Leaf trichome responses to herbivory in willows: induction, relaxation and costs


Author for correspondence: Christer Björkman Tel: +46 18 671532 Fax: +46 18 672890 Email: christer.bjorkman@ekol.slu.se


  • • To circumvent the inherent problem of discriminating between the cost of losing photosynthetic tissue and the cost of producing an inducible defence, the growth response of herbivore-damaged plants was compared with plants damaged mechanically to the same extent but without eliciting the defence.
  • • Two experiments were conducted, studying the response of willows (Salix cinerea) to damage by adult leaf beetles (Phratora vulgatissima).
  • • In the first experiment, willows produced new leaves with an enhanced leaf trichome density 10–20 d after damage, coinciding in time with the feeding of beetle offspring. The response was relaxed in foliage produced 30–40 d after damage. In the second experiment, which also included mechanical damage, willows exposed to beetle feeding showed an increase in leaf trichome density of the same magnitude (> 70%) as in the first experiment. The cost of producing the defence was a 20% reduction in shoot length growth and biomass production. Willows exposed to mechanical damage had an 8% reduction in shoot length growth compared with control plants, that is, a cost of leaf area removal.
  • • The results are the first quantitative estimates of the cost of a plant defence induced by natural and low amounts (3.3%) of herbivory.


Plants defend themselves against pests and abiotic stresses (Karban & Baldwin, 1997; Larsson, 2002; Strauss et al., 2002). Although some defence traits are always present in the plant (constitutive defences), many are turned on, or increased, only when the plant is under stress (induced defences) (Karban & Baldwin, 1997). The prevalence of inducible defences has been taken as evidence that plant defence traits are costly to produce, that is, that they may divert resources from plant growth and reproduction. This is also a basic assumption in most plant defence theories, dealing with how plants balance possible trade-offs between growth, reproduction and defence (Coley et al., 1985; Herms and Mattson, 1992; Stamp, 2003).

Even though there are several studies concerning the costs of plant defences, few have been able to estimate the costs under realistic conditions (Koricheva, 2002). For example, Zavala et al. (2004) showed in an elegant study that genetically transformed Nicotiana attenuata plants, with low or no trypsin proteinase inhibitor activity, grow faster and produce more seed capsules than plants producing trypsin proteinase inhibitors, and that jasmonate elicitation lowers seed capsule production even further. Common to this and similar studies (Baldwin, 1998; Cippolini, 2007) is that the plant defences have been induced chemically. Using chemical elicitors provides a clever way to circumvent an inherent problem when studying the costs of induced plant defences; the difficulty of discriminating between the costs of damage (e.g. leaf area loss) and the costs of producing the defence. However, plants and their defences evolve in response to naturally occurring damage, and plant responses to chemical induction may differ from those in a ‘natural’ stress situation (Heil, 2002).

In one of the few studies in which ‘natural’ herbivory has been used successfully to estimate the cost of a plant defence, Agrawal et al. (1999) showed that feeding by caterpillars on wild radish plants (Raphanus raphanistrum) induced a defence response (increased concentrations of indole glucosinolates), whilst at the same time producing fewer pollen grains. In this and many other studies, including ones with sophisticated means to mimic herbivory mechanically (Mithöfer et al., 2005), high amounts of damage (> 15% leaf area removal) have been used to induce the defence (Karban & Baldwin, 1997; Valkama et al., 2005; Rooke & Bergström, 2007). Although such high percentages may be common in certain systems, most plants usually experience much lower amounts of damage (< 10%) (Larsson & Tenow, 1980). Previous studies using low percentages of leaf area removal to induce defence responses have not been able to detect any costs (Brown, 1988; Karban, 1993).

The present study addresses the above limitations in experimental design. We used low amounts of natural insect damage to induce a defence response in willows. We discriminated between the cost of losing photosynthetic tissue and the cost of producing the defence; plants experiencing the same amounts of mechanical and natural damage showed no defensive response. In addition, we controlled for variation among plant genotypes by using cuttings originating from the same stem. This method to replicate genotypes among treatments reduced the amount of uncontrolled variation and increased the likelihood of detecting treatment differences.

