A complex interplay of three R2R3 MYB transcription factors determines the profile of aliphatic glucosinolates in Arabidopsis.

While R2R3 MYB transcription factors are a large gene family of transcription factors within plants, comprehensive functional data in planta are still scarce. A model for studying R2R3 MYB control of metabolic networks is the glucosinolates (GLSs), secondary metabolites that control plant resistance against insects and pathogens and carry cancer-preventive properties. Three related members of the R2R3 MYB transcription factor family within Arabidopsis (Arabidopsis thaliana), MYB28, MYB29, and MYB76, are the commonly defined regulators of aliphatic GLS biosynthesis. We utilized new genotypes and systems analysis techniques to test the existing regulatory model in which MYB28 is the dominant regulator, MYB29 plays a minor rheostat role, and MYB76 is largely uninvolved. We unequivocally show that MYB76 is not dependent on MYB28 and MYB29 for induction of aliphatic GLSs and that MYB76 plays a role in determining the spatial distribution of aliphatic GLSs within the leaf, pointing at a potential role of MYB76 in transport regulation. Transcriptional profiling of knockout mutants revealed that GLS metabolite levels are uncoupled from the level of transcript accumulation for aliphatic GLS biosynthetic genes. This uncoupling of chemotypes from biosynthetic transcripts suggests revising our view of the regulation of GLS metabolism from a simple linear transcription factor-promoter model to a more modular system in which transcription factors cause similar chemotypes via nonoverlapping regulatory patterns. Similar regulatory networks might exist in other secondary pathways.

The number and structural complexity of secondary metabolites evolved in plants are simply mesmerizing (Dixon and Strack, 2003). Secondary metabolites allow the plant to adapt to the ever-changing environment. Ideally, these costly metabolites should be highly adaptable and precisely distributed within the correct tissue at the appropriate time to balance resources utilized in their synthesis with maximal biological impact. Yet, identification and characterization of transcription factors directly controlling secondary metabolite accumulation is still in its infancy (Grotewold, 2008). The direct regulatory network of the glucosinolates (GLSs) encompasses six R2R3 MYB transcription factors from a single gene family within Arabidopsis (Arabidopsis thaliana). ATR1/MYB34, MYB51, and MYB122 are thought to regulate the Trp-derived (indole) GLS pathway (Celenza et al., 2005;Gigolashvili et al., 2007a;Malitsky et al., 2008), and MYB28, MYB29, and MYB76 regulate the Met-derived (aliphatic) GLSs in Arabidopsis accession Columbia (Col-0; Gigolashvili et al., 2007bGigolashvili et al., , 2008Hirai et al., 2007;Sonderby et al., 2007;Beekwilder et al., 2008;Malitsky et al., 2008). So far, the distinct roles of the R2R3 MYB genes controlling GLS biosynthesis remain largely uncharacterized.
The amino acid-derived GLSs are specific to the order Brassicales and shape plant-pest interactions (Halkier and Gershenzon, 2006;Bednarek et al., 2009). Several advantages make the regulation of aliphatic GLSs an excellent model for the elucidation of regulatory networks (Hirai et al., 2007;Wentzell et al., 2007). First, most genes in the biosynthetic pathway are known (Grubb and Abel, 2006;Halkier and Gershenzon, 2006). The pathway takes place in three stages: a side chain elongation of aliphatic amino acids by incorporation of one to six methylene groups, formation of the core GLS moiety, and finally, secondary modifications of the side chain to generate the plethora of GLS compounds. Second, GLSs can rapidly and reliably be measured, allowing for direct assessment of the connection between homeostasis and induction within numerous conditions and genotypes (Kliebenstein et al., 2001a). Finally, GLSs are present in the model plant Arabidopsis, with all its available genomics tools, and also in important crop plants, where consumption of Brassica vegetables has been correlated with a decrease in the occurrence of cancer (Juge et al., 2007;Traka and Mithen, 2009). Therefore, a systematic understanding of GLS regulation can positively affect agriculture and human nutrition as well as our understanding of regulatory networks.
In recent years, MYB28, MYB29, and MYB76 have been characterized as direct transcriptional regulators of aliphatic GLS biosynthetic genes after identification using an omics-based methodology (Hirai et al., 2007), a focused transactivation approach (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008, and a quantitative systems biology approach (Sonderby et al., 2007). The analysis of plants overexpressing MYBs in wild-type Col-0 and the transactivation potential of the genes toward GLS biosynthetic genes indicate that the three transcription factors carry largely the same functions (Fig. 1A). However, metabolite analysis of single knockouts revealed that the similar potential shown in the overexpression lines does not fully match the in planta activity. T-DNA mutants in MYB29 and MYB76 are decreased in short-chained aliphatic GLSs (one to three cycles of chain elongation; Sonderby et al., 2007;Beekwilder et al., 2008;Gigolashvili et al., 2008), as are T-DNA mutants in MYB28 that, in addition, are almost devoid of long-chained aliphatic GLSs (four to six cycles of chain elongation; Hirai et al., 2007;Sonderby et al., 2007;Beekwilder et al., 2008). Based on these data, an in planta model might regard all three proteins to act as direct biosynthetic transcriptional activators, with MYB28 as the major regulator of aliphatic GLSs followed by MYB29 and MYB76 having minor, accessory roles ( Fig. 1B; Gigolashvili et al., 2009a). However, a double knockout in MYB28 and MYB29 is almost devoid of aliphatic GLSs, which suggests an epistatic effect that requires an interacting regulatory mechanism between the two genes (Sonderby et al., 2007;Beekwilder et al., 2008). Furthermore, changes can occur in the relative amounts of individual GLSs, most obviously shown by the change of GLS profiles throughout the development of the plant (Brown et al., 2003). In addition, feeding of aphids only increased short-chained aliphatic GLSs, whereas feeding of the lepidopteran Spodoptera exigua increased both short and long-chained aliphatic GLSs (Mewis et al., 2006). Thus, there must be mechanisms by which relative amounts of individual GLSs can be adjusted beyond the link of MYB28 to long-chained aliphatic GLSs.
The general model of flavonoid biosynthesis regulation is that the level of biosynthetic transcripts mirrors the level of accumulating metabolites (Quattrochio et al., 2006). By contrast, recent data indicate that a similar linear correlation is not always present for GLSs. For instance, IQ-DOMAIN1 (IQD1), a calmodulin-binding transcription factor, modifies GLS accumulation and is generally thought to integrate signals when the plant is under biotic stress. Curiously, however, while aliphatic GLS levels are increased in the iqd1 mutant, a concurrent decrease in the transcript levels of the aliphatic biosynthetic genes CYP79F1, CYP79F2, and UGT74B1 is observed (Levy et al., 2005). This contrasts with the coordinated regulation of GLSs metabolites and transcripts by SUL-FUR LIMITATION1 in response to sulfur deficiency (Hirai et al., 2005;Maruyama-Nakashita et al., 2006;Falk et al., 2007). Moreover, the distribution of GLSs within the leaf (Shroff et al., 2008) cannot be explained by the expression pattern of MYB28, MYB29, and MYB76 as shown by promoter-GUS analyses (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008Malitsky et al., 2008). This is unlike how the spatial expression pattern of three R2R3 MYB transcription factors specifically directs the accumulation of flavonols (Stracke et al., 2007). Assuming that the MYBs directly control transcript levels of biosynthetic genes in planta, which, in turn, directly reflects the secondary metabolite accumulation, this does not seem applicable to GLS biosynthesis.  Hirai et al. (2007), Gigolashvili et al. (2007bGigolashvili et al. ( , 2008, Sonderby et al. (2007), Beekwilder et al. (2008, and Malitsky et al. (2008). A, Model for potential activities of MYB28, MYB29, and MYB76 based on data from overexpression lines and transactivation assays. B, Model for actual regulation in planta. Circles represent short-or long-chained aliphatic GLSs. Lines leading to circles indicate increasing (single arrowheads) effect on metabolites. Lines to genes represent induction of the genes in moderate overexpression lines or in planta regulation (red single arrowheads), strong overexpression lines (red double arrowheads), various treatments in wild-type Col-0 (pink single arrowheads), or transactivation (purple triple arrowheads). In the case of transactivation, arrows pointing toward circles represent activation of biosynthetic genes directing the production of short-chained (MAM1 and CYP79F1) or long-chained (MAM3 and CYP79F2) aliphatic GLSs. The pink T-line represents repression by salicylic acid (SA). [See online article for color version of this figure.] The indications for a complex regulatory hierarchy within aliphatic GLS regulation prompted us to investigate the individual roles of MYB28, MYB29, and MYB76 in planta and their potential interactions to unravel their roles in shaping the GLS profile of the plant. We show that MYB29 does not affect transcript levels for any of the biosynthetic genes, while MYB28 constitutes the major transcriptional inducer. The combination of transcriptional analysis with GLS profiles of the knockouts revealed an intriguing uncoupling of the level of biosynthetic transcripts from the level of aliphatic GLSs, since similar levels of biosynthetic transcript resulted in surprisingly different chemotypes. We unequivocally demonstrate that MYB76 has a MYB28-and MYB29-independent role in Arabidopsis (accession Col-0) and that MYB76 has transcriptional effects on specific parts of the biosynthetic pathway. Finally, altered aliphatic GLS distribution in the leaves in different myb knockouts and overexpressors indicates that the spatial expression pattern of MYB76 in particular determines the pattern of aliphatic GLS accumulation. These data prompted us to suggest an improved model for the regulation of aliphatic GLSs.

