Does the leaf economic spectrum hold within local species pools across varying environmental conditions?


Correspondence author.


  1. Understanding patterns of trait variation across environmental variability is necessary for development of ecological predictions. The leaf economic spectrum (LES) has demonstrated global trade-offs in leaf traits, but it is unclear whether such patterns are robust in local communities exposed to varying environments.
  2. We conducted separate greenhouse experiments to examine the effects of varying water-table depth and nitrogen availability on leaf-level trait values among a suite of co-occurring wetland species. We then assessed the effects of species-specific trait value responses on relationships predicted by LES and whether species responded similarly to variations in water-table depth and nitrogen availability.
  3. We found that both water-table depth and nitrogen availability had significant species by treatment interactions for specific leaf area, leaf nitrogen and photosynthetic rates, indicating species-specific responses to environmental variability. The responses of individual traits to different treatment levels were relatively consistent across species, but multivariate responses were more variable.
  4. We found that apart from significant relationships between specific leaf area and photosynthetic rate under some treatments, there was little support for the relationships predicted by the LES.
  5. These results suggest that, before trait-based ecology will be able to make progress towards using plant traits to predict responses of communities and ecosystems to changes in environmental drivers, considerable attention needs to be paid to the processes that control intraspecific trait variation.


Global increases in rates of extinctions, invasions and range shifts coupled with changes in the abiotic template that shapes species interactions have challenged the ability of ecologists to predict changes in the composition and function of ecological communities. Trait-based ecology has been proposed as a framework for using basic information about the morphological, physiological and behavioural traits of species to predict changes in community composition in response to global change, as well as how these changes in composition will feedback to ecosystem function (Lavorel & Garnier 2002; Eviner & Chapin 2003; McGill et al. 2006; Westoby & Wright 2006). While this approach has considerable appeal, this appeal is predicated on a number of simplifying assumptions. Here, we examine one of the central assumptions, namely that species can be adequately characterized by a single value for their traits (Albert et al. 2010a).

One of the most striking examples of the use of traits to explore fundamental ecological relationships is the description of what has been termed the ‘leaf economic spectrum (LES)’. Recent studies have demonstrated strong global correlations between leaf-level traits related to photosynthesis (Amass), nitrogen content (Leaf N) and specific leaf area (SLA) (Wright et al. 2004; Reich et al. 2006). These correlations suggest a constrained set of ‘economic’ strategies for plants to ‘choose’ from when it comes to producing leaves, with plants producing leaves that fall along a spectrum from a quick return on dry mass and nutrient investments (i.e. high SLA, high leaf N and high Amass with low life spans) to a slow potential rate of return (i.e. low SLA, low leaf N and low photosynthetic rates with a long leaf life span) (Reich et al. 2003; Wright et al. 2005a; Shipley et al. 2006). While patterns of trait correlations across species are strikingly robust, they are typically based on a single estimate per species for each trait, potentially ignoring critical intraspecific trait variation (Albert et al. 2011; Bolnick et al. 2011).

Analyses of the traits underlying the LES show that each trait can be affected by environmental conditions such as temperature and precipitation (Ordoñez et al. 2009; Messier, McGill & Lechowicz 2010; Sandel et al. 2010), although the nature of the data sets used in these analyses are insufficient to differentiate between shifts of trait values within species or shifts in the composition of species as one moves across environmental gradients (Wright et al. 2005b). Furthermore, the slope of the relationship between, for example, SLA and Amass has been shown to vary in environments with contrasting water and nutrient availability (Wright, Reich & Westoby 2001), although again, the importance of plasticity of trait values within species relative to species turnover is not clear. To date, there has been relatively little attention to how trait values and relationships between traits within a local species pool might respond under changing environmental conditions (Albert et al. 2010a).