Our system consists of the willow Salix cinerea and one of its main insect herbivores, the leaf beetle Phratora vulgatissima (Coleoptera, Chrysomelidae). We have previously shown that plants exposed to feeding by adult beetles produce new leaves with an increased trichome density (Dalin & Björkman, 2003). This induction coincides with the feeding of beetle offspring on the same host plants. Willow plants with an induced increase in leaf trichome density become less damaged by leaf beetle (P. vulgatissima) larvae; the mechanism appears to be that the larvae are forced to move around more in search of acceptable feeding sites on hairy leaves (Dalin & Björkman, 2003). The feeding by the overwintered adults in the spring is normally low compared with the damage caused by their larval offspring a few weeks later.

The present study consists of two experiments. First, a relaxation experiment was performed to get indirect support for the idea that the defence is costly to produce. If a defence trait is costly to produce, one would predict plants would return to the initial defence conditions when no longer under attack (Underwood, 1998; Stamp, 2003). Second, a cost experiment was conducted under two different environmental conditions to increase the likelihood of detecting and estimating costs because growth conditions may affect the likelihood of plants investing in defence (Bergelson & Purrington, 1996; Baldwin, 1998; Koricheva, 2002). We measured the growth of plants experiencing no competition and compared it with plants experiencing above-ground competition. Growth was measured in three ways: (i) above-ground biomass produced after damage; (ii) total root biomass; and (iii) shoot length growth after damage.

We hypothesized that:

  • • Willows no longer exposed to adult beetle damage should cease producing new leaves with increased trichome density if not experiencing subsequent damage from larval offspring.
  • • Increase in leaf trichome density is induced by leaf beetle feeding and not by equal amounts of mechanical damage. Therefore, the cost of the induced defence is the difference in growth between plants exposed to beetle and mechanical damage. Hence, the cost of losing leaf area is the difference in growth between undamaged control plants and plants exposed to mechanical damage.
  • • Willows experiencing above-ground competition should invest less in defence than willows not experiencing any competition, because costs of defence have been shown to be higher for plants growing under competition.

Materials and Methods

Two separate studies were performed. The first study (relaxation experiment) was conducted in 2001–2002 and focused on whether leaf trichome density changes over time in relation to damage. The second study (cost experiment) was done in 2002–2003 and considered the effects of leaf damage and above-ground competition on plant growth. In both studies, cuttings (i.e. 20-cm-long pieces of current year shoots) were collected from natural plants of Salix cinerea L. close to Uppsala, central Sweden, in December. Plants grew at least 1 km apart, to minimize the risk of using plants of the same genotype (clone). We used a high number of clones (24 in the relaxation and 20 in the cost experiment) because both growth (Glynn et al., 2004) and trichome responses (Dalin & Björkman, 2003) can vary among willow clones. From each clone, 10–15 cuttings were collected and placed in water for root growth in the laboratory. After c. 20 d, cuttings were planted individually in separate 3 l pots filled with soil (85% peat, 15% sand) containing NPK fertilizer mixture (51 : 10 : 43 + micronutrients; 1.3 kg m−3). Plants were then left to grow for another 20 d. From each clone, the plants most similar in growth and size were chosen for the experiments (four in the relaxation and six in the cost experiment). At the start of the experiments, the average number of shoots was 5.7 (SE = 0.58), the total shoot length was 68 cm (SE = 6.7) and the mean number of leaves was 67 (SE = 7.0) per plant. All studies were conducted in an environmentally controlled glasshouse (T, 20°C; RH, 80%; L/D, 20 : 4 h).

