1The presence, concentration and composition of plant secondary compounds, which confer plant resistance to herbivores and pathogens, vary greatly both within and among individuals. Optimal defence theory predicts that plant tissues most closely tied to plant fitness should be most defended at the constitutive level, and that more expendable tissues should be inducible with damage.
2We examined variation in glucosinolate content between leaves and petals, as well as among four petal colour morphs of wild radish, Raphanus sativus . We predicted greater levels of constitutive defences in petals, and greater inducibility of glucosinolates in leaves, based on previous studies that could relate leaves and petals to plant fitness.
3While, overall, optimal defence predictions were supported, individual glucosinolates differed in both their degree of inducibility as well as in their distribution between tissue types.
4Petal colour variants differed in their induced responses to damage, but not in their constitutive levels of compounds. Yellow and white morphs, which are preferred by the dominant bee pollinators as well as by herbivores, were generally less inducible than anthocynanin-containing pink and bronze petal morphs.
5Pleiotropic effects between petal colour and defence loci, or tight linkage between these loci, may allow pollinators to maintain variation in secondary chemistry, as well as allow herbivores to influence colour morph fitness and prevalence.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Along with variation in secondary chemistry among tissue types within plants, secondary chemistry may also vary among individual plants. A variety of hypotheses have been proposed to explain the maintenance of variation in secondary chemistry among plants in the same population. Costs of defence are one mechanism through which variation in defence levels may be present in non-equilibrial populations. In one scenario, well-defended plants have higher fitness in years with high herbivore damage, whereas less-defended plants have higher fitness in low-herbivory years, due to reduced costs, as recently reviewed in Bergelson & Purrington (1996), Koricheva (2002) and Strauss et al. (2002). Variation in defence may be similarly maintained in a non-equilibrial state when there are ecological costs of defence, e.g. when different defences work against different herbivores, and herbivore composition varies from year to year.
Here, we consider another hypothesis to explain variation in defence among individuals of Raphanus sativus, wild radish. Variation in defence may be a result of pleiotropic effects of genes at other loci, or of selection on genes tightly linked to defence. For example, herbivores have been shown to discriminate among petal colour morphs in some species (Simms & Bucher 1996; Irwin et al. 2003). Beetle larvae performed better on leaves of pink-flowering morphs (anthocyanin-producing colour morphs) than on leaves of white-flowering morphs of morning glory, Ipomaea purpurea (Simms & Bucher 1996). In addition, some floral and pollen herbivores discriminate among flower-colour morphs, including thrips (Vernon & Gillespie 1990; Gaum et al. 1994; Chyzik et al. 1995) and pollen-feeding beetles (Giamoustaris & Mithen 1996). For R. sativus, many herbivores exhibited better performance on leaves of anthocyanin recessive (white and yellow petal colour) compared with anthocyanin-dominant (pink and bronze petal colour) (Irwin et al. 2003). Armbruster (2002) showed that petal and leaf anthocyanin expression were linked in the Acer and Dalechampia clades; however, in a study of R. sativus, there was no relationship between leaf anthocynanin content and herbivore performance, and only petal anthocyanins, or traits linked closely to petal colour, influenced herbivores (S. Y. Strauss, unpublished data).
One explanation for the links between defensive compounds and petal colour may be that genes controlling flower colour directly influence plant resistance to herbivores, if pleiotropic effects exist between the synthesis of floral pigments and defensive plant compounds (Simms & Bucher 1996; Fineblum & Rausher 1997). Alternatively, if petal colour genes and defence genes are tightly linked, then selection on one trait may cause correlated changes in values of the other trait. Both petal colour and glucosinolates are known to be heritable traits in R. sativus (Panetsos 1964; Carlson et al. 1985; Ishii et al. 1989; Schuetze et al. 1999). In addition, Hemm et al. (2003), using Arabidopsis mutants, showed that altering alkylglucosinolate biosynthesis simultaneously affected phenylpropanoid metabolism, from which anthocynanin pigments are derived. Thus, there is evidence for pleiotropic effects of genes affecting both the glucosinolate and anthocyanin pathways in a related mustard.
