Correlations amongst character traits

To explore the main trends of morphological differentiation across species, we ordinated the matrix of mean values of 14 traits for 40 species (App. 1). Because the dataset consisted of nominal, ordinal and continuous variables, we first computed a species similarity matrix using the Gower coefficient (Gower 1971), a general measure of proximity that can combine different types of variable. We then performed a principal coordinate analysis (PCoA) on this similarity matrix using GenStat 8. PCoAis a general form of principal component analysis (PCA) that allows the use of a wide array of distance measures (Legendre & Legendre 1998). The results of this analysis were (a) equivalent to those of a PCA restricted to continuous variables previously transformed to correct for skewed distributions and (b) identical to those of a non-metric multidimensional scaling analysis of the same similarity matrix. The significance of the extracted axes was tested using a broken-stick model distribution which predicts the partition of the total variance amongst the axes if it is made at random; observed ordination axes are considered interpretable when they explain more variation than expected from this model distribution (Legendre & Legendre 1998). To explore in detail the correlations between pairs of traits and between traits and ordination axes (not provided in PCoA) we used Kendall rank correlations because many of the variables did not fit normal distributions and some of them had natural outliers.

Although our data matrix and replicates were almost complete, in a few cases we could not collect any samples and had to estimate values in order to run the ordination. For six species we estimated seed mass from seed sizes reported by Diglio & Legname (1966) through a seed size - seed mass regression (R² = 0.69, n = 32, p < 0.001), for Xylosma longipetiolata we replaced leaf decomposition rate with the all-species mean and for Eugenia hyemalis we used visual estimations of maximum height. Any error introduced by these estimations was probably minimal compared with the benefit gained by including all species in the ordination analyses. Nominal multistate variables (dispersal and pollination syndromes) were split into dummy variables for analysis.

Measures of performance

To assess the functional significance of traits in terms of use of light or soil resources, we analysed the correlation of traits and groups of correlated traits with distribution-based measures of recruitment under shade and of recruitment over a gradient of soil fertility (App. 1). Both measures were based on a census of trees (> 5 cm DBH) in 46 plots (0.075 ha) (5 in old growth and 41 in secondary closed canopy forest) which were distributed across a wide range of soil types (pH 3.97 − 6.30) and land use histories (Easdale unpubl.) within Los Toldos valley. All plots were in closed canopy forest stands but they differed widely in structure and shade cast; maximum tree stature ranged from 9.5 m to 35.1 m amongst the stands and the density of stems > 5 cm DBH ranged from 747 stems/ha to 5480 stems/ha. Because both measures of performance were based on patterns of recruitment in the field and under competition with other species, they are estimates of the species' realized, rather than fundamental, responses.

Recruitment under shade was estimated from the capacity of each species to support a population of saplings/sprouts under closed canopy cover. It was computed as the quotient between each species' frequency of saplings/sprouts (> 0.3 m height and < 1 cm DBH) amongst the 46 plots and its frequency of trees > 5 cm DBH in the same plots. The measure assumes that if trees > 5 cm DBH of a certain species are present in a forest plot, the species is able to reach the site, grow and possibly produce progeny in that site. Therefore, the more shade tolerant the species the more likely that its saplings/sprouts would be found under the canopy of that stand. We report results for all 40 species but analyses in which we discarded the six resprouting species that could be falsely recorded as functionally shade tolerant (if ‘saplings’ recorded were actually sprouts that were physiologically sustained by their ‘parent’ trees) exhibited similar results in all cases except for a marked increase in the statistical significance of the relationship with pollination syndromes.

Recruitment was assessed over a gradient of soil fertility with the purpose of estimating the capacity of each species to establish in infertile soils. For this, we computed weighted averages (ter Braak & Looman 1986) from the depth of the upper organic soil layer (ca. A horizon), standard chemical analysis and species' abundance for each of the 46 plots. The soil analyses were done on upper soil (5 cm to 15 cm depth) composite samples (three sub-samples per plot) and included pH (potentiometric, 1:2.5 (v/v) soil to water suspension), total N (micro-Kjeldahl), available P (modified Bray-Kurtz), total C (loss on ignition), available K (flame photometry) and available Ca and Mg (EDTA complexometric titration). Exchangeable cations were extracted with 1 N ammonium acetate at pH 7. For each of the 40 species, mean values of each of the nine soil variables were calculated over the plots in which the species occurs, and these were weighted by the species' abundance on those same sites. Two species that were surveyed in only one site were excluded from this analysis. To give an approximation of the conditions where each species established (because current soil conditions in old growth forest could differ substantially from those prevalent at the time of establishment of old trees) we only used counts of trees in old growth forest plots that were smaller or equal in DBH to the largest trees found in secondary forest for that species.

