1Ecologists have long recognized that plants occurring in areas of low rainfall or soil nutrients tend to have smaller leaves than those in more favourable regions.
2Working with a large data set (690 species at 47 sites spread widely through south-east Australia) for which this reduction has been described previously, we investigated the morphology of leaf size reduction, asking whether any patterns observed were consistent across evolutionary lineages or between environmental gradients.
3Leaf length, width and surface areas were measured; leaf traits such as pubescence or lobing were also scored qualitatively. There was no correlation between soil phosphorus and rainfall across sites. Further, there was no evidence that pubescence, lobing or other traits assessed served as alternatives to reduction of leaf size at the low ends of either environmental gradient.
4Leaf size reduction occurred through many combinations of change in leaf width and length, even within lineages. Thus consistent patterns in the method of leaf size reduction were not found, although broad similarities between rainfall and soil P gradients were apparent.
In Australia, rainfall and total soil phosphorus are seen as the most important abiotic factors shaping vegetation structure (Beadle 1962; Beadle 1966; Cunningham et al. 1999). Soil P determines what kinds of species occur at a given location, as total soil P is decided by the geomorphology of the site rather than by traits of the vegetation, as might be the case for soil nitrogen or other measures of available nutrients (Cunningham et al. 1999; Fonseca et al. 2000).
It is generally thought that smaller leaf sizes are favoured at lower rainfall because larger leaves develop thicker boundary layers of still air (Givnish 1978; Givnish 1979; Givnish 1984; Givnish & Vermeij 1976; Parkhurst & Loucks 1972; Schuepp 1993; Vogel 1970). Larger leaves therefore shed heat more slowly and are heated above air temperature more than smaller leaves. This overheating will become increasingly important as rainfall declines, because transpiration is also effective in shedding heat, and because when there is less total foliage per unit ground area, more of the leaves are exposed to direct radiation. Hence, smaller leaves are favoured towards lower rainfall.
The functional basis of leaf size reduction towards lower total soil P is less clear. Some suggestions, for example that a smaller leaf size is favoured because of the slower growth of plants on infertile soils (Rodriguez et al. 1998), can equally be interpreted as favouring smaller investment in total leaf area, as opposed to leaf size per se. Despite this, smaller leaf size at low soil P is consistently observed (Cunningham et al. 1999; Fonseca et al. 2000). Consequently, these possibilities were tested along the soil P and rainfall gradients.
Other leaf traits may also reduce radiative overheating, for example, pubescence or a glossy/waxy cuticle may increase reflectance (Ehleringer 1982; Johnson 1975; Miller 1983; Sandquist & Ehleringer 1997; Smith & Nobel 1977). Similar-sided leaves are indicators of a relatively steep leaf angle, whereas leaves with horizontal orientation typically concentrate photosynthetic tissue on the upper side, and stomata on the underside. Steeper leaf angles may reduce leaf overheating and thus transpiration in dry climates. Lobed leaves have shorter airflow paths across their surfaces relative to entire leaves of the same total area. Lobed leaves should therefore be able to shed heat more rapidly by convection (Givnish 1978; Givnish 1979; Parkhurst & Loucks 1972; Schuepp 1993). Therefore steep leaf angle or lobing might function as alternatives to smaller leaf size – clades with these traits might not reduce leaf size to the extent of those without them. Conversely, it might be argued that the same environments and lifestyles that favoured pubescence or steep leaf angle would also favour smaller leaf size, producing a positive rather than a negative correlation. Thus the relationship between these leaf characters and leaf size reduction was of interest in this study.
Materials and methods
A total of 47 sites throughout New South Wales in south-eastern Australia were selected to span two orders of magnitude of total soil P (range 22–2063 µg g−1) and one order of magnitude of mean annual rainfall (range 206–2034 mm). Soil P and rainfall were uncorrelated across the 47 sites (Pearson's correlation coefficient = −0·20, P = 0·43). Variation in temperature was restricted by confining sites to a latitudinal band of 30–35°S and to elevations below 800 m asl (see Fonseca et al. 2000 for site details). Five soil cores were taken at each site, each 10 cm deep. These were pooled to average out within-site soil heterogeneity, then analysed by the Soil Laboratory at James Cook University, Townsville, Australia, for total soil P and N. Soil P was assessed according to the protocol outlined by Fonseca et al. (2000), while N was determined using the single-digestion method of Anderson & Ingram (1989), after initially grinding soil samples to 0·2 mm and digesting with sulphuric acid and hydrogen peroxide. Nitrogen was then determined colorimetrically by the salicylate–hypochlorite method of Baethgen & Alley (1989).
