Myzus persicae (green peach aphid) feeding on Arabidopsis induces the formation of a deterrent indole glucosinolate

Authors


(fax (607) 254 2958; e-mail gj32@cornell.edu).

Summary

Cruciferous plants produce a wide variety of glucosinolates as a protection against herbivores and pathogens. However, very little is known about the importance of individual glucosinolates in plant defense and the regulation of their production in response to herbivory. When Myzus persicae (green peach aphid) feeds on Arabidopsis aliphatic glucosinolates pass through the aphid gut intact, but indole glucosinolates are mostly degraded. Although aphid feeding causes an overall decrease in Arabidopsis glucosinolate content, the production of 4-methoxyindol-3-ylmethylglucosinolate is induced. This altered glucosinolate profile is not a systemic plant response, but is limited to the area in which aphids are feeding. Aphid feeding on detached leaves causes a similar change in the glucosinolate profile, demonstrating that glucosinolate transport is not required for the observed changes. Salicylate-mediated signaling has been implicated in other plant responses to aphid feeding. However, analysis of eds5, pad4, npr1 and NahG transgenic Arabidopsis, which are compromised in this pathway, demonstrated that aphid-induced changes in the indole glucosinolate profile were unaffected. The addition of purified indol-3-ylmethylglucosinolate to the petioles of cyp79B2 cyp79B3 mutant leaves, which do not produce indole glucosinolates, showed that this glucosinolate serves as a precursor for the aphid-induced synthesis of 4-methoxyindol-3-ylmethylglucosinolate. In artificial diets, 4-methoxyindol-3-ylmethylglucosinolate is a significantly greater aphid deterrent in the absence of myrosinase than its metabolic precursor indol-3-ylmethylglucosinolate. Together, these results demonstrate that, in response to aphid feeding, Arabidopsis plants convert one indole glucosinolate to another that provides a greater defensive benefit.

Introduction

Glucosinolates are defense-related secondary metabolites found in all Brassicaceae (Fahey et al., 2001), including agriculturally important oilseed and vegetable crops in the Brassica genus and the closely related genetic model plant Arabidopsis. Upon tissue damage, glucosinolates come into contact with and are hydrolyzed by endogenous plant myrosinases (β-thioglucosidases, EC 3.2.1.147). Further catalysis and the likely spontaneous degradation results in the production of nitriles, epithionitriles, oxazolidine-2-thiones, thiocyanates and isothiocyanates (Halkier and Gershenzon, 2006). These glucosinolate breakdown products typically have a deterrent effect on generalist herbivores, but can serve as attractants for crucifer-feeding specialists. Well over 100 glucosinolates have been identified in nature (Fahey et al., 2001) and, although it is often assumed that this diversity is necessary for defense against a multitude of herbivores and pathogens, relatively little is known about the specificity of individual glucosinolates in plant defense.

The majority of Arabidopsis thaliana (Arabidopsis) foliar glucosinolates are derived from tryptophan (indole glucosinolates), methionine (aliphatic glucosinolates), and phenylalanine. Modification of the amino acid side chains greatly increases the diversity of the glucosinolate profile (Petersen et al., 2002; Reichelt et al., 2002). Many, perhaps most, Arabidopsis glucosinolate biosynthetic enzymes have been identified using genetic and biochemical approaches. Known enzymes include those involved in the methionine chain elongation cycle (Field et al., 2004; Kroymann et al., 2001; Textor et al., 2004), modifications of the glucosinolate side chains (Kliebenstein et al., 2002b) and the formation of the core glucosinolate structure (Bak et al., 2001; Chen et al., 2003; Hansen et al., 2001; Hull et al., 2000; Mikkelsen et al., 2000; Piotrowski et al., 2004; Reintanz et al., 2001; Wittstock and Halkier, 2000) (Figure 1). Indol-3-ylmethyl glucosinolate (I3M) may be modified by the addition of hydroxy and methoxy groups (Figure 2), although the enzymes catalyzing these reactions have not yet been identified. Double knockout mutations of CYP79B2 and CYP79B3 cause an almost complete lack of indole glucosinolates (Zhao et al., 2002). Indole glucosinolate production is similarly blocked in sur1 and sur2 (cyp83B1) mutants (Bak and Feyereisen, 2001; Mikkelsen et al., 2004; Naur et al., 2003), but these plants also have severe morphological defects as a result of changes in auxin metabolism.

Figure 1.

 Pathways for the biosynthesis of methionine- and tryptophan-derived glucosinolates in Arabidopsis.

Figure 2.

 Possible pathways for the biosynthesis of methoxyindole glucosinolates. Indol-3-ylmethyl (I3M), 4-hydroxyindol-3-ylmethyl (4OHI3M), 4-methoxyindol-3-ylmethyl (4MI3M) and 1-methoxyindol-3-ylmethyl (1MI3M) glucosinolates are shown.

