Global meta-analysis shows that relationships of leaf mass per area with species shade tolerance depend on leaf habit and ontogeny
Christopher H. Lusk
Tel: +61 2 9850 8165
Fax: +61 2 9850 8245
Email: [email protected]
Summary
It was predicted that relationships of leaf mass per area (LMA) with juvenile shade tolerance will depend on leaf habit, and on whether species are compared at a common age as young seedlings, or at a common size as saplings. A meta-analysis of 47 comparative studies (372 species) was used to test predictions, and the effect of light environment on this relationship. The LMA of evergreens was positively correlated with shade tolerance, irrespective of ontogeny or light environment. The LMA of young seedlings (≤ 1 yr) of deciduous species in low light was also positively correlated with shade tolerance; more weakly so in high light. By contrast, size-specific comparisons of deciduous saplings gave negative correlations of LMA with shade tolerance. Independent contrasts and cross-species analyses yielded broadly similar results. We conclude that ontogeny strongly influences the relationship of LMA with shade tolerance of deciduous trees, but has little impact on that of evergreens. Size-specific comparisons reveal opposing trends in deciduous and evergreen taxa: the negative relationship of LMA with shade tolerance of deciduous species is probably dominated by interspecific differences in palisade thickness, whereas patterns in evergreens are probably shaped more by the degree of structural reinforcement, linked to wide variation in leaf lifespan.
Introduction
Comparative ecologists have debated the existence or otherwise of a general stress-tolerant strategy common to plants native to various types of unproductive habitat (Grime, 1979; Tilman, 1988; Grubb, 1998). Plants native to habitats subject to chronic drought or nutrient scarcity share certain traits, such as robust leaf construction (large leaf dry mass per area (LMA)) (Westoby et al., 2002), although more subtle differences between species of these different habitats have also been shown (Wright et al., 2002; Reich et al., 2003). A large LMA appears to enable leaves to live longer on average: a global data set shows a twofold increase in LMA scales with a more than threefold increase in leaf lifespan (Reich et al., 1997; Westoby et al., 2002). By contrast, rapid turnover of lightly built leaves is common in plants native to productive environments (Lambers & Poorter, 1992). Although interpretations of this axis of variation differ in their emphasis, in general terms it can be seen as representing a trade-off of maximum resource pre-emption in favourable environments vs prolongation of the revenue stream obtained from hard-won resources in unproductive environments (Berendse & Aerts, 1987; Kikuzawa, 1995; Westoby et al., 2002).
The deep shade of forest understoreys is another widespread type of unproductive habitat. Late-successional tree species must often survive decades of slow incremental growth in the understorey before acceding to the canopy (Lorimer et al., 1988; Lusk & Smith, 1998). To what extent do the leaf traits involved in tolerating shade resemble those that confer tolerance of other unproductive environments (Grime, 1979; Reich et al., 2003)? Early ecophysiological work assumed that evolutionary responses of plants to shade would be similar to the plastic responses of individual plants (Boardman, 1977; Givnish, 1988). However, whereas plants respond plastically to shade by reducing LMA (Lambers et al., 1994), two extensive reviews found that seedlings of shade-tolerant tropical evergreens usually had larger LMA than light-demanding pioneers, when grown in similar light environments (Veneklaas & Poorter, 1998; Walters & Reich, 1999). Comparative studies of evergreen saplings under common conditions have also reported positive correlations of LMA with species shade tolerance (e.g. Williams et al., 1989; Lusk, 2002; Poorter & Bongers, 2006), although there is some evidence of attenuation of this relationship as plants grow larger (Poorter, 2007). Overall, these findings seem consistent with the proposal that tolerance of shade in evergreen forests, like tolerance of shortages of other resources, involves similar selection for robust leaf construction. The situation in deciduous forests is less clear. The review by Walters & Reich (1999) reported that the LMA of temperate deciduous species (like that of tropical evergreens) was positively correlated with shade tolerance variation; however, although their review encompassed juvenile plants of a wide range of sizes and ages, the focus was primarily on data from seedlings ≤ 1 yr old. The applicability of their findings to later ontogenetic stages has been questioned (Niinemets, 2006), and the results of some studies of deciduous saplings (e.g. Abrams & Kubiske, 1990; Lusk & Reich, 2000) depart quite strongly from the reviews’ conclusions.
