The ecology and evolution of induced resistance against herbivores


Correspondence author. E-mail:


1. Induced resistance is now widely accepted as a potent and widespread ecological force although several pieces of the story remain very poorly known. Theory predicts that induced defences should be favoured in variable environments especially when plants can use cues to reliably predict future conditions; however, this idea has not been seriously evaluated for plants.

2. Theory also predicts that plastic, induced defences should be favoured over permanent, constitutive defences if defences are costly and not always needed. This hypothesis has received considerable attention and limited support; resource allocation costs have been difficult to detect although ecological costs of defence may be more common. Recently, priming has emerged as a mechanism that may further reduce costs. Primed plants do not immediately produce the gene products associated with induced resistance but later respond more rapidly and strongly to severe or repeated attacks. It remains to be determined how common priming is.

3. Much of what we know about induced resistance is from the herbivore’s point of view. Induced resistance will be beneficial from the plant’s point of view if herbivores avoid induced plants, but the behavioural responses of herbivores to induced plant variation are still poorly studied.

4. Recent progress in understanding the detailed spatial and temporal extent of induced resistance has improved our appreciation of the phenomenon. Although some induced responses are systemic throughout entire plant individuals, many others have been found to be localized to damaged tissues. Plant vascular architecture and transpiration rates greatly constrain the distribution of vascular cues. Some plants rely on volatile cues that are active over relatively short distances and may be subject to eavesdropping by other plants, herbivores, and carnivores. Similarly, the temporal duration of induced responses may have important consequences on effectiveness although limited information is available concerning lag times before induction and relaxation times following induction.

5. Limited spatial and temporal scales of induced responses make plants more variable from the herbivore’s point of view. Recent work suggests that plant heterogeneity may be difficult for herbivores to cope with although this hypothesis awaits further empirical testing.

Now, 40 years after the early descriptions of induced resistance in plants to insects (Green & Ryan 1972; Benz 1974; Haukioja & Hakala 1975), the importance of this interaction as an ecological force is no longer controversial. For example, induced resistance is included in most ecology textbooks that discuss plant defences or plant–herbivore interactions. Induced resistance has been found to be widespread in plants (Karban & Baldwin 1997) and can produce large effects on herbivore populations (Karban & Carey 1984; Underwood & Rausher 2002) and herbivore communities (Thaler et al. 2001; Ohgushi 2008). Furthermore, recent progress using genetic manipulations has provided more definitive evidence of significant roles for several biochemical mechanisms and signalling pathways in the induction process. For example, tobacco genotypes engineered for high and low levels of induced proteinase inhibitors were respectively more and less resistant to caterpillars (Zavala et al. 2004), and tobacco plants with genetically impaired inducible jasmonate signalling pathways were more susceptible even to herbivores that normally ignore wild type tobacco plants (Paschold, Halitschke & Baldwin 2007).

However, induced plant resistance is not the necessary or ubiquitous outcome of attacks by herbivores. Many plants are damaged by herbivores without becoming measurably more resistant, and some plants become more susceptible to additional attacks by herbivores (Karban & Baldwin 1997:116–119; Nykanen & Koricheva 2004). Plants may also express tolerance to herbivory although we know less about whether tolerance can be induced (Nunez-Farfan, Fornoni & Valverde 2007). This variation in plant responses to damage begs the question: can we predict the outcome of the plant–herbivore interaction? Under what circumstances do we observe plants inducing resistance to herbivores rather than becoming more susceptible?

The distribution of induced resistance

In general, phenotypes should be more plastic in environments that are more variable (Darwin 1881; Bradshaw 1965), and this prediction has been supported by comparative studies of a variety of organisms (e.g. frogs, Van Buskirk 2002). Similar theory has been applied to plasticity of plant defences (Adler & Karban 1994; Karban & Nagasaka 2004; Ito & Sakai 2009). When the identity or severity of damage caused by herbivory varies for a plant either across space or time, a plastic strategy (induced resistance) may be better than one that is fixed (constitutive resistance). In addition, a plastic strategy should be favoured if a plant can adjust its defences appropriately to match its current environment. An appropriate match to current risks will be facilitated when cues about the past environment are reliable predictors of future environments (Karban et al. 1999). Surprisingly, despite hundreds of catalogued examples of induced responses in plants, these long-standing predictions have not been seriously examined for plant and herbivore systems although testing this theory would significantly advance our understanding of the conditions that favour induced defences.

