Scientia Terrae Research Institute, B-2860 Sint-Katelijne-Waver, Belgium
Laboratory for Process Microbial Ecology and Bioinspirational Management, Lessius University College, Campus De Nayer, Consortium for Industrial Microbiology and Biotechnology (CIMB), Department of Microbial and Molecular Systems, KU Leuven Association, B-2860 Sint-Katelijne-Waver, Belgium
•Nonrandom species–species associations may arise from a range of factors, including localized dispersal, intra- and interspecific interactions and heterogeneous environmental conditions. Because seed germination and establishment in orchids are critically dependent upon the availability of suitable mycorrhizal fungi, species–species associations in orchids may reflect associations with mycorrhizal fungi.
•To test this hypothesis, we examined spatial association patterns, mycorrhizal associations and germination success in a hybrid zone containing three species of the genus Orchis (Orchis anthropophora, Orchis militaris and Orchis purpurea).
•Hybridization occurred predominantly between O. purpurea and O. militaris. The spatial distribution patterns of most pure species and hybrids were independent from each other, except that of O. purpurea and its hybrids. The fungal community composition of established individuals differed significantly between pure species, but not between hybrids and O. purpurea. Seed germination experiments using pure seeds showed that the highest number of protocorms were found in regions where adult individuals were most abundant. In the case of hybrid seeds, germination was restricted to areas where the mother plant was most abundant.
•Overall, these results suggest that the observed nonrandom spatial distribution of both pure and hybrid plants is dependent on the contingencies of the spatial distribution of suitable mycorrhizal fungi.
One of the central goals of ecology is to understand the strength and context dependence of species interactions and how they mediate the distribution and abundance of organisms in natural environments (Agrawal et al., 2007). In plants, it has been shown that significant species–species associations may be common (Coomes et al., 1999; Wiegand et al., 2007) and may arise from a range of processes, including localized seed dispersal, intra-specific and inter-specific interactions, and local heterogeneity in growth conditions (e.g. topography, resource supply or background vegetation). In orchids, because successful germination is critically dependent upon mycorrhizal fungi (Smith & Read, 2008; Rasmussen & Rasmussen, 2009), recruitment and establishment are likely to be sensitive to the contingencies of the spatial distribution and abundance of mycorrhizal fungi (Diez, 2007; McCormick et al., 2009).
However, at present little is known about the actual distribution of mycorrhizal fungi in natural orchid populations and how it affects spatial patterns of recruitment and establishment of orchids (Otero & Flanagan, 2006). In some orchid species, successful seed germination was shown to be directly related to the distance from the maternal plant (McKendrick et al., 2000, 2002; Batty et al., 2001; Diez, 2007), suggesting a declining abundance of essential mycorrhizal partners further away from adult plants, whereas no such relationship was observed in other species (Masuhara & Katsuya, 1994; McKendrick et al., 2000). In addition, there is a growing consensus that, despite the huge amounts of minute, dust-like seeds that are produced each year, seed dispersal tends to be limited (Jersáková & Malinová, 2007), contributing to the significant spatial clustering that has often been observed in natural orchid populations (Chung et al., 2005a; Jacquemyn et al., 2007, 2009; McCormick et al., 2009). Thus, when multiple orchid species occur at a single site, differences in the community of mycorrhizal partners combined with a patchy spatial distribution of fungal lineages and restricted seed dispersal can be expected to result in different species aggregation patterns, ranging from complete overlap to significant segregation (Wiegand et al., 2007). The extent to which seed dispersal and/or fungal availability determines species–species associations in orchids remains, however, largely unknown.
A specific situation arises when orchid species belong to the same genus and cross to form hybrids. In this case, the frequency of hybridization and spatial aggregation patterns of hybrids and pure species will depend on the strength and direction of reproductive barriers between these species, the spatial distribution of maternal plants, seed dispersal distances, and mycorrhizal specificity and abundance. Although numerous studies have investigated the extent of hybridization in orchid species (e.g. Cozzolino et al., 2006; Moccia et al., 2007; Pellegrino et al., 2009; Pinheiro et al., 2010), there are only a few studies that have looked at the spatial components of hybridization (e.g. Chung et al., 2005b). It can be hypothesized that, if hybrid plants are able to associate with the mycorrhizal fungi of one of the parents, as has been shown for Caladenia (Hollick et al., 2005), hybrid plants in the field should display spatial distribution patterns that strongly coincide with that of one of the parents. If, however, hybrid plants associate with the mycorrhizal fungi of both parental species, their spatial distribution pattern should be positively related to that of both parents. Finally, if hybrid plants associate with other fungal strains than the parents, it is possible that hybrid offspring occupy different locations within the population and that their distribution therefore shows little association with that of the parental species. In this case, one can expect that hybrids are spatially independent or segregated from parental plants.