Although the glucosinolate-myrosinase system exists as a pre-formed defense, there are also numerous reports of induced glucosinolate responses in the Brassicaceae. The indole glucosinolate class is most commonly induced during biotic stress, including the response to fungal pathogens (Doughty et al., 1991), bacterial elicitors (Brader et al., 2001), root flies (Griffiths et al., 1994; Hopkins et al., 1998), caterpillars (Traw, 2002), flea beetles (Bartlet et al., 1999; Rostas et al., 2002) and aphids (Mewis et al., 2006). Moreover, mechanical wounding (Bodnaryk, 1992) and treatment with methyl jasmonate (Bartlet et al., 1999; Bodnaryk, 1994; Doughty et al., 1995; Kliebenstein et al., 2002a; Mewis et al., 2005; Mikkelsen et al., 2004; Piotrowski et al., 2004; Sasaki-Sekimoto et al., 2005) can increase indole glucosinolate levels. Two recent Arabidopsis studies showed that, in contrast to I3M and 1-methoxyindol-3-ylmethyl glucosinolate (1MI3M), production of 4-methoxyindol-3-ylmethyl glucosinolate (4MI3M) is induced by a salicylate-mediated pathway (Kliebenstein et al., 2002a; Mikkelsen et al., 2004).

Myzus persicae (green peach aphid) is the most common aphid pest in the Arabidopsis greenhouse and growth-chamber settings (Bush et al., 2006). This aphid has a worldwide distribution, infests hundreds of species from 40 plant families (Blackman and Eastop, 2000) and is commonly found on Brassica crops and weedy crucifers. Mewis et al. (2005) showed that aphid reproduction is negatively correlated with both indole and aliphatic glucosinolate content in defense-related Arabidopsis mutants. However, other studies found no correlation between the glucosinolate content of rapeseed cultivars and M. persicae fecundity (Weber et al., 1986), a positive correlation of aphid reproduction with I3M content and a negative correlation with the 3-butenyl glucosinolate content in wild and cultivated Brassica species (Cole, 1997). Analysis of Arabidopsis transcription during aphid feeding demonstrated the up-regulation of genes involved in salicylate-mediated defense signaling and oxidative stress responses (Moran and Thompson, 2001; Moran et al., 2002). Affymetrix microarray analysis of Arabidopsis transcription during M. persicae infestation showed extensive up- and down-regulation of gene expression (De Vos et al., 2005). Here we show that in response to M. persicae feeding on Arabidopsis, there is an overall decrease in glucosinolate content and a conversion of I3M to the more aphid-deterrent 4MI3M.

Results

Analysis of microarray expression data (De Vos et al., 2005) showed that transcription of almost all Arabidopsis glucosinolate biosynthetic genes was repressed to less than 50% of uninfested levels when M. persicae were allowed to move freely on 5-week-old rosette-stage Columbia-0 (Col-0) plants for 2–3 days (Table S1). Although the rate-limiting step(s) in glucosinolate biosynthesis are not known, it seemed likely that such transcriptional changes would decrease the glucosinolate content. Consistent with the gene expression data, we observed an overall glucosinolate decrease in 2-week-old rosette stage Col-0 and Landsberg erecta (Ler) plants infested with aphids for 3 days (Figure 3a,b). Microarray experiments performed at earlier time points after aphid infestation (2, 8 and 36 h) showed no significant changes in the transcription of known glucosinolate-related genes (J. Pritchard, University of Birmingham, UK personal communication; Table S1). Similarly, we did not observe significant changes in the total glucosinolate content after only 24 h of aphid feeding (Figure S1).

Figure 3.

 Glucosinolate content in whole aphid-infested Arabidopsis thaliana Col-0 (a) and Ler (b) plants, and in cabbage (c), expressed as a percentage of the levels of each glucosinolate (nmol mg−1 dry tissue) in uninfested control plants. Mean ± SD, n = 8; *P < 0.05, **P < 0.01 by two-tailed Student's t-test. Aliphatic, total methionine-derived glucosinolates; DW, plant dry weight at the end of the experiment; Indolyl, total tryptophan-derived glucosinolates; Total, total plant glucosinolate content. Glucosinolate side chain abbreviations: 3MSP, 3-methylsulfinylpropyl; 4MSB, 4-methylsulfinylbutyl; 4MTB, 4-methylthiobutyl; 5MSP, 5-methylsulfinylpropyl; 7MSH, 7-methylsulfinylheptyl; 8MSO, 8-methylsulfinyloctyl; 8MTO, 8-methylthiooctyl; I3M, indol-3-ylmethyl; 4MI3M, 4-methoxyindol-3-ylmethyl; 1MI3M, 1-methoxyindol-3-ylmethyl; 3HP, 3-hydroxypropyl; 4OHI3M, 4-hydroxyindol-3-ylmethyl. Additional runs of this experiment are shown in Figure S2.

