| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Population Biology |
2INRA, Unité de Recherches Forestières Méditerranéennes, Domaine Saint Paul, Site Agroparc 84914 Avignon Cedex 9, France; 3Conservatoire Génétique des Arbres Forestiers, Office National des Forêts, Campus INRA, F-45160 Ardon, France; 4Unité de Biométrie, Domaine St-Paul, Institut National de Recherche Agronomique, F-84914 Avignon Cedex 9, France; and 5Laboratoire Ecologie, Systématique et Evolution, UPRESA 8079, CNRS/Université Paris-Sud, Bât. 362, F-91405 Orsay Cedex, France
Received for publication October 14, 2005. Accepted for publication September 11, 2006.
ABSTRACT
Understanding the role of mother plants as pollen recipients in shaping mating patterns is essential for understanding the evolution of populations and in particular to predict the consequence of habitat fragmentation. Here, we investigated variation in mating patterns due to maternal phenotypic traits, phenological variance, and landscape features in Sorbus torminalis, a hermaphroditic, insect-pollinated and low-density, European temperate forest tree. The diversity and composition of pollen clouds received by maternal trees in S. torminalis were mainly determined by their conspecific neighborhood: isolated individuals sample more diversity through more even paternal contributions, low relatedness among paternal genes, and high rates of long-distance pollen dispersal within their progenies. Maternal phenotypic traits related to pollinator attractiveness also had an effect, but only when competition was strong: in this case, larger mother trees with more flowers sampled more diversity. The floral architecture of S. torminalis, with multiple-seeded fruit, strongly shaped mating patterns, with higher levels of correlated paternity among seeds belonging to the same fruit (30% full sibs) than among seeds belonging to different fruits (14% full sibs). Finally, flowering phenology affected the distribution of diversity among maternal pollen clouds, but the earliest and latest mother trees did not receive less diversity of pollen than the others.
Key Words: correlated paternity • covariance analyses • flowering phenology • insect pollination • pollen flow • Rambouillet forest • Rosaceae
The interaction between ecological processes and mating system and its impact on the evolution of genetic diversity in plant populations is an issue of growing interest in evolutionary biology (Ward et al., 2005
). In sexually reproducing plants, the female (i.e., functionally female) acts first as a pollen recipient during pollination, then provides the environment for seed development, and finally acts as the source for seed dispersal. Together, these features affect the successful development of each progeny, through its own individual genetic makeup (e.g., inbreeding/outbreeding depression) and through competitive interactions among sibs (Hardy et al., 2004
). Thus, unraveling the ecological determinants of maternal pollen cloud diversity and composition is a key step to understand the evolution of populations (e.g., García et al., 2005
). In particular, we will be able to more accurately predict the effects of habitat fragmentation and reduction in population size on the maintenance of genetic diversity in a managed, fragmented, or degraded landscape (Ellstrand and Elam, 1993
).
A growing number of empirical studies have started to address these issues, thanks to the joint development of powerful statistical methods and of highly variable molecular markers (Hardy et al., 2004
; O'Connell et al., 2004
; García et al., 2005
). These authors showed the impact on mating systems of phenotypic traits of the pollen recipient, in particular their sexual phenotype (Barrett, 2002
; García et al., 2005
), male/female floral display and floral architecture (Motten and Antonovics, 1992
; Barrett, 2002
; Hodgins and Barrett, 2006
). Also, they highlighted how temporal variations in flowering patterns induce phenological assortative mating in many plant populations (Weis and Kossler, 2004
), which can strongly shape the diversity and composition within and among maternal pollen clouds (Fuchs et al., 2003
). Finally, they emphasized that mating patterns can be affected by a wide range of ecological factors acting in the vicinity of the pollen-recipient plant, the so-called ecological maternal neighborhood (EMN, see García et al., 2005
). In particular, the density and spatial distribution of conspecific individuals is widely acknowledged as a main determinant of outcrossing rates (Robledo-Arnuncio et al., 2004a
), pollination distance (Goodell et al., 1997
; Robledo-Arnuncio et al., 2004b
), and the effective number of pollen donors (Dick et al., 2003
; Fuchs et al., 2003
). The direct effect of conspecific density can be finely described using modeling approaches (e.g., Robledo-Arnuncio et al., 2004a
). These effects may, however, be either reinforced or compensated by interactions with other ecological factors, notably pollinator abundance and assemblage, vegetation structure, and/or silvicultural practices (Dick et al., 2003
; Fuchs et al., 2003
; Robledo-Arnuncio et al., 2004b
). Thus, another conclusion from these empirical studies is that even using a sophisticated (e.g., spatially explicit) statistical framework, disentangling the relative impact of landscape factors (e.g., relative disposition of pollen sources and recipients) and of maternal phenotypic traits (e.g., number, color, size of flowers) in the shaping of pollen cloud diversity remains difficult.
