Herbivore-induced plant vaccination. Part I. The orchestration of plant defenses in nature and their fitness consequences in the wild tobacco Nicotiana attenuata


For correspondence (fax +49 3641 571102; e-mail Baldwin@ice.mpg.de).


A plant's responses to attack from particular pathogens and herbivores may result in resistance to subsequent attack from the same species, but may also affect different species. Such a cross-resistance, called immunization or vaccination, can benefit the plant, if the fitness consequences of attack from the initial attacker are less than those from subsequent attackers. Here, we report an example of naturally occurring vaccination of the native tobacco plant, Nicotiana attenuata, against Manduca hornworms by prior attack from the mirid bug, Tupiocoris notatus (Dicyphus minimus), which results from the elicitation of two categories of induced plant responses. First, attack from both herbivore species causes the plants in nature to release predator-attracting volatile organic compounds (VOCs), and the attracted generalist predator, Geocoris pallens, preferentially attacks the less mobile hornworm larvae. Second, attack from both mirids and hornworms increases the accumulation of secondary metabolites and proteinase inhibitors (PIs) in the leaf tissue, which is correlated with the slow growth of Manduca larvae. Mirid damage does not significantly reduce the fitness of the plant in nature, whereas attack from the hornworm reduces lifetime seed production. Consequently, plants that are attacked by mirids realize a significant fitness advantage in environments with both herbivores. The combination of growth-slowing direct defenses and predator-attracting indirect defenses results in greater hornworm mortality on mirid-attacked plants and provides the mechanism of the vaccination phenomenon.


Plants respond to attack from their enemies with a bewildering array of metabolic responses. The response to a particular herbivore/pathogen species may cause cross-resistance to one or more subsequently attacking species. Such cross-resistance, broadly characterized as plant vaccination or immunization, has been used commercially, for example, to protect crop plants against harmful pathogens by inoculating them with less damaging pathogens or by applying pathogen-derived elicitors (Moffat, 2001). Competitive interactions between arthropod herbivores may also result from such plant-mediated cross-resistance (Fisher et al., 2000; Karban and Baldwin, 1997). For example, prior inoculation of Zinfandel grapevines by phytophagous Willamette mites (Eotetranychus willamettei) induced resistance against the economically more damaging Pacific spider mite (Tetranychus pacificus), a vaccinating effect that controlled Pacific spider mite populations in vineyards (Karban et al., 1997). The asymmetric competition between these two phytophagous mite species was amplified by the introduction of a shared predator (Karban et al., 1994). Despite the apparent success of this study, vaccinating plants against losses to herbivores by inducing resistance with benign herbivores has not gained acceptance in agriculture (Karban and Baldwin, 1997), nor is it known to occur in natural populations, probably because the requirements are onerous. Two requirements are essential: the more competitive herbivore species should be the less damaging one and the plant should realize a fitness benefit from the initial, vaccinating challenge. Competitive interactions between herbivores have been shown to be frequently, but not always (Faeth, 1992), plant-mediated and asymmetric (Denno et al., 1995). However, the herbivore-induced plant responses and their fitness consequences that provide the mechanisms of the vaccination phenomena are rarely examined (Agrawal, 1998, 1999). Moreover, it is crucial to distinguish between simple induced resistance and vaccination by evaluating the potential fitness costs of the response to the initial vaccinating attacker.

Here, we describe a naturally occurring example of herbivore-induced plant vaccination of the wild tobacco Nicotiana attenuata Torr. ex Watts., which results from the co-activation of the plant's direct and indirect defenses. N. attenuata grows ephemerally in large populations after fires in desert habitats in south-western USA, and germinates from long-lived seed banks in response to factors in wood smoke (Preston and Baldwin, 1999). This ‘fire-chasing’ behavior forces the plant's arthropod herbivore community to re-establish itself with every new population. Inducible plant defenses are thought to be an adaptation to this unpredictable biotic environmental factor (Karban and Baldwin, 1997). Both the herbivore-induced direct and indirect defenses of N. attenuata are known to increase the fitness of plants under attack in natural populations (Baldwin, 1998; Kessler and Baldwin, 2001) and to incur fitness costs when the defenses are elicited in the absence of herbivores (Baldwin, 1998; Glawe et al., 2003). Wild tobacco increases its production of secondary metabolites (nicotine, proteinase inhibitors (PIs), phenolics, volatile terpenoids, etc.) after Manduca hornworm attack or elicitation with methyl jasmonate (MeJA), both of which diminish the plant's palatability and attract generalist predators (van Dam et al., 2000, 2001a,b; Keinänen et al., 2001; Kessler and Baldwin, 2001, 2002a). Many of these metabolic responses are herbivore-specific and require the introduction of herbivore elicitors, such as fatty acid–amino acid conjugates from Manduca caterpillar saliva (Halitschke et al., 2001) into plant wounds.

While studying the herbivore community of many populations of N. attenuata plants, we noticed that the leaf-chewing larvae of the sympatric sibling species, Manduca quinquemaculata Haworth and M. sexta L., tend not to co-occur with a sap-sucking mirid (Tupiocoris notatus Distant), even when both species are found in adjoining host populations. Here, we present field and laboratory data that explain this lack of co-occurrence and highlight the role of the plant in mediating interactions within its arthropod community. We demonstrate that attack from the less damaging, piercing–sucking herbivore vaccinates the plant against attack from the chewing herbivore, which causes more damage and decreases plant fitness in nature. In addition, we describe the mechanisms that account for this vaccination phenomenon.

