Testing the potential for conflicting selection on floral chemical traits by pollinators and herbivores: predictions and case study

Authors


*Correspondence author. E-mail: ak357@cornell.edu

Summary

1. There are myriad ways in which pollinators and herbivores can interact via the evolutionary and behavioural responses of their host plants.

2. Given that both herbivores and pollinators consume and are dependent upon plant-derived nutrients and secondary metabolites, and utilize plant signals, plant chemistry should be one of the major factors mediating these interactions.

3. Here we build upon a conceptual framework for understanding plant-mediated interactions of pollinators and herbivores. We focus on plant chemistry, in particular plant volatiles and aim to unify hypotheses for plant defence and pollination. We make predictions for the evolutionary outcomes of these interactions by hypothesizing that conflicting selection pressures from herbivores and pollinators arise from the constraints imposed by plant chemistry.

4. We further hypothesize that plants could avoid conflicts between pollinator attraction and herbivore defence through tissue-specific regulation of pollinator reward chemistry, as well as herbivore-induced changes in flower chemistry and morphology.

5. Finally, we test aspects of our predictions in a case study using a wild tomato species, Solanum peruvianum, to illustrate the diversity of tissue-specific and herbivore-induced differences in plant chemistry that could influence herbivore and pollinator behaviour, and plant fitness.

Introduction

Plants have evolved many ways of coping with their insect attackers, including physical or chemical defences that directly reduce herbivore performance and survival, indirect defences that actively attract or facilitate the action of predators and parasitoids, as well as tolerance of herbivory (Duffey & Stout 1996; Dicke & Van Loon 2000; Kessler & Baldwin 2002).

Ehrlich & Raven (1964) first deduced that insect herbivory had selected for chemically defended plants. In turn, high plant secondary metabolite production would favour insect herbivores that are resistant to plant defences. All hypotheses that were derived from this original plant defence theory assume both costs of tissue damage in the presence of herbivores and costs of producing chemical defences in the absence of herbivores. The latter can arise from the allocation of limited resources to defences at the expense of other fitness-related traits (i.e. physiological costs, allocation costs, opportunity costs) (Karban & Baldwin 1997; Heil & Baldwin 2002). Alternatively, the production of defensive compounds could be ecologically costly if they compromise interactions with mutualists such as pollinators. Evidence for ‘ecological costs’ is beginning to accumulate, suggesting they are potentially an important force in the evolution of chemical plant defences (Strauss et al. 2002; Kessler & Halitschke 2007; Poelman, Van Loon & Dicke 2008).

Compromised interactions with pollinators are often cited as significant and probably underestimated costs of plant secondary metabolite production and thus important factors influencing selection on plant defensive traits (Lohmann, Zangerl & Berenbaum 1996; Strauss & Armbruster 1997; Strauss et al. 1999; Strauss & Whittall 2006; Adler 2007). The reproductive success of a plant is determined both by investment in reproductive tissue as well as maintenance of a large enough vegetative biomass to provide the necessary resources for reproduction. As a result, pollinator attraction might be compromised by chemical defences in two principal ways. First, allocation of resources to defence may come at the expense of investment in reproductive tissues, and may thereby reduce the attractiveness of flowers by reducing flower size, display and reward (nectar or pollen) quality or quantity (Mothershead & Marquis 2000). Second, a generally higher production of defensive metabolites in the plant may also be expressed in reproductive tissues and floral rewards, and thereby reduce pollinator attraction or the number of suitable pollinator species (Adler, Karban & Strauss 2001; Adler et al. 2006; Kessler & Baldwin 2007; Kessler, Gase & Baldwin 2008). In this scenario we can view toxic or deterrent metabolites in pollinator rewards as costs of plant defences that result from a pleiotropic effect of defence compound production in the leaves and floral tissues that are under selection by folivores and florivores respectively (Strauss et al. 1999; Adler 2000). This pleiotropy hypothesis is a non-adaptive explanation for why we find defensive secondary metabolites in pollinator rewards, and contrasts with a number of other hypotheses that assume that pollinators (Kessler & Baldwin 2007; Kessler, Gase & Baldwin 2008) and nectar robbers or degrading microbes (Baker 1977; Rhoades & Bergdahl 1981) are the primary selective agents on pollinator reward toxins (for a review, see Adler 2000). The mechanisms described by these hypotheses are not mutually exclusive but may represent the extremes of interactive continua that result from reciprocal and potentially conflicting natural selection by organisms interacting with plant chemistry and associated floral traits.

Accepting that both herbivores and pollinators are agents of selection on chemical floral traits allows for a number of predictions concerning the circumstances that would favour the evolution of certain combinations of reproductive and defensive strategies and the composition of the pollinator community in plant populations.