Here, we studied the spatial extent of hybridization between three closely related species of the genus Orchis: O. anthropophora, O. militaris, and O. purpurea. Hybridization between all three species has been described previously (Kretzschmar et al., 2007), but confirmation of hybrid origin based on comparative analyses of molecular and morphological data is still limited. More specifically, we aimed (1) to assess the extent of hybridization at a site where the three species co-occur; (2) to describe the genetic composition of putative hybrids; (3) to assess spatial distribution patterns of established pure and hybrid individuals and (4) to examine the potential role of mycorrhizal associations in determining spatial patterns of pure and hybrid individuals. To address points (3) and (4), seed germination experiments were combined with spatial point pattern analyses that test for departures from spatial independence in all bivariate patterns of the four types of plants (i.e. pure O. anthropophora, pure O. purpurea, pure O. militaris and hybrids). In case hybrid individuals showed a positive association with their parents, we also tested if this association was consistent with the hypothesis that the hybrids were associated with a fungal field of influence from their parents. Seed germination experiments using seed packages allowed us to assess spatial variation in seed germination, and thus to discern the importance of mycorrhizal availability from seed dispersal in affecting spatial patterns of established individuals. Finally, assessment of mycorrhizal associations in established pure and hybrid plants using DNA array technology allowed us to test the hypothesis that mycorrhizal interactions have contributed to the observed species–species association patterns.
Materials and Methods
The three species studied here belong to the anthropomorphic group of the genus Orchis (Bateman et al., 2003) (subgenus Orchis sensuKretzschmar et al., 2007). Although O. anthropophora displays pronounced differences in flower morphology, recent phylogenetic analyses based on DNA sequences (Bateman et al., 2003) have placed the three species within the same genus. Before, O. anthropophora was assigned to the genus Aceras, of which it was the sole member (Kretzschmar et al., 2007). Both O. anthropophora and O. purpurea are species with a clear Mediterranean distribution, whereas O. militaris has a more continental distribution, occurring from the Atlantic coast to Mongolia (Kretzschmar et al., 2007). In general, O. anthropophora and O. purpurea like warmer conditions for growth and survival than O. militaris. However, in grazed dry meadows and calcareous grasslands they can often be found growing together.
All three species flower in about the same period, from the beginning of May to the end of May/beginning of June. The species are allogamous and thus dependent on pollinators to guarantee successful pollination. Little, however, is known about the pollinators of the species (Van der Cingel, 1995). Orchis anthropophora is pollinated by at least two species of sawflies, Tenthredopsis sp. and Arge thoracia (both Hymenoptera; Tenthredinidae), and by two beetle species, Cantharis rustica (Coleoptera: Cantharidae) and Cidnopus pilosus (Coleoptera: Elateridae) (Schatz, 2006). In the western region of France (Brittany), the beetle Isomira murina (Coleoptera: Elateridae) has been reported as the sole pollinator of O. anthropophora. No reports on confirmed pollinators are available for O. militaris and O. purpurea, but our own observations and observations included in Farrell (1985) suggest that both species are regularly visited by bumblebees, bees and occasionally butterflies, which may serve as occasional pollinators for these species.
The study was conducted in May 2008 in a calcareous grassland located in a local nature reserve near Eben-Emael (Belgium) (50°46′N 5°40′E; 210 m asl). This site had been severely overgrown by shrubs and trees, but was recently restored by cutting down all trees and shrubs. As a result, individuals of O. anthropophora, O. militaris and O. purpurea grew closely together, providing possibilities for hybridization to occur at the study site. Within this site, a 20 × 30 m plot was established in the most homogenous part of the site, and each flowering individual within this plot was mapped to the nearest cm using high-precision GPS (Trimble Navigation Ltd., Sunnyvale, CA, USA). A few plants were located in the immediate neighborhood of the plot. In this way, a total of 315 individuals were mapped (Fig. 1).
Hybridization in the field
To assess the extent of hybridization in the field, we used amplified fragment length polymorphism (AFLP) markers. Young leaf material was collected for each mapped plant and immediately frozen in liquid nitrogen for subsequent genotyping. Before DNA extraction, leaf 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 the Nucleospin® 96 Plant Kit (Machery-Nagel, GmbH & Co.KG, Düren, Germany). DNA concentrations were estimated using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, USA).