RESULTS
MYB76 Is Independent of MYB28 or MYB29 for Induction of Aliphatic GLS Biosynthesis Due to no significant impact (Gigolashvili et al., 2008) or only a small change (Sonderby et al., 2007) on GLS levels in myb76 knockdowns, MYB76 has been concluded to carry out a minor, possibly accessory role in aliphatic GLS biosynthesis within the Col-0 accession (Gigolashvili et al., 2009a). Combined with the minute aliphatic GLSs measured in the myb28 myb29 double knockout (Sonderby et al., 2007;Beekwilder et al., 2008), this led us to hypothesize that MYB76 might require MYB28 and MYB29 to induce GLS biosynthesis in planta. To test this hypothesis, we overexpressed MYB76 in the myb28-1 myb29-1 double knockout background by way of the cauliflower mosaic virus 35S promoter, allowing a direct test of the capability of MYB76 to induce aliphatic GLSs in the absence of both MYB28 and MYB29.
Foliar aliphatic GLSs were detected in independent T1 transformants. Contrary to the sole impact on shortchained aliphatic GLS in the myb76 knockdown mutants, both short-chained and long-chained aliphatic GLSs were induced up to 50% and 60% of wild-type Col-0, respectively. Both methylthioalkyl-and methylsulfinylalkyl-GLSs were produced (for GLS abbreviations, see Supplemental Table S1), thereby indicating that MYB76 can induce both core biosynthesis and the secondary modification pathway encompassing FMO GSOX s, which convert methylthioalkyl-to methylsulfinylalkyl-GLSs Li et al., 2008;Supplemental Tables S2 and S3).
Analysis of T2 leaves indicated transgene silencing, since the plants only contained 5% short-chained aliphatic GLSs in comparison with the wild type and undetectable levels of long-chained aliphatic GLSs (data not shown). Similarly, we have observed the loss of chemotype in subsequent generations when overexpressing MYB76 in the wild-type Col-0 background (data not shown).
Previous overexpression analysis showed that MYB76 had the potential to regulate aliphatic seed GLS levels in Arabidopsis Col-0 plants (Sonderby et al., 2007). These experiments, however, did not address if MYB76 requires a functional MYB28 and/or MYB29 to induce GLSs in seeds. In T2 seeds of the myb28-1 myb29-1 knockout plants overexpressing MYB76, the different lines had short-chained aliphatic levels of up to 140% and long-chained aliphatic GLSs of 10% to 30% of the levels in wild-type Col-0 ( Fig. 2; Supplemental Tables S4 and S5). While short-chained methylthioalkyl-and methylsulfinylalkyl-GLSs accumulated to levels above the wild type, the shortchained seed-specific GLSs, 3bzop and 4bzob, did not exceed wild-type levels.
In summary, these data show that MYB76 can function independently of MYB28 and MYB29 in controlling both short-chained and long-chained aliphatic GLSs in leaves as well as in seeds. Furthermore, MYB76 can induce long-chained aliphatic GLSs, an ability previously solely ascribed to the regulatory realm of MYB28. Since myb76 and myb29 single knockout mutants both show decreases only in short-chained aliphatic GLSs (Sonderby et al., 2007;Beekwilder et al., 2008;Gigolashvili et al., 2008) and MYB29 and MYB76 are tandemly duplicated genes, the two genes might functionally overlap in planta. To explore the role of MYB76 and a possible genetic interaction of MYB76 with MYB28 (as shown for MYB28 and MYB29; Sonderby et al., 2007;Beekwilder et al., 2008), we investigated the GLS profile of homozygous myb28-1 myb76 knockouts as well as wild-type Col-0, myb76, and myb28-1 obtained from crossing myb28-1 with either myb76-1 or myb76-2 (Fig. 3). An ANOVA of GLS levels between myb76-1 and myb76-2 as well as between myb28-1 myb76-1 and myb28-1 myb76-2 showed no significant difference between the two myb76 lines, allowing us to combine the lines to test the effect of the absence of MYB76 rather than the effect of one specific insertion allele versus the wild type (Supplemental Table S6).
The myb28-1 myb76 double knockouts retained 33% of the short-chained foliar aliphatic GLSs in comparison with wild-type Col-0. Thereby, the myb28-1 myb76 foliar aliphatic GLS phenotype is an additive combination of the two single mutants ( Fig. 3; Supplemental Table S7), which is in contrast to the emergent quality observed in the almost aliphatic GLS-free myb28 myb29 double knockout mutant (Sonderby et al., 2007;Beekwilder et al., 2008). The decrease in the myb28-1 myb76 knockout in comparison with the single mutants was observed for all short-chained aliphatic GLSs (i.e. 3msp, 4mtb, 4msb, and 5msp; Supplemental Table S1). The only nonadditive interaction effect observed in the aliphatic GLS data was for 4mtb (Supplemental Table S6), suggesting an epistatic interaction of MYB28 and MYB76 upon the FMOs that convert methylthioalkyl-to methylsulfinylalkyl-GLSs. In addition, an epistatic effect for total indole GLS levels was observed in the data set (Supplemental Table S6). The analysis of homozygous F2 progeny was identical to the F3 generation presented here (data not shown).
Seeds are one of the predominant sites for the accumulation for aliphatic GLSs with a unique GLS profile (Brown et al., 2003). No significant difference was found in seed GLS between the myb28-1 and myb28-1 myb76 knockout mutants (Table I; Supplemental Table S8). Since the myb76 single knockdowns do not have a seed GLS chemotype (Table I; Supplemental Table S8), these data again support the absence of an epistatic effect between MYB28 and MYB76 for the vast majority of GLSs. In conclusion, the myb28-1 myb76 knockout mutant data provide additional support that MYB76 has regulatory capabilities on its own. Figure 3. Comparison of foliar GLS levels in wildtype Col-0, myb76, myb28-1, and myb28-1 myb76 knockouts. GLSs were extracted from 30 to 70 mg of leaf material derived from 21-to 24-dold plants and quantified as desulfoglucosinolates by HPLC. n = 11 for wild-type Col-0, n = 32 for myb76, n = 25 for myb28-1, and n = 56 for myb28-1 myb76. The data are sums of two independent experiments and were analyzed via ANOVA (Supplemental Table S6). Individual indole GLSs can be found in Supplemental Table  S1. Different letters indicate significant differences at P , 0.05. FW, Fresh weight; ND, not detected. Figure 2. HPLC analysis of GLSs in seeds of wildtype Col-0 (trace a), T2 seeds of a plant overexpressing MYB76 in the myb28-1 myb29-1 double knockout background (trace b), and seeds of myb28-1 myb29-1 double knockout (trace c). GLSs were extracted from 10 seeds using 6 nmol of sinigrin as an internal standard and analyzed as desulfoglucosinolates by HPLC (for details, see "Materials and Methods"). Peaks representing individual compounds are denoted using the abbreviations in Supplemental Table S1. Longchained aliphatic GLSs are marked by asterisks and indole GLSs by squares. For quantitative results and more lines, see Supplemental Table  S4. mAU, Milli absorption units. MYB28 knockout mutants are strongly reduced in or lack detectable long-chained aliphatic GLSs, indicating MYB28 as the sole regulatory factor for these GLSs. However, the above experiments showed that MYB76 could induce long-chained aliphatic GLSs in the absence of both MYB28 and MYB29 (Fig. 2). Possibly, MYB29 could likewise induce long-chained aliphatic GLSs in the absence of MYB28 and MYB76 in spite of its pure short-chained aliphatic GLS knockout mutant chemotype. To test this, we overexpressed MYB29 in the myb28-1 myb76 double knockout background. Ten independent T1 transformants were generated, all of which showed the production of long-chained aliphatic GLSs, as was likewise observed in both the T2 seeds and their progeny ( Fig. 4; Supplemental Tables S2-S4; data not shown). Thus, MYB29 can control the biosynthesis of these compounds independently of MYB28 and MYB76. Interestingly, the short-chained aliphatic GLSs were elevated in comparison with both myb28-1 myb76 and wild-type Col-0 in both leaves (Supplemental Tables S2 and S3) and seeds (Supplemental Tables S4 and S5). In seeds, the increase in short-chained GLSs was far more pronounced than the increase in long-chained GLSs. This suggests that while MYB29 can induce long-chained aliphatic GLSs, MYB28 may still be necessary for optimal long-chained aliphatic GLS accumulation in seeds.
Genome-Wide Transcriptional Analysis of T-DNA Knockouts Points to Divergent Roles of MYB28, MYB29, and MYB76 in Planta Individual overexpression of MYB28, MYB29, and MYB76 revealed both altered GLS accumulation (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008Hirai et al., 2007;Sonderby et al., 2007;Malitsky et al., 2008) and a significant overlap in GLS transcripts induced by the three MYB genes (Sonderby et al., 2007). Thus, the regulatory model based on overexpression lines and transactivation data (Fig. 1A) suggests a transcriptional analysis of the single knockouts to identify similar regulatory modules for the three MYBs. However, the different additive and epistatic genetic effects observed in the GLS profiles of T-DNA double knockouts suggest a much more complex pattern in planta, with some overlap in regulatory modules but also extensive specificity. To test the transcriptional regulatory capabilities of MYB28, MYB29, and MYB76, we performed Affymetrix ATH1 GeneChip microarrays on the following knockouts: myb28-1, myb29-1, myb29-2, myb76-1, myb76-2, and myb28-1 myb29-1. GLS analysis confirmed the expected chemotypes of all plants used for RNA extraction (data not shown).
As multiple alleles of both MYB29 and MYB76 were included in the analysis, we utilized independent ANOVAs to test whether the difference in each transcript was significantly different between each mutant allele compared with the wild type (Supplemental Table S10). ANOVA of transcript levels suggested that myb76-1 and myb76-2 (Supplemental Table S10) were identical, allowing us to combine the means of the two lines and test the effect of lack of MYB76 rather than the effect of any single insertion allele versus the wild type. In contrast, myb29-1 and myb29-2 occasionally showed opposite effects on genes, and the ANOVA showed that the effect on gene expression of the two alleles was significantly different. As such, although no difference was found in the effect on the GLS biosynthetic genes between the two knockouts, we did not combine the data and focused on the transcript data from the myb29-2 mutant, as the myb29-1 allele did not show a significantly lower transcript of MYB29 Table I. Seed GLSs in wild-type Col-0, myb28-1, myb76, and myb28-1 myb76 T-DNA knockouts Mean shows the average GLS content in pmol per 10 seeds using the abbreviations in Supplemental Table S1. SE is the standard error for the line. These data represent two independent biological replicates. n is the number of replicates per genotype. The data for the two myb76 alleles and the two myb28-1 myb76 alleles were pooled, as there was no significant difference in the GLS phenotype between the different alleles (Supplemental Table S8). Different letters (a, b, c) represent significant differences at P , 0.05 between wild-type Col-0 and the homozygous mutant lines as determined by ANOVA (Supplemental Table S8).