Our objective here is to gauge the impact of these ecological factors in a population of Sorbus torminalis (L.) Crantz, a hermaphroditic, insect-pollinated, low-density European forest tree species. In a previous paper (Oddou-Muratorio et al., 2005
), we investigated patterns of pollen dispersal in the same population, using a source-modeling approach that incorporated the effect of phenotypic factors on male fecundity. By contrast, the focus of the present study is on variation in pollen clouds intercepted by different mother trees across the landscape.
The diversity of maternal pollen clouds can be characterized by the number of siring fathers and the level of balance between their contributions, the level of relatedness among these fathers and the proportion of long-distance pollen flow. We used three kinds of "mating variables" to describe these features. First, we estimated the correlation of paternity within and among pollen clouds (Hardy et al., 2004
). We computed in particular the average level (FS) of relatedness within the pollen clouds. Second, we used offspring assigned to a sampled father by categorical paternity analyses (Oddou-Muratorio et al., 2003
) to estimate the number of effective fathers (Nep) siring seeds on each mother tree. These variables (FS and Nep) provide complementary estimates of the probability of paternal identity within maternal progeny (Hardy et al., 2004
; Smouse and Robledo-Arnuncio, 2005
). The last variable used to characterize pollen clouds was the rate of pollen flow from outside the "close neighborhood," measured as the percentage of pollen coming from >250 m or from outside the study site (PF250).
We examined the role of mother trees as pollen recipient in the shaping of these mating variables, by investigating various ecological processes acting on the mother trees or their neighborhood. In particular, we first tested whether there was a higher correlation of paternity within than among fruits, as observed in other plants with multiple-seeded fruits (Quesada et al., 2001
).
Second, we investigated how flowering phenology affects the correlation of paternity within and among maternal pollen clouds. Phenological assortative mating is expected to increase differentiation among pollen clouds of asynchronously flowering mother trees and to induce genetic drift through reduced Nep, in particular for the earliest or latest bloomers. Until now, however, experimental trials failed to support these expectations (Fuchs et al., 2003
). Additionally, in long-lived species such as trees, the potential interannual variation in phenology and thus in available partners should also be accounted for (Ward et al., 2005
). As a first insight on this temporal component of mating patterns, we assessed here the variation in pollen clouds intercepted by single mother trees across two consecutive years.
Finally, we also evaluated the effects of landscape heterogeneity on pollen cloud diversity. In particular, we first focused on the conspecific density around the mothers. Intuitively, isolated mother trees should have more balanced contributions from pollen donors than do mother trees within dense patches of flowering trees, where the pollen cloud should be saturated by a few, closest neighbors (García et al., 2005
). This expectation assumes that the amount of pollen received by a mother tree from a given pollen donor depends only on the distance they are apart, i.e., that mother trees behave as passive recipients (Klein et al., 2006
). For animal-pollinated species, however, it is questionable whether the probability of pollen dispersal will remain the same for isolated vs. non isolated individuals. Pollinators should indeed spend more time foraging in dense patches of flowering trees, but they will also switch among trees within these patches more often, because the cost for traveling to the next tree is lower (Charnov, 1976
; Pyke, 1984
). By contrast, isolated trees or low-density patches may attract fewer pollinators that should stay longer on the same tree. These two optimal-foraging arguments, however, can lead to various relationships between mother tree isolation and pollen cloud diversity, depending in particular on the spatial scale of pollinator movements.
