Introduction

The Guianan shield is one of the largest continuous areas of pristine lowland tropical rainforest in the world. It is characterized by a high species richness of local and regional endemism, particularly among the flora. Dicorynia guianensis Amshoff (Fabaceae) is an endemic canopy tree (Sabatier and Prevost, 1990) representing up to 10% of the Guianan forest stands (Bena, 1960). It is one of the most exploited species for timber wood in French Guiana. As is the general rule for tropical tree species, D. guianensis is characterized by a low density: seven and 35 trees per ha for adults and saplings, respectively (Bariteau, 1993). Moreover, D. guianensis has a patchy spatial distribution (Cabrera-Gaillard and Gignoux, 1990), which led to the use of a spatial approach to analysing genetic diversity.

Caron et al (1998) studied the mating system using allozymes and demonstrated that D. guianensis is mostly outcrossing. This study also showed heterogeneity of pollen clouds between patches. The same authors found that maternally inherited markers (chloroplast DNA, cpDNA) exhibited a strong spatial structure (Caron et al, 2000) and suggested that the maintenance of cpDNA polymorphisms in a patchy pattern was most likely a result of rapid turnover of extinction–colonization phases. In the present study, this approach was extended by combining a nuclear and chloroplast survey of genetic diversity within two different cohorts (adults and saplings) at the population scale. A total of six simple sequence repeats (SSR) developed by Latouche-Hallé et al (2002) were used to examine nuclear diversity. These multivariate markers are extremely powerful tools for describing fine-scale spatial structures (Streiff et al, 1998) and reconstructing gene flow (Streiff et al, 1999). Comparative analysis of nuclear and cytoplasmic diversity permits the separation of the impact of pollen and seed-mediated gene flow on the differentiation of populations (Ennos, 1994). Considering that the species is thought to be pollinated by large bees and that the seeds are dispersed by wind (Caron et al, 1998), it would be expected that asymmetric gene flow would result in different fine-scale spatial genetic structure for nuclear and chloroplast DNA markers.

The aim of this study was to depict the fine-scale spatial and generational genetic structures to infer the major genetic and ecological processes shaping the distribution of genetic diversity in D. guianensis. Therefore, three comparative approaches we implemented. First, demographic and genetic spatial structures were compared in order to identify spatial processes. Second, nuclear and cytoplasmic spatial structures were compared to evaluate the contribution of pollen and seed to local dispersion. Finally, the spatial genetic structure between the adult and sapling cohorts were compared in order to identify the processes modifying diversity during the lifetime of the trees.

Materials and methods

Field site

The field site is part of the Paracou experimental site (detailed description in Schmitt and Bariteau, 1990), which was established in 1984 in the tropical primary rain forest, 45 km west from Kourou, in French Guiana (5°15′N, 53°W). The original aim of the Paracou experiment was to test different silvicultural regimes, to assess the impact of logging for most of the important exploited Guianan tree species. A total of 12 plots of 9 ha each were established. Systematic inventories (species, mapping, number, diameter) are carried out every two years on all adult trees greater than 10 cm in diameter at breast height (DBH). Since 1988 no silvicultural operations have been undertaken in Paracou.

Sampling

This study was conducted in a subset of the Paracou experimental site within four almost contiguous plots of 9 ha (Figure 1), representing in total an area of 40 ha (650 m by 650 m). The demographic sampling was subdivided into two cohorts and was exhaustive within each cohort. The adult cohort was composed of all trees greater than 10 cm DBH (CIRAD-forêt, personal communication). The sapling cohort consisted in the additional inventory of all the 2211 trees taller than 1 m but less than 10 cm DBH (CIRAD-forêt, personal communication).

Figure 1
figure 1

Distribution of the trees sampled in two cohorts for the analysis of the demographic spatial structure. ○: 172 adults (all trees having a DBH >20 cm)+: 2211 saplings (all trees taller than 1 m but less than 10 cm DBH). Trees are clustered in five aggregates according to their spatial distributions (see text).

Full size image

For genetic analysis, the study material was restricted to two samples (Figure 2).

Figure 2
figure 2

Distribution of the trees sampled in the two cohorts for the genetic diversity survey. •: 172 adults (all trees having a DBH >20 cm), +: 375 saplings (all trees taller than 1 m but less than 10 cm DBH). Saplings were randomly selected, except in the central part of the plot, where all saplings were genotyped.

