• Spatial genetic structure (SGS) holds the key to understanding the role of clonality in hybrid persistence, but multilocus SGS in hybrid zones has rarely been quantified. Here, the aim was to fill this gap for natural hybrids between two diploid, ecologically divergent European tree species with mixed sexual/asexual reproduction, Populus alba and P. tremula.
• Nuclear microsatellites were used to quantify clonality, SGS, and historical gene dispersal distances in up to 407 trees from an extensive Central European hybrid zone including three subpopulation replicates. The focus was on P. × canescens and its backcross parent P. alba, as these two genotypic classes co-occur and interact directly.
• Sexual recombination in both taxa was more prominent than previously thought, but P. × canescens hybrids tended to build larger clones extending over larger areas than P. alba. The 3.4 times stronger SGS in the P. × canescens genet population was best explained by a combination of interspecific gene flow, assortative mating, and increased clonality in hybrids.
• Clonality potentially contributes to the maintenance of hybrid zones of P. alba and P. tremula in time and space. Both clonality and SGS need to be taken into account explicitly when designing population genomics studies of locus-specific effects in hybrid zones.
Although the spatial analysis of allele frequencies at individual genetic loci in hybrid zones has a long tradition (reviewed by Barton & Hewitt, 1985; Barton & Gale, 1993), multilocus spatial genetic structure (SGS; the nonrandom distribution of genotypes in space) in hybrid populations is a little explored topic (but see Cornman et al. (2004) and Valbuena-Carabaña et al. (2007) for two rare examples in plants). This knowledge deficit for most hybridizing species is unfortunate, considering the power of SGS analysis to uncover changes in fine-scale structure as a result of asexual reproduction (Chung & Epperson, 2000; Suvanto & Latva-Karjanmaa, 2005). Also, SGS in hybrid zones will be affected by many of the same factors that also affect genome-wide levels of linkage disequilibrium (LD), a crucial parameter in population genomics and association genetics studies (Neale & Savolainen, 2004; Ingvarsson, 2005; Lexer et al., 2006). Key factors affecting both SGS and LD include levels of inbreeding vs outcrossing, sexual vs clonal reproduction, and effective population size (Flint-Garcia et al., 2003). Clearly, fine-scale analysis of SGS holds the potential to elucidate the extent to which asexual reproduction may contribute to the maintenance of natural hybrid genotypes, and it can provide baseline data for population genomics work in hybrid zones.
Many empirical studies have investigated fine-scale SGS within plant populations, using molecular markers and spatial autocorrelation methods (Epperson & Allard, 1989; Perry & Knowles, 1991; Streiff et al., 1998; Heuertz et al., 2001). Most of these studies describe patterns in a qualitative way, making quantitative comparisons among studies difficult (Vekemans & Hardy, 2004). In order to quantify SGS within populations, Vekemans & Hardy (2004) introduced a new statistic called Sp, which facilitates comparisons among studies, taxa and sites, and at the same time is robust to different sampling schemes. Furthermore, indirect estimates of ‘historical’ gene flow can be obtained from the SGS of populations at drift-dispersal equilibrium. In contrast, approaches that follow the movements of individuals or propagules via in situ monitoring or reconstruction through parentage analysis provide ‘contemporary’ (real-time) gene flow estimates (Vekemans & Hardy, 2004). Plants with wide-ranging dispersal of both pollen and seed via wind exhibit weak SGS with a close-to-random spatial distribution of genotypes (Perry & Knowles, 1991; Knowles et al., 1992; Heuertz et al., 2001; Suvanto & Latva-Karjanmaa, 2005), but the extent of fine-scale SGS in interspecific hybrid zones is largely unknown.
The diploid genus Populus (poplar) is widely regarded as an ‘enfant terrible’ by taxonomists because of high amounts of intraspecific variability, leaf heteroblasty (Cronk, 2004) and widespread hybridization between ecologically differentiated species (Rajora & Dancik, 1992; Eckenwalder, 1996; Martinsen et al., 2001). However, it is exactly the ‘porosity’ of species barriers in this genus which makes Populus an attractive model for microevolutionary studies, together with an almost completely sequenced genome (2n = 38, 440–550 Mb, 2C = 1.1 pg) for one species of the genus (Tuskan et al., 2006). Also, natural Populus hybrids are sometimes persistent (Schweitzer et al., 2002) and both sexual and clonal reproduction are common.
