The Growth-Differentiation Balance Hypothesis

The growth-differentiation balance (GDB) hypothesis is based on the premise that there is a physiological trade-off between growth and differentiation processes in plants (Loomis, 1953). This trade-off exists because secondary metabolism and structural reinforcement are physiologically constrained in growing cells, and because they divert resources away from the generation of new vegetative tissues. Therefore, plants face a dilemma: to grow fast enough to compete with other plants, or to maintain the defences required to ward off attackers (Herms & Mattson, 1992).

This hypothesis integrates the trade-off between growth and secondary metabolism with responses of net assimilation rate (NAR) and relative growth rate (RGR) to the availability of resources, and it predicts that resource availability (nutrients and water) will exert a parabolic effect on the concentration of secondary metabolites (Herms & Mattson, 1992). NAR reflects the balance between carbon gain in photosynthesis and carbon losses via respiration and other processes, per unit leaf area per unit time. Therefore, NAR integrates environmental effects on net carbon acquisition at the whole plant level over a specified growth period. RGR is largely the product of NAR and leaf area ratio (LAR, the ratio of total leaf area to total plant mass) (Lambers & Poorter, 1992).

The GDB hypothesis predicts that rapidly growing plants will have low levels of secondary metabolites, because production of new leaves is supported by export of assimilates from source leaves, leaving little for synthesis of secondary metabolites. Exposure of plants to moderate levels of nutrient deficiency limits the growth of sink tissues such as new leaves, with little effect on NAR. Thus, photoassimilates that would otherwise be exported to young, rapidly growing leaves accumulate in source leaves instead, where they could be used to produce secondary metabolites. Under conditions of severe nutrient deficiency, where NAR is limited, both growth and the production of secondary metabolites are predicted to decrease, since both processes are carbon limited (Herms & Mattson, 1992).

Herms & Mattson (1992) summarized the GDB hypothesis with a mechanistic conceptual model of the evolution of plant resource allocation patterns (Figure 8.1).

The model incorporates the effects of natural selection by both plant competition and herbivory, with the evolutionary outcome mediated by resource availability. In the model, competition selects for growth, while herbivory selects for resource allocation to secondary metabolism and, as a result, gives rise to variation in life history strategies. Plant species that are growth-dominated have adaptations that optimize growth via minimal investment in defence, such as inducible resistance and secondary metabolites that are highly bioactive at low concentrations. In contrast, differentiation-dominated plant species possess adaptations which optimize the benefits of maximal retention and economy of acquired resources.

Optimal Defence Hypotheses

Thus, selection will favour defence when the benefit of that defence exceeds its cost. The optional defence hypothesis (ODH) was formulated to address the question: ‘in what regions of the plant should the limited quantity of defensive compounds be concentrated?’ Herbivore pressure and the fitness consequences of herbivory are assumed to constitute important evolutionary forces that vary among different parts of the plant. According to the ODH, the value of a plant tissue would be determined by the reduction in plant fitness resulting from the loss of that tissue (McKey, 1974).

For the most part, experimental data support this prediction. For example, in wild parsnip, the reproductive parts of the plant were considered to be at most risk of attack and to be of high value to the plant. Interestingly, these parts exhibited high constitutive levels of the furanocoumarin, xanthotoxin, which was not inducible. In contrast, roots of wild parsnip were at much less at risk of attack and had low constitutive levels of furanocoumarins, but furanocoumarins were highly inducible in these tissues (Zangerl & Rutledge, 1996). The most valuable leaves of Arabidopsis thaliana include young leaves on rosette plants nearing bolting, with the least valuable being old leaves of flowering plants (Brown et al., 2003). Patterns in the expression of some defence proteins expressed in these leaves are supportive of ODH, while others are more supportive of GDB (Barto & Cipollini, 2005).

An interest in understanding the mechanistic basis of invasiveness in plants introduced to novel habitats has motivated the development of several hypotheses that are rooted in optimal defence theory. The evolution of increased competitive ability (EICA) hypothesis is an extension of the ODH, which proposes that when plants are introduced to a new geographical area, relaxation of selective pressure from natural enemies will favour the evolution of reduced defences. This, in turn, will lead to greater allocation of resources to competitively advantageous traits such as growth and reproduction (Blossey & Nötzold, 1995).

Experimental studies with fast-growing annual plants like A. thaliana have indicated that exclusion of herbivores can quickly lead to selective relaxation of defences (Mauricio & Rauscher, 1997). However, despite the potential of this hypothesis to explain the success of invasive exotic plants, tests to date have given mixed results (e.g. Willis et al., 2000; Siemann & Rogers, 2003; Vila et al., 2003; Meyer et al., 2005; Cipollini & Lieurance, 2012). This may relate to the potential benefits that some plant defences can confer beyond herbivore resistance, that may act to maintain defence levels in the face of reduced herbivory, including allelopathic effects on competitors, a hypothesis called the defence-stress-benefit (DSB) hypothesis for which some support has been found (Siemens et al., 2003). Further refinements of EICA include the shifting defence hypothesis (SDH), which predicts loss of some defences, coupled with maintenance or gains in others, depending upon the herbivore community composition (i.e., generalists versus specialists) in the novel range of invasive plants (Doorduin & Vrieling, 2011).