Full size image

(i) The adult sample consisted of all trees greatest than 20 cm DBH present in the study area (172 individuals). By doing this, sampling was restricted to only potentially reproductively mature trees, as the minimum diameters of flowering and fruiting trees are 18.3 and 22 cm, respectively (Caron et al, 1998). Prior incomplete knowledge based on the distribution of rare allozymes indicated that pollen can be dispersed over several hundreds of metres; result confirmed using microsatellites (Latouche-Hallé et al, submitted). As a result, the sampling should comprise a subset of a given population.

(ii) The sapling sample comprised two subsamples: a random sample throughout the study area, and an exhaustive sample in the central part of the study area. In the centre of the study area, all saplings present in a 1.8 ha area were genotyped (Figure 2). In total, 375 saplings were sampled.

Cambium tissue and leaf tissue were collected from each adult tree and sapling, respectively. Samples were frozen at −50°C in the laboratory until DNA extraction.

Molecular analysis

Total genomic DNA was extracted from cambium (adults) and leaf (sapling) tissues following the procedure of Doyle and Doyle (1990), modified as described by Dumolin et al (1995). Six microsatellite loci were amplified by the polymerase chain reaction (PCR Latouche-Hallé et al, 2002). The reaction solution (12.5 μl) contained each of the four dNTPs (100 μM), 0.2 μM of each primer, 1.8 mM of MgCl2, 2.5 μl of 10X reaction buffer (670 mM Tris-Hcl pH 8, 160 mM (NH4)2SO4, 0.1% Tween-20), 3 ng of genomic DNA and 0.2 units of Taq DNA polymerase Sylverstar (Eurogentec, France). The primers were labelled with the infra-red fluorescent dye IR800. Cycling conditions were stated by denaturation at 94°C for 4 min, 35 cycles of 45 s at 94°C, 45 s at the annealing temperature (53–60°C according to the locus), and 45 s at 72°C, with a final extension step of 8 min at 72°C. Following amplification, 2 μl of PCR product were mixed with 5 μl of loading buffer (78% formamide, 10 mM EDTA pH 7.6, 0.1% bromophenol blue and 0.1% xylene cyanol), heated for 3 min at 75°C and then set on ice. A measure of 0.8 μl of the denatured SSR solution was loaded onto a LI-COR automated sequencer on a 25 cm 8% acrylamide/bisacrylamide (19:1) denaturating gel.

One PCR-RFLP cpDNA combination (TC2 primer/Taq I), that has revealed polymorphism in D. guianensis in the Paracou stand (Caron et al, 2000), was used following the Dumolin-Lapègue et al (1997a),(1997b) protocol. Amplification problems due to the large length of the PCR fragment were resolved by designing nested primers to obtain smaller fragment (1366 vs 3236 bp, Table 1).

Table 1 Primer pairs for chloroplast PCR-RFLP analysis
Full size table

Demographic spatial distribution of trees

Spatial coordinates of all adult trees and 2211 saplings present in the study area were available. We used the second-order neighbourhood analysis of Ripley (1976) (K function) to analyse the spatial distribution of the trees. The K(r) function is proportional to the average number of neighbours located a distance r from any given tree (Pelissier and Goreaud, 2001). Given a density λ of trees the following expression holds:

The K statistic could be compared to its expectation under the null hypothesis of complete spatial randomness, which follows a Poisson process. For example, under the null hypothesis, the expected number of neighbouring trees at a distance r from any point is λπr2 and K(r) is then πr2. If trees are aggregated there are more neighbouring trees at distance r and K(r) is greater than πr2. Conversely, if trees are regularly distributed there will be fewer neighbours at a given distance r, hence K(r) will be lower than πr2. When the K values are plotted as a function of r, the curve K(r) is expected to be a parabola. Hence the value of K depends on the distance r. To avoid this difficulty, the L transformation has been used (Besag, 1977). The L function (equation 2) has the same expectation under the null hypothesis (L(r)=0) regardless of the r-value (Besag, 1977a,1977b).

L(r) is positive for an aggregated pattern and negative for a regular spatial distribution of trees. As the theoretical distribution of the estimator of the function L(r) is unknown, Monte Carlo simulations were used to test significant deviation at the 5% significance level (Besag, 1977a,1977b). Analyses were performed by ADS in ADE 4 (package ‘Spatial Data Analysis’ in the software ‘Ecological Data Analysis’; Thioulouse et al (1997); http://pbil.univ-lyon1.fr/ADE-4/Download.html).