GLS
Wild-Type Col-0 (n = 8) myb76 (n = 15) myb28-1 (n = 8) in comparison with wild-type Col-0 (Table II; Supplemental Table S9). Using a false discovery rate (FDR) of 0.05, myb76 had the smallest number of changed transcript levels in comparison with Col-0 with 668 altered transcripts (394 down, 262 up), followed by myb28-1 with 1,066 altered transcripts (584 down, 449 up), myb28-1 myb29-1 with 1,412 altered transcripts (666 down, 726 up), and finally myb29-2 with 2,914 altered transcripts (1,115 down, 1,799 up; Supplemental Table S10). This is a rather large number of altered transcripts, but the experimental design likely allowed us to identify both direct and indirect targets. Given the centrality of GLSs to Arabidopsis defense, such a number of direct and indirect links might be expected. Additionally, GLSs have recently been shown to influence stomatal closure via the abscisic acid pathway (Zhao et al., 2008), which indicates that we are just beginning to understand their linkages to the rest of the genome. A Venn diagram of the three single gene knockouts, myb28-1, myb29-2, and myb76, revealed that the overlap between the three mutants was indistinguishable from random ( Fig. 5A; x 2 , P = 0.687). This shows that the three genes have distinct transcriptional roles in planta. Of the five genes showing expression overlap among all three single myb knockouts, only one had any apparent link to the GLS pathway. SULTR2;1 (At5g10180), a low-affinity sulfate transporter, was significantly decreased in all genotypes. The other four genes were GASA5 (At3g02885), FAS1 (At1g65470), a glycosylphosphatidylinositol transamidase component family protein (At5g19130), and an F-box protein (At5g03970) with no known function linked to GLS biosynthesis.

