• Seed dispersal and the subsequent recruitment of new individuals into a population are important processes affecting the population dynamics, genetic diversity and spatial genetic structure of plant populations.
• Spatial patterns of seedling recruitment were investigated in two populations of the terrestrial orchid Orchis purpurea using both univariate and bivariate point pattern analysis, parentage analysis and seed germination experiments.
• Both adults and recruits showed a clustered spatial distribution with cluster radii of c. 4–5 m. The parentage analysis resulted in offspring-dispersal distances that were slightly larger than distances obtained from the point pattern analyses. The suitability of microsites for germination differed among sites, with strong constraints in one site and almost no constraints in the other.
• These results provide a clear and coherent picture of recruitment patterns in a tuberous, perennial orchid. Seed dispersal is limited to a few metres from the mother plant, whereas the availability of suitable germination conditions may vary strongly from one site to the next. Because of a time lag of 3–4 yr between seed dispersal and actual recruitment, and irregular flowering and fruiting patterns of adult plants, interpretation of recruitment patterns using point patterns analyses ideally should take into account the demographic properties of orchid populations.
In general, successful establishment of seedlings depends on seed production and subsequent dispersal on the one hand and the availability of suitable microsites on the other (Turnbull et al., 2000; Münzbergová & Herben, 2005). To elucidate the role of seed limitation vs microsite limitation, in most studies seed addition experiments have been carried out (reviewed in Turnbull et al., 2000 and Münzbergová & Herben, 2005). Turnbull et al. (2000) showed that approx. 50% of all seed augmentation experiments provided evidence for seed limitation, suggesting that seed production is a limiting factor in determining the population dynamics of plant populations. Unfortunately, these experiments mostly investigated whether a species was seed or microsite limited, but largely neglected the spatial component of seedling recruitment. Yet, there may be substantial spatial variation in recruitment within a single population. A better understanding of this spatial variation can be very useful for inferring many aspects of population dynamics and genetics of plant populations, including yearly fluctuations in population size and the origination (Tonsor et al., 1993; Kalisz et al., 2001) or breakdown (Epperson & Alvarez-Buylla, 1997; Chung et al., 2003; Jacquemyn et al., 2006) of spatial genetic structure within plant populations.
This may be especially true for orchids, for which successful germination and seedling establishment have long been recognized as a critical life history stage (Darwin, 1877), despite the fact that most orchid species produce huge amounts of minute, dust-like seeds that are easily dispersed by the wind (Arditti & Ghani, 2000). Surprisingly, few studies have investigated the process of seedling recruitment in orchid populations in a spatial context (McKendrick et al., 2000; Batty et al., 2001; McCormick et al., 2004; Diez, 2007). Because of their dependence on mycorrhiza for successful establishment, it can be reasonably expected that seedling recruitment of orchids is largely determined by the distribution and availability of suitable mycorrhiza (Rasmussen, 1995), but at present little is known about the actual distribution of orchid mycorrhiza within and among sites (Otero & Flanagan, 2006). The few studies that experimentally introduced orchid seeds into the soil demonstrated that at small scales germination decreased with increasing distance from adult plants (McKendrick et al., 2000, 2002; Batty et al., 2001; Leake et al., 2004; Diez, 2007), suggesting a declining abundance of mycorrhiza further away from adult plants, whereas others did not find such a relationship (Masuhara & Katsuya, 1994; McKendrick et al., 2000). In addition, other factors such as soil moisture, organic content and pH can also affect germination success (Diez, 2007).
However, most seed-sowing experiments did not investigate the establishment and subsequent survival of seedlings within the population. To do so, longer term data sets and spatially explicit sampling of seedlings and juvenile plants are needed to fully capture the role of recruitment limitation in determining the population dynamics of plant populations (Clark et al., 1999). Investigation of the spatial structure and recruitment dynamics using point pattern analysis can therefore be a first step in providing important insights into the processes affecting orchid population dynamics (Perry et al., 2002; Wiegand & Moloney, 2004). However, although the analysis of spatial structure can indicate the existence of underlying processes, it remains challenging to infer causation because several ecological processes can generate the same spatial pattern (Wiegand & Moloney, 2004). Therefore, it will probably be most fruitful to combine the spatial pattern analysis with experimental investigations of seed germination and offspring distances inferred from molecular markers.
