Plant Species

Plantago lanceolata L. (ribwort plantain) is a rosette-forming, self-incompatible, perennial plant with a worldwide distribution and large ecological amplitude. In The Netherlands, it is a common plant of roadsides and moist and dry grassland (Haeck 1992). Among the secondary plant compounds produced by P. lanceolata are the two IGs aucubin and catalpol (Duff et al. 1965; Bowers and Stamp 1992; Adler et al. 1995).

Selection Lines

The selection lines used in this experiment were derived from an artificial selection experiment on leaf IG concentration in P. lanceolata for four generations. A brief description of the selection procedure is presented below. More details can be found in Marak et al. (2000). Seeds from 20 pairwise reciprocal crosses between plants from two different source populations were used to constitute a base population of 200 plants (five per maternal parent from each cross) that were completely randomized and grown in a greenhouse under 16h/8h light/dark at 22°C/18°C day/night. At a standardized plant age (five weeks from transplanting), the third and fourth pair of true leaves produced on the main rosette of each plant was harvested and processed for analysis of the concentration of the IGs aucubin and catalpol using high performance liquid chromatography (HPLC). Plants were selected on the basis of the total (aucubin + catalpol) IG content in these leaf pairs. From the 200 plants of the base population, 50 plants with the highest and lowest levels were selected to initiate lines with directional upward (H) and downward (L) selection, respectively, while 50 plants with intermediate levels (i.e., a mean value equal to that of the population mean), were selected to initiate a line with stabilizing (M) selection. Selected plants of each line were transferred to closed Plexiglas cages at anthesis and spikes were shaken daily to ensure that pollen was well distributed among plants within cages. Unidirectional selection within the L and H line and stabilizing selection within the M line was continued for four generations. For each line, seeds representing the fourth generation were collected by maternal parent, resulting in 50 seed families available per selection line, which, given the random open pollination system, are expected to represent predominantly maternal half-sib families. A significant response to selection was observed, resulting in an almost four-fold difference in IG concentration of the leaves that had been under direct selection and a more than two-fold difference in leaf IG concentration on a whole-plant basis.

Fitness Experiment

For each selection line, a random selection of 12 of the 50 available families was made. Forty seeds of each family from the high (H) and low (L) line and 100 seeds of each family from the stabilizing selection (M) line were germinated on moistened glass beads in a growth cabinet (14h/10h light/dark; 25°C/15°C). After 12 days, seedlings were transplanted into plastic pots (diameter 13 cm, height 9.5 cm) with a mixture of sand and compost (95%: 5%, weight ratio). Plants were subjected to four different treatments (environments), consisting of a factorial combination of two competition and two nutrient treatments. The competition treatments consisted of (1) a single treatment, in which one seedling of each family of the H and L line was individually planted per pot and grown singly, and (2) a competition treatment, in which one seedling of each family of the H and L line (target plant) was planted individually in the center of a pot and surrounded by four seedlings obtained from four randomly selected families of the M line (competitors). The nutrient treatment consisted of the application of high- and low-fertilization conditions. During the first two weeks after transplanting, every pot received 50 ml of a 1/32 strength Hoagland's solution twice a week. Full strength solution contained: 5 mM Ca (NO₃)₂, 5 mM KNO₃, 1 mM KH₂PO₄, 2 mM MgSO₄, 174 μM C₁₀H₁₂FeN₂O₈Na, 93 μM H₃BO₃, 18 μM MnCl₂, 1.5 μM ZnSO₄, 0.6 μM CuSO₄ and 1.0 μM Na₂MoO₄. During the next five weeks, half of the pots of each family and competition treatment received 50 ml of a 1/16 strength poor solution three times a week, while the other half received 50 ml of a 1/2 strength rich solution three times a week. During the last seven weeks, pots of the low- and high-nutrient treatment received 100 ml of a 1/8 and 1/2 strength solution three times a week, respectively. Total size of the experiment was 480 pots, five replicates for each of the 24 plant families (H and L) and four treatment combinations. Plants were grown in a greenhouse (16h/8h light/dark and 22°C/18°C day/night) in five physical blocks each containing one replicate per family and treatment combination. Pots were completely randomized within each block.

