In response to arthropod herbivory, plants release volatile organic compounds (VOCs), which are attractive to natural enemies. Consequently, VOCs have been interpreted as co-evolved plant–natural enemy signals. This review argues that, while these data are necessary, they are not sufficient to demonstrate a VOC plant–natural enemy signaling function. We propose that evidence that (1) plant fitness is increased as a consequence of natural enemy recruitment, and either (2A) natural enemies preferentially learn prey-induced VOCs or (2B) natural enemies respond innately to the VOCs of the prey–host plant complex, is also required. Whereas there are too few studies to rigorously test hypotheses 1 and 2A, numerous studies are available to test hypothesis 2B. Of 293 tests of natural enemy responses to VOCs, we identified only 74 that were unambiguous tests of naïve natural enemies; in the remainder of the tests either natural enemies were experienced with their host in the presence of VOCs, or experience could not be ruled out. Of those 74 tests with naïve natural enemies, attraction was observed in 41 and not in 33. This review demonstrates that empirical support for the hypothesized VOC plant–natural enemy signaling function is not universal and presents alternative hypotheses for VOC production.
Although communication is ubiquitous in biology and consensus on the meaning of terms is critical for scientific progress (see West et al., 2007; Scott-Phillips, 2008), ambiguity in the use of the term ‘communication’ and the associated concepts ‘signal’ and ‘cue’ exists. These terms have been discussed in detail elsewhere and we refer readers to Scott-Phillips (2008) for an excellent review of the literature. In brief, definitions of communication are primarily of two types: those that emphasize adaptation and those that emphasize information transfer. Informational definitions have been criticized for being ‘conceptually unsound, and at best derivative upon the adaptationist approach’ (Scott-Phillips, 2008). Adaptationist definitions emphasize reciprocal adaptations for signal production and response. Three of the more significant consequences of emphasizing adaptation are: communication occurs via signals and not cues; information transfer is an emergent property of communication, not a defining property; and traits associated with the production and reception of signals are not evolutionarily independent (Butlin & Ritchie, 1989; Endler, 1992). This means that implicit to the interpretation that VOCs are signals between plants and natural enemies is the assumption that a behaviorally active blend of VOCs is the product of a co-evolved mutualism between plants and natural enemies. This is because signals are stimuli whose perception benefits not only the receiver but also the emitter; by contrast, the perception of cues is beneficial to the receiver independent of any effect on the emitter (Greenfield, 2002). Therefore, parallel changes in traits of signalers and receivers are required for the evolution of signals (Butlin & Ritchie, 1989). This is not the case with traits associated with emission and reception of cues, which are free to follow independent evolutionary trajectories. Thus, in the context of the interactions between plants and natural enemies, the classification of VOCs as cues or signals is contingent (in part) on whether or not plants that attract natural enemies with VOCs have a higher mean fitness than those that do not. In short, if VOCs are plant–natural enemy signals then: (1) plants that produce particular VOC blends will recruit greater numbers of natural enemies than plants that emit other or no VOC blends; (2) VOC production is under genetic control and heritable; (3) increased attraction of natural enemies reduces herbivory and increases plant fitness; therefore (4) natural enemies have imposed selection on plants for more attractive VOC blends (see Faeth, 1994 for a review). Alternatively, if VOCs are cues, then there is no expectation that recruitment of natural enemies would consistently and systematically reduce losses in plant fitness resulting from damage by herbivores. Therefore, if VOCs are signals, the form of the signal (i.e. the absolute and relative amounts released) will have been modified to improve information transfer as a consequence of fitness benefits associated with the attraction of natural enemies. Conversely, if a particular VOC blend is not predictably associated with plant fitness, then the VOC blend would not be modified by natural selection and VOCs must be a cue.
Using the terminology of Otte (1974), if VOCs are plant–natural enemy signals then their function must be to recruit natural enemies to damaged plants; however, if VOCs are cues, the recruitment of natural enemies would only be an effect of VOCs. As Otte (1974) points out, not all cases of stimulus reception and the attendant information transmission constitute examples of signaling: ‘For any given signal emission, there is at least one potential class of legitimate receivers for which the signal is intended; but the information transmitted may be of use to illegitimate receivers as well, for instance predators, parasites, and competitors for whom the information is not intended, but who exploit communicative systems of other organisms’. Unlike illegitimate predator, parasite and competitor receivers, the effect of eavesdropping natural enemies (hereafter ‘fortuitous’ receivers) on plant fitness can be positive, neutral or negative (in some cases parasitism by gregarious koinobiont hymenoptera can result in increased host feeding; e.g. Rahman, 1970). Although the scenario of natural enemy recruitment benefiting plant fitness intuitively seems to suggest a signaling function, VOCs should only be referred to as signals if characteristics of the VOC blend have been selected to inform natural enemy receivers (see Otte, 1974).
Encounters between arthropod herbivores and their natural enemies typically have negative and positive fitness consequences for the herbivore and natural enemy, respectively. This places strong selection on herbivores to reduce, and natural enemies to increase, encounter rates. As a result of selection to optimize their relative encounter rates, herbivores are usually inconspicuous and well hidden, and natural enemies use flexible foraging strategies that incorporate environmental cues from multiple modalities (e.g. Vinson, 1976). Phenotypic plasticity in foraging strategies (i.e. learning) has been demonstrated in several arthropod natural enemies (e.g. Lewis & Tumlinson, 1988; de Boer et al., 2005). For example, while naïve female Microplitis croceipes (Hymenoptera: Braconidae) do not orient towards some volatile stimuli, those given an experience with nonvolatile components of host feces in association with the same volatile stimuli do (Lewis & Tumlinson, 1988). This type of phenotypic plasticity is an example of associative learning. Remarkably, the ability of some arthropod natural enemies to learn is not limited to biologically relevant chemical stimuli (e.g. Lewis & Tumlinson, 1988; Olson et al., 2003). Female M. croceipes can learn to recognize and associate novel and otherwise unattractive odors such as vanilla with nonvolatile host recognition cues (Lewis & Tumlinson, 1988). The expectation is that individual natural enemies experience higher encounter rates with prey as a result of their ability to learn and incorporate volatile cues (e.g. VOCs) into their foraging strategies. Evidence showing the adaptive value of associative learning, that is, a link between associative learning and increased fitness of natural enemies, exists in a few systems (e.g. Dukas & Duan, 2000).