To assess the robustness of the LES within local species pools under different environmental questions, we must examine three null hypotheses about the underpinning assumptions about trait responses. The simplest case is that there are no interactions between species identity and environment conditions in the traits underlying the LES. If all species respond similarly to changes in the environment, the relationships predicted by the LES should hold under all conditions. If we reject this null hypothesis, and there are species by environment interactions, one might hypothesize that there are still significant correlations, across species, in trait values from one environment to another. In other words, values for SLA in different environments might be consistently correlated across all species but the slope of the relationship would differ from unity. Such a pattern would lead to the observed shifts in slopes between, for example, SLA and Amass in different environments; if so, we would see strong relationships but shifts in slopes (Wright, Reich & Westoby 2001). These first two hypotheses deal with the predictability of the responses of individual traits to different environments. However, even if there is little support for them, one might still find that the relationships between traits predicted by LES are unaffected by shifts in individual traits if trait responses are correlated.

For the LES to serve as a useful predictive model for how trait values will shift in response to changes in environmental conditions or resources, we need to address three key questions related to the previous hypotheses. First, to what extent do species change their trait values in response to changes in the environment? Second, if species do shift their trait values, do co-occurring species show shifts of similar magnitude, or do species differ in how responsive their traits are to environmental changes? Finally, if species do differ in their responsiveness, are these differences significant enough to change the correlation between traits that comprise the LES within a local flora exposed to different environmental conditions? Here, we address these questions by examining the effects of changes in water-table depth and nitrogen availability on trait expression within a set of wetland plants.

Materials and methods

Experimental Design

We conducted two independent greenhouse experiments to examine the response of leaf traits to environmental variability. Because we were using species from wetlands, we selected environmental variables that are key determinants of wetland plant composition and ecosystem function: water-table depth and nitrogen availability. In 2005, we manipulated water-table depth, a proxy for flooding stress (WT experiment), and in 2006, we manipulated nitrogen supply (N experiment). For both experiments, seed for all species in the experiment came from a commercial source (Ernst Conservation Seed, Meadville PA, USA), with the exception of Microstegium vimineum seeds, which were locally collected. The WT experiment contained 22 species, and the N experiment contained 23 species, with species selected from a list of species commonly used in wetland restoration in N.C. based on availability and germination success (Tables S1 and 2, Supporting information). Individuals of each species were started from seed and then transplanted to individual pots (6·4 cm diam; 36 cm deep) and allowed to grow for 2 months under conditions designed to replicate local summer temperature (25 °C during day and 18 °C at night) and photoperiod. For the WT experiment, plants were grown in a standard greenhouse potting mix (Metromix, Bellevue, WA, USA). For the N experiment, plants were grown in a 3:1 sand:gravel mix to better control N supply.

For the WT experiment, pots were placed in tubs that kept water at a constant level of 0 cm (High WT), 15 cm (Medium WT) or 30 cm (Low WT) below the surface of the pots. For the N experiment, all pots were grown with a water-table 15 cm below the surface, and pots were randomly assigned to three different N treatments that were applied through a modified Hoagland's solution. The concentrations of N in the solution were 1 (Low N), 5 (Medium N) and 10 (High N) mol-N L−1, with all other nutrients held constant across treatments. In both experiments, five individuals of each species were randomly assigned to each treatment. Pots were watered with 200 mL of Hoagland's three times a week and were watered daily on nontreatment days in both experiments. In both experiments, tubs used to maintain water-table levels were drained and refilled with deionized water weekly to prevent build-up of nutrients.

Trait Measurement

In both experiments, after 2 months of growth, two leaves from each individual were selected to measure leaf traits. All traits were measured on the two youngest fully expanded leaves (Cornelissen et al. 2003). We measured photosynthesis at 1000 μmol s−1 m−2 and at ambient levels of CO2 using a LI-COR-6400 (LI-COR Biosciences, Lincoln, NE, USA). After measuring photosynthesis, we harvested the leaves and measured leaf area using a LI-COR 3100C area metre. Leaves were then dried at 70 °C for at least 48 h and weighed. Specific leaf area (SLA) was calculated as leaf area per mass (cm2 g−1), and we converted areal measurements of photosynthesis provided by the LI-COR (Aarea, μmolCO2 × m−2 s−1) to mass-specific photosynthesis (Amass, nmolCO2 × g−1 s−1) by multiplying by SLA and dividing by 10, so that the values would be reported in similar units to the literature. Values of SLA and Amass from the two leaves from each individual were averaged to obtain a single estimate for each individual. To calculate%N for each individual, we collected sufficient additional leaf material to total approximately 5 g and measured leaf nitrogen content on a FlashEA 1112 Elemental Analyzer (Thermo Scientific, Waltham, MA, USA.).