Results and discussion

Attack from herbivores elicits the accumulation of secondary metabolites, with toxic, antidigestive, or antinutritive activity, which function as direct defenses (Duffey and Stout, 1996), as well as the release of volatile organic compounds (VOCs) that can function as indirect defenses by attracting the enemies of the plant's enemies (Dicke and van Loon, 2000). Hence, the plant–herbivore interaction is influenced by plant responses that function at many spatial scales from the cellular to the whole-plant and community levels (Kessler and Baldwin, 2002b). Attack from different herbivore species can elicit similar responses, but the fitness consequences frequently depend on the ecological context and the natural history of the interacting species. Plant resistance against a particular herbivore may be influenced by additional herbivores that compete for the plant as a resource (Agrawal, 1998; Karban et al., 1997) by members of the third trophic level (Hoballah and Turlings, 2001; Kessler and Baldwin, 2001), or even by neighboring plants (Baldwin et al., 2002; Karban et al., 2000). This context-dependence highlights the importance of studying plant–herbivore interactions in nature.

Our interest in the plant-mediated interaction between hornworms and mirids originated from two observations of natural N. attenuata populations: (i) established mirid populations were correlated with higher hornworm mortality, and (ii) higher hornworm mortality was associated with higher plant fitness (Table 1). These observations prompted three hypotheses: (i) the mirid-induced plant responses directly or indirectly increase the mortality of hornworm larvae; (ii) the plant-mediated interaction of the herbivores results in a fitness benefit for plants attacked by both herbivores; and (iii) a less damaging herbivore (mirids) vaccinates the plant against a more damaging one (hornworm) in nature.

Table 1.  Mirid density, mean (±SEM) M. quinquemaculata hornworm mortality, and mean (±SEM) seed capsule production in two N. attenuata populations (HI91 burn – 1998 (R313: 37°05′18″N, 113°50′34″W; Pahcoon Spring burn – 1998 (R310: 37°13′59″N, 113°50′18″W)) in the Great Basin desert in south-western Utah in June 1999
PopulationnPlants with mirids (%)nHornworm mortality (% per day) ± SEMnSeed capsule number (per plant)
Pahcoon362.76613.4 ± 5.2367.1 ± 1.7
HI912684.63034.4 ± 8.82922.6 ± 10.4

Herbivore-induced indirect defenses increase hornworm predation and decrease moth oviposition in nature

To evaluate the potential role of VOCs as indirect defense traits for the observed cross-resistance, we characterized and compared the VOC emission of N. attenuata plants in response to mirid and hornworm attack in a natural population. The GC–MS analysis of the trapped VOCs revealed that attack from both herbivore species elicited increased emissions of a similar suite of compounds, although the elicited odor profiles differed quantitatively. The identified compounds are derived from three biosynthetic pathways (Figure 1a; Table 2). The green leaf volatiles, cis-3-hexene-1-al and cis-3-hexenyl acetate, are derived from the oxylipin pathway. Their emission did not differ between plants damaged by mirids and those damaged by hornworms (Figure 1; Table 2), but was significantly elevated compared to the emission from undamaged control plants. In contrast, the emission of another green leaf volatile, cis-3-hexenyl butyrate, and the emission of the sesquiterpene, cis-α-bergamotene, differed between caterpillar- and mirid-damaged plants (Table 2). The emissions of other terpenoid compounds, such as the sesquiterpene trans-β-farnesene, and the monoterpenes trans-β-ocimene and linalool, were similar from mirid- and hornworm-attacked plants. Methyl salicylate, which derives from the shikimate pathway, was exclusively emitted after herbivore damage, with similar amounts released from hornworm- and mirid-damaged plants (Table 2). The finding that plants attacked by herbivores from different feeding guilds emit a qualitatively similar suite of VOCs is consistent with recent findings from cotton and maize, which demonstrate that insects from two different feeding guilds, the piercing–sucking tarnished plant bug (Lygus hesperus) and leaf-chewing caterpillars, elicit similar VOC emissions (Rodriguez-Saona et al., 2002). While different herbivore species may cause plants to release similar signals, the fitness consequences for the plant will depend on the natural history of the herbivores and the effects on third trophic level interactions.

Figure 1.

Indirect defense function of VOC emission.

(a) Representative total ion chromatograms of the headspace of an undamaged leaf from an undamaged plant (CTRL) compared to a leaf damaged by a M. quinquemaculata larva (hornworm) and a leaf damaged by five T. notatus adults (mirid). The labels represent the following compounds: 1, cis-3-hexen-1-al; 2, trans-β-ocimene; 3, cis-3-hexenyl acetate; 4, linalool; 5, terpineol; 6, cis-3-hexenyl butyrate; 7, methyl salicylate; 8, cis-α-bergamotene; 9, trans-β-farnesene.

(b) Mean (±SEM) mortality of hornworm eggs experimentally applied to undamaged plants (CTRL) and plants under T. notatus (mirid) or M. quinquemaculata (hornworm) attack.

(c) Mean (±SEM) oviposition of adult Manduca females on undamaged plants (CTRL) and plants damaged by either 10 T. notaus (mirid) or one to four M. quinquemaculata larvae (hornworm).

