The cost-benefit paradigm is central to functional biology and to ecological and evolutionary theory because fitness costs and benefits associated with a trait determine its equilibrium value in a population. If the trait offers fitness benefits to the population rather than costs, then selection should lead to beneficial allele(s) being fixed, which reduces variability 1. Alternatively, when the fitness benefit of the trait also has a cost, intermediate frequencies of the trait may be favored because the benefit varies 123. For example, resistance against natural enemies has costs as well as obvious benefits for fitness, as has been shown in insect-parasite, insect-parasitoid, plant-pathogen and plant-insect systems 4567.

Herbivores can reduce seed production and other correlates of plant fitness, and this reduction can result in natural selection for either constitutively expressed or inducible plant defenses 8910. Current theory predicts that one benefit of induced defenses is to allow a plant to optimize its allocation of limiting resources to defense, growth, and reproduction 9. Although defenses might benefit plants in the presence of herbivores, plant resistance to herbivores can be costly in the absence of enemies, and inducible expression of resistance traits allow plants to forgo or, to pay the potential fitness cost of resistance traits when they are needed 3511121314.

Evidence for the existence of resistance costs and benefits from studies using plant species with constitutive and inducible defenses is increasing 3141516. Experiments on natural populations of plants as diverse as Arabidopsis, Ipomea, Pastinaca and Trifolium have provided evidence for costs 217181920. These experiments typically use quantitative genetic approaches to determine whether, in the absence of enemies, fitness and resistance are inversely correlated. However, attribution of fitness consequences to the expression of a particular defense trait in an environment either with or without herbivory is difficult, because genes that control the expression of defensive traits may have pleiotropic effects 21. Ideally, one should assess the costs and benefits of inducible defenses in plants that differ only in the expression of genes that control (induced) resistance but are otherwise genetically identical 15. Transformation technology provides a means of manipulating traits with unparalleled precision. Although the benefits of plant traits that provide resistance against herbivores are expected to equal or exceed their cost when the system is at evolutionary equilibrium 22232425, very few direct tests have been done. While costs and putative benefits of defense traits have been studied in separate experiments, their currencies are usually not comparable (i.e., plant fitness for the cost; herbivore performance for the benefits). Tests of the cost-benefit model using the same currency are few 5 and these studies do not consider the heterogeneity of the plant.

Ecological interactions can be viewed as the net outcome of a series of cost-benefit optimizations in which both players respond to the variability in each others' defense traits. For example, there is enormous within-plant heterogeneity of defensive secondary metabolites. This heterogeneity could motivate within-plant movement of herbivores, so that they eat leaves of low fitness value rather than leaves of high fitness value, or it could motivate herbivores to move off plants and onto neighboring competitors 2627. Herbivores, in turn, can both readjust their metabolism to cope with the secondary metabolites as well as adjust their feeding positions to maximize their performance 272829. We present here a cost-benefit analysis of a plant-insect interaction in which the costs and benefits of a defensive protein are evaluated in the currency of plant fitness.

Nicotiana attenuata [Torr. Ex Wats. (synonymous with Nicotiana torreyana Nelson and Macbr.)], a postfire annual native tobacco inhabiting the Great Basin Desert, has a number of well-described herbivore-induced direct and indirect defenses 30, which increase the fitness of plants under attack in natural populations 531. Trypsin proteinase inhibitors (TPI) play an important defensive role in addition to nicotine 30. We isolated cDNA from N. attenuata that coded for a TPI precursor belonging to the potato PI-II family with a 7-repeat TPI domain. The normal constitutive expression of this gene increases 4-fold after herbivore attack 3233.