One week after transplanting (i.e., before nutrient treatments were effectuated), we measured the cotelydon length of all seedlings of the H and L lines (i.e., all seedlings in the single treatment and target plants in the competition treatment) and of one of the four competitors of the M line in the treatment. For plants from the H and L line, date of appearance of the first flower bud was recorded during the experiment. After 14 weeks, when leaves started to wither and seeds started to drop from the spikes, we harvested all plants from the H and L line and recorded the length and width of their longest leaf, the number of leaves in the main and side rosettes, and the number of spikes. As plants were harvested before all spikes had reached maturity, each spike was classified according to its developmental stage (bud, immature, female flowering, male flowering, and producing seeds), and its length was recorded. For the longest spike, the number of flowers was counted and 10 seeds resulting from open pollination were used to calculate average seed weight. Aboveground parts of plants were separated into leaves in the main rosette, leaves in the side rosettes, spikes, and stalks and their dry weights (oven-dried for three days at 70°C) were weighed separately. In the competition treatment, roots from target plants and competitors could not reliably be separated. Therefore, roots were only harvested from singly grown plants. For these plants, root dry weight was measured as well. Plants from the stabilizing selection (M) line used as competitors were discarded.

Chemical Analyses

To correlate family means for biomass production with family means for leaf IG concentration, leaf IG levels were analyzed using HPLC. From each plant, the dry leaf material was finely ground using a blender. For each family and treatment combination, leaf material from the five blocks was pooled by mixing 50 mg of ground leaf material from each of the five individual plants. From each mixture, 50 mg was extracted in 10 ml of 70% MeOH while shaking overnight. The crude extract was filtered and diluted with ultrapure water. The concentrations of the IGs aucubin and catalpol were analyzed by HPLC using a Dionex (Sunnyvale, CA) DX 500 equipped with a GP40 gradient pump, a Carbopac PA1 guard (4 × 50 mm) and analytical column (4 × 250 mm), and an ED40 electrochemical detector for pulsed amperimetric detection (PAD). NaOH (1 M) and ultrapure water were used as eluents (10:90%, 1 ml/min). Retention times were 3.12 min and 4.57 min for aucubin and catalpol, respectively. Concentrations were analyzed using Peaknet Software Release 5.1 (Dionex Corp.).

Statistical Analyses

Effects of block, selection line, family (nested within selection line), nutrient level, and competition treatment on plant traits were analyzed using generalized linear models with a normal error distribution for the response variables and an identity link function (SAS ver. 6.12, procedure GENMOD; SAS Institute 1996). Interactions with block were generally not significant and were pooled with the error term (Newman et al. 1997). All independent variables were treated as class variables. Dependent variables were transformed prior to analysis if necessary to improve normality. Leaf number, dry weight of vegetative plant parts, total aboveground plant dry weight, and total plant dry weight were ln-transformed. Number and weight of reproductive parts were ln(x + 1) transformed. Leaf IG concentration and the product of leaf length and leaf width of the longest leaf (hereafter referred to as leaf area) were square-root transformed. Significance of differences was tested using Type III likelihood-ratio statistics. Effects of nutrient or competition treatment on the magnitude of fitness costs of chemical defense were inferred from the significance of line-by-environment interactions for reproductive or total plant biomass. To estimate effect sizes required to detect such interactions, simulations were run in which we varied the ratio of costs experienced at the two levels of an environmental factor, without changing the observed average fitness value across the two lines at each of the two levels of the environmental factor or changing the observed average cost (proportional reduction of the fitness trait at the high as compared to the low line) across the two levels of the environmental factor. Generalized linear models were run for increasing cost ratios until the interaction term was significant.

Because data for plant roots were not available for the competition treatments, root and total plant biomass were analyzed separately. None of the plants reproduced in the presence of competitors at the low nutrient level (poor-competition treatment). Therefore, a new variable, environment, was constructed with three levels, rich-single, rich-competition, and poor-single to replace nutrient level and competition treatment in analyses of reproductive traits. Because IG levels were determined for pooled samples from the five replicates per family and treatment combination, there was no replication at the family level and differences in leaf IG concentration between selection lines were tested directly over the error term.