Observed relationships of LMA with shade tolerance may depend on whether plants are compared at a common age or common size, and may also be influenced by light environment. Field studies of saplings usually compare species at a common size (e.g. Lusk, 2002; Poorter & Bongers, 2006). Studies of young seedlings, in contrast, usually compare species at a common age, generally ≤ 1 yr (e.g. Davies, 1998; Walters & Reich, 2000). Because of the correlation between seed size and successional status of angiosperms (Hammond & Brown, 1995; Hewitt, 1998), young seedlings of shade-tolerant species tend to be larger than pioneers of the same age (e.g. Popma & Bongers, 1988; Walters & Reich, 2000), although the faster maximum growth of the latter will enable them to rapidly overturn this advantage in moderate to high light. These initial size differences between pioneer and shade-tolerant species mean that comparisons of young seedlings are likely to be dominated by ontogenetic influences on LMA (Evans, 1972), rather than constitutive differences between species. Even when light environment is held approximately constant, the LMA of juvenile evergreens increases with plant height (Sack et al., 2002; Lusk, 2004). This is probably linked to ontogenetic increases in lamina size, which will require proportionally greater investment in structural support (Niinemets et al., 2007). There is also some evidence of increasing palisade layer thickness as plants grow taller (Kenzo et al., 2006), although it is unclear how much of this is a direct response to light environment. By contrast, comparisons at a common size are likely to find that shade-tolerant and light-demanding evergreens differ constitutively in LMA: achievement of long leaf lifespan by shade-tolerant species is likely to require intensive physical reinforcement to protect against physical stresses and herbivores, raising LMA. Shade-tolerant and light-demanding deciduous species, however, differ much less in leaf lifespan (Koike, 1988): there is therefore less reason to expect common-size comparisons to show larger LMA in shade-tolerant deciduous taxa.
This paper reports meta-analyses examining the effects of ontogeny, leaf habit and light environment on relationships of LMA with shade tolerance. A potential obstacle to conventional review approaches to these questions is the difficulty in fitting a common scale of shade tolerance to species from diverse regions. In view of ontogenetic drift in LMA, and the sensitivity of this trait to light and other resources, the heterogeneity of plant age, size and growth conditions across studies also poses potential problems for conventional review techniques. Meta-analysis obviates both these problems, by averaging effect sizes across studies (Hedges & Olkin, 1985; Hunter & Schmidt, 2004), rather than pooling data.
Materials and Methods
Study selection criteria
The ISI Web of Science and Google Scholar were searched for comparative studies reporting the LMAs of at least four woody dicot species grown under common conditions. The keywords used in searches were various combinations of ‘LMA’, ‘specific leaf area’ (‘SLA’), ‘seedling’, ‘sapling’ and ‘leaf area’. Only studies that included species differing in reported shade tolerance were of use for our meta-analytic approach: if we were unable to rank species by shade tolerance, it was not possible to measure relationships with LMA.
In order to test the influence of ontogeny, two types of study were sought: those that compared seedlings ≤ 1 yr old at a common age; and those comparing juveniles 20–500 cm tall at a common size. These two categories are referred to hereafter as ‘seedlings’ and ‘saplings’. Studies not satisfying either of these criteria were omitted, for example those using seedlings of widely differing ages, or plants produced from cuttings. In cases of doubts about plant age or size, the original authors were contacted for clarification. There is overlap in size between the two categories, as seedlings growing in high light can reach well over 20 cm tall in < 12 months. As far as we could tell, however, there was no overlap in age: all juveniles used in size-specific ‘sapling’ comparisons were apparently > 1 yr old.
Separate meta-analyses were carried out for deciduous and evergreen species. It was expected that leaf habit would significantly influence relationships of LMA with shade tolerance, in view of the much greater range of leaf lifespan in evergreen forests. The deciduous category includes studies carried out in both winter-deciduous temperate forests and drought-deciduous tropical forests. In both these biomes, seasonal constraints on leaf habit limit the scope for differences in leaf lifespan between shade-tolerant and light-demanding species. We therefore saw no reason to expect relationships of leaf structure with shade tolerance to differ between these two deciduous biomes. One study (Veenendaal et al., 1996) included eight deciduous species and seven evergreens: these two groups were analysed separately.