The elicitation of TPI expression in N. attenuata varies with ontogeny and leaf age 34, as is true for nicotine 35 and volatile emissions 36. The within-plant pattern of systemic TPI induction at the rosette stage of growth suggests that the signal(s) triggering remote TPI induction follows a source-sink relationship; regardless of ontogenetic stage, if young sink leaves are damaged, TPI levels increase only in the attacked leaf, while older leaves are less sensitive to leaf damage and produce a less intense response in the attacked leaf, the systemic responses in young leaves is dramatic 34. The spatial and temporal variability in N. attenuata's ability to deploy certain defenses against herbivores can be correlated with the relative fitness values of leaves growing at particular nodes. Removal of young and mature leaves at the elongation stage in N. sylvestris had a greater negative effect on fitness than did the removal of old leaves, but not at either the rosette or flowering stages, demonstrating the different fitness values of leaves growing at different nodes on a plant. Damage to younger leaves increases nicotine contents more than damage to older leaves does, suggesting that defense allocation is proportional to the fitness value of the tissue, as predicted by Optimal Defense (OD) theory 10233537.

Manduca sexta, a specialized lepidopteran herbivore, prefers elongating N. attenuata plants to rosette-stage plants for oviposition and places eggs on leaves in the middle section of the stem (from S1 to S3; Figure 1; 38). TPIs of N. attenuata leaves reduce the growth of M. sexta larvae 3233. However, insects may adapt to high TPI levels, replacing the inhibited trypsin with the secretion of trypsins that are insensitive to the particular TPIs of the diet 2939. Intra-plant movement of the first instar larvae is very rare but common in the second-to fourth-larval instars 38. Larger instars are heavier and more difficult to handle by insect predators, and also better able to defend themselves against an attack in their natural environment by vigorous movement and regurgitation (A. Kessler, personal communication). Larvae are particularly sensitive to jasmonate-induced defenses during the third instar (approx. 11 days after hatching), and can be motivated to move between adjacently growing plants 26 by the plant's induced defense. When M. sexta larvae were placed on MeJA-induced plants, larvae left the induced plants 1–3 days earlier than did larvae placed on uninduced plants, which dramatically reduced the leaf area consumed and larval weight gain 40.

TPI expression in N. attenuata is known to decrease lifetime seed production in unattacked but competing plants 32 and to decrease M. sexta performance in attacked plants 32. Whether the TPI-mediated decrease in herbivore performance translates into a fitness benefit for the plant is unknown. In other systems, plants expressing high PI levels caused herbivores to grow more slowly, but they compensated by eating more tissue, a potential fitness detriment for the plant 41. Here we provide a critical test of whether the fitness benefits of TPI expression outweigh their costs.

We compare lifetime seed production of N. attenuata genotypes with either low or no TPI production to that of TPI-producing genotypes on which M. sexta larvae were allowed to feed freely for 11 days. TPI and protein content were measured in all genotypes at all leaf positions. M. sexta larval mass and movement were recorded, and we calculated and simulated the amount of TPI and protein consumed by the larvae from the larval movement and the TPI and protein concentration at each leaf position from each genotype. We used two independently transformed N. attenuata lines in which the expression of the pi gene was down-regulated by antisense expression of a 175 bp fragment of the N. attenuata pi gene (AS –, AS-), and untransformed wildtype plants (WT) of the same genetic background (an inbred line collected from Utah). In addition, we used a natural N. attenuata genotype collected from Arizona, which has a mutation in the endogenous 7-domain pi gene and does not produce pi transcripts or TPI activity (A). We transformed this genotype with the full-length cDNA of the 7-domain pi gene in a sense orientation under control of a constitutive promotor (S++), so that after 11 days of caterpillar attack it produced TPIs at 74 % of the level found in the stem leaves of the wildtype Utah genotype. Our analysis demonstrates that the fitness benefits of TPIs production outweigh their cost when plants are attacked.