We also investigated the impact of canopy and vegetation structure around flowering trees. Indeed, some studies showed that forest fragmentation can lead to genetic isolation of relict trees left in forest fragments (Rocha and Aguilar, 2001
), while in other cases, shifts in pollinator assemblage have been shown to prevent disruption of gene flow, and even to lead to increased level of pollen flow across fragmented landscapes (Dick, 2001
; White et al., 2002
).
This study of morphological, phenological, and landscape determinants of the diversity in maternal pollen clouds raises a major issue of reproductive biology, that of the balance between features of the pollen donors and features of the pollen receptors in shaping mating patterns. In particular, we show here how a classical linear model can be used to gauge the impact on mating patterns of spatial heterogeneity in pollen sources distribution vs. that of maternal phenotypic traits (e.g., tree size, quantity of flowers produced) and also to study the interaction between these factors.
MATERIALS AND METHODS
Study species
Sorbus torminalis is a light-demanding, postpioneer forest tree often found growing as a minor component of oak and beech woodlands. In lowlands of northern France, the tree flowers from late April through May (S. Oddou-Muratorio, personal observation). Each flower has 
12 stamens, and two styles, stigmas, and carpels (Fig. 1). Each carpel contains one or two ovules, so each flower can produce up to four seeds even if on average only 2–3 ovules/fruit develop into mature seeds. When flowers open, stamens mature slightly before styles. The flowers turn from white to yellow when pollen is released, and then to black as floral pieces fade (first stamen, then pistil). Pollinators are mainly small flies, but the fairly large and open flowers attract other generalist pollinators as well, including bees and beetles (Griffin and Sedgley, 1989
; S. Oddou-Muratorio, personal observation). The inflorescence of S. torminalis is a compound corymb with 30–50 flowers that do not open simultaneously and can be visited by multiple pollinators. Several hundreds of corymbs can be found on a single tree.
|
) showed that S. torminalis is essentially 99% outcrossed at the study site, which may result either from partial self-incompatibility or from inbreeding depression. We also showed that an average distance of pollen dispersal is 750 m, a median distance is 500 m, and an average effective number of pollen donors per maternal adult (Nep) is 10 in the 2 years of the study. A wide range of variation for Nep was observed among maternal plants (from 1 to 23).
Study site and field sampling
The study site covers 472 ha of mixed stands of oaks and other broad-leaved species. It is a part of the managed Rambouillet forest in France (total area: 15 000 ha). Since 1960, forest management has consisted of a regular cycle of thinning (through selective logging, that spared S. torminalis) ending with clear-cutting (high forest; see details in Oddou-Muratorio et al., 2004
). The study site is thus a mosaic of roughly even-age stands with various tree densities (Fig. 2).
|
for details). These reproductive trees occur in diffuse patches of related individuals (Oddou-Muratorio et al., 2004
). A total of 73 progeny arrays (corresponding to 60 different mother trees) were considered in that study. Among these 73 maternal families, 14 were collected in autumn 1999 (653 seeds; 11–100 seeds/tree, mean = 46.6, SD = 21.1) and 59 in autumn 2000 (1073 seeds; 7–27 seeds/tree; mean = 17.9, SD = 3.9). For 13 mother trees, progeny arrays were sampled for both years of study, allowing us to compare the behavior of these trees as pollen recipient between years. Moreover, for 14 maternal progenies (267 seeds) of the 59 collected in 2000, we recorded the mother fruit for each seed, to study the differences in kinship within and among fruits.
Microsatellite genotyping
The genotypes of the mother trees, the seedlings, and all the sampled potential fathers were scored at six microsatellite loci described in Oddou-Muratorio et al. (2003)
, with 6–21 alleles per locus. The theoretic exclusion probability (EP) per locus ranged from 0.45 to 0.82, while EP across the six loci was 0.983 for the 149 flowering adults of 1999 and 0.987 for the 172 flowering adults of 2000 (Oddou-Muratorio et al., 2003
).
