Plants produce a diverse array of physical and chemical traits that confer various types of resistance to insect pests, and can mediate complex interactions among the members of the plant-associated arthropod community (Kessler & Baldwin 2002b; Kessler & Halitschke 2007; Poelman, Van Loon & Dicke 2008). Among the most striking examples of such interactions are those between plants and species of the third trophic level – predators and parasitoids of herbivores. In a seminal paper, Price et al. (1980) highlighted that ‘…theory on insect–plant interactions cannot progress realistically without consideration of the third trophic level…. as part of a plant’s battery of defences against herbivores’. Since that time, researchers have gained insights into the expression of plant traits that facilitate top-down control of herbivores. These traits have been termed ‘indirect defences’ to contrast them with plant traits such as toxic and anti-digestive compounds, leaf toughness or trichomes, that directly reduce herbivore fitness and thereby increase plant fitness (Duffey & Stout 1996). We follow the nomenclature of Karban & Baldwin (1997) and use the term resistance (e.g. indirect resistance) for traits which affect the preference and performance of herbivores (i.e. the herbivore’s perspective), and the term defence for traits that, through the reduction in herbivore performance, have a positive effect on plant fitness (i.e. the plant’s perspective). Accordingly, indirect defences are plant traits that (i) attract and improve foraging by natural enemies of herbivores, and thereby (ii) facilitate control of herbivore populations and (iii) ultimately increase plant fitness (Karban & Baldwin 1997).
Indirect resistance can theoretically be mediated by a large spectrum of traits, including those that constitutively facilitate the residence of natural enemies on the plant, as well as herbivore-induced traits that may be produced to recruit ‘bodyguards’ (Dicke & Sabelis 1988) in response to herbivore attack. These traits may provide some kind of resource to predatory insects, such as food and shelter, or provide information about the location and identity of herbivores in the form of altered visual and olfactory cues (Fig. 1).
Conventional hypotheses for the evolution of plant defence suggest that these resource-and information-providing traits should be viewed collectively as indirect defences (Karban & Baldwin 1997). However, the phenotypic diversity of these traits suggests that they may be subject to quite different selective forces, and quite disparate ecological contexts, depending on variation in the herbivore and predator communities, and on the strength of the association between plants and their defenders (i.e. facultative vs. obligate). Indeed, we argue here that resource and information based recruitment represent two fundamentally different strategies. Moreover, we suggest that our theoretical and conceptual understanding of plant defence would be improved by a more detailed analysis of the adaptive value of these different traits in their natural ecological context. In this paper, we attempt such an analysis. We first review the current state of knowledge for the major indirect defence strategies, and then synthesize these findings to shed light on which traits actually contribute to a plant’s ‘indirect defence’.
Ant–plant interactions cover the spectrum from facultative to obligate and symbiotic
Ant plants are prime examples of a plant-predator symbiotic mutualism (Fig. 1). Plants provide resources, largely constitutively produced (Heil & McKey 2003), to ant species that may be specialized to nest and feed on these plants (Fiala & Maschwitz 1990). The ants consume extrafloral nectar (EFN) and food bodies (FBs), and may nest in specialized plant parts in return for a protection of the plant from attackers and competitors. These systems can be obligate (myrmecophytes) and represent the association of one or a few plant species with one or very few ant species (Bronstein, Alarcon & Geber 2006) (Fig. 1a), or facultative (myrmecophiles), and may lack a close association (Fig. 1b). The production of ant food rewards increases predator visitations, and thus predation on herbivores may be higher on EFN-producing plants than on plants not producing EFN (O’Dowd 1982; Koptur 1992; Heil & McKey 2003). Similarly, physical shelters (domatia), such as hairy leaf pads or hollow shoots may increase predation pressure on plants with these traits (Agrawal, Karban & Colfer 2000; Heil 2008). Facultative plant associations with predators are often mediated via an inducible production of the rewards. For example, the secretion of EFN can increase in response to the presence of consumers and/or herbivory and lead to increased ant visitation (Heil et al. 2001).
