Highly local environmental variability promotes intrapopulation divergence of quantitative traits: an example from tropical rain forest trees

Abstract

Background and Aims

In habitat mosaics, plant populations face environmental heterogeneity over short geographical distances. Such steep environmental gradients can induce ecological divergence. Lowland rainforests of the Guiana Shield are characterized by sharp, short-distance environmental variations related to topography and soil characteristics (from waterlogged bottomlands on hydromorphic soils to well-drained terra firme on ferralitic soils). Continuous plant populations distributed along such gradients are an interesting system to study intrapopulation divergence at highly local scales. This study tested (1) whether conspecific populations growing in different habitats diverge at functional traits, and (2) whether they diverge in the same way as congeneric species having different habitat preferences.

Methods

Phenotypic differentiation was studied within continuous populations occupying different habitats for two congeneric, sympatric, and ecologically divergent tree species (Eperua falcata and E. grandiflora, Fabaceae). Over 3000 seeds collected from three habitats were germinated and grown in a common garden experiment, and 23 morphological, biomass, resource allocation and physiological traits were measured.

Key Results

In both species, seedling populations native of different habitats displayed phenotypic divergence for several traits (including seedling growth, biomass allocation, leaf chemistry, photosynthesis and carbon isotope composition). This may occur through heritable genetic variation or other maternally inherited effects. For a sub-set of traits, the intraspecific divergence associated with environmental variation coincided with interspecific divergence.

Conclusions

The results indicate that mother trees from different habitats transmit divergent trait values to their progeny, and suggest that local environmental variation selects for different trait optima even at a very local spatial scale. Traits for which differentiation within species follows the same pattern as differentiation between species indicate that the same ecological processes underlie intra- and interspecific variation.

INTRODUCTION

Environmental variation occurring at the local scale creates complex habitat patchiness which has been found to contribute to shaping the great diversity observed in tropical rainforests (Ricklefs, 1977; Wright, 2002; Vincent et al., 2011). A common explanation for these diversity patterns is the divergence of preferences for edaphic conditions among tree species, as repeatedly shown throughout the Neotropics (e.g. ter Steege et al., 1993; Sabatier et al., 1997; Clark et al., 1999; Valencia et al., 2004; Baraloto et al., 2007; John et al., 2007; Kanagaraj et al., 2011). Community-level differences in functional traits have been found to underlie such differences (Kraft et al., 2008): for instance, Lopez et al. (2003) and Engelbrecht et al. (2007) showed that divergence in species distribution between hilltops and bottomlands is determined by variations in tolerance to drought and waterlogging.

It has been shown that bottomland, slope and hilltop habitats actually differ in many ways that may explain their impact on forest community composition. Generally speaking, water availability in lowland tropical forests is strongly associated with topography and soil characteristics (Sabatier et al., 1997). Large variations occur in soil drainage and moisture between hilltops, slopes and bottomlands (ter Steege et al., 1993; Clark et al., 1999; Webb and Peart, 2000). Bottomland soils are subject to frequent periods of flooding and undergo cyclical changes in O₂ availability that strongly affect the metabolism of root tissues and thus tree establishment and growth (Ponnamperuma, 1972; Kozlowski, 1997; Perata et al., 2011). In contrast, thin soils on slopes undergo lateral drainage, which increases their susceptibility to water shortage during dry periods (Sabatier et al., 1997). Finally, hilltops are usually characterized by deep soils and display deep vertical drainage, with strong seasonal variations in soil water availability (Sabatier et al., 1997). Beside differences in water availability constraints, these habitats also differ in nutrient content, with lower nitrogen and higher phosphorus content in bottomlands than on plateaus (Luizao et al., 2004; Ferry et al., 2010). Moreover, soil respiration decreases in bottomlands as root biomass and soil carbon content decreases (Epron et al., 2006). These variations in soil characteristics have an additional impact on forest dynamics, with slopes and bottomlands exhibiting more frequent light gaps than hilltops, and therefore higher irradiance reaching the understorey (Ferry et al., 2010).

The widespread links between gradients of soil properties and species-specific habitat preferences suggest that ecological specialization has recurrently arisen through evolutionary processes such as adaptation and species divergence (Endler, 1977; Schluter, 2001; Rundle and Nosil, 2005; Savolainen et al., 2007). Evolutionary dynamics may play a major role in the build-up of lowland rain forest community diversity, and the role of genetic diversity (including sensu lato both allelic and gene expression variability) in ecological processes has been widely acknowledged (Ford, 1964; Randall Hughes et al., 2008). In other words, if ecological sorting of functional traits has occurred across different habitats and has led to the emergence of ecologically different species, it is sensible to expect that such processes are also currently occurring within species. Therefore, in species with continuous stands growing in different, contiguous habitats, we should be able to observe ‘highly local’ intraspecific divergence (sensu Salvaudon et al., 2008) between sub-populations submitted to divergent local environmental conditions; moreover, we expect that divergence between intraspecific sub-populations growing in different habitats should co-occur with divergence between species with different ecological preferences for those habitats. Here, we use the term ‘highly local’ to characterize patterns observed at scales for which environmental turnover occurs at shorter distances than gene flow (i.e. the average distance between patches of different habitat types is shorter than the average gene dispersal distance, implying that gene flow occurs among different habitats).