Here, patterns of seed germination and seedling recruitment were studied in two populations of the perennial, tuberous orchid Orchis purpurea. More specifically, the following research questions were addressed. How are seedlings distributed relative to the adults? What are observed parent–offspring distances inferred from the genetic markers and parentage analyses and how do these results compare with the results of the point pattern analyses? How is microsite availability affecting spatial patterns of seedling recruitment? To answer these questions, the spatial structure of both recruits and adult plants was first described using univariate and bivariate spatial point pattern analyses (see Wiegand & Moloney, 2004 for details). Secondly, amplified fragment length polymorphism (AFLP) markers and parentage analyses were used to assess parent–offspring dispersal distances and to compare these with distances obtained from the spatial point pattern analyses. Finally, a seed-sowing experiment using seed packages was conducted to investigate the suitability of microsites for germination.
Materials and Methods
Orchis purpurea Huds. (Lady Orchid) is a tall, tuberous, perennial orchid that mainly occurs in forests on limestone and in calcareous grasslands. In the latter, it is mostly found in the immediate vicinity of trees and shrubs, whereas in forest habitat it usually occurs in light gaps or along the forest edge. Orchis purpurea perennates during the winter and the leaves appear above the ground in February. Plants have one to four (sometimes up to seven) basal leaves of elliptic-ovate to lanceolate shape, 2–5 cm wide and 6–20 cm long (Rose, 1948). Flowering takes place at the end of May. Flowering stalks vary in height between 25 and 60 cm (sometimes reaching 80 cm) and carry 10–50 bright white to purple-brown, self-compatible flowers (Jacquemyn et al., 2002). Seed capsules ripen by the end of June, and this is followed by dehiscence and seed dispersal in August. Fruit production is generally low, with < 10% of all flowers successfully setting seed (Darwin, 1877; Jacquemyn et al., 2002). Seeds are small (length 0.38 ± 0.04 mm; width 0.15 ± 0.07 mm (mean ± SD)) and have an average volume of 2.65 (± 1.69) mm3 (mean ± SD; Arditti & Ghani, 2000). Germination of seeds is probably induced by saprotrophic soilborne fungi of the genus Rhizoctonia. From mid-August onwards, no living green parts are found above-ground.
Study sites and sampling
The study was conducted in two natural populations of O. purpurea located in Voeren, Belgium. Here, the species is rare and endangered (Jacquemyn et al., 2005). Both populations have been known for many decades, but had seriously decreased in size as a result of inappropriate management. As a consequence of the restoration of the original management, both populations have expanded substantially during the last 5 yr, providing excellent opportunities to study the spatial factors affecting seedling recruitment in this species. Before restoration of the sites began, the populations consisted solely of adult, mostly vegetative plants and no seedlings were observed. Both sites were intensively monitored from 2001 to present. During this period, all individuals were mapped to the nearest centimetre and their life history stage determined (Fig. 1a,b). At the moment of sampling (2004), the first site consisted of 340 individuals (193 adults and 147 recruits) and was located in a large forest gap that comprised an area of c. 2500 m2 (density = 0.136 individuals m−2) (Fig. 1a). The second site was situated in a species-rich calcareous grassland immediately bordering forest habitat and comprised an area of 375 m2. This site consisted of 172 individuals (density = 0.459 individuals m−2) (Fig. 1b), of which 99 individuals (58%) were adults. The maximum distances between individuals were 57.69 and 20.45 m for the first and second sites, respectively (Fig. 1a,b).
