Species selection

Four congeneric/confamilial species pairs were selected consisting of one species specialized to the sandstone soil type and one to the alluvium. Four generalists and other common unpaired specialists were also selected to obtain data on a wider range of common species (16 in total). Sapling selection was restricted to healthy individuals with unbroken stems and stem height of 0.5–1.5 m. 15–60 individuals of each species were selected and uniquely tagged. Species information can be found in the online supplement (Table S1). The categorization of species as edaphic generalists or specialists was qualitatively based upon observation with numerous transects walked on each soil type searching for ‘wrongly placed’ individuals (specialists) or species that were fairly common on both sandstone and alluvial soil types (generalists) and corresponds with large plot data for the reserve [35].

Functional trait data collection

We measured all six traits included in the LES: mass-based photosynthetic and respiration rates (A_mass and R_mass, respectively), leaf mass per area (LMA), leaf N and P concentrations, and leaf lifespan (LL). For gas-exchange measurements, six individuals per species were sampled in high light locations (for generalists, this resulted in 12 sampled individuals; six per soil type). The range of light environments to which ‘high light’ corresponded was 6–13 mol m⁻² d⁻¹. Individual sapling light environment was measured using hemispherical photographs taken directly above the crown using a Nikon Coolpix 900 and FC-E8 fisheye converter (Nikon, Tokyo, Japan). Photographs were analyzed using the program Winscanopy 2001 (Regent Instruments Inc., Québec, Canada). Using a LI-6400 gas-exchange system (Licor, Inc., Lincoln, NE, USA), gas-exchange measurements were made on recent, fully expanded leaves. All measurements were made before noon with cuvette conditions maintained at 350 ppm CO₂, 60–80% RH and 25–30°C leaf temperature. Measurements were made in 2002 at the beginning of a wet spell. Gas-exchange leaves were harvested, measured for area, dried at 60°C and weighed. Both leaf area and weight excluded the petiole. For compound-leaved species, the measured leaflet was sampled. LMA was calculated and used to convert gas-exchange values to mass-based measures. Leaf tissue, excluding primary and secondary veins, was finely ground and samples wet digested using the sulfuric acid-hydrogen peroxide digest procedure. The digest was analyzed for total N using the phenol blue reaction employing autoanalysis [36] and P by the colorimetric molybdenum blue method [37]. To monitor leaf production, the three most recently expanded leaves on each branch were labeled with non-toxic permanent marker at the base near the petiole every two months from September 2001–March 2002 and again in September and November 2002. Aside from one light-demanding species (Homolanthus populneus), there was never an instance when at least one marked leaf could not be found on each branch (for H. populneus, LL was conservatively estimated at two months in subsequent analyses). At each census, leaves remaining from the previous census and new leaves produced were tallied. LL was estimated using a demographic approach [38]:Where L is leaf lifespan, b is leaf production between censuses, d is leaves dropped between censuses, N_t1 represents initial number of leaves and N_tα represents number of leaves at t₁ plus number of leaves produced between t₁ and t₂. This calculation provides an estimate of average leaf longevity based on cumulative leaf production and deaths.

Additional datasets

To test whether similar patterns exist in other tropical lowland forests, we searched the literature for datasets from other moist tropical forest sites for which comparable data on LES traits as well as quantitative measures of juvenile shade tolerance have been published. We found two sites: La Chonta, Bolivia [18], [39] and Parque Nacional San Lorenzo (hereafter, San Lorenzo), Panama [17]. For the La Chonta dataset, all leaf traits measurements were made on saplings (18) with the exception of foliar P, which was measured on leaves from adult trees (38). The shade tolerance metric used in the Bolivian studies was the Crown Exposure Index (18). For San Lorenzo, we only included the 14 tree species for which shade tolerance was quantified in our analysis. All leaf trait data were collected for adults. The shade tolerance index used in the Panamanian study consisted of factor scores for the first principal component calculated for height growth and proportional survival of first-year seedlings (17).