Simple path analyses (see Fig. 3) based on family means for cotyledon length, leaf IG concentration, and final plant size (aboveground dry weight at harvest) in each of the environments were used to evaluate direct effects of IGs on plant performance from indirect effects of IGs mediated by correlations with cotyledon length. Direct effects of cotyledon length and IGs were calculated as standardized partial regression coefficients from multiple regressions of cotyledon length and leaf IG concentration on final plant size.

Differences in Leaf Iridoid Glycoside Levels between Selection Lines

In accordance with previous results, leaf IG concentrations were approximately twofold higher in plants from the high-selection line than in plants from the low-selection line (F_1,88 = 281.1, P < 0.001; Fig. 1A). The magnitude of the difference in leaf IG level between the selection lines was independent of nutrient level (no line-by-nutrient interaction, F_1,88 = 0.09, P = 0.76) and the presence or absence of competitors (no line-by-competition interaction, F_1,88 = 0.05, P = 0.82), but there was a significant interaction between the effects of nutrient level and competition on leaf IG concentration (F_1,88 = 96.1, P < 0.001). In the absence of competitors, leaf IG concentration increased with decreasing nutrient level in both lines, but under competition leaf IG concentrations increased with decreasing nutrient level (Fig. 1A).

Costs of Iridoid Glycosides

In our experiment, detection of costs of chemical defense requires that plants from the high-IG selection line show lower performance in the absence of herbivores or pathogens than plants from the low-IG selection line. There were no significant differences in total leaf dry weight between the two selection lines (Table 1, Fig. 1D). Plants from the high line had a significantly lower number of leaves per plant (Table 1, Fig 1C), but the leaf area of their longest leaf was significantly larger than for plants from the low-selection line (Table 1, Fig. 1B). The absence of differences in total leaf dry weight is consistent with a perfect trade-off between number and size of leaves among the selection lines. In contrast to the similarity in leaf mass, plants from the high line overall produced fewer spikes, had a smaller total spike length, and had a lower reproductive dry weight than plants from the low line (Table 2, Fig. 1F, G), indicating a significant reproductive cost of IG production. The traits that contributed to lower reproduction of plants from the high-IG line differed among environments. Under nutrient-rich conditions in the presence of competitors (rich-competition), lower spike production of high-IG plants was caused both by a lower proportion of plants that reproduced and by a lower spike production per reproductive plant (Table 3). In the absence of competitors, only differences in spike production per reproductive plant among selection lines were statistically significant (Table 3). Due to their similar leaf mass but lower reproductive dry weight, plants from the high-selection line also had a lower total aboveground dry weight than plants from the low-selection line (Table 1, Fig. 1H). Under noncompetitive conditions, which allowed us to harvest roots from individual plants, root weight did not differ between the selection lines (F_1,22 = 1.23, P = 0.279) and the total dry weight (including roots) of plants from the high-selection line was significantly lower than the total dry weight of plants from the low line under these conditions (F_1,22 = 10.81, P = 0.003).

A potential caveat in the interpretation of these costs is that the lower dry weight of plants from the high line compared to plants from the low line in the absence of herbivores and pathogens could in fact be due to their associated growth form (larger but fewer leaves) rather than to their higher level of IGs (Table 1, Fig. 2). We therefore reran the analysis of aboveground dry weight with total leaf number and leaf area as covariates. Line effects were still significant (F_1,22 = 7.62, P = 0.012), indicating that at least part of the costs is not mediated by differences in growth form.