Sampling for genetic analyses and AFLP protocol
In spring 2004, a total of 389 individuals (adults and seedlings) were sampled from both populations (Table 1). Young leaf material was collected and immediately frozen in liquid nitrogen. Before DNA extraction, plant material was freeze-dried for 48 h and homogenized with a mill (Retsch MM 200, Retsch GmbH, Haan, Germany) to fine powder. Total DNA was extracted from 30 mg of freeze-dried leaf material using methods described in Dendauw et al. (2002). After extraction, DNA concentrations were estimated on 1.5% (weight/volume (w/v)) agarose gels.
Table 1. The proportion of polymorphic loci (P) and expected heterozygosity (Hj) in two Orchis purpurea populations
Results for all individuals and for adults and recruits separately are shown.
AFLP analysis was carried out according to Vos et al. (1995), using commercial kits and following the protocol of Roldán-Ruiz et al. (2000). The enzymes EcoRI and MseI were used for DNA digestion. Each individual plant was fingerprinted with three primer combinations: EcoRI-AGG/MseI-CTGC, EcoRI-AGG/MseI-CTGG and EcoRI-AGG/MseI-CTAG. Fragment separation and detection were performed using an ABI Prism 377 DNA sequencer (AME Bioscience Ltd., Sharnbrook, UK) on 36-cm denaturing gels using 4.25% polyacrylamide (4.25% acrylamide/bisacrylamide 19/1 and 6 m urea in 1 × tris-boric acid-disodium EDTA (TBE)). A GeneScan 500 Rox-labelled size standard (Applied Biosystems, Foster City, CA, USA) was loaded in each lane. The fluorescent AFLP patterns were scored using genemapper version 3.7 (Applied Biosystems). We scored the presence or absence of each marker in each individual plant. Each individual displayed a unique banding pattern (Jacquemyn et al., 2006).
Experimental investigation of seed germination: a field sowing experiment
To study the possible effect of seed vs microsite limitation on seed germination and seedling establishment, seed packages were constructed using a modified design from Rasmussen & Whigham (1993). This technique has the advantage that orchid seeds are subjected to the physical and chemical conditions of the substrate, and contact with small soil organisms is possible, but the seeds are protected from larger animals such as millipedes and earthworms (Whigham et al., 2006). In August 2004, approx. 150 seeds were placed within a square of 53-µm mesh phytoplankton netting, enclosed within a Polaroid slide mount. Seed packets were placed just under the topsoil layer along three transects within each population and marked with a flag. Each transect consisted of 10 sample points that were 4 m (site 1) or 2 m (site 2) apart. At each point, four seed packets were placed horizontally in the ground in the four major directions (north, south, east and west), producing a total of 240 seed packages that were left in the ground for 2.5 yr. When seed packages were retrieved in spring 2007, they were gently washed and maintained moist in paper towel for 1 d until examination. For each slide mount, the number of developed protocorms was determined by visual inspection at ×10 under a dissection microscope.
Point pattern analysis We used second-order statistics and the distribution function of the distances to the nearest neighbour to describe the univariate spatial pattern of recruits and adults and the bivariate relationship between recruits and adults. Second-order statistics such as the pair-correlation function or Ripley's K function are based on the distribution of distances of pairs of points, and are powerful tools to describe the small-scale spatial correlation structure of point patterns (Diggle, 2003). The analysis of nearest neighbour distances, however, provides additional information on aspects of the data not covered by second-order statistics (i.e. different point processes may have identical second-order statistics but different distribution functions of the distances to the nearest neighbour; Stoyan & Stoyan, 1994; Diggle, 2003).
Ripley's K function can be defined using the quantity λK(r), which has the intuitive interpretation of the expected number of further points within distance r of an arbitrary point of the process which is not counted (Ripley, 1976), where λ is the intensity of the pattern in the study area. The pair-correlation function g(r) is related to the derivative of the K function, i.e. g(r) = K′(r)/(2πr) (Ripley, 1976; Stoyan & Stoyan, 1994). Bivariate extensions of K(r) and g(r) follow intuitively (e.g. Diggle, 2003; Wiegand & Moloney, 2004).