Understanding differences in defensive chemistry among colour morphs of wild radish may shed light on the selective forces maintaining both the petal colour polymorphism and variation in defence in this species. Variation in defence may be maintained if defence is somehow linked (sensu lato) to petal colour, and if petal colour is under selection from agents other than herbivores (see Irwin et al. 2003). In Raphanus sativus, dominant bee pollinators prefer yellow petal morphs (anthocynanin recessive) to pink and bronze anthocyanin dominant morphs (Stanton 1987; Stanton et al. 1989), and therefore pollinators may exert selection on defensive chemistry, if petal colour morphs differ in defensive chemistry. A first step to understanding whether conflicting selection could maintain variation among genotypes of wild radish is to determine whether flower colour is associated with consistent differences in secondary chemical variation. While this study focuses on variation in defensive chemistry within and among plants of R. sativus, many of the predictions will also apply to understanding variation in traits that influence important mutualists and antagonists simultaneously for any species.
predictions of optimal defence theory and the value of tissues inr. sativus
Petals are often tightly linked to plant fitness, especially in obligately outcrossing species like R. sativus, because of their important role in attracting pollinators (reviewed in Proctor et al. 1996). The link between petals and fitness in R. sativus is substantive. Young & Stanton (1990) and Stanton & Preston (1988) found that increased petal size was associated with greater components of male fitness (pollen removal). In addition, female fitness was positively associated with petal size in field experiments on a close relative of R. sativus, Raphanus raphanistrum (Conner et al. 1996), whose flowers are virtually indistinguishable from those of R. sativus in morphology. Thus, petal size may be related to fitness through both male and female fitness components in R. sativus. In R. raphanistrum, plants with larger petals often receive more pollinator visits (Strauss et al. 1996; Lehtilä & Strauss 1997). In addition, damage to petals by florivores has been associated with decreased pollinator attraction in a variety of plant species (Karban & Strauss 1993; Krupnick & Weis 1998; Krupnick et al. 1999; Mothershead & Marquis 2000; Adler et al. 2001). Moreover, symmetrical flowers are more attractive to pollinators than asymmetrical flowers (Moller 1996), so damage directly to petals could also reduce attractiveness to pollinators through asymmetry. R. sativus experiences extensive petal herbivory in some locations and years; we have observed woolly bear caterpillars at Bodega Bay feeding extensively on both leaf and petal tissue in outbreak years (S. Y. Strauss, personal observations). Thus, small changes in petal area or shape may have large impacts on pollinators and plant fitness. We therefore predict that petals should be constitutively well defended, especially for self-incompatible annuals such as R. sativus that rely on pollinators for plant reproduction.
Leaves are also important to plant fitness, and high levels of damage to leaves can reduce fitness in Raphanus spp. through both direct and indirect pathways (e.g. Mauricio & Bowers 1990; Strauss et al. 1996; Lehtilä & Strauss 1997; Agrawal et al. 1999). However, Raphanus spp. are relatively tolerant to herbivory and suffer little to no fitness costs with small amounts of leaf damage (Mauricio & Bowers 1990; Lehtilä & Strauss 1999; Strauss et al. 2001). When 25% leaf area was removed from each of the first four leaves of R. sativus by Pieris rapae larvae, reproduction and growth of these damaged plants was indistinguishable from that of controls (Mauricio & Bowers 1990). Damage levels in the field vary among years, and range from a mean of 5% to 20% overall damage in adult plants (Strauss and Irwin, unpublished data). Thus, we expect leaves to exhibit inducible defences. Another hypothesis to explain the greater inducibility of leaves over petals hinges on the costs of defence and the timing of defence expression. Costs of defence incurred early in the ontogeny of the plant, i.e. if leaves are constitutively defended or induced early in the plant lifetime, may have long-lasting impacts on plant fitness through diminished resource acquisition; in contrast, for tissues like petals, which are created and defended later in the lifetime of the plant, costs may have a lesser overall impact on plant resources (P. Klinkhamer, personal communication). This argument is also consistent with the prediction that leaf defences should be inducible with damage. Moreover, we expect that plants should invest more defences in petal tissue than in leaf tissue because low levels of damage to petal tissue may have greater impacts on plant fitness than low levels of leaf damage in this annual plant.