Our experiments demonstrate that the benefits of TPI expression in N. attenuata grown in greenhouse conditions outweigh their costs when plants are attacked by M. sexta larvae. Unattacked plants with low constitutive TPI levels produced more seed capsules (AS–: 44, AS-: 34 and A: 49 capsules) than did plants with high TPI levels (WT: 33 and S++: 35 capsules), and 11 days of M. sexta attack reduced seed capsule production per plant in all genotypes and reversed the pattern of seed capsule production with higher reductions in AS (AS–: 91 % and AS-: 68 %) and A (65 %) than in WT (54 %) and S++ (35 %) plants (Fig. 5 and Table 1). This differential reduction in seed capsule production amongst genotypes correlated negatively with larval mass. Across all genotypes, the larger the larval mass, the lower the number of capsules per plant (Fig. 6). This result is consistent with previous demonstrations that endogenous TPIs decrease the performance of M. sexta 33 and with the central prediction of the Optimal Defense theory, namely that defense is costly 233537. Moreover, the results highlight the heuristic value of the cost-benefit paradigm for functional studies. However, conclusive evidence that TPI expression in N. attenuata outweigh their costs when plants are attacked will require field experiments in which both ecological and allocation costs of defense can arise. For example, constitutive and inducible TPI production incurs large fitness costs in N. attenuata when plants where grown with competitors 3242, one of the dominant selective factors for this species 30. In addition, other factors such as temperature and M. sexta predators can affect feeding damage 38.

Despite the central role of the cost-benefit model of inducible defenses, the vast majority of research in this area examines how inducible defenses influence either herbivore performance or plant fitness in separate experiments and their currencies are usually not comparable (i.e., plant fitness for the cost; herbivore performance for the benefits). Few studies have tested the cost-benefit model by measuring both costs and benefits in the same currency (plant fitness for both the costs and benefits) and have elicted plant defenses by either applying methyl jasmonate or damaging leaves 511. However, because of the pleiotorpic effects of the elicitors, the observed fitness differences do not arise solely from the expression of the resistant trait 3043, and therefore these studies are likely to overestimate the fitness cost of resistance. Direct genetic manipulation of TPI expression allowed us measure the costs and benefits of a defensive protein in a plant-insect interaction in the common currency of plant fitness.

Transformation technology gave us the means to manipulate TPI expression with high precision. Antisense expression of the pi gene reduced constitutive and caterpillar induced TPI levels in AS – and AS-genotypes (by 35–80% of the activity of WT) in S1 and stem leaves without influencing protein contents. Caterpillar attack increased TPI levels 2–2.5-fold in either WT or AS genotypes (Fig. 2; Figs 1-3/Appendix 1) but the absolute levels were substantially lower in the AS genotypes. Transformation of the A genotype with a functional TPI gene under the control of a constitutive promoter (S++ genotype) produced TPI levels that were 74% of the activity found in caterpillar attacked WT plants (Fig. 2; Fig. 4/Appendix 1). Because these transformed lines did not differ in any other measured defense traits 42, they allowed us to examine the defensive function of TPIs by constraining plant responses to herbivore attack and observe unconstrained herbivore behavior in response to these constrained plant responses. In this way, the dynamics of the plant responses, or the lack thereof, are reflected in the herbivore behavior. Low constitutive TPI expression in the host plant may increase proteolytic enzyme activity in the guts of neonates, digestion efficiency and the growth rates 44 (Fig. 4). This early increase in larval growth rate translates into increases in pupal mass, which is an accurate proxy for fecundity in Lepidoptera 4546, but may also profoundly influence larval movement.

Given the large within-plant heterogeneity in food quality, it is reasonable to expect a complex resource-oriented larval behavior that changes with instars 2747. Moving has been shown to be costly during the first 3 instars 2638, but these costs are thought to decrease with size 48. Larvae with larger mass (on either low or no TPI-producing genotypes) left the S1 leaf 1–2 days earlier than did those with lower mass (on high TPI-producing genotypes; Fig. 3). The heavier larvae moved earlier than lighter larvae to young leaves which typically have higher levels of protein and water contents 274849. Based on the calculations, larvae fed on high TPI genotypes consumed 3–4 fold more TPI and 10 % less protein than did larvae feeding low TPI genotypes over the 11d of the experiment (Fig. 3 and 4; Fig. 5/Appendix 1). Since we did not measure the amount of leaf consumed by larvae and the values used for the calculation of protein consumption are from plants with natural TPI levels, the calculations likely underestimate the amount of protein consumed. These results suggest that a high TPI content keeps caterpillars from feeding on the high-protein younger leaves at the TOP of the plants possibly by decreasing larval mass and thereby their ability to move. Larval movement influences the caterpillar's ability to compensate for variation in diet quality.

