Interacting determinants of specific leaf area in 22 herbaceous species: effects of irradiance and nutrient availability


Bill Shipley E-mail:


We measured specific leaf area (SLA) and six of its determinants (the thickness of lamina, mesophyll, epidermis, mid-vein and mid-vein support tissues and leaf water content) in a collection of 22 herbaceous species grown in factorial combinations of high μ1100 (mol m–2 s–1) and low (200) irradiance crossed with high (1 : 1) and low (1 : 6 dilution) concentrations of a modified Hoagland hydroponic solution. SLA increased with both decreasing irradiance and with increasing nutrient availability but there was a strong interaction between the two. Lamina and mesophyll thickness both increased with increasing irradiance and nutrient availability without any interaction. The experimental treatments had complicated effects on mid-vein thickness and its support tissues. Leaf water content (a measure of leaf tissue density) decreased with increasing irradiance levels and with decreasing nutrient supply, but with an interaction between the two treatments. Changes in nutrient supply had no effect on SLA at high irradiance because leaf thickness and leaf tissue density changed in a compensatory way. A path analysis revealed that each of the components affected SLA when the others were statistically controlled but the strengths of the effects of mesophyll thickness, mid-vein thickness and water content differed between treatment groups. The effect of epidermal thickness on SLA was constant across environments and it showed no significant covariation with the other determinants. There was significant covariation between mesophyll thickness, mid-vein thickness and water content and this covariation was constant across the treatment groups.


Specific leaf area (SLA) is a widely used variable in plant ecology because it is easily measured and yet is correlated, often strongly, with many plant attributes of interest to ecologists. These attributes include relative growth rate (e.g. Muller & Garnier 1990; Poorter & Remkes 1990; Garnier 1992), leaf gas exchange (e.g. Field & Mooney 1986, Reich, Walters & Ellsworth 1997), biochemical parameters related to photosynthesis (Niinemets & Tenhunen 1997), leaf longevity (Reich et al. 1997) and leaf palatability (Lucas & Pereira 1990). In short, SLA seems to be a useful summary variable that provides information on many key aspects of a plant's ecology (Westoby 1998). It is therefore important to understand how changes in the external environment translate into changes in SLA through the morphology of the leaf.

In this paper we pose two general questions. First, how do different leaf traits contribute to variation in SLA? Second, how do changing levels of irradiance and nutrient supply – both singly and in interaction – affect SLA through the changes that the environment imposes on these underlying leaf traits? Note that these two questions can be posed both at an intraspecific and at an interspecific level, and the answers at these two levels need not be the same. For this reason, we studied both how changing irradiance (1100 and 200 μmol m–2 s–1 PAR) and nutrient levels (1 : 1 and a 1 : 6 dilution of a hydroponic nutrient solution) change the mean values of SLA and its underlying components, and how differences between species in these underlying components relate to interspecific variation in SLA at constant levels of irradiance and nutrient supply. Both the lowered irradiance level and the 1 : 6 dilution of the nutrient concentration significantly reduced the relative growth rates (RGR) of these species (Meziane 1997). Average RGR in the most productive environment averaged 0·25 d–1 and was reduced to an average of 0·15 d–1 in the least productive environment.

As many authors have pointed out (Witkowski & Lamont 1991; Garnier & Laurent 1994; Garnier et al. 1997), specific leaf area is a function of leaf tissue density (D) and leaf thickness (T). Specifically (Witkowski & Lamont 1991):


Clearly, a leaf can change in thickness in more than one way. Changes in epidermal thickness, the volume of mesophyll cells, the number of layers of mesophyll, the amount of intercellular space in the mesophyll, vein thickness or cell wall thickness can all change leaf thickness. Similarly, tissue density will be affected by each of these same variables as well as by cytoplasmic constituents within the cells or different proportions of cell types (for example, mesophyll versus structural or conducting cells). Changes in any of these leaf attributes could simultaneously change both thickness and tissue density to varying degrees (Garnier & Laurent 1994; van Arendonk & Poorter 1994; Shipley 1995; Garnier et al. 1997). Since these underlying determinants of SLA are themselves interacting, these interactions must be taken into account when studying how changing environments or selective pressures change SLA. The third objective of this study was therefore to separate the overall effects of these determinants on SLA into direct and indirect effects, using path analysis, and to test whether the strength of these direct and indirect effects change with differences in irradiance and nutrient supply.


