Present address and correspondence: Johan Ehrlén, Department of Botany, Stockholm University, S-106 91 Stockholm, Sweden (tel.: +46 816 1202, fax: +46 816 2268, e-mail:firstname.lastname@example.org).
1A trade-off between current and future reproduction, often referred to as the cost of reproduction, is a fundamental assumption in life history theory. In long-lived plants, large absolute differences in size between individuals, storage of resources between reproductive events and organ preformation may make such costs difficult to demonstrate, especially when only natural variation is considered.
2The long-lived legume Lathyrus vernus shows large size differences compared with variation in carbon resource allocation, and is known to store resources in below-ground rhizomes. We therefore followed individual plants over a period of 4 consecutive years. We examined the cost of reproductive investment by comparing the performance of untreated plants that differed in size and herbivore damage. We also compared controls with plants where we experimentally reduced flowering in terms of fitness measured as: survival, growth, flower number, fruit:flower ratio and storage.
3Natural patterns of flowering and fruiting provided no evidence of a negative relationship between current and future reproduction. Individuals that produced fruits did not experience a lower probability of surviving and producing fruits the following season compared with flowering individuals that failed to produce any fruits, even when differences in above-ground size and herbivore damage were taken into account.
4Flower removal in a single season increased the allocation to the rhizome but the size of shoot buds for the next season was not increased. Experimental manipulation of reproductive effort by repeated removal of flowers during 3 consecutive years, however, resulted in a significant increase in vegetative size and the probability of flowering and setting fruit compared with control plants.
5While long-term data on natural variation in fruit production and short-term experimental data provided no evidence of a cost of reproduction, such a cost is still present, although detectable only after repeated flower removal.
One of the fundamental assumptions in life history theory is that reproductive investment carries a cost in terms of future survival and reproduction (Williams 1966; Charnov & Krebs 1974; Bell 1980). This trade-off arises because current reproduction exhausts the nutrients or the energy of an organism, thereby reducing its subsequent ability to invest in reproduction or making it more prone to stress-related sources of mortality. Negative genetic correlations are needed to prove the evolutionary significance of such costs of reproduction (Reznick 1985), but phenotypic correlations and experimental manipulations can provide evidence of the functional relationships between current and future reproduction, and between current reproduction and survival, and indicate which selection pressures act on these trade-offs. Although we know that there must be costs associated with reproduction, they are often difficult to detect at all, or appear very small, in long-lived plants. This is especially so when working with natural variation, and manipulation to induce greater variation in allocation might facilitate their detection.
Although the cost of reproduction may indeed be minimal or even non-existent due to low reproductive effort, compensating increases in photosynthetic activity and a capacity to store resources below-ground (Tuomi et al. 1983; Primack et al. 1994; Thorén et al. 1996), the reason why trade-offs are not detected in data on natural variation may be more fundamental. A negative effect of reproductive effort on future performance may be evident if we compare plants of similar size, but in modular organisms, such as plants, size (in terms of fixed carbon) frequently varies over several orders of magnitude (Harper 1977). Moreover, this variation is much larger than that in resource allocation, which is far less plastic because it represents the ratio between carbon allocated to non-photosynthesizing parts, e.g. reproductive organs (so-called sinks) and photosynthesizing parts such as shoots (so-called sources) (Weaver & Cavers 1980; Samson & Werk 1986). Consequently, there is a positive relation in absolute terms between size and reproductive output. Larger individuals are able to both reproduce more in the current season and grow better in the following season, compared with smaller individuals. As a result, detection of trade-offs in studies of natural variation will be unlikely (van Noordwijk & de Jong 1986) and both negative and positive correlations between current and future reproduction have been reported (e.g. Karlsson et al. 1990; Newell 1991; Autlfinger & Wendel 1997; Geber et al. 1997).
