Growth–survival trade-offs and allometries in rosette-forming perennials

Authors


†Author to whom correspondence should be addressed. E-mail: Metcalf@demogr.mpg.de

Summary

  • 1Demographic rates such as growth and survival may interact directly as a result of allocation constraints, or indirectly through their relationship with structural characteristics.
  • 2We explored the relationship between growth and survival in a range of rosette-forming species across different habitats, and investigated possible mechanistic explanations for the patterns we found.
  • 3Results indicated a negative association between growth and survival in small plants across species in different habitats. There was no relationship for large plants.
  • 4Relative growth rate (RGR) was positively correlated to specific leaf area (SLA), but unrelated to the percentage biomass allocated to roots. This argues against the hypothesized role of allocation to root mass in mediating the growth–survival trade-off.
  • 5The pattern of biomass partitioning was compared with the predictions of Enquist & Niklas (2002a) Global allocation rules for patterns of biomass partitioning in seed plants. Science 295, 1517–1520. In agreement with their predictions, the overall relationship between above- and below-ground biomass was isometric. However, after accounting for species-specific effects it was found that allocation to roots varied widely between species and was size-dependent, suggesting that the conventional statistical analysis (double-log regression) may be insensitive to biologically important sources of variation.

Introduction

Monocarpic species are interesting from an evolutionary perspective because they have a relatively simple life history where reproduction is fatal, yet even closely related species in the same habitat can show enormous variation in flowering size. For example, within six species in a dune system at Leiden (the Netherlands), weight at flowering varied by >380-fold (van der Meijden, Wijn & Verkaar 1988; de Jong, Klinkhamer & de Heiden 2000). A range of models have been used to explain why particular species flower at a certain size (Klinkhamer & de Jong 1983; de Jong et al. 1989; Klinkhamer, de Jong & de Heiden 1996; de Jong et al. 2000; Rees et al. 2000; Rose, Rees & Grubb 2002). However, these analyses considered only single species, so shed little light on what controls variation in flowering size among species. One possible explanation for the range of flowering sizes observed is that demographic rates, such as growth and survival, are linked through a trade-off such that advantages gained through, say, growth are paid for in terms of reduced survival (Metcalf, Rose & Rees 2003). Theoretical work has shown that such trade-offs flatten fitness landscapes of timing of reproduction (Mangel & Stamps 2001; Marks & Lechowicz 2005): in monocarps, different flowering sizes could be associated with different growth–survival combinations and thereby have equal fitness. To understand why similar species flower at such different sizes, we need to understand how growth and survival vary within and between species.

The ways in which growth and survival interact depends on how their underlying physiological and metabolic processes are interrelated. Declining relative growth rate (RGR) with increasing size is a widespread pattern in vascular plants, and early investigations focused on shoot : root allocation (Tilman 1988; Shipley & Peters 1990). Assuming an ideal environment, where no resources are limiting, Tilman argued that RGR would be an increasing function of allocation to leaves (Tilman 1988). However, this is obviously unrealistic, as some level of root allocation is always critical to prevent uptake of soil resources from limiting growth. The optimal growth solution is to allocate biomass towards roots when soil nutrients are limiting, and towards photosynthetic tissues when light is limiting (Iwasa & Roughgarden 1984; McConnaughay & Coleman 1999). Such ‘balanced growth’ has some empirical support (Shipley & Meziane 2002), although investigations of the implications for survival are lacking. Generally, empirical work has focused on seeking morphological and physiological correlates of RGR differences between species, such as specific leaf area (SLA), net assimilation rate, leaf weight fraction, leaf chloroplast density, root weight fraction, seed mass and a variety of others (Poorter & Remkes 1990; Maranon & Grubb 1993; Hunt & Cornelissen 1997; Reich et al. 1998; Shipley & Meziane 2002; McKenna & Shipley 2003; Sack, Grubb & Maranon 2003). Variation in structure of leaf and root appears to influence RGR more strongly than biomass partitioning (Reich et al. 1998), but the pattern is complex, and the critical correlate may depend on resource levels (Meziane & Shipley 1999; Shipley 2000), or on which functional groups are being compared (Maranon & Grubb 1993).