Table 2.  Mean (±SEM) amountsa of N. attenuata VOCs, collected from the headspace of undamaged control plants, plants damaged by T. notatus (mirid), and plants damaged by M. quinquemaculata hornworms after 24 h of herbivore attack
CompoundaControlbMirid damagebIncreaseHornworm damagebIncreaseanova
  • Letters in parentheses represent the results of a Bonferroni post hoc test of an anova (last column) of log-transformed data, whereby different letters designate significantly different means (P < 0.05).

  • a

    Expressed as nanograms per hour.

  • b

    Data derived from n = 8 replicates.

 Cis-3-hexene-1-al3.36 ± 1.1 (A)18.97 ± 3.6 (B)5.6×31.03 ± 8.8 (B)9.2×F = 10.43; P = 0.0011
 Cis-3-hexenyl acetate2.13 ± 1.1 (A)5.51 ± 0.8 (B)2.6×11.98 ± 3.1 (B)5.6×F = 9.84; P = 0.0013
 Cis-3-hexenyl butyrate1.91 ± 1.2 (A)4.58 ± 1.2 (B)2.4×19.88 ± 4.4 (C)10.4×F = 21.45; P < 0.0001
 Trans-β-ocimene1.09 ± 1.1 (A)10.20 ± 4.8 (B)9.4×16.25 ± 6.7 (B)14.9×F = 6.43; P = 0.0096
 Cis-α-bergamotene0.42 ± 0.4 (A)2.79 ± 0.6 (B)6.6×15.13 ± 3.6 (C)36.0×F = 33.83; P < 0.0001
 Trans-β-farnesene0.97 ± 0.7 (A)2.53 ± 0.5 (B)2.6×4.50 ± 0.9 (B)4.6×F = 9.41; P = 0.0016
 Methyl salicylate0 ± 0 (A)12.8 ± 3.5 (B)13.0 ± 6.4 (B)F = 6.45; P = 0.011

The herbivore-induced emission of VOCs has long been proposed as a potential indirect defense trait that facilitates the prey search behavior of predatory/parasitoid insects and thus top-down control of the herbivore community (Dicke and van Loon, 2000). Only one has been shown to function defensively in nature: the VOCs released from herbivore-damaged N. attenuata plants that attract the predatory bug Geocoris pallens, which attacks both mirids and the eggs, and early larval instars (first and second) of hornworms (Kessler and Baldwin, 2001, 2002a). Interestingly, experimental applications of some of the commonly emitted compounds to N. attenuata plants (Kessler and Baldwin, 2001) or sticky traps (James, 2003) are sufficient to attract the generalist predator.

To test for indirect predator-attracting effects of mirid- and hornworm-elicited VOC emission on hornworm mortality, we glued hornworm eggs on herbivore-damaged and undamaged plants, and recorded the subsequent egg predation. We used eggs rather than larvae in these experiments measuring predation rates to avoid eliciting additional inducible responses by hornworm feeding, and thus, to measure exclusively the indirect effects of the plant response. On plants damaged by mirids, we found a 4.6-fold higher predation rate on the experimentally applied hornworm eggs than on those on undamaged plants (Figure 1b; Mann–Whitney U-test z = − 3.667; P = 0.0002). Because the mirids do not feed on hornworm eggs (we tested this possibility by offering hornworm eggs (without leaf material) to 15 mirids; none of the 20 eggs were damaged, and after 2 days, all larvae had hatched), and all dead eggs were pierced and emptied, the egg mortality was most likely caused by G. pallens (Kessler and Baldwin, 2001, 2002a), which was the only predator observed during the experiment. Similarly, in a parallel experiment in the same population, experimentally oviposited hornworm eggs suffered a 4.2-fold higher predation rate on hornworm-damaged plants (Figure 1b; Mann–Whitney U-test z = −2.304; P = 0.0212) than on undamaged control plants.

These results demonstrate that the very similar VOC emissions from mirid- and hornworm-attacked plants result in a similar attraction of the generalist predator (Figure 1b). However, the consequences of attracting the predator are very different for the two herbivore species. Hornworm eggs and young larvae are more easily attacked than the mobile mirids. When G. pallens was allowed to choose between both prey species, we found that 64.3% of the hornworms in the experiment were killed within 1 h, while none of the mirids were attacked. In non-choice experiments, 21.4% of the mirids and 85.7% of the caterpillars were attacked by G. pallens within an hour. These attacks killed 66.7% of the caterpillars, but none of the mirids were killed. This clear preference of the predator for young Manduca larvae and the similar attractiveness of the VOCs emitted from both mirid- and hornworm-damaged plants provide a strong selective force for adult moths to avoid ovipositing on signaling plants, which our field experiments corroborated.

Manduca females avoided both mirid- and hornworm-damaged plants for oviposition (Figure 1c). The oviposition rate of female Manduca moths on undamaged control plants in a natural population was, on average, 5.3-fold higher than on hornworm-damaged plants (Student's t-test t = −2.607; P = 0.014) and fourfold higher than on mirid-damaged plants (Student's t-test t = −2.47; P = 0.018). These results are consistent with the premise that moths use the same predator-attracting volatile signals to avoid ovipositing on plants that are damaged by conspecifics, as shown in earlier studies (De Moraes et al., 2001; Kessler and Baldwin, 2001). In addition, we show here that moths avoid plants that are damaged by herbivores other than their conspecific larvae. By avoiding mirid-damaged plants for oviposition, the adult moths lessen predation pressure on their offspring. Moreover, adult Manduca moths usually oviposit only one egg per plant, perhaps because N. attenuata plants normally provide only sufficient food to mature one hornworm larvae (Kessler and Baldwin, 2001, 2002a). This behavior amplifies both the benefit for the plant and the ability of mirids to dominate the herbivore community when both species are present (Table 1).