Despite the overall decrease in glucosinolate content after three days of M. persicae feeding, the indole glucosinolate 4MI3M was significantly increased in both Col-0 and Ler (Figure 3a,b). Three independent runs of the Col-0 experiment showed similar results (Figure S2), with an anova showing significant differences between runs only for 1MI3M (F2,27 = 3.52, P = 0.04; Table S2). In two runs of the experiment with Ler, significant differences in the induction were observed only for 3-methylsulfinylpropyl (3MSP) (Student's t-test, P = 0.01; Table S2). Cabbage (Brassica oleracea var. Wisconsin Golden Acre) also showed a significant induction of 4MI3M (Figure 3c) in two independent runs of the experiment, suggesting that increased 4MI3M may be a more general crucifer response to M. persicae feeding. The relative induction of 1MI3M and sinigrin, but not of other glucosinolates, varied significantly between the two runs of the cabbage experiment (Student's t-test, P = 0.04; Table S2). The negative effect of aphid feeding on both Arabidopsis and cabbage is demonstrated by the fact that the dry weight of aphid-infested plants was lower than that of controls in each case (Figure 3).

Changes in the leaf glucosinolate profile could result from transport from the roots, which have a relatively high indole glucosinolate content (Brown et al., 2003; Petersen et al., 2002). To test this hypothesis, we measured the glucosinolate content of detached leaves with their petioles inserted in tubes of water. As in the case of whole plants, aphid feeding on detached leaves of Col-0 and Ler increased 4MI3M, decreased the total glucosinolate content and decreased the dry weight relative to control leaves (Figure 4a,b), showing that transport between different parts of the plant is not a prerequisite for the observed changes. Although the levels of 4MI3M in plants infested with aphids were consistently and significantly increased in additional runs of the experiment (Figure S3), the relative reduction of some of the other glucosinolates was different across the three Col-0 and two Ler runs (one-way anova, P < 0.05; Student's t-test, P < 0.05; Table S2).

Figure 4.

 Glucosinolate content in detached leaves of aphid-infested Arabidopsis thaliana Col-0 (a) and Ler (b) plants, expressed as a percentage of the levels of each glucosinolate in uninfested control plants. Mean ± SD, n = 8; *P < 0.05, **P < 0.01 by a two-tailed Student's t-test. Abbreviations are the same as in Figure 3. Additional runs of this experiment are shown in Figure S3.

To determine whether there are differences in the local and systemic plant responses, glucosinolate levels were measured in rosette-stage Col-0 and Ler plants with M. persicae caged on individual leaves (Figure S4). In comparison to leaves with empty cages, aphid-infested Arabidopsis leaves showed significant increases not only in 4MI3M but also in the long-chain aliphatic glucosinolates (Figure 5). A significant induction of 4MI3M and 8-methylsulfinyloctylglucosinolate (8MSO) in Ler was observed in three independent caged-aphid experiments (Figure 5; Figure S5; Table S2). However, the induction of long-chain aliphatic glucosinolates in Col-0 was variable. Unlike the case of whole plants and detached leaves, the total glucosinolate content of leaves with caged M. persicae was either unchanged or even increased relative to controls in all experiments (Figure 5; Figure S5). No significant glucosinolate changes were observed in both the older and the younger uninfested leaves of plants with caged M. persicae (Figure S6), indicating that there is no systemic glucosinolate response to aphid feeding on Arabidopsis.

Figure 5.

 Glucosinolate content in leaves of Arabidopsis thaliana Col-0 (a) and Ler (b) with caged aphids, expressed as a percentage of the levels of each glucosinolate in uninfested control plants. Mean ± SD, n = 8; *P < 0.05, **P < 0.01 by a two-tailed Student's t-test. Abbreviations are as in Figure 3. Additional runs of this experiment are shown in Figure S5.

Salicylate-mediated signaling is a component of the Arabidopsis response to M. persicae feeding (Moran and Thompson, 2001; Moran et al., 2002), and has also been shown to specifically induce 4MI3M production (Kliebenstein et al., 2002a; Mikkelsen et al., 2004). Therefore, we determined which glucosinolate changes occur when aphids are caged on individual leaves of Ler NahG transgenic and Col-0 eds5-1, pad4-1, and npr1-1plants, all of which have defects in salicylate-mediated signaling during pathogen infection (Glazebrook, 2001). In two independent runs of this experiment, a consistent and significant aphid-mediated induction of 4MI3M was observed in all four mutant lines (Figure 6; Figure S7). The aphid-induced changes of some other glucosinolates varied between the two runs (Student's t-test, P < 0.05; Table S2). Salicylate hydrolase encoded by the NahG transgene converts salicylic acid to catechol, a biologically active compound (van Wees and Glazebrook, 2003). However, catechol added to artificial diets at 0.01 and 0.1 mm concentrations did not have a significant effect on aphid reproduction (Figure S8).

Figure 6.

 Glucosinolate content in leaves of Arabidopsis thaliana Col-0 pad4-1 (a), eds5-1 (b) and npr1-1 (c), and Ler NahG (d) mutants with caged aphids, expressed as a percentage of the levels of each glucosinolate in uninfested control plants. Mean ± SD, n = 8; *P < 0.05, **P < 0.01 by a two-tailed Student's t-test. Abbreviations are as in Figure 3. Data for 1MI3M in pad4-1 and eds5-1 mutants are not presented because of a poor separation from background noise in the chromatograms from these particular experiments. An additional run of this experiment is shown in Figure S7.