The experiment consisted of 1320 plants from 22 herbaceous species of Magnoliophyta consisting of 13 Magnoliopsidae (six orders, seven families, 13 genera) and nine Liliopsidae (two orders, three families and nine genera). Of the nine Liliopsidae, seven species were from the Gramineae, one was from the morphologically similar genus Carex, and the other from the Acoraceae. Species names are given in the Appendix. Taxonomy follows Gleason & Cronquist (1991). The experiment involved four factorial combinations of high and low irradiance (1100 and 200 μmol m–2 s–1 PAR) crossed with a 1 : 1 (full-strength) and a 1 : 6 dilution of a modified Hoagland solution. Each species was represented by 15 plants in each factorial combination such that three individuals of each species in each factorial combination were harvested at each of 15, 20, 25, 30 and 35 d after planting. Data presented in this paper come only from 35 day-old plants.

Growth conditions

Plants were grown in a Conviron CMP 3244 growth chamber with a 15 h : 9 h (light : dark) photoperiod, a 25 : 15 °C temperature cycle and a relative humidity above 80%. Germination was timed so that all plants began growing within a one-week period. Since the growth chamber could only hold 165 plants (thus, 11 species), each factorial treatment was split into two sets of 11 species at a time. Plants were randomly positioned within the growth chamber in 15 blocks of 11 species (randomly chosen within each block). The three individuals to be harvested at a given age were randomly chosen from among the 15 blocks.

Each plant was grown individually in a 1·3 dm3 container holding 1 dm3 of fine silica sand. Each container had small drainage holes at the base and at mid-height. Since the surface of the sand was a few centimetres below the top of the container, the leaves of prostrate plants were directed up and away from neighbouring plants. Each plant was approximately 8 cm distant from neighbouring plants.

The individual 1·3 dm3 containers were placed in a larger basin (150 cm × 69 cm × 26 cm) made of inert plastic, which was held 60 cm from the floor of the chamber. Fans below this basin ensured the proper circulation of air and therefore the proper air temperature at plant height. The hydroponic solution was held in an external holding tank and the solution was pumped into the large basin three times per day using a programmable timer. At each flush, the hydroponic solution reached the surface of the sand and saturated it through the drainage holes. After 30 min, a second pump returned the solution to the external holding tank. Because of the drainage holes each 1·3 dm3 container drained to field capacity after each flush. In this way, each plant was supplied with three complete flushes of hydroponic solution daily without the root systems interacting.

Light was supplied by five 400 W high-pressure sodium and five 400 W metal halide lamps, giving a high light (L) treatment of 1100 ± 100 μmol m–2 s–1 PAR at pot height. The low light (l) treatment had a photon flux density of 200 ± 50 μmol m–2 s–1 using neutral shade cloth; the light spectrum was the same in both treatments, as measured by a Kodak Personal spectrometer II. The full strength (1 : 1) nutrient solution (N) consisted of 2 mol m–3 KNO3, 1·5 mol m–3 Ca(NO3)24H2O, 2·01 mol m–3 MgSO47H2O, 1 mol m–3 KH2PO4, 0·5 mol m–3 (NH4)2SO4, 9·07 mmol m–3 MnSO4, 0·73 mmol m–3 ZnSO47H2O, 46·3 mmol m–3 H3BO4, 0·089 mmol m–3 Na2MoO4H2O, 0·38 mmol m–3 CuSO4 and 0·38 mmol m–3 FeSO47H2O with EDTA. The low-nutrient treatment (n) consisted of a 1 : 6 dilution of this solution. These two levels provided 6 mol m–3 nitrogen for the high-nutrient treatment and 1 mol m–3 nitrogen for the low-nutrient treatment. The nitrate concentration was monitored periodically using a nitrate-selective electrode and never deviated appreciably from these levels. The pH was monitored daily and adjusted to pH 5·8. The solution was completely changed every 5 d.

The four treatment combinations will be designated as LN (high irradiance, full strength nutrient solution), Ln (high irradiance, diluted nutrient solution), lN (low irradiance, full strength nutrient solution) and ln (low irradiance, dilute nutrient solution) in all that follows.


All leaf blades (petioles were included with stems as a structural component) of each harvested plant were weighed immediately after harvesting, giving a wet mass (WW). Total leaf area (A) of each plant was measured with an image analyser. Leaves were then dried at 70 °C for at least 36 h and then re-weighed, giving a dry mass (DW). Proportional water content (PWC) was calculated as (WW – DW)/WW and, where appropriate, converted to a percentage. SLA was calculated as A divided by total leaf dry mass of the plant. Note that these measures differ slightly from those of Shipley (1995), who measured these variables on single leaves.