Tree populations in general are known to harbour large amounts of heritable variation for several putatively adaptive characters (Cornelius, 1994; González-Martínez et al., 2006); Neotropical rain forest trees are no exception (Navarro et al., 2004; Scotti et al., 2010). If adaptation contributes to divergence between sub-populations occupying different habitats, these sub-populations should be differentiated at potentially adaptive traits (sensu Howe and Brunner, 2005). The goal of the present study was therefore to test whether populations of tree species growing as continuous stands across different habitats could be sub-divided into habitat-associated sub-populations displaying phenotypic divergence for such traits [i.e. divergence caused by differentiation in (multi-locus) gene frequencies, by maternal effects or by inheritance of stable gene expression patterns (‘epigenetic inheritance’)]. The test was performed in two congeneric rain forest tree species of the Guiana Shield (Eperua falcata and E. grandiflora), that display partially divergent habitat preferences (Sabatier et al., 1997; Baraloto et al., 2007) but occur, even in low abundance, in multiple habitat types. In Eperua species, gene flow is expected to be restricted – mainly due to heavy seeds – but still intense at the distances considered here (estimate of mean parent–offspring distance for E. grandiflora: 166–343 m; Hardy et al., 2006). In spite of such dispersal distances, a recent study, performed partly on the same populations as those studied in the present paper (Audigeos et al., 2013), has shown that molecular divergence occurs (in E. falcata) at a highly local scale for genes involved in response to soil water content-related stress, against an overall background of no genetic differentiation at other loci.

Two specific questions are asked in this study about phenotypic divergence in these two congeneric species. (1) Do seedlings from different local habitats diverge phenotypically? (2) Are patterns of intraspecific phenotypic divergence similar to those observed at the interspecific level?

MATERIALS AND METHODS

Study species

Eperua falcata and E. grandiflora are abundant in the Guiana Shield, and grow sympatrically in different but partially overlapping habitats. This allowed us to compare intraspecific and interspecific patterns of divergence in the same phylogenetic context and ecological background. Eperua falcata (Aubl.) (Fabaceae) has a preference for seasonally waterlogged bottomlands, whereas E. grandiflora (Aubl.) Benth (Fabaceae) is mostly restricted to hilltops and slopes (Baraloto et al., 2007). The two species differ in several morphological and functional traits, but their seedlings display similar degrees of tolerance to drought or hypoxia under controlled conditions (Baraloto et al., 2007), indicating that they are potential generalists for soil water conditions, at least at the younger life stages. Both species are bat pollinated (Cowan, 1975) and disperse their heavy seeds by explosive dehiscence and gravity at short distances of a few metres (Forget, 1989). Gene dispersal distance is about 150–350 m for E. grandiflora (Hardy et al., 2006) and probably similar for E. falcata (O. Hardy, pers. comm.), well beyond the size of the habitat patches studied here. Data from nuclear genetic markers (Audigeos et al., 2013) suggest that E. falcata is allogamous with no significant selfing.

Study site

The experiment was performed in Plot 6 at the Paracou forest inventory site (5°18′N, 52°53′W) (Gourlet-Fleury et al., 2004) located in an undisturbed forest in coastal French Guiana, South America. The sampling area covers 9 ha and is characterized by a rugged landscape formed by the alternation of 40–50 m high hills, slopes and bottomlands, varying in soil drainage type and water table depth (Gourlet-Fleury et al., 2004). In such a habitat mosaic, variations occur on geographical distances of the same order of magnitude as pollen and seed dispersal but do not occur monotonically (i.e. there is no continuous gradient in a given spatial direction). Three habitat types have been identified in the study area (Supplementary Data, Fig. S1) based on elevation, soil drainage and waterlogging characteristics (Ferry et al., 2010): ‘bottomlands’ (B) with hydromorphic soils and a water table between 0 and 60 cm in depth depending on the season (Supplementary Data, Fig. S1); ‘slopes’ (S) with surface drainage conditions, and a water table consistently below 100 cm; and ‘hilltops’ (H) with deep soils, deep vertical drainage and a water table consistently below 150 cm.