Because plants were harvested before all spikes had reached maturity, we could not measure costs in terms of seed production. However, there was no indication that the lower number of spikes produced by plants from the high line would be compensated by a higher reproductive output per spike. First, the length of individual spikes did not differ between plants from the high and low line for any of the developmental stages (line-effect in ANOVAs of stage, line, and family within line: all P > 0.20). Also, plants from the high line did not produce a larger number of flowers per unit mature spike length (mean ± SE: 2.17 ± 0.07 flowers/mm) than plants from the low line (2.13 ± 0.04; F_1,22 = 0.10, P = 0.751), nor did they have a larger individual seed weight (2.10 ± 0.07 and 1.97 ± 0.07 mg, respectively, F_1,22 = 0.85, P = 0.367). Second, there is no indication that the larger number of spikes on plants from the low line mainly consisted of a large number of immature spikes that might never reach maturity. Under nutrient-rich conditions, the stage distribution of spikes was similar for spikes produced by the high and the low line (no competition: χ² = 2.12, P = 0.145; competition: χ² = 0.349, P = 0.554). We therefore expect that differences in spike production under these conditions will be reflected in differences in seed production between the two selection lines. Under low nutrient conditions, plants from the low line did not only produce more spikes per reproductive plant but these spikes were on average also in a more advanced developmental stage (χ² = 6.38, P = 0.012), despite the absence of a significant difference in the onset of reproduction between reproductive plants from the two lines under these conditions (Fig. 1E). Potentially, this may result in an even larger difference in seed production between plants from the high and low line than expected on the basis of differences in spike number, if the length of the season limits maturation of the slower developing spikes.

Environmental Impact on Costs of Iridoid Glycosides

As intended, the nutrient and competition treatments that were used to investigate the impact of environmental stress on fitness costs of chemical defense clearly resulted in very different growth conditions for the plants (Fig. 1). Each of the factors nutrient and competition stress separately resulted in an average reduction of total aboveground dry weight by a factor of about 4, while a combination of both led to an average reduction by a factor of about 16 (Fig. 1H). Under the most unfavorable conditions for growth, that is, under both nutrient and competition stress, none of the plants initiated reproduction.

For vegetative traits, the magnitude of differences between the two selection lines was generally smaller under less favorable conditions. Line-by-nutrient interactions were significant for leaf area of the longest leaf (Table 1), due to the smaller magnitude of differences between the two selection lines under low-nutrient conditions (Fig. 1B). Significant line-by-competition interactions were found for both leaf area of the longest leaf and total number of leaves (Table 1), again reflecting smaller differences between selection lines when plants were grown in competition (Fig. 1B, C). However, line-by-environment interactions for leaf dry weight (Table 1), reproductive dry weight (Table 2), and total aboveground dry weight (Table 1) were not significant. This indicates that we were unable to detect significant differences in fitness costs of defense under different environmental conditions. Only at the family level, differential responses in reproductive dry weight to environmental conditions could be detected (Table 2, environment-by-family interaction for reproductive dry weight).

The results thus do not provide evidence that the magnitude of fitness costs increased under competition or nutrient stress. If we compare the average values for costs under different environmental conditions, it appears that the direction of the (nonsignificant) difference in costs between nutrient levels is even opposite to expectation. At the high-nutrient level, plants from the high line had on average 12.8% and 14.6% lower aboveground dry weight than plants from the low line, in the absence and presence of competitors, respectively. At the low-nutrient level, these reductions were 7.8% and 5.5%, respectively. Costs were thus on average 1.6 and 2.6 times higher at high- than at low-nutrient supply, respectively. In other words, contrary to expectation, costs on average decreased with nutrient stress, though the difference was not statistically significant (line-by-nutrient interaction F_1,22 = 1.99, P = 0.17, Table 1). The lack of significant differences in costs was partly due to the low power to detect such interactions with the current set-up involving families within lines. Analysis of effect sizes required to detect line-by-nutrient interactions revealed that costs should have been 2.6 and 4.1 times higher, respectively, to be detected in the current experiment.

Among-Family Correlations between Iridoid Glycosides, Seedling Size, and Final Plant Size

In accordance with the decreasing statistical significance of costs of chemical defense under increased nutrient and competition stress (Fig. 1H), negative correlations between family means for leaf IG concentration and family means for aboveground dry weight were only significant under the environmental conditions that were most favorable for growth and reproduction (rich-single, Fig. 3A), marginally significant under nutrient or competition stress (Fig. 3B,C) and not significant under simultaneous stress (Fig. 3D).