Study treatments were also sorted by light environment. Light environments were separated into two categories; those equivalent to ≤ 10% of daily photon flux density on open sites in the native habitat of the species concerned, and those < 10%. This approximately separates light environments characteristic of closed understoreys and small treefall gaps from those encountered by seedlings growing on sites opened by recent extensive disturbances (Chazdon & Fetcher, 1984). When more than one light intensity within either or both ranges was included in a study, we either used the light treatment that included most species, or else averaged results across light intensities. Most age-specific comparisons of young seedlings were carried out in artificial environments (glasshouses, nurseries or growth chambers), and most size-specific comparisons were carried out in the field. Field studies were included only if sufficient information was provided to enable comparison of species in a common light environment. In one field study of saplings (Ellis et al., 2000), only 13 out of a total of 21 species had been measured in comparable light environments, so the remaining eight were excluded.
The literature search identified a total of 20 suitable studies of seedlings ≤ 1 yr old, and 25 size-specific comparisons of saplings, spanning the period 1986–2007. Two unpublished data sets were also included (see Supplementary Material Table S1). Although over half of all studies were carried out on either northern temperate or tropical floras from the Americas, all major forested regions had some representation (Table 1). As many studies included measurements in light environments both ≤ 10% and > 10%, we obtained a total of 70 samples for meta-analysis. This data set included a total of 372 species. Despite overlap between the species lists of many studies, no two studies in the same combination (ontogenetic stage ? light environment) comprised exactly the same suite of species.
| Region | Seedling studies | Sapling studies |
|---|---|---|
| Temperate North America | 5 | 9 |
| Tropical America | 5 | 7 |
| Temperate South America | 0 | 2 |
| Tropical Africa | 2 | 0 |
| Europe | 4 | 1 |
| Temperate Asia | 0 | 7 |
| Tropical Asia | 3 | 0 |
| Tropical Australia | 1 | 0 |
| Temperate Australasia | 0 | 1 |
| Total | 20 | 27 |
Quantifying effect size
We measured the Spearman rank correlation (rS) of LMA with interspecific variation in shade tolerance in each study. Rank correlations enabled comparisons across studies that quantified the shade tolerance of species using methods that were diverse, and sometimes incommensurable. Only three of the chosen studies had quantified shade tolerance variation by measuring survival of the study species in low light. Rather more studies used shade-tolerance indices developed from the natural distributions of juveniles in relation to canopy openness or light availability (e.g. Davies, 1998; Poorter & Bongers, 2006). Niinemets & Valladares (2006) developed indices of shade tolerance for 806 woody species from North America, Europe and temperate Asia, obtained by cross-calibrating indices developed independently by various ecologists and foresters in these three regions. This meant that shade tolerance ratings (1.0–5.0) were available for almost all northern temperate species found in our literature search. No such data were available for many tropical studies: in these cases we used subjective shade tolerance ratings provided by the original authors, usually consisting of two or three categories or classes.
Height data were also analysed to verify the initial premise about the influence of seed size variation and study design on interspecific comparisons. Eight of the chosen studies reported mean height of each species in at least one light environment, enabling us to carry out meta-analyses of the rank correlations of the mean heights of species with shade tolerance variation in each combination. Similar relationships between seed size and shade tolerance have been reported in evergreen and deciduous angiosperm trees (Hammond & Brown, 1995; Hewitt, 1998), and height patterns in the data set did not appear to be influenced by leaf habit. We therefore combined studies of deciduous and evergreen species in our meta-analyses of the relationship of plant height with shade tolerance variation.