Because the pair-correlation function is a nonaccumulative version of Ripley's K function it does not integrate the ‘memory’ (= the trend at the beginning) of small-scale second-order effects to larger scales, as does Ripley's K (Wiegand & Moloney, 2004). Additionally, it is more intuitive than an accumulative measure because it has a direct interpretation as the normalized neighbourhood density function (Stoyan & Penttinen, 2000). We therefore used the pair-correlation function in the following analyses.
We followed the grid-based approach of Wiegand & Moloney (2004) for implementation of the pair-correlation function g(r). We used a grid size of 0.1 m × 0.1 m, and a ring width of 0.2 m for estimation of the pair-correlation functions. This is a sufficiently fine resolution compared with the size of our study plots (Fig. 1), is larger than the precision of the data (c. 1 cm) and is sufficient to respond to our objectives. We calculated the accumulative distribution G(y) of distances y to the nearest neighbour without edge correction (Diggle, 2003). This is appropriate here because all plants of the population were mapped.
We used a Monte Carlo approach for construction of simulation envelopes of a given null model. Each of the n simulations of the point process underlying the null model generates a g (or G) function, and simulation envelopes with an approximate α = 0.05 are calculated from the 25th highest and 25th lowest values of 999 simulations (Stoyan & Stoyan, 1994). Note that we cannot interpret the simulation envelopes as confidence intervals because we test the null hypothesis at many scales simultaneously. This may cause type I error (Stoyan & Stoyan, 1994; Diggle, 2003; Loosmore & Ford, 2006). However, because we are performing an exploratory data analysis and not a confirmatory analysis, this is of minor concern.
Null models: the Thomas process Because the univariate patterns are apparently not random but clustered (Fig. 1a,b), we used a simple cluster process, the Thomas process (Thomas, 1949; Stoyan & Stoyan, 1994; Wiegand et al., in press), to describe the basic characteristics of the univariate patterns. The Thomas process assumes that (1) the spatial distribution of the parents follows a homogeneous Poisson process (complete spatial randomness) with intensity ρ; (2) each parent produces a random number of offspring following a Poisson distribution with mean µ = λ/ρ (λ is the intensity of the offspring); (3) the location of the offspring, relative to the parents, has a bivariate Gaussian distribution h(r, σ) with variance σ2 (Stoyan & Stoyan, 1994). The pair-correlation function g(r) of the Thomas process yields:
The unknown parameters ρ and s can be fitted by comparing the empirical ĝ(r) with the theoretical g function using minimum contrast methods (Stoyan & Stoyan, 1994; Diggle, 2003). We used minimum contrast methods for the parameter fit, but both the contrast of the K function and the contrast of the g function because this provided improved results (for details, see Wiegand et al., in press). Note that 86% of all offspring are located within distance rC = 2σ from the parent and the approximate area covered by one cluster is The process is thus more clustered if there are fewer parents (i.e. ρ is smaller) or if the area occupied by one cluster is smaller. However, because formally distinct clusters may coalesce, it may be difficult to identify the sets of offspring with any confidence.
Null models: independence of bivariate pattern The univariate patterns of adults and recruits appeared to show a clustered spatial pattern (Fig. 1a,b), and the clusters appeared to overlap to some extent. To verify that the pattern of recruits shows a significant association with the pattern of adults, we used the toroidal shift null model (Goreaud & Pelissier, 2003; Wiegand & Moloney, 2004), which provides a nonparametric way in which to test independence between two patterns. This null model preserves the spatial structures of both patterns, but breaks the dependence between them by moving the entire pattern 2 to a randomly chosen vector. One way of achieving this is to perform simulations that involve random shifts of the whole of one component pattern relative to the other. In practice, a rectangular study region is treated as a torus where the upper and lower edges are connected and the right and left edges are connected.