Strauss & Whittall (2006) predicted that floral trait values depend on the relative strength of the selective agents and on whether pollinator and non-pollinator agents of selection inflict antagonistic or coincidental selective pressure. Accordingly, pollinators and herbivores would exert coincident selective effects favouring the same trait optimum when they have opposing preferences for floral traits. Alternatively, floral traits may reflect a compromise between values maximizing plant fitness through interactions with antagonists and those maximizing fitness through interactions with pollinators when pollinators and herbivores share the same floral preferences. More specifically for defensive chemical traits, the relative dependence of a plant’s reproductive success on herbivory vs. pollinator availability should determine the evolution of vegetative and floral secondary metabolite production and may result in a correlation between defensive and reproductive strategies among different plant species. We are aware that few data exist with which to test these predictions, but they provide a valuable opportunity to perform comparative studies under various environmental conditions. As a first step, studies are needed that measure selection on floral chemical traits in the presence and absence of pollinators and herbivores respectively (e.g. island studies). Recent studies revealed that herbivory can alter the strength of pollinator selection on morphological traits, but such studies have not been extended to include measuring selection on chemical traits (Gomez 2003). Moreover, more studies are needed that test (i) whether defensive secondary metabolite content in the vegetative tissue is correlated with that of the reward tissues (nectar and pollen) and (ii) whether or not pollinator attraction is altered by herbivore-induced changes in plant chemistry. Along these lines, such studies will need to address the possibility that tissue-specific developmental regulation could sidestep the indirect selective pressures between pollinators and herbivores (as there would be no cross-tissue correlation). Ideally, plants could induce production of either foliar and/or floral chemicals (production of chemical defences only when needed) in order to escape compromising pollinator attraction with herbivore defences when herbivores are absent. The available data on the direction and relative strength of natural selection on floral chemical traits by pollinators and herbivores are very limited (but see Mant, Peakall & Schiestl 2005; Zangerl, Stanley & Berenbaum 2008). However, an increasing number of studies report tissue-specific and induced production of secondary metabolites. In the following two sections we first review tissue-specific production of defensive and signalling metabolites and herbivory-induced plant responses as potential factors influencing pollinator behaviour. We then introduce a case study with a wild tomato, Solanum peruvianum, to test whether allogamous plants will show (i) tissue-specific defensive secondary metabolite production (allogamy/tissue specificity hypothesis), or (ii) will induce changes in both leaf and floral secondary metabolism (allogamy/induction hypothesis) with consequences for pollinator behaviour.

Optimal defence and pollination

McKey (1974) was the first to suggest that if the resources available for the production of chemical defences against herbivores are limited, then the distribution of chemical protection within the plant should parallel the ‘needs’ of the plant as defined by the ‘value’ and ‘vulnerability’ of the different tissues. This hypothesis was later termed ‘optimal defence theory’ according to which the ‘value’ of a tissue is determined by its relative contribution to the plant’s Darwinian fitness (Rhoades 1979). Resources for the production of secondary metabolites as well as the allocation of metabolites should be directed to the most valuable tissues. Flower parts have a strong effect on fitness and should thus be protected by relatively high amounts of defensive plant metabolites compared to leaves, and this prediction is supported by numerous studies (Catalfamo, Martin & Birecka 1982; Hartmann et al. 1989; Detzel & Wink 1993; Bravo & Copaja 2002; Frolich, Hartmann & Ober 2006; Alves, Sartoratto & Trigo 2007; Beninger et al. 2007; Frolich, Ober & Hartmann 2007; Cirak et al. 2008; Zhang et al. 2008).

Correlated secondary metabolite production in flowers and leaves

In a survey of 14 publications that reported quantitative measures of secondary metabolites in leaf and flower tissues, 13 studies reported significantly higher secondary metabolite concentrations in flower tissues than in leaf tissues (Table S1, Supporting information). Additionally, studies in which a broader spectrum of compounds was analysed seem to find compounds exclusively produced in either the flowers or the leaves, suggesting a pattern of tissue-specific biosynthesis or allocation (Bennett et al. 2006). Moreover, finer scale analyses of different floral organs on the same flowers find the highest concentrations of defensive compounds in the stamen or pistil tissues (Bravo & Copaja 2002). It is noteworthy that none of these studies included a community wide or phylogenetic survey, reiterating the need for such studies to evaluate the evolutionary significance of tissue-specific differences in secondary metabolite production.

Does the presence of higher concentrations of defensive secondary metabolites in rewarding floral tissues (nectaries and anthers/pollen) compromise interactions with pollinators? Such pleiotropy-mediated correlations have been found in a number of cases. For example, the concentration of the deterrent and toxic alkaloid gelsemine in the nectar of Gelsemium sempervirens flowers is correlated with the concentration of gelsemine in the leaves. Gelsemine concentration in floral nectar is also negatively correlated with pollinator attraction, suggesting fitness costs of gelsemine production through pollinator deterrence (Irwin & Adler 2006). Similarly, the concentration of herbivore-defensive 1,2-saturated pyrrolizidine alkaloids (T-and iso-phalaenopsine) that are produced in every tissue of Phalaenopsis orchid hybrids is highest in the pollinia (Frolich, Hartmann & Ober 2006). If defences are optimized to maximize reproductive output of the plant, one should also expect plants to be under selection to limit the presence of toxic or deterrent secondary metabolites in pollinator rewards. One principal way for the plant to overcome this is the spatial regulation of the production and/or allocation of secondary metabolites and their secretion into nectar and pollen.