Seed sampling

A total of 267 E. falcata trees and 67 E. grandiflora trees were identified in the study area. Operators visited the plot at least three times a week in February–March 2006, 2007 and 2008 to hand-collect seeds on the ground from 44 fruiting trees. The choice of the mother tree set was based on several considerations: (a) tree fertility; (b) balanced sampling from all habitats; and (c) non-overlapping tree crowns. Pairwise distances between same-habitat fruiting trees were not statistically smaller than between trees in different habitats (Fig. 1).

Fig. 1.

(A) Pairwise spatial distances between mother trees within and between microhabitats. Boxes show the standard deviation of each group. (B) Boxplots showing the distribution of pairwise spatial distance between mother trees within and between all microhabitats.

Seeds collected for our experimental study were assigned to the same habitat as their mother tree, thus forming three different native habitat types (‘B’, ‘S’ and ‘H’). When crowns of conspecific trees overlapped, seeds were collected at opposite sides of the crown. Each seed was assigned to a maternal family corresponding to its mother tree. A total of 3122 seeds were collected over the three seed production years.

Glasshouse experiment

The seeds were weighed and laid down in germination boxes that were filled with a substrate made of river sand which was kept damp using an automatic sprinkler system. Germination success rate was about 60 % for both species. Two months after germination, the seedlings were transplanted into individual 12 L pots filled with a mixture of sand and an A-horizon soil (30/70, v/v), then transferred to a glasshouse. The A-horizon had been collected in the same plot as the seeds and contained about 1·4–1·9 g kg⁻¹ of nitrogen (Ferry et al., 2010).

About 4 % of the seedlings died before transfer to the glasshouse. The remaining seedlings were grown in the glasshouse for 24 months, until the study ended; then they were harvested. The 1637 seedlings (Supplementary Data, Table S1) were randomly assigned to each of 103 blocks of 16 plants. Each block contained four seedlings from each of four randomly drawn maternal families, so that each family was combined randomly with a different set of other families in each of the blocks in which it was represented (see Supplementary Data, Method 1 for details). The seedlings were placed under non-limiting conditions, which prevented both drought and hypoxia (expected to occur in the field on hilltops/slopes and in bottomlands, respectively; see above). Moreover, seedlings grown in the glasshouse experienced higher light levels and milder competition than in natural conditions, favouring optimal growth. A layer of neutral shade-cloth was used to reduce irradiance received to about 13 % of full sun [maximum photosynthetic photon flux density (PPFD) approx. 300 µmol m⁻² s⁻¹] to simulate solar radiation levels received by seedlings in gap openings. Seedlings were watered 2–3 times per week to maintain the substrate close to field capacity (approx. 0·25 m³ m⁻³). The pots were fertilized every 6 months (5 g of complete fertilizer per pot, 12/12/17/2 N/P/K/Mg). Pots were distributed in the glasshouse following an incomplete randomized block layout (for the details of the experimental design, see Supplementary Data, Method 1).

A total of 1637 seedlings survived until month 24. For measures taken at 24 months, the sample used in the present study was restricted to 656 seedlings of E. falcata and 170 seedlings of E. grandiflora (Supplementary Data, Tables S1 and S2), since two-thirds of the seedlings grown in this experiment were set aside for a companion experiment involving different soil water content treatments.

Phenotypic traits

We recorded 23 functional traits (Table 1) related to plant growth, biomass allocation, leaf structure and leaf physiology (photosynthetic capacity and carbon isotope composition). These traits are commonly used as proxies of plant fitness in general (Kraft et al., 2008) and their ecological significance as proxies of fitness in seedlings has been established by several studies (Westoby et al., 2002; Cornelissen et al., 2003; Wright et al., 2004).

Table 1.

List of abbreviations and units of phenotypic traits

Abbreviation	Trait	Units
Growth and biomass allocation
(1) Seedling dimensions
H6	Height at 6 months	cm
H12	Height at 12 months	cm
H18	Height at 18 months	cm
H24	Height at 24 months	cm
H612	Elongation rate from 6 to 12 months	cm month−1
H1218	Elongation rate from 12 to 18 months	cm month−1
H1824	Elongation rate from 18 to 24 months	cm month−1
D18	Diameter at 18 months	mm
D24	Diameter at 24 months	mm
D1824	Radial growth rate from 18 to 24 months	mm month−1
(2) Biomass and allocation
TM24	Total dry mass at 24 months	g
RM24	Root dry mass at 24 months	g
LM24	Total leaf dry mass at 24 months	g
LA24	Total leaf area at 24 months	cm2
LMR24	Leaf/total mass ratio at 24 months	g g−1
RMR24	Root/total mass ratio at 24 months	g g−1
LAR24	Leaf area/total biomass ratio at 24 months	cm2 g−1
Leaf traits
LMA18	Leaf mass/area ratio at 18 months	g m−2
LMA24	Leaf mass/area ratio at 24 months	g m−2
%C18	Carbon content in leaves at 18 months	%
%N18	Nitrogen content in leaves 18 months	%
Asat	Light-saturated carbon assimilation rate at 18 months	μmol m−2 s−1
δ13C	Carbon isotope composition of leaves at 18 months	‰