For each of the three randomly chosen 35 day-old individuals per species and treatment combination, two leaves in the middle of each plant were used for anatomical measurements. Four or five thin sections of these leaves were taken by hand using a razor blade midway between the leaf edge and the mid-vein, placed in 6% sodium hypochlorate for 10 min and then rinsed five times in distilled water. These sections were then placed in a colouring solution which consisted of Carmin (which stains cellulose pink) and Fast Green (which stains lignin green). To prepare the colouring solution, 15 g of aluminium sulphate and potassium where dissolved in 1 dm3 of water after which 2 g of Carmin was added and then heated to reduce its volume to 750 mL. After cooling and filtering, 1·5 mL of a 1 dm3 solution of water and 2 g of Fast Green were added. After colouring, the thin sections were mounted on a microscope slide with a Farrat solution (glycerine and Acacia Gum) to prevent desiccation. Immediately following this procedure, these sections were used for the following thickness measures (μm): lamina thickness, thickness of the mesophyll tissues (which was differentiated into palisade + spongy mesophyll only for the dicots), thickness of the upper and lower epidermis (including cuticles) and thickness of the sclerenchyma (monocots) or collenchyma (dicots) tissue surrounding the conducting tissues of the mid-vein. Note that we did not measure fibres surrounding the vascular bundles or elsewhere in the leaf since these cells were not arranged in a way that allowed an objective measure of total thickness of this tissue type.

Statistical analysis

(I) Univariate methods

Whenever normality was required for statistical tests [analysis of variance (ANOVA) and multigroup path analysis] all variables were transformed to their natural logarithms to better approximate the normality of the distributions, except for PWC which was transformed to its logit since it was a proportion, namely ln[PWC/(1 – PWC)].

The hierarchical sampling design generated three nested sources of random variation, namely, between leaves of the same plant (for the morphological attributes), between individuals of the same species (for all variables) and between species (for all variables). In such nested samples, variation at higher levels confounds variation unique to that level with variation generated by lower levels. We estimated the relative importance of these sources of variation using the VARCOMP function of SPLUS (SPLUS 1995) in a nested design using restricted maximum likelihood. These variance components were estimated separately for each treatment combination.

Effects of the treatments on each variable were evaluated using a three-way factorial ANOVA; due to some missing values (see Appendix) we used TYPE III sums of squares but sequential sums of squares always gave the same conclusions.

(II) Multivariate methods

A graphical summary of the patterns of covariation between percentage water content, SLA, and the five thickness measures (support tissues, epidermis, lamina, mesophyll and mid-vein) for each species in each of the four treatments was obtained using principal components analysis based on the correlation matrix of these seven variables. We plotted the scores of each species for each treatment on the first two principal components. We superimposed the correlation (multiplied by two to increase visual interpretation) of each measured variable on each principal component as a vector arrow. For instance, a variable that had a correlation of 0·7 on axis 1 and –0·4 on axis 2 would be represented as an arrow whose tail was at (0,0) and whose head was at (1·4,–0·8). The length of an arrow projected vertically (for axis 1) or horizontally (for axis 2) is therefore proportional to its correlation on that axis. The angle between any two arrows is proportional to the correlation between them. Two arrows that are perfectly superimposed and in the same direction (0°) have a correlation coefficient of 1. Two arrows that are perpendicular (90°) are uncorrelated. Two arrows that are perfectly superimposed but in the opposite direction (180°) have a correlation coefficient of –1. The means of the seven variables in each treatment (standardized to zero mean and unit variance) were projected onto the two principal components analysis axes. This type of plot is called a biplot (Gabriel 1971).

The SLA can be logically decomposed into tissue density and leaf thickness. We measured three components of leaf thickness (mesophyll, protruding mid-vein and epidermal thickness) and an indirect measure of tissue density (PWC). Each of these can potentially affect SLA and the slopes can potentially vary with the experimental treatments. Furthermore, the patterns of covariation between these components, as well as the residual variation of each, can potentially vary with the experimental treatments. To test for equality of slopes, covariances and residual variances across the treatments, we used multigroup path analysis (Bollen 1989; Bentler 1995). This method, although largely unfamiliar to biologists, has been commonly used in many other disciplines for over 20 years and is thoroughly described in Bollen (1989). See Shipley & Meziane (1998) for a biologically oriented introduction. Only the basic method is described here.