Defensive secondary metabolites in nectar and pollen

While the precise mechanisms of secondary metabolite allocation/production to nectar remain poorly understood (Adler 2000), it is likely that secondary metabolites are transported into the nectar through mechanisms similar to those known to mediate sugar and amino acid transport into nectar, including both passive diffusion and active transport (Luettge & Schnepf 1976; Fahn 1988). Evidence has been found for both secretory mechanisms and, additionally, the active resorption of nectar constituents (Nepi & Stpiczynska 2007, 2008) as well as the passive differential diffusion of floral volatiles into the nectar from surrounding tissues (Raguso 2004). Transport of components within the nectaries, the mechanism of active and passive secretion, and resorption all can influence the composition of nectar (Luettge & Schnepf 1976). Passive vs. active transport for different compounds (or the same compound in different species) may reflect conflicting natural selection by both pollinators and herbivores.

Given this diversity of mechanisms, it seems possible that natural selection could affect different nectar constituents differentially. Compounds that are passively transported into the nectar are more likely subject to compromising selection by both pollinators and herbivores while compounds that are actively secreted into the nectar, under regulatory control, are more likely subject to selection by pollinators and nectar robbers only. As a consequence, plants which are under strong selection by herbivores, and thus have high secondary metabolite production, also should be under strong selection to tightly control secondary metabolites in floral rewards when they depend on an allogamous reproductive strategy (see example of S. peruvianum below). Alternatively, selection might favour an evolutionary shift to an autogamous reproductive strategy if such plants’ defences compromise their interactions with pollinators, or the mitigation of compromising selection by herbivores and pollinators through induced defence metabolite production only when needed after attack (see next section).

Making predictions about the allocation of plant defences into pollen (compared to nectar) is even more difficult because we are aware of even fewer studies reporting pollen secondary metabolites (Pernal & Currie 2002; London-Shafir, Shafir & Eisikowitch 2003; Boppre, Colegate & Edgar 2005). Nectar is exclusively a reward, whereas pollen is only a reward for pollinators if its provision as food results in increased plant fitness, which likely results in more diffuse selection on pollen chemical traits compared with nectar chemical traits. For example, plants exclusively pollinated by nectar-foraging hummingbirds are unlikely to experience pollinator-mediated selection on pollen chemistry. Accordingly, in plant taxa which have pollen as the sole reward for pollinators (e.g. many Solanum and most Begoniaceae), selective forces on pollen chemical traits should be similar to those on nectar chemical traits. In contrast, pollen defensive secondary metabolite content should be relatively high in plants where pollen is not a reward and plants produce both nectar and pollen (Frolich, Hartmann & Ober 2006). As with nectar rewards, future studies on pollen composition should focus on multi-species and multi-tissue comparisons to more fully explore the potential for conflict in plant–pollinator–herbivore interactions.

Similar to leaf tissues, secondary metabolites in nectar and pollen may have multiple functions, such as defence (toxic or repellent) and interspecific signalling. Thus their production may be correlated with other attributes of the reward that are meaningful to pollinators. Flowers produce a large diversity of volatile compounds, many of which can also be found in nectar. As with other secondary metabolites, there is evidence for both a passive diffusion and active secretions of these compounds from the secreting tissues into the nectar (Raguso 2004). Although floral volatiles may not have a direct defensive function, their value as a signal is depending on the association with reward quality by the pollinator (Wright & Schiestl, 2009) and thus should be under very similar natural selection as nutrients and defensive metabolites in pollinator rewards. This may be why very similar hypotheses have been proposed for why plants have volatile organic compounds (VOCs) as well as non-volatile toxic or repellent secondary metabolites in pollinator rewards (Adler 2000; Raguso 2004).

Herbivore-induced plant responses and pollinator behaviour

The above examples illustrate the many ways by which herbivory can influence the interaction between plants and pollinators on an evolutionary scale. In addition, plants have a vast array of phenotypically plastic responses to herbivory that can influence their interactions with other organisms, including pollinators, in ecological time. While phenotypic plasticity itself is subject to natural selection, the extent to which the plasticity of a trait influences selection on the same trait is still debated (Agrawal 2001). An herbivore-induced change that reduces pollinator attraction could be viewed as an additional cost of inducible plant defences (Heil & Baldwin 2002). This would make pollinator attraction an important selective agent in a plant’s adaptive response to herbivory. However, induced responses to herbivory could also be viewed as adaptive strategies of plants to maximize fitness under the altered environmental conditions of herbivory. The herbivory-induced accumulation of a defensive metabolite can save the plant the costs of defences that would accompany the constitutive production of this trait when herbivores are absent, including the ecological cost of being less attractive to pollinators. Plants induce defences in response to an herbivore attack and can therewith limit the potentially fitness-reducing removal of leaf tissue (Karban & Baldwin 1997). Thus, induced plant responses that influence pollinator behaviour could represent a mechanism to cope with the compromising selection by herbivores and pollinators. In the absence of herbivory, plants would escape the antagonistically pleiotropic link between defensive and reward secondary metabolite accumulation to allow maximal pollinator attraction. We suggest that the extent to which induced plant responses to herbivory influence pollinator behaviour will depend on whether the benefits of induction for herbivore resistance outweigh the costs of deterring pollinators for plant fitness.