Univariate anova was used to identify the impact of fixed factors, experimental and aphid treatment, on glucosinolate level as a dependent variable (Table S3). The experiment has a significant effect on most of the glucosinolate levels, indicating a large variation in the levels of glucosinolate (nmol mg−1 dry weight) among independent experimental runs. On the other hand, few significant interactions between experimental and aphid treatment were observed. Although Col-0 detached leaf and Col-0 caged leaf raw glucosinolate data showed a significant interaction in seven out of 14 dependent variables (Table S3), the relative increase of 4MI3M, the glucosinolate of greatest interest, did not vary among three independent runs (Table S2).

Biosynthesis of 4MI3M may occur via the addition of a methoxy group to I3M (Figure 2). To test whether such a pathway is induced by aphid feeding, purified I3M was added to detached Col-0 cyp79B2 cyp79B3 leaves, which are almost completely devoid of indole glucosinolates because of a block in the production of indole-3-acetaldoxime (Figure 1). When leaf petioles were immersed in a solution containing 0.25 mm intact I3M, this glucosinolate accumulated to a concentration of 2 nmol mg−1 dry weight in the leaf as a whole (Figure 7). 4MI3M, which is undetectable in Col-0 cyp79B2 cyp79B3 mutants, also increased significantly above background levels. When M. persicae were allowed to feed on the detached leaves containing I3M, there was a significant (Student's t-test, P = 0.001) seven-fold increase in the 4MI3M content relative to control plants (Figure 7). Three additional independent runs of this experiment showed two- (P = 0.04), three- (P = 0.0003) and seven-fold (P = 0.002) increases in the level of 4MI3M relative to control leaves without aphids. As there is no de novo synthesis of 4MI3M in Col-0 cyp79B2 cyp79B3 mutants, this suggests the direct conversion of exogenously added I3M to 4MI3M. Previous reports showed that I3M is bewteen two- and 10-fold more abundant than 4MI3M in Col-0 leaves (Brown et al., 2003; Petersen et al., 2002). As would be expected if I3M is converted to 4MI3M, the I3M:4MI3M ratio decreased in response to aphid treatment in experiments with whole plants, detached leaves and with caged aphids (Table S4). The only exception were NahG transgenic plants, where a concomitant increase in I3M cause a net decrease in the 13M:4MI3M ratio. No measurable quantities of 1MI3M were produced by the Arabidopsis plants with exogenously added I3M, showing that the production of the two methoxyindole glucosinolates is differentially regulated. As in previous experiments with detached leaves (Figure 4), total glucosinolate content and leaf dry weight were significantly reduced by aphid feeding (Student's t-test P < 0.05; data not shown).

Figure 7.

 Exogenously added indol-3-ylmethyl glucosinolate (I3M) is converted to 4-methoxyindol-3-ylmethyl glucosinolate (4MI3M) in response to aphid feeding. The petioles of Arabidopsis thaliana Col-0 cyp79B2 cyp79B3 mutant leaves were inserted into tubes either with or without I3M. Levels of glucosinolates were measured in aphid-infested and control leaves after 2 days. Mean ± SD, n = 3 experiments. There is a significant increase in the production of 4MI3M in aphid-infested leaves relative to uninfested control leaves (Student's t-test, P = 0.001).

To determine which Arabidopsis glucosinolates are taken up from the phloem and activated during M. persicae feeding, we measured the total glucosinolate content of whole plants, phloem sap, aphid bodies and aphid honeydew. As M. persicae does not sequester glucosinolates, the measurement of glucosinolate content in whole aphids is a good representation of the glucosinolates that are passing through the mouth parts and gut. The abundance of indole glucosinolates as a percentage of total intact glucosinolates is reduced in aphid honeydew compared with aphid bodies, phloem sap and whole plants (Figure 8), indicating that indole glucosinolates are broken down to a greater extent than aliphatic glucosinolates. As glucosinolates are broken down within seconds upon contact with Arabidopsis myrosinase (Barth and Jander, 2006), the presence of intact aliphatic glucosinolates in the honeydew (Figure 8) suggests that M. persicae are able to either avoid or inactivate myrosinase when feeding from the phloem. Interestingly, the relative levels of 8MSO in aphid bodies and honeydew were greatly increased compared with that of the whole leaf and phloem exudates (Figure 8). Three independent runs of the experiment shown in Figure 8 produced quantitatively similar results.

Figure 8.

 Glucosinolate content of whole Arabidopsis plants, phloem sap, aphid bodies and aphid honeydew.

Both the specific induction of 4MI3M (Figures 4–6) and the breakdown pattern of glucosinolates during passage through M. persicae (Figure 8) suggested that indole glucosinolates play a unique role in plant defense against this phloem-feeding insect. Therefore, HPLC-purified indole glucosinolates (Figure S9) were added to cups with artificial diet (Figure S10) in order to determine whether they provide a defensive benefit, either with or without myrosinase. Both 4MI3M and 1MI3M affected M. persicae reproduction at significantly lower concentrations than I3M in the absence of myrosinase (Figure 9a–c), suggesting that a conversion of I3M to 4MI3M would provide a defensive benefit to crucifers. Whereas the addition of myrosinase to the I3M- and 1MI3M-containing diets significantly reduced M. persicae reproduction, this enzymatic cleavage did not further increase the deterrent properties of 4MI3M (Figure 9e–g). Aphid fecundity on diets to which myrosinase without a glucosinolate substrate was added did not differ significantly from controls (Figure S11). Aphid reproduction was also not measurably inhibited by up to 7 mm 2-propenyl glucosinolate (sinigrin), a representative aliphatic glucosinolate and the most abundant glucosinolate in the cabbage on which the aphids were reared prior to the start of the diet assay (Figure 9d). The addition of myrosinase to diets containing sinigrin reduced aphid reproduction (Figure 9h), although to a lesser extent than did the indole glucosinolates. Reduced aphid fecundity in these experiments was correlated with the reduced production of honeydew, as measured by ninhydrin staining of filter paper placed underneath the artificial diet (Figure S12). This shows that the addition of indole glucosinolates deters M. persicae feeding, rather than being merely a post-ingestive inhibitor of growth and reproduction.