Nuclear and cpDNA diversity Analyses

Statistical analysis of microsatellite data was performed using GENEPOP (version 3.3; Raymond and Rousset, 1995). Distributions of the observed allelic frequencies were computed for each locus. For each locus, allele number (A), effective allele number Ae=1/(1–He), mean number of alleles per locus and mean effective number of alleles per locus were calculated. The allelic richness (A) of two cohorts was standardized to remove the bias due to population size using a rarefaction method (El-Mousadik and Petit, 1996). Observed and expected heterozygosities (unbiased estimator of Nei, 1987) were calculated. Fis (fixation index) were estimated according to Weir and Cockerham's (1984) method. A Student's t-test and a z-test were performed to compare allelic richness, He, and Ho between adult and sapling cohorts. Departure from Hardy–Weinberg equilibrium was tested using the exact tests of Haldane (1954) and exact P-values were estimated without bias by the Markov chain method (Guo and Thompson, 1992). A global test across loci was constructed using Fisher's method (Sokal and Rohlf, 1994).

Genetic diversity statistics for maternally inherited cpDNA markers were calculated in similar manner as for nuclear diversity: haplotypic richness (A), effective number of haplotypes (Ae), and within population diversity (Hs). These statistics were estimated by software (written in Fortran) that provide gene diversity and differentiation statistics for uniparentally inherited markers, following Nei's (1987) subdivision of genetic diversity.

Spatial and generational genetic structure analyses

F-statistics were used to measure hierarchical population structure (Weir and Cockerham, 1984). Subpopulations were defined according to (i) generational and (ii) spatial distributions.

  1. i)

    For the generational distribution, the two cohorts (adult vs sapling) were first considered. In addition, the adult cohort was subdivided into three classes according to diameter distributions (20–34, 34–50, and >50 cm).

  2. ii)

    For the spatial distribution, the population was subdivided into five patches according to the spatial distribution of all the 2383 individuals (adults and saplings) in the south block of Paracou (Figure 1).

For nuclear data, analyses were carried out using FSTAT V2.9.1. (Goudet, 1995). Jackknifing over loci was performed to estimate standard errors. Statistical tests were performed for each locus individually using permutations (2000 randomizations). For FIS, alleles were permuted among individuals within samples, for FIT alleles were permuted among samples, and for population differentiation (FST), genotypes were randomized among samples (tested without assuming Hardy–Weinberg). The log-likelihood statistic, G (Goudet et al, 1996) was used to classify genotypic contingency tables. Finally, tests for pairwise differentiation between subpopulations (not assuming HW) were performed by randomly assigning multilocus genotypes between sample pairs.

For chloroplast data, within population (HS), total population diversity (HT) and differentiation (GST) were computed following Pons and Petit (1995). χ2 statistical tests were performed and P-values estimated to compare alleles frequencies.

Autocorrelation analyses were performed using Moran's indexes for both diploid nuclear (monolocus and multilocus) and haploid chloroplast data using SGS software (version 1.0 c Degen et al, 2001). This statistic measures correlation between spatial (sample coordinates) and genetic data. The analysis is based on classes of distances between pairs of individuals. The size (r) and the number of distance classes (k) were chosen such that all classes had approximately the same number of pairs. For a given class i (ranging from 1 to k), Moran's index was calculated for all pairs of trees, whose Euclidean distance was between [(i−1)r; ir]. Deviations from random distribution of alleles were calculated by permutation tests, randomizing the redistribution of genetic data over the spatial coordinates of sampled points. Since the sample size of saplings was larger than that of the adults, and because of the unbalanced sampling of saplings (exhaustive sampling within the central area, and random elsewhere), the sample size of saplings was restricted to that of the adults by sampling the nearest sapling to each adult tree.

Results

Demographic spatial distribution of trees

For both cohorts, the L-function curves exhibited a divergence from the null-hypothesis of a random distribution of trees (Figure 3); positive values suggested that trees are distributed in aggregates. The L-function reached two maxima, at about 100 m, and at about 400 m. Aggregation of trees was therefore suggested at different sizes ranging between these two peak values, meaning that no precise size of aggregates can be distinguished in the data. The L curves indicated that there is significant aggregation between 100 and 400 m. Although there was an overall similarity of the L curves for the two cohorts, the bimodality is sharper for the adults than for the saplings, indicating that aggregation at the two levels is stronger in adults than in saplings.