Plant height and stem diameter at collar were measured every 6 months. Net CO₂ assimilation rate under saturating irradiance (A_sat, μmol m⁻² s⁻¹) was recorded in vivo at 18 months on one leaf per plant with a portable photosynthesis system (CIRAS1, PP-Systems, Hoddesdon, UK) operating in open mode and fitted with a Parkinson leaf cuvette, under the following microclimate: ambient air CO₂ concentration = 380 µmol mol⁻¹; PPFD = 600 ± 20 µmol m⁻² s⁻¹; vapour pressure deficit = 1·0 ± 0·5 kPa; ambient air temperature = 28·7 ± 2·0 °C. Full stabilization was obtained after about 3–5 min. Measurements were conducted between 0900 h and 1300 h to avoid mid-day depression of photosynthesis. After gas exchange measurements, 6–8 mature and fully expanded leaflets were collected per plant close to the top of the stem. Fresh leaf area (LA) was then measured in the laboratory with an area meter (Li-2100, Licor, Lincoln, NE, USA). The leaves were subsequently dried to constant weight at 60 °C for about 3 d, then finely ground to measure carbon (C) and nitrogen (N) content and carbon isotope composition (δ¹³C, ‰) as a surrogate for intrinsic water-use efficiency (WUE_i; Farquhar et al., 1982). Elemental and isotopic analyses were conducted on a sub-sample of about 1 mg of dry leaf powder with an isotopic ratio spectrometer (Delta-S Finnigan Mat, Bremen, Germany). Leaf mass to area ratio (LMA, g m⁻²) was calculated as the ratio of dry mass to LA.

At 24 months, all the plants were harvested and the leaves, stems, and roots were separated for biomass measurements. Total leaf area was measured with the same area meter as above. All three compartments were dried at 60 °C to constant weight for about 3–4 d and then weighed. Leaf area to total biomass ratio (LAR, m² g⁻¹) was obtained by dividing the total LA of a given plant by its total dry weight. Leaf mass ratio (LMR, g g⁻¹) and root mass ratio (RMR, g g⁻¹) were calculated as the ratio of leaf or root dry mass to total plant dry mass (Table 1). Growth rates for height and diameter growth between two dates were calculated as ΔP/Δt = (P_t2 – P_t1)/(t2 – t1), where P indicates the phenotypic value and t1, t2 the times of the two different measurements.

Linear model of character variation

We fitted a classical linear model for the partition of individual phenotypic values, including species, native habitat, maternal family, year of seed collection and seed mass as sources of trait variation in a hierarchical framework. To produce unbiased estimates of progeny and native habitat type effects, interannual variation and seed mass effects were used as cofactors in the model, as they capture, at least partially, environmental effects mediated by maternal allocation to seeds, and thus represent ‘maternal effects’ related to resource availability (Rice et al., 1993; Leiva and Fernández-Alés, 1998; González-Rodríguez et al., 2012). Our hierarchical framework allowed us to estimate the effects of each habitat type for each species, and the effect of each maternal family in each native habitat and each species. The linear model for all traits is as follows:

(1)

where Y_ijklm is the phenotypic value of the i-th individual, μ the global mean, α_j the effect of the j-th year of seed sampling and cultivation, β_k the effect of the k-th species, γ_kl the effect of the l-th native habitat type within the k-th species, τ_klm the effect of the m-th progeny within the l-th native habitat within the k-th species, φ_k the regression coefficient between trait value and seed mass in the k-th species, seed mass_i the fresh mass of the i-th seed, and ɛ_ijklm the residual variation of the i-th individual.

Model parameters and effects were estimated in a Bayesian framework (see Supplementary Data, Methods 2 for details) using the WINBUGS^® software (Lunn et al., 2000). Bayesian methods can easily accommodate unbalanced/incomplete experimental designs (Browne and Draper, 2006) [erratic seed output (Supplementary Data, Tables S1 and S2) made a balanced design impossible in our study].

Conventional hypothesis testing of the significance of effects can be performed using the 95 % posterior distribution of effects (Qian and Shen, 2007). In this context, credible intervals are treated as the Bayesian analogues of confidence intervals: an estimated parameter has 95 % chance to be within the credible interval (Ellison, 1996): parameters for which zero falls outside the credible interval are considered significantly different from zero. The statistical consequences of multiple testing were evaluated by computing the Bayesian analogue of the false discovery rate (FDR; Benjamini and Hochberg, 1995; Miranda-Moreno et al., 2007).