MYB28 Is the Key Transcriptional Activator of Aliphatic Biosynthetic Transcripts
Having established that the different myb single knockouts conferred different transcriptional effects at the overall level, we more deeply queried the distinct roles of MYB28, MYB29, and MYB76.
MYB28 transcript was nearly absent in the myb28-1 knockout (Table II; Fig. 6). Analysis of myb28-1 versus the wild type showed that 13 out of 28 transcripts of known or putative aliphatic GLS biosynthetic genes were significantly lowered ( Fig. 6; Table II). Interestingly, the largest fold changes were centered on the chain-elongation machinery, BCAT4, MAM1, MAM3, the IPMIs, and BAT5, a transporter putatively involved in the chain elongation pathway (Gigolashvili et al., 2009b;Sawada et al., 2009). Additionally, large impacts on late genes in the pathway were present, as shown by down-regulation of FMO GSOX1 and FMO GSOX3 . However, the general effect on the rest of the core biosynthesis and secondary modification enzymes was more modest, thereby still upholding some basic transcriptional level of the pathway (Table  II). Interestingly, MYB29 was decreased 43% in the myb28-1 mutant. While this was not statistically significant in this experiment, it is in accordance with a previously observed significant down-regulation of MYB29 in a different myb28 knockout mutant, as measured by quantitative reverse transcription (RT)-PCR (Beekwilder et al., 2008).
These data point to a key importance of MYB28 for transcriptional regulation of the side chain elongation component of the aliphatic GLS biosynthetic enzymes. . HPLC analysis of GLSs in seeds of wildtype Col-0 (trace a), T2 seeds of a plant overexpressing MYB29 in the myb28-1 myb76-2 double knockout background (trace b), and seeds of myb28-1 myb76-2 double knockout (trace c). GLSs were extracted from 10 seeds using 6 nmol of sinigrin as an internal standard and analyzed as desulfoglucosinolates by HPLC (for details, see "Materials and Methods"). Compounds are denoted using the abbreviations in Supplemental  Table S1. Long-chained aliphatic GLSs are marked by asterisks and indole GLSs by squares. For quantitative results and more lines, see Supplemental Table S4. mAU, Milli absorption units.
Additionally, it shows an intriguing capacity within GLS regulation: even very low expression levels of parts of the GLS biosynthetic apparatus can maintain two-thirds of the wild-type levels of short-chained aliphatic GLS in the myb28-1 mutant (Fig. 3), thereby hinting at an uncoupling of biosynthetic transcripts from GLS chemotypes.
myb29 Is Not Affected in the Transcription of Aliphatic Biosynthetic Genes As MYB29 has been described as the second most important aliphatic GLS regulator, we next investigated the transcript differences between wild-type Col-0 and myb29-2. Transcript of MYB29 was only 35% decreased in the myb29-2 single knockout (Table  II; Fig. 6). However, since previous RT-PCR in the important R2R3 MYB binding domain did not detect transcript in the myb29-2 mutant (Sonderby et al., 2007) and the ATH1 probes are situated farther downstream in the gene, we regard the discrepancy to be caused by aberrant downstream transcript and thus myb29-2 as a true functional knockout. This is supported by its aliphatic GLS chemotype. To our surprise, however, a direct comparison of transcript levels between the myb29-2 knockout and the wild type did not show any decrease in transcript abundance for genes within the aliphatic biosynthetic pathway (Figs. 5C and 6). The myb29-1 mutant also did not alter the accumulation of aliphatic GLS gene transcripts (Table II). On the contrary, there was a tendency for the biosynthetic transcript levels to be elevated, most prominently shown in myb29-2 by a significant increase of 50% in the transcript of SOT18, one of the sulfotransferases in core GLS biosynthesis (Table II; Fig. 6).
At first glance, these data seem to contradict previous understanding of MYB29 as a positive transcriptional activator of aliphatic biosynthetic transcripts. A potential explanation for this discrepancy may come from a significant increase in MYB28 transcript in the myb29-2 knockout, which may lead to the wild-type transcript levels of the other GLS genes (Table II;   The almost aliphatic GLS-free chemotype of the myb28-1 myb29-1 mutant suggests a corresponding abolition of aliphatic biosynthetic transcripts in the mutant. Indeed, the absence of MYB28 and MYB29 in the double knockout did lead to a broad decrease in transcript abundance for all the aliphatic GLS biosynthetic phases: chain elongation, core biosynthesis, and secondary modification. Sixteen of the 28 transcripts of known or putative aliphatic GLS biosynthetic genes were significantly decreased, as were the transcripts of MYB28 and MYB29 themselves ( Fig. 6; Table II). Intriguingly, however, even though several genes became significantly down-regulated in the myb28-1 myb29-1 double knockout in comparison with the myb28-1 knockout, the observed proportional decrease in biosynthetic transcripts was no lower than that observed in the myb28-1 knockout ( Fig. 6; Table II). Thus, the epistatic effect of MYB28 and MYB29 shown at the GLS level is not mirrored as an epistatic effect on the total level of biosynthetic transcripts but only on individual transcripts.
GLS accumulation within a certain tissue is determined by the amount of biosynthesis, catabolism, and transport. Thus, changes in any of the processes might cause the uncoupling. However, no significant differences could be detected in the transcription of genes in the degradation pathways (i.e. the b-glucosidases TGG1, TGG2, PEN2, and PYK10 as well as the specifier proteins ESP, NSPs, and ESM1) involved or putatively involved in GLS breakdown (Supplemental Table  S11). However, myrosinase-binding protein 1 (MBP1; At1g52040), which has been suggested to serve a function in endogenous turnover of GLSs in Brassica napus (Andreasson et al., 2001), was one of the most up-regulated genes in the myb76 knockdowns (59%) and nonsignificantly up-regulated 63% and 52% in the myb29-2 and myb28-1 myb29-1 knockouts, respectively. It still remains unknown what the role of MBP1 is in plants. However, if the MYBs affect breakdown, then MBP1 or other uncharacterized proteins could account for it.
Previous data have shown large transcriptional effects on sulfur metabolism when overexpressing the three MYB genes in the wild-type Col-0 background (Sonderby et al., 2007;Malitsky et al., 2008). Therefore, we queried whether transcripts of the following sulfur utilization pathways were changed in the knockouts: sulfate transport and assimilation, phosphoadenosine- Figure 5. Overlap of genes altered in gene expression levels among different myb knockout mutants. A to C, Venn diagrams in which each ring of the individual Venn diagram shows the number of genes present on the ATH1 array whose transcript level was statistically significantly altered as compared with the wild type by the given knockout mutation. Statistical significance was determined by individual gene ANOVAs using a FDR of 0.05 (Supplemental Table S10). A, Overlapping genes in the myb28-1, myb29-2, and myb76 single knockouts as compared with wild-type Col-0. B, Overlapping genes in the myb28-1, myb29-2, and myb28-1 myb29-1 knockout mutants as compared with wild-type Col-0. C, As in B, but showing only known and putative GLS genes (Table II). D, Plot of changed transcripts in the myb28-1 and myb29-2 single knockouts. The correlation plot was obtained by taking the 170 genes showing differential expression in both the myb28-1 and myb29-2 mutant lines but not in the myb28-1 myb29-1 double knockout (inset and B). These genes identify a negative correlation of effects between the two mutants (slope = 20.24, r = 0.42, P , 0.0001, 5#-phosphosulfate, Cys, S-adenosyl Met and Met production, and homo-Cys conversion. While sulfur utilization was not affected in the myb28-1 and myb76 single knockouts, myb29-2 was up-regulated in a sulfate transporter (SULTR5;1; At1g80310) together with three adenosine-5#-phosphate reductases, APR1 (At4G04610), APR2 (At1G62180), and APR3 (At4G21990; Supplemental Table S11) that all reduce adenosine-5#-phosphate to sulfite. In contrast, sulfate assimilation appeared to be down-regulated in the myb28-1 myb29-1 knockout. APR1 and APR3 were down-regulated 75% along with a significant downregulation of an ATP sulfurylase, ATPS1 (At3G22890), the sulfite reductase, SiR (At5G04590), and the sulfate transporters SULTR1;2 (At1g78000), SULTR2;1 (At5g10180), and SULTR4;2 (At3g12520). In opposition, the ATP sulfylase, ATPS2 (AT1G19920), and SULTR3;1 (At3g51895) were significantly up-regulated in the double knockout. In conclusion, the aliphatic GLS MYBs seems to regulate both biosynthetic transcripts and precursor availability, as suggested from data on overexpression lines (Sonderby et al., 2007;Malitsky et al., 2008).
Hitherto, all data support a role for the MYBs as exclusively positive regulators of aliphatic GLS genes. However, FMO GSOX4 , one of the sulfur-oxygenating enzymes in the secondary modification of GLSs (Li et al., 2008), was up-regulated in the myb28-1 myb29-1 double knockout but was not affected in any of the single knockouts (Fig. 6). This suggests that MYB28 and MYB29 in concert can function as negative regulators. To test if MYB28 and MYB29 can have unexpected combinatorial roles with unidentified genes, we generated a Venn diagram illustrating overlap in altered transcript levels between myb28-1, myb29-2, and myb28-1 myb29-1 (Fig. 5, B and C). A total of 170 genes were significantly changed in both myb28-1 and myb29-2 but not in the myb28-1 myb29-1 knockout (Fig.  5B). This represents genes whose significance disappears in myb28-1 myb29-1. When the changes in transcript levels of the 170 genes were plotted against each other, the correlation plot gave a negative slope (Fig.  5D). This suggests that MYB28 and MYB29 have opposing effects on most genes in this category, such that the myb28 knockout represses a gene that the myb29 knockout activates and vice versa. This antagonistic functionality could explain why the wild type and the myb28-1 myb29-1 double knockout have similar levels of these transcripts. In fact, only a few of these genes appear to be positively regulated by both MYB28 and MYB29.
Accordingly, MYB28 and MYB29 can act as both activators and repressors with a significant level of independence. The repressor effect is particularly conspicuous in the case of MYB29, since approximately two-thirds of the genes significantly changed in the myb29 knockout are up-regulated genes (Supplemental Table S10). It remains to be seen if these are direct or indirect targets of MYB29.