Null models: bivariate cluster process under the antecedent conditions If the two patterns show a significant association, more specific null models can be tested to describe in more detail the relationship between two clustered patterns. There are two bivariate cluster models that describe the two extreme cases of such a situation. First, we may expect that the adults will be the cluster centres of the recruits and that the recruits will be distributed similarly to a ‘seed shadow’ around the adult plants. This null model is similar to the Thomas process, but assumes that the parent pattern is known. If the parent pattern is itself a Thomas process with parameters ρ1 and σ1, Wiegand et al. (in press) showed that the bivariate pair-correlation function yields:
( Eqn 2)
If the parents are distributed in a random pattern, the last term of Eqn 2, which describes the interaction between parent and offspring patterns, disappears and the pair-correlation function simplifies to g12(r, σ2, ρ2) = 1 + exp(–r2/2σ2)/
Null models: bivariate cluster process with shared parents The second bivariate cluster process is a situation where both adults and recruits follow a Thomas process with parameters ρ1, σ1, ρ2, and σ2 and some of the adult and recruit clusters overlap. Thus, some of the points of the parent patterns of adults and recruits (which define the cluster locations) are shared. The pair-correlation function of this process yields:
where and the intensity of shared parents yields ρs = ρ1ρ2/ρ* (see Supplementary Material Appendix S1). Note that Eqn 3 collapses to Eqn 1 if pattern 1 and pattern 2 are identical (i.e. the univariate case). If none of the parents is shared (i.e. ρs = 0) the pair-correlation function collapses to g12(r) = 1, which is the pair-correlation function of two independent patterns.
Intensity estimate To visualize the spatial structure of the univariate patterns, we estimated the intensity of the patterns using an Epanečnikov kernel recommended by Stoyan & Stoyan (1994), which is defined as:
(d, the distance from a focal point; h, the bandwidth.) For a given location (x, y), the intensity λ(x, y) is constructed by using a moving window with circular shape and radius h around location (x, y) and summing up all points in the circle, but weighting them with the factor eh(d) according to their distance d from the focal location (x, y). We used the estimate of the cluster radius rC = 2σ as the bandwidth. Because all plants of the two populations were mapped and located within our study rectangles, we did not use edge correction to account for points outside the study rectangle.
Parentage analysis To obtain a parent–offspring distance distribution directly, a molecular parentage analysis was conducted using methods outlined in Gerber et al. (2000). A likelihood-based approach was used to detect the most likely parents and parent pairs using the logarithm of the likelihood ratio (log-likelihood ratio (LOD) score) (Meagher & Thompson, 1986; Gerber et al., 2000). After evaluation of all genetically possible parents, offspring were assigned to the parental pair with the highest LOD score. Only highly reliable parent–offspring matches (LOD score > 7.7) were used. Because results from previous analyses (Jacquemyn et al., 2006) indicated that pollen flow was larger than seed flow, the parent nearest to the seedling was considered the mother. This allowed calculation of the distance between mother plants and established seedlings, yielding a histogram of seed establishment distances. All calculations were performed using the famoz software package (Gerber et al., 2000).
Experimental investigation of seed germination Seed germination from transect locations was related to the distance to the nearest individual using Spearman rank correlation.
Univariate analyses; site 1
The Thomas process described the data of site 1 reasonably well (Fig. 2a,b). The parameter estimates for flowering adults were Aρ = 8.6 clusters (A is the area of the study site) and a cluster size 2σ = 4.72 m. For recruits we obtained Aρ = 6.1 clusters and a cluster size 2σ = 4.47 m. The estimate of the intensity of the univariate patterns (Fig. 3a) shows that the adults indeed formed readily recognizable clusters.
To illustrate the cluster structure of the adults, we visually placed nine circles representing the area in which 86% of all ‘offspring’ should be located. For the recruits we also found clearly identifiable clusters (Fig. 3b), but the pattern did not follow the cluster structure as closely as that of the adults. Interestingly, the clusters of recruits and adults overlapped to a large extent.
For scales > 30 cm, the empirical pair-correlation functions were well within the simulation envelops of the fitted cluster process. However, for both recruits and adults there was evidence of an additional small-scale clustering at scales of up to 0.3 m. Because of this additional small-scale clustering which was not accommodated by the Thomas process, the empirical distribution functions G(y) of the distance to the nearest neighbours were close to (or slightly above) the upper confidence limit (i.e. there were several points that had their nearest neighbour closer than predicted by the null model).