We thus created a matrix of weighted averages for each of nine soil variables and each of 40 species. We subj ected the matrix to a PCA to summarize these results and took the loadings on the first principal component of the ordination as the measure of recruitment over a soil fertility gradient. This first principal component explained 36.7% of the total variation and summarized species' differential affinities (weighted averages) for pH, K, Mg, depth of organic soil layer and available P (factor loadings 0.83, 0.74, −0.65, 0.63 and 0.61 respectively). Ca, total N and total C had loadings with absolute values lower than 0.41 on the first principal component, indicating that species' affinities for these variables were largely independent from their affinities for the other variables and that they had no weight in this index.

Relationships amongst vegetative traits

The broad-leaved tree species analysed in this study can be discriminated in their morphology according to at least three independent sets of traits. The first axis exhibits close and consistent associations amongst leaf traits: deciduous species tend to have leaves with high specific area, low tensile strength, low dry matter content, large laminas and fast decomposition, whereas opposite attributes tend to be associated with evergreen species. If we consider leaf phenology as a coarse indicator of leaf longevity, and leaf decomposition rate as an indicator of high leaf nutrient content and/or low leaf defence, then the set of leaf attributes associated with deciduous species indicate increased capacity for resource acquisition and those associated with evergreen species indicate resource conservation. In consequence, the observed associations are consistent with the major axis of plant evolutionary differentiation which reflects a fundamental trade-off between rapid acquisition of resources and conservation of resources.

The close association amongst the measured leaf traits reflects an expected integration of form and function within the leaf. However, the correlation of leaf morphological traits with wood density and bark thickness, which apriori had aless obvious linkage, suggests afunctional integration within the whole plant. Indications of a link between leaf and stem traits were originally derived from rather indirect and qualitative observations that fast growing pioneer tree species had high photosynthetic rates and low wood density (Bazzaz & Pickett 1980; Swaine & Whitmore 1988) but there is now growing empirical evidence of an integration between leaf and stem ecophysiology. An association between hydraulic conductivity and photosynthetic capacity has recently been found for 16 angiosperm tree species in tropical New Caledonia and sub-alpine Tasmania (Brodribb & Feild 2000) as well as for 12 deciduous and evergreen tree species in tropical dry forests of Costa Rica (Brodribb et al. 2002). Additionally, Santiago et al. (2004) found that branch wood density is negatively correlated with hydraulic conductivity, and with net CO₂ assimilation rate per unit leaf area and per unit leaf mass, for 20 tree species of wet seasonal tropical forests in Panama. Parallel support can be drawn from the positive association between wood density and cavitation resistance (e.g. Hacke et al. 2001) and the correlation found between phenology and cavitation resistance. As initially hypothesized by Orians & Solbrig (1977), Sobrado (1993) found that drought deciduous species from the seasonal tropics had higher hydraulic conductivity in the short term at the risk of embolisms in the xylem, while evergreen species had lower water transport capacity but maintained relatively constant hydraulic conductivity during the wet and dry seasons. This body of evidence suggests that high hydraulic conductivity, imparted by low density woods, is necessary to sustain leaves with high assimilation potential while cavitation resistance, a property of dense woods, is necessary to support water transport in the dry season in evergreen species. Although we did not measure wood densities of the functional xylem (which are relevant to water transport) and wood densities are known to vary from pith to bark, our measures can be assumed to be proportional to those of the functional xylem because we sampled branches (presumably with a low fraction of heart-wood) of similar sizes and because most species (16 out of 20 in a study of tropical woods, Wiemann & Williamson (1989)) exhibit positive gradients in stem wood density from pith to bark. The high wood density of evergreen species is also consistent with high resource conservation since it increases mechanical resistance to decay, wind and fallen debris and is likely to contribute to longer individual lifespans (Loehle 1988). A thicker bark can result from faster growth and increased accumulation of necromass on the outer side of the bark and/or from thicker phloem on its inner side which may provide an increased transport rate of soluble carbohydrates andpeptides/amino acids within the plant. Therefore, our evidence of the association of thick bark with lower wood density, as well as with traits indicative of productive leaves, is consistent with the more general interspecific trade-off between high resource acquisition and resource conservation.