For plants grown in the absence of competitors, costs of defense on average decreased under nutrient stress even though the absolute difference in IG production between the selection lines was larger under nutrient stress (cf. Fig. 3A and 3C). For plants grown under competition, part of the more modest difference in dry weight between the high- and low-selection line could be explained by the smaller difference in IG level between the selection lines (cf. Fig. 3B and 3D).

Costs were also not enhanced under competition stress. One possible explanation for the absence of increased costs under more competitive conditions could be that high-IG plants on average have a larger seedling size. Cotyledons of plants from the high line (mean ± SE) were significantly longer (61.0 ± 0.8 mm) than cotyledons of plants from the low line (54.5 ± 0.7 mm; Table 1). This implies that plants from the low line had a size disadvantage at the start of the experiment relative to plants from the high line. Cotyledons of plants from the low line were on average 4.8 ± 1.3 mm smaller than cotyledons of plants from the medium line with which they were competing (pairwise t-test, t = 0.367, df = 119, P < 0.001). Cotyledons of plants from the high line were on average 1.4 ± 1.5 mm larger than cotyledons of plants from competitors, but the difference was not significant (t = 0.95, df = 119, P = 0.343).

The larger cotyledons could give plants from the high line an initial size advantage relative to plants from the low line that could especially become important under competitive conditions. Thus, even if direct costs of high leaf IG production are higher under competitive conditions, they might be ameliorated by pleiotropic effects on initial plant size that are expressed under such conditions. A path analysis of the direct and indirect effects of cotyledon size and IG level on aboveground dry weight, based on family means, only partly supports this idea (Fig. 4). Under nutrient-rich conditions, the association between cotyledon size and IG level appeared to be weak, and the indirect positive effects of IGs on aboveground weight through associated larger initial plant size were insufficiently large to counteract the direct costs incurred by higher leaf IG concentrations. Only under nutrient-poor conditions, higher IG levels were significantly correlated with larger cotyledon size (Fig. 4C, D). In the absence of competitors (Fig. 4C), the direct contribution of cotyledon size to final plant size was small, and the direct cost of IGs (path coefficient −0.51) was only partly offset by associated larger cotyledon size (+0.14). In the presence of competitors (Fig. 4D), larger cotyledon size did make a significant direct contribution to final plant size, resulting in a positive indirect effect of high leaf IG concentration on final plant size through associated larger cotyledon size. Due to this positive indirect effect of associated larger cotyledon size and the only marginally significant direct cost of IGs (path coefficient −0.38, P = 0.067, Fig. 4D), no significant correlation between leaf IG concentration and final plant size was observed under these conditions (Fig. 3D).

Costs of Chemical Defense under Benign Conditions

Fitness costs of secondary plant metabolite production are defined as trade-offs between the production of secondary plant compounds and other fitness-enhancing traits in the absence of herbivores or pathogens. In our study, we found that there is a cost of chemical defense, that is, of IG production in P. lanceolata. Plants selected for a high leaf content of IGs had a lower plant dry weight than plants from the low-selection line. The differences in plant dry weight between the two selection lines could mainly be attributed to differences in reproductive dry weight. Plants from the low line produced more inflorescences than plants from the high line, and there was no indication that this would be offset by a lower reproductive output per inflorescence. This result is in agreement with other studies reporting reproductive costs of chemical defense by secondary metabolites such as furanocoumarins (Berenbaum et al. 1986; Zangerl and Berenbaum 1997). Like furanocoumarins, terpenoids rank among the most expensive classes of secondary metabolites in terms of construction costs. Producing 1 g of the IG aucubin requires 2.39 g of glucose; the biosynthetic costs of 1 g of the furanocoumarin psoralen are 3.39 g of glucose (Gershenzon 1994). But also in studies of secondary plant compounds with relatively low construction costs, such as leaf tannins in the Neotropical tree Cecropia peltata, significant costs of defense in terms of rates of leaf production have been detected (Coley 1986).