If hornworm eggs manage to hatch and grow to the third instar (of five instars) without being detected by a natural enemy, the larvae become sufficiently large and capable in their defensive behaviors to escape predation (Kessler and Baldwin, 2002a). As a consequence, plant traits that influence the larvae's rate of growth also influence the time during which the larvae remain vulnerable to the plant's indirect defenses. Attack from herbivores results in a transcriptional and metabolic reorganization in the plant and increase in the accumulation of secondary metabolites that may not only facilitate indirect defense, as demonstrated for the herbivore-induced VOCs, but also have toxic or antidigestive effects on the herbivore.

Mirid-induced direct plant defenses reduce hornworm performance

Field-collected M. quinquemaculata had a faster mass gain and reached the next larval instar earlier on undamaged control plants than on mirid-infested plants. After 8 days of development, the larvae which were fed on undamaged control plants had, on average, 85% larger body mass than those fed on plants that had been previously attacked by mirids (Figure 2a; repeated measures anovaF1,62 = 7.972; P = 0.0064). Moreover, after 8 days of feeding, 89% of the hornworms on control plants, but only 35% of the hornworms on mirid-attacked plants, had reached the fourth larval instar. These results suggest that mirid attack elicits plant resistance, which, in turn, negatively influences hornworm performance. Earlier studies had shown that attack by hornworms or the application of their regurgitant to mechanical wounds elicits changes in gene expression and secondary metabolism (Halitschke et al., 2001, 2003; Hermsmeier et al., 2001; Schittko et al., 2001; Winz and Baldwin, 2001). Some of the herbivory-induced compounds are also elicited by MeJA application (van Dam et al., 2001b; Keinänen et al., 2001) and may partially account for the reduced hornworm performance observed on the herbivore-attacked plants. Here, we compared the effects of both mirid and hornworm attack on resistance to subsequent hornworm feeding and production of secondary metabolites.

Figure 2.

Performance of Manduca hornworms on T. notatus (mirid) damaged and undamaged control plants.

(a) Comparison of the mean (±SEM) M. quinquemaculata hornworm mass 0, 4, 6, and 8 days after feeding on control leaves from undamaged plants (CTRL, circles) versus leaves of plants previously damaged by the mirid T. notatus (triangles).

(b) Mean (±SEM) mass of M. sexta hornworms on day 4 after feeding on damaged/uncaged (local) or undamaged/caged (systemic) leaves of undamaged (CTRL), mirid-, and hornworm-attacked plants. Different letters designate mean hornworm weights that differed significantly (P < 0.05, in a Fisher's PLSD of anova for comparison of different treatments, and P < 0.05 in a paired Student's t-test for comparison of local and systemic effects).

A laboratory experiment with the M. quinquemaculata sister species, M. sexta, in which mirids as well as hornworms were used to experimentally infest plants, revealed a similarly negative effect on hornworm growth and performance, and excluded the possibility that mirids in the field experiment had selected plants of low quality. Already after 4 days of development, larvae feeding on undamaged control plants were significantly heavier than those feeding on mirid-damaged plants (anovaF2,32 = 17.542; P < 0.001, Fisher's protected least-significant difference (PLSD) P = 0.0002; Figure 2b) and previously hornworm-attacked plants (Fisher's PLSD P = 0.0001; Figure 2b). The mass of hornworms feeding on both mirid- (paired Student's t-test t = 0.96; P = 0.37) and hornworm-damaged (paired Student's t-test t = 2.3; P = 0.06) plants was similar to the mass of larvae reared on leaves that were excluded from herbivory on otherwise damaged plants, demonstrating that the responses were systemically transmitted throughout the plant (Figure 2b). The lack of significant differences between the masses of hornworms reared on caged and uncaged control plants ruled out effects of caging as a possible explanation (paired Student's t-test t = 1.78; P = 0.13; Figure 2b). Interestingly, the decrease in M. quinquemaculata (Figure 2a) and M. sexta hornworm (Figure 2) weight gain was comparable to that observed in similar experiments with N. attenuata genotypes differing in trypsin protease inhibitor (TPI) production (Glawe et al., 2003), and with plants elicited with MeJA (van Dam et al., 2000).

Here, we found that the reduced growth rates of M. quinquemaculata and M. sexta larvae were correlated with increased accumulations of a suite of secondary metabolites and TPIs in the leaf tissue of mirid- and hornworm-attacked plants (Table 3). The concentrations of several secondary metabolites, such as chlorogenic acid, an unknown compound (RT 13.2), and four diterpene glycosides (DTGs) 6–9 significantly increased after both mirid and hornworm damage. Others, such as cryptochlorogenic acid and caffeoylputrescine, significantly increased only after mirid attack (Table 3). These findings are consistent with results from a transcriptional analysis of the responses to attack by the two herbivore species, which found a set of transcripts whose expression was upregulated by both species, as well as a set of transcripts that was differentially regulated (Voelckel and Baldwin, 2004). The greater resistance of herbivore-damaged plants compared to undamaged control plants was also positively correlated with the production of TPIs in the leaf tissue (Table 3). Both mirid- and hornworm-damaged plants had significantly elevated TPI concentrations compared to undamaged plants (Table 3).