Plant Apparency Hypothesis

On the other hand, specialist herbivores can evolve tolerance to certain defence compounds, and this is more common for qualitative defences than for quantitative defences. Qualitative defences would reduce herbivore growth rates, thereby subjecting the herbivores to increased predation risk.

Although there is some support for the apparency hypothesis, the true apparency of a plant to its herbivores is extremely difficult to quantify and it changes with changing biotic contexts (such as season, surrounding vegetation, etc.). Being apparent can also have some benefits that may counterbalance costs. For example, some volatile chemical traits of plants that contribute to their apparency to herbivores also attract their natural enemies (Halitschke et al., 2008). Apparent plants may also generally grow in less competitive environments, in which induced resistance and tolerance mechanisms may function more optimally.

The Carbon-Nutrient Balance Hypothesis

The carbon-nutrient balance (CNB) hypothesis provides a framework for the influence of carbon and nutrient supply on defence expression in plants (Bryant et al., 1983; Tuomi et al., 1988 1991). According to this hypothesis, the carbon-nutrient ratio of a plant controls the allocation of resources to various functions of the plant, which affects the ability of the plant to express its genetic potential for defence. The hypothesis assumes first that a plant will preferentially invest in carbon-based defences when nutrients limit growth more than photosynthesis; but, in situations in which photosynthesis is limited by factors other than nutrients, the ‘free’ nutrients will be allocated to defence. Second, the hypothesis assumes that herbivory is a primary selective force for constitutive secondary metabolites, and that herbivory is reduced by defences. It does not assume, however, that either the total amount of defence, or the type of defence (e.g. nitrogen-containing vs. non-nitrogen containing compounds), is selected for by herbivory.

Under the umbrella of the CNB hypothesis, a number of predictions are made concerning the way in which stressful environments affect the amount and general type of defence (Bryant et al., 1983). Thus, for plant genotypes with phenotypic plasticity in defence, it is predicted that any effects of resource availability on the carbon-nutrient ratio can result in a change in the total level of defence. For example, fertilizer application or shade would decrease the carbon-nutrient ratio, thereby reducing excess carbon production. This would lead to a reduction in non-nitrogen-containing defences and an increase in the availability of nitrogen for defence. For genotypes with little or no phenotypic plasticity in defence, the prediction is that effects of resource availability on the carbon-nutrient ratio in the plant do not lead to altered levels of defence. Therefore, plants adapted to low resource environments would have low intrinsic growth rates and little capacity for re-growth following herbivory. As a result, this would favour selection for maintaining high levels of defence.

Photosynthetic rates can be reduced when leaf tissue is lost as a result of herbivory and, in turn, this might affect the plant's stores of carbon and nutrients. According to the CNB hypothesis, the resulting alteration in carbon-nutrient balance might then lead to a decline in leaf protein and a non-specific accumulation of non-nitrogen-containing defences (Bryant et al., 1983; Tuomi et al., 1984). However, accumulation of such defences is dependent on the availability of soil nutrients and where carbon is stored in the plant (Tuomi et al., 1988). Therefore, because deciduous trees store their reserves of carbon in roots and stems, leaf damage should lead to marked and long-term increases in non-nitrogen-containing defences if nutrients are limiting. Any increase in these defences would be less marked if nutrient supply was not limiting and the carbon-nutrient balance can be restored quickly to pre-herbivory levels (Stamp, 2003).

It is worth remembering that the CNB predicts that plant species can have some combination of fixed and flexible allocation to defence, which can vary between a completely fixed allocation and a completely flexible allocation (Stamp, 2003). It appears that in some plant species, the level of secondary metabolite production is fixed (Holopainen et al., 1995). Some species or plant populations exhibit much less variation in terpene production in response to changes in the environment than others (Muzika et al., 1989). As a result, any test of the CNB hypothesis should first establish the genetic baseline of defence, i.e. the level and range of defences in plants grown at optimal nutrition and growth rate (Stamp, 2003).

Many experimental studies have examined the CNB hypothesis, and the results are equivocal (see Koricheva et al., 1998; Stamp, 2003). However, because it focused attention on the influence of resources on defences expression, it contributed significantly to the development of the growth rate hypothesis (see next section).

The Growth Rate Hypothesis (Also Known as the Resource Availability Hypothesis)

The growth rate (GR) hypothesis suggests that plants that have evolved in low resource or stressful environments exhibit inherently slower growth rates than plants that have evolved in more productive environments (Grime, 1977 2001; Coley et al., 1985). Consequently, it is predicted that stress-adapted species would evolve particular suites of resistance traits, including higher levels of constitutive resistance and lower levels of induced resistance, than faster-growing species.