For adaptive evolution of the VOC blend emitted to occur (i.e. for a VOC signal between plants and natural enemies to evolve), heritable variation among individuals in their VOC blends must exist and variation in the VOC blend must correlate positively with variation among individuals in their lifetime reproductive success (fitness). A positive correlation between VOC blend variation and fitness can only exist if natural enemies respond innately to VOCs or preferentially learn a subset of VOC blends. This is because plasticity in the use of VOCs by natural enemies (i.e. the ability to learn and respond to novel VOC blends) would result in natural selection on the VOC blend that is not consistent and systematic over successive generations. An association between a VOC phenotype and increased fitness that may come to exist in ecological time is unlikely to be stable in evolutionary time. One predicted outcome of this type of selection is that the blend of VOCs released should vary spatially among and temporally within populations. The ability to learn to associate diverse volatile chemical stimuli with the presence of prey items, coupled with the absence of either innate responses or preference learning, would facilitate the existence of interactions between plants and natural enemies where increases in plant fitness as a consequence of recruitment of natural enemies do not result in systematic and directional natural selection on plant VOCs.
While necessary, the observed differences in the VOCs of herbivore-damaged and undamaged plants and natural enemy responses to VOCs of damaged plants are not sufficient to demonstrate signaling between plants and natural enemies. As Williams (1966) pointed out, ‘Major difficulties also arise from the current absence of rigorous criteria for deciding whether a given character is adaptive, and, if so, to precisely what it is an adaptation’. The identification of rigorous criteria to facilitate discrimination between recruitment of natural enemies as a function or as an effect of VOCs (sensuOtte, 1974) remains a major challenge. We propose that, in addition to the observed differences in the VOCs of herbivore-damaged and undamaged plants and natural enemy responses to those VOCs, demonstration that VOCs are plant–natural enemy signals requires evidence that (1) plant fitness is increased as a consequence of natural enemy recruitment, and either (2A) natural enemies preferentially learn a subset of the potential VOCs consistent with those released by the host plant(s) of their prey or (2B) natural enemies respond innately to particular blends of VOCs induced by prey. To discriminate between the mutually exclusive interpretations that recruitment of natural enemies is an effect or a function of VOCs, we propose the above criteria (1 and either or both 2A and 2B) as predictions unique to the hypothesis that recruitment of natural enemies is a function of VOCs. These predictions can be tested empirically, the results of which would allow discrimination between recruitment of natural enemies as a function or as an effect of VOCs. Hypothesis 1 has rarely been tested at the level of the individual plant. We are also unaware of any work to determine if natural enemies learn preferentially (hypothesis 2A). By contrast, there is a substantial body of work addressing the learned and innate responses of natural enemies to plant odors to begin to evaluate hypothesis 2B. Below we review the literature on responses of natural enemies to VOCs. Specifically, using a dichotomous scoring system (naïve or experienced) we determined the state of experience of natural enemies for all studies demonstrating responses of natural enemies to induced VOCs that met our criteria (see section II. Literature). We then compared the responses to VOCs of naïve and experienced natural enemies. In particular we wanted to know what proportion of the literature involved experienced natural enemies, and, for naïve natural enemies, the proportion that were (or were not) attracted to VOCs.
Our literature search began with the citations from several reviews of tritrophic interactions mediated by VOCs (e.g. Turlings et al., 1995; Dicke, 1999; Hunter, 2002; Turlings & Wäckers, 2004). We subsequently expanded our search to include citations discovered in the primary literature. The more recent literature (2004–2009) was searched with the ISI Web of Knowledge (databases: SCI Expanded, SSCI and A&HCI) using the search terms ‘natural enem* and induced volatil*’, ‘tritrophic interaction* and induced volatile*’, and ‘herbivore induced and volatile*’. The citations from these and other studies subsequently discovered were searched as above (concluded February 2009).
For inclusion in the database a paper had to: (1) examine some aspect of natural enemy behavior related to host location mediated by VOCs; (2) have an appropriate test (e.g. damaged vs undamaged plant) to allow a response to general plant odors to be ruled out (attraction of natural enemies to uninfested plant material has been observed (e.g. Takabayashi & Dicke, 1992)); and (3) if it was a laboratory study, provide enough information to allow classification of the natural enemy as experienced or naïve.
The level of natural enemy experience was scored by consulting the text of the study. Studies that were reported as employing naïve natural enemies were examined in more detail to verify the lack of experience with VOCs and to determine if preimaginal learning could be ruled out. Contact with pupal cocoons or mummies developed in the presence of induced plant volatiles, for example, has been demonstrated to result in preimaginal learning of induced plant volatile cues (Hérard et al., 1988; van Emden et al., 1996; Storeck et al., 2000). Natural enemies were only considered naïve if they had no experience with any part of the host–plant complex, including pupal mummies or cocoons developed in the presence of infested plants. This can be achieved in many if not all cases by rearing natural enemies on herbivores reared on artificial diets free of plant material or by surgically removing natural enemies from pupal cocoons or mummies before emergence (Hérard et al., 1988; van Emden et al., 1996; Storeck et al., 2000). Natural enemies were considered experienced if they encountered infested plant material or VOCs either directly from plants infested with hosts or indirectly from their pupal cocoons or the mummies of their hosts (in a few studies natural enemies were allowed to oviposit in hosts in the absence of VOCs; these natural enemies were considered experienced). Natural enemies from field studies were assumed to be experienced. Their survivorship to collection suggests that they either recently emerged or had foraged successfully and thus probably experienced prey in the presence of VOCs.