Traits that affect RGR will also affect survivorship, and the relationship will be negative if both functions are competing for a limiting resource. For example, it has been hypothesized that survival will increase with allocation to storage, as this allows the plant to regrow following damage, for example by herbivores (Iwasa & Kubo 1997; van der Meijden, de Boer & van der Veen-van Wijk 2000). However, stored resources cannot be used for resource capture, thereby generating a growth–survival trade-off. Structural changes that affect RGR and survival are also likely to drive them in opposite directions, because natural selection will always favour those that increase both; so that no heritable variation will persist. Traits such as SLA (leaf surface area per unit mass) show high levels of variability, and are generally positively correlated with RGR (Poorter & Remkes 1990). However, thinner leaves may be more vulnerable to desiccation (Reich 2001), or to herbivory because of reduced levels of indigestible fibre or defence chemicals (Cunningham, Summerhayes & Westoby 1999), which could decrease survival. In agreement with this, seedlings from large-seeded species tend to have a lower RGR, which has been attributed to lower SLA values (Maranon & Grubb 1993; Reich et al. 1998), and are also generally better able to survive and establish under a range of conditions (reviewed by Westoby, Leishman & Lord 1997).

This extensive body of literature suggests that plant growth results from a complex integration involving the biomass and structure of plant organs and allocation decisions, which may be partly determined by trade-offs with survival. However, a recently developed framework suggests that, across all plant species, size and metabolism (which is equated to annualized rates of biomass production) are linked by an invariant functional relationship (Enquist et al. 1999; West, Brown & Enquist 1999; Niklas & Enquist 2001). Extension of the theory to plants predicts static scaling relationships between leaf or root biomass and growth (Enquist & Niklas 2002a), so that structural changes are irrelevant. The functional relationship that emerges is a consequence of physical and biological constraints on the distribution of resources, through fractal-like networks, such that the strategy that maximizes metabolic capacity is determined by very general scaling rules (West et al. 1999). Predicted plant morphologies are obtained by additionally assuming that biomass production scales isometrically with leaf mass (Enquist & Niklas 2002a). This relationship is justified by the assumption that leaf mass scales with effective leaf surface area (Enquist & Niklas 2002b; Sack, Maranon & Grubb 2002), but ignores the complex relationship discussed above between biomass, leaf structure and functioning, and RGR. Despite such broad assumptions, the framework is supported by considerable evidence (Niklas & Enquist 2001; Enquist & Niklas 2002a; Niklas & Enquist 2002; Niklas 2003). Scaling relationships between root, leaf and stem biomass appear to be invariant across species (Niklas 2003). It is therefore of interest to know whether biomass allocation patterns in rosette-forming plants conform to this general framework – if so, then differential root/shoot allocation cannot explain any trade-off between growth and survival.

This paper had three objectives. (1) By measuring growth and survival in a range of species in different sites, we aimed to test for a trade-off. Rosette-forming perennials are a good system for this, as they have a simple growth form and many species are monocarpic, so complications related to the costs of reproduction do not arise. (2) By comparing biomass allocation to above- and below-ground components and measures of leaf structure, we aimed to explore the mechanistic basis of the trade-off. (3) By considering the trade-off's mechanistic basis, and its relationship to the morphological assumptions underlying the fractal scaling model (Niklas & Enquist 2001), we aimed to clarify the role played by the latter in this system. We conclude by discussing implications for the evolution of flowering size in monocarpic plants.

Materials and methods

growth and survival

Field data

Eight sites were chosen, encompassing a broad range of habitat types, to maximize the chance of detecting variation in allocation strategies. In total eight species were investigated: Arctium minus Bernh.; Cirsium arvense (L.) Scop.; Cirsium palustre (L.) Scop.; Cirsium vulgare (Savi.) Ten.; Digitalis purpurea L.; Senecio jacobaea L.; Verbascum blattaria L.; Verbascum thapsus L. In May 2003 we marked a broad size-range of individuals from all species present at each site (a total of 1100 individuals). For each individual we measured root crown diameter, length of the longest leaf, and maximum rosette diameter. We also recorded whether plants had flowered or not between May and August 2003. In May 2004 we recorded which individuals had died (excluding those that had flowered), and measured surviving individuals. This is referred to as the between-year data set.