The rationale for slow-growing species possessing high levels of constitutive defence is that such species are less able to replace tissue lost to herbivory than faster-growing species, and that they might also lack the metabolic capacity for effective induced resistance (Karban & Baldwin, 1997; Grime, 2001). Moreover, due to more intense competition experienced in productive habitats, compared to stressful habitats (Grime, 1977 2001; Herms & Mattson, 1992), it is predicted that plant species from productive environments should allocate more of their resources to growth than to constitutive defence and, instead, should use inducible defences in order to minimize allocation costs (Coley, 1987; Karban & Baldwin, 1997).

A number of studies provide evidence in support of the GR hypothesis. For example, experiments on saplings of 47 tree species found a negative correlation between growth rate and tannins (Coley, 1988). Long-lived leaves of slow-growing species had higher concentrations of tannins than the short-lived leaves of fast-growing species. A negative correlation was not observed, however, between the rate of herbivore damage and tannin concentration.

Relatively recent work, using species from resource-rich habitats, where competition is likely to be intense, and from other habitats that are likely stressful for plants, found that species from stressful habitats generally grew more slowly than their counterparts from resource-rich habitats (Van Zandt, 2007). Stress-adapted species also tended to have greater levels of constitutive resistance and lower levels of inducible defences than species from the resource-rich environments.

Interestingly, most support for the GR hypothesis has come from studies conducted in tropical habitats, which are expected to have high and consistent herbivore pressure (e.g. Coley, 1987; Bryant et al., 1989; Fine et al., 2006). In contrast, evidence from studies in temperate habitats has been mixed, with some work supporting the GR hypothesis (e.g. Shure & Wilson, 1993; Fraser & Grime, 1999), but many studies failing to provide support (e.g. Almeida-Cortez et al., 1999; Hendriks et al., 1999; Messina et al., 2002). From an evolutionary perspective, it would appear that the effects of limited resource supply on growth are considerably greater than costs due to defence production (Coley, 1987).

The Resource Exchange Model of Plant Defence

Since the central tenets of the aforementioned hypotheses were constructed, a much greater understanding and appreciation of the abundance and role of mutualistic microbes in the life of plants has developed. The vast majority of plants form mutualistic associations with bacterial and fungal partners. These mutualisms are nutritionally based and include mycorrhizal associations and nitrogen-fixing symbioses. The resource contributions and requirements of mutualistic microbes can exert profound changes in patterns of resource allocation in plants, but they were not considered in models of plant defence until recently.

In an attempt to rectify this omission, Vanette & Hunter (2011) proposed the resource exchange model of plant defence (REMPD). In this model, the costs and benefits associated with mutualistic interactions affect plant resource status and allocation to growth and defence. The REMPD model predicts quadratic relationships between mutualist abundance and expression of plant defences. Plant biomass and defence expression are maximized when nutrient exchange between partners is optimal, with the result that the two are positively associated.

The authors tested the model by growing milkweed with two species of arbuscular mycorrhizal fungi, Scutellopsora pellucida and Glomus etunicatum. They found that increasing colonization of milkweed by S. pellucida resulted in quadratic responses in plant growth and defences (latex exudation and production of cardenolides), providing good support for the REMPD model. However, infection by G. etunicatum led to an exponential decline in both plant growth and latex exudation, suggesting that the increasing carbon cost associated with colonization by the mycorrhizal fungus outweighed any nutrient benefits received from the interaction (Vanette and Hunter, 2011). Although infection by G. etunicatum decreased resource availability for allocation to both growth and latex production, these results suggests that growth and defence are coupled, as predicted by the REMPD model.

The authors recognize that defence is not a univariate trait, and that suites of traits may co-occur or trade-off (Rasmann & Agrawal, 2009) and, for example, constitutive and induced resistance, and resistance and tolerance, should be included in the broad definition of defence. They predict that if defence is considered in this broad sense, it will respond in a non-linear fashion to fungal colonization and resource exchange (Vanette and Hunter, 2011).

Costs

As we have seen above, producing defences is an expensive business, requiring energy and metabolites. Minimizing allocation and ecological costs of defence expression remain the central explanation for the evolution of inducible defences (Heil & Baldwin, 2002; Zangerl, 2003; Cipollini et al., 2003; Cipollini & Heil, 2010). In fact, the prevalence of inducible defences itself has been taken as evidence that plant defence traits are costly to produce (Karban & Baldwin, 1997). Inducible defences are produced upon herbivore attack, thus ensuring that energy and resources are only used when needed.

Recent studies have indicated that the diversion of resources away from growth during inducible defence reactions is an active process of reprogramming metabolism, as evidenced by important herbivore-induced hormones such as jasmonic acid directly suppressing genes related to growth (Yan et al., 2007; Bilgin et al., 2010; Yang et al., 2012). Once the attack is over, induced levels of defence are expected to return to baseline levels reasonably quickly, except in the case of inducible physical defences (i.e., thorns, trichomes) that appear on newly grown tissues.

Rapid defence induction was first demonstrated clearly by Green and Ryan in 1972, when they found the induction of proteinase inhibitors (PI) in young leaflets of wounded tomato plants within 12 hours of damage. Their data suggested the rapid passage of a signal out of wounded leaves to unwounded leaves on the same plant. The movement of this signal out of wounded leaves had a half-life of 3.5 hours at most. When leaves were detached immediately after wounding, there was no PI accumulation in unwounded leaves but, when they were detached ten hours or more following wounding, there was no effect on PI accumulation in unwounded leaves (Green & Ryan, 1972).