Meta-analysis of correlations
The methods described by Hedges & Olkin (1985, Chapter 11) were used to compare correlation coefficients. This involved first calculating the Fisher z-transformation of each correlation coefficient: if ri is the ith correlation coefficient, the Fisher z-transformation is
(Eqn 1)
The z-transformed values were then analysed using weighted least squares, where the weight is ni – 3 (the inverse of the error variance of zi) and ni is the sample size used in calculating ri. We tested for homogeneity of the K correlation coefficients using
(Eqn 2)
(zi−zi, the fitted value for zi from the model fitted by weighted least squares.) Under the null hypothesis that the model for data is correct, Q has a χ2 distribution with (K – p) degrees of freedom, where p is the number of parameters in the fitted model. Q is a simple generalization of the test statistic in equation 16 of Hedges & Olkin (1985, Chapter 11).
If all our models were to have significant Q values, this would suggest the presence of interstudy variability that is unaccounted for by factors included in the model. Hence, if required, we were prepared to respond to evidence of heterogeneity by including in analyses a random effects term for study (as in Hedges & Olkin, 1985, Section 11F).
The (transformed) correlation coefficients were modelled using a three-way factorial design. The factors were leaf habit (evergreen or deciduous), ontogeny (seedlings vs saplings) and light environment (low or high). The test statistic used to compare nested models was the change in Q, which has the appropriate χ2 distribution. This analysis is based on an analogue of Type II sums of squares: we test for the effect of a factor by considering the change in Q when it is removed, while retaining in the model all other factors of the same order (i.e. all main effects when testing for a main effect, and all main effects and two-way interactions when testing for a two-way interaction). We conducted multiple comparisons of average correlations using a sequential Bonferroni approach (Quinn & Keough, 2002), using Wald statistics constructed as in Hedges & Olkin (1985, Section 11E.2). To construct 95% confidence intervals for average correlation, we back-transformed the weighted average z values and their corresponding confidence limits.
Availability bias?
Biases against nonsignificant results could lead to published findings being unrepresentative of true effect sizes. In our case, publication bias of this sort seems unlikely, as many of the studies included did not explicitly address hypotheses about relationships of LMA with shade tolerance. Nevertheless, we checked for evidence of publication bias in the some of the better-represented combinations, using funnel plots to examine relationships of sample size with size and direction of effects (Light & Pillemer, 1984). If effect size does not influence likelihood of publication, then the plot of rS vs sample size should be roughly funnel-shaped: variation in effect size should be widest in studies with small sample sizes, with increasing numbers of species bringing convergence on the true effect. As a result, rS should be independent of sample size. In contrast, if the likelihood of publication depends on the statistical significance (P-values) of results, then studies of few species reporting small effect sizes will be underrepresented, because such studies are unlikely to attain statistical significance.
We also investigated another possible source of bias, suggested by a referee. Researchers choosing just a few species for a comparative study might focus on species of strongly contrasted shade tolerance, perhaps yielding inflated trait correlations that are not representative of wider patterns. Measuring the correlation of rS with sample size also served as a test for this type of bias.
Independent contrasts
Taking phylogenetic relatedness into account in comparative studies can sometimes aid understanding of trait interrelationships, and of the adaptive significance of traits (Harvey & Pagel, 1991). Cross-species correlations are usually interpreted in terms of biophysical trade-offs and/or environmental selection pressures (e.g. Shipley et al., 2006). However, if hard constraints on trait evolution are associated with one or a few old divergences, these can have an inordinate influence on ecological and/or physiological variation in the data set. Such phylogenetic idiosyncrasies can either (1) obscure trait relationships associated with more recent divergences (Irschick et al., 1996), or (2) inflate the strength of existing adaptive correlations (Ackerly & Reich, 1999), or (3) produce spurious correlations that do not reflect inevitable linkages resulting from natural selection or trade-offs (Ackerly & Reich, 1999). Comparative biologists sometimes attempt to assess the influence of phylogeny on their study outcome using phylogenetic comparative methods such as analysis of independent contrasts (Felsenstein, 1985).
The compare software (Martins, 2004) was used to analyse independent contrasts of LMA and shade tolerance in the studies found in our literature search. Regression lines of independent contrasts were forced through the origin (Grafen, 1992). We analysed independent contrasts of the same ranked values of LMA and shade tolerance that were used for cross-species analyses, facilitating direct comparison of mean correlations obtained from our two different approaches. As in the case of the cross-species analyses, the use of ranks enabled comparisons across studies that used diverse methods to quantify the shade tolerance of species.