Several authors have argued that environmental conditions may drive or constrain the evolution of various modes of plant defence (Coley, Bryant & Chapin 1985; Grime 2001). Plants that are well defended will have to replace less tissue lost to herbivory. Plants that are adapted to stressful or resource-poor environments generally grow more slowly than those adapted to resource-rich environments. Slower growing plants are predicted to invest more heavily in constitutive defences than rapidly growing plants. Mechanisms responsible for this association include an inability to replace tissues that are consumed by herbivores, and a lack of the machinery or resources needed to mount effective induced defences (Karban & Baldwin 1997). This theory makes intuitive sense and has been supported by some of the studies that have attempted to test it (e.g. Fine et al. 2006), although most of these have not explicitly considered induced vs. constitutive defences. One interesting study compared pairs of congeners in resource-poor glades vs. richer non-glade habitats (Van Zandt 2007). In general, congeners from resource-poor sites grew more slowly and relied more on constitutive rather than induced defences. The generality of these results remains unclear.

These important predictions relating variable and productive environments to induced plant defences are still in need of testing. Other hypotheses involving costs of defences have been proposed to explain induced resistance and these have received more attention.

Induced resistance as a means of reducing costs

Induced resistance has also been theorized to be favoured as a means of saving costs (Rhoades 1979; Karban & Baldwin 1997; Stamp 2003). When herbivores are not present, a plant can allocate resources to growth or reproduction rather than to defence. When herbivores are present, a plant can increase its allocation to defence in a manner that will protect it from the particular herbivores at hand. Allocation costs of plant defences have proven exceedingly difficult to detect under natural field conditions (Koricheva 2002; Agrawal 2005). One reason for this difficulty is that allocation costs are often considered in terms of carbon (C) or nitrogen (N). If plant fitness is not limited by these particular currencies that happen to be easy to measure, then there is no reason to expect that allocation costs measured in C or N will be detectable. Focusing on C or N may reduce our ability to understand the relationship of induced defences to plant fitness and more direct measures of plant fitness are preferred. Plants can invest in many different processes and tissues that ultimately increase lifetime fitness; as a result, there is no necessary correlation between resources spent on defence and those available for the particular fitness correlates that any one study measures (Karban & Baldwin 1997:196–207). For example, tradeoffs were undetectable between induction of defences and female reproduction (a common measure of plant fitness) but were seen when male reproduction was also considered (Agrawal, Strauss & Stout 1999).

In addition, it is becoming increasingly clear that ecological costs of defence may exceed costs measured in terms of resources (Simms 1992; Karban & Baldwin 1997; Strauss et al. 2002). Defending against one herbivore often makes the plant more susceptible to other attackers or ecological threats. For example, induction caused by some pathogens made tomato plants better defended against these pathogens but more vulnerable to caterpillars (Thaler et al. 2002). However, defences can have complicated consequences that affect the interactions of the focal plant with other species and abiotic challenges and the net effects can be difficult to predict (Stout, Thaler & Thomma 2006). For example, defending against herbivores may affect a plant’s attractiveness to pollinators or other mutualists (Adler & Bronstein 2004; Kessler, Diezel & Baldwin 2009).

Priming and costs

Plants that are exposed to damage by herbivores may be ‘primed’ to respond rather than mounting a full induced response (Conrath et al. 2006; Frost et al. 2008). Plants that are primed by previous cues do not express measurable changes in their phenotypes unless they are attacked subsequently by herbivores. If plants that have been primed are subsequently attacked, they respond more rapidly and more strongly than plants that have not been primed. Priming has been known to occur in response to pathogen infection for several decades and has been considered an important part of vertebrate immunity for much longer. However, only in the past few years we have become aware of a similar phenomenon involving plant responses to herbivores (Engelberth et al. 2004). These workers found that maize plants that were not damaged but had been primed by herbivore-induced volatiles responded more quickly and more strongly to caterpillar attack. Priming is relevant to this discussion because this mechanism may allow plants to further reduce costs and to adjust defences to current risks.