The present study focuses on a large Central European hybrid population in the Austrian Danube valley. Much of this hybrid zone is located within an extensive national park, which facilitates fine-scale sampling in well defined ‘replicate’ plots or subpopulations along the river. Highly variable nuclear microsatellite markers were used for an accurate classification of P. alba (backcross parent)-like and P. × canescens-like trees sampled via a molecular hybrid index and thus to estimate their proportions and spatial distributions in the hybrid zone. Furthermore, the microsatellites allowed us to quantify fine-scale SGS and recognize clones. In particular, we asked the following questions:
• Do P. × canescens hybrids and their sympatric backcross parent P. alba differ in their propensity to form clones?
• Do the two taxa exhibit different or similar SGS in the Central European hybrid zone, and to what extent does clonality affect SGS?
• How similar or different are SGS-based estimates of historical gene dispersal in P. alba and P. × canescens?
We discuss the potential of asexual reproduction to maintain recombinant genotypes in hybrid zones of P. alba and P. tremula. We also comment on practical implications of multilocus SGS for population genomics work aimed at the detection of locus-specific effects in hybrid zones.
Materials and Methods
Study site and taxa
The study area is situated in riverine (lowland floodplain) forests of the Danube between Krems (Austria) and Bratislava (Slovakia). In the past, the unhampered dynamics of the Danube defined life rhythms in the floodplain and shaped the landscape. This together with the pannonic climate has led to extraordinary degrees of habitat diversity in the floodplain, ranging from tributaries, canals, marshy pools and sloughs to gravel banks, riverine forests (wet and moist wetlands) and well-drained, dry habitat patches with vegetation reminiscent of savannahs (German: ‘Heißlenden’). This wide range of habitats is patchily distributed in the floodplain. Although the regulation of the Danube (1870–75) has resulted in correction and permanent arrest of the riverbed, the river in this area still has the characteristics of an alpine stream with high waters in spring and summer when snow and ice melt in the mountains.
Based on morphological characters, the presence of hybrid P. × canescens in this area has been known for several decades (Lazowski, 1997). Their existence here was later confirmed using nuclear markers (Lexer et al., 2005). Human influence on P. alba and P. × canescens in the floodplain comprised mainly plantations of P. euramericana (a hybrid between P. deltoides and P. nigra) in native habitats of P. alba and P. × canescens and occasional clearings of forest patches for hunting (Jelem, 1974). Subsequent natural regeneration of cleared patches may have supported ‘recolonization’ by P. alba and P. × canescens through root sucker shoots and may thus, at least in theory, have influenced present-day patterns of clonality in some locations.
Leaves of 407 P. alba-like, P. tremula-like and P. × canescens-like trees were collected in the floodplain forests between Krems and Bratislava, covering a stretch of approx. 90 km of the river valley. This dataset was used for studying clonality in all genotypic classes sampled, whereas a subset of the sampled trees (of which the exact geographical location could be determined) was used for studying SGS and historical gene dispersal. The latter analyses were restricted to all 340 ramets collected within the Danube Floodplain National Park (http://www.donauauen.at/html/english/index.html) covering a strech of ∼35 km of the river valley. The sampling of three spatially separated subpopulations (Fig. 1) within the sampling area allowed us to check for differences and commonalities between different localities before combining the data. In more detail, subpopulation 1 was located at the western boundaries of the park (‘Lobau’ floodplain; n = 80). Subpopulation 2 was located near Orth an der Donau in the middle of the park (n = 105). Subpopulation 3 was located near Hainburg at the eastern boundaries of the national park (n = 155). Trees were collected nonexhaustively but randomly with a minimum distance of 8 m between them. Individuals were sampled not only along transects, but also in groups of nearby individuals at different locations. Sampling along transects yields a broad distribution of pairwise geographic distances. As the dispersal of both pollen and seed in poplars is wide-ranging, sampling over large distances is particularly important for a meaningful estimation of SGS. The inclusion of groups of nearby individuals was important for estimating clone size in terms of area covered. Locations of individual trees were identified using a global positioning system (GPS) device. The accuracy of the coordinates was on average between 5 and 6 m. The age of most collected trees was estimated to be between 50 and 100 yr. Specimen vouchers are available at the Jodrell Laboratory, Royal Botanic Gardens Kew, Richmond, UK.