Data Analysis

To improve normality, values for Amass and leaf N were log-transformed prior to all analyses. To assess main experimental effects on traits, we conducted separate standard two-way anovas for each experiment to examine the effects of species, treatment and their interaction on each trait. To determine whether there was consistency across species in the trait responses to changes in environment, we averaged values across individuals of the same species and treatment and then calculated the Pearson correlation coefficient between species average trait values for different treatments within experiments (e.g. SLA under low N vs. SLA under high N). While these correlations capture the degree of consistency with which all species shift individual components of the LES in response to changes in the environment, we were also interested in the consistency of shifts through multivariate trait space. To assess this, we conducted a PCA ordination on Z-transformed values of the species means under each treatment of the three traits. We used a Euclidean distance measure to accommodate both positive and negative values that result from the Z-transformation. This ordination reduced the variables to two dimensions. All statistical analyses described previously were conducted using SYSTAT 11 (Spss 2004).

To examine whether the relationship between components of the LES varied across experimental treatments, we used standardized major axis (SMA) slope-fitting techniques. SMA is considered a more appropriate method for analysing differences in slopes and intercepts than traditional approaches such as ANCOVA when both X and Y variables may have variation associated with them due to both measurement error and species sampling (Warton et al. 2006). SMA slope analyses were used to quantify pairwise relationships of SLA-%N and Amass-%N across treatments within each of the experiments. Because SLA was used to calculate Amass, we quantified the relationship between SLA and Aarea. The DOS–based SMATR package ( used for SMA analyses allows testing for both homogeneity of slopes (i.e. whether the fundamental relationship between traits was constant across treatments) via a permutation test and difference in SMA elevation (intercept) via the SMA analogue of standard ancova (Warton et al. 2006). We only tested for differences in intercept when slopes did not differ significantly.


Main effects of experimental treatments on traits

Varying water-table depth resulted in significant differences in leaf N (Fig. 1b; P < 0·001) and SLA (Fig. 1c; P = 0·015) but had no significant effects on Amass (Fig. 1a; P = 0·81). For all three traits, there were significant differences across species (Amass P < 0·001; leaf N P < 0·001; SLA P < 0·001) and highly significant species by water-table interactions (Amass P < 0·001; leaf N P < 0·026; SLA P = 0·001)

Figure 1.

Responses of leaf economic spectrum traits to experimental variability in depth to water-table (a–c) and nitrogen fertilization level (D-F) (Means ± 1 SE).

Varying nitrogen fertilization levels led to significant differences in all three traits (Fig. 1d–f; Amass P < 0·001; leaf N P < 0·001; SLA P = 0·017). In addition to significant nitrogen effects, there were significant species effects (Amass P < 0·001; leaf N P < 0·001; SLA P < 0·001) and species-by-nitrogen interactions (Amass P < 0·001; leaf N P < 0·001; SLA P = 0·004).

Consistency of species in trait response to changes in environment

Despite significant species by treatment interactions, there were significant correlations between mean values of species traits across levels of water-table depth and nitrogen additions for all traits, although the correlation between Amass at medium and high nitrogen levels was only marginally significant (P = 0·056) (Table 1).

Table 1. Pearson correlation coefficients between species means of traits across different experimental treatments
 Water-table experiment
Low WTMed WTLow WTMed WTLow WTMed WT
Med WT0·741*** 0·713*** 0·783*** 
High WT0·654**0·741***0·565**0·781***0·597**0·795***
 Nitrogen availability experiment
Low NMed NLow NMed NLow NMed N
  1. *P < 0·01; **P < 0·05; ***P < 0·001.