Bayesian estimation of maternal family variance effects

We computed the ratio of maternal family variance (which includes truly genetic, epigenetic and possibly non-genetic maternal effects, and which we summarize as σ²_M) to total phenotypic variance (σ²_P). To estimate variances, we used a reduced version of linear model (1) restricted to family variations within each species. Phenotypic values were broken down as follows:

(2)

and the ratio of maternal family variance to total variance was estimated as:

Maternal family effects were estimated by fitting a quantitative–genetic hierarchical model by a Bayesian inference method of variance partitioning (Supplementary Data, Methods 3). This simplified model was preferred to the full model to compute variance components because (1) it is designed to estimate variance components directly, thus saving computation time and (2) the maternal family-level component (σ²_M) we wished to obtain included all sources of among-family variation, including habitat, but did not include species effects (each species is treated separately).

Phenotypic correlations between traits

We estimated phenotypic correlations both at the individual (seedling) and at the maternal family level, using observed individual phenotypic values of seedlings and Bayesian estimates of maternal family values, respectively. The latter were computed as the sum of all sensu lato ‘genetic’ factors from eqn (1): Y'_klm = μ + β_k + γ_kl + τ_klm.. The sum of these factors conveys the mean phenotypic value of each progeny free from seed mass and year effects (which represent ‘environmental maternal effects’). Phenotypic correlations were calculated using Pearson's coefficient. Significance at two-tailed α = 0·05 was tested by the cor.test function in R (R Development Core Team, 2008). The FDR (Benjamini and Hochberg, 1995) was computed for all correlation matrices.

RESULTS

At the intraspecific level, native habitat had a significant effect on 18 out of 23 traits in E. falcata, and 15 out of 23 in E. grandiflora (Fig. 2; Supplementary Data, Tables S3–S6). Both species displayed significant variation among native habitats for growth traits (including height and diameter; height and diameter growth rates; total, root and leaf mass): seedlings from bottomlands grew faster and produced more biomass than those from slopes and hilltops. Growth rate varied significantly among native habitats at early stages in both species, but this effect vanished after 12 and 18 months for E. grandiflora and E. falcata, respectively. In both species, δ¹³C, leaf area and leaf mass were larger, and N content smaller, in seedlings from bottomlands than those from the other two habitats. Eperua falcata seedlings from bottomlands showed lower LAR, but higher LMA, than those from slopes and hilltops. For E. grandiflora, A_sat was higher in bottomland seedlings than in those from hilltops. We did not find any significant variation in RMR among native habitats. We estimated the expected rate of false positives (FDR) as 0·8 % with a single test α = 5 % as used here.

Fig. 2.

Phenotypic differentiation among habitat types for growth, biomass allocation and leaf traits for Eperua falcata (top) and E. grandiflora (below) sampled at Paracou, French Guiana. Bayesian estimates of departures of each group from the global mean are shown as boxes; t-bars show the 95 % Bayesian credible interval of the estimated parameters. Figures above each plot provide the within-species trait means, which correspond to the zero value in the plots. Units for each trait are provided in Table 1. Hilltop, slope and bottomland are as indicated in the key in the top panel. Asterisks indicate a significant effect of habitat type (95 % CI intervals do not overlap 0).

The two species displayed significant differences for a sub-set of the recorded traits (Supplementary Data, Tables S3–S6): E. falcata seedlings had significantly smaller stems, higher LA and LMR, higher %N and lower LMA than E. grandiflora. No difference was detected for growth rate, biomass accumulation, A_sat or δ¹³C. Nine traits (LMR₂₄, LAR₂₄, D₁₈, D₂₄, H₆, H₁₂, H₁₈, LMA₂₄ and %N₁₈; Supplementary Data, Tables S3–S6) had significant differences at both the intra- and interspecific level. For these traits, intraspecific trends ran contrary to the interspecific traits (Fig. 3; Supplementary Data, Tables S3–S6), i.e. the overall direction of change between same-species hilltop and bottomland sub-populations was contrary to the change between hilltop-preferring E. grandiflora and bottomland-preferring E. falcata. None of the traits showing significant differences among hilltop and bottomland sub-populations also showed significant differences in the same direction between hilltop-preferring E. grandiflora and bottomland-preferring E. falcata. Four traits (RMR₂₄, D₁₈₂₄, H₁₂₁₈ and A_sat; Supplementary Data, Tables S3–S6) showed such a trend, but for none of them were the effects significant both at the species and at the sub-population level. Cofactors representing maternally transmitted environmental effects (year of fruit set and seed mass) also influenced several traits (Supplementary Data, Tables S3–S6).