Paternity assignment
We used the maximum-likelihood method implemented in the software CERVUS 2.0 (Marshall et al., 1998
) to assign paternity to the most-likely father for each genotyped offspring. CERVUS was calibrated with the following parameters: 25 000 simulated mating events (used for determining the significance level of the alleged paternities), a scoring error rate of 0% and a sampling fraction (i.e., the fraction of males that were sampled among the whole set of potential fathers) of 52% in 1999 and 47% in 2000. Father–offspring pairs were reconstituted at the 80% level of statistical confidence. These settings were based on a preliminary simulation study, in which these values allowed us to detect a large fraction of the paternity events without an excessive number of false positives (Oddou-Muratorio et al., 2003
). This procedure allowed us to assign paternity to 373 offspring (57%) among the 653 analyzed in 1999 and to 579 offspring (54%) among the 1075 analyzed in 2000 (Oddou-Muratorio et al., 2003
).
Ecological variables
A set of 12 descriptive "ecological variables" were used to characterize phenotypic traits, flowering phenology, and ecological/silvicultural neighborhood of the sampled mother trees (Table 1).
|
1.30 m), (2) the quantity of blooming corymbs (QBC) as an ordered qualitative variable with five visually assessed values (from 1 to 5: absent, missing data, few, significant, and massive); (3) the quantity of fruits (QF) as an ordered qualitative variable with four roughly assessed values (from 1 to 4: absent, missing data, few, significant).
Flowering phenology
In spring 2000, flowering schedules were recorded for 58 of the 59 mother trees sampled in 2000. We described flowering phenology with two quantitative variables: the first flowering day (FFD) and the length of flowering period (LFP). To determine these values, we observed each tree with binoculars every 3 days between 5 and 25 May and assessed whether it carried flowers at one or several of the five following stages: 1, green, closed flower; 2, white, closed flower; 3, open flower with white corolla; 4, open flower with yellow corolla; 5, faded flower. To be conservative, we considered that the pistil was receptive from stage 2 to 4, even if it is in fact unlikely that flowers are receptive at stage 2 (B. Demesure.-Mush., personal observation). The FFD of a given tree was set at the first day when at least one of its flowers was at stage 2 and the LFP as the number of days from FFD to last flowering day (that is last flower[s] observed at stage 4). Because each tree was observed only every 3 days, however, we also accounted for a 2-day uncertainty before the first flowering and after the last one, and computed the corrected value of FFD as FFD – 2, and the corrected value of LFP as LFP + 4.
The flowering overlap between two mother trees g and h (Ogh) was simply defined as the number of days when both were flowering.
Ecological maternal neighborhood (EMN)
For the 2 years of the study, we measured a set of five variables characterizing the EMN of each sampled mother tree. First, the density of conspecific trees was estimated through the distance to the nearest conspecific flowering neighbor (DNCN), and through the number of conspecific flowering neighbors found in concentric areas around the mother tree in a radius of 50, 100, and 150 m (respectively, NCN50, NCN100, and NCN150).
Additionally, as in Oddou-Muratorio et al. (2005)
, we characterized the neighborhood density (independently of species) through a class variable (denoted ND) with five levels (isolated tree, edge tree, dominant stand-tree, codominant stand-tree, suppressed stand-tree). The three last levels aimed at accounting for differences in light exposure among trees with crowns above, within, or below the surrounding canopy, respectively.
Silviculture
To assess the effects of silvicultural practices, the 59 maternal families analyzed in 2000 were split into three zones corresponding to three silvicultural stages (Fig. 2). Each tree belonged to a given zone, and the zone was treated as a class variable (ZONE) with three possible states in the subsequent analyses. Zone A (seven maternal trees) corresponds to young, mixed stands (S. torminalis DBH 20 cm) with a high density of S. torminalis (2 trees/ha) mixed with oak and beech (500 trees/ha in total). Zone B (34 maternal trees) corresponds to mature stands (S. torminalis DBH 40 cm) with a moderate density of S. torminalis (0.7 trees/ha) mixed with mature oaks (160 oaks/ha, DBH 50 cm). Zone C (18 maternal trees) corresponds to regeneration stands (S. torminalis DBH 40 cm) with a moderate density of S. torminalis (0.8 trees/ha) mixed with mature oaks in low density (60 oaks/ha, DBH 50 cm). The main difference between zones B and C was thus oak tree density (from 9 to 1 using a basal area criterion), whereas zone A had a higher density of S. torminalis, but an oak/beech density roughly comparable to zone B using a basal area criterion.