Not all stem traits are associated with leaf traits however; spines and maximum tree height are grouped by the second independent axis rather than the first. The observed lack of association between spinescence and leaf traits does not support the hypothesis of Coley (1983) and Coley et al. (1985) that plants which grow in resource poor sites invest more heavily in immobile defences but seems to follow Grubb's (1992) observation that both fast and slow growing species (i.e. the extreme syndromes represented by the first morphological axis) can commonly be found to bear spines.

In Malaysian tropical rain forests Thomas (1996) and Thomas and Bazzaz (1999) showed an association between leaf traits and maximum tree height, a probable result of evolutionary responses in shade tolerant species to the selective forces of vertical light gradients (analogous to the differentiation between pioneer and shade tolerant tree species in their response to horizontal gradients of light). Saplings of small stature species growing under similar light conditions were found to have an inherently lower area-based photosynthetic capacity (A_max) than species of the same genus that achieved greater heights (Thomas & Bazzaz 1999). The associations seemed more evident for physiological than for morphological traits but amongst shade tolerant trees they also detected inverse associations between the SLA of adult tree leaves and maximum height (Thomas & Bazzaz 1999) and between wood density and maximum height (Thomas 1996). Interestingly, the links mediated by maximum height are opposite to those exhibited by the fundamental trade-off of plant differentiation described above, where high A_max and low wood densities are associated with high SLA rather than with low SLA. When all species in our dataset were compared, there was only a positive association between maximum height and leaf dry matter content but when the comparison was restricted to shade tolerant species, taller species showed a significant tendency to lack spines and a marginal tendency to have lower SLA, as previously shown by Thomas & Bazzaz (1999).

Reproductive traits

Our results did not show strong associations amongst the reproductive traits measured: seed mass, pollination and seed dispersal syndromes and clonal growth. They also indicate a lack of association between (1) reproductive traits and (2) leaf traits, wood density and bark thickness. This concurs with predictions by Loehle (2000) and results from several studies carried out with herbaceous or shrubby floras that found only weak or no evidence of coupling between ‘regeneration’ and vegetative traits (Montalvo et al. 1991; Leishman & Westoby 1992; McIntyre et al. 1995; Grime et al. 1997; Díaz & Cabido 1997). The few exceptions to this are indications of associations between maximum height and some reproductive traits in forest communities. Momose et al. (1998) found that large trees in Malaysian tropical rain forest can produce large floral displays that attract more generalist pollinators and increase pollination success while small understorey plants have limited displays and therefore ‘employ’ specialist pollinators. This is in agreement with observations of Bawa et al. (1985) at La Selva, Costa Rica, that canopy trees were pollinated by bees and small diverse insects, while hummingbird, sphingid moth and beetle pollination were mainly restricted to sub-canopy trees. This non-random distribution of plant-pollinator interactions is expected on the basis of vertical stratification of animal communities (Smith 1973). However, in the present study we found no significant evidence of an association between pollination syndrome and species' maximum height. A comparable relationship that has been detected by most surveys of seed dispersal type among tropical tree species (as reviewed by Turner (2001)) is the preponderance of anemochory amongst taller trees, and our results do support this association.

Tolerance of shade versus tolerance of low soil fertility

Species' realized potential to recruit under conditions of low soil fertility was orthogonal to their potential to recruit under shade. This is in agreement with experiments on temperate woody species (Latham 1992; Grubb et al. 1996) that indicate an important range of niche differentiation among species which permits each species to be favoured in a different combination of light and soil nutrient conditions in natural forests. Under apatchy distribution of resources, this would provide a basis for species' coexistence. This contrasts, however, with evidence from montane tropical forests in Jamaica, where the results of Goodland et al. (1998) and McDonald et al. (2003) indicate that species which had their highest tree density in the least fertile soils (the Morridge forest type) had a significantly greater tendency to be light demanding than those species with highest tree density in the more fertile Mull ridge and Col forest types. The results from Jamaica could be attributed to soil infertility being correlated with a more open canopy (even in mature forest) in these conditions of very low soil pH (Bellingham et al. 1996).