For each study that met these criteria we recorded: (1) the identity of the first, second, and third trophic levels; (2) whether the natural enemy was a predator or parasitoid; (3) whether the study was conducted in the field, laboratory, or glasshouse; (4) the assay used to measure natural enemy behavior; (5) natural enemy experience state; and (6) if the natural enemies had no experience with VOCs, whether or not naïve natural enemies were attracted to induced volatile stimuli. Most studies involved multiple tests of responses of natural enemies to induced volatile stimuli. The unit of replication in our database was the species of natural enemy. Studies that examined the responses of multiple natural enemy species to the VOCs of a plant–herbivore complex were entered into the database once for each species. Similarly, studies that examined the responses of a natural enemy to multiple plant–herbivore VOCs were entered into the database once for each plant–herbivore complex. Many studies had multiple tests of both naïve and experienced natural enemies and observed both positive and negative responses to the VOCs tested. In studies with multiple tests of a natural enemy to the VOCs of a plant–herbivore complex, a single test of naïve individuals observing attraction resulted in that natural enemy being scored as naïve and attracted. A few studies involved both laboratory and glasshouse (n =15), laboratory and field (n =1) or laboratory, glasshouse and field (n =1) tests and thus were scored as both or all three. Some tests used synthetic compounds to induce plants, or plants genetically engineered to constitutively release VOCs (n =47) and thus did not involve a herbivore. Finally, a plant was only considered wild if it was not an agronomic or ornamental plant and thus had not been artificially selected.
We measured our effectiveness in exhaustively sampling the literature with the Web of Science (SCI-Expanded; time-span 1900–2007) and the search terms ‘tritrophic interaction*’ (n =502) and ‘induced plant volatil*’ (n =694). A haphazardly chosen sample of 100 studies from each search (200 total) was examined to determine whether they met the criteria for inclusion in the database. For those studies that met the criteria we determined whether they were already present in the database.
III. The database
We reviewed over 450 studies. Our search of the literature recovered 293 tests of natural enemy responses from 160 studies. The literature search used to measure the completeness of our database suggested that the database represented a robust sample of the literature. The sample of 200 haphazardly chosen studies from both searches identified an additional nine studies involving 11 tests of natural enemy responses that were not in our initial sample. These studies are included in the 293 tests. Of the 293 tests of natural enemy responses, 176 were reported as experienced and 117 as naïve. Scrutiny of the 117 tests of ‘naïve’ natural enemies identified 47 tests where preimaginal learning could not be ruled out because of contact with pupal cocoons, or mummies developed in the presence of induced plant volatile cues; however, in 11 of these tests the natural enemies were not attracted to VOCs. In studies that demonstrate attraction and where preimaginal learning cannot be ruled out, it is not possible to discriminate between the possibilities that: (1) natural enemies are truly naïve and are innately attracted to VOCs, and (2) natural enemies are experienced and have learned to respond to VOCs. While the former provides support for plant–natural enemy signaling the latter does not; because of this confusion, these studies were excluded. By contrast, studies in which preimaginal learning cannot be ruled out but natural enemies were not attracted to VOCs do not suffer from this limitation. Even though preimaginal learning could not be ruled out, the lack of attraction means that in those tests not only do these natural enemies not respond innately, they may not be capable of preimaginal learning. Therefore, we consider these valid tests and included them in the database. Similarly, several papers were excluded from the database because they compared VOCs versus clean air, different doses of VOCs, VOCs from plants in different stages of induction, VOCs from different plant–host complexes, or VOCs from different cultivars. Although the results of these comparisons are consistent with the paradigm of natural enemy recruitment by VOCs, the treatments used mean that a response to general plant odors cannot be ruled out. Tests of naïve individuals involving VOCs versus clean air that did not observe attraction were included in the database as ‘naive and VOCs not attractive’ because they did not have the same limitation of ambiguity of interpretation (n = 4; Reed et al., 1995; Wyckhuys & Heimpel, 2007). Of the remaining 81 good tests of responses of naïve natural enemies (the remaining 70 tests plus the 11 tests that did not observe attraction but may have involved preimaginal learning), seven appear to involve physiological anomalies (attraction was observed in natural enemies reared on a carotenoid-deficient diet but not in natural enemies reared on a complete diet) and were excluded. Of the remaining 74 tests, attraction was observed in 41 tests and not in the remaining 33 (Fig. 1). These 74 tests involved 27 different natural enemy species, with attraction, no response to VOCs and both attraction and no response observed in 11, 10 and 6 of these, respectively (see Supporting Information Table S1). Excluding those species that are both attracted and not responsive (n =6), 11 out of 21 (c. 52%) of the naïve species were attracted to VOCs compared with 41 out of 74 total tests (c. 55%). The two ratios appear to be equivalent.
Among the 74 tests of naïve natural enemies, several species were tested more than once. In six of these species (two parasitoids and four predators) both positive (n =16 tests) and negative (n =15 tests) responses to VOCs were observed. These conflicting results are probably attributable to differences in the plant–herbivore VOCs tested, methodological differences among studies, or both. For example, most of the studies in our database used plant tissues induced before assay as a source of volatiles. The length of time between induction and assay ranged from hours to days among studies, and this suggests one possible explanation for variable results among experiments. Several studies have observed that natural enemy responses are not independent of the time post-induction that tissues are assayed. For example, broad beans (Vicia faba) infested by Acyrthosiphon pisum for 24 and 48 h were not more attractive to Aphidius ervi (determined using the percentage of flights that resulted in oriented flights or source landings) than uninfested plants, whereas plants infested for 72 h were (Guerrieri et al., 1999). Similarly, mite-infested tea (Camellia sinensis) leaves were not attractive to Neoseiulus womersleyi 1 and 4 d post-infestation but were attractive 7 and 10 d post-infestation (Maeda et al., 2006). By contrast, maize (Zea mays) only needs to be induced for a few hours before becoming attractive (Turlings & Tumlinson, 1992). Of the 15 tests observing negative responses, five of them probably involved plant material assayed too soon after induction to be releasing the full complement of VOCs (e.g. 2, 15 and 24 h post-induction; for three of the tests the time between induction and assay was not given). This may also explain some of the 33 tests of naïve natural enemies that were not attracted to VOCs (e.g. Potting et al., 1999). Future studies may require a more thorough understanding of the time course of induction when contemplating experimental design.