Figure 3
figure 3

Variation of the L(r)-function according to distance (r): (a) adult trees, (b) saplings. The line corresponds to the observed data. The two dotted lines correspond to the 95% confidence interval envelope calculated by Monte-Carlo simulations.

Full size image

Following these observations, the total population was subdivided into five aggregates for further genetic investigations, corresponding to an intermediate level of aggregation suggested by the L. function (200–300 m). The intermediate level was chosen as a compromise, and was also imposed as a constraint to maintain sufficient sample size per aggregate for testing differentiation among them. It does not necessarily correspond to a biological subdivision but should be seen as a first approach to test for spatial differentiation. There are a number of different ways to subdivide the population into aggregates of approximately 200–300 m. We tested different subdivisions into five contiguous aggregates of about equal size, and the differentiation levels calculated were of similar magnitude in each case. For each subdivision tested, we compared the differentiation values to the value obtained by randomly assigning the genotypes to the trees distributed in the same spatial aggregates. As expected, the differentiation was near zero when genotypes were randomly redistributed.

Genetic diversity within and between cohorts

Allelic richness (Table 2) was lower for adults than for saplings (7.33 vs 8.17). However, after standardization using the rarefaction method, allelic richness was of the same magnitude in the two cohorts (7.33 vs 7.40). Observed heterozygosity displayed a similar range for adults (0.37–0.78) and saplings (0.37–0.73). This was also the case for expected heterozygosity: adults, 0.39–0.83 and saplings, 0.41–0.78. Mean expected heterozygosities, were also similar (0.629 vs 0.623) in the two cohorts, but observed heterozygosity was slightly higher in adults than saplings (0.621 vs 0.598). The frequency profiles were the same in the two cohorts and the most frequent alleles at each locus were shared by the two cohorts, (except Dg5, where two common alleles were present). The Student's t-test (used for allelic richness) and the z-test (used for He and Ho) confirmed that diversity was of similar magnitude in the two stage-classes over all loci.

Table 2 Nuclear genetic diversity statistics
Full size table

The ratio Ae/A varied from 0.27 to 0.80 in the adult population (average=0.44), and from 0.24 to 0.61 in saplings (average=0.37). For the two cohorts only one locus exhibited a ratio greater than 0.50 (Dg3 for adults and Dg5 for saplings). These values indicated an important heterogeneity of the allele frequency profiles; for most loci (except Dg5) there was one frequent allele and many rare alleles. Nevertheless, 11 private alleles (ie present in only one stage-class) were found over all loci, at which three were only present in adults. The three adult-specific alleles may not have been detected in the sapling class because all saplings were not genotyped (Figure 1 and Figure 2) or because they have been lost because of genetic drift. On the other hand, the eight sapling-specific alleles reflect gene flow from outside the study plot.

There is a general trend towards an excess of homo-qzygosity in adult and sapling populations, as indicated by the predominance of positive FIS values (Table 2). Deviation from Hardy-Weinberg equilibrium was more pronounced in the sapling cohort (three loci with significant deviations) than in the adults (only one locus with significant deviations). The fixation index ranged from −0.07 to +0.06 (adults) and from −0.04 to +0.09 (saplings); for multilocus analysis, a slight but significant heterozygote deficit was detected, which was higher for saplings than for adults (0.037 and 0.012, respectively). For chloroplast diversity (Table 3), the number of haplotypes was the same in the two cohorts (A=3), and diversity was also of similar magnitude (0.59 vs 0.57). The three haplotypes were almost evenly distributed in the two cohorts (Ae/A=0.8).

Table 3 Chloroplast diversity statistics
Full size table

Generational vs spatial genetic structure

To test for genetic differentiation between different age classes, the total population was subdivided into two different ways: between the two cohorts (adults and saplings) and within the adult cohort there was an additional subdivision into three classes according to diameter distribution.

Whatever the generational subdivision used, F-statistics showed that differentiation among groups was extremely low (Table 4), and the total fixation index was of the same magnitude within each age class. The lack of generational differentiation was also observed for chloroplast markers (Table 5).

Table 4 F-statistics for generational and spatial subdivisions
Full size table
Table 5 Diversity statistics of chloroplast genetic marker
Full size table

These results contrast with the spatial differentiation. Adult and sapling cohorts were subdivided into five patches (Figure 1). There was a significant differentiation among patches in both adult and sapling cohorts, with FST values ranging from 0.02 to 0.03 for nuclear markers (Table 4) and from 0.29 to 0.42 for chloroplast markers (Table 5 and Figure 4). The differentiation remained significant when FST values were calculated for pairs of patches (all pairwise values for saplings, and 8/10 pairwise values for adults).