Univariate analyses; site 2
The Thomas process described the data of site 2 reasonably well (Fig. 2c,d). The parameters that fitted the data best yielded for adults Aρ = 7.0 clusters and a cluster size 2σ = 4.2 m. For recruits we obtained Aρ = 3.8 clusters and a cluster size 2σ = 4.6 m. The intensity estimates for the adult and recruit patterns show that the clusters coalesced more than at site 1 and that individual clusters were not as clearly identifiable as at site 1. Again, the areas of high intensities of adults and recruits overlapped to a large degree.
For scales > 30 cm, the empirical pair-correlation functions were well within the simulation envelops of the fitted cluster process. However, for both recruits and adults there was evidence of an additional small-scale clustering at scales of up to 0.1 m.
Bivariate analyses; site 1
The toroidal shift null model showed clearly that the patterns for recruits and adults were not independent (Fig. 4a). We therefore proceeded with the cluster processes. The univariate analyses showed that the adults and the recruits both followed a Thomas process with basically the same cluster size and nine and six clusters, respectively. It is therefore unlikely that the spatial relationship between recruits and adults can be described by the bivariate cluster process under the antecedent condition (Eqn 2) where the adults are the cluster centres of the recruits. We therefore proceed with the other bivariate cluster process with shared parents (Eqn 3) where adults and recruits both show a clustered univariate pattern and where some of the clusters are shared. This model was also suggested by our visual reconstruction of the clusters (Fig. 3).
The fit with the bivariate cluster process with shared parents (Eqn 3) yielded estimates of σsum = 2.4 m, which agrees well with the expected value of = 2.3 m based on the parameters of the two univariate component processes. The fit of ρ* yielded ρs = ρ1ρ2/ρ* = 5.9 shared parents; thus all recruits clusters are contained in adult clusters. The visual reconstruction of the clusters (Fig. 3a,b) confirmed this result. The Monte Carlo simulations of the fitted cluster process showed that the data were well within the simulation envelopes (Fig. 4c). There was only aggregation at scales smaller than 0.2 m. Figure 4(e) shows that most of the recruits (83%) were not further away than 2 m from an adult and 96% were located within 5 m of an adult.
Bivariate analyses; site 2
The toroidal shift test showed a weak but significant attraction of recruits around adults (Fig. 4b). We therefore proceeded with the cluster process (Eqn 3). The fit with the model (Eqn 3) yielded an estimate of σsum = 20.9, which agrees well with the expected value of = 22 based on the parameters of the two univariate component processes. The fit of ρ* yielded ρs = ρ1ρ2/ρ* = 6.1 shared parents, which was more than the univariate estimate of 3.8 clusters of the recruits. This discrepancy indicates that the bivariate cluster process with shared parents does not describe the pattern well. Simulations of the fitted process, where the number of shared parents was reduced to 3.8, described the data reasonably well (Fig. 4d); however, the recruits were more clustered around the adults than assumed by the model. This was probably because the distribution of the seven clusters of the adults was not a random pattern as assumed by the Thomas process, but probably somewhat more aggregated (see Fig. 3d). Figure 4(e) shows that none of the recruits was located further away than 2.1 m from an adult.
Genetic diversity and the proportion of polymorphic loci were similar between the sites (Table 1). A total of 64 and 29 seedlings could unambiguously be assigned to parents, resulting in two distance curves. In site 1, parent–offspring distances ranged from 0.21 to 24.67 m (median, 7.04 m; Fig. 5a). In site 2, parent–offspring distances varied between 0.02 and 7.01 m (median, 4.16 m; Fig. 5b) and largely overlapped with the distance distribution from the spatial point pattern analyses.