MYB76 Controls Secondary Modification Enzymes and MYB29
Analysis of the microarray data for the myb76 mutants did not show a significant down-regulation of MYB76 transcript ( Fig. 6; Table II), which is already very lowly abundant in the wild type. However, previous RT-PCR analysis of the myb76 knockdowns showed considerably reduced MYB76 transcript (Sonderby et al., 2007). In accordance with the modest 30% decrease in short-chained aliphatic GLSs (Fig. 3), the myb76 knockdown alleles were only affected in three aliphatic biosynthetic transcripts: two in the secondary modification pathway, FMO GSOX1 and AOP2 (the protein is nonfunctional in the Col-0 accession but the gene is nevertheless transcribed; Kliebenstein et al., Figure 6. Altered transcript accumulation for GLS biosynthetic genes in different myb knockouts. The scheme denotes the GLS biosynthetic pathway, with each arrow representing a biosynthetic step in the pathway and the corresponding gene listed on the side. Boxes in the arrows illustrate changes in transcription levels of the respective biosynthetic gene in the various knockouts, myb28-1, myb29-2, myb76, and myb28-1 myb29-1, versus wild-type (WT) Col-0. Green shades represent significant decreases in expression, pink represents significant increases in expression, and white represents no significant change. A, The Met chain-elongation machinery. B, Synthesis of the core methylthioalkyl-GLS structure. C, Secondary modifications of the side chain. D, The MYB transcription factors. [See online article for color version of this figure.] 2001b), and the sulfotransferase SOT17 in core biosynthesis ( Fig. 6; Table II). These data agree with the myb76 knockdown having its largest effect on methylsulfinylated short-chained aliphatic GLSs (Fig. 3), which are predominantly sulfur oxygenated by FMO GSOX1 in the leaves . Interestingly, MYB29 transcript was significantly decreased in both myb76 knockdowns, suggesting that part of the effect in the myb76 knockdowns might be conferred through decreased expression of MYB29.
Altogether, the microarray data draw a picture of an interdependency of the three MYB genes, with each gene controlling its independent module to uphold GLS biosynthesis. Furthermore, the data reveal that the transcriptional level of biosynthetic genes does not fully reflect the level of aliphatic GLSs in the plant.

Knockout Mutants of MYB28, MYB29, and MYB76 Show Altered Distribution of Aliphatic GLSs within the Leaf
Results based on dissection and ion intensity maps from matrix-assisted laser-desorption ionization time of flight of leaves have shown that GLSs are particularly abundant in the edge and midvein (Shroff et al., 2008). The abundance of GLSs in the edge is in contrast to the expression patterns obtained by promoter-GUS fusions of genes from chain elongation, core biosynthesis, and aliphatic regulation genes, which all indicate that biosynthesis takes place in and around the veins of leaves (Reintanz et al., 2001;Chen et al., 2003;Grubb et al., 2004;Schuster and Binder 2005;Gigolashvili et al., 2007bGigolashvili et al., , 2008Malitsky et al., 2008).
To test whether the three MYBs play a role in controlling the spatial distribution of GLSs in Arabidopsis leaves, we dissected leaves of T-DNA knockout mutants and overexpression lines of MYB28, MYB29, or MYB76 into three portions (vein, inner lamina, and edge) and measured GLS levels ( Fig. 7; Supplemental Tables S12-S18). The data obtained in three independent experiments were pooled, since the ANOVA showed no significant difference between experiments. Generally, GLS chemotypes were as previously observed, except that myb76-1 did not display a statistically significant decrease in 4msb in this experiment, even though it was still slightly lower in aliphatic GLSs than in the wild type (Supplemental Tables S12 and S13), and the chemotypes of the MYB overexpression lines were closer to that of the wild type in comparison with previous observations (Supplemental Tables S12 and S13). These differences are probably due to different growth conditions as compared with previous experiments (Sonderby et al., 2007). There was no difference in total leaf weights or weights of leaf parts among the genotypes (Supplemental Table S14), thereby allowing us to compare relative distributions of GLSs instead of absolute concentrations.
In wild-type Col-0 leaves, 52% of short-chained aliphatic GLSs was found in the leaf edge, with the rest being divided equally between the vein and inner lamina ( Fig. 7; Supplemental Tables S15 and S16). This distribution reflects that more than two-thirds of 4mtb and approximately half of 4msb accumulated in the edge ( Fig. 7; Supplemental Table S15). The longchained aliphatic GLS, 8mso, was more equally distributed throughout the leaf, with the highest portion in the lamina ( Fig. 7; Supplemental Table S15). Conversely, 39% of indole GLSs were found in the vein, slightly less in the lamina, and the least in the edge ( Fig. 7; Supplemental Table S15).
If the MYBs are controlling GLS distribution, then ectopic expression of the MYBs driven by the ubiquitously expressed 35S promoter would result in aliphatic GLSs being produced equally in the three leaf sections. As expected, an even distribution of total aliphatic GLS levels was observed across the sections  Table S1) on top. The number to the right of each leaf corresponds to the total amount of GLS in nmol in the whole leaf. The colors denote proportions of a GLS in each leaf section in relation to total amount per leaf. Red squares and green circles indicate significant increases and decreases, respectively, in GLS distribution in comparison with the wild type. For exact relative distributions, see Supplemental Table S15 of the leaves overexpressing MYB28, MYB29, and MYB76 (Sonderby et al., 2007;Supplemental Tables S17 and S18). This suggests that the spatial distribution of GLSs within the wild-type leaf is at least partially due to the spatial expression pattern of the MYBs.
In the myb knockout mutants, the relative distribution of short-chained aliphatic GLSs in the leaves was also changed (Fig. 7). For myb28-1 and myb29-2, we observed decreased partitioning of all short-chained aliphatic GLSs in the vein and lamina, with elevated fractions in the margins. In contrast, for myb76-1 and myb28-1 myb76-2 mutants, decreased levels of 4mtb were observed in the margin but elevated levels were seen in the vein and lamina, suggesting that 4mtb is trapped in these tissues when MYB76 is nonfunctional. Interestingly, 4msb in the MYB76 mutants followed the trend to margin accumulation found in the myb28-1 and myb29-2 knockouts, which resulted in myb76 having a distribution of short-chained aliphatic GLSs similar to wild-type Col-0. Our data suggest that the MYBs contribute to the determination of GLS distribution within the leaf and that MYB76 may play a specific role in the distribution of 4mtb.
In spite of their effect on the distribution of shortchained aliphatic GLS distribution, none of the knockouts affected the distribution of the long-chained aliphatic GLS, 8mso ( Fig. 7; Supplemental Table S15). This suggests a different regulation of the spatial distribution of long-chained aliphatic compared with short-chained aliphatic GLSs. As expected, distribution of indole GLSs in the leaf was not altered between wild-type Col-0 and any of the lines with altered levels of MYB28, MYB29, or MYB76, which corroborates the lack of impact of the three aliphatic MYBs on indole GLSs ( Fig. 7; Supplemental Tables S15 and S17). The only exception to this observation was the myb28-1 myb29-2 mutant, in which a slightly decreased proportion was present in the vein, which suggests that indole GLS distributions may be slightly affected when aliphatic GLS are heavily decreased.