Figure 1 shows the basic structural model corresponding to the decomposition of SLA in the four treatment groups. This model assumes that each of proportional water content, mesophyll thickness, protruding mid-vein thickness and epidermal thickness are direct causes of SLA; these are shown by the single-headed arrows and the parameters (αik for coefficient i of group k) are similar to partial regression coefficients. These measure the effect of each component on SLA after statistically holding constant the values of the other components. Since the variables are ln-transformed, they are linearly additive (eqn 1). This follows directly from the definition of SLA as the product of average leaf thickness and leaf tissue density. These four components of SLA are not necessarily independent. To avoid making any further causal assumptions concerning the relationships between these components, we allow them to freely covary; these covariances (σijk for the covariance between variables i and j in group k) are shown by double-headed arrows. Finally, each variable has its associated error variance (ɛik for the residual error of variable i in group k). There are therefore 15 free parameters in this model for each treatment combination giving a potential total of 60 different parameters to estimate.

Figure 1.

. Multigroup path model to be tested. One path diagram which is repeated in each of k = 1,2,3,4 groups, representing the four treatment groups. Parameters: path coefficients (αik for path i in group k), covariances (σijk between variables i and j in group k) and (residual) variances (ɛik for variable i in group k).

The most restrictive hypothesis that can be tested using this model is that every one of the 15 parameters is the same in all four treatment groups (i.e. only the mean values might change). To test this hypothesis, the 15 parameters per group are estimated using maximum likelihood techniques in which the estimates are forced to be the same in the four treatment groups. If this hypothesis is true then the maximum-likelihood χ2 statistic will follow a χ2 distribution with 15 degrees of freedom. A significant result means that the 15 parameters are not equal in the four groups and this restricted hypothesis must be rejected. To determine which parameter varies between groups (if the first hypothesis is rejected) one then re-fits the model allowing only one parameter to be estimated independently in the four treatment groups, all others still being constrained; this produces a model with 18 degrees of freedom. If the independently estimated parameter was actually the same in the four groups, then the decrease in the χ2 statistic relative to the fully constrained model will, itself, be distributed as a χ2 statistic with three degrees of freedom. A significant value means that the parameter was significantly different in at least one group. This is repeated for each parameter. The EQS statistical package (Bentler 1995) was used to carry out this analysis.


Components of variance

Although the treatment levels were fixed in this experiment, the choice of species, the individuals within each species, and the leaves within each individual were randomly chosen. Table 1 lists the percentage of the total variance of each component of SLA that is attributable to each of three hierarchical levels. It is clear that a large majority of the variation arises through species-level differences for all of the measured attributes, although the degree to which these interspecific differences manifest themselves varies between treatments. Since leaf water content was not measured for each leaf, the variation attributable to each individual in Table 1 actually includes both individual-level and leaf-level variation.

Table 1.  . Percentage of the total variance in each leaf attribute attributable to differences between species, differences between individuals of the same species and (where applicable) differences between leaves of the same plant. Plants were grown in hydroponic sand culture at two levels of nutrient concentration (N = 1 : 1, n = 1 : 6 dilution) crossed with two levels of irradiance (L = 1100 μmol m–2 s–1, l = 200 μmol m–2 s–1). Analyses were done separately for each treatment combination. All thickness measures had units of μm. SLA had units of cm2 g–1. Leaf water content was in proportion of wet mass. Analyses based on transformed data for normality Thumbnail image of

Effects of light and nutrient concentration on mean values

The Appendix presents the mean values of each variable for each species in each treatment combination. Figure 2 presents the interaction plots for each measured variable and gives the overall mean values and their standard errors per treatment combination (i.e. the mean of the species means and the standard errors of the species means).

Figure 2.

. Interaction plots (untransformed data) for each measured variable (SLA cm2 g–1,% water content of the leaf and thicknesses (μm) of lamina, mesophyll, epidermis, mid-vein and mid-vein support tissues). Circles represent treatment means and vertical lines are standard errors. Open circles represent values in full-strength hydroponic solutions and black circles represent 1 : 6 dilution treatments. The two irradiance treatments (200 and 1100 μmol m–2 s–1 PAR) are shown on the abscissa.