Herbivore damage and pollinator behaviour

Evidence for the effects of herbivore damage itself on pollinators is found in an increasing number of recent studies documenting the dramatic, immediate effects on a plant’s interactions with pollinators (Strauss 1997; Adler & Bronstein 2004; Irwin, Adler & Brody 2004; Bronstein, Huxman & Davidowitz 2007). To summarize, there are two principal ways in which herbivory can influence pollinators: (i) immediately, when pollinators avoid herbivore-damaged flowers because of the directly altered floral display (Murawski 1987; Karban & Strauss 1993; Cunningham 1995; Krupnick & Weis 1999; Krupnick, Weis & Campbell 1999; Adler, Karban & Strauss 2001) or avoid herbivores on the flowers (Lohmann, Zangerl & Berenbaum 1996), or (ii) subsequent to the attack, when herbivores induce growth and/or metabolic changes in the plant that mediate the effects of herbivory on pollinators.

Herbivore-induced effects include reduced flower numbers (Karban & Strauss 1993; Lehtila & Strauss 1997; Krupnick, Weis & Campbell 1999; Mothershead & Marquis 2000; Rathcke 2001), reduced flower size (Strauss, Conner & Rush 1996; Lehtila & Strauss 1997; Mothershead & Marquis 2000), the alteration of flowering phenology (Freeman, Brody & Neefus 2003; Sharaf & Price 2004) and sexual expression (Hendrix 1984; Mutikainen & Delph 1996; Strauss 1997; Krupnick & Weis 1998; Strauss, Conner & Lehtila 2001).

A meta-analysis of 16 studies that measured or manipulated herbivory and analysed pollinator responses revealed a dramatic negative effect of herbivory on pollinator behaviour (Fig. 1, Table S2). Interestingly, the effects of herbivory on pollinator behaviour differ, depending on where the plants were damaged, e.g. flowers, shoots or roots (Fig. 1). The strongest reduction in pollinator visitation was caused by damage inflicted directly to the floral tissue, whereas root herbivory actually resulted in increased pollinator attraction in two published studies (Poveda et al. 2003, 2005). Most of these studies only measure morphological changes in flowers or inflorescences of herbivore-attacked plants, but do not consider herbivore-induced changes in the chemistry of floral rewards or display as alternative factors influencing pollinator behaviour. However, relatively minor herbivore-damage to leaf tissue may result in a more rapid and significant reconfiguration of the plant’s secondary metabolism and reward quality, while flower size and number remain constant.

Figure 1.

 Impact of herbivory on pollinator visitation. Hedge’s d-values were calculated in a meta-analysis and the effect of flower, leaf and root herbivory was calculated as mean effect sizes (d+). Sample sizes n for each herbivory category are indicated and 95% bootstrap confidence intervals are shown as error bars. Comparison of the impact of the different herbivory categories by between-group heterogeneity statistics revealed significant differences between the effects of flower, leaf and root herbivory on pollinator visitation (Q = 8·4225, = 0·015). Studies included are listed in Table S2.

Herbivore-induced changes in pollinator reward quality

Recent studies find significant herbivory-induced changes in the pollen and nectar reward quality and quantity of attacked plants that can influence pollinator attraction and constancy (see Quesada, Bollman & Stephenson 1995; Aizen & Raffaele 1996; Strauss, Conner & Rush 1996; Krupnick, Weis & Campbell 1999). Even small herbivory-induced reductions in nectar production and concentration can significantly affect pollinator visitation and may shorten or eliminate visitations to flowers (Zimmerman 1983; Cresswell & Galen 1991; Mitchell & Waser 1992; Ackerman, Rodrigues-Robles & Melendez 1994; Hodges 1995). However, it remains unknown how such herbivore-induced changes in pollinator attraction affect male and female fitness of the studied plant systems.