AFLP analysis was carried out according to Vos et al. (1995), using commercial kits and following the protocol of Roldán-Ruiz et al. (2000). Briefly, the enzymes EcoRI and MseI were used for DNA digestion. Each individual plant was fingerprinted with four primer combinations: EcoRI-ACGT/MseI-CGA, EcoRI-ACCC/MseI-CTA, EcoRI-AGG/MseI-CTGG and EcoRI-AGG/MseI-CTAG. Fragment separation and detection took place on 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 × TBE). GeneScan 500 ROX-labeled 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. Monomorphic markers were excluded from all further analyses.
To assess reproducibility, three independent DNA extractions were carried out for 10 individuals and AFLP fingerprints were generated for each replicate (a total of 120 AFLP fingerprints). Mean reproducibility values (calculated as the percentage of markers that were identical in the three repeats) were high (98.7, 98.5, 98.8, and 99.5%, for primer combinations EcoRI-ACGT/MseI-CGA, EcoRI-ACCC/MseI-CTA, EcoRI-AGG/MseI-CTGG and EcoRI-AGG/MseI-CTAG, respectively), demonstrating the robustness of the assay.
Root samples of both pure and hybrid plants were collected to compare mycorrhizal associations between pure and hybrid plants. The hybrid status of individuals was determined based on the AFLP profiles. Because hybridization involving O. anthropophora was very rare (see Results), only hybrids between O. purpurea and O. militaris were collected. To minimize damage to the population, root parts of five individuals per species were collected, thus yielding 20 individuals in total. To detect and identify mycorrhizal fungi, the same procedures were followed as in Jacquemyn et al. (2011a,b). Briefly, roots were surface-sterilized (30-s submergence in 1% sodium hypochlorite, followed by three 30-s rinse steps in sterile distilled water) and microscopically checked for mycorrhizal colonization. Subsequently, DNA was extracted from 0.5 g of mycorrhizally colonized root pieces per plant using the UltraClean Plant DNA Isolation Kit as described by the manufacturer (Mo Bio Laboratories Inc., Solana Beach, CA, USA). The mycorrhizal fungi colonizing the roots were then assessed using a basidiomycete-centered DNA array enabling the simultaneous detection and identification of previously detected fungal operational taxonomic units (OTUs) defined at 97% internal transcribed spacer (ITS) sequence identity (see Lievens et al., 2010 and Jacquemyn et al., 2010 for more details). In total, the array allowed for a screening of 23 fungal OTUs (Jacquemyn et al., 2011b).
To investigate spatial variation in seed germination, we used both pure and hybrid seeds from previous experiments (Jacquemyn et al., 2011a). In May 2008, both hybrid and pure seeds were created for each species by crossing 30 plants (10 plants per treatment) with pollinia of one of the two other species and with pollinia from the same species (see Jacquemyn et al., 2011a for more details on the pollination experiment). When seeds were ripe (mid-August 2008), fruits were harvested and seeds were buried using the modified seed package method of Rasmussen & Whigham (1993). Per seed package, c. 250 seeds were placed within a square of 53-μm mesh phytoplankton netting, enclosed within a Polaroid slide mount. Packages were buried into the soil along four transects within the 20 × 30 m study plot. Transects were 7.5 m apart, and consisted of seven sample points per transect that were 2.5 m apart. At each point, nine seed packets (one for each cross) were placed vertically in the ground, leading to a total of 252 seed packages which were left in the ground for c. 2 yr. In April 2010, seed packages were retrieved, gently washed and maintained moist in paper towel for 1 d until examination. Packages were first rinsed with tap water, and then opened with a mini-knife under a dissecting microscope connected to a digital camera. As orchid seed germination stages can be variable (Ramsay et al., 1986), germination was considered when a seed had developed into the protocorm stage, clear signs of mycorrhiza formation were present and the leaf primordia had developed (stage 3 sensuRamsay et al., 1986). For each package, the successful development of protocorms was determined by counting the number of protocorms.
Hybridization in the field To assess the extent of hybridization, the AFLP data were subjected to Bayesian assignment analyses using the methods implemented in structure version 2.2 (Pritchard et al., 2000; Falush et al., 2007). structure uses a model-based clustering method to assign individuals to groups in which deviations from Hardy–Weinberg equilibrium and linkage equilibrium are minimized. Individuals assigned to two sources with nontrivial probabilities are putative hybrids. In the structure model, the posterior probability (q) describes the proportion of an individual genotype originating from each of K categories. In our case, setting K =3 corresponds to the assumption of three species contributing to the gene pool of the sample. The AFLP data were entered as Ploidy = 2 for all individuals, with the second line as missing data. Calculations were carried out under the admixture model of ancestry assuming correlated allele frequencies. We used a burn-in of 500 000 steps followed by 1 000 000 iterations, after verifying that the results did not vary significantly across multiple runs and with longer cycles of burn-in and iterations.