Specific leaf area

Both irradiance level and nutrient concentration affected the specific leaf area (P < 0·001) and there was also an interaction between the two treatments (P < 0·001). The quantitative values of these effects differed between species (P < 0·001). In the full strength nutrient concentration, decreasing light intensity increased the mean specific leaf area by 123 cm2 g–1. In the dilute nutrient solution a decrease in light intensity actually increased the mean specific leaf area but only by 36 cm2 g–1. Comparing plants at the same high irradiance level, a decrease in nutrient concentration increased specific leaf area by only 8 cm2 g–1 while, in the low-light environment, decreasing the nutrient concentration decreased specific leaf area by 79 cm2 g–1.

Leaf water content

Leaf water content (an indirect measure of tissue density) was affected by both irradiance and nutrient availability (P < 0·001) and these two treatments showed an interaction (P < 0·001). Species-specific differences to these factors also existed (P < 0·001). Water content was highest (and tissue density was lowest) in the lN treatment. Decreasing the irradiance level in the full strength nutrient solution increased leaf water content by 6%. The same decrease in irradiance level in the dilute nutrient solution did not change water content at all. Decreasing nutrient levels in the high irradiance treatment had only a modest effect, decreasing water content by 2%, but the decrease in nutrient levels in the low irradiance treatment decreased water content (therefore increased tissue density) by 8%.

Lamina thickness

Increases in both irradiance and nutrient availability increased the lamina thickness (P<0·0001) but without any detectable interaction between the two factors (P=0·65). There were obvious differences between species (P<0·00001) and interactions between species and each of the experimental treatments. Changes in mesophyll thickness paralleled the results for lamina thickness; this is to be expected since lamina thickness is overwhelmingly due to the mesophyll. Increases in both irradiance and in nutrient levels increased lamina and mesophyll thickness.

Epidermal thickness

Epidermal thickness, namely the sum of the superior (adaxial) and inferior (abaxial) epidermal layers, was affected by both irradiance (P = 0·0001) and nutrient availability (P=0·03) and these two treatments showed an interaction (P<0·0001). Species-specific differences existed for each treatment (P<0·0001). An increase in nutrient supply increased epidermal thickness at low irradiance but had no effect at high irradiance.

Mid-vein thickness

The thickness of the mid-vein was affected by the irradiance level (P<0·001) and nutrient concentration (P<0·001). In the full strength nutrient solution, decreasing irradiance (LN to lN) decreased the average thickness of these tissues by 107 μm but in the dilute nutrient solution, decreasing irradiance (Ln to ln) had no effect. Since these same manipulations did increase both mesophyll and lamina thickness, this means that the ratio of mid-vein to lamina thickness decreased in the high irradiance, low-nutrient conditions. Again, there were obvious differences between species (P<0·00001) and species-specific responses to the treatments, as shown by significant interactions.

Mid-vein collenchyma and sclerenchyma tissues

The thickness of the mid-vein support tissues was affected by both treatment variables and there was also a significant interaction between the two as well as between treatments and species means. The absolute and proportional effects of the treatments on this variable were less dramatic than those associated with lamina and mesophyll thickness. In the full strength nutrient solution, decreasing irradiance (LN to lN) increased the average thickness of these tissues by about 4 μm but in the dilute nutrient solution, decreasing irradiance (Ln to ln) decreased the average thickness by 12 μm. Again, there were obvious differences between species (P < 0·00001) and species-specific responses to the treatments, as shown by a significant two-way interaction. The patterns of change were not the same when comparing treatment effects on the mid-vein thickness and on the thickness of the mid-vein support tissues.

Multivariate patterns

Figure 3 shows the bivariate patterns between selected pairs of variables, transformed to normality, as well as the correlations between each pair for each treatment combination. SLA was positively correlated with leaf water content in all treatments. SLA was negatively correlated with lamina thickness, but this negative correlation was much stronger in the high irradiance treatments (–0·61 and –0·66) than in the low irradiance treatments (–0·23 and –0·47). Leaf water content was positively correlated with lamina thickness, but this positive correlation was much stronger in the low irradiance treatments (0·58 and 0·51) than in the high irradiance treatments (0·02 and 0·20). Mid-vein thickness was always positively correlated with lamina thickness irrespective of treatment.

Figure 3.