Similarly little is known about the mechanisms that lead to induced reduction in pollinator reward quantity and quality. Interestingly, and contrary to floral nectar, extrafloral nectar production is usually increased in herbivore-attacked plants that bear such nectaries (Heil et al. 2001), which suggests that very similar damage and the elicited wound signalling pathways affect nectar production differentially in floral and extrafloral nectaries. Wound-induced jasmonate signalling has been shown to mediate the increased extrafloral nectar production, but its effect on floral nectar production remains unknown. Jasmonate signalling is known to be crucial for induced plant defences to herbivory, such as increased secondary metabolite accumulation, and is also involved in flower development (Creelman & Mullet 1997). Therefore, many of the induced phenotypic changes in floral display and pollinator reward quality and quantity could be subject to jasmonate-mediated regulation. This can certainly be true for induced changes in pollen and nectar quality. Pollen from staminate flowers on partially defoliated Cucurbita texana plants is less likely to sire seeds than pollen from undamaged plants, demonstrating that leaf damage can not only decrease pollen production but also pollen performance (Quesada, Bollman & Stephenson 1995). If the reduced performance of pollen is a result of the altered chemical composition of pollen, this may affect the behaviour of pollen-collecting pollinators. A similar reduction in pollen quality and size has been observed in herbivore-attacked Alstroemeria aurea(Aizen & Raffaele 1996, 1998) and Lobelia siphilitica plants (Mutikainen & Delph 1996). While we know very little about the secondary metabolite content of pollen and even less about its herbivory-induced changes, recent studies find induced changes in nectar constituents with ecological consequences. Adler et al. (2006) found that leaf and nectar alkaloid production is phenotypically correlated in Nicotiana tabacum plants and that herbivory induces an increased alkaloid production in nectar that could decrease pollinator preferences (but see Baldwin 2007).

Pollinator behaviour and herbivore-induced changes in VOC emission

Herbivore-induced changes in secondary metabolism also alter the volatile signalling of plants in ways that can have profound influence on pollinator attraction. There are only few examples of herbivore-induced changes in floral volatile emission but there is a large body of literature describing the specifically induced production of leaf VOCs on herbivore-attacked plants (reviewed in Janssen, Sabelis & Bruin 2002). While these herbivore-induced leaf volatiles can function as attractive signals for predators and parasitoids, and thus as indirect defences, they may also have signalling function for pollinators, as they affect the composition of the odour plume emanating from the plant, as well as the context in which flowers are perceived (Beker et al. 1989). Because induced volatile signalling is correlated with changes in plant quality, pollinators could use these same volatile signals to gain information about the reward quality of the plant as well as about the risks (predation) associated with approaching a plant if previous herbivory attracted more predators that could pose a risk to the pollinators (Dukas & Morse 2003; Munos & Arroyo 2004). In the following case study we found such a correlation between herbivore-induced floral and leaf volatile emission and the behaviour of the pollinators as well as a flower-specific secondary metabolite production, and discuss these findings in the context of the plant’s reproductive strategy.

Herbivory-induced pollinator limitation in Solanum peruvianum

The wild tomato plant S. peruvianum is native to the pacific slopes of the Andean mountains of South America and occurs in seasonal dry forest habitats. Solanum peruvianum, like all tomato plants, does not produce floral nectar and the sole reward for visiting pollinators is pollen stored inside the anther cone surrounding the stigma. This morphological feature found in most Solanum species limits the potential pollinators to certain bees that are able to ‘buzz’ flowers to release the pollen from the anther cone for collection (Luo, Zhang & Renner 2008). In the native habitat of S. peruvianum in Peru, we observed bees of the Apidae, Colletidae and Halictidae families as the most abundant pollinators. While working on a number of different native tomato species in Peru we observed that leaf tissue damage by a diverse herbivore community was generally correlated with reduced pollinator visitation. This reduced pollination could be the result of an altered pollinator behaviour in response to a changing reward quality and/or floral signalling in herbivore-attacked plants compared with unattacked plants. Here we test the above hypotheses by analysing pollinator behaviour and the volatile and non-volatile secondary metabolite production in herbivore-attacked and undamaged plants.

Materials and methods

Our initial field observations in Peru motivated the characterization of the underlying mechanisms in more controlled manipulative experiments. The study was performed in Ithaca, NY, rather than in Peru in order to control plant history and development, and to manipulate herbivore damage. Furthermore, the utilization of a single pollinator species, Bombus impatiens (Apidae), reduced the complexity of the study system and allowed a focused behavioural analysis. Bombus impatiens is behaviourally and functionally very similar to the bee species that we observed in the native habitats of S. peruvianum and would allow subsequent insect physiological experiments that we can not perform with the native bee species in the field.

To investigate the functional link between herbivory and pollination we planted 16 S. peruvianum plants (accession LA1616) in field plots at Cornell University (Ithaca, NY, USA). The seeds were provided by the TGRC collection (UC Davis, CA, USA) and sown in mid-April. Seedlings were grown in the greenhouse for 6 weeks before they were planted into the field plots. All experiments were performed on flowering plants at the end of September. For chemical analysis we used tissue from flowers that opened at least 3 days after the initial damage by an herbivore (10–15% of leaf tissue removed) or from similarly developed flowers of undamaged plants.

As a first step to characterize the chemical profiles and to evaluate the defensive and semiochemical function of volatile and non-volatile compounds of foliar and floral tissue, we harvested leaves, anther cones and pollen from individual plants. Samples were weighed, flash frozen in liquid N2 and extracted with 80% MeOH. Phenolic compounds were analysed by HPLC using a standard method (Keinanen, Oldham & Baldwin 2001).

To analyse herbivore-induced resistance, six plants each were exposed to herbivory by 6 third-instar Manduca sexta larvae (10–15% leaf tissue removed). Another six plants remained undamaged as controls. After 10 days larvae were removed and new M. sexta neonate larvae were placed on plants of both treatment groups. The neonates were allowed to feed for 7 days and were then removed and weighed.