DNA extraction and microsatellite genotyping
Total genomic DNA was extracted from leaves using the DNeasy Plant Mini Kit (Qiagen, Hilden, Germany), and DNA was quantified using an Eppendorf BIOphotometer. Nineteen independent nuclear microsatellite loci (Table 1) were used to obtain multilocus individual genotypes. The microsatellites developed by Tuskan et al. (2004) and Van der Schoot et al. (2000) are available at the following web site: http://www.ornl.gov/sci/ipgc/ssr_resource.htm. In polymerase chain reactions (PCR), either a traditional two-primer approach (in which the forward primer was fluorescent-labelled) or a three-primer protocol (including unlabelled M13-tagged forward and untagged reverse primers and a ‘universal’ fluorescence-labelled M13-primer) were used as in Lexer et al. (2005). Microsatellite genotypes were resolved on an ABI Prism™ 3100 DNA Genetic Analyzer (PE ABI, Tokyo, Japan) making use of multiplexing of PCR products labelled by the fluorescent dyes FAM and JOE (MWG Biotech, Ebersberg, Germany). Sizing of fragments was carried out with Genescan 3.7 and Genotyper 2.0 software (PE ABI) utilizing the internal GENESCAN™-500 ROX™ Size Standard (PE ABI).
Table 1. Genetic variability for 19 nuclear microsatellite loci at the genet level for Populus alba (169 genets) and P. × canescens (123 genets), respectively, estimated from the entire set of 407 ramets
P. × canescens
A, number of alleles; HO, observed heterozygosity; HE, expected heterozygosity; FIS, inbreeding coefficient.
Marker information content and molecular hybrid index (HI)
To quantify the power of the 19 markers to distinguish between genotypes, the probability of identity (PID) was estimated over all loci in the entire data set using the gimlet software (Valière, 2002). The microsatellite loci were also characterized in P. alba and P. × canescens via the number of alleles (A), expected heterozygosity (HE), and observed heterozygosity (HO) using the msa software (Dieringer & Schlötterer, 2003). All parameters were calculated for the ramet- and the genet level. To check for departures from random mating in the two taxa, we tested for departures from Hardy–Weinberg equilibrium (HWE) using exact tests in genepop 1.2 (Raymond & Rousset, 1995). Subsequently, both P. alba and P. × canescens were characterized using HE and HO calculated by msa, and the within-population inbreeding coefficient FIS for microsatellite loci using genepop.
In European poplars and aspens, identification of hybrids and pure parental species based on morphology is difficult (Adler et al., 1994). In addition, a large proportion of late-generation backcross hybrids was reported for the study region (Lexer et al., 2005), which would make hybrid classification based on morphology even more problematic. Thus, we decided to discriminate between P. × canescens and its sympatric backcross parent P. alba using molecular markers. Maximum-likelihood estimates of HI based on 19 polymorphic microsatellites were used to classify trees as either P. alba or hybrid P. × canescens using the hindex software (Buerkle, 2005). This program calculates HI with values ranging from 0 (parental species P. tremula) to 1 (parental species P. alba). Individual trees were classified as P. alba whenever HI > 0.95, and as P. × canescens when 0.05 < HI < 0.95. An arbitrary cut-off point of 0.95 was chosen over 0.90, as unidirectional gene flow to P. alba and a high proportion of late-generation hybrids (Lexer et al., 2005) easily lead to underestimates of introgression unless many loci in the genome have been sampled. In order to perform an accurate analysis of HI, reference samples from Lexer et al. (2005) were used, representing pure (nonintrogressed) populations for each parental species (P. alba and P. tremula, n = 48 each).
To analyse clonal structure, all 407 ramets sampled were classified as either P. alba or P. × canescens using HI. Within each group, ramets with an identical multilocus genotype (IMG) were identified using Gimlet. Observed IMGs can be either a result of sampling the same clone/genet at different spatial coordinates or products of distinct sexual reproduction events sharing identical alleles by chance. For each taxon, equation (1) of Parks & Werth (1993) was used to calculate the probability Pgen of observing the same genotype by sexual reproduction. Then, number of genets/clones, as well as the number of multiramet and single-ramet genets were determined using Gimlet. Differences between P. alba or P. × canescens were tested for significance using binomial tests in SPSS 14.0 software (SPSS inc., Chicago, IL, USA). Genotypic diversity (clonal diversity) was determined as the G/N ratio, the proportion of different genotypes found for each of the two taxa, where G is the number of observed genets, and N is the total number of ramets analysed (Pleasants & Wendel, 1989).