Wounding of first (youngest) leaves of Vicia faba was shown to induce resistance to rust infection in unwounded second leaves (Walters et al., 2006). Not only was this effect rapid, so too was the accumulation of antifungal trihydroxy oxylipins in unwounded second leaves. In Nicotiana attenuata, the movement of a signal that leads to the amplification of jasmonate in response to oral secretions of Manduca sexta larvae was shown to be too rapid (> 3 mm/min) for larvae to circumvent induction by consuming signal-producing tissue (Schittko et al., 2000). Rapidly induced chemical defences include not only direct defences, but also volatile chemicals, extrafloral nectar and other mechanisms that serve as indirect defences by attracting bodyguards in the form of predators and parasitoids that attack herbivores (Dicke et al., 2003).

Possession of both constitutive and inducible defence strategies would appear to be redundant, and it is predicted that, among plant species, there should be a trade-off between levels of constitutive and inducible defences. Although some experimental data have provided support for this prediction, in many cases this prediction has not held true (see Karban & Baldwin, 1997). This might reflect the evolutionary and ecological implications of possessing a system of defence that needs to deal with a range of different attackers, but which also incurs costs (Agrawal & Karban, 1999; Agrawal, 2000).

Although induced defence may be less costly than constitutive defence, inducibility will still incur costs. Costs of inducible defence can be estimated, but measuring them is difficult because they are not always manifested as detectable reductions in plant fitness (Zangerl, 2003). Plants can achieve their fitness in different ways, so some plants might be well-defended or tolerant of herbivory (or disease), or they might be good competitors but poor at defending themselves (Fineblum & Rausher, 1995; Cipollini & Bergelson, 2001).

Therefore, defence costs are entangled with the costs and benefits of alternative strategies of enhancing fitness (Simms & Triplett, 1994) and, as Zangerl (2003) points out, disentangling these costs can only be achieved using isogenic lines, which are rarely available for wild plants. Nevertheless, numerous studies have attempted to measure the costs of inducible defence, and we will examine some of these studies below (see section 2.2).

Targeting of Inducible Defences

It is likely that the broad range of plant defences will be differentially effective against different types of attackers. Ensuring that the most appropriate defences are induced when a plant is attacked by a particular herbivore or pathogen could, therefore, be highly beneficial. If all of the defences to a particular class of attacker (e.g. chewing herbivores) are positively correlated, the cost savings of not producing the defence until required would be the main factor favouring inducibility. If the defences are not correlated, inducing only those defences that would be effective against the particular attacker would minimize costs. If, however, the defences are negatively correlated, the situation is complicated by the existence of an additional cost – that associated with increased susceptibility to a different herbivore. Here, inducing the defence only when a susceptible herbivore is attacking the plant minimizes both the cost of the defence and the cost of increased susceptibility to other herbivores (Zangerl, 2003). This requires the means to recognise attack by different types of herbivore. Attack recognition is mediated by herbivore-associated molecular patterns (Bonaventure, 2014; and see Chapter Herbivore Oral Secretions are the First Line of Protection Against Plant‐induced Defences).

While several positive and even synergistic interactions between plant defences and natural enemies of insect herbivores have been found (e.g. Hare, 1992; Duffey et al., 1995), other research has demonstrated that plant defence can adversely affect parasitoids or predators of herbivorous insects (e.g. Krips et al., 1999; Havill & Raffa, 2000). For example, parasitoids suffered high mortality due to plant trichomes on wild tomato (Kauffman & Kennedy, 1989), and consumption of herbivores feeding on resistant plants resulted in decreased survival, fecundity and developmental rates of predators and parasites (Barbosa et al., 1991; Stamp et al., 1997). Inducibility of defences might be favoured as a strategy to reduce such negative impacts of plant defence on predators and parasitoids of herbivorous insects.

Plants can emit volatile signals to attract predators and parasitoids of herbivorous insects (see Chapter Plant‐Mediated Interactions among Insects within a Community Ecological Perspective). For example, feeding by spider mites on lima bean induces the emission of several terpenoids and methyl salicylate, and these volatiles attract the predatory mite Phytoseiulus persimilis (Van den Boom et al., 2004). Arthropod predators and parasitoids are known to associate non-host cues with the presence of hosts (e.g. Turlings et al., 1993; Hu & Mitchell, 2001; Halitschke et al., 2008). The cost of constitutive expression of these volatile signals is the likelihood that predators and parasitoids would learn to ignore signals that provide no useful information, because herbivorous insects will not be there all of the time. In contrast, emission of the signals only when the plant is under attack provides the predators with useful information, alerting them to the presence of herbivorous insects (Vet et al., 1990).