Med N0·424** 0·835*** 0·781*** 
High N0·714***0·404*0·690***0·613**0·803***0·902***

The results from the ordination of species showed that there is some degree of variability between species in how they move through multivariate trait space in response to changes in the environment. In the water-table experiment, moving from the low water-table to the intermediate water-table caused species to significantly shift their trait values in a negative direction along Axis 1 on average (Mean ± 95% CI: 0·38 ± 0·22), consistent with a decrease in% N (Table 2), and no significant shift along Axis 2 (−0·28 ± 0·29). While most species exhibited shifts similar to the mean, some species showed strong shifts in the opposite direction (Fig. 2a). Moving from the intermediate water-table to the high water-table showed a broad range of trait value shifts, resulting in no significant shifts in the mean along either axis on average (Axis 1: −0·05 ± 0·18; Axis 2: −0·15 ± 0·32). Moving from low water-table to high water-table caused a similar pattern as from low to intermediate, with significant shifts towards lower values along both Axis 1 (−0·43 ± 0·31) and Axis 2 (−0·43 ± 0·35) on average, and less variability in the direction of the shifts across species.

Table 2. Correlation coefficients between traits and PCA axes. For both experiments, a 2-dimensional ordination was optimal resulting in a cumulative R2 = 0·98 between the ordination distances and distances in original 3-dimensional space (with 98·9% orthogonality between the axes) for the water-table experiment and a cumulative R2 = 0·965 (with 97·2% orthogonality) for the nitrogen experiment
 Water-table experimentNitrogen experiment
Axis 1Axis 2Axis 1Axis 2
Figure 2.

Shifts of species in trait space in response to changes in water-table (a–c) and nitrogen level (d–f). Shifts for each species are illustrated with grey zero-centred vectors linking the species position in trait space under low levels of the experimental variable to medium levels (a & d); medium levels to high levels (b & e) and low levels to high levels (c & f). The means of all vectors are shown by the black vector.

In the nitrogen experiment, moving from the low nitrogen levels to the intermediate nitrogen levels caused species to shift their trait values in a negative direction along Axis 1 and a positive direction along Axis 2 on average, although neither of these changes were significant (Axis 1: −0·09 ± 0·16; Axis 2: 0·20 ± 0·21), consistent with an increase in% N (Table 2). There appear to be two clusters of species, one of which shows strong shifts in the direction of the mean shift, and the other of which shows small shifts in the opposite direction (Fig. 2d). Shifting from medium to high levels caused a much more consistent and significant trait response across species with decreases along both axes (Axis 1: −0·21 ± 0·18; Axis 2: −0·97 ± 0·34), consistent with increases in all three LES traits. Shifting from low to high showed the same pattern as from medium to high, again with decreases along both axes (Axis 1: −0·30 ± 0·24; Axis 2: −0·77 ± 0·28).

Consistency of Leaf Economic Spectrum Relationships

We examined three bivariate relationships that should be significant according to the LES: Nmass vs. Amass, SLA vs. Aarea and SLA vs. Nmass. In the water-table experiment, SMA analysis indicates that the slopes of all three of the aspects of the LES were consistent across treatments, but not significant (Table 3; Fig. 3). However, for the relationships between Nmass vs. Amass and SLA vs. Nmass, there was a significant shift along the slope of the relationship between the different groups, reflecting the significant treatment effects on mean trait values (Table 4). For Nmass vs. Amass, the high water-table depth was significantly different from the medium water-table depth treatment and the low water-table depth (P = 0·001 and P < 0·001, respectively).

Figure 3.

Variation in the leaf economic spectrum across experimental variability in water-table depth (a–c) and nitrogen fertilization level (d–f). In all panels, low levels of the resource are shown by solid circles with the solid light line; medium levels by grey circles and the dark grey long-dash line and high levels by the open circles and the black, short-dash line.

Table 3. Results from standardized major axis analysis for correlations between traits predicted by LES. First, different treatments within experiments were tested for significant differences in slope. In cases where there was no significant difference in slope between treatments, we tested for shifts in elevation (i.e. intercept terms) or shifts along a common slope. Numbers in bold represent significant differences at P < 0.05
Common slopeShift in elevationShift along slopeCommon slopeShift in elevationShift along slope
Nmass vs. Amass0·0750·055 0·003 0·004 N.A.N.A.
SLA vs. Aarea0·2460·2290·920 0·001 N.A.N.A.
SLA vs. Nmass0·8860·082 <0·001 0·134 <0·001 <0·001
Table 4. Slopes and intercepts for the relationships between traits predicted by LES for individual treatments within experiments. Numbers in bold represent significant differences at P < 0.05
SlopeIntercept P R 2 SlopeIntercept P R 2
Nmass vs. Amass
Low1·7631·5260·1780·0381·8532·063 0·005 0·075
Medium1·9551·6460·9870·0002·2721·907 0·022 0·059
SLA vs. Aarea
Medium0·0020·0860·3920·011−0·0031·664 0·002 0·099
High0·0020·053 0·027 0·077−0·0021·491 0·027 0·060
SLA vs. Nmass