Spatial genetic structure (SGS)
Spatial genetic structure was analysed for 340 ramets collected in the Danube Floodplain National Park (Fig. 1). The SGS analysis was carried out for both P. alba and P. × canescens in each of three subpopulation replicates separately, and for the replicates combined at two different levels: the ramet level, including all sampled ramets; and the genet level, including each distinct genotype just once, using one ramet from the centre of each clone. To test for SGS and to assess the effect of clonality on SGS, pairwise kinship coefficients (Fij) between individuals were computed for both taxa and levels (ramet and genet), and their relationship with the spatial distance separating individuals was analysed using Spagedi 1.2 (Hardy & Vekemans, 2002). Kinship coefficients were computed for all pairs of individuals using the statistic of Loiselle et al. (1995). The kinship Fij estimator of Loiselle et al. (1995) was used because the logic behind its construction makes no assumption regarding Wright's inbreeding coefficient; it possesses good statistical properties (it is less biased and has a lower variance than other kinship measures); and, most importantly, all Sp values of different taxa cited in this publication have used this estimator too, which allows us to compare our results with a large body of published literature (see reviews by Hardy & Vekemans, 1999; Vekemans & Hardy, 2004). Pairwise kinship coefficients were regressed on the logarithm of spatial distance dij (d is the distance between i and j) to estimate the logarithmic regression slope blog. The significance of blog was tested by permuting the spatial positions of individuals 10 000 times to obtain the frequency distribution of b under the null hypothesis that Fij and dij were uncorrelated (cf. Mantel test). All 19 microsatellites were used for the SGS analysis. Because both taxon discrimination and the SGS analysis were based on the same data (nuclear microsatellites), the degree of independence between taxon classification and estimation of SGS was investigated by randomly dividing markers into four distinct groups. A combination of three of the groups was used to identify taxa and markers of the fourth group were used for estimating SGS at both the ramet and genet levels in P. × canescens. This was repeated for all possible combinations of marker groups to check if SGS parameters remained robust.
The extent of SGS was estimated using the Sp statistic following Vekemans & Hardy (2004). Sp was quantified by Sp = –blog/(1 – F(10,m)), where blog is the regression slope and F(10,m) is the mean kinship coefficient between individuals belonging to the first distance interval (0–10 m). The blog standard errors were obtained by jack-knifing over loci. For graphical visualization of SGS in correlograms, average kinship coefficients were estimated for the following nine distance classes (in m): 20–40; 40–80; 80–160; 160–320; 320–640; 640–1280; 1280–2560; 2560–3000; 3000–4000 m. The kinship coefficients were computed relative to the entire sample set used for the SGS analyses.
For comparative purposes, SGS was also estimated using a relationship coefficient (Moran's I) in addition to Loiselle's kinship coefficient (Fij). The relationship coefficient has advantages when comparing two species that differ in levels of inbreeding or ploidy, both of which may affect amount of genetic drift (Hardy & Vekemans, 1999). Using Moran's I, Iij for the shortest distance interval (I(10,m)) with standard errors (SE), and blog with SE were estimated at the genet level using Spagedi 1.2 (Hardy & Vekemans, 2002) and the results compared with those calculated using Loiselle's kinship coefficient Fij.
Indirect estimation of historical gene dispersal
Historical gene dispersal (σg) was estimated at the genet level for all P. alba and P. × canescens collected in the Danube Floodplain National Park. To estimate σg, an iterative procedure proposed by Vekemans & Hardy (2004) using Spagedi 1.2 was performed. To apply this procedure, effective population density (De) is assumed to be known. However, De is not simply the adult population density, as it also depends on the variance of lifetime reproductive success among individuals (effective density) (Hardy et al., 2006). The actual density of trees (D) in the study area was estimated to be 200 adult trees (P. alba and P. × canescens together) per hectare. Actual densities (D) of P. alba and P. × canescens per hectare were obtained after genotype classification using the 19 microsatellites. As our study taxa are dioecious clonal plants, and thus effective population size and density are additionally reduced by replication of genets and the presence of separate sexes if sex ratio is unbalanced, we chose to estimate σg using a range of assumed De values, namely 0.25D, 0.5D, and 0.1D.