the study system
Raphanus sativus L. (Brassicaceae) is a naturalized, herbaceous annual, which is common along roadsides and disturbed areas in valley and coastal areas of California, USA. Seeds germinate early in the rainy season (October/November) with plants blooming in March for approximately 3–4 months. In California, R. sativus individuals possess one of four different petal colours: yellow, white, pink or bronze. Petal colour is determined by two independently assorting loci, each with two alleles controlling the expression of carotenoids and anthocyanins ( Panetsos 1964 ) Carotenoid pigments produce yellow petals with yellow (presence of carotenoid) recessive to white (absence of carotenoid). Anthocyanin pigments produce pink petals with white (absence of anthocyanin) recessive to pink (presence of anthocyanin). Bronze-flowered plants express both anthocyanin and carotenoids and thus have at least one dominant allele at the anthocyanin locus and only recessive alleles at the carotenoid locus.
All plants were glasshouse-grown progeny whose grandparents were collected as seed from a naturalized R. sativus population at Bodega Bay, California. After field collection, seeds were grown and all pollen donors were crossed with yellow (double recessive) mothers. We crossed all plants into a yellow background to try to homogenize plants for traits other than flower colour. Thus, non-yellow plants were heterozygous (at least one locus) for flower colour because they were the result of a mating between a yellow parent and another pigmented parent. These heterozygous families produced progeny with multiple flower colours (see below).
Experimental plants were grown in the glasshouse in individual 10 cm square pots using University of California glasshouse soil mix. Plants were watered using a subirrigation system ad libitum and fertilized at the two-leaf stage with 2 g of Osmocote Plus 15-11-13 slow release fertilizer (Scott’s, Marysville, Ohio, USA).
At the four-leaf stage, plants were randomly assigned to one of two treatments: 50% of all leaves, except the fifth and eighth true leaves, consumed by caged Pieris rapae larvae, or unmanipulated controls. Pieris rapae are naturalized specialist herbivores of R. sativus and are a dominant herbivore in many CA populations of R. sativus. In the leaf removal treatment, we caged third to fifth instar larvae in clip cages. Cages were placed along the mid-vein of a leaf, and caterpillars fed on the leaf tissue in the cages. We moved the cages along the mid-vein until one-half of the leaf was consumed. This general pattern of damage mimics natural damage by P. rapae larvae in the field (S. Y. Strauss, personal observation). On the leaves of unmanipulated control plants, we placed clip cages with no larvae to control for clip-cage effects. As plants initiated flowering, the fifth (i.e. undamaged) true leaf was removed with a razor blade, weighed, and immediately microwaved for approx. 30–45 s to denature endogenous myrosinases. Samples were then dried for 48 h at 60 °C and stored at 0 °C until further chemical analysis. Larvae damaged plants over the course of c. 3 weeks, and the damage treatment was completed by the time plants started flowering.
To sample petal tissue for glucosinolate analysis, we removed the petals from at least 25 flowers per plant. Petals were removed from the base of flowers using fine-point forceps and care was taken to ensure that petal samples did not contain calyces or pollen. We only sampled petals from flowers that were 1–2 days old, and petal samples were collected over several dates. We had to combine the petals across multiple flowers on the same plant to obtain enough petal tissue for chemical analysis. Samples within plants were combined, weighed and processed, as described above. Because we required large numbers of flowers of an appropriate stage from a single individual to accumulate sufficient biomass for petal analysis, not all plants could be used for petal analyses. We used a total of 21 maternal families in the experiment. Sample sizes for chemical analysis ranged from one to seven samples per tissue type per family. Each sample came from a single plant. A total of 139 tissue samples were analysed; 68 petal samples and 71 leaf samples.