To measure pollinator preferences, we generated arenas of individual plants with Manduca-attacked and undamaged branches. Previous experiments (A. Kessler, unpublished data) had shown that branches on S. peruvianum plants respond independently to damage and do not result in systemic plant responses that can be measured in distant branches. Flower preference and remaining time on the flowers newly opened after the damage were recorded for individual B. impatiens bumblebees within the arena, so that each bee observation resulted in a proportion of visits to flowers in the two treatment categories as well as an average residence time for each bee on flowers of damaged and undamaged branches. A total of 12 bees with a minimum of 12 flower visits within the experimental arena were recorded and used in the analysis.

To determine whether a change in plant volatile emissions in response to herbivory could be used by the pollinators as a cue to avoid flowers on damaged plants, we collected VOCs from the headspace of leaves and flowers on untreated and herbivore-damaged S. peruvianum plants. VOCs were collected by enclosing single leaves in 500 mL polyethylene cups and pulling air through ORBO-32 charcoal adsorbent tubes (Supelco, Bellefonte, PA, USA), using a 12 V vacuum pump (GAST®), Gast Manufacturing Inc., Benton Harbor, MI, USA). The ORBO-32 tubes were desorbed with dichloromethane and samples were analysed by GC-MS (Kessler & Baldwin 2001). Selected ion chromatograms were integrated and peak areas of individual compounds were normalized by the area of the internal standard (tetraline).

Results and discussion

Floral and vegetative secondary metabolite production in Speruvianum

Solanum peruvianum is an allogamous species and sexual reproduction is fully dependent on bee-assisted cross-pollination (Rick 1963). Contrary to the ‘allogamy/tissue-specificity hypothesis’ our HPLC analysis found correlations between the content of several non-volatile phenolic compounds in leaves and pollen, including compounds with reported defensive functions such as rutin (Spearman Rank Correlation, Z = 1·846, P = 0·06) and chlorogenic acid (Spearman Rank Correlation, Z = 1·056, P = 0·039) (Fig. 2); this result mirrors a similar study on the nectar of tobacco (Adler et al. 2006). However, a number of unidentified coumaroyl derivatives were only present in pollen and anther cones, but not in leaf tissue, while a number of caffeic acid derivatives (e.g. neo-chlorogenic acid) were found only in leaf tissue but not in pollen. This result suggests that the plants, at least in part, are able to control secondary metabolite production or accumulation in a tissue-specific manner.

Figure 2.

 Phenolic compound production in leaf and floral tissue of Solanum peruvianum. (a) Representative HPLC chromatograms of leaf, anther cone and pollen samples (1, neo-chlorogenic acid; 2, chlorogenic acid; 3, quercetin glycoside; 4, 7, 8, unidentified flavonoids; 5, rutin; 6, quercetin glycoside; 9, coumaroyl derivatives). (b) Correlation between chlorogenic acid (y = 0·12x − 0·06; r2 = 0·453) and rutin (y = 0·04x + 0·04; r2 = 0·212) concentrations in leaves and pollen.

Herbivore-induced VOC emission of Speruvianum

A very similar conclusion can be drawn from the GC-MS analysis of the volatile emissions of flowers and leaves (Fig. 3). Several leaf- and flower-specific VOCs could be identified, while a large proportion of the headspace bouquet was shared between leaves and flowers. Interestingly, the VOC emission patterns of flowers and leaves of Manduca-attacked plants differed significantly from those of undamaged plants (Fig. 3, Table S3). Some of the compounds were induced in both flowers and leaves, which, again suggests that leaf and flower secondary metabolism are closely linked. However, other compounds were leaf-specifically and flower-specifically induced, which suggests that a common wound signal is transmitted throughout the plant but triggers tissue-specific responses.

Figure 3.

 Volatile organic compound emission from leaves and flowers of Manduca-attacked and undamaged (Control) plants of Solanum peruvianum. Numbers designate identified compounds that illustrate the difference between leaf and flower as well as herbivore-induced and un-attacked tissues. 1, α-pinene; 2, unknown monoterpene; 3, β-myrcene; 4, β-pinene; 5, unknown monoterpene; 6, limonene; 7, cis-β-ocimene; IS, internal standard, tetraline; 8, methyl salicylate; 9, 3,5-dimethoxy toluene; 10, unknown sesquiterpenes; 11, 3,5-dimethoxy benzaldehyde/geranyl acetone; 12, unknown sesquiterpene.