Information content of markers and sampled genotypes
As expected, all nuclear microsatellites were variable, with up to 24 alleles observed per locus and taxon. Four loci displayed significant deviations from HWE in P. alba and eight did so in P. × canescens at the ramet level (data not shown). At the genet level, however, one locus and three loci departed significantly from HWE in P. alba and P. × canescens, respectively (Table 1). As no null-alleles were observed in the two parental species for these loci previously (Lexer et al., 2005), the deviations from HWE may be explained by clonality and population subdivision. In addition, the more frequent departures from HWE in P. × canescens likely indicate more frequent departures from random mating in the hybrid population. The probability of identity (PID) in the entire dataset representing 407 ramets revealed a high amount resolution: PID calculated over all loci was 7.58e−11, and thus the 19 markers are suitable for genetic studies at the individual level. Maximum-likelihood estimates of hybrid index (HI) based on 19 markers revealed 222 ramets of P. alba (HI > 0.95), and 185 P. × canescens hybrids (HI < 0.95), the latter comprising primarily backcrosses to P. alba, plus 16 apparent F1s, and three recombinant backcrosses to P. tremula. The observed proportions of genotypes were very similar to those observed in the initial characterization of the hybrid zone (Lexer et al., 2005) where hybrid genomic composition was reported in detail.
The probability of encountering individuals with identical multilocus genotypes (IMGs) as a result of distinct sexual reproduction events was extremely low in both taxa (Pgen < 1.3e−9). Therefore, all ramets of recurring genotypes were assumed to belong to the same genet. The distribution of clone size was highly skewed in both P. × canescens hybrids and P. alba (Fig. 2a). That is, many small and few large clones were found for either taxon, the largest group being single-ramet genets in both taxa; 39.6% and 49.7% of ramets were identified in multiramet genets of P. alba and P. × canescens, respectively. Furthermore, in P. alba, 134 (79%) genotypes were found only once, and 35 (21%) genotypes were represented by at least two ramets. The results were similar for P. × canescens, where 93 (76%) single-ramet genets and 30 (24%) multiramet genets were identified (Table 2), and the differences between taxa were not significant. Nevertheless, it is the maximum distance between ramets of a clone, which informs us about differences in clonality between taxa. This distance was larger for P. × canescens (186 m) than for P. alba (132 m) (Fig. 2b). Genotypic diversity (G/N) was 0.76 in P. alba and 0.66 in P. × canescens, indicating a prominent role for sexual reproduction in both taxa. Ramet and genet numbers in P. alba and P. × canescens at the level of subpopulation replicates are shown in Table 2.
Table 2. The number of individuals (ramets) of Populus alba and P. × canescens identified in the entire dataset of 407 ramets and three subpopulation replicates via a molecular hybrid index, as well as the number of genets, and multiramet genets and their percentages
No. of ramets
No. of genets
No. of multiramet genets/%
P. × canescens
P. × canescens
P. × canescens
Spatial genetic structure
The spatial distribution of P. × canescens and P. alba ramets is shown in Fig. 1. No taxon clustering was evident; rather, P. alba and P. × canescens hybrids were intermingled in all subpopulation replicates throughout the study area.
At the level of subpopulation replicates, nonrandom SGS was detected at both the ramet and genet levels in all three replicates, except for the genet level in subpopulation 1. The following patterns were observed across all three replicates: P. × canescens was found to be consistently more structured than P. alba in all sites; and more structure (greater Sp) was found at the ramet than at the genet level in both taxa.