Damage treatments

Experimental plants were randomly assigned to one of two (relaxation experiment) or three (cost experiment) damage treatments: beetle (adults; Phratora vulgatissima L.) damage (BD), mechanical damage (MD), and control (C). In the relaxation experiment, two plants of each clone were exposed to the BD treatment, whereas the other two plants served as undamaged controls (C). In the cost experiment, two plants (one for each competition treatment) were exposed to BD treatments, two plants were exposed to MD treatments, and two served as undamaged controls. The length of each shoot was measured at the start of the experiments and marked at the tip with a felt pen. All plants were kept inside transparent plastic cylinders (height, 70 cm; diameter, 25 cm) with a textile net on the top. Cylinders prevented beetles in the BD treatment from escaping. In the BD treatment, four adult beetles collected from the field were allowed to feed on the plants for 9 d. In the MD treatment, leaves were exposed to clipping with a perforator (diameter, 1 mm). The purpose of the MD treatment was to simulate beetle damage in terms of the amount and distribution but without the induction of trichome responses. We avoided clipping through larger leaf veins on MD plants because adults of P. vulgatissima avoid feeding through large veins (Dalin & Björkman, 2003). MD was inflicted on two occasions (at days 4 and 9) within the same 9 d period as adult beetles were feeding on BD plants. Two occasions were used to simulate beetle herbivory realistically and avoid excessive disturbance of the plants. The total leaf area removed and the distribution of damage on MD plants were determined based on the corresponding damage of BD plants for each clone. To inflict the same amount of damage with the perforator, we first (on day 4) estimated the total amount of damage on BD plants by eye, and then on day 9 we used graph paper to calculate the total leaf area removed in the BD treatment. We then inflicted the same amount of damage to the corresponding MD plant. On average, the total leaf area removed was 3.3 ± 1.5% (mean ± SD).

After 9 d, the cylinders and beetles were removed and the length of each shoot was measured again and the total number of leaves produced was counted. Shoots were again marked at the tip to be able to measure plant growth and to identify the leaves produced during and after damage treatments. Plants in the relaxation experiment were left to grow for 40 d and those in the cost experiment for 20 d.

Relaxation experiment

The purpose of this experiment was to study how leaf trichome density changes in response to damage over a long time period (i.e. 40 d). Leaf trichome density was measured on the underside of leaves by counting the number of trichomes crossing a 2 mm line close to the middle of the leaf (see Dalin & Björkman, 2003 for details). Trichome density was measured on five fully expanded leaves from three levels on each plant: leaves present during damage treatments; leaves produced 10–20 d after damage; and leaves produced 30–40 d after damage. All leaves were sampled at the end of the experiment and classified into three categories based on markings added to the plants.

Plant responses with respect to leaf trichome densities in the relaxation experiment were analyzed with a repeated measure GLM ANOVA with ‘damage’ treatment as fixed and ‘clone’ as the random factor. The MS of the interaction term ‘damage × clone’ was used as error term for the treatment effect. The error MS of the whole model was used when calculating F for the clone effect. The mean number of trichomes from the five leaves at each level from each plant was used as the independent observations. Data were normally distributed and generally met the assumptions for ANOVA.

Cost experiment

Sets of plants, representing one BD, one MD and one C plant, were distributed randomly to two environments. One set of plants from each clone was placed in an environment where they did not compete for resources. There the plants, growing in separate pots, were placed at a distance from each other to avoid any shading effects. The other set of plants from each clone was placed in an environment where they experienced above-ground competition. The three plants, growing in separate pots, were placed together inside an open top cylinder (height, 105 cm; diameter, 40 cm) made of green shade cloth suspended from the ceiling around the plants. The textile cylinders reduced light by c. 80% inside the cylinders (mean was 30 µmol m−2 s−1 inside and 137 µmol m−2 s−1 outside of cylinders; n = 6, SD = 11 and 15, respectively). This amount of light reduction is similar to the relative difference in light experienced by S. cinerea plants in natural habitats, forest vs open farmland (cf. Dalin, 2006). The environmental conditions, except light conditions, were similar for willows growing inside and outside the shading cylinders (21.6°C (SD = 0.6) inside vs 21.3°C (SD = 0.6) outside; and 77.3% (SD = 1.8) inside vs 76.0% (SD = 3.6) outside; n = 3 cages vs paired outside, 12 measurements per cage vs outside). To avoid differences in plant growth resulting from variation in environmental conditions inside the cylinders, all plants were moved one position clockwise once every day. The experiment was terminated after 20 d to increase the likelihood of detecting costs because the relaxation experiment had shown that the induced defence response was relaxed when feeding stopped.