As a result of management works, one of the transects at site 1 was completely destroyed and no seed packages could be retrieved from this transect. The remaining seed packages were almost entirely located in the large area where no orchids occurred, which did not allow testing of the effect of adult proximity on seed germination success. Nonetheless, the data provided interesting information for interpretation of the observed seedling patterns, as in this area no protocorms at all were found, indicating that conditions were not suitable for germination. However, the few seed packages that were situated in the neighbourhood of adult individuals did contain protocorms, suggesting that suitable microsites for germination occurred near existing individuals. At the second site, all seed packages were retrieved and a broad spectrum of germination stages was observed, ranging from protocorms to seedlings with the onset of the first leaf and roots clearly visible (Fig. 6). As we were only interested in germination success, we did not further discriminate between these stages. The number of successfully germinated seeds within individual seed packages varied between 0 and 12 (mean 2.4). When averaged over sampling points, there was, however, no clear trend with spatial location or with adult density, indicating that local conditions did not limit germination at this site (Fig. 7).
Point pattern analyses
Our analyses provide a clear and coherent picture of the spatial structure of recruits and flowering adults of O. purpurea. Although recruits had fewer clusters than adults (6.1 vs 8.6 at site 1 and 3.8 vs 7.0 at site 2), both showed a clustered spatial distribution with cluster radii of c. 4–5 m and they were both distributed inside a few clusters. At both sites, we were able to visually reconstruct the clusters using an estimate of the intensity of the patterns. The reconstruction agreed well with the results obtained by fitting univariate cluster processes, especially at the large site 1. At the smaller site 2, the adult clusters overlapped and made reconstruction less clear.
The null model (Eqn 2), which assumed that the recruits were directly clustered around the adults in a seed-shadow manner, was implausible although the nearest adult from a recruit was mostly within a distance of 2 m. Under such a null model we would expect a structure with two critical scales of clustering, where the parent clustering defines the first (larger) scale of clustering and the clustering of recruits around adults defines the second smaller scale of clustering (e.g. Wiegand et al., in press). This double-cluster structure should be noticeable in the results for the univariate analysis of the recruits. However, we did not find indications of two scales of clustering (except for a very small-scale clustering); rather, adults and recruits showed approximately the same cluster size.
More likely, however, was the second alternative null model (Eqn 3) of a bivariate pattern where the two component patterns follow a Thomas process. In this null model, the clusters of recruits and adults may partly overlap. If all clusters are shared, a maximal degree of association is reached and if no cluster is shared the patterns are independent. We found that this null model described the pattern found at site 1 very well. All recruits overlapped adult clusters. This finding was also clearly supported by the visual reconstruction of the clusters (Fig. 2a,b). At site 2 the model did not describe the data as well as at site 1, but approximated them reasonably well. The reason for this is that the centres of the adult clusters were probably not a random pattern as assumed for this cluster process. These results confirm earlier findings of Jacquemyn et al. (2006), who found that the extent of spatial genetic structure of recruits was less stringent than that of adults, suggesting overlapping seed shadows and mixing of genotypes within sites suitable for germination. However, it should be noted that this null model (Eqn 3) is descriptive and does not allow us to infer, without additional information, whether the recruits in a given cluster are the offspring of the overlapping adult cluster or whether the clusters form because the habitat is only suitable in the proximity of adults. In the first case, dispersal distances should be short, but in the second case they may considerably exceed the size of the clusters.
Limited seed dispersal?
Overall, both the spatial point pattern analysis and the parentage analysis based on AFLP markers suggested rather limited seed dispersal distances. These results are in relatively good accordance with data obtained for other orchids. For example, most studies investigating spatial genetic structure within terrestrial orchid populations found significant spatial genetic structure, which in most cases was explained by limited seed dispersal (e.g. Caladenia tentaculata, Peakall & Beattie, 1996; Spiranthes spiralis, Machon et al., 2003; Cephalanthera longibracteata, Chung et al., 2004; Liparis makinoana, Chung et al., 2005). Using wind-tunnel dispersal experiments with seeds of the neotropical orchid Brassavola nodosa, Murren & Ellison (1998) found modal dispersal distances of 2, 2.30, 3.30 and 5.90 m, depending on wind velocities. Carey (1998), studying spatial spread of the long-lived orchid Himantoglossum hircinum, found that dispersal distances of a few metres described the expansion of a large population in southern England quite well.