As parents of families were heterozygous, progeny from any single family often expressed multiple flower colours. For both flower and leaf samples, we maximized the use of families with diverse progeny. Blocking on family allows us to detect differences among flower colour types while controlling for genetic background. Of the 68 plants that provided petal samples, 14 were bronze, 14 pink, 20 white and 23 were yellow, from a total of 20 families. Because we needed copious petal tissue of specified stages, we could not use all the progeny each maternal plant produced. We collected petal samples of a single colour from four families, of two different coloured progeny from nine families, and of three different colours from seven families. Families often had more diverse progeny than the ones we sampled, but sufficient petal tissue may not have been available for progeny of all colours.
Because we could readily collect leaf tissue, we sampled more plants per family in order to take advantage of the diversity of progeny produced by families, but sampled only 13 families in total. These were a subset of the same families used for petal colour. Of 71 plants, 16 were bronze, 11 pink, 20 white and 21 yellow. Leaf samples were collected from two families represented by two colours in progeny, from four families represented by three colours, and from seven families represented by four colours.
To quantify glucosinolate content, we followed the basic sephadex/sulphatase glucosinolate extraction and purification protocols described in Hogge et al. (1988). Samples were placed into deep-well microtiter tubes. We added four 2.3-mm ball bearings, and the samples were ground into a fine powder in a paint shaker by high-speed agitation. To extract glucosinolates, we added 400 µL of methanol, 10 µL of 0.3 m lead acetate, and 120 µL of water. The samples were mixed for 1 min and then allowed to incubate for 60 min at 180 g on a rotary shaker. The tissue and protein were pelleted by centrifugation, and the supernatant was used for anion-exchange chromatography.
We loaded 96-well filter plates from Millipore (model MAHVN4550) with 45 µL of DEAE Sephadex A-25. We then added 300 µL of water to each column and allowed the mixture to equilibrate for 2–4 h. We removed the water with 2–4 s of vacuum and then added 150 µL of the supernatant to the 96-well columns. The liquid was removed by 2–4 s of vacuum, and this step was repeated once to bring the total volume of plant extract to 300 µL. The columns were washed four times with 150 µL of 67% methanol, three times with 150 µL of water, and three times with 150 µL of 1 m sodium acetate. To desulphate the glucosinolates on the columns, we added 10 µL of water and 10 µL of sulphatase solution to each column, and the plates were incubated overnight at room temperature (Hogge et al. 1988). To elute the desulphoglucosinolates, the DEAE Sephadex was washed twice with 100 µL of 60% methanol and twice with 100 µL of water. We ran 40 µL of the glucosinolate extract on a Hewlett-Packard 1100 series HPLC with a Hewlett-Packard Lichrocart 250–4 RP18e 5-µm column. Glucosinolates were detected at 229 nm and separated and identified using the following programmes with aqueous acetonitrile: (i) a 6-min gradient from 1.5 to 5.0% acetonitrile; (ii) a 2-min gradient from 5 to 7% acetonitrile; (iii) a 7-min gradient from 7 to 25% acetonitrile; (iv) a 2-min gradient from 25 to 92% acetonitrile; (v) 6 min at 92% acetonitrile; (vi) a 1-min gradient from 92 to 1.5% acetonitrile; and (vii) a final 5 min at 1.5% acetonitrile.
Total glucosinolate concentration was estimated by adding the concentration of all compounds, after conversion to SI units of µg/mg leaf tissue. Conversion to µg/mg from milli-absorption units was not possible for the two unknown compounds, but these comprised only 0.7% of the total investment in glucosinolates prior to conversion to SI units; therefore, these compounds were not included in our analysis of total glucosinolates. Concentrations were log-transformed to meet assumptions of normality.