. Scatterplot matrix of four leaf attributes (SLA, leaf water content, lamina thickness and mid-vein thickness, all ln-transformed) measured on each of 22 species in each of four different treatment conditions. Open symbols represent high (1100 μmol m–2 s–1 PAR) irradiance levels and filled symbols represent low (200 μmol m–2 s–1 PAR) irradiance levels. Squares represent high hydroponic nutrient levels and circles represent low (1 : 6 dilution) hydroponic nutrient levels. Lines are regressions in high (thick) or low (thin) irradiance treatments. Diagonal panels show the distributions of each variable. Subdiagonal panels list the Pearson correlation coefficients between the two variables in each treatment group. The rightmost column shows the treatment means (larger symbols) and the species means (smaller symbols) in each treatment. In each plot, the variable named in that row gives the ordinate and the variable named in that column gives the abscissa. Thus, the plot in row 1, column 3 shows ln(SLA) on ordinate versus ln(lamina thickness) on the abscissa.

Because it is difficult to visualize from pairwise comparisons how the leaf attributes vary together as a whole, we next conducted a principal component analysis based on the data in the Appendix. Variables were standardized to zero mean and unit variance to remove the gross effects of different units. The biplot of these data is shown in Fig. 4. The first two axes accounted for 64% of the total variance. The loadings of each variable on these first two axes, as well as the correlations between the variables and each axis, are given in Table 2. We projected the standardized means of these variables by treatment (Table 3) onto these axes for visual comparison.

Figure 4.

. Biplot of each species in each treatment combination plotted onto the first two axes of a principal component analysis, based on standardized untransformed variables. Closed circles represent ‘graminoid’ species, namely grasses and Carex crinita. Correlations (multiplied by two for visual ease) between each measured variable and each axis are shown by the arrows.

Table 2.  . Loadings and Pearson correlations between each leaf attribute and the first two axes of a principal component analysis based on standardized variables (i.e. the correlation matrix). Bold type indicates those variables correlating most strongly with the given axis. Thickness measures are in μm, SLA is in cm2 g–1 and water content is in percentage of wet mass. Analyses based on untransformed data Thumbnail image of
Table 3.  . Means (± SD) and ranges of various leaf attributes measured in four combinations of irradiance (L = 1100 and l = 220 μmol m–2 s–1 PAR) combined with dilutions of a Hoagland hydroponic solution (N = 1 : 1 and n = 1 : 6 dilution). Each mean is based on the species means for each combination. Thickness measures are in μm, SLA is in cm2 g–1 and water content is in percentage of wet mass. Data have not been transformed Thumbnail image of

The first axis clearly contrasts differences in SLA with differences in leaf thickness that are largely independent of tissue density (leaf water content and the thickness of the support tissues and epidermis). The SLA was negatively correlated with this axis while each of lamina, mesophyll and mid-vein thickness was strongly positively correlated. The second axis just as clearly contrasts SLA with differences in tissue density that are largely independent of leaf thickness. SLA and water content correlated strongly and negatively with this second axis while the thickness of the mid-vein support tissues and epidermal thickness correlated positively with this second axis. There is also a clear difference between the ‘graminoid’ species (grasses and Carex crinita, shown by solid black circles) and the other species. The graminoid species tend to have thinner but more dense tissues which mostly cancel each other out in relation to SLA. The effects of the experimental treatments on the mean values, as seen in Fig. 4, largely corroborate the trends in Fig. 2. The high irradiance treatments produce leaves that are both thicker and with denser tissues. The higher nutrient treatments tend to produce leaves that are both thicker and with less dense tissues.

Path analysis

Path analysis was used to study the interspecific relationships between the variables, and how these relationships might change as a function of the experimental treatments. The most obvious causal hypothesis is that each of water content, mesophyll thickness, protruding mid-vein thickness and epidermal thickness are independent causes of SLA; that is, each of these determinants cause interspecific SLA but they do not affect each other. Given this hypothesis, the covariances involving all of these four determinants are zero. This multigroup path model, with free parameters allowed to vary between groups and only these covariances constrained to be the zero in the four treatments, was rejected (χ2 = 38·739, 24 d.f., P = 0·03). These path analyses were done on the transformed variables in order to insure multivariate normality, which is an assumption of the method. The determinants of SLA were not independent of one another. We used protruding mid-vein thickness, that is the mid-vein thickness minus lamina thickness, in order to remove the obvious allometric dependence of the first on the second. The rejection of this first model required that we explore the structure of the dependency relations between the determinants.

We began the modelling exercise with a fully saturated path model in which epidermal thickness, the protruding mid-vein thickness, mesophyll thickness and proportional water content (each transformed as described previously to normality) were direct causes of SLA but with each determinant allowed to freely covary with the others. In this way we did not impose any causal structure on the determinants of SLA (see Fig. 1). As is usual in path analysis, each variable was also centred to a zero mean within each treatment group so that the units were logarithmic deviations from the mean of that variable in that treatment group. This highly constrained model was also rejected (χ2 = 64·55, 45 d.f., P = 0·03) indicating that at least one parameter differs between the four treatment groups.

