolism in A. thaliana were mapped and found to correspond to 45% of the interacting loci, which were involved in 344 epistatic interactions. This implies that a large number of unknown genes are also involved in epistatic interactions. The relationships between epistatic interactions and known genes underlying the known metabolic pathway supports the principle that genes underlying the epistatic interactions are also involved in the biochemical pathways (Table S4). For example, one epistatic interaction, EI(A2-12.4/C1-39.4) involved two interacting loci, one of which corresponded to the region containing MYB29 and MYB76 (A2-12.4); AOP2, BCAT3, CYP79F1 and CYP79F2 also mapped within the confidence intervals of the other interacting loci (C1-39.4) (Fig. 7). These results show that this epistatic interaction is likely to be involved in the core glucosinolate biosynthesis pathway (pathway B) or glucosinolate side-chain modification (pathway D). The function of the 344 epistatic interactions that involved orthologous genes might also be expected to have comparative functions in the glucosinolate regulatory network of B. napus (Table S4).


Comparative analysis of QTL in Brassica

In this study, 21 mQTL were identified for total glucosinolate in seeds, in which nine mQTL (q.mcG-A9a, q.mcG-A9b, q.mcG-A9c, q.mcG-C2b, q.mcG-C2c, q.mcG-C2d, q.mcG-C7e, q.mcG-C9a and q.mcG-C9b) correspond to QTL identified in previous studies of B. napus (AACC, 2n = 38) (Toroser et al., 1995; Uzunova et al., 1995; Howell et al., 2003; Quijada et al., 2006). Twenty-seven mQTL for aliphatic and indolyl glucosinolates in leaves corresponded to the A genome, in which four mQTL (q.mcG-A3a, q.mcG-A5a, q.msG-A5a and q.mcG-A7d) for aliphatic and indolyl glucosinolate accumulation in the leaves of B. rapa have been reported (AA, 2n = 20) (Lou et al., 2008). A further 25 mQTL for aliphatic glucosinolates in seeds were identified on A genome linkage groups; only four mQTL (q.mcG-A2b, q.mcG-A9a, q.mcG-A9b and q.mcG-A9c) are thought to have been identified previously in B. juncea (AABB, 2n = 36) (Ramchiary et al., 2007; Bisht et al., 2009).

The QTL region identical to q.mcG-A9a, q.mcG-A9b and q.mcG-A9c was mapped previously as GLN1 in B. napus (Howell et al., 2003) and J9Gsl3 in a DH population of B. juncea (Ramchiary et al., 2007; Bisht et al., 2009). Another QTL (A9: Glu_S7/S8/K8) for total seed glucosinolate has also been mapped to this region in the CK (Chiifu × Kenshin) population of B. rapa (data not shown). However, no genes known to be involved in glucosinolate biosynthesis or transport have so far been identified within this region. Therefore, fine mapping of these mQTL in conjunction with genome sequence information of B. rapa Chiifu ( should make it possible to identify these genes and, in a similar approach, genes underlying other QTL, such as q.mcG-A3a, which corresponds to Ali-QTL3.2, identified from a DH38 population of B. rapa by Lou et al. (2008). The remaining 91 mQTL are thought to be novel, which could be attributable to the high density of the TN DH linkage map, increased polymorphism between the parents, different tissues and different field trials and it is interesting to consider which genes may underlie these.

Presumed function of the genes underlying mQTL and epistatic loci

Only 39% of the mQTL and 45% of the interacting loci were coincidental with orthologs of genes involved in glucosinolate metabolism in Arabidopsis. This suggests that either the genes that underlie most of the mQTL and the interacting loci are of unknown function or many of the paralogous loci in B. napus have been shuffled within the genome (Fig. 8). Further investigation of the relationships of the QTL and epistatic interactions with their role in glucosinolate metabolic pathways would help to reveal the functions of the genes that underlie the QTL and the epistatic loci (Barker et al., 2007; Zhao et al., 2008). During the analysis of the QTL, we found that almost half of the mcQTL were involved in the core pathway (pathway B), > 30% of the mcQTL were involved in the glucosinolate side-chain modification (pathway D) and < 20% of the mcQTL were related to the side-chain elongation pathway (pathway A). Similarly, we found that > 55% of the 344 epistatic interactions were involved in the core pathway (pathway B) of glucosinolate biosynthesis, and < 45% of the 344 epistatic interactions were related to other pathways of glucosinolate metabolism (pathway A, D and E). Direct comparisons between the QTL and epistatic interactions showed that most of genetic variation detectable via QTL was caused by variation in the core pathway (pathway B) and glucosinolate side-chain modification (pathway D); however, for the epistatic interactions, most of genetic variation was attributable to variation in the core pathway (pathway B). The problem with any analysis of QTL or epistatic interactions is that it is dependent on the variation between the parents of the cross used. The TN population has been useful in identifying some novel loci; however, a different population would be better suited for identifying QTL and the pathways involved in the synthesis of indolic glucosinolates.