Dispersal of Damage

Localized, as opposed to systemic, induction of defences might be beneficial if it results in herbivorous insects moving around the plant, thereby dispersing their damage, or moving to neighbouring plants (Edwards & Wratten, 1983; Van Dam et al., 2001). In fact, there is some evidence that dispersed damage has less effect on plant fitness than concentrated damage (Marquis, 1988; Mauricio et al., 1993; Meyer, 1998), although it is not clear if this is a more widely applicable result. This could be one argument for why even slow-growing and apparent plants, like long-lived trees, can still have inducible defences. However, dispersed damage might not always equate to less impact. For example, insects moving around the plant might also be transmitting bacterial, fungal and viral pathogens (Garnier et al., 2001).

But what about evidence that localized damage influences insect feeding patterns? Although this is more difficult to determine, some workers did find that herbivores were likely to move away from damage sites on leaves (Bergelson et al., 1986). The lack of concentrated herbivore damage observed on many trees (e.g. Marquis, 1988), is suggestive that localized induced responses forced herbivores to move away from damage sites.

Elicitation Studies: The Approaches

If induced resistance is costly, induced plants are expected to exhibit lower fitness than uninduced plants in the absence of herbivores. Manipulations of inducible resistance have provided some of the best examples of both costs and benefits of plant resistance, in part because of the ability to manipulate the defence phenotype of single plant genotypes in controlled settings (Cipollini & Heil, 2010). Plants can be induced by herbivore feeding or simulations that typically involve some combination of mechanical wounding, herbivore elicitors and plant hormones. Although plant responses to natural and artificial elicitation can be similar in many cases (Thaler et al., 1996), induction by any of these means will often cause a different set of changes in the plant. For example, the response to mechanical damage can be different from the response to damage inflicted by herbivores, due to differences in the pattern (see Chapter Plant Transcriptomic Responses to Herbivory) and timing of damage and the lack of herbivore salivary elicitors in mechanically damaged plants. When tissue is mechanically damaged in a way closely resembling herbivore damage, however, similarities in some induced responses become more apparent (Mithöfer et al., 2005).

Tissue loss during the induction event comes with its own fitness consequences, termed ‘induction costs’ (Cipollini and Heil, 2010), and this needs to be controlled for when estimating costs of resistance with mechanical or herbivore damage. Specificity in induction also exists among herbivores (Agrawal, 2002), sometimes related to the degree of specialization of the herbivore on the host or to feeding mode (see Bidart-Bouzat & Kliebenstein, 2011), suggesting that no single approach may be able to capture the average induced plant response. While induced plants typically display increases in a number of potential defences, specificity in induction among damage types can be exploited to assess costs and benefits of different induced defence traits in the same plant genotypes (Bjorkman et al., 2008).

Defence hormones or other chemical elicitors have been used to explore costs of resistance for nearly 25 years (Brown, 1988). Use of chemical elicitors enables careful control over tissue loss and issues with spatial and temporal variation in induction that may be due, for example, to differential herbivore feeding rates on different plant tissues or genotypes. However, it can be difficult to simulate herbivory adequately, because of the lack of plant- and herbivore-derived elicitors in wounds, variation in chemical uptake through leaves or roots, metabolism, inappropriate localization and other methodological issues.

Since its identification as an important mediator of inducible defence pathways (Farmer & Ryan, 1990), jasmonic acid and its methyl ester have been used in numerous studies of costs of induced resistance (Baldwin, 1998). Jasmonates have been applied in cost-benefit studies in various ways, e.g. through foliar sprays, root drenches, and lanolin pastes (Baldwin, 1998; Cipollini, 2002). Methyl jasmonate has been used in many studies, in part because of its high absorption rates and its volatility, but evidence indicates that it must be de-esterified to jasmonic acid and conjugated to isoleucine in order to function in induced resistance (Stitz et al., 2011).

Jasmonic acid and its conjugates act downstream of herbivore elicitors (Halitschke et al., 2001). Thus, their use to examine costs of resistance skips a number of potentially important steps that may incur costs (some of which may be unique to inducible resistance versus constitutive resistance mechanisms). In addition to increasing resistance, the use of defence hormones may cause a multitude of traits other than resistance to change (e.g. developmental traits, photosynthesis, tolerance, source-sink relationships). However, natural induction may also cause these sorts of traits to change (Hermsmeier et al., 2001; Bilgin et al., 2010), which are all part of the mechanisms generating pleiotropic costs of induced resistance in plants.

A number of herbivore salivary factors that both upregulate and downregulate direct and indirect defences have been identified (Bonaventure, 2014; Felton et al., 2014). Some of the best approaches to manipulating induced defences have used minimal mechanical damage (e.g. pinpricks) to liberate plant-derived elicitors while minimizing leaf area loss, coupled with the application of jasmonates and/or herbivore elicitors (Halitschke et al., 2001; Voeckel et al., 2001).