Figure 9.

 Aphid fecundity on artificial diets containing glucosinolates either with or without myrosinase. Aphid numbers were normalized relative to reproduction on a control diet (sucrose + 20 amino acids), which is set to 100%. Each data point in the plots is the mean ± SD, n = 10, at that glucosinolate concentration. The IC50 values calculated based on the linear regression lines shown are indicated for each graph. The numbers in parentheses are confidence intervals at α = 0.95. IC50 values of sinigrin and sinigrin + myrosinase are not shown because the slopes of the respective regression lines are not significantly different from zero.

Discussion

There is a reduction in overall glucosinolate content and an increase in 4MI3M in response to M. persicae feeding on Arabidopsis and cabbage (Figures 3–6). Similarly, treatment with salicylate analogs (Kliebenstein et al., 2002a; Mikkelsen et al., 2004; J.H. Kim and G. Jander, unpublished data), as well as mutations (cpr1 and mpk4) that increase endogenous salicylate levels (Mikkelsen et al., 2003), increased the production of 4MI3M but not of other glucosinolates. This, together with the observation that some salicylate-induced genes are also induced by M. persicae feeding (Moran and Thompson, 2001; Moran et al., 2002), made it tempting to speculate that salicylate signaling mediates the induction of 4MI3M by aphids. However, the aphid-induced production of 4MI3M was unaffected in eds5-1, pad4-1, npr1-1 and NahG transgenic Arabidopsis plants (Figure 6). Therefore, if salicylate signaling is involved in the observed glucosinolate changes, this occurs by a pathway that is different from those associated with pathogen defense.

The NahG gene product, which converts salicylic acid to catechol, has an indirect negative effect on non-host resistance to Pseudomonas syringae in Arabidopsis (van Wees and Glazebrook, 2003). In contrast, the NahG transgene did not affect aphid reproduction on Arabidopsis (Pegadaraju et al., 2005) and tomato (Li et al., 2006). Aphid-induced changes in the Arabidopsis glucosinolate profile (Mewis et al., 2005) and I3M to 4MI3M conversion (Figure 6) were also unaffected in NahG transgenics. Catechol itself is not deleterious to aphids at physiologically relevant concentrations (Figure S8). Therefore, unlike in the case of P. syringae, the production of catechol per se does not appear to have unexpected effects on Arabidopsis–M. persicae interactions.

The M. persicae-induced glucosinolate changes in Col-0 that we observed differ from those reported previously. In one study (Mewis et al., 2005), the relatively large standard errors of the 4MI3M data may have precluded the detection of the aphid-induced increases that we observed. However, opposite to our results, there were small (approximately 10%) but significant increases in short-chain aliphatic glucosinolates in response to M. persicae feeding. In another report with somewhat different results (Mewis et al., 2006), the total aliphatic and the total indole glucosinolate content, as well as the transcription of five glucosinolate biosynthetic genes (Table S1), were increased after one week of M. persicae feeding. However, as the glucosinolate categories were not subdivided, it is not possible to determine which individual glucosinolate species made significant contributions to the total increase. Differences in the plant growth conditions, the length of aphid exposure, the particular isolate of ecotype Col-0 or the aphid strains used for the respective experiments could explain these dissimilar results. Another possible confounding factor is that the aphid strains used in these experiments might carry different plant-pathogenic viruses that influence glucosinolate production. Although our M. persicae culture was passaged on an artificial diet, which would break the virus transmission cycle, we cannot rule out later reinfection from plants raised in the greenhouse.

It is possible that the 4MI3M increase and the decrease of all other foliar glucosinolates during M. persicae feeding on Arabidopsis (Figure 3) is not a direct alteration of plant defense, but rather a symptom of premature senescence in the infested leaves. A somewhat similar overall glucosinolate decline and a slight increase in the level of 4MI3M as a percentage of the total indole glucosinolate content occurs during the normal Arabidopsis leaf senescence (Brown et al., 2003; Petersen et al., 2002). However, a counter argument to this scenario is that M. persicae-induced leaf senescence is compromised by pad4 mutations (Pegadaraju et al., 2005), whereas induction of 4MI3M is pad4-independent (Figure 6).