Fig. 3.

Comparison of the direction of trait value change between habitats (within species) and between species for traits with significant differences both at the intraspecific and at the interspecific level (see Supplementary Data, Tables S3–S6 for raw results). Trait names and units are as described in Table 1. B, bottomlands; S, slopes; H, hilltops; sp, species-level values. Bayesian posterior medians for Eperua falcataand E. grandiflora as indicated in the key in the first graph. Error bars are Bayesian 95 % credible intervals; non-overlapping credible intervals between two values imply significant differences (95 % CI intervals do not overlap 0).

The maternal family effect (which is obtained independently from the native habitat effect described above) was significant for all traits in both species (Supplementary Data, Tables S3–S6). Ratios of maternal family-to-total variance (σ²_M/σ²_P) ranged between 1·2 % [H₁₈₂₄; 95 % credible interval (c.i._0·95) = 0·007–7·2 %] and 10·1 % (H₁₈; c.i._0·95 = 4·9–20·5 %) in E. falcata, and from 0·02 % (LMA₁₈; c.i._0·95 = 0·00003–2·61 %) to 25·4 % (LA₂₄; c.i._0·95 = 6–58 %) in E. grandiflora (Fig. 4; Supplementary Data, Table S7). Credible intervals were larger in E. grandiflora than in E. falcata (Supplementary Data, Table S7) probably due to differences in sample size (Supplementary Data, Table S1).

Fig. 4.

Boxplots of Bayesian posteriors of σ²_M/σ²_P for all traits for Eperua falcata and E. grandiflora. Trait names are as described in Table 1.

Correlation matrices were very similar between the two species (Supplementary Data, Figs S3 and S4). Most traits showed significant correlation at the individual seedling level (raw phenotypic data), but not at the maternal family level (capturing maternally inherited effects on traits). Seedling-level and progeny-level trait correlation matrices, if both significant, always had the same sign; we did not observe any significant family-level correlation without matching significant seedling-level correlation. At the seedling level, two main correlation groups emerged: dimensions, biomass and leaf traits (Table 1; Supplementary Data, Figs S3 and S4) were tightly correlated; allocation traits were all negatively correlated with the remaining traits and had a mixed pattern of correlation to each other. Leaf mass per area (LMA) was somewhat intermediate, showing both positive and negative correlations with dimension, leaf and biomass traits and positive correlation with RMR. At the maternal family level, traits such as A_sat and δ¹³C retained their positive correlation with biomass traits (but not with dimension traits) and their negative correlation with allocation traits; the latter globally retained their negative correlation with all other traits and the positive correlation between LAR and LMR (although fewer correlations were significant in E. grandiflora than in E. falcata). The FDR was <2 % for all matrices for both the 5 % and the 1 % significance threshold (Supplementary Data, Figs S3 and S4).

DISCUSSION

Divergence among sub-populations and maternal families was apparent for several traits, indicating the presence of maternally inherited variability in both species, in agreement with existing estimates of quantitative trait diversity in wild tree populations (Cornelius, 1994; Coutand et al., 2010; Scotti et al., 2010).

After removal of environmentally derived maternal effects (as described by seed mass and year of fructification), native habitat explained a significant fraction of phenotypic differentiation for several leaf- or plant-level traits. These effects are relatively small (Supplementary Data, Fig. S2) but significant, which is quite surprising, considering the small spatial scale at which they occur. A sub-set of these traits may show divergence between sub-populations only because they are correlated with traits that are involved in some adaptively meaningful divergence (Lande and Arnold, 1983). The analysis of phenotypic correlations at the progeny level actually reveals that 20 of the 23 traits (61 %) showing some degree of divergence are correlated to at least another divergent trait. Because maternal family-level correlations were estimated on mean maternal family phenotypic values (which do not include seed mass and year of production effects), the correlations between traits is probably driven by several factors [including epigenetic effects, pleiotropy, and physical quantitative trait locus (QTL) linkage], which we cannot break apart with the current data set.

Nine traits (Fig. 3) displayed divergence both between species and between sub-populations within species. For all these traits, the intraspecific patterns ran opposite to the interspecific one. This suggests that intraspecific trait distributions may be unimodal functions of environmental variables with peak positions that differ between species (‘reaction norm shift’: Fig. 3; fig. 5 in Albert et al., 2010; Crispo, 2007). In such conditions, if the span of environmental conditions sampled is limited relative to the extent of such unimodal distributions, one may observe the kind of patterns reported here, with intraspecific trends contrary to interspecific ones (Albert et al., 2010). Four additional traits (RMR₂₄, D₁₈₂₄, H₁₂₁₈ and A_max18; Supplementary Data, Tables S3–S6) had monotonic intraspecific trends that were concordant with interspecific ones, but without significant effects at either the species or the population level, or both. These results show that, at least for a relatively large sub-set of traits (nine out of 23, or 39 %), it is possible to detect intraspecific variation for those traits showing interspecific variation along the same environmental gradients. This is in agreement with the hypothesis that the differentiation processes currently affecting within-population diversity may be the same as those that caused species divergence, although our observations require confirmation by functional–ecological experiments.