The point has been made that, “although vast, the literature on plant–insect interactions has a precariously narrow base because it is based on a handful of plant families, of chiefly agricultural interest, and that these families are hardly random samples of plant diversity” (Berenbaum & Zangerl, 2008). The literature on VOC-mediated tritrophic interactions could be similarly described. In fact it has an even narrower base, with the majority of the work focused on an even narrower group of plants, pest herbivores and natural enemies primarily of applied interest (i.e. biological control). At all three trophic levels a broader phylogenetic context is needed, particularly for natural enemies, in order to address issues of phylogenetic weighting and how best to interpret natural enemy responses to VOCs (i.e. single vs multiple evolutionary events). Of the 293 tests of natural enemy responses to induced volatile cues, 262 involved tests of cultivated plants (agronomic or ornamental plants) and 31 tests of wild plants. A total of 56 different plant species were assayed in the 293 tests. The most common cultivated and wild species were the lima bean (Phaseolus lunata; n =47) and Arabidopsis thaliana (n =12), respectively. Approximately 82% (n =241) of the tests involved laboratory assays (olfactometer and flight chamber trials), 14% (n =42) involved field assays (predation/parasitism rates and trapping trials) and 10% (n =28) involved glasshouse assays (flight chamber and release recapture trials). The sum of the percentages totals more than 100% because 17 of the tests in the database involved more than one type of assay. The database includes tests with 65 different herbivore species. The most commonly tested herbivore was the spider mite Tetranychus urticae (n =60). A total of 86 natural enemy species were tested, 37 parasitoids and 50 predators (one natural enemy was scored as both a predator and a parasitoid). Of the 293 tests, 141 were of predator responses and 152 were of parasitoid responses. The most commonly tested predator and parasitoid were Phytoseiulus persimilis (n =49) and Cotesia glomerata (n =20), respectively (Table S1).
Despite the wealth of primary literature examining natural enemy responses to VOCs, we found a surprisingly limited literature to test hypothesis 2B (natural enemies respond innately to particular blends of VOCs induced by prey) (74 tests in 27 natural enemy species). The literature we did find was equivocal, with 41 tests supporting and 33 tests refuting hypothesis 2B. These 74 tests involved 27 different natural enemy species, with ‘attraction, no response to VOCs’ and both ‘attraction’ and ‘no response’ observed in 11, 10 and 6 of these, respectively (see Supporting Information Table S1). Excluding those species that are both attracted and not responsive (n = 6), 11 out of 21 (c. 52%) of the naive species were attracted to VOCs compared with 41 out of 74 total tests (c. 55%). The two ratios appear to be equivalent. These results suggest that at least in some cases the recruitment of natural enemies to VOCs is an effect of VOCs (Otte, 1974). In the following sections we consider the relationship between the recruitment of natural enemies and plant fitness (see IV. Recruitment of natural enemies), the expectation for learned vs innate natural enemy responses to VOCs (see V. Learning and VOC evolution) and other explanations for the existence of VOCs (see VI. Additional VOC functions).
IV. Recruitment of natural enemies
To date, the study of the adaptive function of VOCs has focused on their indirect effects on herbivores; however, as Price et al. (1980) point out, plant fitness should be the focus of studies of the adaptive function of plant defenses. Many studies have demonstrated that natural enemies are attracted by VOCs (Turlings & Wäckers, 2004) but few have demonstrated increased attack rates on herbivores as a consequence (De Moraes et al., 1998; Thaler, 1999; Bernasconi-Ockroy et al., 2001; Kessler & Baldwin, 2001) or have measured the impact of recruitment of natural enemies on plant fitness under realistic conditions (but see Karban, 2007 for a counter example). Although choice experiments can clearly demonstrate the ability of natural enemies to respond to VOCs, they do not provide information about the functionality of VOC emission (i.e. observed effects of attraction of natural enemies on plant fitness). Despite this, increases in plant fitness are widely believed to result from top-down effects mediated by VOCs. Successful biological control programs (e.g. cottony-cushion scale (Icerya purchasi) (Debach, 1974, p. 92); ash whitefly Siphoninus phillyreac (Gould et al., 1992)) and factorial exclusion experiments (e.g. Costamagna & Landis, 2006) suggest that in some cases natural enemies can suppress herbivore populations. However, mechanisms that lead to top-down effects on herbivore populations do not necessarily result in increased plant fitness. The effects of herbivory on plant survival and reproduction differ significantly among and within species depending on biotic and abiotic factors (Strauss & Agrawal, 1999). For example, while low levels of herbivory (leaf area loss of 10%) can significantly reduce plant fitness in some species (e.g. Marquis, 1984), in others, high levels of herbivory (leaf area loss of 25%) have no observable effects on plant fitness (e.g. Lehtilä & Strauss, 1999). In extreme cases of tolerance, herbivore-damaged plants can have higher fitness than undamaged plants (e.g. Paige & Whitham, 1987). As a consequence, without measuring plant fitness directly, a priori it is difficult to predict what (if any) consequences top-down effects on herbivore populations will have on plant fitness.