Statistical analysis

Analyses followed methods in Wright et al. [13]. Log-10 transformed species trait values were used to ensure normality and homogeneity of variance. A principal components analysis was run on the entire data set to assess whether the LES signal was evident. Generalist species are represented in all analyses twice to quantify shifts in leaf traits between sandstone and alluvial populations. Species' loading scores were then divided by habitat and a Kruskal-Wallis test was used to assess whether habitat type was a good predictor of species' placement along significant axes (loading data were not normally distributed and transformations failed). A regression analysis was run using the primary axis (the ‘LES’) as the predictor of species' whole-plant light compensation point (WPLCP) taken from Baltzer and Thomas [19]. WPLCP is a quantitative measure of shade tolerance in which the X-intercept of the light-growth relationship is estimated for each species and corresponds with a physiologically and ecologically relevant minimum light requirement [19]. To test whether similar patterns exist in other tropical lowland forests, we conducted corresponding analyses for the two additional moist tropical forest sites. Type II regression analysis was used to test whether the slopes differed from unity or the intercepts differed from zero for pairwise relationships of loading values among sites. Differences in these relationships would suggest changes in the relative importance of functional traits among sites. Only five of the six leaf traits were available for San Lorenzo (no R_mass); to ensure differences in trait loadings were not simply due to this discrepancy we conducted an additional PCA for the Sepilok and La Chonta datasets excluding the R_maas data. Results of the corresponding regression anayses can be found in Table S2. Analyses were conducted using R (v. 2.9.0, R Foundation for Statistical Computing).

Axis 1: the LES and shade tolerance

The primary axis, which corresponds with the LES of Wright et al. [13] proved to be a good predictor of a quantitative measure of shade tolerance, the whole-plant light compensation point [WPLCP; 19]. This finding is in keeping with previous studies examining LES-related leaf traits and shade tolerance in tropical trees [e.g.], [ 17], [18,20]. Corresponding analyses for La Chonta and San Lorenzo demonstrated that the first principal component was likewise identical to the LES. Correlations between pairs of axis 1 loadings were very high (Table 2). Furthermore, as was the case for the Sepilok dataset, the first component was a significant predictor of available measures of juvenile light requirements for each site. This provides strong evidence for a functional linkage between foliar trait variation comprising the LES and minimum light requirements in tropical forest trees, consistent with the theory that trees that evolved to regenerate in the light-limited environment of a forest understory employ a conservative strategy to ‘pay back’ leaf construction costs through slow turnover rates and long residence times of key nutrients such as nitrogen [13], [21].

At Sepilok, there are differences in the understory light environment associated with soil variation, with the sandstone understory having slightly greater light availability due to the lower basal area and canopy stature [19], [35]. Baltzer and Thomas [19] demonstrated differences between congeneric species pairs in WPLCP with sandstone species having slightly higher light requirements. It is therefore reasonable to hypothesize that there might be some separation of the two groups along the primary axis; however, in spite of the slight difference in WPLCP detected using the phylogenetic framework described above, there were no differences detected in traits associated with the LES (Fig. 1).

Axis 2: Edaphic specialization and phosphorous

We demonstrate a second significant axis to the LES onto which foliar P loaded most strongly, resulting in near-complete separation of species associated with two distinct edaphic habitats at Sepilok. Sandstone ridges show much lower P availability than the alluvial valleys at this site [32]. Similarly, Paoli [8] demonstrated that foliar P was the strongest predictor of habitat associations in eight closely related Shorea species an Indonesian lowland forest. The patterns observed in the present study may thus be broadly applicable to P-limited tropical forests that show strong resource gradients and clearly defined habitat associations.

Initially, we hypothesized that foliar P should play a prominent role in the LES due to P-limitation in tropical lowland forests [25], its critical role in a wide range of physiological functions in plants, and its contribution as a component of molecules such as nucleic acids, phospholipids, ATP and sugar phosphates [29]. If the contribution of foliar P to the LES was primarily through its linkage to photosynthesis, we would expect P to load much more strongly on the primary axis, which it does not; in fact it shows the weakest loading of all traits examined. Previous studies show that the structure of the LES is modified by P via changes in N-A_mass relationship, but that P does not contribute directly as a primary determinant of photosynthetic rates, even in P-limited tropical forests [16], [ but see 40].

Why then might P be loading orthogonally to the LES in our analysis? The separation of species associated with sandstone vs. alluvial soils corresponds with a large difference in P availability between the two soil types but total soil N likewise is lower on the sandstone soils [32].