Unlike the situation where the M. persicae were moving freely either on the whole plant (Figure 3) or on detached leaves (Figure 4), glucosinolates other than 4MI3M, in particular 8MSO, also accumulated in single leaves with caged aphids (Figure 5). One possible explanation is that glucosinolates are transported to the site of aphid feeding. Several previous studies have demonstrated glucosinolate transport in Arabidopsis and other crucifers (Brudenell et al., 1999; Chen et al., 2001; Du and Halkier, 1998; Durner and Klessig, 1995; Gijzen et al., 1994; Lein, 1972; Macgrath and Mithen, 1993). Aphid bodies and honeydew have markedly higher 8MSO levels than either whole Arabidopsis plants or phloem sap (Figure 8), suggesting a relatively increased concentration at the site of aphid feeding. Except in the case of Col-0 npr1-1 and Ler NahG, the dry weight of leaves with caged M. persicae was not significantly reduced compared with control leaves (Figures 5 and 6), also indicating that metabolites are transported from other parts of the plant during aphid feeding. As the caged leaf represents only a small portion of the total plant mass, it is perhaps not surprising that we did not observe a significant decrease of long-chain aliphatic glucosinolates in systemic leaves of these plants (Figure S6).

Myzus persicae feeding induces the production of 4MI3M from I3M, which is taken up via the leaf petiole (Figure 7). Although it is likely that I3M serves as a direct precursor for 4MI3M in the absence of de novo indole glucosinolate biosynthesis in the Col-0 cyp79B2 cyp79B3 mutant, we cannot rule out the possibility that the methoxy group is added to either the desulphoglucosinolate or some other transported intermediate. The conclusive demonstration of an in vivo role for 4MI3M in plant defense will require identifying mutations affecting the as yet unknown enzymes that catalyze the conversion of I3M to 4MI3M. A targeted approach to studying candidate genes may result in the identification Arabidopsis genes encoding these enzymes. For instance, three putative O-methyltransferases (AT1G21120, AT1G21130 and AT1G21100) showed higher expression after both 2 and 3 days of aphid feeding (De Vos et al., 2005), and could encode enzymes catalyzing the final step in the biosynthesis of 4MI3M from I3M (Figure 2).

Arabidopsis myrosinase is produced in cells that are immediately adjacent to the phloem (Andreasson et al., 2001; Husebye et al., 2002; Thangstad et al., 2004). However, the mainly intercellular path taken by aphid stylets on the way to the phloem (Tjallingii and Hogen Esch, 1993) may allow M. persicae to ingest phloem glucosinolates without bringing them into contact with myrosinase. Aphid reproduction was not significantly affected by the absence of foliar myrosinase in Arabidopsis tgg1tgg2 mutants (Barth and Jander, 2006). Unlike aliphatic glucosinolates, indole glucosinolates still break down in crushed leaves of tgg1 tgg2 mutants, with a half-life of about 5 min (Barth and Jander, 2006). Similarly, the excretion of intact aliphatic glucosinolates and reduced levels of indole glucosinolates in aphid honeydew (Figure 8) supports the hypothesis that aphids are avoiding contact with myrosinase. The relative instability of indole glucosinolates may contribute to their breakdown during passage through the aphids. However, we cannot rule out a scenario whereby an as yet unknown phloem-localized enzyme specifically catalyzes indole glucosinolate, but not aliphatic glucosinolate breakdown during aphid feeding.

On artificial diets where I3M, 4MI3M and 1MI3M are broken down by myrosinase, aphid reproduction is affected similarly (Figure 9). However, both 4MI3M and 1MI3M are more deleterious to M. persicae reproduction than I3M in the absence of myrosinase. One possible explanation is that the methoxyindole glucosinolates are more easily broken down in the aphid gut than I3M, resulting in as yet unknown products that reduce fecundity. Another scenario is that the intact methoxyindole glucosinolates are stronger pre-ingestive feeding deterrents than I3M. As Col-0 rosette leaves prior to flowering have significantly more I3M than methoxyindole glucosinolates (Brown et al., 2003; Petersen et al., 2002; Table S4), it is likely that I3M also makes a contribution to aphid deterrence in vivo. Nevertheless, the conversion of I3M to 4MI3M decreases the I3M:4MI3M ratio (Table S4) and would be expected to have a net deterrent effect.

The results presented here represent a significant step towards understanding how the diversity of glucosinolates found in nature can provide a targeted defense against specific herbivores. Although glucosinolates are commonly grouped as ‘aliphatic’, ‘aromatic’ or ‘indolyl’ in published insect bioassays, the results presented here show that there can be significant differences in the insect-deterrent effects of individual glucosinolate species in these categories. Arabidopsis plants appear to mount the correct defense response against M. persicae by converting I3M to 4MI3M, which provides greater protection in the absence of myrosinase-mediated activation. As most cultivated Brassica species contain indole glucosinolates, it may be possible to increase their aphid resistance by selectively altering the basal or induced indole glucosinolate profile through either breeding or biotechnology.