Table 3.  Mean (±SEM) concentrationsa of N. attenuata secondary metabolites and TPI in undamaged control plants, plants damaged by T. notatus (mirid), and plants damaged by M. sexta (hornworms) after 4 days of herbivore attack
CompoundaControlbMirid damagebIncreaseHornworm damagebIncreaseanova
  • Letters in parentheses represent the results of a Bonferroni post hoc test of an anova (last column) of log-transformed data, whereby different letters designate significantly different means (P < 0.05).

  • a Expressed as micrograms per gram fresh mass; caffeoylputrescine and cryptochlorogenic acid are expressed as chlorogenic acid equivalents; unknown compounds and DTGs are expressed as peak areas at 210 nm mg−1 fresh mass; TPI are expressed as micromoles per milligram protein.

  • b

    Data are means 

  • ±

    SE (n = 8).

Nicotine144.4 ± 30.9 (A)139.4 ± 24.9 (A)0.97×126.2 ± 29.7 (A)0.87×F = 0.97; P = 0.42
Caffeoylputrescine1.8 ± 0.2 (A)8.9 ± 1.1 (B)4.9×4.1 ± 0.9 (A)2.3×F = 15.79; P < 0.0001
Chlorogenic acid97.1 ± 13.1 (A)213.8 ± 18.5 (B)2.2×230.5 ± 33.1 (B)2.4×F = 10.95; P = 0.0002
Cryptochlorogenic acid17.3 ± 2.2 (A)105.8 ± 17.3 (B)6.1×16.8 ± 4.2 (A)0.97×F = 32.32; P < 0.0001
Rutin548.5 ± 17.3 (A)539.2 ± 56.5 (A)0.98×482.3 ± 29.4 (A)0.88×F = 0.83; P = 0.44
Unknown RT 10.322.0 ± 1.8 (A)39.6 ± 7.8 (A)1.8×30.3 ± 2.6 (A)1.4×F = 1.86; P = 0.17
Unknown RT 13.20 ± 0 (A)24.9 ± 4.4 (B)13.2 ± 5.2 (B)F = 16.29; P < 0.0001
DTG 6235.9 ± 25.2 (A)1182.9 ± 296.9 (B)5.0×635.8 ± 113.1 (B)2.7×F = 13.17; P < 0.0001
DTG 7515.6 ± 31.5 (A)951.7 ± 61.9 (B)1.8×914.3 ± 67.1 (B)1.8×F = 18.84; P < 0.0001
DTG 8419.9 ± 26.2 (A)911.1 ± 53.3 (B)2.2×948.1 ± 81.9 (B)2.3×F = 28.41; P < 0.0001
DTG 91867.4 ± 106.4 (A)3252.0 ± 195.9 (B)1.7×3757.6 ± 279.4 (B)2.0×F = 23.05; P < 0.0001
TPI0.182 ± 0.018 (A)2.184 ± 0.71 (B)12.0×1.645 ± 0.259 (B)9.0×F = 30.30; P < 0.0001

In summary, as attack from both species elicits rather similar VOCs and other secondary metabolite responses, it is likely that the ecological context of these similar responses determines the dramatically different fitness consequences for the plant.

Direct defenses amplify the defensive function of mirid-induced indirect defenses and provide a fitness benefit for plants in nature

One fitness benefit for the plant arises from the natural history of interaction between hornworms and N. attenuata plants. Hornworms are the most damaging insect herbivores on N. attenuatta. One hornworm can consume three to five plants before it reaches the pupal stage. Larvae typically remain on the plant they were oviposited on until they reach the fourth larval instar; by then, they consume up to 90% of the plant's leaf tissue (Kessler and Baldwin, 2002a). The hornworms usually depart before the plant is completely consumed, and the amount of leaf tissue lost to hornworm feeding is negatively correlated to lifetime seed capsule production, as measured in different N. attenuata plant populations (Figure 3). Consequently, the fitness costs of hornworm damage for the plant depend strongly on the developmental stage in which the hornworm leaves the plant or is removed from the plant by natural enemies, such as predators or parasitoids (Kessler and Baldwin, 2001). The growth-reducing effect of TPIs (Zavala et al., 2004) and secondary metabolites elicited by previous herbivore attack (mirid or hornworm) causes the hornworms to remain longer in the first two more vulnerable larval instars (Figure 2), which may expose them longer to potential predators. Such predators, in turn, are attracted by the herbivore-induced VOC emission (Figure 1; Table 2). The direct effects of mirid-induced plant responses on hornworm performance amplify the defensive effect of predator attraction, which is consistent with the slow growth–high mortality hypothesis (Clancy and Price, 1987; Williams, 1999).

Figure 3.

Herbivore damage and seed capsule number.

(a) Mean (±SEM) number of seed capsules produced by undamaged control plants (CTRL; n = 18), plants attacked by T. notatus (mirid; n = 20), plants under continuous attack by M. quinquemaculata (hornworm; n = 15), and plants under attack by both species (hornworm + mirid; n = 5).

(b) Mean capsule number (±SEM) of N. attenuata plants after various levels of M. quinquemaculata damage. Different letters designate treatments that differ significantly in capsule number (P < 0.05; Bonferroni post hoc test of an anova).