In all, 1190 samples of three leaves were collected from each evergreen perennial species present in plots: a total of 690 species, 109 of which occurred at more than one site. Epiphytes, parasites and annuals were excluded from analyses, as these species could not be expected to have the same relationship as evergreens with year-round environmental parameters. For brevity, ‘leaf’ is used here to refer to any green surface carrying out most of the plant's photosynthesis. Leaves sampled were adult, free of herbivore damage, and had recently expanded fully. Leaves were taken from as near to a full sun position, and from as many different individuals, as possible. Within-species variation was small relative to variation across species (data not presented).
In the laboratory, images of individual leaves were scanned using an HP DeskScan II Scanner (Hewlett Packard, USA) and quantified using delta-t scan image analysis software (Delta-T Devices, Burwell, UK). Attributes quantified were leaf area (mm2), length (mm) and width (mm). Length was measured along the midrib of the leaf, rather than the linear distance between base and apex, to account for falcate leaves. Width was defined as the diameter of the largest circle that could be fitted within the leaf's margin. Individual leaflets of compound species were treated as equivalents of simple leaves.
The categorical variables of pubescence, lobing and presence of a glossy or waxy cuticle were scored as present/absent from field samples, from herbarium specimens from a nearby locality, or from descriptions in Harden (1993). Leaves were scored as different-sided if veins were prominent on one side only and the two sides were different colours. As specimens/descriptions could not be obtained for all species, analyses of leaf traits were restricted to 474 of the 690 species sampled.
Continuous variables were log10-transformed to improve normality. If species occurred at multiple sites, means were used, including site properties. If categorical traits differed between sites, median scores were chosen (this was rare). Cross-species statistical tests were assessed at α = 0·05 using spss for Windows version 8·0 (SPSS Inc., Chicago, IL).
Evolutionary divergences were assessed using phylogenetic regressions (Grafen 1989; Grafen 1991; Grafen 1992). In these analyses the item of evidence is the node or radiation in a phylogeny, rather than the present-day species (Grafen 1989). The working phylogeny used in assessing evolutionary divergences was constructed to family level following APG (1998) and Bremer, Bremer & Thulin (1998); lower nodes followed Wright et al. (2000). Families of unknown affinity at order level were placed at the most fully resolved point in the phylogeny. This method retains all data, yet adds no extra divergences and thus degrees of freedom, making it unlikely to bias regressions. Similarly, all samples not completely identified were incorporated in a polychotomy at the most fully resolved point.
Phylogenetic regressions were carried out using phylo8.glm version 8 for IBM (Royal Statistical Society, London, 1997) run using glim 4. As the working phylogeny is incomplete with respect to the actual phylogeny (not all species within clades were sampled), we used the fully general method to determine branch lengths (Grafen 1991). The method of branch length determination did not significantly affect results.
Nodes were identified that experienced a residual divergence in either soil P or rainfall greater than 0·3 orders of magnitude, after partialling out the effect of the other factor. These two groups of nodes define evolutionary divergences along soil P and rainfall gradients, respectively. Divergences were expressed such that divergences in leaf length were positive. Only species identified at least to genus were incorporated in these analyses. Species classified to the family level or less would have biased calculations of means, as each branch in a polychotomy was given equal weighting regardless of the number of species sampled within it. Leaf shape changes at divergences of more than 0·5 orders of magnitude of length or width were then summarized into four classes. In length-only change, divergence in log length was more than twice as great as divergence in log width. In width-only change, divergences in log width were more than twice as great as divergences in log length. In isometric change, log length and log width divergences were within twofold of each other. In opposite change, log width increased while log length decreased.
Across species, leaf area, length and width tended to decline towards lower soil P and lower rainfall, with considerable scatter (Fig. 1). This was expected, and was previously reported for site means of leaf width by Fonseca et al. (2000). More variation was explained by rainfall than by soil P for all three leaf size attributes (Fig. 1). Multiple regressions showed that soil P and rainfall had largely additive effects. Close to the same amount of variation in these relationships was explained as the sum of simple regressions for each gradient alone (leaf area, R2 both gradients = 0·246, soil P R2 = 0·074, rainfall R2 = 0·137; length, R2 both gradients = 0·106, soil P R2 = 0·021, rainfall R2 = 0·072; width, R2 both gradients = 0·25, soil P R2 = 0·088, rainfall R2 = 0·126; all P < 0·0005).