Experimental procedures

Plants, insects and growth conditions

Col-0 and Ler wild-type plants were obtained from the Arabidopsis Biological Resource Center. The Col-0 cyp79B2 cyp79B3 mutant (Zhao et al., 2002) was kindly supplied by J.L. Celenza (Boston University, Boston, MA, USA). Col-0 atr1D seeds (Bender and Fink, 1998; Celenza et al., 2005) were provided by J. Bender (Johns Hopkins University, Baltimore, MD, USA). The eds5-1, pad4-1 and npr1-1 mutants, all in the Col-0 background, were obtained from J. Dewdney (Massachusetts General Hospital, Boston, MA, USA) and the Ler NahG transgenic line was from D. Klessig (Boyce Thompson Institute for Plant Research, Ithaca, NY, USA). Arabidopsis plants were grown in a chamber in standard nursery flats (approximately 20 cm × 40 cm) using Cornell Mix (Landry et al., 1995) with Osmocoat fertilizer (Scotts, http://www.scotts.com) at 23°C with an intensity of 180 μmol m−2 sec−1 photosynthetic photon flux density and a 16-h photoperiod at 60% relative humidity.

All experiments were conducted with a tobacco-adapted red lineage of M. persicae that was obtained from S. Gray (USDA Plant Soil and Nutrition Laboratory, Ithaca, NY, USA). Aphids were raised on cabbage (B. oleracea var. Wisconsin Golden Acre; Seedway, http://www.seedway.com/) with a 16-h day (150 μmol m−2 sec−1 at 24 ± 1°C) and an 8-h night (19 ± 1°C) at 50% relative humidity. Unless otherwise mentioned, all aphid experiments were performed under the same environmental conditions.

Glucosinolate induction by aphid feeding

For glucosinolate induction in whole plants, 30 aphids (mixed instars) were placed on a plant (14–16 days after sowing for Arabidopsis; at the two true-leaf stage for cabbage) for three days. For the detached-leaf and caged-aphid experiments, 20 aphids were placed on either a 6th (Ler) or a 7th (Col-0) leaf (harvested 18–20 days after sowing) for 3 days. For detached-leaf experiments, petioles of leaves were inserted in a microcentrifuge tube containing water. For caged-aphid experiments, clear plastic cups (30-ml Portion Cups; Comet Products, http://www.wna-inc.com/), with a bottom of fine-mesh gauze, were fitted around the petiole of each leaf (Figure S4). Control plants received cages without aphids. For examining glucosinolate changes in systemic leaves, a younger and an older un-caged leaf were collected after 3 days of aphid feeding. After the removal of aphids, plant tissues were frozen in liquid nitrogen and stored at −80°C prior to glucosinolate analysis.

Glucosinolate extraction and analysis

Glucosinolate-containing extracts of plant tissue were prepared as described by Barth and Jander (2006). Desulfoglucosinolates were separated using a Waters 2695 HPLC and detected using a Waters 2996 photodiode array detector and a Micromass Platform LC mass spectrometer (Waters, http://www.waters.com/), with minor modifications of a previously described protocol (Kim et al., 2004). For HPLC separation, the mobile phases were A, water, and B, 90% acetonitrile, at a flow rate of 1.2 ml min−1 at 23°C. Column linear gradients for samples were: 0–1 min, 98% A; 1–6 min 94% A; 6–8 min, 92% A; 8–16 min, 77% A; 16–20 min, 60% A; 20–25 min, 0% A; 25–27 min hold 0% A; 27–28 min, 98% A; 28–37 min, 98% A.

Isolation and purification of indole glucosinolates

Glucosinolates were isolated using a modification of a previously described method (Thies, 1988). Sixty 4-week-old Col-0 atr1D Arabidopsis mutant plants were macerated in 500 ml 80% methanol, followed by filtration through Whatman paper (Grade 1; Whatman, http://www.whatman.com/) and centrifugation at 500 g for 10 min. The supernatant was transferred to a 2 × 25 cm glass column filled with Sephadex A-25 resin (Amersham Biosciences, http://www1.amershambiosciences.com/) to a height of 5 cm, and the column was washed out with the following: 80% methanol (20 ml), water (10 ml), 3:2:5 formic acid:isopropanol:water (20 ml) and water (50 ml). Most aliphatic glucosinolates were eluted with 25 ml of 0.5 m K2SO4 in 5% isopropanol. Indole glucosinolates were eluted by the addition of a further 100 ml of 0.5 m K2SO4 solution to the columns. The eluent from the columns dripped directly into 100 ml of 100% ethanol and the collected solution was dried using a Rotavapor (Büchi Flawil, http://www.buchi.com/). The dried residue was washed with 5 ml of 100% methanol to extract glucosinolates. After centrifugation at 17 000 g for 10 min, the supernatant was evaporated using a Rotavapor and the residue was resuspended in water prior to storage at –80°C.

A semi-preparative LiChrospher RP-18, 250 × 10mm, 5 μm end-capped column (Varian, http://www.varianinc.com/) was used to purify intact indole glucosinolates (I3M, 1MI3M and 4MI3M). Mobile phases were A, water with 0.01% trifluoroacetic acid, and B, 90% acetonitrile with 0.01% trifluoroacetic acid, at a flow rate of 3 ml min−1 at room temperature. Column linear gradients for samples were: 0 min, 85% A; 0–5 min 60% A; 5–8 min, 100% A; 8–9 min, 85% A; 9–15 min, 60% A. Fractions representing each individual indole glucosinolate (Figure S9a) were collected. After lyophilization, the residue was resuspended in water and purified again using the same HPLC method. The purified intact glucosinolates from each fraction were confirmed by HPLC as desulfoglucosinolates (Figure S9b).