Fully resolved phylogenetic trees were obtained for most of the studies found in our literature search. Trees were obtained to family level from Soltis et al. (2000), and more detailed information was sought for families represented by multiple taxa (Table 2). All branches were assumed to be of equal length. Three studies were omitted completely because of largely or completely unresolved phylogenies, and a few species had to be omitted from two other studies because of incomplete resolution (Kitajima, 1994), or incomplete identification (Poorter & Bongers, 2006). As analysis of independent contrasts of n species provides n – 1 contrasts, and as studies were weighted by n – 3 in our meta-analysis, only studies with at least five species could be used for independent contrasts. These constraints together reduced the number of eligible studies from 45 to 30, and the total number of correlations measured from 70 to 53.
| Family/subfamily | References |
|---|---|
| Betulaceae | Chen et al. (1999); Jarvinen et al. (2004) |
| Caesalpinioideae | Bruneau et al. (2001) |
| Cunoniaceae | Bradford & Barnes (2001) |
| Dipterocarpaceae | Kamiya et al. (2005); Indrioko et al. (2006) |
| Euphorbiaceae | Blattner et al. (2001) |
| Fagaceae | Manos et al. (2001) |
| Malvaceae | Alverson et al. (1999); Bayer et al. (1999); Baum et al. (2004) |
| Meliaceae | Muellner et al. (2003) |
| Moraceae | Datwyler & Weiblen (2004) |
| Myrtaceae | McVaugh (1968) |
| Phyllanthaceae | Tokuoka & Tobe (2006) |
| Proteaceae | Johnson & Briggs (1975) |
| Salicaceae | Hamzeh & Dayanandan (2004) |
| Sapindaceae | Ackerly & Donoghue (1998) |
Meta-analysis of independent contrasts was carried out using the same approach described above for cross-species analyses.
Results
Cross-species analyses
Most variation in our data set was explained by leaf habit, ontogeny and their interaction (Table 3a). There was no significant effect of light environment, or of any interactions involving light. The nonsignificant error term means that there was no evidence of any unexplained variation in correlation coefficients across studies, that is, variation beyond that attributable to the distinctions of leaf habit, ontogeny and light environment that we established a priori.
| Effect | Weighted SS | d.f. | P-value |
|---|---|---|---|
| (a) Cross-species analyses | |||
| Leaf habit | 55.390 | 1 | < 0.001 |
| Ontogeny | 7.560 | 1 | 0.006 |
| Light | 0.130 | 1 | 0.721 |
| Leaf habit ? ontogeny | 12.940 | 1 | < 0.001 |
| Leaf habit ? light | 0.490 | 1 | 0.482 |
| Ontogeny ? light | 1.570 | 1 | 0.210 |
| Leaf habit ? ontogeny ? light | 0.350 | 1 | 0.551 |
| Error | 35.530 | 62 | 0.997 |
| (b) Correlated divergences | |||
| Leaf habit | 25.228 | 1 | < 0.001 |
| Ontogeny | 2.288 | 1 | 0.130 |
| Light | 0.003 | 1 | 0.959 |
| Leaf habit ? ontogeny | 3.192 | 1 | 0.074 |
| Leaf habit ? light | 0.041 | 1 | 0.840 |
| Ontogeny ? light | 1.709 | 1 | 0.191 |
| Leaf habit ? ontogeny ? light | 0.564 | 1 | 0.453 |
| Error | 39.983 | 45 | 0.684 |
- d.f., degrees of freedom; SS, sum of squares.
The LMA of young seedlings grown was positively correlated with shade tolerance of both deciduous and evergreen species (Fig. 1a). Mean correlations ranged from 0.19 to 0.59 and, except for studies of deciduous species in high light, their confidence intervals did not overlap zero. These positive correlations show that young seedlings of shade-tolerant species have higher LMA than those of pioneers of the same age, irrespective of leaf habit. This pattern was accompanied by a strong positive correlation of shade tolerance with seedling height in low-light environments (Table 4), confirming the existence of initial average size differences between shade-tolerant species and pioneers.