When the experiment was stopped, the length of shoots was measured and the number of leaves counted. Leaf trichome density was measured on five fully expanded leaves from two levels of each plant: leaves present during damage; and leaves produced within 10–20 d after damage. We also measured the leaf areas (see Dalin & Björkman, 2003 for details). Because base leaves had developed before damage treatments, we expected no effects of damage or shading on trichome density or area of the base leaves. As expected, damage and shading did not affect trichome density (damage: F2,40 = 0.01, P = 0.99; shading: F1,40 = 0.66, P = 0.43) or area (damage: F2,40 = 0.60, P = 0.56; shading: F1,20 = 0.06, P = 0.81) of base leaves (see following discussion on statistical analyses). However, we found significant variation among clones in the density of trichomes (F20,40 = 10.60, P < 0.0001) and area (F20,40 = 10.86, P < 0.0001) of base leaves.

Based on the markings made on the shoots, we divided the different plant parts into growth before, during damage treatment and after damage treatments. The length of each part was measured before they were dried in an oven (70°C, 48 h) to measure dry weights. Root biomass (dry weights) was also measured by first removing soil from roots and then removing the roots from the cutting to be dried in an oven (70°C, 48 h).

Plant responses in the cost experiment were analyzed with ANOVA using ‘damage’ and ‘above-ground competition’ treatments as fixed factors, and ‘clone’ as a random factor. In the analysis of trichome responses, ‘leaf area’ was used as covariate. From each plant, we used the mean number of trichomes from the five leaves as independent observations. In the analyses of growth responses, no covariate factor was used. The mean square for the interaction between the fixed and the random factor was used as error term when calculating F-values for the fixed factors. Data were normally distributed and generally met the assumptions for ANOVA and planned pairwise comparisons.

To further illustrate the variation in response to damage treatments (i.e. among willow clones), we present a figure showing the reaction norms for above-ground biomass produced after treatment. The results are presented separately for the two competition treatments.


The relaxation experiment showed that beetle feeding induced an increase in leaf trichome density in leaves produced 10–20 d after initial damage and that willows no longer exposed to herbivore damage stopped producing new leaves with increased trichome density after 30–40 d (Fig. 1, Table 1; significant time × treatment interaction). The leaf trichome levels of control and beetle-damaged (BD) plants appeared to be similar before and 30–40 d after damage treatment (Fig. 1). Clones responded differently to damage treatment. No significant interaction was found between time and clone (Table 1), indicating that the time course of the response was not different. We found a significant effect of time on leaf trichome density in this experiment (Table 1), showing that the response of BD plants 10–20 d after damage ceased was strong enough to affect the whole model.

Figure 1.

Mean number of trichomes (+ SE) crossing a 2 mm line on (five) leaves of undamaged (control, C) Salix cinerea plants and plants damaged (3.3% leaf area removal) by adult leaf beetles (Phratora vulgatissima) (BD). Leaves are divided into three groups based on when they were produced in relation to beetle damage.

Table 1.  Effects of treatment (control vs beetle damage), clone (n = 24) and time (0, 10–20 and 30–40 d after damage) and their interactions on leaf trichome density of Salix cinerea plants exposed to low amounts of leaf beetle (Phratora vulgatissima) damage (3.3% leaf area removal) as revealed by a repeated-measures analysis of variance
Source of variationdfms or Wilks’λ*FP
  • *

    MS for between subjects and Wilks’λ for within subjects.

Between subjects
Treatment 1205.726.3< 0.0001
Clone2365.269.9< 0.0001
Treatment × clone237.838.40< 0.0001
Within subjects
Time 2, 470.065339.0< 0.0001
Time × treatment 2, 220.099100.1< 0.0001
Time × clone46, 940.3451.440.07
Time × treatment × clone46, 940.3401.460.06

The cost experiment showed that beetle feeding induced an increase in leaf trichome density whereas mechanical damage did not (Fig. 2; Table 2). The response was stronger for plants experiencing competition (Fig. 2), as revealed by a significant interaction between damage and competition (Table 2). Pairwise comparisons revealed that plants experiencing BD had a leaf trichome density that was 115% higher than MD plants in the competition treatment (F1,37 = 31.1, P < 0.0001), whereas BD plants had 75% higher trichome density than MD plants in the no competition treatment (F1,37 = 30.2, P < 0.0001). Trichome density did not differ between C plants and MD plants in either competition treatment (F1,37 = 1.07, P = 0.31 and F1,37 = 1.46, P = 0.23, respectively). Leaf trichome density was affected by clone (Table 2). No significant damage × clone interaction was found (Table 2), suggesting that all clones responded to beetle feeding in the same way – by increasing the density of trichomes. The significant competition × clone interaction (Table 2) suggests that clones differed in their response to the competition treatment.