Ant–plant associations mediated through the provision of food and/or shelter appear to be widely distributed and can involve up to one-third of all woody plant species in a given habitat (Schupp & Feener 1991). EFN is secreted by species in at least 332 plant genera from 93 Angiosperm families as well as in 11 genera of ferns (Koptur 1992; Keeler 2008). Domatia suitable for ants are found in at least 200 plant species and are particularly common and diverse in form and structure in the Rubiaceae (Beattie 1989). Moreover, very close and specialized ant-plant associations are found in over 100 genera of tropical angiosperms in at least 20 plant families (O’Dowd 1982; Davidson & McKey 1993). The group of ant species engaged in specialized plant associations comprises five subfamilies of the Formicidae (Davidson 1997), while facultative ant plants often recruit a larger diversity of insect taxa into the indirect defensive arsenal.
Plant volatiles mediate facultative interactions with the third trophic level
The close relationships between tropical ants and ant plants discovered in the nineteenth century were widely accepted as mutualistic by the 1970s (Janzen 1966; Bentley 1977; Bronstein 1998; Heil & McKey 2003; Bronstein, Alarcon & Geber 2006). In contrast, the emission of herbivory-induced volatile organic compounds (HIVOCs) was only recently identified as a trait that can attract predators and parasitoids to herbivore-damaged plants. Classic examples of HIVOC-mediated indirect resistance include the attraction of predatory mites to spider mite-attacked lima beans (Dicke & Sabelis 1988), and the attraction of the parasitoid braconid wasp, Cotesia marginiventris, to maize seedlings that are attacked by the beet armyworm, Spodoptera exigua (Turlings, Tumlinson & Lewis 1990). Meanwhile a whole range of folivorous mites (Dicke & Sabelis 1988; Dicke et al. 1990), insects (Turlings, Tumlinson & Lewis 1990; Drukker, Scutareanu & Sabelis 1995; Du et al. 1998; Halitschke et al. 2000; Kessler & Baldwin 2001), stem borers (Potting, Vet & Dicke 1995; Khan et al. 2008), seed feeders (Steidle, Fischer & Gantert 2005), and root feeders (Van Tol et al. 2001; Rasmann et al. 2005) have been shown to induce VOCs. They attract predators and parasitoids from at least five animal orders, including entomophagous nematodes (Rasmann et al. 2005), predaceous mites (Dicke et al. 1990), Heteroptera (Kessler & Baldwin 2001, 2004; Halitschke et al. 2008) (Fig. 1c) and Coleoptera, and particularly parasitoid Hymenoptera (Turlings, Tumlinson & Lewis 1990; Mattiacci, Dicke & Posthumus 1995; De Moraes et al. 1998). More recently birds have also been shown to be attracted to HIVOCs and use them to find their prey (Mantyläet al. 2008). Leaf injury caused by caterpillar movement (Peiffer et al. 2009) and insect oviposition (Hilker & Meiners 2006; Kopke et al. 2008) can also increase volatile emission and attract members of the third trophic level.
The composition and abundance of volatiles can change dramatically in response to herbivory, and such changes are likely ubiquitous among many, if not all, plant species. Many of these compounds can be perceived by insect olfactory receptors, thus they may provide cues for prey-searching predators and host-searching parasitoids. In contrast to direct resource provisioning, HIVOCs are information, and supposedly function in plant resistance only when inducible, since the HIVOC cue has to be associated with prey/host by the foraging natural enemies (Heil & Karban 2010).
Depending on whether or not predator attraction is primarily mediated through the provision of additional resources or through the provision of information, we should expect significantly different natural selective forces directing their evolutionary trajectories. In the following paragraphs we try to identify the likely factors and agents influencing natural selection on resource- and information-mediated indirect defensive traits.
Plant defensive function of predator attraction
A defensive function of a plant trait is only evident if its expression increases the relative fitness of the plant under herbivory compared to a plant that does not express this trait growing in the same environment (Karban & Baldwin 1997). In order to prove that a trait has an indirect defensive function, we must present data that support the following three hypotheses: (a) Indirect defensive traits make plants more attractive to predators and parasitoids, and increase the residence time and/or number of the natural enemies on a plant compared to a plant not expressing the trait. (b) Increased enemy abundance decreases performance and survival of herbivores. (c) The difference in enemy abundance observed in (b) causes plant genotypes with these traits to show higher fitness compared to plant genotypes without the traits: that is, there is selection on heritable variation in indirect resistance traits. We review empirical support for these hypotheses as well as evidence for coevolution (i.e. reciprocal adaptation) between plants and third trophic level organisms.