Another fitness benefit of this interaction results from the very different reproductive consequences of hornworm and mirid attack for the plant. While the plant responds similarly to hornworm and mirid attack (and gains resistance to hornworms), attack by mirids does not reduce the reproductive success of plants, although the damage resulting from the feeding of these piercing–sucking insects can be substantial. In the field experiment, the lifetime seed capsule number did not differ between control and mirid-damaged plants (Figure 3a; Bonferroni post hoc test of anovaP > 0.05), but it was significantly reduced in hornworm-damaged plants (Figure 3a; anovaF3,53 = 4.74; P = 0.0079; Bonferroni post hoc test P < 0.05). Therefore, plants experience mirid attack without severe fitness consequences; yet hornworm damage reduces lifetime seed production by 51.2% compared to undamaged control plants. Recent experiments with Datura wrightii reported similar neutral effects of T. notatus attack on plant fitness (Elle and Hare, 2000; Hare and Elle, 2002). Here, seed production was not reduced, and the authors suggested that damage by T. notatus may reduce photosynthetic capacity less than equivalent damage by chewing insects (Hare and Elle, 2002).

Compensatory photosynthetic responses may play an important role in the tolerance of plants to herbivore attack. A recent differential display-reverse transcriptase (DDRT)-PCR and subtractive library study of mirid-attacked N. attenuata plants (Voelckel and Baldwin, 2003) identified a series of mirid-specific transcriptional responses, which collectively suggest that an adjustment of photosynthesis is involved in N. attenuata's ability to tolerate mirid attack. The targets of transcriptional regulation ranged from proteins in photosynthetic electron transport, CO2 fixation, and ribulose bisphosphate regeneration. Of particular note was the mirid-induced increase in ribulose 1,5-bisphosphate carboxylase (RuBPCase) activase transcripts, which code for a stromal, regulatory protein that regulates the activity of the key enzyme in CO2 assimilation, RuBPCase (Portis, 1995).

The severe negative effect of hornworm damage on N. attenuata fitness results from both the allocation of resources to the defense responses and from the loss of photosynthetically active leaf tissue. An analysis of all hornworm-damaged plants from this study (anovaF3,56 = 4.327; P = 0.0082) and the earlier analysis of 100 Manduca-attacked plants from another population (Apex mine) revealed that lifetime capsule production was negatively correlated with the amount of damage inflicted by hornworms (Figure 3b; anovaF3,88 = 7.014; P = 0.0003). Plants that were under continuous attack by both mirids and hornworms had an intermediate fitness level. The number of seed capsules was lower than on plants damaged exclusively by mirids, but higher than on exclusively hornworm-damaged plants (Bonferroni post hoc test P > 0.05; Figure 3a), which resulted from the slower hornworm growth and, consequently, decreased consumption of leaf tissue.

The neutral effects on plant fitness and the induced responses to herbivory that function as direct and indirect defenses in the context of the particular life history traits of the competing herbivore species provide the mechanism for the plant vaccination phenomena. Specifically, plant VOC-mediated alterations in indirect defenses accounted for the vaccination effect. Our study emphasizes the central role of induced plant defenses in structuring arthropod communities, and thereby confirms earlier studies (Denno et al., 1995, 2000; Fisher et al., 2000; Inbar et al., 1995). It shows that a suite of rather similar plant responses to attack from different herbivores can result in dramatic differences for the plant's fitness in nature, and thereby illustrates the value of studying plant–herbivore interactions in the complexity of the natural environment. Moreover, the fact that under certain circumstances (top-down control of the herbivore community, prior vaccination by the less harmful herbivore, etc.), these ecological interactions can produce a fitness benefit for the plant suggests that herbivore vaccination may have agricultural uses.

Experimental procedures

Census of arthropod communities

Our formal investigation of the interactions of N. attenuata's arthropod community in the Great Basin desert in south-western Utah started in June 1999. We censused 30–40 cm tall plants (n = 96; Table 1) in linear transects (4–6 m distance between plants) at two different burns (HI91 burn – 1998 (R313: 37°05′18″N, 113°50′34″W); Pahcoon Spring burn – 1998 (R310: 37°13′59″N, 113°50′18″W)), and recorded the occurrence and dynamics of the two most abundant herbivores, M. quinquemaculata and T. notatus, over 2 weeks at 2-day intervals. In addition, we monitored the lifetime seed capsule production of N. attenuata plants in the two different populations.

Volatile organic compound emission

To examine the difference in VOC emission between hornworm- and herbivore-attacked plants, we chose plants from a 100 m2 portion of a population (HI91 burn – 2000 (37°07′28″N, 113°49′46″W)) and sampled them simultaneously for 7 h after the herbivores were allowed to feed on the plants for 24 h. We used an open-flow trapping design (Kessler and Baldwin, 2001) to collect VOCs individually from 24 plants, each of which had one leaf attacked by either one second-instar Manduca caterpillar or by five T. notatus adults, or had remained undamaged. To restrict insects to a single leaf and to trap volatiles from this leaf, leaf and insects were enclosed in 400 ml polystyrene chambers fitted with holes at both ends. Air was pulled through the chamber at 450–500 ml min−1 (measured by a mass flow meter; Aalborg Instruments, Orangeburg, NY, USA) and subsequently through a charcoal air-sampling trap (ORBO™-32; SUPELCO, State College, PA, USA) by a portable 12 V DC vacuum pump (Gast Mfg, Benton Harbor, MI, USA). The VOCs were eluted from the charcoal trap using dichloromethane (SIGMA, Taufkirchen, Germany). The analysis was performed at the field site with a Shimadzu (Model 5000) quadrupole GC–MS programmed as described earlier by Kessler and Baldwin (2001). Compounds were identified by comparing retention times and mass spectra with those of authentic standards. We statistically evaluated the correlation between the VOC emission and the different treatments by using log-transformed data in an anova, and compared the results from different treatments with a Bonferroni post hoc test.