DISCUSSION
Prior to this study, MYB28, MYB29, and MYB76 were already firmly established as players in the regulation of aliphatic GLS biosynthesis (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008Hirai et al., 2007;Sonderby et al., 2007;Beekwilder et al., 2008). Based on data from overexpression lines, clearly all three genes, MYB28, MYB29, and MYB76, were capable of inducing aliphatic GLSs, and they seemed to confer their regulation by positively regulating the transcription of aliphatic biosynthetic genes (Fig. 1A). Based on the chemotypes of the single knockouts, MYB28 was regarded as the major important player in planta, MYB29 as a lieutenant, and MYB76 bearing an accessory role ( Fig. 1B; Gigolashvili et al., 2009a). In this paper, we challenged the existing model, added significant information to the puzzle of the interaction between the genes (Fig. 8A), and suggest an improved model for the role of MYB28, MYB29, and MYB76 in the regulation of aliphatic biosynthesis in planta (Fig. 8B).

MYB76 Plays an Important Role in the Regulation of Transcription of Genes in Secondary Modifications and MYB29
Previously, myb76 single T-DNA knockdowns were shown to have a small reduction in short-chained methylsulfinylalkyl-GLSs (Sonderby et al., 2007). In accordance with this chemotype (Fig. 3), the transcriptional analysis showed that transcription of FMO GSOX1 , responsible for the conversion of methylthioalkyl-to methylsulfinylalkyl-GLSs, was decreased in the myb76 knockdowns ( Fig. 6; Table II). However, the role of MYB76 is not limited to FMO GSOX1 , as it also decreases transcripts of AOP2, another secondary modification enzyme, the core biosynthetic sulfotransferase SOT17, as well as MYB29 (Table II). One might speculate that the effect of MYB76 is purely performed through MYB29, but the apparent dependency is not absolute, since overexpression of MYB76 in the almost aliphatic GLS-free myb28 myb29 double knockout background induced significant production of aliphatic GLSs ( Fig.  2; Supplemental Table S2), thereby showing independence of MYB76 from MYB28 and MYB29 for the induction of aliphatic GLSs. This observation is consistent with the detectable induction of GLSs in the myb28 myb29 double knockout when challenged with the lepidopteran insect Mamestra brassicae (Beekwilder et al., 2008) or under variable growth conditions (data not shown), which might be explained by induction of the normally lowly expressed MYB76 (Sonderby et al., 2007;Gigolashvili et al., 2008).
Transactivation assays performed in cultured Arabidopsis cells left no doubt that MYB76 can activate the promoters of both chain elongation and core biosynthetic enzymes, MAM1, MAM3, CYP79F1, CYP79F2, CYP83A1, and SUR1 (Gigolashvili et al., 2008), whereas the capability of MYB76 to activate the enzymes doing secondary modifications was not tested. Therefore, we can only speculate whether the specific effect on these genes is caused by a comparably higher binding affinity in comparison with aliphatic core biosynthetic enzymes in planta. Another proof for the role of MYB76 in planta is added by the additive effect observed in the myb28-1 myb76 double knockout on foliar aliphatic GLS levels ( Fig. 3; Table I). This is in contrast with the epistatic effect observed in the myb28 myb29 double knockout, which stresses that, even though myb29 and myb76 single knockout mutants have similar decreases in levels of short-chained aliphatic GLSs (Sonderby et al., 2007), MYB29 has a more prominent role in upholding GLS accumulation. The only epistatic effect on single aliphatic GLSs observed in the myb28-1 myb76 double knockout data set was on 4mtb, which agrees with a combined interaction of MYB28 and MYB76, leading to an epistatic effect on the FMO GSOX s. In addition, there is an epistatic effect of MYB76 on total indole GLSs. The doubling of total indole GLSs in the myb28-1 mutant (Fig. 3) is repressed in the myb28-1 myb76 mutants, which indicates a possibly repressive effect of MYB28 on indole GLSs but also that MYB76 might be necessary for the induction of indole GLSs. Of the three aliphatic MYBs, only MYB76 shows induction of CYP79B2 in transactivation assay in Arabidopsis cells 2 to 3 d after induction (Gigolashvili et al., 2008). Further experiments are needed to explore a putative link of MYB76 to indole GLS accumulation.
MYB76 does not play a role in seeds, since the aliphatic seed GLS level in myb28-1 myb76 knockouts was not different from the myb28-1 knockout (Table I). This agrees with a MYB76 promoter-GUS analysis, in which MYB76 expression was not found in siliques but only in inflorescences (Gigolashvili et al., 2008). In summary, we have shown a significant role for MYB76 in determining GLS profile and leaf distribution.