Many of the marked plants flowered in the first year, and relatively few individuals died, making it difficult to estimate species-specific survival schedules. We therefore marked an additional 1300 non-flowering individuals of five species (C. arvense, C. vulgare, D. purpurea, S. jacobaea, V. blattaria) at three sites in June 2004, and re-measured them in late August 2004 to explore growth and survival. We also recorded flowering, and counted the number of flowering heads for ≈30 individuals of each species to explore reproductive allometries. This is referred to as the within-year data set.

The life history of S. jacobaea may range from a strict monocarp to an iteroparous perennial (Gillman & Crawley 1990), and this is also the case for D. purpurea (van Baalen & Prins 1983). All flowering individuals were, however, excluded from the within-year data set, so reproduction did not affect the allocation processes. Cirsium arvense reproduces vegetatively via lateral rhizomatous roots (Kluth, Kruess & Tscharntke 2003), and between-clone allocation may obscure the growth–survival allocation patterns of interest. However, the identity of genetic individuals could not be determined without disturbing the study population, so we assumed this effect was negligible.

Statistical analysis

Demographic functions are typically size-dependent (Metcalf et al. 2003), and in all analyses the size measure used was transformed using natural logarithms. The length of the longest leaf generally accounted for the greatest proportion of the variance/deviance; for brevity, we present results only for the longest leaf length. For the within-year data set, all species were present in all sites, so site, species and size were initially included in models. For the between-year data set, sites were not distinguished. Growth was described using a least-squares linear regression relating size the following year, or size later in the season, to initial size. The probability of survival and flowering were analysed using logistic regression.

Because growth and survival are size-dependent, to allow comparisons between species, each was calculated for small and large plants using the smallest and largest size common to all species. RGR was calculated according to:

RGR = ag + (bg − 1)Ls( eqn 1)

where ag and bg are the intercept and slope of the linear regression relating size at t to size at t + 1, and Ls is size (smallest or largest log longest leaf measure common to all species) on a log scale. Asymptotic size was also calculated from the fitted growth equations, and is given by ag/(1 − bg). The probability of survival was calculated according to:

image( eqn 2)

where m0 and ms are the intercept and slope of the logistic regression of survival on size.

allometric relationships between organs

Field data

In 2003, at least 30 non-flowering plants of all species except V. blattaria were selected across the range of sites from the between-year data set, and were dug up. In the laboratory plants were cut at the base of the rosette, and roots were washed thoroughly. Total leaf area (not including petioles) of each plant was measured using an ADC AM100 area meter. The petiole was considered to begin at the lowest extent of the leaf blade. The roots, dead leaves and rosette were then dried in a drying oven at 60 °C for 72 h, and weighed separately. For C. arvense each rosette was treated as an individual, and the vertical root was cut where it joined a common lateral root, and weighed separately. In many of the species stems are partially photosynthetic, making the distinction between leaf and stem difficult. Stem biomass was not separated from leaves, and the two are presented as ‘shoot biomass’. With the exception of C. arvense, vertical stem growth is absent in these species until plants start to bolt, so above-ground biomass is approximately equal to leaf biomass in pre-reproductive individuals. Approximate specific leaf area (SLA) was calculated as total leaf area/shoot biomass.

Statistical analysis

Rosette-forming plants can be described as roughly disc-shaped. Simple geometric considerations dictate that, if volume is proportional to disc area, so that biomass is proportional to area, measures of diameter such as longest leaf or rosette diameter should scale with the square root of biomass. We fitted regression models to test these and other allometric relationships using both least squares (LS) and standardized major axis (SMA), which is more suitable for testing whether a slope has a particular value (Warton & Weber 2002). The SMA routines and test were performed using (s)matr (version 1: D.S. Falster, D.I. Warton and I.J. Wright, http://www.bio.mq.edu.au/ecology/SMATR).