Defensive function of resource-providing traits
There is strong evidence that plants benefit from investing in resources for predators. The exclusion of ants from myrmecophytic plants almost always results in dramatic increases of herbivory on the plants (Heil & McKey 2003; Heil 2008). For myrmecophilic plants and their more diverse and ephemeral predator community, the benefits of investing in EFN or food body production are not always obvious. For example, studies on bracken fern (Pteridium aquilinum) and Helichrysum spp. could not find experimental evidence for an EFN-mediated resistance (O’Dowd & Catchpole 1983). The majority of ant-exclusion studies, however, clearly demonstrate a defensive effect of ants using extrafloral nectaries (Chamberlain & Holland 2009; Rosumek et al. 2009) and this interpretation is strongly supported by experimental manipulations of the quantity of EFN on Macaranga spp. (Heil et al. 2001) or lima bean (Phaseolus lunatus) plants (Kost & Heil 2005, 2008). Moreover, predatory ants were identified as the agents of selection in a definitive demonstration of the defensive function of EFN of wild cotton, Gossypium thurberi. When ant visitation was experimentally reduced, increased herbivore abundance caused greater damage and reduced plant fitness. Experimental reduction in EFN provision reduced ant recruitment, increased herbivory and decreased fitness. Moreover, natural variation in EFN production correlated positively with plant fitness (Rudgers & Strauss 2004).
Defensive function of HIVOCS
In contrast to resource-mediated indirect resistance, natural selection on HIVOC traits has never been demonstrated. There are some studies that collectively imply a defensive function of plant volatiles and support hypotheses (a) and (b) above. For example, differential HIVOC emission results in differential attraction of predators and parasitoids (Rasmann et al. 2005; Halitschke et al. 2008). These predators and parasitoids can in turn have significant effects on herbivore performance, herbivore populations and herbivory (De Moraes et al. 1998; Thaler 1999; Kessler & Baldwin 2001, 2004). However, observations of plant fitness benefits of volatile traits are rare (Kost & Heil 2008; Allison & Hare 2009). We propose two major reasons why HIVOC-mediated reductions in herbivory may not result in increased plant fitness. First, the effects of HIVOC-attracted predators or parasitoids on herbivores may not influence plant fitness, perhaps because many of these natural enemies do not immediately kill their hosts (e.g. parasitoids), which results in continued damage to the plant. In addition, plants may simply have a high tolerance for herbivory, which overrides potential fitness effects mediated by attracted predators. Secondly, because of the facultative nature of the attraction, complex interactions within the arthropod community may result in no net effect.
A synthesis of the available studies is difficult because the data are often from simplified agro-ecosystems, where natural selection by natural enemies of herbivores is less likely, because there is usually virtually no genetic variation within a single agricultural field. In other cases, there are missing causal links. For example, in wild tobacco Nicotiana attenuata, the major herbivore, tomato hornworm (Manduca quinquemaculata) has significant negative effects on plant fitness (Kessler & Baldwin 2002a, 2004). Plants emit HIVOCs that are attractive to a major heteropteran predator, Geocoris pallens (Kessler & Baldwin 2001; Halitschke et al. 2008) (Fig. 1c). The predator has significant positive fitness effects for the plant, suggesting that HIVOCs function as indirect defences in this system (Kessler & Baldwin 2001, 2004). However, it is unclear if natural genetic variation in HIVOC emission (Halitschke et al. 2000) is correlated with predator effects on plant fitness. Allison & Hare (2009) proposed and reviewed a number of additional requirements to demonstrate that HIVOCs are adaptive as predator signals (synomones). Specifically, natural enemies should (a) preferentially learn a subset of HIVOCs associated with their host/prey and (b) respond innately to the particular HIVOCs induced by their prey/host. A review of over 450 studies revealed no study testing hypothesis (a) and weak evidence for (b) (Allison & Hare 2009). The latter hypothesis is based on the expectation that a more intimate association (higher dietary specialization) of natural enemies with the plant (here through the herbivore) should result in a higher specialization of the traits mediating the interaction. However, a review of 140 papers on 95 species of parasitoids and herbivores revealed only weak evidence that specialist natural enemies use more specialized volatile cues than generalists, and no significant difference of innate responses between specialists and generalists was found (Steidle & Van Loon 2003).
In conclusion, although we know of a large number of plant species specifically emitting HIVOCs in response to herbivore damage, and an equally large number of natural enemy species responding to these signals, evidence is weak that these HIVOCs in fact function as signals or as indirect defences by the above definitions. One emerging theme however is that the probability of reciprocal natural selection on plant traits that attract natural enemies of their herbivores, and on predators to utilize those resources and/or information, may be higher the more intimate the interaction between plants and the predators/parasitoid of their herbivores.