Reproductive effort may be manipulated by flower removal or hand pollination, respectively, reducing or increasing the sink:source ratio. When there is a resource allocation trade-off between current and future reproduction, such manipulations will, all else being equal, result in a compensating response the next year. This is usually expressed as dry mass differences in sources (green shoots) and sinks (reproductive tissue), but becomes more complicated when reproductive structures and storage organs act as competing sinks. This is especially important in seasonal habitats where the amount of stored resources has a direct effect on next year’s regrowth and such a carry-over effect might seriously affect our ability to detect a compensating response following manipulation of reproductive sinks (e.g. Snow & Whigham 1989; Cunningham 1997).
We investigated whether reproductive investment carries costs in terms of reduced future performance in the long-lived legume Lathyrus vernus (L.). Such effects have proved difficult to detect in this species because the relative variation in carbon resource acquisition is largely due to marked size differences in the plants, that are also subject to herbivore damage and are known to store resources in below-ground rhizomes. Previous studies with L. vernus have demonstrated that several types of herbivory have important and long-term effects on various components of plant fitness (Ehrlén 1995a,b, 1997). This study focused on the different issue of how investment in seed production influences future plant performance, and data on herbivory and size variation were used only to facilitate detection of such effects. We followed individual plants over a period of 4 consecutive years and compared controls with plants where we experimentally manipulated flower sinks. Five components of fitness (survival, growth, flower number, fruit:flower ratio and storage) were recorded in order to assess how potential carbon costs were expressed and to ask (i) whether there is a trade-off between current reproduction and future survival and reproduction, (ii) to what extent a trade off is affected or obscured by variation in size, and (iii) whether storage organs buffer these effects.
Materials and methods
Lathyrus vernus (Fabaceae) is a long-lived forest understorey herb distributed from central and northern Europe to east of the Ural mountains. It lacks specialized organs for vegetative spread and individuals are therefore usually well delimited and equivalent to separate genets. One to several erect shoots, 5–40 cm tall, sprout from a subterranean rhizome early in spring. Growth terminates a few weeks after shoot appearance, except for regrowth following severe damage. The purple-red flowers (mean total number per individual ± SD = 12.9 ± 17.8, n = 1527) open c. 2 weeks after shoot emergence. Average fruit set within a plant is usually 10–25% and average seed set within fruits is 35–40%. Fruit set is sometimes limited by pollen availability and floral herbivory (Ehrlén 1992). There is an estimated reproductive threshold of 230 mm3 above-ground volume, below which no plants produce fruits (Ehrlén 1995a). The relatively large seeds (mean ± SD = 12.0 ± 3.5 mg, n = 200) mature after 50–80 days. Seeds are dispersed up to a few metres when the dry pods dehisce explosively. Next season’s shoots are initiated and flower buds differentiate during the summer, and future size and flower number is therefore determined before fruit maturation. Above-ground structures normally dieback in October. The average life-span of L. vernus individuals that reach maturity has been estimated at 44.1 years (J. Ehrlén and K. Lehtilä, unpublished data).
Natural allocation patterns within and among years
The data for this study were collected during 1988–91 in 11 permanent study plots in south-east Sweden, mainly on the Tullgarn peninsula (only 10 plots were followed in 1988, see Ehrlén 1995a for further details of study site). Individuals were tagged with small numbered flags and their locations mapped. All individuals were followed from 1988 (or 1989 for one plot) through 1991. Height and basal stem diameter of each shoot were measured, flower number was recorded in all fertile individuals and the number of mature fruits counted. Monitoring of flowering frequency was continued to 1995 in two of the plots.
The effect of reproduction on the resource budget of a plant is likely to depend on its capacity to acquire resources. We used the sum of the products of height and basal stem cross-sectional area for each shoot to give an estimate of above-ground volume and therefore of the plant’s ability to acquire carbon. This non-destructive estimate of plant size was well correlated with dry mass of above-ground tissues (Ehrlén 1995a). The best-fitting relation between mass and vegetative volume was: log mass (mg) = 0.779 × log volume (mm3) (SE = 0.036, R2 = 0.914, P < 0.001, n = 50, least square regression without constant).