Consistent trends across subpopulation replicates allowed us to combine subpopulations and re-run SGS analyses at the ramet and genet levels for all P. × canescens (n = 161) and P. alba (n = 179) identified in the Danube Floodplain National Park. In agreement with models of isolation by distance, a significant linear decrease of pairwise kinship coefficients Fij with the logarithm of geographical distance was detected in both taxa at both the ramet and the genet levels. For visualization of SGS in P. alba and P. × canescens, correlograms of spatial autocorrelation with nine distance classes are presented in Fig. 3 (last class up to 4000 m). The first seven Fij values in P. × canescens at both the ramet and the genet levels were significantly larger than the values encompassed by the permutation envelope (P ≤ 0.05), despite the fact that confidence intervals for the first three distance classes were large because of a small number of pairs in these classes. As the spatial scale studied and the sampling sizes are essentially the same for both taxa, we can compare their Fij (and Iij) values directly. The mean of the kinship coefficient for the first seven distance classes was higher in P. × canescens than in P. alba at both the ramet and genet levels. The fact that the Fij values obtained were almost exclusively positive in both taxa is the result of the choice of the overall data set as a reference. The slopes of kinship coefficients (blog) at the ramet level were clearly steeper (had higher negative values) than those at the genet level in each taxon (Table 3). In addition, if taxa are compared, the blog at the genet level for P. × canescens was three times higher than for P. alba. A similar trend was found at the ramet level, where blog in P. × canescens was 2.5 times higher than in P. alba. Consequently, the steeper slopes for P. × canescens than for P. alba resulted in higher values for the Sp statistic in P. × canescens regardless of the level of analysis (Table 3).
Table 3. Summary of kinship autocorrelations in Populus alba and P. × canescens analysed at two different levels (ramet and genet)
F(10,m) (± SE)
blog (± SE)
Kinship coefficients (Fij) for the shortest distance interval (F(10,m)) with standard errors, the slope of the regression of mean kinship with the logarithm of spatial distance (blog) with standard errors, and the Sp statistic. Significant values of F(10,m) and blog are shown in bold (α = 0.025).
P. alba_ramet level
P. alba_genet level
P. × canescens_ramet level
P. × canescens_genet level
Analyses including distinct, nonoverlapping marker sets for hybrid classification and SGS analysis revealed consistently high values for SGS in P. × canescens (regression slopes blog ranging from –0.0089 to –0.0107 and from –0.01712 to –0.02081 at the genet and ramet level, respectively). Thus, the increased SGS found in P. × canescens (Table 3; Fig. 3) is robust and not an artefact of the marker set used for hybrid classification.
Analyses of SGS at the genet level using Moran's I relationship coefficient, which takes into account increased amounts of drift in the P. × canescens hybrid swarm, led to similar results as Loiselle's kinship coefficient. More specifically, the I(10,m) was greater for P. × canescens (0.4 ± 0.089) than for P. alba (0.263 ± 0.082) and the blog in P. × canescens (0.019 ± 0.003) was 2.75 times higher than in P. alba (0.007 ± 0.002).
Historical gene dispersal distances
The iterative procedure used to estimate the gene dispersal converged in all cases. For all three assumed densities (De), historical gene dispersal was shorter in P. × canescens than in P. alba. Estimates of σg ranged from 28 to 137 m and 51 to 281 m for P. × canescens and P. alba, respectively. These estimates should be treated with caution because they are affected by assumptions regarding effective population density. In summary, P. × canescens displayed stronger SGS than P. alba, resulting in more restricted gene dispersal (lower σg) in the former.
Similar numbers of P. alba and P. × canescens were found by random sampling and subsequent classification of genotypes with a molecular hybrid index, and both taxa occurred intermingled in the lowland floodplain forest (Fig. 1). As expected, based on its known ecological preferences, the second parental species P. tremula is not present here, but occurs within easy pollen dispersal distance several kilometres away in mixed upland forest communities. Thus, a comparative analysis of P. × canescens and its sympatric backcross parent P. alba is of particular relevance to our understanding of the spatial structure of this Central European hybrid zone.
Clonal reproduction and hybrid persistence
Populus species produce numerous wind-dispersed seeds, and high germination rates as well as rapid growth in riverine Populus species are commonly reported as strategies of establishment after flooding (Schweitzer et al., 2002). However, seedling mortality is often high, because seedlings require specific geomorphic niches until the roots are long enough to reach the water table (Siegel & Brock, 1990; Mahoney & Rood, 1998). Thus, a high propensity to form clones and high survival of ramets are crucial to the establishment of riverine Populus species.