Given the suite of correlated morphological responses to selection on leaf IG concentration in P. lanceolata, the significantly lower aboveground dry weight of high-IG plants under benign environmental conditions should be interpreted with care. Plants from the upward-selection line produced fewer but larger leaves (Fig. 1). This confirms results from a previous study of correlated responses to selection using plants from these selection lines (Marak et al. 2000). Leaf IG levels are genetically correlated with a suite of morphological traits that are involved in differences between previously described “hayfield” and “pasture” ecotypes in P. lanceolata (Van Groenendael 1986; Wolff 1987; Van Tienderen and Van der Toorn 1991; Van Hinsberg and Van Tienderen 1997). Selection for high levels of IGs resulted in plants in which fewer meristems developed into organs of larger size (hayfield type), whereas selection for low IG levels resulted in plants with many small leaves and many side rosettes (pasture type) that occur in more open, grazed areas, even though the base population for selection constituted neither distinct hayfield nor distinct pasture plants. The lower biomass of high-IG plants could therefore either reflect a direct cost of IG production or a cost associated with one of the genetically correlated traits. However, because costs were still significant when the two main correlated traits, leaf number and area, were used as covariates, we conclude that there is a significant cost of IG production that is not mediated by associated differences in growth form.

A second potential concern in the interpretation of the observed cost of chemical defense in P. lanceolata is that this species is a perennial. Fitness costs of defense in terms of reduced first-year reproduction may therefore be offset by increased reproduction in ensuing years, if there is a cost of reproduction. Because the experiment was limited to first-year plants, we can only conclude that production and maintenance of high leaf IG concentrations incurs a cost up to this point in time. However, the vegetative biomass of plants from the high- and low-IG line was quite similar. Hence, there was no indication that the higher reproduction of low-IG plants was due to an altered allocation pattern in favor of investment in reproductive rather than vegetative biomass, which often underlies such costs of reproduction.

Our results are at variance with those of previous studies in P. lanceolata, in which no costs of IGs were detected in terms of a reduction in aboveground biomass or plant relative growth rate using phenotypic (Bowers and Stamp 1992; Darrow and Bowers 1997) or among-genotype correlations (Adler et al. 1995) of plants that were grown outdoors. The reason for this discrepancy is unclear. The range of genotype means (about fivefold variation in IGs) among the 30 genotypes used in the study of Adler et al. (1995) was roughly similar to the range among our 24 families. Possibly, the differential results reflect differences in genetic background of the material, their use of a field study (in contrast to our greenhouse study) that potentially results in higher variances in both biomass and defense levels, or their lower overall extent of reproduction, the stage in which costs in our study were most clearly expressed.

Environmental Effects on Fitness Costs

We found no evidence that costs of chemical defense were significantly affected by the environment. On average, costs in terms of the proportional reduction in aboveground dry weight (of plants from the high vs. low line) were lower under combined nutrient and competition stress (5%) than under the most favorable environmental conditions (13%), but none of the line-by-environment interactions for fitness-related traits were statistically significant, partly as a result of the low power to detect such differences. Our data therefore do not support the hypothesis that costs of chemical defense are higher under stressful conditions (Bergelson and Purrington 1996). The prediction of increased costs under environmental stress is based on the arguments that a given investment in defense more strongly constrains other fitness-enhancing traits at low than at high resource levels and that stress conditions themselves can induce the production of higher levels of secondary metabolites, further increasing the constraint on other fitness-enhancing traits. However, the precise effects are likely to depend on the type of stress and the type of chemicals involved. As stated earlier, IGs are carbon-based secondary metabolites. According to the carbon-nutrient balance hypothesis (Bryant et al. 1983, 1988; Tuomi et al. 1988), increased production of carbon-based secondary metabolites is expected under low-nutrient conditions. Indeed, we observed induction of IGs under nutrient stress, as previously observed in P. lanceolata (Darrow and Bowers 1999), but only under noncompetitive conditions (Fig. 1A). However, despite the fact that under these noncompetitive conditions the induction of IGs in response to nutrient limitation was on average larger in plants from the high line (10.7 g IGs/g leaf) than in plants from the low line (4.9 g IGs/g leaf), the cost in terms of reduced aboveground dry weight of high-IG plants was on average not larger but actually smaller under nutrient limitation than under high nutrient conditions. This is consistent with the idea that under low nutrient conditions, when growth and reproduction are nutrient rather than carbon limited, the induced IGs can be produced mainly from excess carbon and do not significantly affect growth or reproduction, unless nutrients are required for associated processes such as transport, storage, self-detoxification, or activation of these compounds. Other studies suggest that even for nitrogen-containing secondary metabolites costs are not necessarily increased under nutrient stress. Vrieling and Van Wijk (1994) suggested that energy rather than nutrients are limiting the production of the nitrogen-containing pyrrolizidine alkaloids in Senecio jacobaea and in accordance detected costs under light but not under nutrient limitation. Second, contrary to the assumption that nutrient limitation would impose a stronger constraint on reproduction in plants producing nitrogen-containing chemical defense compounds, Van Dam and Baldwin (2001) observed that costs of induced chemical defense in Nicotiana attenuata were lower under nutrient limitation than under high-nutrient supply due to plastic allocation responses of plants, increasing allocation of nitrogen to seeds rather than defense under nutrient limitation.