Mating variables
The diversity and composition of all maternal pollen clouds were characterized using three different mating variables (Table 1).
Relatedness among pollen gametes
The relatedness between the two fathers of every pair of offspring of the same or of different mothers was estimated using pairwise kinship coefficients (Hardy et al., 2004
). This method consists of (1) finding out the paternal alleles of each offspring by subtracting, for each locus, the maternal allele from the offspring genotype (Hardy et al., 2004
) and (2) computing the pairwise kinship coefficient (Fij) between the paternal gametes of offspring pairs (i and j denote two offspring from the same or from different mothers). Fij measures the probability that a paternal gene taken at random in i and a paternal gene taken at random in j are identical by descent. The Fij-values were estimated using the method described in Loiselle et al. (1995)
, implemented in the software SPAGEDI, version 1.1 (Hardy and Vekemans, 2002
).
We then estimated average Fij values within maternal families (denoted as Fs) and the average level of relatedness among the pollen clouds of a pair of pollen recipients g and h as:
|
| (1) |
Effective number of siring fathers
We used the pool of offspring assigned to a single most-likely father to obtain the effective number of pollen donors (Nep). Nep was estimated using equation (16) in Nielsen et al. (2003)
, which provides an unbiased estimator of the effective number of pollen donors for small sample sizes.
FS and Nep may appear redundant because in theory we expect 1/Nep = 2FS in the absence of inbreeding (Hardy et al., 2004
). Although paternity-based estimates (Nep) are theoretically more precise because they contain all information regarding correlated paternity, they may in practice be poorly estimated when the potential fathers are not sampled exhaustively (between 47% and 52% of fathers were sampled in our study) (Oddou-Muratorio et al., 2003
). By contrast, relatedness-based estimates obtained with the method of Hardy et al. (2004)
may be biased, for instance under biparental inbreeding, but they may also be more accurate because the estimation procedure uses all the available genetic information (including that of nonsampled fathers).
Pollen flow
To estimate pollen flow from outside the close neighborhood, we used the percentage of pollen coming from >250 m or from outside the study site. This threshold of 250 m corresponds to a compromise between the size of the neighborhood, where potential fathers have been effectively sampled around each mother tree, and the estimated distribution of pollen-dispersal distance. From the dispersal curves estimated in Oddou-Muratorio et al. (2005)
, we know that on average at least 35% and at best 60% of the pollen donors are expected to fall in these 250 m. With this threshold, PF250 can thus be compared among mother trees and is also expected to have some variation
We did not use the results of paternity analyses directly to estimate PF250 because this could yield an underestimate due to cryptic gene flow or an overestimate due to border effects. Instead, assuming a neighborhood model with a given pollen dispersal function fd within the neighborhood, the likelihood function of PF250 was written for each mother tree j as:
|
| (2) |
jl is the expected proportion of the pollen of a given father l in the local pollen cloud of mother j for a dispersal function fd (see Eq. 5 in Oddou-Muratorio et al., 2005
). For fd, we used an exponential power function with 
= 743 m and b = 0.31, previously determined to best fit data for the year 2000 (Oddou-Muratorio et al., 2005
). The background allelic frequencies (BAF) were estimated from the inferred paternal gametes of offspring for which no compatible father parent was found within the study site (Oddou-Muratorio et al., 2003
).
Analyses of the variation in mating traits
In all the analyses described in this section, mean Fij estimates were computed using SPAGEDI (Hardy and Vekemans, 2002
). Approximate standard errors and 95% confidence intervals were obtained using a jackknife procedure that consists of deleting each locus one at a time. Doing this, we assumed that the different loci provided independent replicates of the process determining genetic structure.
Impact of fruit multiple seededness
For 14 maternal progenies, we computed the average Fs value for each progeny, as well as the average Fs values within fruit (Fs WF), and among fruits within maternal progeny (Fs AF).
Impact of spatial distances between mother trees
The variation of
Impact of flowering phenology
Using the 58 mother trees collected in 2000 and monitored for flowering schedules, we studied the relationship between the level of relatedness among maternal pollen clouds (as measured by
).