Figure 4
figure 4

Spatial distribution of the cpDNA haplotypes in the two cohorts. Each of the three haplotypes is represented by a shaded (white, black, and grey) symbol, for the adults (•) and the saplings (Δ). Borders of the aggregates are in dotted lines.

Full size image

The spatial differentiation among patches was further confirmed by spatial autocorrelation analysis. There was a clear trend of significant positive autocorrelations at short distances for each SSR locus, despite interlocus variation (Figure 5). This trend was also supported by multilocus analysis. The correlogram for nuclear data (Figure 6), showed a significant nonrandom distribution of genotypes at short distances (approximately less than 110 m) for both cohorts. Moran's index for the adult cohort was larger than for the sapling one (0.18 vs 0.08) within the first distance-class. As expected, considering differentiation values, the chloroplast autocorrelations were larger than the nuclear ones at short distances for adults (0.9 vs 0.18, respectively) and for saplings (0.7 vs 0.08, respectively). For adults, the autocorrelations were also significant over larger distances for cpDNA (up to 140 m) than for nuclear DNA (up to 110 m). The same trend was observed for the saplings. Hence, these analyses show that spatial genetic structure is more pronounced for chloroplast markers than nuclear markers. Significant negative values of Moran's index at distances larger than 250 m suggest that pairs of trees are less genetically similar than expected. Indeed, they always exhibit different haplotypes when they are compared at these distances. This is also confirmed by the high GST values for cpDNA among patches (Table 5).

Figure 5
figure 5

Spatial genetic autocorrelation for the six microsatellite loci in the adult cohort. Each solid line corresponds to the observed data for one locus. The grey area corresponds to the 95% confidence interval envelope calculated by 500 Monte-Carlo simulations.

Full size image
Figure 6
figure 6

Spatial genetic autocorrelation for cpDNA markers and six microsatellite loci in the two cohorts. Moran's indexes were calculated in the adult cohort (▪) and in the sapling cohort (□). The solid line corresponds to observed data. The two dotted lines correspond to the 95% confidence interval envelope of calculated by 500 Monte-Carlo simulations.

Full size image

Discussion

Spatial genetic and demographic genetic structures suggest efficient and rapid colonization dynamics

The data of this study confirm the aggregated spatial distribution of trees of D. guianensis, even though a precise size of aggregates is not clearly shown by the L-function. The variation of the L-function rather suggested various possible aggregates ranging in size from 100 to 400 m, with peak values reached at these extremes. Interestingly, the same pattern of spatial distribution was observed in the two cohorts of adults and saplings. Previous demographic studies estimated an average area of aggregates of 0.9 ha (ie a circle of radius 50 m) at the same field site (Eymond, 1992), which correspond to the lower level of aggregation detected here. Goreaud et al (1997) found two smaller peaks (50 and 150 m) but they discarded marsh and low-ground zones in their survey; unfavourable environments for D. guianensis (Kokou, 1992). The lower level of aggregation (100 m or below), groups trees that are very similar genetically, as suggested by the spatial autocorrelation analysis (Figure 6), and illustrated by the spatial distribution of cpDNA haplotypes (Figure 4). Hence, there is strong correspondence between demographic and spatial genetic structure in small distance classes.

The congruence between the demographic and genetic structures is likely to be the consequence of restricted seed dispersal, meaning that the smaller size of aggregates would correspond to the dispersion distance from a single source (ie maternal tree). Larger demographic aggregates most probably result from dispersion of seeds originating from different sources (Figure 4). They could also be constituted by an assemblage of smaller aggregates. As a result, chloroplast variation in large aggregates is more diverse than small ones, even though a significant genetic differentiation between larger aggregates can still be observed (Table 4). An interesting feature of these dynamics is that they permit high levels of diversity to be maintained. It is striking to observe that the level of cpDNA diversity within and between larger aggregates in our study is of the same magnitude as the level observed within and between different stands across French Guiana (Caron et al, 2000).