By moving, caterpillars can exploit the high within-plant heterogeneity in food quality to compensate for nutritional imbalances. For example, Helicoverpa zea larvae feed on multiple plant structures to balance their amino acid requirements 50. M. sexta larvae fed low protein and nutritionally unbalanced diets compensated not only for the decreased protein intake 51 but also for unbalanced nutrition by selecting diets high in the missing nutrients which increased larval growth rates 5253. Growth depends on nutrient ratios, and insects may use behavioral and post-ingestive mechanisms to compensate for nutrient imbalances 5455. To estimate the consequence of higher caterpillar mass on movement, we transposed the larval location data of caterpillars from those observed on low-to high-TPI genotypes, and found increased larval protein (10 %) and TPI (12 %) consumption (Fig. 6/Appendix 1). Transposing daily larval location data in the opposite direction decreased (by 10 %) protein consumed but did not influence TPI consumption (Fig. 6/Appendix 1). These calcuations suggest that M. sexta caterpillars may adjust their feeding positions to minimize TPI consumption and maximize protein intake. Hence the naturally occurring high TPI levels delay larval growth and prevent caterpillars from feeding on high-quality younger leaves, which may have a high fitness value for the plant 355056.

The interaction between N. attenuata and M. sexta starts with moths ovipositing on leaves at the bottom of the plant; oviposition is influenced by temperature, food quality and quantity, and predation risk 38. Plants respond by increasing TPI levels, which decreases larval mass and survivorship 33, and by increasing the emission of volatile organic compounds, which alters oviposition choices and attracts the generalist predator Geocoris pallens to feeding larvae 31. Geocoris is size selective and preferentially attacks eggs and larvae in the first three instars. The up-regulation of TPIs by herbivore attack slows larval growth and keeps larvae in stages that are more vulnerable to the predator, thus increasing larval mortality 57. Interestingly, the volatile signals that function as indirect defenses by attracting Geocoris to feeding larvae are elicited by the same signals that elicit TPI production 343658, providing the mechanism of coordination among these defense system.

Once larvae reach a mass that can compensate for the cost of movement, they leave the leaf with high TPI levels and move upward within the host plant and feed preferentially on young leaves with high levels of protein and nicotine, which increases larval mass and decreases plant fitness 353851. A starvation period during the firsts instars was found to reduce M. sexta larval development more than feeding on fully JA-induced (high TPIs) N. attenuata leaves 40. Thus for these larval instars, the costs of movement, which include increases in starvation and predation risks are likely greater than the costs of coping with a plant's induced defenses. Other generalist herbivores on N. attenuata, namely noctuid larvae and weevil beetles, usually attack older leaves that are lower in nutrients as well as nicotine 303538. Nicotine is not an efficient defense against M. sexta, because this insect is adapted to feed on N. attenuata and larvae can detoxify nicotine 596061. Moreover, its attack down-regulates nicotine production which could be sequestered by the herbivore and maybe co-opted and used as a defense against parasitoids 306263. Hence the plant relies on other defenses when attacked by M. sexta larvae: TPIs, for example, decrease larval mass and prevent caterpillars from feeding on leaves with high fitness value for the plant. This delayed in caterpillar movement upward within the plant, maybe a result of larvae adaptation to high leaf-TPI levels by increasing the production of insensitive gut proteases to TPIs 29. Eliciting only those defenses that confer resistance to the attacking herbivore (targeting), rather than the entire defensive repertoire, may minimize the cost of resistance 14.

We conclude that despite the ongoing evolutionary interaction between N. attenuata and M. sexta, TPI-mediated decreases in herbivore performance translates into a fitness benefit for the plant.

JAZ carried out the experiments and analyzed the data, while planning of the experiment and writing of the manuscript was a joint effort by JAZ and ITB.