To investigate the differences between tissue types, the relationship between colour morph and glucosinolate content, and the effects of induction via herbivore feeding on glucosinolates, we used maximum likelihood estimation (type III; PROC MIXED; SAS V.8) with colour morph (bronze, pink, white, yellow), damage treatment (50% removal on all leaves but 5th and 8th/no damage), and plant tissue (petals/leaves) as main effects, and plant family, family × damage and family × flower colour as random effects. Satterthwaite's approximation was used to account for unequal sample sizes of colour morphs and families. To test the significance of random effects in PROC MIXED, the best approach is to run the model both with and without the random factor(s) included in the model and then to use the likelihood ratio statistic (Littel et al. 1996). This statistic is computed by taking difference between the REML log-likelihood of the model containing the random effect and the log-likelihood of the model without the random effect. The critical value for this difference is half the probability of a greater chi-squared distribution from a chi-squared distribution with one degree of freedom (Littell et al. 1996); i.e. the difference in REML with and without the random factor in the model must exceed 2.71 at alpha = 0.05. For our data on total glucosinolate content, the difference between REMLs of models with the family × flower colour and the family × damage interactions compared with the model with just family as a random factor was less than 1.5 and thus neither interaction was significant. Family main effects were, however, significant (REML difference between models with and without random family effect = 4.6). Although random family effects are included in all the models reported, F-statistics in the tables are only reported for fixed effects and random effects will not be included in tables.
We also wanted to explore tissue-specific differences in the expression of individual glucosinolates; unfortunately, multivariate, maximum likelihood methods to explore the overall changes in compounds are not available in PROC MIXED. Instead, we examined the two most common glucosinolates: indol-3-ylmethyl glucosinolate (I3MTRP) and 4-methylthio-but-3-enyl (MTBUT). Together, these comprised 71% of the total glucosinolates produced by plants. In addition, I3MTRP is an indole glucosinolate, a class of compounds with known defensive properties against insects and non-ruminant mammals (e.g. McDanell et al. 1988, 1989), and thus represents a good choice for examination a priori.
The following glucosinolate compounds were identified as present in leaves and petals: 4-methylsulphinylbutyl (MSO), 4-methylsulphinyl but-3-enyl (MSOBUT), 4-methylthiobutyl (MT), 4-methylthio-but-3-enyl (MTBUT) and indol-3-ylmethyl (I3MTRP) glucosinolate, plus two unknowns. All compounds were found in both leaves and petals of all colour morphs.
total glucosinolate production
Constitutively, colour morphs did not differ in total glucosinolate content, and petal tissue contained about 20% higher overall levels of glucosinolates than did leaves (Table 1, Fig. 1a). When both damaged and undamaged plants were included in the model, there was a highly significant three-way interaction among flower colour, damage and tissue type (Table 2a, Fig. 2a). Glucosinolates were generally not induced in petals, except for in pink morphs, but were highly inducible in leaves (Figs 1a and 2a). Damage tended to increase glucosinolate content by 28% compared with undamaged plants (back-transformed LS means, Table 2a), although this trend was only marginally significant (P = 0.07). Overall, pink-flowered plants tended to show the greatest post-damage induction, and yellow morphs the least, in fact, a 5% decrease (Fig. 2a). Overall, there was a marginally significant damage ¥ flower colour interaction (Table 1a).