Spatial point pattern analyses We used techniques of spatial point pattern analysis for two purposes. In a first analysis (‘testing for independence’) we tested for positive spatial association in the various bivariate patterns between the different orchid species and the hybrids by assessing departures from the null model of independence (Dixon, 2002). In a second analysis (‘spatial pattern of hybrids in relation to potential field of mycorrhizal fungi’) we tested the four specific hypotheses formulated in the Introduction for the spatial pattern of hybrid plants in relation to their parents, given that these patterns showed a positive association. In this analysis we make the assumption that for each parent plant a ‘field of a mycorrhizal fungal influence’ exists which declines with distance from each parent plant. In the null model, the hybrid plants are then randomly relocated in space but retained with a probability proportionally to the superposed field of a mycorrhizal fungal influence of all parent plants.
Summary characteristics To quantify the spatial structures we used two summary characteristics designed for completely mapped bivariate point patterns. The first is the pair correlation function gij(r) (Stoyan & Stoyan, 1994; Wiegand & Moloney, 2004; Illian et al., 2008). The quantity λjgij(r) can be interpreted as the density of type j points at distance r from type i points, where λj is the density of type j points in the study area (= the number of plants of pattern j divided by the area of the study plot). To estimate the pair correlation function for bivariate patterns we used the extension of the estimators for univariate patterns proposed by Ohser & Mucklich (2000), which are detailed in Illian et al. (2008: p. 230 and pp. 352–354), together with adapted estimators of the intensity function (Illian et al., 2008, p. 194). Important additional information is provided by the distribution function Dij(r) of the distances r from type i points to their nearest type j neighbor (Illian et al., 2008). Nearest neighbor statistics are ‘short-sighted’ and sense only the immediate neighborhood of the typical point, which makes them especially sensitive to local cluster structures. To estimate Dij(r) we used the extension of the estimator suggested by Hanisch (1984) and Stoyan (2006), which is detailed in Illian et al. (2008, p. 322).
Null model for independence We tested for independence of the 12 possible bivariate patterns, conditionally on the local scale defined by our plot. A test for independence of a bivariate pattern must be conditional on the structure of the two univariate component patterns (i.e. the univariate structures must be conserved), but any relationship between the component patterns needs to be removed (Dixon, 2002; Goreaud & Pélissier, 2003). We achieved this by using techniques of pattern reconstruction (Tscheschel & Stoyan, 2006; Illian et al., 2008). This technique generates patterns with approximately the same spatial structure as the observed univariate patterns and can therefore be used as a null model for independence. Pattern reconstruction is described in detail in Tscheschel & Stoyan (2006) and Illian et al. (2008: p. 407ff). Basically, an annealing algorithm is used to minimize the differences in several summary statistics between the observed and the reconstructed patterns (in our case up to distances of 6 m). Thus, the reconstructed pattern shows up to this distance the same spatial structure as the observed pattern. We used the pair correlation function, the K-function (Ripley, 1981), the distribution function of the distances to the nearest neighbor (Diggle, 2003) and the spherical contact distribution (Illian et al., 2008) for pattern reconstruction. The spherical contact distribution can be estimated as distribution function of the distances to the nearest neighbor calculated for a bivariate pattern where the focal pattern is a ‘test pattern’ with points located on the nodes of a regular grid and the second pattern is the pattern to be reconstructed. This function measures basically the gaps in the pattern (Diggle, 2003; Illian et al., 2008). Because these summary characteristics capture different aspects of spatial structure (i.e. neighborhood density, nearest neighbors and gaps) the reconstructed patterns approximate the small to intermediate scale (i.e. up to 6 m) spatial structure of the observed patterns well (Supporting Information Fig. S1).
Spatial pattern of hybrids in relation to potential field of mycorrhizal fungi We tested four competing hypotheses concerning the spatial pattern of hybrid plants in relation to their parents: hybrid plants are associated with the mycorrhizal fungi of (1) one of the parents, (2) both parents, (3) fungal strains other than the parents, or (4) all studied orchid species. Assuming that successful germination, mediated by mycorrhizal fungi, is directly related to the distance from the maternal plant (Diez, 2007), a field of a mycorrhizal fungal influence was derived. This field is the additive superposition of the fields of influence of individual plants, each of them modeled as a two-dimensional normal distribution with parameter σ and centered in the individual. Following the four hypotheses, we derived (1a) one field for O. militaris, (1b) one for O. purpurea, (2) one joined field for O. purpurea and O. militaris, (3) one field for O. anthropophora, and finally (4) one field for all three species together. The joined fields were constructed additively from the fields of the individual species.