In the nitrogen fertilization experiment, only the relationship between SLA and Nmass showed a consistent slope across treatments, and this slope was not significantly different from 0 (Table 3). There were significant shifts both in elevation and along the slope between groups (Table 4), with, in both cases, the high N treatment different from the medium and low N treatments (elevation: P < 0·001, P < 0·001; shift along slope: P < 0·001 and P < 0·001 respectively). For the relationships between Nmass vs. Amass and SLA vs. Aarea, treatments differed significantly in their slope. For Nmass vs. Amass, the relationship was always weak (explaining between 4 and 8% of the variance) and was only significant in the low and medium N treatments. The relationship between SLA and Aarea was weak and negative in the high and medium N treatments and not significant in the low N treatments.


These experiments were designed to determine the extent to which traits that comprise the LES vary in response to changes in water-table depth and nitrogen availability within a community of wetland plants, and whether species-level differences in trait plasticity would affect our ability to observe the ‘universal’ correlations expected by the LES. In general, we found that trait values varied strongly in response to variation in the environment and that species varied in their trait plasticity. This means that one of the assumptions of trait-based ecology, and of LES, may not be valid as one trait value does not necessarily adequately characterize a species across changing environmental conditions. The response of species to environmental variation for individual traits was generally predictable, although species' multivariate responses were more varied. In addition, the magnitude with which individual species' traits responded to changes in one environmental factor did not predict that species' response to a different driver. Most strikingly, although we observed some of the expected LES relationships under some experimental treatments, these relationships were not universal.

With the exception of water-table not affecting photosynthetic rates, all traits were affected by the experimental manipulations of environmental conditions, so, answering our first question, trait values are not consistent across environmental variation. These results are consistent with a number of studies that have shown that plant traits can vary as a function of both water availability (Wright, Reich & Westoby 2001; Wright et al. 2005b; Sandel et al. 2010) and nutrient availability (Xia & Wan 2008; Ordoñez et al. 2009). While the slight decrease in SLA with increasing water availability is counter the typically observed relationship of SLA increasing with increasing rainfall (Reich et al. 2003), it should be noted that to date most comparisons have focused on much broader gradients of water availability. Our manipulation of water-table depth was unlikely to have created any drought stress, particularly, as all individuals were watered regularly. Instead, this manipulation created a gradient in flooding stress, an equally important environmental factor in wetland habitats. Furthermore, significant interactions between species identity and treatment indicate that species differ in their response to environmental conditions, a result that reinforces other findings that intraspecific variability in traits can be large and species specific (Albert et al. 2010b).

However, despite these significant interactions, responses amongst individual traits were generally in the same direction, just of different magnitudes. Thus, correlations between individual traits across different treatments were generally high. This suggests that differences in species-level responses to changes in environmental conditions may not be significant enough to alter the relative positions of species along trait axes. In other words, species with relatively low SLA under high nitrogen still had relatively low SLA under low nitrogen. Furthermore, in the water-table experiment, trait values in the high and low treatments showed consistently lower correlations than between more similar treatments (high vs. medium or medium vs. low). This suggests that small changes in water-table depth elicit small changes in the trait, but larger changes in the environment produce larger and less predictable responses in the trait. However, this pattern was not present in the N experiment. Thus, to answer our second question, while species differ in how responsive their traits are to environmental changes, it may be feasible to reliably predict trait values for a broad community across environmental gradients based on a limited sampling of the community, particularly across relatively narrow gradients. While predicting traits across broad environmental gradients may not be advisable, these broad environmental changes might be more likely to induce changes in species composition than trait shifts within species.