Table 1. Results from PROC MIXED analysis of effects of tissue type and petal colour morph on the constitutive concentration of glucosinolates (ln-transformed) in undamaged plants. Family and Family × flower colour were included in the model as random factors, but those effects are not presented below (see Materials and methods). Convergence criteria were met. Satterthwaite's approximation to estimate degrees of freedom was used because of unbalanced representation of flower colour among families and treatments
(a) Total glucosinolates
Tissue × flower colour
Tissue × flower colour
Tissue × flower colour
Table 2. Results from PROC MIXED of effects of tissue type, petal colour morph and damage on the concentration of glucosinolates (ln-transformed). Family, Family × damage and Family × flower colour were included as random factors in the model; only family main effects were significant and are presented in the Materials and methods. Convergence criteria were met. Satterthwaite's approximation to estimate degrees of freedom was used because of unbalanced representation of flower colour among families and treatments
(a) Total glucosinolates
Damage × flower colour
Tissue × flower colour
Tissue × damage × flower
Damage × flower colour
Tissue × flower colour
Tissue × damage × flower
Damage × flower colour
Tissue × flower colour
Tissue × damage × flower
mtbut glucosinlate production
Patterns for MTBUT were generally similar to those of total glucosinolate content (Table 2b, Fig. 1b) because MTBUT was the most abundant glucosinolate. Constitutively, petals had 40% greater levels of MTBUT than did leaves, there were no differences in constitutive levels among flower colour morphs, nor was the interaction between flower colour and tissue type significant. When both damaged and undamaged plants were included in the model, only the three-way interaction among tissue, damage and flower colour was significant (Table 2b). For all but the pink morphs, damage generally caused a decrease in the amount of MTBUT in petals, but an increase in the amount of MTBUT in leaves; decreases were particularly pronounced in yellow and bronze petal morphs (Fig. 2b).
indole glucosinolate production
Indole glucosinolates behaved very differently from MTBUT. Petals, overall, had 48%lower constitutive levels of indole glucosinolates when compared with leaves (undamaged plants; Table 1c, Figs 1c and 3). There was also a marginally significant interaction between tissue type and flower colour; constitutively, petals of pink flowers had the lowest indole glucosinolate content; in contrast, leaves of the pink morph had the greatest concentration of this glucosinolate.
Leaf indole glucosinolates were highly inducible, with pink and bronze morphs inducing more indole glucosinolates than white and yellow morphs in leaves (Fig. 3, see also Irwin et al. 2003). Petals induced indole glucosinolates less than leaves, and in bronze morphs there was no induction in petals (Fig. 3). Divergent behaviours among tissue types and colour morphs in induction resulted in the highly significant three-way interaction among damage, tissue type and flower colour (Fig. 1c, Table 2c). There was also a trend for damage to increase indole glucosinolate levels overall (P = 0.07).
Optimal defence theory predicts that tissues linked to reproduction should be more highly defended than leaf tissue because of their closer ties to plant fitness (McKey 1979). We found that, constitutively (i.e. in the undamaged state), total glucosinolate concentrations in petals were indeed generally greater than in leaves; this effect was due primarily to the concentration of the single, most abundant glucosinolate, MTBUT; however, petals had lower overall constitutive and induced levels of indole glucosinolates. Neither the leaves nor the petals from which we took our measurements were damaged, so changes in glucosinolate content reflected systemic responses to damage. Our results are in general agreement with other investigations of overall concentrations of defensive chemicals in leaf and reproductive tissues. For plants in the diverse families Solanaceae, Apiaceae, Asteraceae and Boraginaceae, floral parts all had greater levels of secondary compounds than did leaves (Nitao & Zangerl 1987; Van Dam et al. 1996; Zangerl & Rutledge 1996; Ohnmeiss & Baldwin 2000). Overall, these results suggest that patterns of total glucosinolate expression match predictions of optimal defence theory; however, the behaviour of individual glucosinolates varies, and the key will be to understand the function of individual compounds in relation to plant fitness.
Another prediction of optimal defence theory is that less valuable tissue, i.e. leaf tissue, should be more inducible with damage. This prediction was also generally upheld. Damage to leaf tissue increased both indole and MTBUT glucosinolates in leaf tissue much more than it did in petal tissues. However, an alternative explanation is that leaf tissue samples were collected much closer in time to the damage event than were petal samples, which were necessarily collected over a 2–3-week period in order to obtain sufficient biomass for chemical analysis.