The maternally transmitted component of both trait divergence and trait correlations may have multiple origins.

(1) Environmentally driven maternal effects (i.e. variation in resource availability transmitted to seedlings through seed resources) can influence seedling growth (González-Rodríguez et al., 2011, 2012). In our study, these were controlled through modelling of the effect of both seed mass and year of seed set, which are estimated separately from maternal family effect; therefore, we suggest that these effects should be negligible in our estimation of sensu lato genetic factors, although some cases of maternal background × environmental effects have been reported (Rice et al., 1993; González-Rodríguez et al., 2011).

(2) ‘Epigenetic’ maternal effects (mainly due to the transient transmission of gene expression states through the embryo) can contribute to similarity of traits within maternal families, thus inflating maternal family effects. Epigenetic inheritance has been proven to occur in trees (Rix et al., 2012), although its overall impact on trait variance was negligible. It is not possible to estimate the importance of such effects in our study, and they can clearly contribute to trait divergence among maternal families from different native habitats, if mother trees transmit environmentally induced gene expression states to their progeny. These variations in epigenetic state may have an adaptive meaning, if epigenetically inherited trait values confer higher fitness in the maternal habitat.

(3) Truly heritable (additive and non-additive) genetic effects may also contribute to trait divergence, and also have an adaptive meaning, for the same reasons as in (2). Two arguments let us think that ‘truly genetic’ effects may account for at least part of the observed divergence between sub-populations. First, the same E. falcata adult tree population used for the present study displayed molecular–genetic divergence between habitats for genes involved in response to stresses related to soil water content (Audigeos et al., 2013); this supports the possibility that genetic structuring can occur in these populations. Secondly, we have shown that there are significant phenotypic differences between maternal families within habitats. If habitat-driven differentiation were only caused by epigenetic effects related to environmental differences, variation between same-habitat maternal families should be negligible, which is not the case in our results. Traits that had large maternal family variance components (σ²_M/σ²_P) in our study (e.g. height and biomass traits, and LA; Fig. 4) often also showed high heritability in other tropical or temperate tree species (Vásquez and Dvorak, 1996; Hodge et al., 2002; Carnegie et al., 2004; Navarro et al., 2004; Scotti-Saintagne et al., 2004; Costa e Silva et al., 2005; Sotelo Montes et al., 2007; Callister and Collins, 2008; Ward et al., 2008a, b; Scotti et al., 2010), suggesting that a non-negligible part of the phenotypic divergence among maternal families may be due to true genetic factors; it has to be noted that heritability estimates are generally obtained at the species or at the whole-population level, without considerations for environmental sub-division, and therefore our σ²_M/σ²_P estimates are properly comparable with previous studies. Finally, it has been proven that plant populations can show genetic divergence at functional traits even if they are potentially connected by migration (Hovenden and Vander Schoor, 2004; Byars et al., 2007) or have been shown to undergo strong gene flow (Gonzalo-Turpin and Hazard, 2009).

Whatever the mechanistic base of phenotypic divergence between sub-populations from different native habitats, how likely is it that these differences have arisen because of neutral processes, e.g. spatial genetic structure (due to local inbreeding)? Our study plot is a 300 m sided square, and the largest possible distance between trees is approx. 425 m, within Eperua gene dispersal distance (Hardy et al., 2006); gene flow is thus possible between the different habitat types. Moreover, seeds were sampled in a habitat mosaic, and mother trees inhabiting the same habitat type are not on average closer than trees inhabiting different habitats (Fig. 1). Thus, neutral divergence induced by neutral spatial genetic structure seems unlikely.

Several studies on plants have shown divergence in adaptive traits along environmental gradients (Kawecki and Ebert, 2004; Carlson et al., 2011), particularly with respect to edaphic factors and waterlogging conditions (Silva et al., 2010). The existence of sensu lato heritable traits showing highly local divergence between sub-populations suggests that local adaptation at short geographical distances may occur (Ehrlich and Raven, 1969; Schemske, 1984; Jump et al., 2006; Turner et al., 2010) in the presence of gene flow, which is precisely the sense given by Kawecki (2004) to the term ‘local adaptation’. Conditions for highly local adaptation are not unlikely in tropical rain forests, based on evidence about local species distribution (ter Steege and Hammond, 2001) and the association between functional traits and habitats (Baraloto et al., 2005) over short spatial scales (<50 m) (Kraft and Ackerly, 2010).