The capacity of L. vernus to acquire carbon resources may be reduced by herbivores due to: (i) damage of emerging shoot buds by molluscs early during the season, (ii) removal of part or all of the above-ground shoot by vertebrate grazers, and (iii) damage to individual leaves by insects or molluscs. Source–sink balance and therefore, potentially, future survival and reproduction will be differently affected with the first two types of herbivory affecting both sources and reproductive sinks whereas the third type influences only source strength. Damage of emerging shoots always occurs before canopy closure, but types 2 and 3 can occur after canopy closure when it may be less costly for the plant (Ehrlén 1995a). The numbers of intact and damaged shoot buds per individual were counted in 1989–91 after shoot elongation had been initiated. During flowering, the proportion of above-ground volume that had been removed by grazers and the proportion of leaf area removed by leaf herbivores was estimated by eye to the nearest 10%.
Experimental sink variation
Allocation patterns were manipulated independently from acquisition patterns by removing flowers from individuals outside the permanent plots in the L. vernus population containing most of the permanent study plots. Fertile individuals were selected in 1992 and randomly assigned to the flower removal treatment (n = 45), from which all flowers were removed in 1992, 1993 and 1994 in early spring (when flower buds were visible but before flowers opened) or control (n = 46). All plants were allowed to set fruit freely in 1995. Size, flower number and fruit number were recorded for each individual each year up until 1995.
An additional 25 plants were selected to investigate the effect of fruit production on allocation to below-ground storage and randomly assigned to treatment or control. Flower buds were removed (discussed earlier) in May 1997 and when plants started to dieback (October), they were excavated and brought to the laboratory. Soil, roots and old shoots were removed and the remaining tissues were separated into above-ground vegetative tissue, preformed shoot buds for the next season and rhizome. Above-ground volume was estimated as before and plant parts were then dried to constant mass at 70 °C and weighed.
In the correlative study, total fruit mass was used to represent the amount of carbon allocated to reproduction (calculated as the product of fruit number and estimated average dry mass of fruits, 173.5 mg, Ehrlén 1993) and therefore excluded allocation to flowers and aborting fruits. Reproduction was controlled by removal of flower buds, and this experiment therefore examined the costs of both fruiting and flowering beyond the bud stage. However, the dry weight of flowers is very small compared with that of fruits (Ehrlén 1991), and fruits usually abort at an early stage of development when only a very small fraction of the total carbon resources needed for maturation have been allocated. We therefore assumed that the costs of reproduction result mainly from production of mature fruits and that the costs associated with flowering and aborted fruits are insignificant, but comparing the performance of those individuals that did not flower with those that flowered but did not produce mature fruits, allowed us to test this assumption.
Although treating each flowering event as an independent observation might be considered pseudoreplication (Hurlbert 1984), using a repeated-measures design introduces a bias because only a fraction of plants flowered in consecutive seasons. We therefore chose to analyse data for each of the three annual transitions (1988–89, 1989–90 and 1990–91) separately.
The effects of plot, size (year t), fruit production (year t), grazing damage (year t), leaf damage (year t) and shoot bud damage (year t + 1) on size, flower number and fruit:flower ratio in year t + 1 were examined by ancova models. The effects of the same factors on flowering state were examined by multiple logistic regression models with plot treated as a dummy variable.
The effects of flowering state (considering only flowering individuals that had not produced mature fruits vs. non-flowering, i.e. excluding fruiting individuals to remove the effects of fruit production) in year t on size in year t + 1 were examined in ancova models including also plot, size (year t), grazing damage (year t), leaf damage (year t) and shoot bud damage (year t + 1) as predictors. The effects of the same factors on probability of flowering were examined by multiple logistic regression models. In these analyses, only individuals larger than the threshold size of reproduction in year t were included.
To analyse effects over more than one season, the effects of initial size, three types of herbivory and fruit production during 3 years on size and probability of flowering in the fourth year were examined by logistic and ordinary multiple regression models in all individuals that flowered in the initial year. Predictors that did not significantly (P > 0.15) increase the fit of the model were eliminated from an initial model in a backward stepwise procedure.