Mean correlations of leaf mass per area (LMA) with species’ shade tolerance, from meta-analyses of (a) cross-species analyses, and (b) independent contrasts. Error bars show 95% confidence intervals, back-transformed after z-transformation. Letters on the left show results of multiple comparisons: mean correlations do not differ significantly between combinations sharing the same letter. Numbers of studies included in each analysis are shown in parentheses on the right.
| Combination | Mean | 95% confidence interval | n |
|---|---|---|---|
| Seedlings, low lightb | 0.66 | 0.32, 0.85 | 3 |
| Seedlings, high lightab | 0.29 | –0.24, 0.69 | 2 |
| Saplings, low lightab | 0.24 | –0.12, 0.55 | 5 |
| Saplings, high lighta | –0.17 | –0.59, 0.32 | 3 |
- Data from evergreen and deciduous taxa were combined. Results of multiple comparisons across combinations are given as indices: combinations sharing the same letter do not differ significantly (P > 0.05).
Relationships of LMA with shade tolerance of saplings depended on leaf habit, in both light environments (Fig. 1a), in contrast to the more homogeneous results obtained for age-specific studies of seedlings. LMA was strongly positively correlated with shade tolerance of evergreen saplings, with confidence intervals not including zero. By contrast, LMA of deciduous saplings was negatively correlated with species shade tolerance: although these relationships were not as strong as the positive correlations shown by evergreens, confidence intervals again did not include zero. The available height data suggested no relationship between sapling height and species shade tolerance in either light environment (Table 4), showing that the sapling data had been successfully restricted to studies comparing species at a similar size.
Testing for biases
There was no clear evidence that small sample sizes gave biased (inflated) correlations. A significant relationship of rS with sample size was not found for any combination of leaf habit, ontogenetic stage and light environment (P > 0.15 in all cases). In five out of eight combinations the sign of the correlation of rS with sample size was actually opposed to the trend expected if small sample sizes did give inflated estimates of effect size.
The weak relationships of rS with sample size also suggest little availability bias. There were ≤ 10 studies in most combinations of leaf habit, ontogenetic stage, and light environment, and funnel plots were constructed only for the two best-represented combinations: deciduous saplings in low light, and deciduous seedlings in high light (Fig. 2). Although only the latter plot closely approximated a funnel shape, in neither case was there a significant correlation between effect and sample size.
Relationship between number of species and effect size in studies of (a) deciduous saplings grown in low light and (b) deciduous seedlings grown in high light.
Independent contrasts
Meta-analysis of independent contrasts yielded broadly similar patterns to those found in the cross-species meta- analyses (Fig. 1b). Our factorial model was less successful in explaining overall variation in the results of independent contrasts: although the effect of leaf habit was again dominant, the effect of ontogeny was nonsignificant, and the interaction of leaf habit ? ontogeny was reduced to marginal significance (Table 3b). Although some correlations were weaker than those yielded by meta-analysis of cross-species analyses, in no case was there a change in the direction (sign) of the relationship (Fig. 1b).
Discussion
This global study shows that, when juveniles are compared at a common size, relationships of LMA with shade tolerance differ fundamentally between evergreen and deciduous taxa (Fig. 1). As leaf lifespan is correlated with shade tolerance in both groups (albeit more weakly in deciduous taxa), this pattern seems at variance with the idea of universal scaling of leaf functional traits (Reich et al., 1997; Wright et al., 2004). In attempting to explain this discrepancy, it is probably helpful to think about the respective contributions to LMA of (1) photosynthetic tissue and (2) structural reinforcement, and how these might be involved in adaptation to light availability. As we shall see in the following paragraphs, the strong effect of leaf habit on the relationship with shade tolerance probably reflects the differential contributions of components (1) and (2) to interspecific variation in LMA in deciduous and evergreen taxa.