Figure 2.

Mean number of trichomes (+ SE) crossing a 2 mm line on (five) leaves of Salix cinerea plants previously exposed to three types of damage treatments: control (no damage, C), mechanical damage (MD) and beetle (Phratora vulgatissima) damage (BD). Open bars, plants not experiencing competition; closed bars, plants experiencing above-ground competition.

Table 2.  Effects of leaf area (covariable), damage treatment (control vs mechanical vs beetle damage), competition treatment (no competition vs above-ground competition), clone (n = 21), and two-factor interactions on leaf trichome density of Salix cinerea plants exposed to low amounts of leaf beetle (Phratora vulgatissima) damage (3.3% leaf area removal) as revealed by a GLM-ANOVA
Source of variationdfmsFP
  1. The three-factor interaction was not significant and excluded. Model R2 = 0.929.

Leaf area
Damage treatment 2437.540.0< 0.0001
Competition treatment 189.75.430.030
Damage × competition 228.93.430.042
Clone2087.510.4< 0.0001
Damage × clone4010.91.300.209
Competition × clone2016.51.960.035

When above-ground biomass produced after induction was used as a measure of fitness in the cost experiment, we found a significant effect of damage treatment (Fig. 3a; Table 3). We found that BD plants produced 20% (95% confidence interval: 2–37%) less biomass than MD plants (Fig. 3a; F1,39 = 7.90, P = 0.0077). However, MD plants did not differ significantly from C plants (Fig. 3a; F1,39 = 0.09, P = 0.76). For above-ground biomass production damage treatment was not found to interact significantly with any other traits. However, clone did interact significantly with competition (Table 3), indicating that clones differed in growth responses to competition. On average, competition decreased above-ground biomass production by 77% among control plants (Fig. 3a; Table 3). The reaction norms of individual clones with respect to above-ground biomass production are presented in Fig. 4, separated by competition treatments. It should be noted that including or excluding the outlying clone did not affect the statistical outcomes of the analysis and, thus, not the conclusions.

Figure 3.

Growth responses of Salix cinerea plants exposed to three types of damage treatments: control (C, no damage), mechanical damage (MD) and beetle (Phratora vulgatissima) damage (BD). Open bars, plants not experiencing competition; closed bars, plants experiencing above-ground competition. Three measurements of growth are presented: above-ground plant biomass (g DW) produced after damage; total root biomass produced (g DW); shoot length growth after damage (cm). Error bars indicate SE.

Table 3.  Effects of damage treatment (control vs mechanical vs beetle damage), competition treatment (no competition vs above-ground competition), clone (n = 21), and two-factor interactions on above-ground biomass production (a), below-ground biomass (b) and shoot length growth (c) of Salix cinerea plants after being exposed to low amounts of leaf beetle (Phratora vulgatissima) damage (3.3% leaf area removal) as revealed by a GLM-ANOVA
Source of variationdf(a) Above-ground biomass(b) Belowground biomass(c) Shoot length
  1. The three-factor interactions were not significant and excluded. Model R2 = 0.975 (a), 0.954 (b) and 0.928 (c).

Damage treatment 269.620.5< 0.00011.3916.3< 0.0001 35 42821.2< 0.0001
Competition treatment 1280563.1< 0.000146.4155.3< 0.0001460 91366.9< 0.0001
Damage × competition 223.26.380.00410.899.670.0004   16670.900.41
Clone2062.217.1< 0.00010.687.33< 0.0001 10 5865.70< 0.0001
Damage × clone403.400.930.590.0850.920.60   16690.900.63
Competition × clone2044.512.2< 0.00010.2993.240.0005   68863.710.0002
Error39138  0.092     1857  
Figure 4.

Reaction norms of Salix cinerea clones with respect to above-ground biomass produced after damage in response to three damage treatments (control, C; mechanical damage, MD; and beetle (Phratora vulgatissima) damage, BD) separated by two competition treatments: (a) no competition; (b) above-ground competition. Each datapoint is based on a single clonal replicate in each treatment.