In the flower removal experiment, differences in size and flower number were examined by a repeated-measures analysis of covariance. Measurements of size and flower number in flowering individuals for the same plants in consecutive years were treated as repeated measurements on the same subject. The effect of treatment (between-subject factor), time (within-subject factor) and treatment × time interaction were examined by models also including initial size as a covariate. Differences in flowering state each year were examined by log-linear models. The fruit:flower ratio was compared in only the fourth year when experimental individuals were allowed to set fruit freely.
All proportions were arcsine square root-transformed before statistical analysis. Above-ground volume, flower number and fruit number were log-transformed to achieve approximately normal error distributions. Means presented below are back-transformed values. All analyses were carried out using systat statistical package (systat 1996).
Natural variation in fruit production
Of all reproductive size individuals (n = 2686, pooled across all years), 56.3% flowered. In the two plots that were followed from 1988 to 1995, the proportion of flowering individuals differed among years but no trend was apparent (Fig. 1).
The size of flowering individuals varied across almost two orders of magnitude (mean ± SD = 1360 ± 1231 mm3, range = 230–15 652, n = 1512, pooled across all years) and was positively correlated with size in the next year, although many individuals varied considerably between years (r = 0.585, n = 1193, P < 0.001, transitions 1988–89, 1989–90 and 1990–91 pooled). The size of individual plants was not significantly smaller or larger in the year following flowering (ratio [year t + 1] / [year t]: mean = 0.958, 95% CI = 0.906–1.010, n = 1193).
Fruits were formed by 62.4% of flowering plants (n = 1512), with most producing 1 (36.1%), 2 (21.5%) or 3 (14.5%) fruits and less than 3% producing 10 or more fruits (Fig. 2). The average ratio of total fruit mass to total above-ground mass (fruits plus vegetative tissues) in individuals that produced at least one fruit was 0.309 (SD = 0.138, n = 943).
Individuals that produced fruits were less likely to die during the following year compared with individuals that flowered but did not set fruit (Pearson χ2 = 16.7, d.f. = 1, P < 0.001). This indicates that a high reproductive effort does not directly increase mortality, and suggests that death is preceded by a period of failed fruit production.
When size differences among individuals were ignored, the flowering and fruiting history of individuals was positively correlated with their current reproductive state. The probability of flowering in individuals larger than the threshold size for reproduction in 1991 increased with the number of consecutive years that an individual had already flowered (0 years: 52.4%, 1 years: 67.8%, 2 years: 74.5%; 3 years: 91.3%, χ2 = 65.44, d.f. = 3, P < 0.001). The probability of flowering and producing fruits differed significantly among plots in all years (Table 1a). Individuals that flowered but produced no fruits in one season had the same probability of flowering in the following season as fruiting individuals, but for the 1989–90 transition their probability of producing fruits was lower and there was a significant plot × past fruiting interaction in 1988–99, again in favour of those that had produced fruits in the previous season (Table 1b).