A crucial aspect of clonal reproduction in hybrid zones is that it will effectively shelter interspecific gene combinations from further recombination. Indeed, clonal reproduction is one of several mechanisms that may allow hybrid genotypes to persist in the face of meiotic difficulties (Ellstrand et al., 1996; Schweitzer et al., 2002). Also, a ramet-producing genet increases its expected time to extinction (Cooper, 1984) and can exploit new habitats as a result of its interconnected ramets. Thus, clonal reproduction may contribute to the persistence of hybrid zones in time and space. Note, however, that the extent to which the accumulation of deleterious mutations may offset fitness benefits in clones is largely unknown (Orr, 2000).
A role for asexual reproduction in hybrid persistence has been demonstrated for both natural hybrid populations and experimental crosses of Populus fremontii and P. angustifolia (Schweitzer et al., 2002). In their study, hybrids produced approximately two and four times as many ramets as their parental species P. angustifolia and P. fremontii, respectively, and mortality of asexually derived ramets was low. Increased hybrid fitness was attributed in part to genetic contributions of asexual traits from one parent (P. angustifolia). Also, cloning was a transgressive character in backcross hybrids (Schweitzer et al., 2002), reinforcing the idea that increased asexual reproduction may contribute to hybrid persistence.
In our study, there was no statistically significant difference in the number of multiramet genets in P. alba and P. × canescens. Nevertheless, the proportion of ramets that occurred within multiramet genets was higher in P. × canescens than in P. alba (Fig. 2a). Also, clones studied in P. × canescens extended over larger areas than in P. alba (Fig. 2b; maximum distance was 186 vs 132 m, respectively). Our estimates of clone size are conservative, as trees in the study area (8 × 90 km) were not sampled exhaustively. Thus, a role for clonal reproduction in shaping the spatial structure of P. × canescens hybrid zones seems likely. Common-garden experiments with experimental crosses would help to confirm the observed patterns. In addition, as the geographical positions of all P. alba and P. × canescens are mapped for the entire study area, clonal reproduction can be studied in situ by exhaustive surveys of clonal propagates from different genotypic classes.
Spatial genetic structure in P. × canescens and its sympatric backcross parent P. alba
In this study, SGS was estimated using the Sp statistic developed by Vekemans & Hardy (2004), an approach which has been used in many publications. Although this statistic has, to our knowledge, not proven to be misleading, more theoretical work comparing its statistical properties with other analytical approaches would be beneficial to guide experimental workers.
Calculations using Loiselle's kinship estimator and Moran's relationship estimator yielded comparable results and identical patterns in P. alba and P. × canescens, and thus only results based on Sp estimates (for which Loiselle's kinship estimator was used) are discussed from here onwards. In both P. alba and P. × canescens, the correlograms revealed clear differences between the spatial patterns of genets and ramets. As expected for a tree with wind-based pollen and seed dispersal, the observed Sp for the P. alba genet population (0.00414) was among the smallest values reported for SGS in plants to date. For comparison, Sp estimates for wind- or animal-pollinated outcrossing trees range from 0.00196 to 0.00460 (Vekemans & Hardy, 2004). Extremely weak SGS was also found in Larix decidua (Sp = 0.0045), Fraxinus excelsior (Sp = 0.00196) (Vekemans & Hardy, 2004) and in P. tremula (Suvanto & Latva-Karjanmaa, 2005). Although the latter study did not describe SGS in terms of the Sp statistic, the authors provided us with b-values (regression slope with linear distance). Utilizing this information and kinship coefficients F(20,m) read off from the published correlograms, we calculate Sp to be < 0.0035 at the genet level in P. tremula, which is remarkably similar to that observed for P. alba in our study.
Although SGS can result from a variety of factors, gene flow through pollen and seed dispersal is a key determinant in its establishment (Wright, 1943; Schnabel & Hamrick, 1995; Streiff et al., 1998; Vekemans & Hardy, 2004). In Populus, both seed and pollen are dispersed by wind (Reim, 1929), which provides a simple explanation for the weak SGS observed in P. alba. Notably, the observed Sp (0.01413) in the hybrid P. × canescens genet population was 3.4 times higher than that for genets of P. alba, which requires additional explanations. We shall discuss the most important key factors that may be responsible for this conspicuous pattern one by one below.