Plants become primed in response to cues indicative of herbivores or to low levels of attack. In the event of higher levels of herbivore attack or repeated exposure to cues associated with herbivory, they respond by producing the more costly metabolites required to induce resistance. Priming is assumed to accrue some small costs or else all plants would remain constantly in the primed state. This difference in the benefits and costs caused by priming vs. induced resistance has been best characterized for Arabidopsis infected with pathogens (Van Hulten et al. 2006). Priming caused no measurable costs in terms of growth or seed production for Arabidopsis relative to unprimed controls. By contrast, induction of resistance reduced growth by as much as 71% and reduced seed production by more than 30% relative to controls. This is a convincing example of minimal costs of priming but it involves only one plant species, Arabidopsis attacked by pathogens in the lab, and it would be informative to repeat with other species attacked by herbivores under natural conditions.

There are many potential costs of producing, storing and metabolizing the morphological structures and chemicals used to provide resistance to herbivores (Karban & Baldwin 1997). These processes or their end products may be damaging to the plant as well as its herbivores. Another potential ‘cost’ of induced defences is that the plant is vulnerable to herbivores during the time between initial attack and expression of the induced defence. A constitutively defended plant does not experience this period of vulnerability. Priming can also help to reduce this ‘cost’ since plants may become primed prior to actual herbivore attack in response to cues that predict a high risk of herbivory. Priming allows plants to produce the induced response more rapidly, perhaps because the genes involved in signalling are already up-regulated although the defensive metabolites have not been produced. In summary, plants that are primed but do not actually induce resistance save costs if herbivore attack does not materialize and save time in responding if the attack does occur.

Priming has the potential to shift the cost–benefit ratio in favour of induced rather than constitutive resistance. As such, it would be very informative to know how common priming is in plant–herbivore interactions. Since it was first reported in 2004 for maize resistance against caterpillars (Engelberth et al. 2004), workers have looked for and found evidence for priming in other systems including lima beans (Heil & Silva Bueno 2007), poplars (Frost et al. 2007) and blueberries (Rodriguez-Saona, Rodriguez-Saona & Frost 2009). Cues released by experimentally damaged sagebrush primed neighbouring wild tobacco to induce resistance to some of its herbivores (Kessler et al. 2006). Priming has been shown to induce both ‘direct’ resistance mechanisms and ‘indirect’ defences involving other trophic levels. For example, volatiles from experimentally damaged corn or lima bean plants primed neighbouring conspecifics to produce more volatiles or extrafloral nectar when attacked by herbivores (Heil & Silva Bueno 2007; Ton et al. 2007). The volatiles or extrafloral nectar attracted arthropods that parasitized or preyed upon herbivores. The list of plants for which priming has been demonstrated is still relatively small although it includes a variety of plant families, growth forms, and habitats. In the future we will certainly get a much better sense of whether priming plays a role in most induced responses or only a limited representation of examples.

Induced plant responses and herbivore behaviour

From the plant’s point of view, induced responses will be most cost effective if herbivores are sensitive to plant traits that provide resistance and avoid defended plants. Most examples of induced resistance have been measured from the point of view of the herbivores – reduced herbivore growth rates, survival, etc. However, these ‘antibiotic’ effects on herbivores may not necessarily provide an advantage to the plant and few workers have attempted to determine the expected fitness associated with traits that are presumed to be plant defences. Herbivores that feed on induced plants and require a longer time to complete development may be more damaging to plants than herbivores that grow quickly. Herbivore behaviour is the potential link between herbivore performance and plant fitness although herbivore choice has largely been ignored in our thinking about plant defences (Adler & Grunbaum 1999). Examining herbivore behaviour may help us determine the costs and benefits of induction. What do we know about herbivore choice and movement in response to changes in plant quality?

Early studies of induced responses to herbivory noted that herbivores sometimes moved away from locally damaged plant parts and fed preferentially on undamaged tissues (Edwards & Wratten 1983). These responses can occur before the herbivores have even physically contacted the damaged leaves. For example, plants emit different volatile chemicals depending upon whether they have been damaged or not and herbivores use these volatile profiles as cues to choose sites for oviposition (De Moraes, Mescher & Tumlinson 2001). Neonate caterpillars have also been found to use volatiles induced by conspecific feeding to inform their decisions to stay on suitable host plants (Carroll, Schmelz & Teal 2008).