Table 1. Effect of past fruiting, plot and their interaction on: (a) probability of flowering and (b) probability to produce at least one fruit in Lathyrus vernus. Effects were assessed by applying log-linear models to three-way contingency tables. Past fruiting was classified as ‘fruiting’ or ‘non-fruiting’ (flowering individuals that had produced at least one fruit and no fruits, respectively, in the previous season). Relative flowering and fruiting frequencies for each category are pooled for all plots. All effects significant at P < 0.05 are also significant following Bonferroni correction for multiple comparisons, n = 3 tests
(a) Probability of flowering
χ2 = 0.6, d.f. = 1, P = 0.440 (Fruiting: 67.4%, n = 233, Non-fruiting: 57.6%, n = 170)
χ2 = 0.5, d.f. = 1, P = 0.477 (Fruiting: 59.7%, n = 313, Non-fruiting: 46.6%, n = 189)
χ2 = 0.2, d.f. = 1, P = 0.671 (Fruiting: 76.7%, n = 223, Non-fruiting: 70.0%, n = 193)
χ2 = 34.5, d.f. = 9, P < 0.001
χ2 = 63.5, d.f. = 10, P < 0.001
χ2 = 56.7, d.f. = 10, P < 0.001
Past fruiting × plot
χ2 = 6.4, d.f. = 9, P = 0.700
χ2 = 12.4, d.f. = 10, P = 0.258
χ2 = 12.1, d.f. = 10, P = 0.280
(b) Probability of fruiting
χ2 = 1.8, d.f. = 1, P = 0.180 (Fruiting: 48.1%, n = 233, Non-fruiting: 30.0%, n = 170)
χ2 = 9.9, d.f. = 1, P = 0.002 (Fruiting: 36.1%, n = 313, Non-fruiting: 22.2%, n = 189)
χ2 = 0.0, d.f. = 1, P = 0.956 (Fruiting: 30.5%, n = 223, Non-fruiting: 34.7%, n = 193)
χ2 = 70.7, d.f. = 9, P < 0.001
χ2 = 57.7, d.f. = 10, P < 0.001
χ2 = 49.0, d.f. = 10, P < 0.001
Past fruiting × plot
χ2 = 27.3, d.f. = 9, P = 0.001
χ2 = 13.0, d.f. = 10, P = 0.223
χ2 = 1.5, d.f. = 10, P = 0.998
The substantial variation in size of the above-ground parts and the damage to leaves among reproducing plants may mask reproductive costs. However, even after accounting for this variation, the analyses failed to detect consistent effects of fruit production on performance the following year. Size, probability of flowering and flower number were all correlated with size in the previous year. Shoot bud damage resulted in a significantly smaller size and lower probability of flowering in that season. There were no effects of fruit number on size, probability of flowering or fruit set (Table 2), but in 1 of 3 years, there was a negative effect on the number of flowers in the following year. Results, following incorporation of current size as a covariate in models of probability of flowering, flower number and fruit:flower ratio were very similar; fruit production in the previous year had no significant effects.
Table 2. Summary test statistics of the effects of plot, shoot bud damage, grazing, leaf damage, size 1 year previous and fruit production on size, probability of flowering, flower number in flowering individuals and fruit:flower ratio, for three time intervals in Lathyrus vernus. Only individuals that flowered in the previous year were included in the analyses. Presented values are test statistics and probability values for significant predictors. Alpha levels were set to P < 0.017 (0.05/3) after Bonferroni corrections for multiple comparisons. Examined effects that were not significant are denoted with a bar. Effects of grazing and leaf damage were not significant in any case and are not presented in the table. Effects for size, flower number and fruit:flower ratio were assessed by ancova models. The partial correlation coefficient (b) is given for effects of covariates. Effects on flowering state (flowering = 1 or non-flowering = 0) were examined by logistic regression models. In these analyses t-ratios (T) and the odds ratio (O) are given for effects of predictors. The odds ratio of the response is given by Q / 1 – Q where Q is the probability of response and the odds ratio is the multiplicative factor by which the odds change when the independent variable increases by one unit. Plot was treated as a dummy variable in logistic regression models
Probability of flowering
1989 n = 330
1990 n = 403
1991 n = 382
1989 n = 319
1990 n = 404
1991 n = 384
1989 n = 192
1990 n = 217
1991 n = 167
1989 n = 192
1990 n = 217
1991 n = 167
F = 8.89
F = 2.79
F = 2.61
F = 2.39
F = 2.88
F = 2.43
F = 2.74
P < 0.001
P = 0.004
P = 0.007
P = 0.014
P = 0.004
P = 0.012
P = 0.005
F = 45.22
F = 33.25
F = 33.83
T = 3.53
T = 5.35
F = 9.17
F = 24.34
F = 52.16
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
b = 1.136
b = 0.848
b = 0.834
O = 3.56
O = 7.73
b = 0.425
b = 1.045
b = 1.186
Shoot bud damage
F = 164.03
F = 180.80
F = 248.44
T = −7.14
T = −6.39
T = −8.04
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
P < 0.001
b = −2.125
b = −1.880
b = −2.452
O = 0.06
O = 0.16
O = 0.03
F = 6.67
P = 0.011
b = −0.64
The comparison of flowering individuals that did not set seed with non-flowering individuals of similar size did not indicate any cost of flowering. Size did not depend on the flowering state in the previous year (1989: F1,406 = 0.03, P = 0.871; 1990: F1,331 = 2.91, P = 0.089; 1991: F1,394 = 1.62, P = 0.203). Flowering state in the previous year significantly influenced the probability of flowering in 1989 (t = 3.32, P = 0.001) but not in the 2 other years (1990: t = 0.90, P = 0.368; 1991: t = 0.50, P = 0.565). The significant effect was once again in the opposite direction to that expected (i.e. lower in individuals that had not flowered previously: odds ratio = 0.36, 95% CI = 0.20–0.60).