Attraction of natural enemies does not guarantee a reduction in herbivory and a reduction in herbivory does not guarantee an increase in plant fitness (reviewed by Faeth, 1994). Surprisingly few studies have examined the fitness consequences of VOC-mediated natural enemy responses to VOCs. The results of manipulative experiments with A. thaliana (van Loon et al., 2000) and maize (Fritzsche Hoballah & Turlings, 2001) have been interpreted as evidence of a defensive function of VOCs mediated by natural enemy responses. Both studies involved significantly reduced environmental and genetic heterogeneity and experimentally controlled trophic interactions. Parasitized or unparasitized caterpillars were confined on undamaged plants and the effect of herbivory on plant fitness measured using the metrics of seed production (both studies) and plant dry weight after caterpillar pupation or death (Fritzsche Hoballah & Turlings, 2001). Caterpillars were experimentally parasitized (i.e. natural enemy attack was not mediated by VOCs) and plants were initially kept in climate chambers (maize) or in a glasshouse (A. thaliana). Although consistent with the hypothesis that VOCs are plant–natural enemy signals, these types of study are not sufficient to demonstrate increased plant fitness as a consequence of the attraction of natural enemies by VOCs. A few studies have examined the forces that determine community structure under more realistic, natural conditions. While some found support for increased plant fitness as a consequence of top-down effects (e.g. Gomez & Zamora, 1994; Tooker & Hanks, 2006), some did not (e.g. Karban, 2007) and others suggested that variation among factors that influence bottom-up effects may influence plant fitness more strongly (e.g. McMillin & Wagner, 1998). In fact, it has been argued that bottom-up population structuring forces may be more common than top-down forces in parasitoid–host food webs (Hawkins, 1992). Fitness benefits mediated by natural enemy responses to VOCs are critical to the proposed plant–natural enemy signaling function. The lack of empirical evidence supporting the signaling function represents a serious deficiency in the literature. Clearly, additional field studies exploring the effects of natural enemy attraction to VOCs are needed to resolve this issue.
The impact of natural enemy (specifically parasitoids) recruitment on plant fitness is further complicated by natural enemy life history. While predation and parasitism by idiobiont hymenoptera terminate herbivore damage, increased and decreased feeding in larvae parasitized by koinobiont hymenoptera has been reported. It has been suggested that parasitism by solitary koinobiont hymenoptera results in decreased feeding, while the effect of parasitism by gregarious koinobiont hymenoptera on host feeding varies and appears to be influenced by parasitism level and host plant quality (e.g. Rahman, 1970; Thompson & Redak, 2007).
In addition to natural enemy life history (e.g. solitary or gregarious parasitoids), the complexity of natural food webs may interfere with the realization of increases in plant fitness induced by top-down effects. To date, empirical work on chemically mediated tritrophic interactions has focused on linear agroecosystem food webs (e.g. crop–herbivore–natural enemy). Natural food webs are much more complex (e.g. Rosenheim et al., 2004). With respect to trophic cascades induced by top-down effects, increased food web complexity has two consequences of interest. The first involves the effect of high prey diversity on top-down effects. A recent review concluded that higher prey diversity often dampened top-down effects as a result of an increased probability of including nonhost prey and reduced efficacy of specialist natural enemies when confronted with diverse prey (Duffy et al., 2007). The second involves increased interactions among natural enemies. Herbivores usually possess diverse natural enemy assemblages (e.g. the leafhoppers Prokelisia dolus and Prokelisia marginata in mid-Atlantic salt marshes are fed on by hunting and web-building spiders and predaceous bugs and beetles; Finke & Denno, 2002). Thus VOCs have the potential to promote interactions among natural enemies and potentially alter the cascade of top-down effects among trophic levels via increased intraguild predation (e.g. Holt & Polis, 1997; Rosenheim, 1998; Finke & Denno, 2004). Mesocosm experiments with variable predator assemblages have documented the potential for dampened top-down effects resulting in diminished trophic cascades with increased natural enemy diversity (e.g. Finke & Denno, 2004, 2005; Finke & Snyder, 2008). These experiments also demonstrate that high predator diversity can result in the complete absence of top-down effects.
V. Learning and VOC evolution
It has been hypothesized that the level of dietary specialization of natural enemies will influence whether or not natural enemy responses to chemical stimuli are innate or learned (Vet & Dicke, 1992). This hypothesis predicts that the optimal character state of natural enemies will be correlated with the degree of host specialization of those natural enemies. Selection is predicted to favor learned responses in natural enemies with broad host ranges and innate responses in natural enemies with narrow host ranges, the underlying assumptions being that (1) because of the static nature of the diet of specialist natural enemies, the informational content of VOCs would be high in evolutionary time; and (2) because of the dynamic nature of the diet of generalist natural enemies, the informational content of VOCs would be low in evolutionary time. A recent review of the literature found mixed support for the predicted relationship between the level of dietary specialization and the type of responses of natural enemies (Steidle & van Loon, 2003).
The informational value of a stimulus is the product of its reliability (i.e. the correlation with available and suitable prey) and detectability (i.e. apparency in the foraging environment). Foraging natural enemies face what has been referred to as the reliability–detectability dilemma (Vet et al., 1991). As a result of strong and continuous selection on herbivores to be inconspicuous, the reliability and detectability of prey-related stimuli are expected to be inversely correlated. This imposes a significant constraint on the evolution of long-distance foraging strategies by natural enemies. As a consequence, to optimize their foraging efficiency natural enemies are predicted to learn highly detectable, but less reliable, stimuli associated with prey host substrates (i.e. VOCs). Theoretical models predict that selection will only favor phenotypic plasticity (e.g. learning) if: (1) populations experience variable environments, (2) environments produce reliable cues that favor different phenotypes, and (3) no phenotype is superior in all environments (Ghalambor et al., 2007). The blend of VOCs has been shown to vary among plant populations (or cultivars) (e.g. Takabayashi et al., 1994; Loughrin et al., 1995; Gouinguene et al., 2001; Hare, 2007), with plant age, the tissues sampled and host plant condition (e.g. Takabayashi et al., 1994; Scutareanu et al., 1997), with herbivore infestation level and instar (e.g. Takabayashi et al., 1995; Scutareanu et al., 2003), with the stage of attack (e.g. Röse et al., 1996), with light intensity and water and fertilization regime (e.g. Takabayashi et al., 1994; Gouinguene & Turlings, 2002; Lou & Baldwin, 2004) and depending on the presence or absence of plant pathogens (e.g. Cardoza et al., 2002). This variation suggests that natural enemy populations do experience variable environments and that no natural enemy foraging phenotype may be superior in all environments (support for conditions 1 and 3 favoring phenotypic plasticity); however, support for the reliability of VOCs in ecological time is equivocal (condition 2). While some studies have observed similar VOC blends in response to herbivore damage and artificial damage (e.g. Mattiacci et al., 1994) and in response to different herbivores (e.g. Kessler & Baldwin, 2001), others have observed the necessary qualitative and quantitative variation in the blend of VOCs induced among herbivore species attacking the same host plant (e.g. Takabayashi et al., 1994; Ozawa et al., 2000; Shiojiri et al., 2001).