We suggest that the strong loading of foliar P onto the second axis may be due to the availability of interchangeable forms of each nutrient and the relative energetic and carbon costs associated with uptake. There are more interconvertable forms of N than P available to plants. At Sepilok, nitrate availability is lower in sandstone soils while ammonium concentrations are higher [32]. Therefore it may be the case that N is not as strongly limiting as P in this forest, as suggested by high N∶P ratios (15∶1 and higher; data not shown). N uptake is thought to be closely linked to demand whereas P uptake frequently continues even when P is not limiting [41]–[43]. Therefore we expect foliar N to correspond closely with the LES; the proteins necessary to support high photosynthetic and respiration rates have high N demands [44]. With respect to P uptake, plants only have direct access to dissolved phosphate [43], which is replenished by slow diffusion. As the rooting zone becomes devoid of dissolved phosphate, root proliferation or enhanced reliance on mycorrhizal associations and root exudates become necessary to maintain P supply [43]. Root proliferation is considered a primary mechanism for increasing uptake on P-deficient soils whereas on richer soils, up-regulating physiological activity can increase uptake rates [45], [46]; however, root proliferation is costly both in terms of C and N. Recent studies show that in P-deficient soils, the costs of P uptake are very high and rather than maintaining high rates of P uptake, that increased efficiency in resource use is a better strategy [47], [48]. Improved P-use efficiency (i.e., higher cumulative carbon assimilation per unit P investment over the lifetime of the leaf) may be achieved by such means as longer P residence times and preferential allocation of P to photosynthetic tissues (e.g., 46). The clear separation of sandstone vs. alluvial specialists may thus correspond to inherent differences in P use between the two groups that are not directly linked with the LES, but further work is necessary to directly test for such differences.

Sepilok, and Bornean forests more generally, display marked patterns of habitat association driven by edaphic characteristics [30], [33]. A second axis of leaf trait variation might be expected to be strongly expressed under these conditions. Conversely, if species distributions at the local scale are not largely being driven by edaphic limitations, then the contribution of that second axis may disappear. By comparing our findings with LES analyses of La Chonta and San Lorenzo, we can examine these predictions. Loadings of traits onto the second axis were virtually identical for Sepliok and La Chonta. At San Lorenzo, in contrast, all available traits loaded similarly onto the primary axis and no signficant second axis was detected. This qualitatively supports our predictions: soils in Sepilok show the strongest habitat differences in P associated with distinct soils on sandstone ridges and alluvial valleys [32] with correspondingly strong patterns of habitat specificity [35]. La Chonta shows somewhat less pronounced edaphic variability, with anthropogenic dark earths enriched in P [49] that impact soil-vegetation relationships (Peña-Claros et al., submitted). In contrast, recent analyses for Barro Colorado Island, another Panamanian moist lowland forest proximate to San Lorenzo with substantial species overlap, show relatively weak effects of soil P levels on community niche structure [3].

Another interesting feature of the Sepilok analysis is the placement of generalist species along the second axis (Fig. 1). For both sandstone and alluvial habitats, the generalist species occupying those habitats cluster closely with specialists associated with each habitat. This is in keeping with the idea that generalist species are more plastic in their responses to soil resource availability [50] or that some degree of local adaptation has occurred. However, it could also simply be an indication that individuals will take up resources when available (e.g., high P values in generalists occurring on the alluvium) and that sandstone specialists may well be capable of increased P uptake were they to establish and survive on the P-rich alluvial soils. In other words, plasticity could be underlying observed patterns. However, evidence from a reciprocal transplant experiment involving two alluvial and two sandstone specialists at Sepilok suggests that adaptive responses to soil conditions can be pronounced. Although sandstone species showed small increases in foliar P when grown on alluvial soils, foliar P of alluvial specialists was almost double that of the sandstone specialists [32]. When the two groups were grown on sandstone, alluvial specialists maintained higher foliar P concentrations [32] suggesting an important role of differential P-use efficiencies. Whole-plant or cumulative P-use efficiencies may similarly differ among specialists and generalists such that sandstone specialists have a competitive advantage over generalists on P-deficient soils.

In conclusion, our data indicate that the first axis of the leaf economics spectrum (LES) describing patterns of trait covariation in leaf functional ecology corresponds to shade tolerance as quantified by the whole-plant light compensation point, and that this relationship is consistent across tropical lowland forests. However, foliar P is relatively uncorrelated with the first LES axis at sites where soil P strongly influences distribution of vegetation, and instead correlates with a “leaf edaphic habitat spectrum” axis that is quite strong in a tropical tree community characterized by high habitat differentiation among species. Further tests of this leaf edaphic habitat spectrum across substrate-based resource gradients are necessary to assess its generality in tropical forests and other plant communities.