Genetic Variation Studies: The Approaches

To test for the existence of costs, some studies attempt to demonstrate genetic variation in defence traits with a subsequent establishment of negative correlations between defence level and plant fitness in the absence of herbivores. This approach was used in some of the first studies of costs of constitutive defence traits. The advantage of this approach is that naturally occurring variation in plant defence can be exploited, and variation can be found and studied in even long-lived woody plants by using clones. However, many traits other than resistance can vary among the different plant genotypes used in such studies, so it is difficult to assign unequivocally the mechanism of variation in fitness to defence production. Costs may be more difficult to detect in natural genotypes in which they have been optimized by selection. Better experimental control and partitioning of variance can be achieved using plants derived from quantitative genetics breeding designs or inbred lines. While reducing genetic variability, these studies can be subject to fitness effects that are caused by genetic combinations that would not be naturally maintained.

Artificial Selection, Mutant and Transformant Studies: The Approaches

As the goal is to attribute fitness consequences to polymorphisms at resistance loci, rather than other fitness-related loci, comparisons of susceptible and resistant genotypes need to control for genetic background (Bergelson & Purrington, 1996; Strauss et al., 2002; Vila-Aiub et al., 2011). One approach that has been taken is to alter the levels of certain defence traits through artificial selection designs that specifically target defence traits. One disadvantage of this technique is the speed at which experimental material can be produced (along with issues such as genetic linkage of traits), and the successful attempts at this approach have been restricted mostly to rapid-cycling annual plants (Agren & Schemske, 1993; Stowe, 1998; Stowe & Marquis, 2011).

Control over genetic background can also be achieved through the introgression of a resistance allele into a susceptible background, followed by repeated backcrosses to produce near-isogenic lines, or through genetic engineering of specific defence traits. Natural or chemically-induced mutants that over- or underexpress specific defence traits or key pathway enzymes have been identified in several species, most importantly A. thaliana. These mutants have been used more frequently to explore potential benefits of defence expression, especially to pathogen resistance, but they have also been employed in some cost of resistance studies (Cipollini, 2002; Züst et al., 2011). Furthermore, studies of costs of and benefits of resistance have benefited both from over-expression of genes for specific defence traits or inducible defence pathways and from silencing of important genes (Steppuhn et al., 2004). Still, these sorts of studies have been restricted to fast-growing species with rapid generation times that are easily manipulated with molecular techniques.

Elicitation Studies: The Evidence

Some of the earliest attempts at using either natural or artificial induction to estimate costs of resistance failed to detect them. For example, in the first study to assess costs of induced resistance, Brown (1998) used chitin injections to induce trypsin inhibitor production in Lycopersicon esculentum and failed to detect a significant cost to growth or reproduction across a range of nitrogen levels. However, the fact that the plants were nearly two months old before induction occurred was raised as a possible explanation for the lack of an observable effect.

In a natural experiment in the field, Karban (1993) failed to find a negative impact on fitness of induced resistance caused by early season herbivory in Gossypium thurberi, but no defence traits were measured; also, such field studies can be confounded by unknown benefits of defence trait induction obscuring the detection of costs. However, induction of nicotine by mechanical wounding in Nicotiana sylvestris was used by Baldwin et al. (1990) in one of the first examples showing costs of induced defence. Although reporting bias may play a role, the majority of studies of costs of induced resistance have since reported significant direct or ecological costs of resistance (Cipollini & Heil, 2010).

Most studies of physiological or allocation costs of inducible defence have been done on fast-growing herbaceous plants in a few families, the advantage of this being that fast-growing plants offer the ability to examine quickly relationships between defence and reproduction. In the Solanaceae, for example, exogenous application of jasmonates have been used to reveal costs of induced defence in N. attenuata (Baldwin, 1998), Lycopersicum esculentum (Redman et al., 2001), Solanum carolinense (Walls et al., 2005) and Brugmansia suaveolens (Alves et al., 2007).

Similar studies have been done with several species in the Brassicaceae, including Raphanus raphanistrum (Agrawal et al., 1999), Brassica kaber (Cipollini & Sipe, 2001), A. thaliana (Cipollini, 2002), and Alliaria petiolata (Cipollini & Lieurance, 2012). Accamando & Cronin (2012) recently added soybean, a member of the Fabaceae, to the list of studies on costs of jasmonate-induced resistance. A rare study on a plant in the Umbelliferae, Pastinaca sativa (Zangerl et al., 1997), utilized minimal mechanical damage to reveal respiratory costs of wound-inducible furanocoumarins.

In these studies, induced plants generally displayed upregulation of at least some chemical and/or physical defence traits and generally grew more slowly, reached smaller sizes, had lower male or female fitness (or both) and had delayed development, compared to uninduced plants. In most cases, the benefits of induced resistance were assumed, but clear fitness benefits of induced resistance in natural settings have been revealed in a few studies (Baldwin, 1998; Agrawal et al., 1999). In some cases, costs of induced responses were larger under resource limitation due to plant competition or nutrient deprivation (Baldwin, 1998; van Dam & Baldwin, 1998; Cipollini & Lieurance, 2012), but this was not always the case (Cipollini, 2002; Walls et al., 2005; and see section 3.1). In a study using natural herbivory, shifts in allocation from roots to above-ground biomass (rather than changes in total biomass) were detected in induced Lepidium virginicum plants, but only at high plant density (Agrawal, 2000).