Plants can limit the damage that they receive from choosy herbivores by getting these herbivores to leave or to be removed by predators and they may also be able to influence herbivores to feed on neighbours with which they compete. For example, hornworm caterpillars are most damaging to their host plants as late instars (older stages). Wild tobacco plants that were attacked by these caterpillars delayed mounting their maximal induced response until hornworms were entering their most destructive stages (Van Dam, Hermenau & Baldwin 2001). At this stage, the estimated cost–benefit ratio favoured caterpillars that left their natal host plant in search of a neighbour of the same plant species.

While we have some (surprisingly few) examples of herbivores that respond to plant quality and make choices that benefit their own performance or those of their offspring, the scale at which they respond to cues seems very important and often unknown. There are examples of herbivores that respond to variation at very local scales; for some systems, chemical and nutritional differences within individual leaves have been found to be large and to have profound effects on herbivore performance (Whitham 1983; Orians, Ardon & Mohammad 2002; Shelton 2005). Variation at this scale has been found to affect the spatial distribution of damage within a plant (Edwards & Wratten 1983; Rodriguez-Saona & Thaler 2005; Underwood, Anderson & Inouye 2005), and this distribution of damage can have as large an impact on plant fitness as the total amount of damage that the herbivore inflicts (Lowman 1982; Watson & Casper 1984; Marquis 1992; Mauricio, Bowers & Bazzaz 1993; Lehtila 1996).

Herbivores may also make choices at coarser scales; entire plants can become systemically resistant to their herbivores and patches of plants may also influence herbivore decisions. For example, plants may become relatively more or less preferred by herbivores depending upon the identity and patch size of their neighbours (associational susceptibility and resistance, reviewed by Barbosa et al. 2009). We have paid far too little attention to the role of behaviour as a mechanism that determines the costs and benefits of induced plant changes and when we have measured behaviour we often have not distinguished processes that occur at different spatial scales.

The spatial extent of induced resistance

Many of the early reports of induced responses were systemic; e.g. experimental damage to one part of the plant induced resistance throughout birch trees (Haukioja & Neuvonen 1985), tomato plants (Pearce et al. 1991), or cotton seedlings (Karban & Carey 1984). However, for each of these model systems, subsequent work suggested that the induced resistance was stronger in some regions of the plant, particularly those close to the site of damage (birches –Tuomi et al. 1988; tomato –Orians, Pomerleau & Ricco 2000; cotton –Karban & Niiho 1995). Many plants are highly sectored such that the exchange of nutrients, secondary chemicals, and hormones that mediate plant–herbivore interactions is limited to those plant tissues that share active vascular connections (Orians & Jones 2001; Schittko & Baldwin 2003). In other words, the limited ‘plumbing’ of plants can severely constrain transfer of the vascular cues and defensive compounds that determine levels of induced resistance (Viswanathan & Thaler 2004; Orians 2005; Rodriguez-Saona & Thaler 2005).

Another limitation of vascular communication is that it may require active transpiration. Woody plants from dry environments are particularly likely to be highly sectored, perhaps as a means of conserving moisture or avoiding hydraulic failure (Waisel, Liphschitz & Kuller 1972; Zanne et al. 2006; Schenk et al. 2008). These plants are poorly equipped to use vascular signals to coordinate induced responses and other physiological processes. These physiological constraints raised the question of whether plants induce resistance using non-vascular signals.

Volatile cues of induced resistance

Recent evidence indicates that some plants use volatile signals to coordinate systemic responses. For example, when a branch of sagebrush is experimentally clipped, other branches on the same plant become more resistant to subsequent herbivore attack, but only if air flow among branches is permitted (Karban et al. 2006). Plastic bags that blocked air flow between the experimentally clipped branch and assay branches on the same plant prevented communication even though the clipped and assay branches were physically connected and had the unrealized potential for vascular communication (Karban et al. 2006). Conversely, air transfer was sufficient to induce resistance between plants that were separated by several metres (Karban et al. 2010). Similar dependence on volatile cues for systemic resistance has been found recently in lima beans (Heil & Silva Bueno 2007), poplars (Frost et al. 2007), and blueberries (Rodriguez-Saona, Rodriguez-Saona & Frost 2009).