Transcription of Biosynthetic Genes Does Not Reflect the Chemotype
The general view of positive regulators of metabolic pathways is that they act through the induction of biosynthetic transcripts and transporters, channeling intermediates between different compartments. However, in all the myb knockout mutants, there is a lack of correlation between total levels of biosynthetic tran-scripts and chemotypes. GLS biosynthetic transcripts are heavily down-regulated in the myb28-1 single knockout ( Fig. 6; Table II), which points to MYB28 as the major transcriptional activator of aliphatic biosynthetic transcription in Arabidopsis (accession Col-0), as suggested previously (Gigolashvili et al., 2007b;Hirai et al., 2007). However, it remains a puzzle how myb28-1 can retain as much as 73% of short-chained aliphatic GLSs in comparison with the wild type (Fig.  3), with its relatively low levels of biosynthetic transcripts (Table II).
Likewise, in spite of its almost GLS-free chemotype, biosynthetic transcripts in the myb28-1 myb29-1 double knockout are remarkably close to the level measured in myb28-1 ( Fig. 6; Table II). Similarly, no change in biosynthetic transcripts was observed in the myb29-2 mutant in comparison with the wild type (Table II; Fig.  6) in spite of its short-chained aliphatic GLS chemotype (Sonderby et al., 2007;Beekwilder et al., 2008;Gigolashvili et al., 2008). As such, the metabolic chemotype is uncoupled from the total level of biosynthetic transcripts, which could indicate that a decrease in one or a few biosynthetic transcripts is enough to change total aliphatic GLS accumulation. Alternatively, other mechanisms besides direct transcriptional regulation of aliphatic GLS biosynthesis are at play when determining aliphatic GLS accumulation.
The lack of effect on biosynthetic transcripts in the myb29-2 mutant could indicate that it is not a positive regulator of biosynthetic transcripts. However, previous transactivation assays showed that MYB29 alone activates transcription of the promoters of aliphatic biosynthetic genes in cultured Arabidopsis cells (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008. Together with the remaining short-chained foliar and seed aliphatic GLSs in the myb28-1 myb76 double knockout, this supports a role for MYB29 as a positive regulator of aliphatic biosynthetic transcripts. Therefore, the missing effect on transcripts in myb29-2 is probably due to a concurrent increase in MYB28 transcript (Table II; Fig. 6). However, since the myb29 knockouts have altered chemotypes on short-chained aliphatic GLSs (Sonderby et al., 2007;Beekwilder et al., 2008;Gigolashvili et al., 2008), MYB29 must affect other determinants for GLS accumulation besides the transcription of biosynthetic genes.
Changes in transcript levels related to sulfur assimilation were observed in the myb29-2 mutant and the myb28-1 myb29-1 double knockout (Supplemental Table S11). This may indicate that a concomitant decrease in aliphatic GLS biosynthetic transcripts and limitation in sulfur availability may be the cause of the uncoupling observed in the myb28-1 myb29-1 knockout. Further studies are necessary to explain the uncoupling of aliphatic biosynthetic transcripts from GLS levels.
transcript of MYB29 was significantly down-regulated and substantially decreased in the myb76 and myb28-1 knockouts, respectively (Table II), the latter supported by quantitative PCR data on a different myb28 mutant (Beekwilder et al., 2008). Similarly, out of the three transcription factors, only MYB29 was up-regulated in moderate overexpression lines of all three MYBs (Sonderby et al., 2007), which is further supported by independent data on overexpression lines in which MYB76 was concurrently up-regulated (Gigolashvili et al., 2008). This seems to point toward a positive regulation by MYB28 and MYB76 on MYB29, which consequently plays an important role in the integration of signals from MYB28 and MYB76. Possibly, MYB29 constitutes a positive feed-forward loop in aliphatic GLS regulation similar to the indole GLS regulator, MYB122, which activates the other indole GLS regulator, MYB51 (Gigolashvili et al., 2007a). Other examples of MYB feed-forward loops exist within anthocyanin biosynthesis, where the positive anthocyanin regulator PRODUCTION OF ANTHOCY-ANIN PIGMENT1 feed forwards on MYB114, an R2R3 MYB transcription factor present in the same subclade (Dare et al., 2008). It might be a general feature of these small subclades of R2R3 MYB factors to primarily feed forward on one transcription factor that thus amplifies the signal from the other activators.
MYB29 cannot control indispensable regulatory functions, though, since the lack of MYB29 only results in moderate decreases in aliphatic GLS levels (Sonderby et al., 2007;Beekwilder et al., 2008;Gigolashvili et al., 2008). The compensation of MYB28 in the myb29-2 mutant can be interpreted in two different ways: MYB29 acts as a repressor of MYB28 in wild-type Col-0, or alternatively, the plant may sense the decrease in aliphatic GLSs and subsequently act to compensate by up-regulating MYB28. This points to a possible feedback mechanism on MYB28 by the lack of end products, which is a feature previously suggested by Mugford et al. (2009) and further supported by the up-regulation of CYP79B2 in the absence of CYP79B3 and visa versa (Celenza et al., 2005). Curiously, such a compensatory mechanism only feeds back on MYB28 in the absence of MYB29, since neither MYB76 nor MYB29 is up-regulated in the GLS-lacking myb28-1 knockout, nor is MYB28 in the myb76-1 knockdown. These data seem to indicate different signaling pathways for the activation of MYB28, MYB29, and MYB76.

The Presence of Unknown Interacting Partners
The in planta induction of MYB29 by MYB28 and MYB76 (Table II; Beekwilder et al., 2008) is inconsistent with transactivation data performed in Nicotiana benthamiana, in which only MYB76 could be induced by MYB28, MYB29, and MYB76 whereas neither MYB29 nor MYB28 showed any transactivation (Gigolashvili et al., 2008). Possibly, MYB28 and MYB76 activate MYB29 through another mechanism than direct transcriptional activation, maybe through the induction of an unknown transcription factor that subsequently induces MYB29. Alternatively, an unknown transcription factor, not present in N. benthamiana, interacts with MYB28 and MYB76 to facilitate the induction of MYB29. A common feature of R2R3 MYB transcription factors is the combinatorial control with bHLH and WD repeat proteins, as exemplified in the anthocyanin and proanthocyanidin biosynthesis (Davies and Schwinn, 2003). Recently, a bHLH transcription factor interacting with the indole GLS regulator MYB51 was identified (T. Gigolashvili, personal communication), which suggests that a similar bHLH may interact with the aliphatic MYBs.
So far, the induction of long-chained aliphatic GLSs has largely been ascribed to the regulatory realm of MYB28, due to its long-chained aliphatic GLS chemotype in the myb28 knockouts. However, trace amounts of long-chained aliphatic GLSs occasionally occur in both the myb28-1 mutant (Hirai et al., 2007; Table I) and the myb28-1 myb76 mutants (Table I), which indicates that under certain conditions at least MYB29 can induce long-chained aliphatic GLSs in the absence of MYB28. Similarly, ectopic expression of MYB29 and MYB76 results in increased levels of long-chained aliphatic GLSs both in the absence (Figs. 2 and 4) and the presence of MYB28 (Sonderby et al., 2007;Gigolashvili et al., 2008). Ectopic overexpression of MYB28 in wild-type Col-0, on the other hand, did not result in increased long-chained aliphatic GLSs (Gigolashvili et al., 2007b;Sonderby et al., 2007). This speaks for the presence of an additional factor needed for the induction of long-chained aliphatic GLSs localized outside of the tissue where MYB29 and MYB76 are normally expressed and with which only these two MYBs can interact.

Roles of MYB28, MYB29, and MYB76 in Leaf GLS Distribution
The primary accumulation sites for GLSs have been shown to be in the edge and midvein (Shroff et al., 2008), which is supported by our GLS distribution analysis. However, the accumulation in the edge is in contrast to the expression pattern of the regulatory proteins, which indicates that production takes place in the veins (Gigolashvili et al., 2007b(Gigolashvili et al., , 2008Malitsky et al., 2008). This suggests that the edge constitutes a "GLS sink" to which GLSs are transported from the veins (the source) through a yet unknown mechanism. Our analysis on the different MYB overexpression lines clearly showed that the putative source-sink relation between the veins and the edge is interrupted if the MYBs are ectopically expressed (Supplemental Tables S17 and S18). Thus, the specific spatial expression of the MYBs is at least partially responsible for the distribution of aliphatic GLSs in the leaf.
Since total GLS levels go down, the increase in the proportion of short-chained aliphatic GLSs in the leaf edge observed in the myb28-1 and myb29-2 knockouts is expected under the assumption of unchanged trans-port processes (Fig. 7). However, the distribution of aliphatic GLSs in the myb76-1 and myb28-1 myb76-2 knockouts did not conform to this tendency. Instead, a smaller proportion of 4mtb is present in the edge in myb76-1 and myb28-1 myb76-2 in comparison with the wild type ( Fig. 7; Supplemental Table S15). This cannot be explained solely by the presence of less GLSs, since the GLS levels in myb76-1 under these growth conditions are similar to wild-type Col-0 ( Fig. 7; Supplemental Table S12). Furthermore, given the small amount of 4msb in the myb28-1 myb76-2 double knockout and presuming unchanged transport, one would expect a higher proportion of GLSs in the edge, as observed in the myb28-1 and myb29-2 knockouts. Additionally, since 4mtb follows the same pattern in the myb28-1 myb76-2 double knockout as in the myb76-1 single knockdown, this reveals a dominance of MYB76 over MYB28 in controlling spatial patterns. This indicates that MYB76 changes the distribution of shortchained aliphatic GLSs from the vein to the edge in wild-type leaves, possibly by the regulation of a transporter. So far, no transporters of GLSs have been identified, even though several putative Suc transporters have been proposed to be involved in GLS transport (Nour-Eldin and Halkier, 2009). However, since the distribution of GLSs is still biased toward the leaf edge in the myb76-1 and myb28-1 myb76-2 knockouts, MYB76 cannot be the only player in the regulation of distribution.