Although the above variation is apparent to our instrumentation, its significance regarding the reliability of VOCs depends on whether or not it is apparent to natural enemies. Some studies have observed discrimination in the orientation behavior of natural enemies (e.g. Takabayashi et al., 1995; Du et al., 1996; De Moraes et al., 1998; Powell et al., 1998) while others have not (e.g. Turlings et al., 1993; Röse et al., 1998; van Poecke et al., 2003). Interpretation of the available literature is complicated by the fact that discrimination among blends by natural enemies does not appear to be independent of experience. Several studies have observed that, while naïve individuals are not attracted to host-induced VOCs, experienced individuals are (e.g. Reed et al., 1995; Potting et al., 1999; Schnee et al., 2006; Blande et al., 2007). Experimental work also suggests that the scale at which natural enemies are able to resolve differences among stimuli is dependent on their level of experience. For example, Vet et al. (1998) observed that the parasitoid Leptopilina heterotoma did not discriminate between volatile chemical stimuli with subtle qualitative differences when they only had a single rewarding experience with the treatment stimuli. They did discriminate when one rewarding experience with the treatment stimuli was coupled with an unrewarding experience with the control stimuli. These results suggest that not only are natural enemies more responsive after experience, but they also may be better able to discriminate among reliable and unreliable VOC blends. Thus VOCs may only be a solution to the reliability–detectability problem for appropriately experienced natural enemies. Ultimately, the evaluation of the reliability of VOCs in evolutionary and ecological time will require an understanding of (1) which VOCs in the blend are important for attraction; (2) the ability of natural enemies to detect differences in VOC blends (see Vet et al., 1998); (3) the duration of induction of release of the VOC blend relative to the presence of vulnerable prey; and (4) the repeatability/heritability of the VOC blend induced. All of these areas remain virtually unexplored. With regard to (1), in some field trapping studies, partial blends are attractive and individual components interchangeable (e.g. Kessler & Baldwin, 2001; James, 2003a,b, 2005; James & Grasswitz, 2005). These results are potentially inconsistent with plant VOC–natural enemy co-evolution.
An older view of phenotypic plasticity is that it is a mechanism to facilitate the persistence of populations until selection is able to fix a phenotype (i.e. innate response) via genetic assimilation (Baldwin, 1896; Waddington, 1942). Genetic assimilation occurs when environmentally induced phenotypic variation becomes fixed in response to natural selection (reviewed by Pigliucci et al., 2006). With respect to VOCs this would mean that, if VOCs are reliable in evolutionary time, learning could facilitate the evolution of innate responses (assuming learning is the plesiomorphic character state). Using a simple paradigm where the behavioral response of the parasitoid is defined as an expression of the rate individuals learn, the component of the behavior that is instinctual, and the number of trials, Papaj (1993) used simulations to explore the relationship between learned and congenital responses. Interestingly, the rate of learning was inversely correlated with the rate at which innate responses evolved. A limiting condition was that, when learning is relatively rapid (L = 0.05), then innate responses no longer evolve. Although parameter values of this magnitude are high, they represent a rate slower than that of single trial learning, a phenomenon that has been observed in parasitic hymenoptera (e.g. Kaiser et al., 2003). The ability to learn rapidly would essentially eliminate any selective advantage associated with congenital responses by significantly reducing the fitness differences between genotypes with and without innate responses. Papaj (1993) speculates that this may explain the existence of learned responses when theory predicts innate ones (e.g. in feeding and oviposition behavior in specialist insects; Papaj, 1986; Papaj & Prokopy, 1989). These simulations suggest that learning is most adaptive in and maintained by variable conditions, and that the ability to learn rapidly may retard the evolution of innate natural enemy responses.
VI. Additional VOC functions
The VOC blends emitted by individual herbivore-infested plants are complex, containing terpenoids, phenylpropanoids/benzenoids and fatty and amino acid derivatives among other classes of compounds and in some cases may contain > 200 compounds (Dudareva et al., 2006). Complexity is not limited to the VOC blend; the ecology of VOC-mediated interactions is similarly complex and potentially influenced by herbivore and natural enemy community structure on all plant parts (reviewed by Dicke et al., 2009). The diversity of compounds suggests that the VOC blend may be multifunctional and the type and intensity of natural selection acting on these compounds may vary spatially and temporally. Several recent reviews discuss additional VOC functions that do not involve natural enemy recruitment (Holopainen, 2004; Peñuelas & Llusià, 2004; Dudareva et al., 2006; Vickers et al., 2009). Below we discuss three of these hypothesized additional VOC functions. This discussion is not intended to be exhaustive and we refer readers to those reviews and included citations for more detail.