Reliance on volatile rather than vascular cues has several possible consequences. First, an external volatile signal becomes public information and other organisms in the vicinity of the damaged plant may respond to this cue. Some of the organisms that respond, such as predators or parasites of the herbivores, may benefit the plant that released the volatile cue (Dicke & van Loon 2000; Kessler & Heil 2011). However, herbivores and other enemies may also respond to the volatile cue and cause more subsequent damage to the emitter plant (refs in Heil & Karban 2010). For example, parasitic dodder has been found to respond to volatile cues to locate suitable host plants (Runyon, Mescher & De Moraes 2006). Neighbours that compete with the plant that emitted the cue may alter their own defences such that the neighbours become well matched for their expected risk of damage. For example, wild tobacco plants in close proximity (within 15 cm) to clipped sagebrush became more resistant to their shared generalist herbivores and experienced increased reproductive success compared with tobacco near unclipped sagebrush (Karban & Maron 2002). A similar benefit has also been found for lima bean tendrils exposed to cues from experimentally induced neighbours (Heil & Silva Bueno 2007).

Secondly, reliance on volatile cues limits the spatial extent over which the cues will be active. For example, cues between sagebrush individuals caused changes in resistance of neighbours that were detectable only over distances up to 60 cm (Karban et al. 2006). Limited signal distance may have many consequences, negative as well as positive. Limited dispersal of volatile signals may make it unlikely that a large plant can communicate with all of its tissues but may also reduce the likelihood that unrelated neighbours will be able to eavesdrop on messages since adjacent leaves often belong to the same individual. Thirdly, vascular cues are constrained by the particular architecture of the vascular connections (Orians 2005). Leaves that do not directly share vascular channels have been found in many plants to have poor physiological integration. Volatile cues allow plants to overcome this limitation. Fourth, volatile cues move very rapidly, which reduces lag times involved in signalling relative to vascular cues (Heil 2009). We currently have evidence for volatile communication from approximately 10 model systems (Heil & Karban 2010) and information about other plant systems is needed to determine the distribution and commonness of volatiles as cues that coordinate induced plant defences.

If volatile cues are commonly used by plants and eavesdropping by neighbouring plants is a significant risk, then selection may favour cue specificity. Both emitting and receiving plants may be sensitive to the specific details of the attacker. We know that different herbivores and different levels of attack cause differential emission of volatiles, although it remains unknown if plants are sensitive to these chemical nuances.

We know that other organisms detect the variation in cues. Parasitic wasps that use volatile cues released by damaged host plants to locate herbivorous prey have been found to exhibit high sensitivity to specific cues. For example, specialized wasps only respond to the plant cues associated with their particular host herbivore; this sensitivity allows them to distinguish the most profitable foraging patches (De Moraes et al. 1998; Hoballah, Tamo & Turlings 2002). In addition, different herbivores cause different responses that affect subsequent herbivory. For example, different induced plant responses to initial attacks by different early season herbivores caused changes that ultimately resulted in predictably different late-season herbivore communities on milkweeds (Van Zandt & Agrawal 2004).

As plant susceptibility to herbivory has a strong genetic component (Karban 1992; Mopper & Strauss 1998; Johnson & Agrawal 2005), families of related individuals may share susceptibilities. Risk of attack should be more highly correlated among close relatives, and it makes sense that they may be particularly sensitive to the cues emitted by their close kin. The volatiles produced by plants that are damaged by herbivores vary considerably among individuals and also have a strong genetic component (Hare 2007; Karban & Shiojiri 2009; Schuman et al. 2009). As we know relatively little about which parts of the volatile profiles are biologically meaningful, it is premature to attempt to evaluate the function of different profiles or the levels of variability among them. It will be very interesting to determine whether plants that are close relatives are more effective communicators than plants that share fewer genes.

The timing of induced resistance

The temporal progression of induced responses will greatly influence their effectiveness. A repeated conclusion of almost all models of induced defence (and cycling populations) is the importance of time-lags (Karban & Baldwin 1997:159–160). One disadvantage of induced defences relative to constitutive defences is that plants experience a period of vulnerability during the time that it takes to activate their induced defences. Both the lag between damage and the onset of defence and the lag between the cessation of damage and the relaxation of defence are critical parameters that determine the effectiveness and the costs of induction.