Similarly, evolutionary divergences for all three leaf size attributes were significantly correlated with divergence in soil P after controlling for divergence in rainfall and vice versa (Table 1). Leaf area, length and width all tended to be smaller on the lower P or lower rainfall side of divergences. Phylogenetic regressions generally explained more variation than cross-species analyses.
Table 1. Effects of soil phosphorus and mean annual rainfall on leaf surface area, length and width as determined by phylogenetic regressions (P < 0.001 for all regressions)
After controlling for the soil P and rainfall gradients, soil N, mean annual temperature and site elevation added little further predictive power explaining variation in leaf area in cross-species analyses (multiple regression predicting leaf area from all site properties: R2 = 0·253 vs R2 = 0·246 for soil P and rainfall alone, residuals not significantly greater; F3,684 = 2·34; Fcrit 3,∞ = 3·12; P > 0·05). This was expected as the sampling regime minimized variation in elevation and mean annual temperature, and soil N was correlated with soil P. Similarly, evolutionary divergences in all three leaf size attributes were not explained further by divergences in mean annual temperature, elevation or soil N after controlling for divergences in soil P and rainfall (data not presented, all P > 0·35). Hence leaf divergence was studied in relation to soil P and rainfall only, following the original design of the sampling.
are other leaf traits alternatives to leaf size reduction?
For most leaf traits (such as pubescence or lobing) examined in cross-species analyses, there was no evidence of an interaction between trait presence/absence and leaf size, after controlling for main effects (Table 2). Leaf size did decline towards lower rainfall among different-sided leaves (R2 = 0·119, F1,179 = 55·3, P < 0·0005), but not among similar-sided leaves (R2 = 0·041, F1,293 = 2·64, P = 0·11; significant interaction in Table 2). However, species with different-sided leaves averaged larger leaf areas than similar-sided species across most of the rainfall range, which does not support the idea that similar-sided leaves (vertical leaf angles) function as an alternative to small leaf size.
Table 2. Results of ancovas examining the relationship between leaf surface area along the soil phosphorus and mean annual rainfall gradients and the presence/absence of the categorical leaf characteristics: pubescence, lobing, waxy/glossy cuticle and similar-sided leaves
Similarly, the presence or absence of leaf traits largely did not affect evolutionary divergences in leaf area along the soil P or rainfall gradient (Table 3). The only significant interaction observed involved the presence/ absence of lobing along the rainfall gradient. However, the direction of this interaction was again contrary to the hypothesis that increased lobing is an alternative to decreasing leaf area towards lower rainfall. Clades static with respect to the level of lobing had the smallest decrease in surface area, regardless of whether they had unlobed or lobed leaves. Further, the interaction with lobing was only marginally significant, and it should be borne in mind that 10 different tests at α = 0·05 were carried out (Table 3). It therefore appears unlikely that lobing is an alternative to reducing leaf size at lower rainfall.
Table 3. Interaction term statistics for leaf surface area with pubescence, lobing, similar-sided leaves and glossy/waxy cuticle status in a multiple phylogenetic regression along soil phosphorus and mean annual rainfall gradients
evolutionary divergences in leaf size: evidence for clade-specific responses
Given that leaf area tended to decrease towards lower rainfall and soil P, we investigated how this was accomplished in terms of changes in leaf length and width. Of 263 higher nodes in the phylogenetic tree, 122 represented divergence of more than 0·3 in log soil P after partialling out the effect of the rainfall gradient (Fig. 2a). Among these 23, 39, 42 and 18 were classified as changing shape in isometric, width-only, length-only or opposite fashion, respectively (Table 4). There were 85 nodes representing divergence in log rainfall of more than 0·3 after controlling for the effects of soil P (Fig. 2b); among these 15, 23, 36 and 11 were classified as changing shape in isometric, width-only, length-only or opposite manner, respectively (Table 4). These distributions were significantly different from a 1 : 1 : 1 : 1 ratio (soil P χ2 = 13·67, df = 3, P = 0·003; rainfall χ2 = 17·17, df = 3, P = 0·001). However, the most striking feature of the distribution of changes was that all four were commonly observed, with no strategy dominating along either gradient (Fig. 2). Further, mixtures of leaf shape adjustments were not significantly different between soil P and rainfall divergences (χ2 = 1·40, P = 0·71; Table 4), with nodes tending to follow the same broad patterns along both gradients.
Table 4. Method of leaf size reduction utilized by higher nodes in the working phylogeny in response to changes in soil phosphorus or in mean annual rainfall greater than 0·3 orders of magnitude, after controlling for the effects of the other gradient
Clades were classified as using either opposite (O); isometric (I); width-only (W); or length-only (L) change, see Materials and methods for definitions).