Studies of costs of induced resistance have only recently included woody plants, but such studies have not yet assessed effects on reproduction. Bjorkmann et al. (2008) separated the costs of trichome production from those of leaf area removal in Salix cinerea by comparing the difference in growth of plants exposed to beetle and mechanical damage. Small amounts of beetle damage induced increases in trichome production and had significant costs to growth, whereas an equivalent amount of mechanical damage had no effect on trichome production and much smaller effects on growth. Sampedro et al. (2011) showed that costs of methyl jasmonate-induced increases in foliar phenolics on growth in half-sib families of Pinus pinaster only appeared under low soil phosphorus conditions, where the induced production of these traits was increased.

Plants that constitutively overproduce defence signals have been used to circumvent some potential criticisms of external manipulations of inducible defences. For example, Corrado et al. (2011) used a single line of Lycopersicon esculentum that overexpresses a gene for prosystemin, a polypeptide signal involved in the upregulation of jasmonate-dependent inducible responses. These plants constitutively expressed normally wound-inducible responses, and they also grew more slowly, flowered later and produced fruits with significantly fewer seeds than wild type plants. Prosystemin-overexpressing plants also photosynthesized at a lower rate than wild type plants, which is consistent with studies using exogenous jasmonates.

Plants that overproduce methyl jasmonate via overexpression of jasmonic acid : carboxy methyltransferase (JMT) have been used to address similar questions. In A. thaliana, transgenic JMT plants constitutively express a number of normally inducible defence responses, including a variety of inducible defence proteins and heightened resistance to some pathogens and herbivores (Seo et al., 2001; Jung et al., 2007). Across several independently transformed lines, JMT Arabidopsis plants had lower total seed mass, reduced seed germination and a substantial delay in the onset of reproduction compared to empty-vector control plants, suggesting that constitutive expression of methyl jasmonate-mediated responses was costly in this species (Cipollini, 2007; Cipollini, 2010). However, transgenic N. attenuata overexpressing JMT were compromised in the production of several defence metabolites and were more susceptible to various herbivores (Stitz et al., 2011). JMT tobacco plants had reduced seed capsule production, indicative of a cost, but this effect was caused by impaired self-pollination in these plants rather than allocation costs, an effect not apparent in JMT Arabidopsis.

These contrasting results for the consequences of methyl jasmonate overproduction suggest significant differences in how defences and other traits are mediated by jasmonate and its conjugates in these species. Therefore, a thorough characterization of elicitor effects is necessary when using elicitation as a tool to study costs and benefits of resistance. Importantly, jasmonic acid and methyl jasmonate may not be interchangeable as experimental tools, and they may not function the same way in all species.

Genetic Variation Studies: The Evidence

Some of the earliest studies on costs of resistance exploited naturally occurring variation among plant populations or genotypes and found, for example, a negative correlation between tannin concentrations and plant growth rate for species such as Cecropia peltata (Coley, 1986), Psychotria horizontalis (Sagers & Coley, 1995) and Bauhinia brevipes (Cornelissen & Fernandes, 2001). However, similar attempts failed to reveal significant costs of resistance or resistance traits in Ipomoea purpurea (Simms & Rausher, 1987 1989) and in several species in the Asteraceae (Almeida-Cortez et al., 1999).

In a series of studies that illustrates some of the inconsistencies in results that can be obtained using this approach, Osier & Lindroth (2006) showed negative correlations between foliar phenolic glycoside concentrations and growth of Populus tremuloides genotypes that were not apparent under conditions of high nutrients and high light. Using several of the same genotypes, along with some others, Stevens et al. (2007) reported the existence of a negative genetic correlation between foliar phenolic production (including phenolic glycosides) and growth in Populus tremuloides genotypes that was only apparent under conditions of high nutrients and high light. However, ontogenetic changes in defence expression (Donaldson et al., 2006), along with the use of trees of different ages in these studies, may have led to such apparently conflicting results. These sorts of effects can only be revealed by repeatedly measuring relationships between plant defence and growth in time course studies.

Inbred lines have been used on numerous occasions to study costs of resistance, with a high degree of success. Studies with inbred lines of A. thaliana that varied quantitatively in trichome density and total glucosinolate concentration (Mauricio, 1998) and Datura wrightii lines that varied qualitatively in glandular trichome production (Elle et al., 1999), each revealed substantial costs of resistance. Similar relationships between fitness and trichome densities were found recently by Sletvold et al. (2010) in half-sib families of Arabidopsis lyrata. Holeski et al. (2010) found a negative genetic correlation between constitutive and induced trichome production in inbred lines of Mimulus guttatus, evidence of a trade-off among defence strategies. Zangerl & Berenbaum (1997) exploited heritable variation in constitutive furanocoumarin content in seeds of half-sib families of Pastinaca sativa to demonstrate significant fitness costs of these compounds. In their study of half-sib families of Pinus pinaster, Sampedro et al. (2011) found a negative genetic correlation between constitutive expression of stem diterpenes and growth. In one of the only studies to assess costs of ‘inducibility’, half-sib families of wild radish exhibited genetic variation in the inducibility of glucosinolates, and more plastic families had lower fitness than less plastic families in the absence of herbivory (Agrawal et al., 2002).