To test these hypotheses, we used stochastic point process models that redistribute the individuals of the hybrid plants following a heterogeneous Poisson process (Wiegand & Moloney, 2004) where the field of a mycorrhizal partner serves as an intensity function. In other words, the probability that a given location receives a hybrid plant was proportional to the field of a mycorrhizal fungal influence, but hybrid plant locations were otherwise independent. We then compared the observed summary characteristics of the bivariate pattern of hybrid plants relative to the parent plants (of the respective field of mycorrhizal fungal influence) with that simulated by the different point process models corresponding to our hypotheses. Because the exact distance dependence of the mycorrhizal fungi field is unknown we repeated our analyses for different parameters σ to find the value of σ and the hypothesis that describes the observed association pattern between hybrids and orchids best. To compare observed and simulated patterns we used pair correlation functions gij(r) and distribution functions Dij(r) to the nearest neighbor for bivariate patterns as summary characteristics.
Significance tests To assess the fit of a given point process model, we generated 999 (or 199) simulated data sets and used the 25th (or 5th) lowest and highest values of our test statistics (i.e. gij(r) or Dij(r)) at distance r as simulation envelopes to depict the range of possible values under the point process model. The simulation envelopes provide approx. 5% intervals but are prone to type I error because we conduct with the same data multiple tests at various distances (Diggle, 2003; Loosmore & Ford, 2006; Illian et al., 2008). To assess the overall fit of a point process model, we therefore used a goodness-of-fit (GoF) test proposed by Diggle (2003) and Loosmore & Ford (2006). This curve-wise test reduces the distance-dependent information of a given summary statistic for the observed data (k = 0) and the simulated data (k = 1, ..., 199 or 999) into one single test statistic uk. We used a Cramer–von Mises type statistic that compared the entire range of a summary statistic (0–3 m in our case) with that expected under the null hypothesis. The test then calculates the rank of the observed uk (k = 0) within all uk. For example, if the rank of u0 is smaller than 950 (990) the point process model cannot be distinguished from the data at a 5% (1%) level.
The presence–absence of 23 previously detected fungal OTUs (Jacquemyn et al., 2011b) was assessed for each of the sampled individuals. Based on presence–absence data of fungal virtual taxa in each of the sampled individuals, the fungal community composition associating with the different orchid species was analysed by detrended correspondence analysis (DCA). A species-label reallocation scheme using the multiple response permutation procedures (MRPP; Biondini et al., 1988) test was implemented to test the hypothesis that overall fungal composition differed among species. In the case of significant differences, pairwise comparisons were performed to see whether fungal composition differed between particular species. All analyses were performed using pc-ord for Windows, version 5 (MjM Software, Gleneden Beach, OR, USA).
To investigate whether the probability of germination was affected by the presence of conspecific individuals, for each seed package the distance to the nearest individual of the same species in the case of pure seeds or the mother species in the case of hybrid seeds was calculated based on the geographic positions of both plants and seed packages. Additionally, we also calculated for each seed package the distance to the nearest individual regardless of whether it was the same or one of the other species. We used the distance to the mother species in the case of hybrid seeds, as previous research showed that mycorrhizal fungi associating with hybrid seeds resembled those of the mother species more than those of the father species (Jacquemyn et al., 2011a). Presence–absence of protocorms in seed packages was then related to these distance using generalized linear mixed models (GLMMs) using the glmer() function from the lme4 package in R (Bates et al., 2009). In the case of overdispersion, a categorical variable with a different level for each location as a random effect was added to the model (Warton & Hui, 2011). Both the distance to the nearest conspecific individual or the nearest individual of the mother species (for pure and hybrid seeds, respectively) and the distance to the nearest individual were included in these models. However, because inclusion of the distance to the nearest individual did not reveal any significant effects, this variable was not further investigated.
Hybridization in the field
The four primer combinations generated a total of 168 polymorphic bands. Each individual displayed a unique banding pattern. Analyses using structure showed that hybridization between O. anthropophora and the other two species was limited, but that frequent hybridization between O. purpurea and O. militaris had occurred in this population (Fig. S2). Assuming a threshold Tq = 0.90 (Pritchard et al., 2000; Vähä & Primmer, 2006; Burgarella et al., 2009), individuals with q ≥ Tq are assigned to the purebred category and individuals with q < Tq are assigned to the hybrid category. Using this threshold, it was found that the majority of O. anthropophora plants (171 plants or 54.3% of all sampled plants) were purebreds, and that only six (1.9% of all sampled plants) individuals showed signs of hybridization, two involving hybridization with O. purpurea and four with O. militaris (Fig. S2). Orchis militaris and O. purpurea, in contrast, hybridized frequently. Analyses of morphometric data confirmed the hybrid status of hybrid plants (H. Jacquemyn, unpublished results). Nineteen (6.0%) individuals were assigned to pure O. militaris, 78 (24.8%) individuals were assigned to pure O. purpurea and the remaining 41 (13.0%) plants were considered as hybrids between O. purpurea and O. militaris.
