Leaves at low versus high rainfall: coordination of structure, lifespan and physiology

Site and species selection

Traits of fully expanded, outer-canopy leaves were measured on perennial species sampled from nutrient-rich and nutrient-poor sites in each of two rainfall zones in eastern Australia (site details given in Table A1, species list in Appendix). The same set of 75 species was used here for which LL and LMA were reported previously (Wright et al., 2002), with the exception that leaves of Eutaxia microphylla (Fabaceae) were too small for work to shear to be measured. For Exocarpos aphyllus (Santalaceae) and Bossiaea walkeri (Fabaceae), both leafless stem-photosynthesisers, ‘leaf’ traits were actually measured on the youngest cohort of stems. For Callitris glaucophylla (Cupressaceae), traits were measured on terminal ‘branchlets’ of closely appressed leaves.

Measurement of leaf traits

Average LL for each species was estimated as the inverse of the rate of annual leaf turnover (Southwood et al., 1986; Ackerly, 1996) measured over 2 years (June 1998 to June 2000). For the two leafless, stem-photosynthesising species, ‘leaves’ were defined as unbranched twigs, ‘branches’ as twigs with twigs coming off them. Both ‘leaf’ death and ‘leaves’ morphing into a ‘branch’ were defined as mortality; leaf longevity was then calculated as above. Note that Wright & Cannon (2001) estimated LL for species at the high rain, nutrient-poor site based on 1 year of leaf mortality censuses whereas the data presented here for that site include a second year of census data, and thus should be a more robust estimate of average LL for these species.

LMA was calculated from leaf dry mass (oven-dried for 48 h at 65°C) and one-sided leaf area (flat bed scanner; needle leaves assumed to have circular cross-section and leaf area adjusted by π/2). For LMA, five leaves from each of five individuals were sampled per species for all sites except the drier, nutrient-rich site, for which one leaf from each of five individuals was sampled in December 1997 and a further 6–10 leaves collected in November 1998. Leaf tissue density (dry mass per unit volume) was calculated as LMA divided by leaf thickness.

Leaf work to shear (W_shear), thickness and toughness were measured for one leaf from each of three–five individuals per species (generally five), except at the high rainfall, nutrient-poor site, for which two leaves from each of 3 individuals per species were measured (Wright & Cannon, 2001), giving an overall average of 5.2 samples per species. Leaf thickness was measured at two–five points per leaf with a dial gauge micrometer. Major veins and the midrib, where present, were avoided on all but the smallest leaves; otherwise placement of the micrometer was haphazard. The cross-section of needle leaves was assumed to be circular, and average leaf thickness was adjusted accordingly in these cases (multiplied by π/4).

Leaf work to shear (W_shear) was determined using a custom-built machine, which was not intended to simulate any particular herbivore, but to provide a generalised measure of physical defence (Wright & Cannon, 2001). The machine measures the work required to cut a leaf at a constant cutting angle (20°) and speed, and is similar in principle to systems described by Darvell et al. (1996) and Aranwela et al. (1999). W_shear and tissue toughness (= average W_shear/thickness) measure different aspects of how robust a leaf or leaf tissue is. Both would likely be correlated with other physical properties such as leaf stiffness, tensile strength and resistance to puncture by a penetrometer (Vogel, 1988; Choong et al., 1992; Wright & Illius, 1995; Edwards et al., 2000).

One cut at right angles to the midrib was made per leaf for measurement of W_shear. Leaves were cut at the widest point along the lamina, or half way between the leaf base and tip if the leaf had no obviously widest point. For species with a prominent midrib, this feature could be discerned on the machine output (Wright & Cannon, 2001), and the W_shear and toughness of the lamina were calculated separately in addition to that integrated over the whole leaf. This was possible for 33 species (24 at high rainfall, nine at low rainfall); for the remainder it was not because the midrib was not sufficiently prominent relative to other features in the graph, or because no midrib was present (e.g. needle leaves). Whole-leaf and lamina-only measures of W_shear and toughness were tightly correlated (across all species, r² > 0.95, slopes not different from 1). Consequently we report whole-leaf measures only, because this makes no qualitative difference to the results, and because these measures are more analogous to other whole-leaf properties, such as LMA and density.