To make comparisons across species, root/shoot biomass and SLA of small plants were calculated using species-specific parameters for the LS linear regression relating log shoot mass to log root mass, and log SLA to log total mass, for the smallest log shoot or log total mass common to all species (as described above).

Results

models of growth and survival

Growth was described using a linear regression relating size the following year, or size later in the season, to initial size. For the between-year data, size in year t + 1 increased with size in year t (F1,184 = 727, P < 0·0001). The intercept of this relationship varied among species (ancova, F6,178 = 25·9, P < 0·001), but the relationship with size at t did not (F6,172 = 1·54, P > 0·05): a model with a common slope and site-specific intercepts was adequate to describe the data (F6,184 = 1·50, P > 0·05; Fig. 1). For the within-year data set, final size also increased with initial size, but the absolute amount varied across species according to which site they were in, and this relationship also varied across size (the size × site × species interaction was significant, F8,818 = 3·41, P < 0·0007): the relationship between consecutive sizes required separate intercepts and slopes for each species in each site.

Figure 1.

Relationship between size in May 2003 and May 2004 from the between-year data set. Lines indicate fitted models obtained from the full model used for prediction of relative growth rate and asymptotic size, which includes differences in species intercepts but not slopes, r2 = 0·82.

Survival increased with size and the pattern of increase depended on the specific site (the size × site interaction was significant, inline image, P < 0·05). Within different sites, species identity affected survival differently (the site × species interaction was significant, inline image, P < 0·001).

trade-off between growth and survival

In the within-year data set there was a negative relationship between RGR and survival in small plants (F1,13 = 31·0, P < 0·0001), and the effect varied across sites (F2,11 = 15·7, P < 0·0005); Fig. 2. For large plants, survival showed no relationship with RGR (F1,13 = 3·19, P = 0·10); but survival varied across sites (F1,13 = 3·70, P = 0·046).

Figure 2.

Relationship between probability of survival and relative growth rate (RGR) in small and large plants for five species in three different sites, left to right: (1) dry, open habitat; (2) shaded, undisturbed habitat; (3) disturbed, open habitat. The RGR had a significant effect on survival probability in small plants, and each habitat had a significantly different intercept (see text). Survival probability was uniformly high in large plants, and RGR did not have a significant effect, but the effect of habitat was significant (see text).

allometric relationships and site effects

Least-squares allometric slopes conformed to geometrically predicted values where longest leaf was the explanatory variable, or total mass the target variable (Table 1); SMA slopes were generally significantly different; however, overall, slopes were close to predicted values. For scaling between root and shoot, ignoring all species and site effects, the LS allometric slope was 0·99 ± 0·02 ( indicates 95% confidence interval throughout; here n = 213, r2 = 0·81; Fig. 3a) and the SMA allometric slope was 1·10 ± 0·06 (r2 = 0·81). Incorporating differences in slopes significantly improved the LS model (F6,199 = 6·70, P < 0·005), and we detected significant heterogeneity in the SMA slopes among species groups using a permutation test (Manly 1997) via (s)matr, P < 0·001. Two species-specific slopes were significantly <1, indicating that allocation to root decreases with size (Table 2). As 81% of the variation of root mass can be accounted for using just shoot mass, and the estimated slope is close to unity, we might assume that all species have a similar, constant allocation to roots, as predicted by the fractal scaling model. In fact the full model indicates considerable variation in allocation strategies at different sizes in different species (Fig. 3b).

Table 1.  Predicted and observed relationship between different metrics when fitted on a log–log scale such that log(Y) =a + b log (X)
XY
Total biomassLongest leafRoot crown diameterRosette diameter
  1. Values in bold, predicted b obtained by assuming plants are disc-shaped. Upper value in brackets, observed LS slope ± 95% CI; below, SMA slope ± 95% CI. Total number of observations (n) varies between 200 and 213. Estimates where predicted value is outside the CI: *, P < 0·05; **, P < 0·01; ***, P < 0·001.