In a study that used a diallel breeding design, Ivey et al. (2009) found additive genetic variation for spittlebug biomass among Mimulus guttatus plants and also found that more resistant genotypes were smaller in the absence of spittlebugs, indicating a potential cost of resistance. A similar negative relationship was found between fitness and resistance in a study of goldenrod genets (Hakes & Cronin, 2011). In each of these studies, resistance was measured indirectly through herbivore performance, and the inverse of herbivore damage and either plant biomass or inflorescence biomass was used as proxies for plant fitness, respectively.

Yet another study that made use of potato cultivars differing in trichome densities found leafhopper damage to be negatively correlated with trichome density, but did not find trichomes to be correlated with potato yield in the absence of leafhoppers (Kaplan et al., 2009). This apparent lack of allocation costs may have been a consequence of insufficient trichome variation, or the contribution of unmeasured variables on the net effect of trichomes. In contrast, glandular trichome production incurs significant costs in Datura wrightii plants as glandular trichome-producing genotypes (sticky plants) were found to produce less seeds per volume of canopy area than non-glandular trichome producing genotypes (velvety plants) (Hare et al., 2003).

Natural variation in proteinase inhibitors and herbivore-induced volatiles, two potent plant defences that may work synergistically, led to the discovery of costs associated with trypsin proteinase inhibitor (TPI) production in N. attenuata. TPI-deficient genotypes produced more seed capsules than competing TPI producing-genotypes, and elicitation by jasmonate decreased capsule production more in TPI-producing, as opposed to TPI-deficient genotypes (Glawe et al., 2003).

Artificial Selection, Mutant and Transformant Studies: The Evidence

Agren & Schemske (1993) used artificial selection to produce lines of Brassica rapa with both increased and decreased foliar trichome densities, and found positive phenotypic and genetic correlations between trichome density and flowering time. In a subsequent study, delayed flowering was again observed in high trichome lines, indicative of a cost to reproduction, but high lines actually grew larger and produced more flowers than low trichome lines by the end of the experiment (Agren & Schemske, 1993). Stowe (1998) used a similar breeding design to produce lines of B. rapa with high and low foliar glucosinolate concentrations. High glucosinolate lines produced fewer flowers with fewer seeds per fruit than lines selected for low glucosinolate concentrations (Stowe & Marquis, 2011). High glucosinolate lines also compensated more weakly for herbivore damage than did low glucosinolate lines, indicative of a trade-off between defence and tolerance (Stowe, 1998).

A variety of A. thaliana defence pathway mutants were used by Cipollini (2002) to explore costs of constitutive and induced defences. In this study, mutants that either lack salicylate or jasmonates, or are insensitive to some of their effects, generally had high seed production, while a mutant that overexpresses salicylic acid had a dwarf phenotype and low seed production. Furthermore, reduction in fitness of plants induced with either salicylate or jasmonate correlated with the degree of induction of some marker defences. A. thaliana knock-out mutants, with altered trichome and glucosinolate production, experienced increased size-standardized growth rates compared to wild type plants, which translated into a fitness advantage for the myb28 mutant when grown in competition (Züst et al., 2011).

A species for which costs and benefits of traits involved in defence and pollination have been studied extensively with transgenic plants is N. attenuata. For example, comparisons between TPI-producing genotypes and isogenic lines that had antisense-mediated reductions in TPI production, and between TPI-deficient genotypes and isogenic lines that had TPI production restored, revealed TPI production to incur costs in competing N. attenuata plants (Zavala et al., 2004). Importantly, TPI-mediated decreases in the performance of M. sexta larvae translated into a fitness benefit for the plants (Zavala & Baldwin, 2004).

Using lines independently silenced for the production of putrescine methyl transferase (pmt), and thus having up to 95% reduced constitutive and induced nicotine levels, Steppuhn et al. (2004) established nicotine's defensive function in nature; PMT-silenced plants were preferred in herbivore choice tests and experienced less damage from native herbivores. Also, when using genetic techniques to manipulate both nicotine and TPI production in two N. attenuata accessions, it was demonstrated that the preference of mammalian herbivores for nicotine and the preference of flea beetles for the absence of TPIs depended on the genetic background in which these defences were or were not produced, respectively. Similarly, fitness consequences of metabolite manipulations in the presence of herbivores depended on plant genotype (Steppuhn et al., 2008).

Further experiments with nicotine-silenced plants and plants blocked in the synthesis of the floral attractant benzyl acetone demonstrated that both nicotine and benzyl acetone were required to maximize flower visitation by native pollinators, as well as capsule production and seed siring in emasculated flowers (Kessler et al., 2008).

These examples illustrate the great potential of transgenic tools to advance cost-benefit studies of resistance traits. However, such studies need to take precautions to control for the transformation procedure, e.g. by using empty vector controls (but see Schwachtje et al., 2008) and by comparing several independently transformed lines to account for positional effects. However, positional effects may lead to gradients in the manipulated phenotype which may be desirable in ecological studies (Schwachtje et al., 2008).