Leaves were sampled for determining average total nitrogen concentration over a 2.5-year period at the drier, nutrient-rich site (December 1997–June 2000) and for c. 2 years at all other sites (June 1998–August 2000). Between two and four leaf collections were made per species over that time (average 3.0). One collection consisted of leaves used for photosynthesis measurements (Wright et al., 2001). For other collections, five leaves were collected from each of five individuals for each species. Only fully expanded outer canopy leaves were sampled. Leaves were finely ground, and leaf N concentration was measured by Kjeldahl digestion followed by colorimetric analysis (undertaken at CSIRO Plant Industry, Canberra).

Data considerations

Individual leaf trait measurements were averaged for each species at a site since we were primarily interested in cross-species trait relationships, and differences in these relationships among sites. Since the pattern we were trying to understand involved a shift in trait relationships according to site rainfall, for brevity we focus just on that aspect rather than also examining shifts according to site nutrient status within rainfall zones (cf. Wright et al., 2002). Species-mean data are listed in Appendix. Variance components analyses (ANOVA, type I sums of squares, log transformed variables) indicated that > 84% of variation in leaf thickness, W_shear, toughness, leaf size and N concentration was associated with differences between species as opposed to within-species. Thus, treating these variables as species means in subsequent analyses was strongly supported. Species-means of all variables showed approximately log-normal distributions and were deemed normal following log transformation (Kolmogorov-Smirnov test, α = 0.05).

Cross-species analyses

Data were analysed primarily by fitting Standardised Major Axis (SMA) slopes since each variable had variation associated with it due to both measurement error and species-sampling, hence it was inappropriate to minimise sums of squares in the Y dimension only (Sokal & Rohlf, 1995). SMA analysis is also known as Geometric Mean Regression because a SMA slope is equal to the geometric mean of the model 1 regression of Y on X, and of the reciprocal of the regression coefficient of X on Y. SMA ‘scaling’ slopes, calculated on log-transformed variables, give the proportional relationship between variables. Slopes were first fitted across the species within each site, with confidence intervals (95%) calculated following Pitman (1939). Tests for homogeneity of slopes and calculation of common slopes used a likelihood ratio method (Warton & Weber, 2002). The ability to calculate common slopes allows one to test for elevation (i.e. intercept) differences between individual slopes, as in standard analyses of covariance (ANCOVA). Where nonheterogeneity of slopes was demonstrated, we tested for elevation shifts in SMAs between high and low rainfall sites at each soil nutrient level (low or high) by transforming slopes such that the common slope was 0 (Y′= Y –βX, where β is the common slope) and then testing for differences in group mean Y′ with t-tests (Wright et al., 2001).

Phylogenetic analyses

Key trait relationships were also explored with ‘phylogenetic’ analyses in order to ascertain whether cross-species relationships were being driven by differences between higher taxonomic groups. By superimposing the species-mean data onto a phylogenetic tree of the study species (Appendix Table A1), inferred evolutionary divergences in one trait can be tested for correlation with divergences in another. In these ‘correlated divergence analyses’ (Westoby et al., 1998), each independent divergence (radiation) contributes a single item of evidence. This independence has led to these types of analyses being known as ‘Phylogenetically Independent Contrast’ or PIC analyses (Felsenstein, 1985; Harvey & Pagel, 1991).