M. quinquemaculata mortality

To test for indirect predator-attracting effects of mirid damage on hornworm mortality, we conducted a field experiment in June 2001 in a natural population of about 100 000 plants growing in a recently burned area (HI91 burn – 2000 (37°07′28″N, 113°49′46″W)). We chose 40 elongated but not yet flowering plants at the same developmental stage (30–40 cm tall) along a linear transect across the population, and infested every second plant (n = 20) with 10 mirid adults. The rest of the plants (n = 20) remained undamaged. The minimum distance between two plants receiving the same treatment was 6 m and that between different treatments was 3 m. Two days after the infestation, we glued five hornworm eggs on the second and five on the third leaf position of each of the 40 plants using a water-α-cellulose (SIGMA) solution as glue, which neither influences the development of eggs nor attracts predators (Kessler and Baldwin, 2001). In a similar experiment in June 2000 (burn Apex mine – 1999 (W246: 37°05′09″N, 113°46′13″W)), we recorded egg predation on previously Manduca hornworm-damaged plants (n = 15), and compared it with predation on undamaged plants (n = 15). As with the mirids, we allowed two second-instar caterpillars to damage the plants for 2 days before we glued 10 eggs to the second and third stem leaf of each plant. In both experiments, predation on the eggs was monitored over 2 days, until hatching. Because G. pallens pierces and empties eggs, leaving the eggshells nearly intact, its predation is easy to distinguish from other mortality sources. The mortality rates were ln-transformed and compared with Mann–Whitney U-tests.

G. pallens prey preference

Under field conditions, the predatory bug G. pallens seemed to prefer young Manduca larvae over mirids as food. To test this hypothesis, we conducted a laboratory choice experiment in which one G. pallens individual was offered one T. notatus adult and one first-instar M. sexta hornworm in a soufflé cup (Solo 29.6 ml; P100, Urbana, IL, USA). This set-up was replicated 14 times with new insect individuals, and the behavior of G. pallens as well as the mortality of the prey species were recorded for 1 h. Similarly, in a non-choice experiment, G. pallens bugs were placed with either one first-instar hornworm (n = 14) or one adult T. notatus bug (n = 14) in individual soufflé cups, and their behavior and mortality, respectively, were recorded for 1 h.

Manduca oviposition

In two simultaneous field experiments in May 2000 (burn Apex mine – 1999 (W246: 37°05′09″N, 113°46′13″W)), we examined the oviposition preferences of adult M. quinquemaculata females. In the first experiment, we infested 20 plants each with 10 T. notatus adults, and compared moth oviposition on these plants with those on 20 uninfested plants. In the second experiment, we infested 20 plants each with one to four M. quinquemaculata first- to third-instar larvae per plant, and compared natural moth oviposition on these plants with those on another 20 uninfested plants. The spatial separation of the experiments ensured the independence of the moth selection. The plants in both experiments were similar-sized, elongated (30–40 cm) but not yet flowering, and were chosen along two transects across the population. The plants were examined for eggs every second day over a 2-week period. All freshly laid eggs were removed to avoid their potential repelling effects on moths searching for oviposition sites. Oviposition by itself does not elicit VOC release (Kessler and Baldwin, 2001); larvae must hatch and begin feeding to elicit VOCs. Oviposition rates on damaged and control plants were compared with Student's t-test.

M. quinquemaculata performance

To test for direct plant-mediated effects of mirid damage on hornworm performance, we collected the third stem leaves of elongated N. attenuata plants that had either been previously attacked by 5–10 mirid adults or not attacked at all. Twelve damaged and 12 undamaged leaves were placed in individual 1-l polyethylene boxes and infested with one M. quinquemaculata hornworm egg, freshly collected from the field in May 1999. Hornworms were weighed just after hatching, and 4, 6, and 8 days thereafter. To control for environmental and genetic factors that might have influenced host plant susceptibility, we chose developmentally similar leaves from plants in the same developmental stage and population. The leaves were replaced every other day by developmentally similar leaves from new plants to avoid any influence of leaf removal. The hornworm mass performance was analyzed with repeated measures anova of log-transformed data.

Secondary metabolite elicitation and M. sexta performance

Soil-grown N. attenuata plants of an inbred line that originated from a seed collection from Utah were placed pair-wise in glass and mesh cages (L × B × H = 30 cm × 30 cm × 60 cm), and either remained undamaged (n = 10), or were attacked by 10 mirids (n = 12) or by two freshly hatched M. sexta hornworms (n = 14). After 4 days of hornworm or mirid feeding, two additional, freshly hatched hornworms were placed on the source–sink transition leaf (first fully expanded leaf) and on a leaf of the next younger position of each of the damaged or undamaged plants. The hornworms' mass gain was monitored every second day over an 8-day period. To compare the effects of mirid/hornworm feeding on the performance of subsequently attacking hornworms, we compared the mass gain of the hornworms using an anova.