CONCLUSION
To our knowledge, this is the first comprehensive genome-wide transcriptional analysis of regulatory single knockout mutants of transcription factors all involved in the same metabolic pathway and without whose presence the metabolites of the corresponding pathway are not produced. This analysis showed that the level of biosynthetic transcripts was uncoupled from the levels of metabolites in the knockouts. This feature might hold true for other pathways as well and has implications for the way we view regulation of secondary metabolism. Our findings enabled us to decipher the individual roles of MYB28, MYB29, and MYB76 and to identify new positive and negative cooperative interactions among these transcription factors that shape the aliphatic GLS profile. In combination with new insights about MYB76 controlling the spatial distribution in the plant, this has led to an improved model for aliphatic GLS regulation (Fig. 8B). The next step in the field will be to identify putative interaction partners of the MYBs to explain their complex interactions and to find what regulates the regulators.

GLS Extraction and Analysis
GLS extraction was performed as described previously (Sonderby et al., 2007). For analysis of leaves, 30 to 70 mg of leaves was harvested just before bolting (22-25 d after germination). For analysis of seeds, 10 seeds were used for the extraction. HPLC analysis was performed as described previously .

GLS Statistical Analysis
GLS contents were analyzed via ANOVA utilizing SAS proc glm. All comparisons contained multiple independent experiments that were tested for significance as a fixed effect to allow for testing an interaction with genotype. For MYB76, there were two independent T-DNA mutant lines. Both lines were assayed, and the difference between the independent mutants was tested as a nested factor using the factor MYB76 (line). This allowed us to test if the effects on GLS accumulation were due to the presence of any T-DNA within MYB76 (line) or if there was a difference between the separate T-DNA alleles. Sums of squares, F values, and P values are presented. More details on specific models and sample numbers can be found in the tables of the individual experiments.

Dissection Experiment
Wild-type Col-0, transgenic lines, and T-DNA mutants (see above) were grown in Pindstrup 2 sphagnum medium (Pindstrup Mosebrug) in a growth chamber (HEMZ 20/240/S; Thermo Heraeus) with 100 mE light intensity, 8/16-h light/dark cycle, 17°C to 20°C, and 70% relative humidity. At 6 to 7 weeks after germination, two leaves including the entire petiole were cut from each plant. One leaf was immediately submerged in 300 mL of 85% (v/v) methanol containing 0.02 mM sinigrin as internal standard. The other leaf was dissected with a scalpel with two cuts 1 mm on either side of the midvein and three cuts along the edges, about 1 mm from each edge, according to the drawings in Figure 7. Each section was weighed and immediately submerged in 300 mL of 85% methanol supplemented with 0.02 mM sinigrin. Four replicates were made per genotype. GLSs were extracted and measured as described previously (Sonderby et al., 2007). The experiment was repeated three times.

Dissection Statistical Analysis
To compare GLS partitioning among individual genotypes, we added up total GLS amounts in each dissected leaf and calculated the percentage of each individual GLS in each separate section. The three sections were then utilized as three variables within a single section factor in the ANOVA. Three independently replicated experiments for the dissection experiment each with four plants per genotype were done. Experiment 3 experiment and genotype 3 experiment interactions were included in the original model, but Complex Aliphatic Glucosinolate Regulation Plant Physiol. Vol. 153, 2010 as the genotype 3 experiment term showed no significance, it was dropped from the final model. To test if GLSs are lost in the dissection, we individually summed up all GLSs in the dissected sections and statistically compared the levels of dissected and undissected leaves. No difference was observed in GLS levels between the dissected and intact leaves as tested by ANOVA (data not shown). This was done for all genotypes. As such, we combined all leaves to calculate the average total GLS level in a single leaf from each genotype.

Microarray Analysis of MYB Knockouts
Plants for the various genotypes were grown as described previously (Sonderby et al., 2007). At 25 d post germination, a fully expanded mature leaf was harvested, weighed, and analyzed for total aliphatic GLS content via HPLC. The remaining plant material was collected and flash frozen, and total RNA was extracted via RNeasy columns (Qiagen). Two independent plants were combined to provide sufficient starting material for a single RNA extraction. Two independent samples were obtained per mutant, thus providing 4-fold replication. Eight wild-type Col-0 RNA samples were obtained. This provided a total of 32 independent microarrays. Labeled copy RNA was prepared and hybridized, according to the manufacturer's guidelines (Affymetrix), to whole genome Affymetrix ATH1 GeneChip microarrays, containing 22,746 Arabidopsis transcripts. The GeneChips were scanned with an Affymetrix GeneArray 2500 Scanner, and data were acquired via the Microarray Suite software MAS 5.0 at the Functional Genomics Laboratory (University of California, Berkeley). Robust Multichip Average normalization was used to obtain gene expression levels for all data analyses (Irizarry et al., 2003).

Microarray Statistical Analysis
The gene expression data were analyzed via individual gene ANOVA for each transcript. This was done by conducting ANOVA on each gene using the independent samples for both mutants and the wild type. The ANOVA calculations were programmed into Microsoft Excel to obtain all appropriate sums of squares and the F values for the effect of the genotype (wild type versus mutant) and replicate effects for each transcript. The P values for genotype (wild type versus mutant) are presented after a FDR adjustment to the 0.05 level (Supplemental Table S10; Benjamini and Hochberg, 1995). For downstream analysis, biosynthetic pathways were obtained from AraCyc version 3.4 (http://www.arabidopsis.org/biocyc/) and modified to better organize the pathways based on metabolites of importance for GLS synthesis (Supplemental Table S10).

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Table S1. GLS abbreviations and chemical structures.
Supplemental Table S9. GLS gene microarray analysis comparing Col-0 versus various MYB loss-of-function knockouts.
Supplemental Table S11. Microarray data, selected transcripts, and sulfur utilization biosynthetic pathways.
Supplemental Table S12. GLS levels in whole leaf and sum of dissection portions.
Supplemental Table S13. ANOVAs for leaf GLSs in whole leaf and sum of dissection portions.
Supplemental Table S15. Fractions of GLSs in dissected leaves of MYB knockouts.
Supplemental Table S16. ANOVAs for fractions of GLSs in dissected leaves of MYB knockouts.
Supplemental Table S17. Fractions of GLSs in dissected leaves of plants overexpressing MYB transcription factors.
Supplemental Table S18. ANOVAs for proportions of GLSs in dissected leaves of plants overexpressing MYB transcription factors.