Two key factors shaping the spatial structure of hybrid populations, in particular, are the amount of gene flow and the strength of selection against hybrids (Barton & Hewitt, 1985). Most genetic marker-based hybrid zone studies have utilized spatial ‘clines’ of allele frequencies at individual loci (reviewed by Barton & Hewitt, 1985; Barton & Gale, 1993), rather than multilocus SGS (but see Cornman et al., 2004, and Valbuena-Carabaña et al., 2007). The latter study addressed the effect of hybrids on multilocus SGS in mixed oak populations containing Quercus pyrenaica and Q. petraea. In their study, interspecific gene flow tended to dilute SGS present in the parental species by lowering pairwise kinship for shorter distance classes. However, their study cannot easily be compared to ours because of the greater difficulty of detecting interspecific hybrids in Quercus spp.
In our study, the much stronger SGS in P. × canescens than in P. alba found at the genet level may be explained by the spatial aggregation of backcrossed hybrids resulting from hybridization and introgression. Considering the estimated age of the hybrid zone (100–200 generations; Lexer et al., 2006), however, it is difficult to imagine how the observed spatial structure in P. × canescens could be maintained without assortative mating. The reduced gene dispersal distance observed for P. × canescens may in itself be a consequence of the combined action of these factors. Of course, an increased propensity of hybrids to clone will contribute to increased SGS in genets of P. × canescens as well, since high historical amounts of cloning will affect spatial patterns even after all present-day ramets for each genet have been removed.
In addition, although P. × canescens and P. alba trees were found in similar numbers (Fig. 1), P. × canescens hybrid clones extended over larger areas than clones of P. alba. This may lead to lower effective density in hybrids and thus contribute to the differences in SGS observed here. It is known that the Sp values tend to be higher in low-density compared with high-density populations (Vekemans & Hardy, 2004). Ecology may also shape the spatial structure of P. × canescens hybrid zones via microenvironmental selection, but more refined measurements of ecological amplitudes will be required to test this hypothesis, including the recording of fine-scale pedological mapping information, affiliation with different plant communities, and soil core profiles.
In summary, interspecific gene flow and increased clonality in hybrids both maintain increased spatial structure in a P. × canescens hybrid zone, and assortative mating and microenvironmental selection may contribute to this pattern to an unknown extent.
Implications for ‘admixture mapping’ studies in European Populus hybrid zones
Population genomics studies aimed at detecting locus-specific effects (Luikart et al., 2003) and admixture mapping studies aimed at the genetic mapping of phenotypic trait differences in hybrid zones (Rieseberg & Buerkle, 2002; Lexer et al., 2006) require sampling strategies that minimize LD as a result of cloning or the spatial aggregation of related genotypes. Linkage disequilibrium induced by these factors would result in spurious false-positives and decreased resolution. Considering the well-known effect of asexual reproduction on LD and known amounts of background LD in these two European Populus species and their hybrids (Lexer et al., 2006), our results indicate that asexual reproduction needs to be accounted for when designing admixture LD-related experiments. This can be achieved in a simple way by detecting clonal propagates in an initial marker screen as carried out here and reducing subsequent genomic scanning to the genet level. Our finding of increased SGS in the P. × canescens genet population (clones removed) also indicates a high potential to moderate the strength of LD experienced in genome-wide marker scans by widening the spatial level of sampling. Population genomics and admixture mapping-related studies in interspecific plant hybrid zones are still in their early days (Rieseberg & Buerkle, 2002; Lexer et al., 2006). Our results on a Central European hybrid zone of P. alba and P. tremula indicate that studies of multilocus SGS and clonality can inform such experiments.
We thank the International Populus Genome Consortium (IPGC) for making microsatellite PCR primers available, Hans Herz, Wilfried Nebenführ, and Edwin Herzberger of the BFW Vienna for help with field and literature work, Christian Fraissl and Franz Kovacs of the Danube Floodplain National Park, Austria, the Forstverwaltung Lobau of the Vienna City Council, Herbert Tiefenbacher and several other private landowners in Austria for their continued support during sample collection, Myriam Heuertz and Olivier J. Hardy for advice on SGS analysis, and two anonymous referees for their comments on the manuscript. This work was supported by Erwin-Schrödinger grant J2476 of the Austrian Science Foundation (FWF) to MvL and NERC project NE/C507037/1 to CL.