The positive relationship of LMA with shade tolerance in evergreens is probably dominated by variation in the degree of structural reinforcement (Westoby et al., 2002; Reich et al., 2003). Despite some overlap, leaf lifespan is usually greater in evergreens than in deciduous species (Reich et al., in press), and this is presumably made possible by a greater average investment in structural reinforcement in the former. More important, however, is the range of variation in evergreen forests: whereas the most shade-tolerant species can retain leaves for c. 4 yr on well-lit branches, and somewhat longer in the shade (Lusk, 2002; Reich et al., 2004), short-lived pioneers in evergreen forests often have leaf lifespans of < 1 yr (Williams et al., 1989; Lusk, 2002). Large interspecific differences in structural investment, associated with leaf longevity variation sometimes spanning close to an order of magnitude, presumably cause the strong positive relationship of LMA with shade tolerance variation in evergreen forests.
The negative correlation of LMA with shade tolerance of deciduous saplings (Fig. 1), paralleling plastic responses to shade (Lambers et al., 1994), is probably driven mainly by variation in palisade mesophyll thickness. Although leaf lifespan is much more constrained in deciduous trees than in evergreens, it can be 50–100% longer in shade-tolerant species than in their light-demanding associates (Koike, 1988; Matsuki & Koike, 2006), suggesting that the former are likely to have slightly greater investment in cell walls, cuticle, and other structural features (Koike, 1988). However, the negative correlation of LMA with shade tolerance suggests that any differences in structural reinforcement are outweighed by the countervailing effects of thicker palisade mesophyll in the light-demanding taxa, giving an extensive mesophyll surface area per unit leaf area, and a large photosynthetic capacity (Koike, 1988; Hanba et al., 2002). The generally thinner mesophyll of shade-tolerant species reflects the insufficiency of understorey light intensities to enable leaves to meet the energetic cost of maintaining large pools of photosynthetic enzymes, and of paying off the construction of a deep layer of photosynthetic tissue. Interspecific variation in leaf lifespan of deciduous species may reflect an effect of photosynthetic investment on leaf carbon balance in low light, as well as the effect of structural investment on leaf survival prospects.
Our other major finding is that the effects of initial size on seedling LMA completely mask the constitutive relationship of this trait with shade tolerance of deciduous species. At least in low light, the pattern found in age-specific comparisons of deciduous seedlings is diametrically opposed to that found in size-specific comparisons of saplings (Fig. 1). A combination of initial size differences and ontogenetic drift in LMA (Fig. 3b) probably explains why researchers focusing mainly on seedlings of deciduous trees (Walters & Reich, 1999) have arrived at very different findings from those working on saplings (e.g. Koike, 1988; Abrams & Kubiske, 1990). The importance of ontogenetic shifts in plant traits was recognized some time ago by ecologists studying adaptive plasticity in response to resource availability (McConnaughay & Coleman, 1999). However, the implications for comparative physiological ecology have perhaps not been so widely appreciated (Lusk, 2004; Niinemets, 2006). We note that the observed relationship of LMA with shade tolerance in common-age comparisons of deciduous species will depend on the duration of the growth period. Although comparisons at < 1 yr generally give a positive relationship (Fig. 1), a comparison of 7–10-yr-old open-grown plants of 11 species gave a negative correlation (Niinemets, 2006), presumably because by that age the faster growth of the light-demanding species had overturned the initial size hierarchy. Despite the nonsignificant effect of light environment in our analysis, this result reported by Niinemets (2006) shows that light availability will also influence the relationship of LMA with shade tolerance, if experiments run long enough. High-light environments will enable seedlings of light-demanding species to fully express their potential growth rate advantage over their shade-tolerant associates, rapidly reducing and eventually overturning the initial size differences (Table 2) and thus neutralizing the initial ontogenetic influence on age-specific comparisons of LMA (Fig. 1). In low light, however, where the growth rates of species are less differentiated (Walters & Reich, 1999), the legacy of the initial size hierarchy could persist for several years if plants survive (Poorter & Rose, 2005).
Likely influences of initial size differences and ontogeny on comparisons of leaf mass per area (LMA) between shade-tolerant (solid line) and light-demanding (broken line) species. Seedlings are usually compared at a common age (A1); saplings, at a common size (Sx). Patterns for (a) evergreens and (b) deciduous species are shown.