We found a significant effect of damage treatment on below-ground biomass in the cost experiment (Fig. 3b, Table 3). The effect differed between competition treatments (significant damage × competition interaction; Table 3). For competing plants, there were no significant differences in pairwise comparisons (P > 0.15). Among uncompeting plants, BD plants had a significantly lower below-ground biomass than MD plants (F1,40 = 13.9, P = 0.0006), whereas MD and C plants did not differ (P = 0.14). We found a significant difference among clones, between competition treatments and for the interaction between competition and clone (Table 3). Clone did not significantly interact with damage treatment (Table 3).

The shoot length growth differed between damage treatments (Fig. 3c, Table 3) but there was no significant interaction between damage and competition (Table 3). The overall mean reduction in shoot length growth among BD plants in comparison with MD plants was 22% (95% confidence interval, 12–32%; F1,39 = 18.36, P < 0.0001). Expressed in length, the reduction was, on average, 32 cm. Contrary to what was found for above-ground biomass, we estimated a reduction in shoot length growth of MD plants in comparison with control plants, which on average was 8% (95% confidence interval, 0–20%; F1,39 = 7.53, P = 0.0091). This corresponds to a reduction in total shoot length of 24 cm. Competition reduced shoot length growth by 49% viewed over all three damage treatments (Fig. 3c; Table 3). Clones differed significantly in shoot length growth (Table 3). Different clones responded similarly to damage (nonsignificant damage × clone interaction; Table 3) but not to competition (significant competition × clone interaction; Table 3).


We found that feeding by adult leaf beetles induced an increase in leaf trichome density, whereas the same amount of mechanical damage did not (as previously shown in Dalin & Björkman, 2003). Studies in other systems have yielded both similar results (Agrawal, 1999, 2000) and results showing an equally strong response to mechanical damage (Pullin & Gilbert, 1989; Traw & Bergelson, 2003). Variation in the amount of damage may be one reason for these discrepancies. The relatively low amount of damage used here (3.3%) has seldom been used elsewhere, although damage percentages in nature do not commonly exceed 10% (Barbosa & Schultz, 1987; Coupe & Cahill, 2003). In contrast to the results presented here, previous studies using low amounts of natural herbivory have not been able to demonstrate any costs of defence (Brown, 1988; Karban, 1993). It has been suggested that the response of willows to adult beetle feeding serves as a defence against subsequent larval damage, which is normally more severe than the damage by adults (Dalin & Björkman, 2003).

Despite using naturally low amounts of herbivory, the costs found here were high compared with costs found in many other systems (Strauss et al., 2002). One reason for high costs could be that damage often not only removes photosynthetic tissue but also indirectly affects photosynthesis negatively (Zangerl et al., 2002). In addition, other induced responses not measured by us, for example, some chemical change not affecting leaf beetle performance negatively (cf. Dalin & Björkman, 2003; Dalin et al., 2004), may have contributed to the cost measurements. Moreover, the high costs in the present study could result from the fact that we studied a structural plant defence (cf. Skogsmyr & Fagerström, 1992) whose production directly depends on growth processes. Further studies are therefore needed to evaluate the benefits of the induced response and whether the benefits can outweigh the costs detected in this study.

The stronger defence response found among willows experiencing competition is not as expected since the costs of defence are likely to be higher for plants growing under competition (Koricheva, 2002). The observed response may, however, be adaptive because plants may have a limited ability to tolerate herbivory when light (or other) resources are in short supply and, instead, invest resources into defence to reduce further damage (Herms & Mattson, 1992). The ‘competition’ treatment used here to attain severe resource depletion may have enhanced the difference among treatments if the control plants were able to ‘outcompete’ the other plants in their search for light. How commonly such situations occur in nature is not known to us.