Maintenance of diversity requires that the founders of different aggregates were diverse, and that the time elapsed since foundation was too short to allow erosion of diversity by drift. Genetic multiplicity of founders is expected for species that have a composite dispersion mechanism, such as diffusion and rare long distance dispersal (Le Corre et al, 1997). An avoidance of fixation of haplotypes is expected in recently formed aggregates for a colonizing species or in nonequilibrium situations where aggregates are ephemeral in time and are subject to rapid extinction–recolonization dynamics. Our results fit within current knowledge about the efficient colonizing ability of Dicorynia. Seeds could germinate in shade or luminous environment (Meer et al, 1998) but saplings need light to grow (Forget, 1988). The species is known to be opportunistic and rapidly colonizes gaps that occur because of accidental trees falls, except in marsh and low-ground zones (Kokou, 1992).

Comparative analysis of spatial nuclear and cytoplasmic genetic structure indicates asymmetry of pollen and seed flow

Nuclear and cytoplasmic autocorrelations show similar patterns and were significant only over short distances (ie less than 100 m). However, there is an important difference in absolute values of Moran's index between these genomes, as I values approach 0.9 at 50 m for cpDNA and 0.18 for nuclear genes (Figure 6). These data suggest that the fine-scale spatial genetic structure is mainly the consequence of restricted seed dispersal, and that pollen flow has limited impact. These conclusions are also supported by the coexistence of low fixation index at the whole population level and significant genetic relatedness at short distances (shown by the significant I values). Genetic relatedness at short distances is therefore likely to result from uniparental relations (half-sibs) rather than biparental relations (full sibs). Hence, neighbouring trees most probably share only one parent in common (the maternal), the male parent being more distantly located. We may therefore conclude that the comparative analysis of fine-scale nuclear and cytoplasmic spatial structure suggests a strong asymmetry between pollen and seed flow. These results can be compared to the current knowledge of the dispersion mechanisms of the species. Seeds of D. guianensis are thought to be wind-dispersed with pollen dispersed by Hymenoptera species (Caron et al, 1998). Previous studies have found average and maximum distances of seed dispersal of 30 and 60 m, respectively, based on seed tagging experiments (Forget, 1988). The maximum distance of effective pollen flow evaluated so far for D. guianensis was 200 m as inferred by the spatial distributions of rare allozymes (Caron et al, 1998). Our genetic results support the biological observations suggesting that pollen is dispersed over larger distances than seeds, resulting in a stronger genetic structure for chloroplast markers than nuclear markers.

Changes of genetic structures between the juvenile and adult cohort

In this study, D. guianensis exhibited similar levels of microsatellite diversity (He=0.63) to other neotropical tree species, Vouacapoua americana (He=0.5, Dutech, 2001) Swietenia humilis (He=0.546, White and Powell, 1997), and Pithecellobium elegans (He=0.64, Chase et al, 1996) but lower than in Symphonia globulifera (He=0.83, Aldrich et al, 1998). The computation of diversity and differentiation statistics has shown that allelic frequencies (either for nuclear or cytoplasmic markers) are more differentiated among aggregates than among cohorts, and has been commented on earlier as the result of dispersion and colonization dynamics of Dicorynia guianensis.

Fixation indices and genetic autocorrelation showed clear genetic structure differences between adult and sapling cohorts. Fixation indices for SSRs were significant for both cohorts, but indicated a higher heterozygote deficiency for the saplings than for the adult trees. This was found in other species comparing different cohorts (review in Murawski, 1995). Maintenance of positive fixation indices in the adult cohort may be the consequence of the tolerance of selfing by Dicorynia (Caron et al, 1998), or the existence of Wahlund effects caused by generation overlapping in the adult cohort. As already mentioned, the adult cohort was composed of trees exhibiting a wide range of dimensions (DBH varying between 20 and 50 cm) and may be composed of two or even three generations. The decrease of fixation index in the adults may result from the preferential elimination of selfed trees because of the high selection pressures in dense tropical tree forests. More interesting is the reinforcement of spatial genetic structure in the adult cohort at short distances (less than 100 m), as compared to saplings, despite the decrease of the fixation index between sapling and adult cohort. The stronger genetic relations between neighbouring adults was shown for both nuclear and cpDNA markers (Figure 6). Overlapping of generations within the adult cohort could be invoked as one hypothesis, suggesting that neighbouring trees may actually be a mixture of parents and offspring. However, when the autocorrelations were computed within separate diameter classes, Moran's index for all adult subgroups remained larger than the saplings (data not shown). Hence, stronger genetic relatedness was present within a given generation. A likely explanation to this observation is selection. If the competitive ability is highly heritable, then selection may favour progenies that share this ability, and within a given neighbourhood only offspring of these progenies will survive to the adult stage.