In order to examine the effects of reproduction and herbivory over more than one season, final size and flowering state (i.e. in 1991) were regressed on reproduction and each of the three types of herbivory in the 3 previous years (Tables 3 & 4). Fruit production never had a negative effect on final size and flowering state, and the number of fruits had a significant or marginally positive effect on flowering state in 2 of 3 years. Shoot bud damage had a negative effect the same year whereas leaf damage had a significant negative effect on the probability of flowering only in the second year after damage (Table 4).
Table 3. Generalized linear model of the effects of herbivory and fruit production during 3 years on size the fourth year in Lathyrus vernus. Data represent a cohort of 351 individuals that flowered in 1988. The response variable was size (above-ground volume) in 1991. Predictors that did not significantly (P > 0.15) increase the fit of the model were eliminated from the initial model (which considered plot, initial size (1988), and shoot bud damage, leaf damage, grazing damage and fruit number in 1988, 1989 and 1990) in a backward stepwise procedure. Multiple R2 of the final model = 0.449
Shoot bud dam 1991
Fruit production 1990
Table 4. Logistic regression analysis of the effects of herbivory and fruit production during 3 years on probability of flowering in the fourth year in Lathyrus vernus. Data represent a cohort of 351 individuals that flowered in 1988. The response variable was flowering state (flowering or non-flowering) in 1991. Predictors that did not significantly (P > 0.15) increase the fit of the model were eliminated from the initial model (as in Table 3 and with plot treated as a dummy variable) in a backward stepwise procedure. The odds ratio of the response is given by Q / 1 − Q where Q is the probability of response and the odds ratio is the multiplicative factor by which the odds change when the independent variable increases by one unit
Odds-ratio (95% CI)
n = 315 (107 flowering), G = 151.1, d.f. = 14, P < < 0.001, McFadden’s rho2 = 0.374.
Size in 1988
Leaf damage 1989
Shoot bud damage in 1991
Fruit production in 1989
Fruit production in 1990
Experimental removal of flower buds
Severe grazing (more than 50% of above-ground tissues removed) or shoot bud damage (more than 50% of shoot buds damaged) was rare in the flower removal experiment and did not differ between treatments (grazing; control: 9.8%, n = 122, pooled across three 1 year intervals, removal: 7.4%, n = 121, χ2 = 0.44, d.f. = 1, P = 0.506; shoot bud damage; control: 3.4%, n = 145, removal: 3.1%, n = 129, χ2 = 0.02, d.f. = 1, P = 0.881). Survival was high and only one experimental (0.9%, n = 116) and two control (1.7% year−1, n = 115, pooled across three 1 year intervals) individuals died during the study.
Final size differed significantly between treatment and control (Fig. 3a, Table 5a). Individuals in the flower removal group maintained their size during the study (final size was on average 97.8% of initial size vs. 60.8% in individuals that were allowed to set fruit freely). This difference was primarily because a larger proportion of experimental plants experienced increases of 20% or more: large decreases (60% or more) were present in both groups.