At the start of this no-choice experiment, we harvested the source–sink transition leaf of one plant from each pair in the cages. The leaf samples were weighed, processed, and analyzed for secondary metabolites with an HPLC-based screen described by Keinänen et al. (2001). We quantified nicotine, chlorogenic acid, and rutin by external standard techniques, and expressed the contents as microgram per gram fresh mass. Caffeoylputrescine and cryptochlorogenic acid were expressed as chlorogenic acid equivalents, while DTGs and unknowns were expressed as peak areas at 210 nm mg−1 fresh mass. Student's t-test was used to compare the values from herbivore-damaged and control plants.

In a similar experiment, the source–sink transition leaves of 32 plants were enclosed in individual clip cages. The rest of the plants either remained undamaged (cage control, n = 8), or were attacked by two M. sexta larvae (n = 12) or 10 mirids (n = 12). After 4 days of feeding activity by the two herbivore species, two freshly hatched M. sexta larvae were placed on each of the plants, one in the clip cage on the undamaged leaf and one on the younger damaged leaf. After 4 days of feeding, their mass gain was measured. Paired Student's t-test were used to compare the effects of mirid/hornworm feeding on the performance of subsequently attacking hornworms on the enclosed undamaged and damaged source–sink transition leaves of attacked plants.

Elicitation of proteinase inhibitors

Similar to the experiment described above, plants were placed in mesh cages (L × B × H = 30 cm × 30 cm × 60 cm), and either remained undamaged (n = 8) or were attacked by 10 mirids (n = 8) or two second-instar M. sexta hornworms (n = 8). After 4 days, we harvested approximately 0.1 g leaf material from the source–sink transition leaf of each plant. Protein content and trypsin PI activity of the flash-frozen samples were determined as described by van Dam et al. (2001b). Student's t-test was performed to compare the values from herbivore-damaged and control plants.

Plant fitness after herbivore attack

To determine the fitness consequences of hornworm and mirid attack for the plant, we chose 20 triplicates of undamaged, elongating N. attenuata plants of the same phenology, size, and proximity – at least 4 m from each other – and added either no herbivore, 10 adult mirids, or one hornworm to one of the three plants to simulate the characteristic herbivore densities found in nature (Kessler and Baldwin, 2001). The number of mirids applied was informed by a census of 100 plants in the studied population, which showed that plants attacked by mirids hosted between 3 and 16 individuals. The plant fitness study was performed from June to July 2001 in a natural population (HI91 burn – 2000 (37°07′28″N, 113°49′46″W)), and the plants' damage levels and herbivore loads were monitored every second day during a 4-week period. We removed naturally oviposited hornworm eggs and mirid colonizers from control plants. From the plants infested with hornworms, we removed mirid colonizers to observe only the effects of hornworms. None of the mirid-treated and hornworm-treated plants received naturally oviposited hornworm eggs, supporting the hypothesis that mirid as well as hornworm damage repels ovipositing moths (Kessler and Baldwin, 2001). However, one mirid-treated plant and two control plants were slightly damaged by migrating fifth-instar hornworms, and therefore were excluded from the analysis. By adding a new hornworm of the respective instar when a previous hornworm died, we ensured that the hornworm-treated plants were under continuous attack until the hornworms reached the stage at which they naturally emigrated from the plant (Kessler and Baldwin, 2002a). In this experiment, mirids colonized hornworm-treated plants with the same frequency as control plants. Mean numbers of mirids per plant, which were censused and removed over the course of the experiment, did not differ between the control (0.98 ± 0.35; n = 20) and hornworm-damaged plants (1.7 ± 0.39; n = 20; Student's t-test t = −1.39; P = 0.17). Co-infestation with both herbivore species are rarely observed in nature (see above). Because of the high mortality of hornworms on mirid-damaged plants, only five plants were under continuous attack from both herbivores over the course of the experiment. These were analyzed separately as a fourth treatment group (hornworm + mirid). On July 7, when plants had stopped growing and senesced, we harvested, counted, and measured all seed capsules. The seed capsule volume was calculated by multiplying their measured lengths and widths, and by treating the capsules as cylinders. Capsule size and seed number were significantly correlated (n = 19; R2 = 0.81; P < 0.0001). As capsule size did not vary between the treatments (anovaF2,57 = 0.319; P = 0.73), capsule number remained a reliable proxy of N. attenuata's fitness, as it has been in earlier experiments (Baldwin, 1998). We performed an anova of the log-transformed capsule numbers of each treatment, and compared the treatment groups using a Bonferroni post hoc test of anova.

In another natural N. attenuata population (burn Apex mine – 1999 (W246: 37°05′09″N, 113°46′13″W)), we categorized 100 flowering plants of similar phenology and size in a natural population by the tissue loss they incurred from hornworms. We identified four groups of damage (undamaged – 0%; leaf damage that only involved parts of the laminae – 1–30% tissue loss; removal of complete leaves, including the midrib – 30–60% tissue loss; and damage to the apical meristem and the stem – >60% tissue loss) and compared the average seed capsule numbers of the plants in the four damage groups using a Bonferroni post hoc test of anova.


This research was supported by Max-Planck-Gesellschaft. We thank J. H. Tumlinson, A. Roda, C. DeMoares, J. Thaler, J. C. Schultz, N. M. van Dam, E. Wheeler, R. Halitschke, and K. Sime for help on an earlier draft. We thank R. Baumann for assistance with species determinations, and Brigham Young University for use of their Lytle Preserve as a field station.