Ontogeny would also be expected to influence the relationship of LMA with shade tolerance of evergreens, albeit to a lesser extent (Fig. 3a). Whereas ontogenetic and constitutive variation have opposite influences on common-age comparisons of LMA in deciduous seedlings (Fig. 3b), these two sources of variation should reinforce each other in age-specific comparisons of evergreen seedlings (Fig. 3a). Although a positive correlation of LMA with shade tolerance of evergreens would thus always be expected, age-specific comparisons of seedlings in low light should give a stronger relationship than size-specific comparisons of saplings (Poorter, 2007). Despite this expectation, mean correlation coefficients differed minimally between seedling and sapling studies (Fig. 1). This could reflect a limitation of the use of rank correlations: if, despite decreasing differences in LMA, few species changed ranks between the seedling and sapling stages, there would be little effect on statistical outcomes. Alternatively, the representation of ontogenetic variation in LMA in Fig. 3 might not be accurate. This model assumes that the slope of the relationship of LMA with height is similar in light-demanding and shade-tolerant taxa, but if the slope were steeper in the latter, this would offset the waning of the influence of initial size differences. However, Poorter (2007) found that the relationship of LMA with shade tolerance of 58 neotropical evergreens was indeed strongest at the seedling stage, and weakest in adults – in agreement with the model presented in Fig. 3.
The overall similarity of results from meta-analyses of cross-species analyses and independent contrasts suggests that the former analyses were not strongly influenced by phylogeny. Meta-analysis of independent contrasts gave noticeably weaker mean correlations for both deciduous seedlings and deciduous saplings in high light (Fig. 1). These two changes could indicate that their equivalent mean correlations obtained from the meta-analysis of cross-species analyses were inflated to some degree by phylogenetic idiosyncrasies (e.g. Ackerly & Reich, 1999), although in these cases any such bias would not lead to substantially different conclusions. The factorial analysis had less explanatory power when applied to independent contrasts, with the effect of ontogeny losing significance (Table 3b). However, the similarity of the overall patterns yielded by the two approaches (Fig. 1) suggests that the loss of statistical significance can be attributed mainly to a decrease in statistical power as a result of the exclusion of 17 correlations from the meta-analysis of independent contrasts.
Our findings are supportive of proposals to complement research on LMA with measurements of other aspects of leaf structure (Niinemets, 1999; Roderick et al., 1999). Strong relationships of LMA with seedling growth rates (Poorter & Remkes, 1990), and the overall correlation of this readily measured parameter with other leaf functional traits in global data sets (Reich et al., 1997; Wright et al., 2004), have led to an emphasis on LMA as a core aspect of plant functional diversity (Westoby et al., 2002; Karst & Lechowicz, 2007). However, the failure to find a universal relationship of LMA with shade tolerance, plus the fact that other aspects of leaf structure are sometimes better correlated with plant strategies or positions on environmental gradients (Wilson et al., 1999; Roche et al., 2004), suggests that natural selection does not act on LMA per se. If our interpretation is correct, adaptation to shade in general would involve a relatively large investment in leaf structural reinforcement and a small investment in photosynthetic components: this gives variable correlations with LMA, depending on ontogenetic considerations and climatic constraints on leaf lifespan. Studies of nutrient and moisture gradients also provide evidence of differential selection on leaf structural and photosynthetic components, the optimal relative and absolute investments in these components apparently varying between gradients (Wright et al., 2002). Theoretical advantages of decomposing LMA into approximations of photosynthetic vs structural elements have been demonstrated (Roderick et al., 1999), and empirical studies have shown that these two leaf components respond differentially to environmental gradients (Cornelissen et al., 1996; Niinemets, 1999). Further use of such approaches is likely to advance our understanding of the physiological determinants of species sorting on gradients of light availability, and on other environmental gradients.
Acknowledgements
Thanks to Macquarie University for support through a New Staff Grant, to Garreth Kyle and Andrés Rolhauser for help with data collation, to Lourens Poorter for constructive criticism, and to Dan Falster, Yusuke Onoda, Lawren Sack, Fiona Scarff and Mark Westoby for helpful discussion. This work arose from the Working Group on ‘Leaves: size, shape, economics, palaeobiology and evolutionary radiations’, of the ARC-NZ Research Network for Vegetation Function.