To determine which parameters differ, we then tested a series of 15 models, each nested within the fully constrained model, as described in the Methods section. There was strong evidence that the direct effect of the protruding mid-vein thickness on SLA differed between treatment groups (P = 0·0002) and moderately strong evidence that two other direct effects on SLA also differed (P = 0·01 and 0·05 for the direct effects of mesophyll thickness and water content, respectively). It therefore appears that the experimental treatments change the slopes of the relationships between these three thickness measures and SLA.

Given these results, we fitted a final multigroup model in which all parameters except for the three significantly different path coefficients were constrained to be equal in all groups. The covariances between epidermal thickness and each of the other three determinants were never significantly different from zero (P > 0·3). The covariances between each of the other three determinants were always significantly different from zero (P < 0·05). Once controlled for the other determinants, each determinant significantly affected interspecific variation in SLA. This model provided a very good fit to the data (χ2 = 28·182, 36 d.f., P = 0·82) and always explained a large proportion of the interspecific variation in SLA. Figure 5 shows the result; the non-significant covariances involving epidermal thickness have been removed from this figure for visual clarity.

Figure 5.

. Final multigroup path model (centred variables, transformed to normality) relating leaf water content, mesophyll, protruding mid-vein and epidermal thickness to SLA in each of the four experimental treatments. Arrows represent direct effects and double-headed arrows represent unstructured covariances. Broken arrows indicate coefficients that do not vary significantly between treatment groups, and solid arrows represent coefficients that do vary significantly between treatment groups. The covariances between epidermal thickness and the three other determinants of SLA were not significantly different from zero and have been omitted in this diagram. This model shows no significant lack of fit to the data (χ2 = 28·182, 36 d.f., P = 0·82).

This final path model shows that the importance of leaf water content in explaining interspecific differences in SLA (the path coefficients) increases with decreasing light intensity and with decreasing nutrient supply. In other words, the importance of leaf water content (thus tissue density) increases as the environment becomes more light- and nutrient-limited. The importance of mesophyll thickness in explaining interspecific differences in SLA decreases as light intensity decreases but increases as nutrient supply decreases. In other words, the importance of mesophyll thickness increases in more light-limited environments but decreases in more nutrient-limited environments. The importance of the protruding mid-vein thickness in explaining interspecific differences in SLA decreases as both light intensity and nutrient supply decreases. In other words, the importance of the mid-vein decreases as the environment becomes more light- and nutrient-limited.


Effects of irradiance and nutrient concentration on SLA

It has been known for a very long time that the morphology of sun and shade leaves differ (Hanson 1917; Wylie 1951; Carpenter & Smith 1981; Abrams & Kubiske 1990). Tree leaves developing in shade tend to be thinner with thinner palisade and spongy mesophyll layers. Smith, Bell & Shepherd (1998) showed that the same trends occur in interspecific comparisons using species which are typical of habitats that differ in irradiance levels. On the other hand, much less is known about the effects of differing nutrient availabilities on leaf morphology. Hirose, Freijsen & Lambers (1988) and van Arendonk et al. (1997) found that SLA decreased with decreasing nitrogen supply but van der Werf et al. (1993) did not find this result. Witkowski & Lamont (1991) found that plants of two Australian tree species growing in drier or more nutrient-poor soils had a lower SLA and thicker, denser leaves. The effects of other environmental conditions on the functional morphology of leaves are equally poorly known (Körner et al. 1989).

We found that both irradiance and nutrient supply affected SLA but the strength of the interaction between these two variables means that treating each factor independently is misleading (Fig. 2). Although the well-known effect of decreasing irradiance on increasing SLA (by an average of 123 cm2 g–1) is clearly seen in the full strength nutrient solution (LN to lN), the same change in irradiance in the 1 : 6 diluted solution increased SLA by only a small amount (36 cm2 g–1 on average). The less-well known effect of nutrient availability had even more complicated effects on SLA. A nutrient stress in the high irradiance treatment (LN to Ln) had virtually no effect but the equivalent nutrient stress in the low irradiance treatment (lN to ln) reduced SLA on average by 79 cm2 g–1. Note that the effect of decreased nutrient supply under low-light conditions reduced SLA by more than twice the amount of the effect of decreased irradiance under the low-nutrient conditions. Put another way, whenever SLA had already been reduced by one of the treatments (either by increased irradiance or by decreased nutrient levels), then changing the other treatment had little effect. It is as if these species have a lower limit to the degree of plastic change in SLA and, once reached due to one treatment (high irradiance or low nutrient supply), this lower limit is little affected by an additional treatment.