Spatial point pattern analysis
Bivariate species associations At the small neighborhood of 20 cm, we only found significant (and symmetric) positive association of O. purpurea with O. militaris and with the hybrids; all other bivariate relationships did not depart from the null model of independence. The positive small-scale association between O. purpurea and the hybrids is obvious from the species map (Fig. 1) and persists also into the larger neighborhoods of 1 and 3 m. In general, however, the GoF test indicated independence for all pairs except the two pairs between O. purpurea and the hybrids (Table S1).
Spatial pattern of hybrids in relation to mycorrhizal fungi Fig. 1 suggests that hybrid plants were not related to a potential field of a mycorrhizal fungal influence of O. militaris (Fig. 1, Table 1). Consequently, the GoF test yielded independence between the two patterns (Table S1), which indicates that the observed association could arise just by chance. Furthermore, the GoF test (and visual inspection of the data) indicated a strong positive spatial association of the hybrids of O. purpurea and O. militaris to O. purpurea individuals. A test of hypothesis 1b with the heterogeneous Poisson null model showed that the spatial pattern of hybrid plants can be explained well by a point process model where the field of a mycorrhizal fungal influence was modeled as a normal distribution centered in the O. purpurea individuals with parameter σ = 0.3 m (Table 1, Fig. 2b). The point process model with parameter σ = 0.2 m produces too strong an attraction of hybrid plants around O. purpurea individuals (cf. expected (gray line) with observed pair correlation function (disks) in Fig. 2a), whereas the point process model with parameter σ = 0.4 m produces too weak an attraction (Fig. 2c). Comparison of the expected and observed pair correlation functions for the point process model with parameter σ = 0.3 m nevertheless shows that the normal kernel function does not provide a perfect fit; the ‘real’ kernel function shows a stronger attraction at small distances, which, however, declines somewhat more quickly than the normal distribution (Fig. 2b). The distribution function of the distances to the nearest neighbor agrees well for smaller distances (i.e. r <1 m), but intermediate nearest neighbor distances are somewhat less probable than suggested by the point process model (inset in Fig. 2b). This indicates that some of the hybrid plants occurred further away from O. purpurea individuals than expected in the point process model. Inspection of Fig. 1 shows that there are a few hybrid plants located close to the O. militaris cluster.
Table 1. Rank of goodness-of-fit (GoF) test for the different hypotheses concerning the spatial pattern of hybrid plants in relation to their parents conducted over distances up to 0.5 m
If the rank is < 950 (990), the point process model is consistent with the data for this measure at the 5% (1%) level. In the different hypotheses, hybrid plants were associated with fields of influence of the mycorrhizal fungi of (1a) Orchis militaris, (1b) Orchis purpurea, (2) O. purpurea and O. militaris, (3) O. anthropophora, and (4) all three species together. Values in bold denote weak or non-significant (P > 0.01) differences between the null model and the data.
The GoF test also indicated a strong positive association between the patterns of both parents and the hybrids (Table S1), and using individuals of both parents for construction of the field of a mycorrhizal fungal influence (hypothesis 2) yielded results similar to those for hypothesis 1b (Fig. 2e; Table 1). Finally, hypotheses 3 and 4 were clearly rejected (Table 1), which was not unexpected given the data displayed in Fig. 1.
A total of nine different fungal OTUs were detected at the study site (Table S2). The phylogenetic relationships between these OTUs were presented previously (Lievens et al., 2010; Jacquemyn et al., 2011b). As a consequence of the rarity of hybrids involving O. anthropophora, no hybrids between this species and the other two species were sampled. Although plants sometimes associated with several fungal lineages at the same time, fungal composition differed between the three species and hybrids (Fig. 3). The first axis of the DCA (explaining 46.3% of the total variation) separated O. militaris from O. anthropophora, whereas the second axis (27.9% of the total variation) distinguished O. purpurea and hybrids from O. militaris and O. anthropophora. The average within-group distance (δobs = 0.394) was significantly smaller (P =0.0001) than the value based on random reallocation of groups (δexp = 0.546), yielding a chance-corrected within-group agreement A =0.278. This discrimination was mainly attributable to OTU1, which was not found in O. purpurea and hybrids, and OTU4, which was found in all sampled individuals of O. purpurea and hybrids. OTU6 and OTU7, in contrast, were the dominant fungal lineages in O. militaris and O. anthropophora, respectively, but did not occur in O. purpurea. Multiple comparisons using MRPP showed that fungal composition of all species combinations was significantly different at α = 0.05, except for O. purpurea and hybrid specimens (T =0.78, P =0.77) (Table 2).