Response of a node at the base of an entire clade.
We could find few clear patterns among lineages in the shape adjustments to soil P or annual rainfall gradients. Asterids and caryophyllales tended to show width-only change, while rosids as a group tended to show length-only change along both gradients. However, this commonality was not representative of the data set as a whole (Table 4), with a mixture of adjustments present in most clades.
As expected, a reduction in leaf size towards lower soil P and rainfall was observed in both cross-specific analyses and among evolutionary divergences. However, the aim of this study was to determine the manner of this reduction in leaf size. Reduction in leaf length and width (isometric change) could easily be achieved by abbreviation of leaf expansion. On this basis, isometric leaf size reduction was expected to be the most common method. However, leaf narrowing with little shortening, or shortening with little narrowing, were at least as common along both gradients.
It might be expected that developmental propensities in particular clades would favour one sort of leaf size adjustment over another. Similarly if herbivory intensity (Brown & Lawton 1991; Obeso 1997; Rivero-Lynch, Brown & Lawton 1996; Stone & Bacon 1995); architectural considerations (Ackerly & Donoghue 1998; Brouat et al. 1998; White 1983); leaf chemistry (Cunningham et al. 1999); or leaf orientation (Valladares & Pugnaire 1999) varied between clades, any of these factors might affect the fitness of particular leaf shapes in a given environment. However, we found little evidence for clade-specific patterns in the shape changes associated with leaf size reduction. While asterids and caryophyllales often adjusted leaf width alone, and rosids leaf length, these responses were universal for only four nodes sampled within the caryophyllales. Moreover, common responses among other clades were not observed.
If smaller leaf size is selected at low rainfall, and perhaps soil P, principally to decrease the likelihood of leaf overheating in the sun, other leaf traits such as pubescence, lobing and a steep leaf angle (similar-sided leaves) might be expected to interact with the evolution of smaller leaf size. Nevertheless, most such interactions were absent in this flora. The two interactions that were statistically significant were not consistent with substituting these traits for smaller leaf size to avoid overheating. Thus these two interactions should probably not be regarded as biologically meaningful.
A long-standing issue in plant ecology is to what extent different types of environmental stressors favour a common suite of traits (Chapin, Autumn & Pugnaire 1993; Grime 1977) vs distinct traits (Grubb 1985; Harper 1982). Some relationships among leaf traits hold true across a wide range of environments, for example, leaf area per mass (specific leaf area, SLA), leaf lifespan, potential photosynthetic rates (Amax) and leaf N concentrations are interrelated (Reich 1993; Reich, Walters & Ellsworth 1997; Reich et al. 1999). These relationships persist at lower rainfall and humidity, but are shifted to shorter leaf lifespan and higher leaf N at a given SLA (Wright, Westoby & Reich 2002). Cunningham et al. (1999) investigated chemical and structural traits of leaves in species pairs that represented independent evolutionary divergences and were contrasted in relation to either rainfall or soil P. Some trait shifts were similar at low rainfall compared to low soil P; others were different. Fonseca et al. (2000), reporting the field work on which the present paper also draws, examined site means for leaf width, SLA and potential canopy height. They showed not only that leaf width declined towards both lower rainfall and lower soil P, but also that it declined in the same relationship to SLA. In this sense, common leaf width–SLA combinations were found under both low rainfall and low soil P. On the other hand, leaf width declined in a different relationship to potential plant height at low soil P compared to low rainfall.
Our analyses found no evidence that leaf size reduction occurred in different ways along rainfall gradients compared to soil P gradients, with a mixture of each type of shape change spread throughout all lineages. One possible interpretation of this result is that both gradients favour similar leaf-trait combinations. However, leaf size reduction occurred in many ways, at different phylogenetic divergences, along each gradient. Further, patterns of common methods of leaf size reduction were not found within lineages, nor were correlations of shape change and other leaf traits such as pubescence detected. Taken together, this suggests that many mechanisms are possible for reducing leaf size. The similarities between the soil P and rainfall gradients observed in this data set are more likely to be caused by averaging across many different reductions, and are not evidence for strong evolutionary convergence.
We thank Barbara Rice for her assistance in data collection and specimen identification. David Warton provided statistical advice, as did Alan Grafen, who also gave valuable advice in running the phylogenetic regressions. Ian Wright assisted with compiling the working phylogeny and numerous tips on using glm. The research was supported by Australian Research Council funding to M.W. Two anonymous reviewers provided helpful comments on an earlier draft. This is contribution 373 from the Research Unit for Biodiversity and Bioresources, Macquarie University.