However, the results of the parentage analysis indicated that offspring distances tended to be larger than the cluster sizes and nearest neighbour distances obtained by the point pattern analyses. Although offspring distances derived from the point pattern analysis and parentage analysis showed a fairly good agreement (median offspring distance 4.15 m) in site 2, median offspring dispersal distances in site 1 were larger (median offspring distance 7.04 m), indicating that some of the seedlings growing close to adult plants originate from other, more distant plants. This can be explained by the fact that some flowering individuals did not set any fruit (Jacquemyn et al., 2006) and thus did not contribute to seed production. Nevertheless, microsites around these individuals may be suitable for germination. However, management works at the beginning of this study (August 2001 and 2002) with removal of all above-ground vegetation may have caused accidental long-distance seed dispersal. Hence, the nearest neighbour distances between seedling and adult in the point pattern analysis do not necessarily coincide with the distances between mother plant and offspring. Alternatively, given the 4-yr time lag between seed production and seedling emergence from the soil, many parents that contributed to recruits could have died before the 2004 sample collection took place. However, our demographic analyses showed that mortality is very low (H. Jacquemyn & R. Brys, unpublished), so that the impact of loss of potential parent plants on dispersal distances estimated from genetic markers should be minimal.
Spatial distribution patterns of recruits not only depend on the mechanisms of seed dispersal, but are equally affected by the probability of seed germination and further establishment of seedlings. We have shown that seed germination rates differed among sites. Although our results for site 1 were only partial because of the loss of a large number of seed packages, the results of the seed germination experiment indicated that the clustered distribution of both adults and seedlings at site 1 was to a large extent attributable to the inability of seeds to germinate in a large part of this population. We can only speculate on the reasons why no protocorms were observed in this area. It may be that the appropriate fungi were missing as a result of unsuitable abiotic conditions or the lack of a suitable substrate. At any rate, our results confirm earlier findings of Diez (2007), who also found a lack of germination in sites that had no adult individuals of Goodyera pubescens. However, the few seed packages that could be retrieved in the immediate neighbourhood of individuals did contain protocorms, suggesting a declining probability of germination with distance from adults and thus positive density dependence for recruitment (Diez, 2007). These results also explain why all recruits almost perfectly overlapped adult clusters in site 1. In site 2, however, this was not the case (there was seed germination everywhere), which may explain why not all seedlings overlapped adult clusters. The exact reasons for the pronounced differences in seed germination rates between the two sites remain unclear. Although both populations have been known for many decades, it might be that the effects of habitat deterioration were more pronounced in site 1 than in site 2, the former being overgrown by Hedera helix for many years, whereas the latter suffered only a short abandonment of the traditional management regime.
As early as the 19th century, Darwin (1877) recognized that, of all life history stages, successful germination and seedling establishment are the most critical in orchids. A better understanding of the multiple factors that affect patterns of seedling recruitment in orchids therefore remains a central goal in orchid biology. In this study, we used spatial point pattern analyses, parentage analyses and seed germination experiments to describe spatial variation in seedling recruitment in the terrestrial orchid O. purpurea. Seed dispersal is limited to a few metres from the mother plant, whereas the availability of suitable germination conditions may vary strongly from one site to the next and can greatly affect spatial patterns of recruitment. Because of a time lag of 3–4 yr between seed dispersal and actual recruitment, and differences in flowering frequencies of adult plants, interpretation of recruitment patterns using point patterns analyses ideally should take into account the demographic properties of orchid populations. Overall, our results indicate that, in order to fully comprehend recruitment dynamics of orchid species, a combination of genetic, demographic and experimental investigations is definitely needed.
This research was funded by the Fund for Scientific Research (FWO). Three anonymous referees provided very useful comments that significantly improved this manuscript.