Table 2. Pairwise tests using multiple response permutation procedures (MRPP) comparing differences in fungal community composition between Orchis anthropophora, Orchis militaris, Orchis purpurea and O. purpurea × O. militaris hybrids
Note that P values were not corrected for multiple comparisons. T, test statistic that describes the separation between the groups; A, chance-corrected within-group agreement.
O. anthropophora – O. militaris
O. anthropophora – O. purpurea
O. anthropophora– hybrids
O. militaris – O. purpurea
O. militaris– hybrids
O. purpurea– hybrids
The probability of germination was significantly related to the distance to the nearest conspecific plant in four out of nine crosses (Table 3). Strong effects were observed for both pure O. anthropophora and O. militaris seeds, whereas the probability of germination of O. purpurea seeds was not related to the location of conspecific plants. When plotting the number of protocorms observed in seed packages in the function of the location in the plot, it was found that seeds of O. anthropophora germinated predominantly on the left-hand side and in the middle of the plot, but no germination was observed on the right-hand side, where the species was absent (Fig. 4a). Similarly, pure O. militaris seeds germinated predominantly in the cluster of O. militaris plants at the lower left-hand side of the plot, whereas in the central part of the plot, where the species was completely lacking, no germination was observed (Fig. 4a). Seeds originating from crosses between O. militaris(♀) and O. purpurea(♂) also showed the highest germination on the lower left-hand side of the plot, where O. militaris was most abundant. Although germination of O. purpurea(♀) × O. militaris(♂) seeds was found throughout most of the population except in the central-left part, the highest densities of protocorms were observed on the right-hand side of the plot (Fig. 4b). Similarly, although the probability of germination was not significantly related to spatial location for O. purpurea seeds (Table 3), protocorm densities were highest on the right-hand side of the plot (Fig. 4a), where the species was most abundant.
Table 3. Results of the seed germination experiment
Equations give the probability of seed germination (p) in relation to the distance of the nearest conspecific plant (x). Parameters were obtained using logistic regression analyses. Values in bold denote weak or non-significant (P > 0.01) differences between the null model and the data.
A, Orchis anthropophora; M, Orchis militaris; P, Orchis purpurea.
Logit (p) = 2.705 − 0.563x
Logit (p) = 1.114 − 0.014x
Logit (p) = 0.333 − 0.031x
Logit (p) = 1.333 − 0.328x
Logit (p) = 1.779 − 0.344x
Logit (p) = 2.686 − 0.333x
Logit (p) = − 0.019 + 1.459x
Logit (p) = 1.799 − 0.277x
Logit (p) = − 0.014 + 1.459x
Hybridization in the field
Hybridization between O. anthropophora on the one hand and O. purpurea or O. militaris on the other was only found in six individuals, indicating that hybridization between these species was limited. By contrast, AFLP analysis clearly showed that hybridization between O. purpurea and O. militaris had occurred very frequently in this population. These results are consistent with the findings of Kretzschmar et al. (2007), who stated that hybridization between the three investigated species has been reported before, but that hybrids between O. anthropophora and the other two species tend to be very rare in natural populations. In contrast, O. militaris and O. purpurea were reported to hybridize to such an extent that hybrid swarms arise and the parental species may ultimately disappear from the population (Kretzschmar et al., 2007). Our findings also corroborate earlier investigations of prezygotic and postzygotic reproductive isolation between O. anthropophora, O. militaris and O. purpurea. Experimental crosses revealed that in none of the crosses was reproductive isolation total, and that hybridization between the investigated species should be possible in natural populations (Jacquemyn et al., 2011a,b). However, consistent with the patterns of hybridization observed here, significant differences in the strength of reproductive isolation were found. Using data on both pre- (pollinator sharing) and post-mating barriers (fruit set, seed viability and seed germination), the average reproductive isolation (RI) between O. militaris and O. purpurea (RI = 0.55) was lower than that between O. purpurea and O. anthropophora (0.88), and O. militaris and O. anthropophora (0.79).