A ‘contrast’ dataset was created for each site, in which the value assigned to each contrast was calculated as the difference between the trait values for the two nodes or species descending from the contrast-node. Node values were themselves calculated as the average of trait values for the two immediately lower nodes or species. The direction of subtraction in calculating contrasts is unimportant, providing all traits are treated in the same manner. Hence, in a graph of divergences in one trait against divergences in another, a data point indicating a positive divergence in both traits would have indicated negative divergences in each trait had the subtractions been performed the other way around. Due to this symmetry, regressions of contrast data have no intercept term (they are ‘forced’ through the origin; Grafen, 1989; Westoby et al., 1998). SMA slopes were calculated for correlated divergence analyses as the geometric mean of the model 1 regression of Y on X, and of the reciprocal of the regression coefficient of X on Y. Confidence intervals (CIs) from the regression of Y on X were used to estimate CIs for these SMA slopes (Sokal & Rohlf, 1995). Classification of species into families and higher groupings followed phylogenies published by the Angiosperm Phylogeny Group (Soltis et al., 1999), while generic delineation followed (Harden, 1990) and Flora of Australia (–, 1981). Additional resolution was obtained from consensus cladistic trees utilising either molecular or morphological data, both from recent publications and from unpublished trees provided kindly by taxonomists studying the relevant groups (further details available from the authors).

LL vs LMA and its components (leaf thickness and density)

Leaf lifespan (LL) varied c. nine-fold, and leaf mass per area (LMA) six-fold, across the 75 perennial species. LL was positively associated with LMA within each site (Fig. 1a, Table 2), as were evolutionary divergences in the two traits (Table 3). Individual cross-species SMA slopes ranged from 1.2 to 2.1 but were deemed nonheterogeneous (P = 0.072), with a common fitted slope of 1.3 (95% CI, 1.1–1.6). SMAs for high and low rainfall sites were clearly separated, such that LL at dry sites tended to be c. 40% shorter at a given LMA, or 30% higher LMA was seemingly required at dry sites to achieve a given LL (t-tests of Y′: nutrient-rich sites, P= 4 × 10⁻⁵; nutrient-poor sites, P= 0.003).

Of the two components of LMA, leaf thickness was more tightly correlated with LL than was tissue density. This was the case in both cross-species and correlated divergence analyses (Fig. 1b,c, Tables 2 and 3). Similarly, the elevation shifts in LL – LMA relationships between high and low-rain sites were driven more strongly by variation in leaf thickness than in density. That is, dry-site species tended to have thicker leaves at a given LL (although, since SMA slopes were heterogeneous, P < 0.001, shifts in elevation with rainfall could not be formally tested). There was no apparent separation of LL – density combinations according to site characteristics (Fig. 1c). On average, dry-site species had thicker leaves than high rainfall species (mean 0.45 vs 0.34 mm; t-test P = 0.003), but rather similar mean density (0.44 vs 0.40 g cm⁻³; P = 0.110).

W_shear vs LL and LMA

Average work to shear (W_shear) varied 87-fold between species. Within each site, leaves with higher W_shear had higher LMA and achieved longer lifespans, both in cross-species and correlated divergence analyses (Fig. 2a,b; Tables 2 and 3). Individual LL – W_shear slopes were heterogeneous (P = 0.032). However, after pooling species by rainfall zone, the slopes were nonheterogeneous (P = 0.089), and did not differ in elevation (P = 0.300). That is, the clouds of high and low rainfall points were not clearly separated in elevation as was the case for the LL – LMA or LL – leaf thickness relationships. By contrast, the SMA slopes for W_shear vs LMA were separated by rainfall (Fig. 2b), with dry-site species having lower W_shear at a given LMA (test for common slopes, P = 0.888, β= 2.7; t-tests for elevation differences within either soil nutrient class, P < 1 × 10⁻⁵). These key results contradicted our original assumption (see Introduction) that LMA could be considered a fair index of leaf strength, irrespective of habitat, thus the LL – W_shear slopes would be separated by rainfall, similar to the LL – LMA slopes. The common slopes calculated across sites indicated that, within a given rainfall zone, a doubling in LMA was associated with a c. six-fold increase in leaf strength (common log-log slope of 2.7) yet only a c. 2.5-fold increase in LL (common slope of 1.3).