1. Function 1
VOCs (or a subset of the VOCs released) increase plant fitness via direct effects on herbivores. Some of the induced volatiles of maize are repellent to aphids (e.g. Bowers et al., 1972; Chapman et al., 1981; Gibson & Pickett, 1983; Hardie et al., 1994; Pettersson et al., 1994; Mostafavi et al., 1996; Bernasconi et al., 1998). Similarly, spider mites of the species Tetranychus urticae have been observed to disperse away from the volatiles of infested plants (Dicke, 1986; but see Pallini et al., 1997). Negative effects on the oviposition behavior of lepidopteran herbivores have also been observed (e.g. Landolt, 1993; De Moraes et al., 2001; Kessler & Baldwin, 2001). Aphids exposed directly to several C6 alcohols and aldehydes (green leaf volatiles (GLVs)) (Hildebrand et al., 1993) and spider mites reared on previously induced plant material (Dicke & Dijkman, 2001) had reduced fitness. Vancanneyt et al. (2001) used an antisense strategy to reduce the levels of GLVs released by treated plants and observed that aphids had increased fecundity on those plants. Conversely, the attraction of herbivores to GLVs has also been reported (e.g. Halitschke et al., 2008). Similarly, Meldau et al. (2009) demonstrated that, although plants with their direct defenses disabled (the ability to produce both nicotine and trypsin protease inhibitors in response to herbivory was reduced by silencing two mitogen-activated protein kinases in Nicotiana attenuata) experienced severe defoliation, they did not if GLV emission was simultaneously disabled. While these studies primarily involved GLVs, they illustrate: (1) the potential for direct effects of some VOCs on herbivores; and (2) that the consequences of VOC emission are complex. The observed direct effects that reduce herbivory could be mediated by (1) increased predation and parasitism risks associated with natural enemy recruitment (e.g. ovipositing females are avoiding habitats characterized by high predation/parasitism risk) or (2) differences in herbivore fitness on induced and noninduced plants in the absence of the third trophic level. While the former is consistent with the null hypothesis of plant–natural enemy signaling, the latter is not. The observed physiological effects cannot be explained by the predicted increased predation and parasitism risk associated with the null hypothesis that VOCs are adaptations for the recruitment of natural enemies. Direct effects on herbivores mediated by differences in the fitness of herbivores on induced and noninduced host plants in the absence of the third trophic level would obviate the need for trophic cascades mediated by natural enemy recruitment. Support for this hypothesis would require empirical evidence that the direct effects result in a trophic cascade with positive effects on plant fitness in the absence of natural enemy recruitment.
2. Function 2
VOCs increase plant fitness by protecting plant tissues from stressful abiotic conditions. Multiple protective functions of VOCs have been proposed in the literature, including: (1) protection from UV-B radiation; (2) protection from oxidative damage; and (3) protection from heat damage (reviewed by Peñuelas & Llusià, 2004; Holopainen, 2004; Dudareva et al., 2006; Vickers et al., 2009). Although preliminary, empirical support of these adaptive functions is accumulating. (1) Aerosols have been demonstrated to be an effective protection against UV-B radiation (Palancar & Toselli, 2004). Bonn & Moortgat (2003) proposed that some VOCs (sesquiterpenes) may be involved in fine particle aerosol formation. The relationships between VOCs and aerosol formation under natural conditions are not well understood; however, there is some evidence that in response to exposure to damaging levels of UV-B radiation the emission of VOCs increases, which may promote aerosol formation and reduce UV-B radiation tissue damage (Johnson et al., 1999; Holopainen, 2004). (2) Relative to other VOCs, sesquiterpenes are highly reactive with ozone (Bonn & Moortgat, 2003). Ozone-resistant and susceptible tobacco (Nicotiana tabacum) plants have different temporal patterns of VOC (sesquiterpene) production in response to exposure to ozone. Ozone-resistant tobacco produces sesquiterpenes immediately after exposure while ozone-susceptible tobacco does not (Heiden et al., 1999). Work with A. thaliana suggests a relationship between ozone resistance and the production of VOCs that quench ozone (Holopainen, 2004; Dudareva et al., 2006). It has also been suggested that some VOCs (isoprenoids) may protect plants from oxidative damage via reactions with membrane bilayers that increase membrane stability, or oxidants (Peñuelas & Llusià, 2004; Vickers et al., 2009). (3) Interactions between isoprene and the thylakoid membrane and/or large membrane-bound protein complexes are a putative mechanism for isoprene- or monoterpene-induced thermotolerance (Sharkey & Yeh, 2001; Dudareva et al., 2006; Vickers et al., 2009). Rosenstiel et al. (2004) proposed an adaptive VOC function related to thermotolerance, VOC production and emission as a metabolic safety valve to remove excess energy and carbon the plant is unable to process (which can occur during stressful conditions such as high temperatures and drought). This metabolic safety valve involves isoprene synthase, dimethylallyl diphosphate and avoidance of the sequestration of phosphate. Peñuelas & Llusià (2004) and Vickers et al. (2009) review these functions but come to different conclusions. If the function of VOCs is to protect plant tissues from abiotic stress, the relationship among the production and release of VOCs, insect herbivory and this VOC function remains to be clarified.
3. Function 3
The release of VOCs is an incidental consequence of cell damage and not a defensive adaptation (van der Meijden & Klinkhamer, 2000). Niinemets et al. (2004) suggest that the principal physicochemical characteristics of the VOCs (e.g. solubility, volatility and diffusivity) play a major role in VOC emission and that the VOC blend released in response to changes in the plant environment is the result of the effects of these changes on plant physiology and the physicochemical properties of the VOCs. The observation that mechanically damaged plants release far fewer VOCs (both quantitatively and qualitatively) than herbivore-damaged plants seems to contradict this hypothesis (e.g. Turlings et al., 1990). One limitation of studies like these that report differences in the VOC profiles of mechanically and herbivore-damaged plants is potentially confounding differences between artificial and herbivore damage. Experimental work using simulated herbivore damage (MecWorm) more similar in physical appearance and duration to natural herbivore damage observed VOC blends qualitatively similar to herbivore-induced VOCs for some plant species (Mithöfer et al., 2005) but not all (Maffei et al., 2007). Connor et al. (2007) observed that the response of natural enemies to VOCs induced by artificial damage was not independent of the rate that plant tissues were damaged. While natural enemies did orient preferentially to herbivore-damaged plants versus plants that experienced rapid artificial damage, they did not discriminate between herbivore-damaged plants and plants that experienced artificial damage on a time scale similar to that of herbivore damage. These results are consistent with the nonadaptive, incidental consequence of cell damage hypothesis.