Another non-adaptive hypothesis to explain responses of chemicals in floral tissues is that they change as a by-product of leaf induction (Adler 2000). Induction in R. sativus petals was both colour morph- and compound-specific. For MTBUT, in all but the pink morphs, petal concentrations decreased or remained the same while leaf concentrations increased with damage; this result suggests independent control of petal and leaf concentrations of MTBUT. In contrast, both petals and leaves showed induction of the indole glucosinolates, with petals inducing much lower concentrations than leaves. In this case, we cannot rule out non-adaptive induction in petal tissues of indole glucosinolates. The herbivores damaging leaves and petals may differ, and induced responses may vary with herbivore. In our study, the P. rapae larvae we used to damage plants typically damage only leaves and fruits; in contrast, woolly bears and mollusk herbivores can damage both petals and leaf tissues of R. sativus (S. Y. Strauss, personal observation). How plant tissues respond to different herbivores may also explain differential tissue-specific induction. In general, very few studies have examined induction in petal tissue. This study, and one showing nicotine induction in Nicotiana attenuata corollas (Euler & Baldwin 1996), show that there can be induction in petal tissue in response to leaf damage.
Another important aspect of our results is that colour morphs differed significantly from one another in the degree to which indole and MTBUT glucosinolates were induced after damage, although there were no differences in constitutive levels of either compound among colour morphs. This result is consistent with some of our previous work (Irwin et al. 2003), in which we found that a range of herbivores (Agriolimax reticulatus slugs, Pieris rapae butterflies, and Brevicoryne brassicae aphids) exhibited no preference among colour morphs when presented with undamaged plants in the rosette stage. However, slugs and aphids exhibited increased performance on yellow and white colour morphs (those with lower levels of induction) in no choice performance trials. P. rapae larvae, which are crucifer specialists, however, performed best on pink morphs (bronze was not included in the study), which had higher induced glucosinolate content after damage. This result is consistent with previous findings that P. rapae specialists have either better performance on plants containing glucosinolates or are indifferent to variation in glucosinolate concentrations (e.g. Blau et al. 1978; Agrawal & Sherriffs 2001). However, we also did trials with a generalist herbivore, Spodoptera exigua, which should have been negatively affected by induction of glucosinolate compounds (Blau et al. 1978), but surprisingly, it also performed better on leaves of pink morphs. These results suggest that colour morphs may differ nutritionally as well as defensively.
Differences among petal morphs in leaf defensive chemistry imply that herbivores could act as selective agents on this flower colour polymorphism (see Irwin et al. 2003). As a corollary, strong and consistent preferences by pollinators for particular colour morphs, as documented by Stanton (1987) and by Irwin and Strauss for this population (unpublished data), suggest that pollinators could also act to maintain variation in defensive chemistry. In this case, pollinators strongly preferred less-defended, yellow petal morphs. These yellow petal variants exhibited the smallest degree of induction of glucosinolates in our study, and therefore may be selected against by herbivores. Both petal colour for this species and glucosinolates are known to be heritable traits (Panetsos 1964; Carlson et al. 1985; Ishii et al. 1989; Schuetze et al. 1999). Thus, conflicting selection pressures exerted by pollinators and herbivores, coupled with pleiotropy or tight linkage between colour and defence loci, could maintain variation in both traits.
We thank Dylan Burge for help with damage treatments and tissue collection. Lynn Adler, Rick Lankau, Peter Klinkhamer and anonymous reviewers provided helpful comments on the manuscript. The work was supported by U.S. National Science Foundation grant DEB 98–07083 to SYS. Partial support was also provided by the Department of Genetics and Evolution, Max-Planck Institute of Chemical Ecology, Max-Planck Gesellschaft, as well as by travel funds provided by the Bodega Marine Laboratory, UC Davis.