Functional significance of leaf traits, wood density and bark thickness

The finding that species with characteristics indicative of high resource conservation (evergreen leaves, low SLA, high leaf tensile strength, low leaf decomposition rate and dense wood), as summarized by the first PCoA axis, are better able to recruit where light or soil resources are in short supply, and that species with opposite traits recruit less commonly under such conditions, is consistent with our functional interpretation of this first PCoA axis. We conclude that it indicates an evolutionary trade-off between high resource acquisition and resource conservation, and that species that recruit under shade and in infertile soils tend to have similar morphology. However, there is still considerable scatter in the relationships between species' traits and their capacity to recruit where there is a shortage of these two resources (Fig. 2b, c), and it is precisely this dispersion that accommodates the observed independence of species' recruitment in shade and over the soil fertility gradient. The first PCoA axis scores are a coarse simplification of species morphology and the results indicate that, although species which score close to one extreme of this axis tend to be tolerant of shortage of light and/or soil resources, they are not necessarily tolerant of shortage of both resources simultaneously. Similar interpretations are applicable to species at the other extreme of the axis.

The morphological variables condensed by the first PCoA axis also show some noise in their pairwise associations and it is this ‘play’, combined with the independence of the response variables, that results in different sets of individual traits being associated with the species' recruitment under shade from those associated with the potential to colonize infertile soils, when the traits are analysed individually. Therefore, our results agree, to some extent, with both sides of the debate on whether the same suites of traits confer tolerance of shortages of different resources (Grime 1977; Grime et al. 1989; Chapin et al. 1993) or whether different suites of traits are advantageous according to the most limiting resource under consideration (Grubb 1985, 1998;Tilman 1987) and show that results and interpretations can vary considerably depending on the resolution of the analysis or the degree of generalization allowed.

The connections observed between species' above-ground traits and their recruitment under shade or in infertile soils are consistent with previous studies on broad-leaved woody species. Our finding that evergreen species are common colonizers of infertile soils and shaded conditions match the recurrent observations that evergreen woody plants are associated with nutrient poor soils (Goldberg 1982; Grubb 1985; Givnish 1987; Aerts 1995), and that shade tolerant tree species have a longer leaf lifespan than sympatric intolerants in tropical moist (Coley 1988) and rain forests (Williams et al. 1989; Reich et al. 2004). Other leaf traits have also been found in association with shade tolerance. In a study of six to seven sympatric Piper shrubs, the more shade tolerant species had lower leaf N concentration, SLA and A_max than the shade intolerant species (Chazdon & Field 1987; Williams et al. 1989), which indicates that related taxa varying in shade tolerance are arranged along the trade-off surface of leaf resource conservation and high resource exploitation (however, it should again be noted that there is usually a wide variation in these associations for individual traits, e.g. some tropical species that are shade tolerant have high SLA, Grubb (1998)). Our finding that wood density also forms part of this trade-off is consistent with the positive association between wood density and shade tolerance initially described by Wells (1976), and later confirmed quantitatively by Augsburger (1984). Large seeds have often been associated with regeneration under low resource conditions, but the many field studies that have examined this relationship seem to find little empirical support (e.g. Westoby et al. 1992). We found weak evidence of a positive association between seed size and recruitment under shade and no evidence of an association between seed size and recruitment on infertile soils.

Other axes of morphological variation and their meaning

The tree species of this montane forest can be differentiated according to no less than three main axes based on shoot form. However, only the first axis has a consistent internal structure (i.e. it links variables that are significantly correlated amongst each other through consistent signs), whereas the remaining axes group variables that are not significantly related to each other and their grouping by the ordination most likely results from articulated signs (trends) in the relationships among them. This implies that there is even more morphological diversification than that condensed by the three axes and suggests that there is much room for functional differentiation across species if different axes or traits do really imply different ecological functions (cf. Loehle 2000). For instance, the second axis of morphological differentiation, which integrates maximum height, dispersal mode and spinescence, is also markedly associated with recruitment under shade and, at least partly, seems to represent the differential responses to vertical gradients of light availability amongst shade tolerant species, as judged from the negative association between maximum height and recruitment in shade observed in this group of species. Reproductive traits, which are mainly integrated in the third axis of morphological variation, could also be associated with response to disturbance as suggested by several other studies (e.g. Noble & Slatyer 1980).