Conversion of I3M into 4MI3M

Petioles of a detached Col-0 cyp79B2 cyp79B3 mutant leaves (either the 5th or 6th leaves from 20–27-day-old plants) were inserted in 300-μl centrifuge tubes containing 0.25 mm I3M in water. Control leaves were treated with only water. Twenty mixed-instar aphids were placed on each individual leaf for 2 days. Petioles and other portions of the leaf that were immersed in the I3M solution were not used for the glucosinolate analysis. Three leaves receiving the same treatment were combined as one replicate for glucosinolate analysis.

Artificial diet assays

The artificial diet used in a fecundity assay consisted of sucrose (440 mm) and 20 amino acids (Ala, 10 mm; Arg, 16 mm; Asn, 20 mm; Asp, 10 mm; Cys, 3.3 mm; Glu, 10 mm; Gln, 10 mm; Gly, 10 mm; His, 10 mm, Ile; 6 mm; Leu, 6 mm; Lys, 10 mm; Met, 5 mm; Phe, 3 mm; Pro, 7 mm; Ser, 10 mm; Thr, 12 mm; Trp, 4 mm; Tyr, 2 mm; Val, 7 mm). The sucrose + 20 amino acids solution, either with or without 1 mg ml−1 myrosinase (thioglucosidase; Sigma, http://www.sigmaaldrich.com/), was added to either lyophilized purified indole glucosinolates or sinigrin (Sigma). Four wingless adults were placed in a cup (30-ml Portion Cups; Comet Products), which was covered with a Parafilm sachet containing 50 μl of the liquid diet (Figure S10). The aphid nymphs were counted after 3 days. Aphid fecundity on either indole glucosinolate- or sinigrin-containing diets, regardless of any myrosinase treatment, was normalized relative to reproduction on the control diet. Aphid numbers in the control diet (sucrose + 20 amino acids) are given as 100% (Figure 9). Ten replicates were performed at a given glucosinolate concentration, and data from 16 independent aphid reproduction experiments were normalized and combined for the results shown in Figure 9.

Glucosinolate extraction from phloem sap, aphids and honeydew

Phloem sap was collected from rosette-stage Col-0 plants (15–17 days after sowing). Roots were cut from the rosettes, leaving 1 mm of the shoot tissue, and put in 100 μl of 15 mm EDTA solution (pH 7.7) for 4 h in the dark with high humidity (King and Zeevaart, 1974). Twenty samples were pooled for each replicate of the experiment.

For the measurement of glucosinolates in aphid bodies, aphids were collected from Col-0 plants and transferred into a 2-ml centrifuge tube containing three 3-mm steel balls (Abbott Ball Company, http://www.abbottball.com/). A 1-ml volume of ground aphids was mixed with 500 μl of 100% methanol by vigorous shaking. After centrifugation at 17 000 g for 10 min, the supernatant was transferred into another centrifuge tube, ground aphid tissues were washed with 500 μl water, followed by another centrifugation at 17 000 g for 10 min. The aqueous supernatant was pooled with the methanol extract for glucosinolate analysis.

For aphid honeydew collection, 30 aphids were placed on a Col-0 plant for 3 days, and honeydew was collected on aluminum foil beneath a plant. After 3 days, the aluminum foil was washed with 80% methanol to extract glucosinolates.

Ninhydrin staining of aphid honeydew

To compare the levels of honeydew excretion from aphids feeding on different diets, honeydew was collected on a Whatman filter paper (Grade 1; Whatman) placed in the bottom of a cage used in an artificial diet assay (Cooper and Goggin, 2005). The filter paper was immersed in 1% ninhydrin in ethanol and heated in an oven for 15 min at 55°C prior to scanning to record the spots of aphid honeydew.

Data analysis

IC50 (the concentration required for 50% inhibition of aphid fecundity) was calculated based on the relative aphid fecundity at each concentration using a simple linear regression model (inverse prediction) using jmp v.6 (SAS Institute, http://www.sas.com/). anova and Student's t-tests were performed using Spss v.14 (SPSS Inc., http://www.SPSs.com/). To better represent the change of glucosinolate levels caused by aphid treatment, glucosinolate levels of aphid-infested plants were calculated as a percentage of the levels found in control plants without aphid treatment. Significant differences in glucosinolate levels either with or without aphid treatment was tested using Student's t-tests. Either one-way anova (three independent runs) or Student's t-tests (two independent runs) were used to identify the variation in the glucosinolate content between runs.

Acknowledgements

We thank J. Celenza, J. Bender, J. Dewdney and D. Klessig for providing mutant seeds, S. Gray for the aphid lineage and S. Despa for consultation on statistics. M. Reichelt and J. Gershenzon provided valuable advice and training for glucosinolate purification. This research was funded by the Boyce Thompson Institute, Atlantic Philanthropies, and the Triad Foundation.

Ancillary