Table 5. Effects of repeated flower removal on size (above-ground volume) and flower number in Lathyrus vernus. Repeated measurements of size (a) and flower number (b) in 1993, 1994 and 1995 were examined by a repeated-measures analysis of variance. The effect of treatment (between-subject factor), time (within-subject factor) and treatment × time interaction were examined by models that included size (a) and flower number (b) in 1992 as covariates. Analysis of flower number was only performed with individuals that flowered all three seasons
Between subject variation
Within subject variation
Time × Treatment
Time × Initial size
(b) Flower number
Between subject variation
Initial flower number
Within subject variation
Time × Treatment
Time × Initial flower number
The proportion of flowering individuals increased in the population from 1992 to 1995 (Fig. 1) and the decrease in flowering frequency (Fig. 3b) in both experimental groups reflects the fact that plants often enter a vegetative phase after flowering. The proportion of individuals that flowered did not differ between groups in the first 2 years, but after 3 years of flower removal a higher proportion of treated individuals flowered compared with controls (Fig. 3b). Examination of flowering state in the fourth year according to both treatment group and flowering history (flowered 1, 2 or 3 of the previous experimental years) revealed that although there were no significant effects of flowering history (G = 0.54, d.f. = 2, P = 0.763) there were effects of treatment (G = 4.60, d.f. = 1, P = 0.032) and of the treatment × flowering history interaction (G = 11.06, d.f. = 2, P = 0.004). Thus, differences for individuals that had flowered for three consecutive seasons – where only 42.9% (n = 14) of controls flowered the fourth season compared with 86.7% (n = 15) of experimentals – accounted for most of the overall effect (Fig. 3b). Flowering state is correlated with size (Table 2) and experimental individuals were larger than controls (Fig. 3a), suggesting that the effect of treatment on flowering state in the fourth year was mediated by differences in size. Such an interpretation is supported by the failure of addition of flower removal treatment as a predictor to a logistic regression model that also included size in the third year to improve significantly the fit of the model (G = 2.3, d.f. = 1, P = 0.136).
Throughout the study there was a marginally significant effect on the number of flowers (Fig. 3c, Table 5b, P = 0.055). The average fruit:flower ratio in the final year was significantly higher in plants that had their flowers removed in the previous seasons (10.5%) than in controls (3.8%) (F1,42 = 4.35, P = 0.043).
Individuals in which flowers were removed allocated significantly more to below-ground storage in rhizomes (Fig. 4), although allocation to shoot buds did not differ. Hence, removal of flowers did not seem to influence the differentiation and early growth of shoots for the next year but increased the storage of resources for subsequent growth.
The results of this study demonstrate that costs of reproduction in perennial plants may be detected only by following extreme manipulations, and suggest that large size differences, storage of resources and organ preformation may mask trade-off patterns in natural populations. In L. vernus, no costs were found when patterns of natural variation in reproduction, size and herbivory were examined and measurable trade-offs were observed only after several years of experimental manipulations.
We first asked whether natural variation in flowering and fruiting provides evidence of a negative relationship between current and future reproduction in L. vernus. No such trade-off was detected from patterns of flowering and fruiting during four consecutive seasons. On the contrary, there was a positive correlation between current reproduction and future performance. Flowering state in one season was positively correlated with flowering state the next season, and individuals that produced fruits had a higher probability of surviving and producing fruits the following season compared with flowering individuals that failed to produce any fruits.
Second, we investigated whether a comparatively large variation in plant vegetative size relative to variation in reproductive effort explains why reproductive trade-offs are not detected in data on natural variation. In L. vernus, the probabilities of surviving, flowering and fruiting are all positively correlated to size (see also Ehrlén 1995a). This study also shows that the size of individuals is positively correlated among years, suggesting that positive correlations between current and future reproduction are mediated by size. Assuming that current shoot size depends on past carbon resource acquisition, the ratio of current to previous size can be used as a size corrected estimate of carbon resource acquisition during the previous seasons. In L. vernus, the variation among individuals in carbon acquisition, measured as the standard deviation in size ratios, was 0.923. In contrast, the estimate of the variation in the proportion of resources allocated to reproduction was small (SD = 0.138). These results are consistent with the view that a negative correlation is not observed because variation in resource acquisition is large compared with the variation in resource allocation. Furthermore, damage by herbivores can make fundamental reproductive trade-offs even more difficult to detect if the same individuals are consistently attacked year after year, or if herbivores show preferences with respect to the flowering state of individuals, and this is the case in L. vernus (Ehrlén 1995a, 1997).