The functional explanation for this interaction is less clear. The total leaf area of the plant determines the potential amount of photons that the plant can intercept at any instant. However, for this potential to be realized the plant must invest in sufficient photosynthetic machinery per unit leaf area to transform this energy. The amount of photosynthetic machinery per unit leaf area will depend both on the concentration of photosynthetic enzymes and pigments per unit mesophyll biomass and on the thickness of the mesophyll (Niinemets & Tenhunen 1997). Thus, the strong interaction between irradiance and external nutrient supply highlights the potential importance of leaf thickness versus leaf density in determining variation in SLA.

Leaf thickness versus leaf density

SLA can be logically partitioned into the average leaf thickness and the tissue density of the leaf. Garnier & Laurent's (1994) study of 14 grass species showed that leaf water content was positively correlated with the percentage of the leaf cross-sectional area occupied by mesophyll protoplast (i.e. mesophyll cell area minus the cell walls) and negatively correlated with the cross-sectional area occupied by epidermis, vascular tissues or sclerenchyma. The overall correlation between leaf water content and leaf density was –0·77 in their study. On the other hand, leaf water is found overwhelmingly within the cytoplasm and vacuoles and the rest is in the wet cell walls and in the vessel elements (Canny 1995); changes in leaf volume due to changes in intercellular spaces would not be much reflected in changes in water content. Therefore, leaf water content is not strictly a measure of leaf density (i.e. leaf mass per unit of leaf volume) but rather a measure of leaf tissue density (i.e. leaf mass per unit of leaf tissue volume). It is clear from our data that the degree to which different species show a trade-off between leaf thickness and leaf tissue density varies with environmental conditions. In light-limited environments species with thicker leaves tend to have less dense tissues (i.e. a larger water content) but this trade-off essentially disappears when comparing species growing in a high-irradiance environment.

Interspecific trends

The interspecific patterns observed within a given treatment between the variables represent differences between species that are not due to changing irradiance or nutrient levels. Since none of the covariances between the four determinants of SLA differed significantly between treatment groups, this suggests that the slopes of the causal links between these four determinants (however these might be structured) are constant. We say ‘suggests’ because larger sample sizes may reveal differences that were too small to be detected in the present study. The strength of the direct effects of leaf water content, mesophyll thickness and protruding mid-vein thickness on interspecific variation in SLA did differ between treatments and these gave rise to the interactions that were detected in the ANOVA’s.

The notion of an ecological strategy implies that certain suites of plant traits are more compatible than others in a given environment (Grime 1979; Tilman 1988; Westoby 1998). SLA is often invoked as an important trait in this regard. If species survive or not in a given habitat based partly on their SLAs, then the path model should be able to predict how the selective importance of the underlying components change as well. All of the various analyses converge in concluding that leaf water content, and therefore tissue density, was the most important determinant of SLA. This was true both when comparing across changing environmental conditions, and interspecifically, when comparing species with inherently different leaf water contents in a constant environment. The path coefficients give the importance of each component when it alone varies. Since three of the components have significant covariances between them, and since different species will vary simultaneously in all of these components, the values of the path coefficients can not be used alone. However, since the covariances appear to be constant across environments, the ratios of the path coefficients can indicate the relative importance of each component in determining SLA. Thus, the path model predicts that interspecific differences in leaf water content would be especially important in low-light environments while interspecific differences in leaf thickness would increase in importance (but would still be secondary to leaf water content) in the high-light environments. Comparisons of the levels of interspecific variation in each trait in such contrasting environments could therefore provide independent tests of our results.


This study was financially supported by the Natural Sciences and Engineering Research Council of Canada and by scholarships to D.M. from the Canadian International Development Association and the Université de Sherbrooke.


Mean values of eight leaf attributes on 22 species of herbaceous angiosperms in each of four experimental treatments. Variables: SLA (cm2 g–1), water content (%), lamina thickness, superior and inferior epidermal thickness, palisade + spongy mesophyll thickness, mid-vein thickness and thickness of the mid-vein support tissues; all thickness measures are in μm