Interannual variation
Using the 13 mother trees collected both in 1999 and 2000, we first tested the among-year correlations for each ecological/mating quantitative variable using Spearman rank coefficient (
). We then evaluated average correlation of paternity (
Relations between mating and ecological variables
For these analyses, we used only 57 of the 59 maternal progenies collected in 2000. One mother tree on the study site border and another with atypical values for FS were excluded. We used analyses of covariance (ANCOVA) to investigate the impact of the various ecological variables (Table 1) on Fs, Nep, and PF250. We included mother tree diameter, quantities of blooming corymbs and fruits, first flowering day, flowering period, distance to the nearest neighbor, numbers of neighbors as quantitative variables and silvicultural context and neighborhood density as class variables. For each mating variable, we successively tested different models with or without interactions. Nonsignificant effects were successively removed from the model, until the simplest model was accepted (including the lowest number of effects while still explaining a large part of the variability). These ANCOVAs were carried out with Proc GLM (SAS Institute, 1998
) using type III sums of square. For each significant explanatory quantitative variable, the differences in slopes among zones were tested in the full model using the corresponding contrasts. Additionally, we tested differences in Fs, Nep, and PF250 values across zones using multiple comparisons of means at 95% with Bonferroni corrections. Simple means or least-squares means were used depending on the model.
RESULTS
The values of mating and ecological variables for each maternal family are given in Appendix S1 (see Supplemental Data accompanying online version of this article).
Within- and among-fruit Fs values
The average Fs value in the 14 maternal progenies with seeds from identified fruits (Fig. 3A) was estimated as Fs = 0.082 (SE = 0.011). The average level of relatedness within fruits (within maternal progeny) was estimated as Fs-WF = 0.147 (SE = 0.015). The among-fruit component of Fs was also significant, but just half of that within fruit (Fs-AF = 0.070, SE = 0.010). As expected, Fs-AF was very close to the average Fs value because there were 2067 "among-fruits" pairs of seeds vs. 365 "within fruit" pairs.
|
30% within fruits and 14% among fruits, which translated into 3.4 effective pollen donors for a fruit on average (for 3.6 seeds per fruit in our sample), vs. 7 effective pollen donors for a whole tree on average (for 19.1 genotyped seeds per maternal progeny). These estimates were consistent with the results of paternity assignment, which yielded average proportions of full sibs within fruits of 30% and among fruits of 11% (Table 2).
|
Impact of flowering schedules
Levels of relatedness for paternal genes sampled among maternal pollen clouds (the Fgh's) increased significantly with increasing flowering overlap between mother trees (Fig. 3C) as shown with the significant Mantel test (Pearson correlation coefficient r = 0.038; P = 0.026). This effect increased slightly with the partial Mantel test that corrected for the effect of the spatial distances between mother trees (r = 0.058; P = 0.009).
Interannual variation
Ecological variables describing conspecific density (DNCN and the NCN's, see Table 1) were stable between 1999 and 2000, indicating similar patterns of flowering around the mother trees (Spearman > 0.8, P value < 0.001 for DNCN, NCN50 and NCN100; Spearman = 0.60, P value = 0.03 for NCN150). Fewer blooming corymbs were recorded in 2000 as compared to 1999, probably because of a sampling bias. Mother trees were selected in 1999 for their high production of blooming corymbs, whereas mother trees already sampled in 1999 were systematically re-sampled in 2000, regardless their production of flowers.
On the other hand, mating variables showed contrasting patterns of interannual variation: Nep and PF250 were significantly correlated between 1999 and 2000 (Spearman = 0.57 and 0.62, respectively, with respective P values of 0.041 and 0.026), whereas no significant interannual correlation was found for Fs. Moreover, levels of relatedness among paternal genes sampled by a given mother tree were significantly lower between years (Fgg' = 0.064, SD = 0.004) than within years (Fs = 0.091, SD = 0.003; see Fig. 3D), showing the limited repeatability of mating patterns between years. Additionally, this Fgg' value was not significantly higher than the correlation of paternity among the pollen clouds of different mother trees (Fgh= 0.045, SD = 0.001, in the [1; 50 m] distance interval; Fig. 3D).