Herbivory-induced changes in pollinator behaviour

The herbivory-induced changes in the secondary metabolism of S. peruvianum were correlated with reductions in subsequent herbivory (Fig. 4a; Student’s t-test, = 7·66, = 0·0003), pollinator visitation (Fig. 4c; Student’s t-test, = 10·32, < 0·0001) and residence time (Fig. 4d; Student’s t-test, = 4·43, = 0·0006). Hence, foliar herbivory increased the resistance of the plant to the attacking herbivore species and reduced its attractiveness to bumblebee pollinators. These results complement findings of earlier studies with similar negative effects of herbivory on pollinator attraction (Fig. 1). However, our results differ from most previous reports on changes in pollinator visitation, which have been linked to changes in floral display as a result of reduced flower number or size (Steets & Ashman 2004; Poveda et al. 2005; Steets, Hamrick & Ashman 2006). We analysed inflorescences on damaged and undamaged control branches and found no changes in flower numbers (Student’s t-test, = 1·7, = 0·11, = 10) and size (Student’s t-test, = 0·9, = 0·37, = 10), suggesting that nutritional and/or infochemical changes in the floral tissue mediate the reduced pollinator attraction. Although floral chemical traits have been linked to pollinator behaviour, only one other study so far showed induced changes in nectar secondary metabolite constituents with an effect on pollinators (Adler et al. 2006).

Figure 4.

 Induced resistance in Solanum peruvianum. (a) Average Manduca sexta caterpillar mass (+SEM) after 7 days of feeding on undamaged control plants and previously Manduca caterpillar-attacked plants. (b) Bombus impatiens bumblebee ‘buzz’-pollinating a tomato flower. (c) Bombus impatiens preference [average proportion (+SEM) of undamaged control and Manduca-damaged flowers visited per observation] for and (d) average visitation time (+SEM) on flowers of plants that are attacked by M. sexta caterpillars or undamaged (Control).

The avoidance behaviour of B. impatiens to flowers on damaged plants in our study is strongly correlated with the altered floral chemistry and is the major correlate we could identify. Although this is strong evidence that the herbivore-induced chemical changes in the plant caused this altered pollinator behaviour, we do not yet know why the pollinators avoid the chemically altered flowers. We note that bumblebees barely land on flowers of damaged plants (Fig. 4c), suggesting remote evaluation of their quality. The average residence time on those flowers is less than 2 s, which is certainly not enough time for the typical ‘buzzing’ behaviour, which takes longer than 3 s for B. impatiens (Fig. 4d). These observations suggest that the pollinators use volatile and/or colour visual cues to differentiate between flowers on damaged and undamaged plants. The latter seems less likely because corollas and anther cones of damaged and undamaged plants showed no obvious colour differences in the visible light and UV spectrum in preliminary experiments (data not shown). Whether or not bumblebees use visual or olfactory cues to avoid flowers of damaged plants, we propose three basic and non-mutually exclusive hypotheses for why they avoid these flowers.

  • 1The cue is associated with pollen quality, which decreases in flowers of damaged plants. An increased secondary metabolite production in the pollen could decrease the pollen quality, but bees may not be able to assess the quality of the pollen during foraging (Roulston, Cane & Buchmann 2000). Unless the ability to avoid pollen of lower quality (Robertson et al. 1999) from herbivore-damaged plants is inherited or there is a feedback loop from the offspring, signal–quality association is unlikely to directly explain the bumblebee preferences.
  • 2The cue is associated with increased predation pressure. Herbivore-induced VOC emission can attract predators and parasitoids to the attacked plant and may increase predation pressure on mutualist insects such as pollinators. In our experiments we observed no predators on the plants or flowers, but ‘damaged plant odours’ may be generally associated with increased predation pressure in their native habitats if the arthropod populations are strongly ‘top-down’ controlled.
  • 3The pollinator avoids an ‘unfamiliar’ floral cue. Bee species typically have high flower constancy, the mechanisms of which remain highly contested (Wright & Schiestl 2009). However, if bees are using VOC cues as part of their search image to find flowers that were previously rewarding, flowers on damaged plants could be avoided because they emit a fundamentally different VOC bouquet. Floral and vegetative VOC emissions are likely integrated by the insect given their mutual proximity, which would mean that an induced change in vegetative VOC emission alone (floral emission unchanged) might be sufficient to trigger the avoidance response in the bees. This hypothesis is supported by preliminary data from another wild tomato species, S. habrochaites, where only vegetative VOC emission changes in response to herbivory (floral VOC emission unchanged) but bees do also avoid flowers on damaged plants.

Herbivory-induced changes in Speruvianum herkogamy

In our study the herbivore-induced changes in plant metabolism were associated with reduced pollinator attraction, which results in an herbivore-induced pollinator limitation and thus in a potential ecological fitness cost of plant defences. However, the plant may have ways to alleviate these fitness costs by also changing herkogamy (stigma–anther separation; SAS) in response to herbivore damage. While floral display and the size of the pollen-containing anther cones remained constant, a significant reduction in pistil length was induced by M. sexta herbivory (Fig. 5). This disproportionate change in size of different flower parts results in a change in SAS in flowers of damaged tomato plants. Stigma exertion or SAS is associated with the evolution of autogamy in the Lycopersicon section of the genus Solanum (Chen & Tanksley 2004) and has been identified as a crucial factor determining selfing rates in several plant genera including Mimulus (Carr, Fenster & Dudash 1997), Ipomoea (Parra-Tabla & Bullock 2005) and Datura (Stone & Motten 2002). If the observed reduction in SAS in wild tomato plants changes the selfing rate in herbivore-damaged plants it could represent a mechanism of reproductive assurance allowing the plant to alleviate fitness consequences of the herbivory-induced pollinator limitation. The adaptive function of this proposed mechanism should be highly dependent on the type of self-fertilization barrier.