Williams (1966) succinctly summarized one of the major difficulties with the concept of adaptation, namely deciding to precisely what a given character is an adaptation. Organisms are complex and traits often have multiple functions, all of which have the potential to contribute to individual fitness. It is unlikely that all of the trait functions evolved contemporaneously; rather, one probably represents the primary function and the others secondary functions. These ideas (i.e. adaptive primary and secondary functions) were developed into the concepts of adaptation (primary adaptive functions) and exaptation (secondary adaptive functions) (Gould & Vrba, 1982). Recruitment of natural enemies is but one way in which the production of VOCs may contribute to plant fitness. In those systems where multiple functions contribute to plant fitness, determining which are primary and secondary functions (i.e. adaptations vs exaptations) remains a major challenge (and may not be possible). Unlike the determination of primary and secondary functions, it is possible to measure the relative contribution of each function to plant fitness. Confirmation and quantification of selection acting on the VOC blend await determination of which components of the blend are involved in each function. The fact that natural enemies possess remarkable plasticity in their ability to learn chemical stimuli (e.g. Lewis & Tumlinson, 1988; Olson et al., 2003) would significantly facilitate the evolution of a secondary function of VOCs as a plant cue for natural enemy recruitment. This is not a new idea. Turlings et al. (1995) also suggested that the production and emission of terpenoid VOCs may have evolved because of their deleterious behavioral and physiological effects on herbivores and that their use by foraging natural enemies may be a secondary function (also discussed by Harrewijn et al., 1994; Janssen et al. 2002). The significance of this idea seems not to have been fully appreciated.
We propose the following predictions to test whether the recruitment of natural enemies is an adaptive function of VOCs. Our first testable prediction is that an individual plant should be able to attract adequate natural enemies to reduce herbivore damage and increase plant fitness relative to nonemitting genotypes. To date this is an area of research that has not received adequate empirical attention, particularly in nonagronomic systems where the interpretation of results is not potentially confounded by correlated responses to artificial selection for agronomic traits. Alternatively, if VOCs benefit plants directly, then individual plants should release high enough levels of VOCs to significantly improve fitness even in the absence of natural enemies. In theory, this prediction could be tested by comparing plants that vary in VOC emission in the presence and absence of natural enemies. For the reasons outlined in our Introduction regarding learned responses as a selective force, our second prediction is that natural enemies should either respond innately to VOC blends induced by prey herbivores or preferentially learn the VOCs (or some subset) induced by prey herbivores.
The activity of natural enemies can induce trophic cascades that positively affect plant fitness and consequently select for plant traits that promote the occurrence of indirect ecological links between the first and third trophic levels. In response to herbivory by arthropods, plants systemically emit VOCs. Since the seminal publications by Dicke & Sabelis (1988) and Turlings et al. (1990), chemically mediated tritrophic interactions have played a prominent role in the field of plant–insect interactions, and responses of natural enemies to VOCs have been interpreted as evidence of signaling between plants and natural enemies. The hypothesis that the VOC blend is an adaptation for the recruitment of natural enemies has largely been accepted on the basis of congruence between observed differences in the VOCs released by herbivore-damaged and undamaged plants and a putative selective force (i.e. potentially increased recruitment of natural enemies inferred primarily from Y-tube bioassays). This conclusion represents an emergent property of the adaptationist program (e.g. Gould & Lewontin, 1979; Pigliucci & Kaplan, 2000) and occurs as a consequence of the absence of rigorous criteria for evaluating the null hypothesis of adaptation (see Williams, 1966). Additional adaptive functions of VOCs have been proposed (see section VI. Additional VOC functions). We propose the following set of testable conditions as criteria for evaluating the null hypothesis that VOCs are adaptations for natural enemy recruitment: (1) field studies must demonstrate that recruitment of natural enemies results in trophic cascades that increase plant fitness, and natural enemies must either (2A) preferentially learn the VOC blends released by appropriate plant–herbivore complexes, or (2B) respond innately to the VOC blends released by appropriate plant–herbivore complexes, but not others. Although the effort required to obtain truly naïve natural enemies for assay is not insignificant, we believe that the only way to critically evaluate the hypothesis that the function of VOCs is to recruit natural enemies and discriminate among VOC functions is to assay naïve individuals. Although we believe these conditions are necessary and sufficient to support the hypothesis that VOCs are adaptations for natural enemy recruitment, in cases where multiple adaptive functions exist it may not be possible to determine which are primary and which are secondary adaptations. When multiple adaptive functions exist, measurement of the relative contributions of each function to plant fitness will be the most profitable approach to understanding the evolutionary ecology of VOCs. Our review of the literature identified 293 tests (from 160 studies) of natural enemy responses to VOCs, in 219 of which either individuals were experienced or experience could not be ruled out (7 of these 219 cases involved naïve individuals but appear to be physiological anomalies). In 41 of the 74 tests of naïve natural enemies, attraction to the VOCs was observed, demonstrating that support for the hypothesis of plant–natural enemy signaling is not universal. If this ratio is representative it suggests that in many systems VOCs are not adaptations for the recruitment of natural enemies, that these natural enemies learn appropriate VOC blends preferentially, or that, in terms of plant fitness, the natural enemies assayed are unimportant members of the food web (see Karban, 2007).
We thank Ian Baldwin, Ring Cardé, Jocelyn Millar and Dan Papaj for helpful comments and discussion during the early stages of conception and development of the manuscript and Ian Baldwin and Ring Cardé for critical reviews of the manuscript (acknowledgement does not imply agreement with all of the ideas as presented). Our research is supported by the National Science Foundation under Grant no DEB 0414181 to JDH.