Spatial associations of species and hybrids
Results of the spatial point pattern analyses showed that the three species showed no spatial association with each other (Fig. 1, Table S1). Visual inspection indicated segregation among the three species, but when a given species j was randomly placed relative to another species i (but with the univariate spatial structure of species j retained thanks to the pattern reconstruction algorithm) many of the random patterns showed similar segregation patterns. Thus, the appearance of segregation could arise by chance alone. However, hybrids between O. purpurea and O. militaris showed a positive association in their distribution with one of the parental species (O. purpurea), but not with the other (O. militaris). Similar results have been reported by Chung et al. (2005b), who investigated patterns of hybridization in two Liparis species. Consistent with the results presented here, they found that the parental species (Liparis kumokiri and Liparis makinoana) were spatially segregated, but the spatial distribution of hybrids was associated with only one of the parental species (L. kumokiri).
Mycorrhizal associations of species and hybrids
Investigation of mycorrhizal associations in both pure and hybrid individuals showed that, although all three species associated with a relatively broad range of mycorrhizal fungi, the dominant fungal lineage differed between pure species. When the total community was considered, fungal composition was also significantly different between pure species (Table 2, Fig. 3). However, the dominant fungal lineage and community composition of fungi associating with hybrids were nearly identical to those of fungi associating with O. purpurea. Possibly, this can be explained by the fact that, genotypically, most hybrids were more skewed toward O. purpurea than toward O. militaris. Similar results were obtained in the case of Liparis (Chung et al., 2005b), where hybrids were also, genotypically, more skewed to the parent with which they were spatially associated. These results thus suggest that the genomic composition of hybrid plants may affect preferences toward mycorrhizal partners and therefore the spatial distribution patterns of hybrid plants relative to that of pure species.
Spatial variation in germination success
Further evidence for the hypothesis that mycorrhizal fungi contributed to the observed species–species associations comes from the seed germination experiment. This experiment showed that the germination success of pure seeds was significantly dependent on the distance to the nearest conspecific plant in at least two out of three species. Moreover, protocorm densities in seed packages were almost invariably highest in regions where the species was most abundant, indicating that germination is spatially restricted to regions where the species occurs. These results confirm those of earlier studies that showed that successful seed germination was directly related to the distance from the maternal plant (McKendrick et al., 2000, 2002; Batty et al., 2001; Diez, 2007).
Interestingly, there was also large spatial variation in the germination success of hybrid seeds depending on the identity of the mother plant. Hybrid seeds originating from controlled crosses with O. militaris as mother plant and O. purpurea as the father plant almost exclusively germinated in the region where adult O. militaris plants were most abundant. Similarly, seeds originating from crosses with O. purpurea as mother plant and O. militaris as father plant showed the highest germination success in regions where O. purpurea was most abundant. Although the mechanisms of cellular signaling among orchid seedlings and fungal partners remain to be investigated, these results suggest that there may be a substantial maternal impact on establishment success of hybrid seeds that is probably linked to patterns of mycorrhizal specificity of the parental species. Further, maternal differences in seed germination can be expected to affect the genetic make-up of hybrid zones, as significant associations between hybrid plants and one of the parental plants are likely to increase the possibility of gene introgression.
The results of the seed germination experiment point to the importance of mycorrhizal associations in affecting species–species associations in this study. However, it is likely that restricted seed dispersal has contributed to the observed aggregation patterns as well. Previous research on O. purpurea, for example, using molecular markers and parentage analyses has shown that seed dispersal is limited to a few meters from parental plants (Jacquemyn et al., 2006, 2007). However, the importance of restricted seed dispersal is probably limited compared with that of germination, as most parts of the site where successful germination was possible were already occupied by the different species.
Our data have demonstrated significant species–species associations in a hybrid zone containing three orchid species. Seed germination experiments have indicated that spatial variation in germination success has contributed to the observed spatial distribution patterns of both pure and hybrid plants. To unambiguously determine the role of orchid mycorrhizal fungi in determining spatial distribution patterns of orchids and their hybrids, future research should sample mycorrhizal fungi directly from the soil, which, combined with detailed soil analyses, will allow elucidation of the factors determining the spatial distribution of fungi in the soil. From a conservation point of view, complete assimilation of one of the species is not likely to occur as long as different species occupy different areas of the population. Therefore, the patchy distribution of mycorrhizal fungi may limit hybridization to particular areas within populations and therefore contribute to the maintenance of the genetic integrity of extant populations.
We would like to thank the Editor and three anonymous reviewers for their constructive remarks which significantly improved the quality of this manuscript. This research was funded by the Flemish Fund for Scientific Research (grant: G.0592.08). H.J. and T.W. acknowledge funding from the European Research Council (ERC starting grant 260601 – MYCASOR and ERC advanced grant 233066 – SPATIODIVERSITY, respectively).