Figure 8.

Comparison of homologous regions between Arabidopsis and Brassica napus by in silico mapping. The 19 sectors of the disc indicate the 19 linkage groups of the TN doubled-haploid (DH) linkage map. The boxes of different colors perpendicular to the linkage group bars at the outer edge represent the different pseudochromosome fragments of Arabidopsis that are aligned with chromosome fragments of B. napus, and the green, black and red lines perpendicular to the linkage group bars at the inner edge indicate the confidence intervals of mQTL for glucosinolate concentration in leaves, seeds and both, respectively. The long, thick lines of different colors represent homologous regions that are aligned with the different pseudochromosome fragments of Arabidopsis.

Most of the mQTL and epistatic loci (> 60%) could be assigned to the A genome linkage groups, but the corresponding loci within the C genome were not identified (Table 2). One explanation is that low glucosinolate loci have been inherited from the A genome (Sharpe & Lydiate, 2003). Another possible explanation is that, during breeding, a greater amount of polymorphism has been introgressed into the A genome of Ningyou7. Whatever the reason, it appears that there is more genetic variation in the A genome than in the C genome of the two parents of TN DH population.

Advanced metabolic network for glucosinolate

Using current knowledge about glucosinolate metabolism (Grubb & Abel, 2006; Halkier & Gershenzon, 2006) and the MetaNetwork software, we have constructed an advanced metabolic network for aliphatic and indolyl glucosinolates in leaves and seeds of B. napus (Fig. 6). In this network, five potential novel regulatory relationships were uncovered. Four potential novel regulatory relationships between aliphatic or 3-indolyl-methyl glucosinolate in both leaves and seeds were detected. This suggests some genetic loci (such as q.mcG-A4c) can control the production of these compounds in both leaves and seeds. These loci have not been identified in previous Arabidopsis studies (Kliebenstein et al., 2001a,c). Ali-5C-GNL was found to have a close correlation with Ali-4C-PRO in leaves and seeds, which implied that GS-OH or a novel enzyme is involved in the synthesis of both compounds from their respective alkenyl precursors in either or both organs. Another possible explanation for some of the relationships detected in the regulatory network analysis is close linkage of two or more genes with different function.

No negative relationship was detected between the aliphatic and indolyl glucosinolates. This suggests that undetected genetic loci with opposite additive effects that control aliphatic and indolyl glucosinolates have yet to be identified. Moreover, there are difficulties in distinguishing multi-copy genes, complex epigenetic regulation, and differential expression of the same gene in different growing periods, environments, and tissues (Keurentjes et al., 2008). More significantly, it should be noticed that strong positive correlations have been demonstrated among glucosinolate, storage protein and phenolic acid concentrations in seeds of both Arabidopsis and B. napus (Hemm et al., 2003; Wittkop et al., 2009). This suggests that potential competition at the level of common amino acid precursors may warrant future investigations into genes involved in amino acid metabolism, partitioning and transport. Therefore, additional field trials would be required to verify such potential regulatory relationships and new mapping populations with different origins of alleles for low glucosinolate concentration would help to refine and enrich the metabolic network further.


This work was supported by the National Basic Research and Development Program (973 Program) (no. 2006CB101600) and National 863 High Technology Program, P. R. China (No. 2006AA10A113). The authors thank Dr Guusje Bonnema, Dr Genyi Li, and Dr Jitao Zou for their critical reading of the manuscript. The authors also thank Ms Xiuli Sun and Ning Zhang for technical help.