The stronger defence response to beetle damage than to mechanical damage may indicate an adaptive response; that is, beetle defoliation triggered an induced response, resulting in decreased growth, whereas mechanical defoliation did not. The fact that the same pattern was observed in both competition treatments strengthens this conclusion. However, the relatively high variability among clones within the BD treatment suggests that the selection pressure for these responses is not exceedingly high. Furthermore, the significant interactions between competition and clone for all growth parameters and trichome production show that clones respond differently to environmental conditions. The ability of plant genotypes to respond differently to environmental conditions (e.g. herbivory or competition) has recently generated considerable interest (Bradshaw, 2006). The current interest in phenotypic plasticity is partly arising as a result of evidence showing that trait-mediated interactions may be as important as density-mediated interactions in many ecological systems (Callaway et al., 2003; Bolnick & Preisser, 2005) and partly as a result of an urge to predict the response of species to global change (Thuiller et al., 2005; Carroll et al., 2007). Our results suggest that the responses to herbivory and competition in S. cinerea have not become fixed within populations and that the selective advantage probably varies with environmental conditions.

The relaxation experiment gave strong indirect support for the hypothesis that costs are associated with the induced defence response in the willow species S. cinerea. A more rigorous test of relaxation would, however, need also to include a treatment where plants experiencing continued herbivory continued to produce leaves with an enhanced trichome density. Results similar to the ones presented here have been reported from grey alder (Alnus incana) fed upon by leaf beetles (Baur et al., 1991), but sometimes the relaxation may take several years (Huntzinger et al., 2004). No general patterns can be discerned, though, as there are few studies of relaxation time. The mechanism behind the observed plasticity in trichome density in S. cinerea may be that the number of epidermal cells that produce trichomes changes in response to environmental conditions when leaves are developing or that the number of epidermal cells is constant and the proportion of cells that become trichomes varies with the conditions, as has been shown in Arabidopsis (Traw & Bergelson, 2003).

Growth is a good indirect measure of plant fitness (Karban, 1993; Agrawal, 2000). In fact, Koricheva (2002) showed that fitness costs of plant defences did not depend on whether fitness was measured as growth or reproduction. Growth can be measured as both biomass gain and shoot length. Biomass gain represents resources accumulated for use in reproduction, whereas shoot length affects the competitive ability of plants when light is a limiting resource. Reductions in both biomass and shoot length, caused by herbivory or environmental conditions, is therefore likely to have significant negative effects on plant fitness. In this study, we detected a cost of mechanical defoliation for shoot length growth but not for biomass gain. One reason behind this difference could be that the plants gained similar biomasses but with reduced shoot lengths in the leaf area removal treatment. Although there are other plausible reasons, the results show that the cost of leaf area removal was relatively small compared with the costs of induced trichome productions.

The lack of evidence for costs of leaf area loss (MD) in terms of above-ground biomass production could have several explanations: the cost of leaf area removal is too small to be detected by our methods; the timing of the mechanical leaf area removal is of importance (Mithöfer et al., 2005); and plant responses vary according to different kinds of damage and, instead of allocating resources into defence, compensate for certain kinds of damage (tolerance). The first explanation is less likely because real herbivory of the same magnitude elicited a strong response; the second is plausible, but the importance of the time-lag used here of 4 d (over a 40 d period) between the start of beetle and mechanical damage can only be evaluated in further experiments; the third explanation appears adaptive but there are no data available to substantiate it. A fourth possible explanation, that below-ground growth was more affected by mechanical damage than above-ground growth (Agrawal, 2000), was not supported by our data. The overall low investment in root biomass among competing plants may have limited our possibilities of detecting differences among damage treatments. Similarly, because we could not discriminate between root biomass produced before and after damage, noteworthy effects may have been missed. Nevertheless, we found evidence of costs in terms of reduced root biomass of induced plants.

The cost estimates presented here are unique because it is the first study in which the following aspects have been controlled for simultaneously: First, and primarily, the system allowed separation of the effects of damage and induction since leaf area removal, per se, did not induce trichome defence in S. cinerea. Second, it was possible to detect a cost induced by a low amount of damage, comparable to that usually found in many plant species under natural conditions. Third, we were able to consider variations among plant genotypes by using plants raised from clonal cuttings. We thus conclude that the cost of producing defence is higher than that of losing leaf area.


We thank Karin Eklund and Astrid Klementsson for assistance; Staffan Karlsson for statistical advice; Anurag Agrawal, Helena Bylund, Carolyn Glynn, Richard Hopkins, Julia Koricheva, Stig Larsson, Tod Osier, Sonia Sultan and anonymous reviewers for helpful comments on the manuscript. This work was supported by the The Carl Trygger Foundation, The Swedish National Energy Administration and The Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (FORMAS).