We find some resemblance between these dimensions and observations from studies encompassing different plant life forms in that SLA (a surrogate for our first axis), height at maturity (second axis) and seed mass (third axis) vary widely amongst species at any given level of the other two (Westoby 1998). In several temperate floras, 70%–83% of the variation in log seed mass is independent of variation in log height (Leishman et al. 1995); similarly SLA varies widely amongst species and is only weakly correlated with the other two variables (unpubl. in Westoby 1998). The ecological meaning of these traits and their independence across species led Westoby (1998) to propose a leaf-height-seed plant ecology ‘strategy’ scheme (based on the CSR strategies of Grime (1977) and Grime et al. (1989)) as a framework for clas sifying plant' strategies' in comparative research. Our results also support Loehle's (2000) suggestion that vegetative reproduction is largely independent of shade tolerance and maximum height, which led him to propose it as another main dimension of tree ‘strategy space’.

Limitations of the study and future challenges

An association between species' traits and realized distribution in the field does not necessarily mean that those traits are functionally responsible for the observed distribution. Although the recent literature in plant ecology, physiology and evolutionary ecology is providing possible explanations for the causes and consequences of some trait associations, many of the mechanisms behind theses links remain unclear. In this respect, it is still uncertain to what degree observed interspecific trait correlations result from cost-benefit compromises or from other processes such as simple ‘trait co-ordination’ or the stochastic effect of genetic drift to which the evolutionary process is exposed (Reich et al. 2003). ‘Trait co-ordination’ could arise because particular trait values function more successfully in combination (driven by reciprocal relationships or due to common causation from a third variable) rather than because some traits drive others mechanically or physiologically (Shipley & Lechowicz 2000; Reich et al. 2003).

The functional value of other traits should also be tested. For instance, root properties (especially of mycorrhiza) are expected to be very important for tolerance of nutrient shortage. It would also be interesting to examine whether our findings on trait associations are similar in other habitats and woody floras. Borchert(1994) showed that tropical dry forests in Costa Rica contain both evergreen trees with light wood (that ‘avoid’ drought by storing water in their stems) and deciduous species with dense wood. This suggests that the associations we found between wood density and leaf traits may not hold in drier forests.

Our measures of species' regeneration in natural field conditions provide just one coarse estimator of their response to light and to soil fertility. Our measure of recruitment under shade gives an approximation of species' capacity to tolerate shade but does not consider (a) their potential for net carbon fixation and growth under shade or (b) ontogenetic changes in their light response such as those known for non-pioneer light demanding trees. There might also be sources of error in the computation of recruitment over the soil fertility gradient: (1) soil analyses were done on upper layer soil samples (5–15 cm depth) and although this seems appropriate for assessing tree establishment, it may not be representative of the soil volume that large trees exploit; (2) soil samples were taken from established forest stands between 12 and 80 years old, and although soil properties change at slower rates than vegetation, physicochemical properties may be different from those prevalent when the trees established; (3) species recorded in only a few sites are vulnerable to biased estimations of their tolerance ranges and (4) the effect of differential dispersal was controlled for in the computation of recruitment under shade, but not for recruitment over the soil fertility gradient where performance is directly assessed on the basis of tree distribution patterns. Field experiments that evaluate species' growth/survival under conditions of controlled resource availability with competition should be conducted to test our findings. Finally, phylogenetic linkages amongst species could bring a problem of pseudo-replication, especially for traits with low levels of convergence (Ackerly 1999).

Irrespective of its limitations, this study does provide solid evidence that wood density and bark thickness integrate the ‘construction costs’ axis commonly reported for broad-leaved plants. This is also the first quantitative attempt to test the correlations (and functional value) of various morphological characteristics with species' realized tolerances of (a) shade and (b) nutrient shortage. The results indicate that the functional value of morphological traits differs with the resolution of analysis: when traits are considered as part of sets of correlated traits or ‘strategies’, species that recruit under shade and on infertile soils tend to have similar morphology but when traits are considered individually they show different correlations with each ecological response.