If a negative relationship between current and future reproduction is hidden by size differences among plants or herbivory, it should show up when we adjust for such differences. Evidence of negative effects of current reproductive effort on growth, flowering state, flower number or fruit:flower ratio the following season did not, however, appear when we took size differences into account. Even when the cumulative effects of fruit production during the 3 last years were examined, there was no evidence of a negative relationship and thus there was no evidence of a trade-off at natural levels of variation in fruit production. Flowering individuals that did not set seed did not differ from non-flowering individuals of similar size, showing that this is not primarily due to ignoring the costs associated with flowering.
On the other hand, experimental manipulation of reproductive effort by repeated removal of flowers during three consecutive seasons had a significant effect on plant size. Size decreased in the control group but experimental individuals were able to maintain their average size because they more frequently grew larger during the study. This suggests that repeated prevention of reproduction allows vegetative growth and thus increases probability of subsequent flowering.
In other plant species, experimental manipulations of sink strength through flower removal have more often demonstrated effects on future reproduction (Ågren 1988; Inghe 1989; Syrjänen & Lehtilä 1993; Ågren & Willson 1994; Cipollini & Whigham 1994; Geber et al. 1997) than on growth (Syrjänen & Lehtilä 1993; Jackson & Dewald 1994). In L. vernus, vegetative size and fruit:flower ratio responded to flower removal whereas probability to flower and flower number at a certain size were not influenced by flower removal. There is no evidence of a negative trade-off between current reproduction and future survival, although mortality events are rare in the species that have been studied. However, we would also expect selection to favour the strategy with least negative effects on fitness, and fitness in L. vernus is indeed much more sensitive to changes in survival than to changes in fecundity (Ehrlén 1995b).
Allocation of resources to storage and organ preformation may buffer the effects of a reproductive trade-off. In L. vernus, costs of reproduction were not expressed until after 3 years of manipulations of reproductive effort. This does not necessarily imply that costs are small relative to the total resource budget of the plant, because if organ preformation and storage occur, resource depletion may not be sufficient to affect the size and state of above-ground parts until at least 2 years after the reproductive event. The key role of storage in between-year re-allocation of resources has been demonstrated in several studies (e.g. Snow & Whigham 1989; Cunningham 1997). In L. vernus, shoot buds for the next year usually differentiate in May–June (J. Ehrlén, unpublished data) and altered resource status and reproductive investment later during the season therefore do not affect next years’ shoot bud formation. The fact that fruit filling mainly takes place in June–July suggests that manipulating reproductive effort will affect flowering probability no earlier than the second year after the manipulation. In this study, the size of shoot buds for the next season did not differ among treatment groups suggesting that the early growth of next year’s shoots was not affected by altering fruit sink strengths. However, flower removal did increase allocation to rhizomes that indicates that fruit production may influence the growth of shoots and bud formation in the next season via an increase in storage. The correlative data also lend some support to the notion that altered resource acquisition in one season may not influence plant performance for 2 years. Thus, leaf damage had no effect on the probability of flowering in the following season but there was a significant negative effect 2 years after damage (Table 4). The fact that there was no indication of a negative impact of fruit production on size or probability to flower 2 or 3 years later in the same data set (Tables 3 & 4) suggests a greater effect of reducing the carbon-fixing capacity than reducing the carbon expenditure on reproduction.
Taken together, the results for L. vernus in this study suggest that observational studies and short-term experiments investigating reproduction costs in perennial plants must be interpreted with great care.
This research was partly carried out when J. Ehrlén was in receipt of postdoctoral fellowship from the Swedish Natural Science Foundation (NFR). The study was supported financially by the Swedish World Wide Fund for Nature (to J. E.).
Received 29 February 2000 revision accepted 6 October 2000