While the consistency in responses across species was notable when looking at the response of individual traits, the movement of species in multivariate trait space was considerably less consistent. Even in cases where there were strong general trends in the movement of the community in trait space (e.g. from medium to high and from low to high nitrogen), there were examples of species that showed conspicuously different trait shifts than the average species. Thus, species' overall trait responses to shifts in environmental conditions appear to be idiosyncratic. This finding was surprising given that the LES predicts coordinated shifts amongst traits due to the fundamental biological trade-offs that underpin the LES (Reich et al. 2003; Shipley et al. 2006). At least, within this local community, there appears to be significant latitude in the ability of different species to alter different aspects of their leaf economy in uncoordinated fashions. It seems unlikely that this lack of trait coordination would be unique to this set of species, but further studies are necessary to determine the generality of this pattern.

At best, there was weak support for the predictions of the LES in these two experiments. There were significant relationships between photosynthesis (on a per-area basis) only under the medium and high N treatments; these relationships were weak and, contrary to the predictions of LES, negative rather than positive. The relationship between per cent N and photosynthesis was only significant at low and intermediate levels of N in the N experiment, and even in those treatments the relationship was weak, suggesting that in general, plants in these experiments increased leaf nitrogen content without a concomitant increase in photosynthetic rates. This response is consistent with luxury consumption of N under high N availability, either for future use or for allocation to nonphotosynthetic uses such as defensive compounds (Chapin 1980). Luxury consumption of N has been observed in a number of ecosystems (Padgett & Allen 1999; Lawrence 2001; Van Wijk et al. 2003) and represents a decoupling between leaf nutrient content and function that might explain a lack of support for this element of the LES. Under no treatments did we observe a significant relationship between leaf nitrogen and SLA. Unlike other studies that have shown that the relationships in the LES typically show similar slopes but differ in elevation in sites with varying precipitation or resource availability (Wright, Reich & Westoby 2001; Wright et al. 2005b; Liu et al. 2010), we found either significant slopes and no shifts in elevation or nonsignificant slopes, in which case, significant shifts in elevation simply represent a significant treatment effect of the experimental treatments on traits. This suggests that the answer to our final question is that within a local flora, relationships between traits predicted by the LES do depend on the environment but are often quite weak.

The lack of support for the prediction of the LES within this local community was surprising. Grime (2006) suggests that the environment acts as a strong filter on traits involving resource acquisition such as those associated with LES. As a result, he predicted that within local communities, one should observe trait convergence and thus a limited range of variability in LES traits and trait divergence in traits associated with reproduction. Given these predictions, one might also predict that correlations between LES traits might be weaker in local communities simply due to low variability and narrow ranges in the component traits. However, such a mechanism does not appear to be operating in this case. The traits measured on the species in our experiments cover the global ranges in traits for herbaceous species in a global plant trait database (GLOPNet) (Fig S1, Supporting information). A similar breakdown of LES relationships was observed by comparing leaf nitrogen to leaf life span within individuals of the same species grown across a latitudinal gradient (Pensa et al. 2010). Relationships between SLA and leaf size have been shown to have different patterns within and between habitats (Ackerly & Reich 1999; Fonseca et al. 2000). The hypothesis that coexisting species should exhibit limiting similarity has strong theoretical support (Macarthur & Levins 1967). Recent work has shown that within habitats, coexisting species can show significant trait dispersion (e.g. Cavender-Bares, Kitajima & Bazzaz 2004; Paoli 2006; Jung et al. 2010). It may be that local competition serves to select for species with more divergent multivariate strategies as well, reducing the power of the LES to generate useful predictions about relationship between traits within a local community. These results suggest that, before trait-based ecology will be able to make progress towards using plant traits to predict responses of communities and ecosystems to changes in environmental drivers, considerable attention needs to be paid to the processes that control intraspecific trait variation (Suding et al. 2008; Messier, Mcgill & Lechowicz 2010; Albert et al. 2011; Bolnick et al. 2011).


We wish to thank S. Arora and B. McGill for assistance in the laboratory and with data management. Thanks to I. Wright and P. Reich for access to the GLOPNet database. This work was funded in part by National Science Foundation (NSF) Grant DEB0508763 to A.S-G. and by Duke University. The authors have no conflicts of interest.