Toughness – density – leaf N relationships

Since leaf thickness is a component of both LMA and work to shear, the separation in LMA – W_shear relationships according to site rainfall logically indicated a separation in density – toughness relationships between the rainfall zones (Fig. 3). Density and toughness were clearly correlated in three of four sites (r² = 0.38–0.62), but not at the nutrient-rich, low rain site (r² = 0.002). Still, fitting a common SMA slope across the sites (justified on the basis of a test for slope heterogeneity, P = 0.120), indicated that tissue toughness was 2–3-fold greater at a given density for high rainfall species (t-tests for elevation shifts, all P < 0.001).

On average, leaves of low rainfall species were half as tough as those of high rainfall species (t-tests, P < 0.05 within either soil nutrient class), but had much higher leaf N, whether considered per unit mass (N_mass, 1.5-fold higher, P < 0.01 within either soil nutrient class) or per unit area (N_area, two-fold higher, P < 0.001 within either soil nutrient class). At all sites, leaves with higher N_mass were less tough (Fig. 4), as found previously by Coley (1983) and Loveless (1962), although weaker associations in correlated divergence analyses suggested that these trends were driven to some extent by differences between higher taxonomic groups (Table 3).

Leaf tissue composition of high and low rainfall species

Many of the leaf traits common to species in dry habitats (e.g. thick, hard leaves with thick cuticles and small, thick walled cells) have been interpreted as increasing the ability of leaves to withstand high levels of desiccation without wilting, thereby avoiding the tissue and airspace compression which would slow or stop photosynthesis, or lead to leaf death (Small, 1973; Orians & Solbrig, 1977; Grubb, 1986; Turner, 1994; Cunningham et al., 1999; Niinemets, 2001). Previously it has been shown that the concentration of cell wall, vascular tissue, fibre or sclerenchyma in a leaf is correlated with both tissue toughness (Choong et al., 1992; Wright & Illius, 1995; Lucas et al., 2000) and density (Garnier & Laurent, 1994; Roderick et al., 1999). Thus, leaves of low rainfall species might be expected to be both tougher and more dense than leaves of high rainfall species. This was not what was found for this set of species. Instead, we found that low rainfall leaves were on average about half as tough but similar in density to high rainfall leaves, and thus less tough at a given density. That a given density can result in a wide range of tissue toughness (Fig. 3) indicates that the configuration of cells and mixture of cell types can affect density and tissue toughness somewhat independently. For example, the toughness of tracheids and fibres is ten-fold greater than that expected from their cell wall concentration alone, since individual cells may be elongated with microfibrils directed uniformly at a small angle to the axis of the cell (Lucas et al., 2000).

The similar mean density but lower toughness of the dry-site species under study here suggests that they had a greater proportion of low toughness tissue type(s) in their leaves. Their higher leaf N, considered on either a mass or area basis, suggests that this might predominantly have been mesophyll tissue. Some tentative support for this prospective explanation can be found from overlap with datasets in which the depths of different tissue types in leaves were estimated from transverse leaf sections (Cunningham et al., 1999, D. H. Duncan & M. Westoby, unpublished results). For the 23 species in common between these datasets and ours (17 at high rain, six at low rain), low rainfall species had approximately twice as high leaf N per area (N_area; t-test, P < 0.0001) and twice as thick palisade mesophyll (P = 0.036), although the proportion of palisade was not significantly higher (59% vs 45%, P = 0.321).

It may be that the trends we report for species in eastern Australia are quite common differences between high and low rainfall perennial species. Although leaf strength has not been measured for large numbers of species worldwide, the tendency for leaf density, thickness and LMA to increase with site aridity is well known (Maximov, 1929; Mooney et al., 1978; Specht & Specht, 1989; Schulze et al., 1998; Cunningham et al., 1999; Niinemets, 2001). Higher mean N_area with increasing aridity is also probably quite general, being reported by Cunningham et al. (1999), for example, and implicit in the fact that low rainfall species tend to have high LMA in combination with similar (if not higher) N_mass to high rainfall species (Killingbeck & Whitford, 1996). In the next section we focus on the apparent benefits of deploying high N_area leaves in low rainfall, low humidity habitats.