The precise time course of induction is difficult to characterize since the state of an individual plant at one time is auto-correlated with states at other times. Studies that have characterized the time course suggest that it does not necessarily follow a simple trajectory. Soybeans attacked by Mexican bean beetles became more resistant to subsequent attack after a lag of up to 3 days (Underwood 1998). This induced resistance was relaxed 15 days following the initial damage and after 20 days the plants became more susceptible to attack than uninduced plants. The greater the induced resistance observed over the early portion of the time course, the greater the induced susceptibility observed later in the time course.

We still have relatively few systems for which the complete time course of induction and particularly, relaxation, are known. For those systems that have been examined, relaxation lags are much longer than induction lags. This is true for systems that induce relatively rapidly as well as systems that are much slower. Trifolium repens required 38–51 h after damage to mount a systemic defence but at least 28 days for it to relax (Gomez, van Dijk & Stuefer 2010). Acacia drepanolobium subjected to browsing by large mammals produced larger spines than those produced by unbrowsed trees within approximately 2 months (Young & Okello 1998). However, relaxation for trees protected from browsing was far more gradual than induction and continued for more than 5 years (Young, Stanton & Christian 2003). The time-lag between initial damage and the onset of the induced response is presumably set by the physiological constraints of growing new tissues. In this example involving spines, the time-lag between cessation of damage and relaxation of the induced response is presumably set by the reliability of the cue that risk of damage has passed (Young, Stanton & Christian 2003). In a related Acacia system that had more reliable damage, spine length decreased within 2 years after damage stopped (Gomez & Zamora 2002). Time-lags involving chemical defences will generally be shorter and less dependent on the growth of new tissues although reliability of cues may still be important.

Norm of reaction models of induced resistance

Induced responses are changes in plant phenotype that occur following attack. Plasticity that is dependent on environmental conditions has been represented by norm of reaction diagrams. In the case of plasticity that involves induced resistance against herbivory for plants of a single genotype, two resistance states are represented: induced plants exhibit greater levels of resistance than uninduced control plants. Norm of reaction models emphasize differences in the mean phenotypes of the two categories of plants. When the diagrams include measures of variance (and this should always be the case), that variance generally reflects differences among the means for multiple individuals within a population rather than variation within each individual. Norm of reaction models implicitly ignore variation around the mean resulting from small scale variation among the different organs within an individual.

This is an important shortcoming because plants are made up of repeated, modular units, such as leaves on a branch or branches on a tree (White 1984; de Kroon et al. 2005; Herrera 2009). These modules vary considerably and measures of variability within an individual commonly exceed variability among individuals. Most of this variation within individuals can be explained by plastic responses of modules to environmental conditions (light, resources, damage), interacting with internal controls (hormones) and tissues at different stages of development (Herrera 2009).

Induced responses as a source of heterogeneity

As discussed above, induced responses are often highly localized with little integration at the scale of whole plants. Induced responses that occur as scales smaller than whole plants increase the heterogeneity of plant traits that are important to herbivores in time and space. This heterogeneity occurs over spatial scales ranging from microsites within a leaf to variation among individual plants as well as over temporal scales from minutes to years (see sections above). Several detailed maps of the induction have been instructive. Young tomato plants damaged by different herbivores responded by altering levels of defensive oxidative enzymes measured at the scale of leaflets and leaves (Stout, Workman & Duffey 1996). When the mean levels of enzyme activity increased for whole plants, the induced responses were highly localized so that the intraplant variance increased even more than the mean levels. Some of the spatial pattern was probably due to limitations imposed by vascular connections in these tomato plants (Orians, Pomerleau & Ricco 2000). For example, following damage to leaf 2, induction was noted in leaf 5 but not in adjacent leaf 3; similarly, the side of the leaf closest to the direct vascular connection responded more than the side farther away (Stout, Workman & Duffey 1996). Similarly, leaf veins and midribs of wild tobacco acted as barriers creating small-scale heterogeneity in the defensive jasmonate burst produced by plants attacked by caterpillars (Stork et al. 2009). Glucosinolates involved in defence of wild radish also have been found to respond to damage at very fine scales (Shelton 2005). Caterpillar feeding on one leaf increased glucosinolate concentrations for the entire plant but in an extremely patchy manner. Differences among leaves, and especially among regions within leaves, were greater than variation among whole plant responses. No spatial autocorrelation was found among leaf tissues as close as 1–2 cm apart (Shelton 2005).