Total biomass0·50·50·5
(0·32 ± 0·02)***(0·39 ± 0·03)***(0·34 ± 0·02)***
(0·45 ± 0·03)***(0·45 ± 0·03)**(0·39 ± 0·02)***
Longest leaf211
(2·17 ± 0·19)(0·91 ± 0·10)(0·96 ± 0·04)
(2·60 ± 0·20)***(1·18 ± 0·11)***(1·03 ± 0·05)
Root crown diameter211
(1·88 ± 0·15)(0·66 ± 0·07)***(0·70 ± 0·06)***
(2·20 ± 0·16)**(0·84 ± 0·07)***(0·85 ± 0·07)***
Rosette diameter211
(2·16 ± 0·17)(0·90 ± 0·04)***(0·96 ± 0·09)
(2·52 ± 0·18)***(0·96 ± 0·04)(1·17 ± 0·10)***
Figure 3.

(a) Allometric relationship between shoot and root over all species and sites. Equation of the standardized major axis (SMA) fitted line: log(root biomass) =−0·53 + 1·10 log(shoot biomass). (b) Predicted percentage of mass comprised of root from the SMA model, parameters as in Table 2 (○, A. minus; ▿, S. jacobaea; +, C. palustre; ×, V. thapsus; •, C. vulgare; ▵, C. arvense; ◊, D. purpurea). Similar patterns are obtained with the LS model parameters. (c) Relationship between SLA and total mass over all species and sites. The equation of the SMA fitted line is log(SLA) = 9·36 − 0·32 log(total biomass).

Table 2.  Standardized major axis (SMA) allometric relationship between shoot and root mass in seven rosette-forming species
SpeciesSlope ± CIIntercept ± CI
  1. Overall slope = 1·10 ± 0·06, n = 213, but there was significant heterogeneity [P < 0·001 using (s)matr]. Species-specific SMA slopes and intercepts ± 95% CI.

Arctium minus0·78 ± 0·10 0·13 ± 0·03
Cirsium arvense0·89 ± 0·09−0·80 ± 0·01
Cirsium palustre1·04 ± 0·08−0·55 ± 0·01
Cirsium vulgare1·12 ± 0·12−0·67 ± 0·02
Digitalis purpurea0·94 ± 0·08−0·73 ± 0·01
Senecio jacobaea0·88 ± 0·12−0·22 ± 0·02
Verbascum thapsus1·05 ± 0·09−0·69 ± 0·01

The overall LS relationship between log SLA and total mass is negative (r2 = 0·29); likewise the SMA relationship (r2 = 0·29; Fig. 3c). Least-squares regression indicates that differences between slopes were significant (F6,187 = 3·43, P < 0·005, r2 = 0·65), and there was significant heterogeneity in the SMA slopes (P < 0·001 from (s)matr).

Least-squares site comparisons are shown in Table 3. For root/shoot allocation models, a maximum of 8% (for A. minus) of the residual variation in LS regression was explained by including site as a factor. In contrast, inclusion of site improved the LS SLA model by up to 44% (for A. minus and S. jacobaea). The direction of site effects was in line with expectations based on site characteristics, for example the predicted SLA for a C. palustre individual with a total mass of 2 g was 175 cm2 g−1 under closed canopy vs 99 cm2 g−1 in open habitat.

Table 3.  Site effects on fitted functions at species level obtained using least-squares regression and logistic regressions
SpeciesSitesGrowthFlowerSitesRoot–shootSLA–total
  • *

    Not enough plants survived to allow a growth function to be fitted, or not enough plants were found flowering to fit a flowering function.

  • Top line, r2 or percentage of deviance explained (D) of the model with no site effects; lower line, whether site effects significantly improved the model and the new r2 or D. Outcomes are (a) no effect of site (0); (b) significant differences in intercepts only (1); (c) significant differences in slopes and intercepts (2). Sites: first the number of sites available for demographic functions (growth and flowering), then for the remaining functions.