Costs and benefits of a high LMA, high leaf N strategy at low rainfall

In general, the construction cost per gram of leaf does not vary systematically with either LMA (Chapin, 1989; Poorter & De Jong, 1999) or LL (Villar & Merino, 2001). This implies that the (generally) higher LMA leaves of dry-site species are more expensive to build per unit leaf area than leaves of high rainfall species. Constructing leaves with high N_area may incur additional costs – in terms of higher respiratory costs for protein turnover (Wright et al., 2001), higher costs from N acquisition and, quite possibly, additional costs related to the increased attractiveness to herbivores of high N leaves. On the other hand, deploying high N_area leaves in low rainfall habitats seems to have substantial benefits. First, in terms of maximising use of the typically high irradiance in such places (Cunningham et al., 1999; Roderick et al., 2000; Niinemets, 2001), but – and we would argue, more importantly – high N_area appears to have particularly significant benefits in terms of enhancing water conservation during photosynthesis (Field et al., 1983; Mooney et al., 1978; Wright et al., 2001). If high N_area is generally linked with low leaf tissue toughness, as was found for low rainfall species in this study, this implies that the economics of N and water use are intrinsically linked with the dry-mass economics of leaf construction (LMA) and LL.

The way in which adaptations of low rainfall species can be understood in terms of this argument is illustrated in Fig. 5. Both transpiration and photosynthesis approximately follow Fick's Law for the diffusion of gases through a surface: transpiration (E) = leaf-to-air vapour pressure deficit (VPD) × stomatal conductance to water (G_s); and photosynthesis (A_area) = G_c(c₂ − c_i), where G_c is stomatal conductance to CO₂, and c_a and c_i are ambient and leaf internal CO₂ concentrations, respectively (Lambers et al., 1998). Low rainfall sites are characterised by low atmospheric humidity, hence high leaf-to-air vapour pressure deficits (high VPD), and thus higher transpiration rate (E) at a given stomatal conductance to water (G_s; box 2 in Fig. 5). Since G_c and G_s are related by a constant (the molar diffusion ratio between CO₂ and water), this implies higher transpirational water loss for a given G_c also. Consequently, more water would be used for a given A_area : G_c, that is all else being equal, the water use efficiency (A_area : E) of dry-site species would be lower (Fig. 5, box 3).

On the other hand, we have found that low rainfall species have c. two-fold higher N_area than high rainfall species (Fig. 5, box 4), and substantially higher A_area at a given G_c (but not higher average A_area; Wright et al., 2001). Higher A_area : G_c indicates a larger drawdown of internal CO₂ (Fig. 5, boxes 5, 6) which, presumably, was possible via higher photosynthetic enzyme content, as indicated by high N_area. All else being equal, this would confer higher water use efficiency (Fig. 5, box 7). Instead however, this tendency and the tendency towards lower WUE as a result of the higher leaf-to-air VPD at low rainfall (Fig. 5, box 3) appear to cancel each other out, resulting in little or no net difference in WUE between low and high rainfall species (Wright et al., 2001; Fig. 5, box 8).

As described above, thicker mesophyll in leaves of dry-site species, inferred from higher leaf N, led to lower tissue toughness (Fig. 5, box 9). At a given leaf thickness, lower toughness translates into lower W_shear (since W_shear= thickness × toughness; Fig. 5, box 10), which presumably should lead to shorter LL (Fig. 5, box 11). However, dry-site species tended to have thicker leaves (Fig. 5, box 12). At a given tissue toughness, thicker leaves would have higher W_shear (Fig. 5, box 13), which would tend to increase LL (Fig. 5, box 14). These countervailing trends (increased thickness, decreased toughness) apparently buffer each other, resulting in little or no net difference in mean W_shear and LL, comparing high and low rainfall species. However, dry-site species tended to have higher LMA because of their thicker leaves, and thus higher LMA at a given W_shear or LL or, equally, lower W_shear and LL at a given LMA (Fig. 5, box 15).