The heterogeneity caused by induced responses is predicted by several models to reduce herbivore performance (Adler & Karban 1994; Karban, Agrawal & Mangel 1997; Shelton 2004; Underwood 2004). According to this hypothesis, variability per se may be defensive, rather than any particular phenotypic state. Although this hypothesis has received relatively little attention, several lines of empirical evidence support it. Gypsy moth caterpillars that were reared on diets with a constant mean level of nitrogen suffered reduced pupal masses and extended development as the variation in daily nitrogen increased (Stockhoff 1993). Similarly, armyworm caterpillars reared on a constant diet that included cyanide were able to habituate to that toxin far better than caterpillars that were given diets with variable levels of cyanide even though cyanide concentrations in the variable diets never exceeded the dose in the constant diet (Brattsten et al. 1983).

Variation among bites in space and time is expected to increase the foraging costs experienced by herbivores (Herrera 2009). In addition, non-linear benefits associated with different levels of plant quality will reduce the expected value of plants that vary to herbivores (Karban, Agrawal & Mangel 1997; Ruel & Ayres 1999).

Herbivores may be able to alter their physiologies to match those of their host plants as long as they can predict their host’s ‘defensive phenotype’. Many of the enzymatic systems that herbivorous insects employ to detoxify induced plant secondary chemicals are themselves inducible (Lindroth 1991). Variability in the secondary chemicals that these insect systems must face reduces their effectiveness (Berenbaum & Zangerl 1993, 1996). Corn earworms eavesdrop on the plant signals that induce production of secondary chemicals which allows the caterpillars to activate enzyme systems that can detoxify the induced plant responses (Li, Schuler & Berenbaum 2002). Herbivorous insects can adjust their feeding strategies in an attempt to match the defences of their host plant as long as those defences are predictable. For example, clutches of pipevine swallowtail caterpillars that start on undamaged leaves are able to tolerate their plant’s induced responses. Clutches that are experimentally added to plants that have already begun inducing defences are unable to tolerate them (Fordyce 2006). In these cases, herbivore counter measures to induced plant responses are only effective if the herbivores are able to match their host plants and become ineffective if host quality varies unpredictably.

These are examples of situations in which plant variability is difficult for herbivores. However, reducing herbivore success does not necessarily increase plant fitness. For example, caterpillars that were fed nutritionally poor plants were able to compensate by eating more of their lower quality hosts (Slansky & Feeny 1977). Unfortunately, we know more about the effects of induced plant responses on herbivore success than their effects on plant fitness.

There are many possible benefits to plants of using plastic inducible defences that were outlined by Agrawal & Karban (1999). If heterogeneity is inherently difficult for herbivores (see above), advantages of heterogeneous host plants that are unpredictable to herbivores may occur in ecological time. Heterogeneous host plants may also present inconsistent targets of selection for herbivores over evolutionary time (Whitham & Slobodchikoff 1981; Whitham 1983; Roslin et al. 2006). These two classes of mechanisms are often confused. The first class, which posits that induced defences are more effective in ecological time frames, should be more generally applicable although it has received less attention. The second class, involving a moving target over evolutionary time that herbivores cannot adapt to, is more intuitive and has been more readily accepted although empirical evidence is difficult to collect (as is the case for all evolutionary hypotheses).


Induced defences are important because they are widespread and cause large effects on herbivores. It is worth evaluating their consequences for the plants that deploy them; plant fitness is the best (and one of the most elusive) currencies to measure the relevant costs and benefits. Recent advances in our understanding of the mechanisms of induced responses such as priming and volatile signalling have changed our views of their relative costs and benefits. A few decades ago induced resistance was viewed as an unlikely phenomenon. Although this is no longer the case, there is still a lot that we do not understand about induced resistance. A realization that plant defences are dynamic puts plants on equal footing in terms of sophisticated behaviours with the herbivores that attempt to exploit them. This realization has made possible the nascent fields of plant communication, plant behaviour, and plant immunity. These less well accepted fields are where induced resistance was a short time ago.


I thank Martin Heil, Mikaela Huntzinger, Marc Johnson, Ariel Novoplansky, and anonymous reviewers for comments that improved versions of this manuscript.