Arctium minus2r2 = 0·57D = 0·153r2 = 0·85r2 = 0·14
 00 2, r2 = 0·932, r2 = 0·58
Digitalis purpurea1r2 = 0·61D = 0·172r2 = 0·93r2 = 0·61
 nana 01, r2 = 0·66
Cirsium arvense6r2 = 0·47D = 0·024r2 = 0·92r2 = 0·75
 1, r2 = 0·772, D = 0·32 1, r2 = 0·952, r2 = 0·95
Cirsium palustre2r2 = 0·71D = 0·053r2 = 0·95r2 = 0·42
 1, r2 = 0·781, D = 0·12 01, r2 = 0·58
Cirsium vulgare4r2 = 0·37D = 0·273r2 = 0·91r2 = 0·67
 01, D = 0·37 1, r2 = 0·972, r2 = 0·87
Senecio jacobaea6r2 = 0·49D = 0·266r2 = 0·83r2 = 0·20
 1, r2 = 0·922, D = 0·34 1, r2 = 0·891, r2 = 0·64
Verbascum blattaria1r2 = 0·49D = 0·07Not takenNot taken
 nana   
Verbascum thapsus3na*na*3r2 = 0·94r2 = 0·44
    1, r2 = 0·972, r2 = 0·69

predictors of rgr

In the between-year data set, for small plants RGR was positively related to SLA (ρs = 0·94, P < 0·05; Fig. 4a), but unrelated to root/shoot allocation (ρs = 0·25, P > 0·1; Fig. 4b).

Figure 4.

Relationship between RGR of small plants and (a) predicted SLA; (b) percentage root mass. (c) Threshold flowering size and predicted asymptotic size (dashed line, least-squares linear regression, y = 0·59 + 0·87x, r2 = 0·86, P < 0·003).

timing of flowering and reproductive allometries

In the between-year data set the probability of flowering depended on size, and the relationship with size differed across species (inline image, P < 0·01). In the within-year data set the probability of flowering depended on the particular site × species combinations, and this relationship varied across size (inline image, P < 0·003). Flowering probability always increased with size. We summarized this using the threshold flowering size, given by µ = –β0s where β0 and βs are the intercept and slope of the logistic regression describing how the probability of flowering varies with size. This can be interpreted as the mean threshold flowering size (Metcalf et al. 2003) or the size at which the probability of flowering is 0·5. The slope of the LS linear regression relating asymptotic size to threshold flowering size was not significantly different from 1 (0·88 ± 0·38, r2 = 0·86, P < 0·003; Fig. 4c).

Analysis of covariance was used to analyse the relationship between seed-head production and plant size, both log-transformed, using LS due to the number of covariates. Seed-head production increased with size, and the pattern of increase depended on the site (the size × site interaction was significant: F7,125 = 4·52, P < 0·001). Within different sites, species affected seed production in different ways (the site × species interaction was significant: F2,125 = 4·98, P < 0·01). The fitted slopes and 95% CI were 2·83 ± 0·92 for the first site; 1·35 ± 0·60 for the second; 1·95 ± 0·72 for the third; the overall SMA slope was 2·13 ± 0·37. In no case were the slopes significantly different from 2.

Discussion

A trade-off between growth and survival was found across species within sites for a subset of smaller plants within each species. The SLA declined with size in all species, reflecting decreasing RGR with size, and also mirrored changes in RGR of small plants across species (Shipley & Almeida-Cortez 2003). By contrast, percentage root mass did not increase with size consistently within species, so declining RGR could not be attributed to this (Tilman 1988), and there was no simple relationship between root allocation and RGR across species. However, separation of biomass into above- and below-ground compartments may fail to capture underlying complexities (Tilman 1991). For example, Canham et al. (1999) indicate that survival of tree seedlings is linked to relative amounts of carbohydrate reserve rather than to root mass per se.

Our simplified biomass allocation measure must, nevertheless, capture some dimension of plant strategy, as biomass allocation did vary significantly across sites within species. However, changes in leaf structure also occurred with species across sites, and the percentage of variance explained by site differences was much higher for this, suggesting that species are more likely to adjust to habitat differences by changing the structure of their leaves rather than biomass in above- and below-ground compartments. This is consistent with work on changes in biomass allocation at different nutrient levels (Gedroc, McConnaughay & Coleman 1996; Muller, Schmid & Weiner 2000), and the observed plasticity of leaf area (Meziane & Shipley 1999). However, a recent study also points to considerable flexibility in biomass allocation between different nutrient conditions (Moriuchi & Winn 2005). Variability detected in SLA or biomass allocation may therefore depend on the range of environments considered, and the greater flexibility of SLA recorded in our study may simply reflect the range of habitats included.