There was also no evidence that costs of IG production were higher in the presence of competitors. We hypothesized that the positive among-family correlation between IG level and cotyledon size might contribute to relaxation of higher costs under competition, as plants from the high line could particularly benefit from the initial size advantage incurred by their larger cotyledon length when competitors are present. The hypothesis was only partly supported by the data. Under nutrient-rich conditions, associations between initial plant size and leaf IG concentration appeared to be weak. Under nutrient-poor conditions, two factors contributed to the absence of higher costs under competition. First, the negative direct effect of IGs on final plant size, that was significant under noncompetitive conditions, was only marginally significant under competition. Second, as hypothesized, the positive direct effect of initial size on final size was not significant when plants were grown singly but was significant under competition, and it offset the weak negative direct effect of high IG on final size under competitive conditions. This indicates a positive pleiotropic effect of high IG content (larger initial size), of which the fitness consequences are more pronounced in the presence of competitors and contribute to a reduction rather than to an increase in fitness costs of these secondary plant compounds in a competitive environment. Such positive pleiotropic effects, expressed under competition, can result from a variety of mechanisms. A study by Siemens et al. (2002) has shown that costs of defense in B. rapa are ameliorated when grown in competition with L. perenne, most likely as a result of allelochemical effects of products of the secondary metabolites from B. rapa on these competitors. Interestingly, some IGs are also known to inhibit germination and growth of seeds and seedlings (Bowers 1991), but whether a similar allelopathic mechanism operates in P. lanceolata is unknown. These results indicate that both associations between defense compounds and morphological traits and multiple functions of secondary metabolites can modulate costs of chemical defense under competition.

Using the same selection lines as the current study, considerable benefits of defense have been shown for high-IG lines of P. lanceolata. Leaf consumption by caterpillars of a generalist insect herbivore was reduced by 55% on plants from the high selection line (A. Biere, H. B. Marak, and J. M. M. van Damme, unpubl. data). In addition, Marak (2000) observed a significant 5% reduction in lesion size on plants from the high selection line after infection by a common stalk pathogen. We expect, therefore, that under competitive conditions such as hayfields, high levels of leaf IGs in P. lanceolata may result both from selection by generalist insect herbivores and, indirectly, from selection on a growth form that is characterized by fewer but larger leaves with a high IG concentration, which are not counteracted by an increase in costs of defense under such conditions.

In summary, we conclude that chemical defense by IGs in P. lanceolata incurs a fitness cost that might contribute to the maintenance of genetic variation in the levels of these defense chemicals previously observed in several studies. Contrary to prediction, the magnitude of the fitness cost is not increased under competition. The latter can partly be explained by an association between defense levels and growth; the initial size advantage associated with high IG production partly negates costs under competition. The results therefore suggest that fitness costs of defense may be modulated by environment-specific fitness effects of genetically associated traits.