Evidence for reciprocal natural selection: high intimacy and high specialization
The broad taxonomic distribution and the ontogenetic diversity of resource traits, such as EFN, domatia and FBs (see above) suggest multiple gains and losses of those traits over evolutionary time. For example, in the Southeast Asian genus Macaranga, the ant-associations range from highly specialized to generalized (Fiala et al. 1994). Highly specialized associations have likely evolved two to four times and been lost one to three times (Blattner et al. 2001; Davies et al. 2001). In contrast, a study on 13 Central American Acacia spp. and related genera suggested that specialized myrmecophytes are monophyletic and obligatory ant–ant plant interactions have only evolved once (Heil et al. 2004). However, anatomical and functional differences make independent origins of myrmecophytes within the African Acacia spp. likely (Palmer et al. 2000). Moreover, myrmecophytes are known from plant genera that belong to many different plant families (Cecropia: Moraceae; Piper: Piperaceae etc.). In short, available evidence supports multiple origins and perhaps also multiple losses of obligate myrmecophytism in the angiosperms.
It has been hypothesized that more long-lived, constant and exclusive associations between ants and ant plants are more likely when the ants reside within the plant, which may result in higher specialization of both partners (Heil & McKey 2003). Such higher specialization would include a larger allocation of resources to supply the ants with food and shelter and a higher level of protection by the ants. Both trends have been found in phylogenetic comparisons. For example, myrmecophytic Macaranga species produce more lipids and proteins in their food bodies, while facultative myrmecophiles produce carbohydrate-rich food bodies (Heil et al. 1998). EFN also tends to have a more complex composition in myrmecophytes than in myrmecophilic plants and is secreted constitutively by the obligate Central American Acacia myrmecophytes, whereas the myrmecophilic species of the same group secrete EFN only in response to herbivory (Heil et al. 2004; Heil, Rattke & Boland 2005; Gonzalez-Teuber & Heil 2009). All these observations indirectly support the hypothesis that more intimate mutualistic interactions result in more specialized traits.
In addition to the altered provision of food via food bodies and EFN along a gradient from obligate to facultative ant-plant associations, selection has likely acted on morphological and chemical traits. Prominent examples include preformed domatia in hollow thorns and stems, petioles and leaf pouches, but also olfactory cues that allow a rapid recruitment of ants and, in at least some cases, alternative chemical defence traits (Bronstein, Alarcon & Geber 2006). Although there are some examples of facultatively defended ant plants producing domatia (Maschwitz & Fiala 1995) and/or food bodies (O’Dowd 1982), there is broad support for the hypothesis that a more intimate interaction between ant plants and their ant partners resulted in the evolution of highly specialized traits and a greater investment in these traits. Accordingly, there is far greater evidence for reciprocal natural selection in myrmecophyte than in myrmecophile systems (Bronstein, Alarcon & Geber 2006).
Similarly, the broad taxonomic distribution of HIVOC predator attractants suggests either a very ancient origin or multiple phylogenetic origins of the ability to produce certain compounds and the ability to change relative rates of their emission. HIVOC emission has been shown in at least 13 plant families (Dicke, Van Poecke & De Boer 2003). However, there are a number of significant differences between HIVOC-mediated and resource-mediated indirect resistance. HIVOC emission has been found in virtually every plant species tested. Part of the reason that HIVOCs are universally produced by probably all plant species is that some of the herbivore-induced volatiles are emitted as a direct consequence of chemical breakdown processes in plant tissue ruptured by feeding herbivores (Noordermeer, Veldink & Vliegenthart 2001). Moreover, even the compounds specifically induced and produced de novo after herbivore damage derive from compound classes (e.g. terpenoids and phenylpropanoids) that are commonly expressed in all plant taxa. Therefore HIVOC emission represents a suite of traits that are diverse, ubiquitous and which may be utilized by any organism with an appropriate perception system. Moreover, in contrast to resource-mediated indirect resistance, HIVOCs can have multiple functions other than indirect resistance (e.g. coping with physiological stresses, direct resistance, plant–plant signaling). This potentially increases the number of agents of selection, which would result in more diffuse natural selection. The analysis of the costs of producing a defensive trait can help to evaluate its functionality and reveal evolutionary processes. Therefore we compare potential costs of resource and signal-mediate indirect resistance in the following paragraphs.