For a growth–survival trade-off to operate via changes in leaf structure, as suggested above, the fractal model of scaling's assumption that there is no effect of leaf structure on growth rate must be violated (Enquist & Niklas 2002b; Sack, Maranon & Grubb 2002). What do the data suggest? In support of the framework, biomass scaling relationships are relatively invariant across habitats, and the overall slope of the relationship between root and shoot is remarkably close to 1 (West et al. 1999; Enquist & Niklas 2002b; Niklas 2003). Recently, however, it has been shown that an allometric slope of 1 will emerge if the regressed variable is a subset of the x variable (Nee et al. 2005) and sample sizes are large enough. A similar effect might be operating here, as root and shoot are proportions of total biomass. Closer examination, however, shows that species differences in intercept and slope are significant, and that this leads to major differences in the fraction of biomass stored in roots at any given size, indicating highly variable allocation strategies both within and between species. This substantial biological variation is obscured in double-log plots, suggesting that this form of analysis may be extremely insensitive. Additionally, the suggested role of SLA in determining RGR indicates that the assumption of an isometric scaling relationship between leaf mass and metabolic rate may be inappropriate. Plants with the same leaf mass, but differences in leaf structure, have very different RGRs, and our results suggest that RGR may covary negatively with differences in survival, which will have critical implications for life-history evolution (Roff 1992) and, potentially, coexistence (Bonsall & Mangel 2003).

As well as the existence of a growth–survival trade-off that challenges the fractal model of scaling, our results also support a number of straightforward relationships between demographic and physical attributes of rosette-forming plants. The allometric relationships between different measures of plant size largely conform to simple geometric predictions based on assuming that plant shape can be approximated as a disc (Table 1). Also, assuming reproductive effort is proportional to biomass leads to the prediction that reproductive effort is proportional to the square of the length of the longest leaf, which is the case in this study, and is in agreement with the studies reviewed by Metcalf et al. (2003). The species considered generally behave as monocarps, or at least have monocarpic ramets, and simple models of the evolution of flowering size (Rees et al. 2000; Rose et al. 2002; Childs et al. 2003) predict that the threshold size for flowering (above which plants always flower) is given by:

image( eqn 3)

where exp(–d0) is the probability of survival; B the allometric slope of the relationship between seed production and size; and σ2 the variance of the residuals about the fitted growth functions (Fig. 1). This suggests that a linear regression of threshold flowering size on asymptote size should have a slope of 1, which is indeed the case (slope = 0·88 ± 0·32). Assuming large plants have a probability of survival of ≈0.9 (Fig. 2), and substituting in the values of B (2), bg (0·56) and σ2 (0·073), we find the values of the mortality and variance terms are ≈−0·12 and ≈0·17, respectively. This gives a combined value of 0·05. In agreement with this, the intercept of the fitted relationship between threshold flowering size and asymptotic size is not significantly different from 0 (0·6 ± 1).

Why is the observed growth–survival trade-off in small plants important for understanding the evolution of flowering size? At first sight it seems that it should not affect flowering decisions, as only large plants flower and these all have high survival rates (Fig. 2). However, this is not the case: when species have a common growth slope (bg), species that grow quickly when small have large asymptotic sizes, because RGR when small and asymptotic size are both increasing functions of ag. Seed production increases with size, so we expect strong selection for rapid growth of small plants because of the consequent large asymptotic size. However, because of the trade-off, rapid growth when small comes at a cost in terms of survival. Strategies corresponding to smaller flowering sizes might, therefore, be able to compete and coexist with larger flowering sizes, despite the fact they produce fewer seeds, because so few individuals for the larger flowering strategies survive to reproduce. In this way the growth–survival trade-off may underlie the diversity of flowering size across species.

Acknowledgements

C.J.E.M. was the recipient of a NERC studentship.

Ancillary