Figure 5.

 Herbivore-induced changes in herkogamy. Anther cone and pistil length were measured in flowers on Manduca sexta-damaged Solanum peruvianum plants (Manduca damage) and plants that remained undamaged (Control). Only pistil length changed significantly in response to Manduca attack (*Student’s t-test, = −2·69, = 0·02), while the size of the anther cone remained constant (= −0·313, = 0·76).

These hypotheses can be tested in detailed behavioural and comparative field studies and allow analysing the causal links between herbivory and reduced pollinator attraction. Furthermore, such studies provide an opportunity to integrate functional floral morphology and chemistry in a more balanced and dimensional concept of floral ‘phenotype’. Additional chemical analyses of pollinator reward quality and herbivore-induced changes, in a comparative context, should allow us to identify common patterns in the production of plant secondary metabolites and their impact on the mediation of herbivore–plant–pollinator interactions.

Conclusion

We have discussed predictions arising from theory on how herbivores and pollinators interact to shape the evolution of chemical floral traits. The literature analysis and our data from S. peruvianum partially support the allogamy/tissue specificity hypothesis. While a large proportion of secondary metabolite production in vegetative tissues is correlated with the content in the pollinator reward, other proportions can also differ tissue-specifically. The second hypothesis (allogamy/induction hypothesis) is supported by our data from S. peruvianum that show that secondary metabolite production in floral tissues can be induced by herbivory. The literature review and our own findings identify major gaps in our understanding of how plant defence and pollination are coordinated through secondary metabolism.

First, we need a more systematic comparative analysis of pollinator reward chemistry and its relationship to whole plant secondary metabolism. The analysis should be performed across and within different taxonomic groups in the light of the defensive and reproductive strategies of each included species. In order to accomplish this goal, systematic studies of defence chemistry in the vegetative and reproductive tissues of taxonomically related groups should be integrated with detailed case studies of the pollination and defence ecology of some of the constituent species. At present, these two fields of research are just beginning to connect (Adler & Irwin 2005; Irwin & Adler 2006; Andrews, Theis & Adler 2007; Bronstein, Huxman & Davidowitz 2007). While making this connection, particular attention should be paid to the identity and tissue-specific production of the compounds and compound classes that similarly or differentially affect pollinator and herbivore behaviour (e.g. is the production of certain compound classes more likely than others to be selected on by both, herbivores and pollinators). This will provide crucial information about the phylogenetic constraints that may limit plants in coping with compromising natural selection by herbivores and pollinators. At this point we can only speculate on this question because studies of the natural selection on floral chemical traits by pollinators and herbivores are almost entirely missing. Moreover, there is a need to better understand the physiological mechanisms of how defensive secondary metabolites are secreted into the nectar and pollen rewards.

Second, we found only a few examples of herbivore-induced changes in floral chemistry and their effects on pollinators (Adler et al. 2006), although there are examples of changing nutrient content and quality in nectar and pollen (Quesada, Bollman & Stephenson 1995; Aizen & Raffaele 1996; Strauss, Conner & Rush 1996; Krupnick & Weis 1999). Here we propose induced changes in whole plant and floral chemistry as a potential plant strategy to alleviate fitness costs that might derive from an antagonistic pleiotropic interaction between whole plant and floral chemistry. Changes in plant VOC emission in response to herbivory may provide the information for both herbivores and pollinators to assess the quality of the plant as a food source. With respect to plant defensive chemistry, pollinators are herbivores too (Praz, Mueller & Dorn 2008), but we do not know the extent to which pollinators use the same cues as herbivores to assess plant quality, nor do we know much about the reliability of plant VOC signals. Thus, it seems important to consider both the herbivore-induced changes in floral- as well as leaf-VOC emission as potentially informative cues for pollinators. Studies of phenotypic plasticity of floral chemical traits in general, and VOC signalling in particular (e.g. Majetic, Raguso & Ashman 2009), should be included in the systematic analysis of floral and defensive leaf chemistry and should be viewed as interconnected entities.

Through their direct and immediate effects on fitness, pollinators and herbivores are the agents of selection that most likely influence pollen and nectar chemistry in most cases. Future studies that combine phylogenetic information, whole-plant and pollinator reward chemistry, and functional molecular biology in ecologically relevant settings (Kessler, Gase & Baldwin 2008) will provide us with the tools to better predict evolutionary outcomes between plants, pollinators and herbivores.

Acknowledgements

We thank Robert A. Raguso, Lynn Adler, Anurag Agrawal, Stuart Campbell, Katja Poveda, Ian Kaplan, Amy L. Parachnowitsch, Robert F. Bode and three anonymous reviewers for helpful suggestions on the manuscript and inspiring discussions when developing the ideas presented in this paper. The study was funded by the National Science Foundation (DEB-0717139).

Ancillary