Functional considerations can help the interpretation of the observed differences among seedlings native from different habitats. A higher productivity of seedlings from bottomlands as compared with the other two habitats is consistent with larger leaf area and higher A_sat, since these seedlings are therefore able to assimilate more carbon, use it to synthesize more biomass and, eventually, allocate it to growth. This is consistent with the results of previous studies revealing a trend towards increasing growth performances from drier to wetter habitats (Russo et al., 2005; Kariuki et al., 2006; Sanchez-Gomez et al., 2006; Ferry et al., 2010). The LMR and LAR were slightly lower in E. falcata seedlings from bottomlands, suggesting that they invest more biomass in roots and stems than in leaves. This is consistent with frequent waterlogging events that drastically reduce O₂ availability in the soil and decrease hydraulic conductivity of roots, with consequences similar to those of drought (Ponnamperuma, 1972). Lower LMR and LAR would also contribute to reducing water loss through a lower LA per unit of plant mass (Poorter and Markesteijn, 2008). Higher LMA in bottomland seedlings also permits a reduction of water loss through the reduction of transpiring LA at the leaf level (Poorter et al., 2009). In parallel, higher investment in root biomass would enhance water capture ability during dry periods as well as root O₂ absorption during wet periods. Furthermore, bottomland seedlings of both species display higher WUE (i.e. less negative δ¹³C) than slope or hilltop seedlings, which means that, during photosynthesis, they use less water for the same amount of CO₂ assimilation (Farquhar et al., 1982). This trade-off in water and carbon use at leaf level is an efficient strategy when soil water resources are limiting (e.g. Ehleringer and Cooper, 1988), not only on hilltops but also in the bottomlands (Baraloto et al., 2007). Finally, variations of N content are well identified as a determinant of photosynthetic capacities (Reich et al., 1994), as revealed by the strong correlations between leaf nitrogen and A_sat. In natural conditions, leaf nitrogen and foliar N:P ratios are known to be highly dependent upon soil chemical properties (Townsend et al., 2007), and the dependence of A_max on N is expected to be stronger in N-limiting habitats than in P- or Ca-limiting habitats. Bottomlands have higher N content and lower P content than hilltop habitats (Luizao et al., 2004; Ferry et al., 2010), and we observe here lower %N in bottomland than in hilltop seedlings. This suggests that the faster growing bottomland seedlings, which also have higher photosynthetic rates, have lower nitrogen content, contrary to what is expected – at the interspecific level – according to the World Leaf Economic Spectrum (Donovan et al., 2011).

Conclusions

We detected phenotypic divergence for growth and physiological traits occurring over very short spatial distances within a habitat mosaic. This suggests that large reservoirs of within-species adaptive potential are maintained by trait filtering caused by niche partitioning and habitat associations (Russo et al., 2005; Kraft and Ackerly, 2010), and possibly by local adaptive processes. Species displaying such variation may respond more easily to environmental changes through microevolution (by being able to react adaptively to the expected impact of global change), if at least part of the variation is heritable or is caused by adaptive plasticity. It is worth remembering that epigenetic (maternal) effects can be considered as heritable in the broad sense (Bossdorf et al., 2008; Klironomos et al., 2013). The mechanisms underlying such local intraspecific divergence may also turn out to play a major role in the generation of the outstanding diversity in tropical forest ecosystems and, more generally, to be a fundamental mechanism in the maintenance of trait variation in natural populations.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following. Method S1: design of the incomplete randomized block experiments. Method S2: Bayesian model of phenotypic value partition. Method S3: estimation of σ²_M/σ²_P with a Bayesian two-way analysis of variance. Figure S1. three habitats were defined according to elevation, waterlogging and soil drainage conditions. Figure S2: relative size of different effects on trait variability. Figure S3: Pearson's correlation between phenotypic traits at seedling and maternal family level for Eperua falcata. Figure S4: Pearson's correlation between phenotypic traits at seedling and maternal family level for Eperua grandiflora. Table S1: sampling size. Table S2: sampling size by mother tree and year of fructification. Table S3: parameters estimated by the Bayesian analysis of phenotypic value partition for seedling biomass and energy allocation. Table S4: parameters estimated by the Bayesian analysis of phenotypic value partition for seedling height and growth rate. Table S5: parameters estimated by the Bayesian analysis of phenotypic value partition for seedling diameter and radial growth. Table S6: parameters estimated by the Bayesian analysis of phenotypic value partition for leaf traits. Table S7: σ²_M/σ²_P (%) for plant-level and leaf-level traits under the three hypotheses, full-sibs, mixture and half-sibs.