Impact of ecological variables on Fs
The best ANCOVA model for Fs (R2 = 0.61; P < 0.0001) had a significant main effect of silvicultural context (Table 3A), with higher average Fs values (multiple comparison of means at 95%) in zone A (Fs 0.17) than in zones B and C (Fs 0.06). Distance to the nearest neighbor and the flowering intensity each had a significant negative main effect (P = 0.0001 and 0.017, respectively), with a trend toward decreasing Fs values for isolated and massively flowering trees.
|
|
9 in zones A, B, C). This apparent contradiction was due to important differences in means of explanatory variable among zones (DNCN = 12 m on average in zone A vs. 65 m on average in zones B and C; see Fig. 4B). Consistent with the results for Fs in zone A, we observed increasing Nep values with increasing distance to the nearest neighbor in zone A (slope ± SE = 0.71 ± 0.23; P = 0.004) but not in zones B and C (Fig. 4B). No significant effects of flowering intensity or mother tree diameter were observed. The variable DBH, though nonsignificant, was kept in the model because removing it led to a much lower R2 value (R2 = 0.19).
Impact of ecological variables on PF250
The best ANCOVA model for PF250 (R2 = 0.31; P = 0.001) included four main effects and no interactions (Table 3C). The proportion of long-distance pollen flow increased with DNCN (slope ± SE = 0.001 ± 0.0005; P < 0.037) and tended to decrease with increasing mother tree DBH (slope ± SE = –0.0049 ± 0.0025; P = 0.057) (Fig. 4C). Least-squares means for PF250 were lower in zone A (lsmean PF250 0.34) than in zone B (lsmean PF250 0.70; P = 0.01) or C (lsmean PF250 0.60; P = 0.06). This ranking was the same for simple means. Additionally, PF250 decreased with increasing distance to the border of the study site (slope ± SE = –0.0002 ± 0.0001; P = 0.042).
DISCUSSION
This study examines the role of mother trees as pollen recipients in the shaping of mating patterns in a scattered, insect-pollinated species. Our results highlight the respective effects of fruit morphology, phenological variance, maternal phenotypic traits, and landscape factors on the diversity and composition of maternal pollen clouds. Phenotypic and landscape factors in particular have important implications for the long-term maintenance of genetic diversity in managed, fragmented, or degraded landscape.
Within- and among-fruit correlation of paternity
Sorbus torminalis floral architecture, with multiple-seeded fruit, had a significant impact on mating patterns, as shown by the higher level of correlated paternity among seeds belonging to the same fruit (30% full sibs) in comparison to level among seeds belonging to different fruits of the same tree (14% full sibs). Within-fruit correlation of paternity was thus more likely to result from correlated dispersal events by a single pollinator. However, note that these values translate into 3.4 effective pollen donors for a fruit on average (for 3.6 seeds per fruit in our sample), versus 7 effective pollen donors for a whole tree on average (for 19.1 seeds sampled per maternal progeny). Thus, on average, the number of effective pollen donors contributing to a single fruit was close to its maximum. Also, among-fruit correlated paternities were more frequent than expected under random mating. This pattern could result either from correlated dispersal events (e.g., movement of a single pollinator between adjacent flowers on the same corymb) or from limited mate availability, i.e., low number of pollen donors (independent dispersal events), as suggested by the high Fgh value observed in the [0–50 m] distance class (see Fig. 3B).
Levels of correlated paternity within and among S. torminalis fruits were lower than those obtained by Quesada et al. (2001)
in a bat-pollinated tropical tree (Pachira quinata) with 20 seeds/fruit, with a correlation of paternity of 60% within fruit and 25% among fruit. These differences may result from different resolution achieved by the molecular markers used to assess relatedness (isozymes vs. microsatellites). However, they are also likely to reflect the impact of the size and abundance of pollinators: bats have a lower density than insects and probably deposit much more pollen on a pistil, potentially saturating the surface.
Significant impact of flowering phenology
Among the factors contributing to limited mate availability, flowering schedules were highlighted by this study as an important temporal source of isolation between pollen sources and recipients, as shown by the positive correlation between flowering overlap and correlated paternity among maternal pollen clouds (Fig. 3C), which further increased when we accounted for geographic distance between trees. Thus, the nonsynchronization with the neighboring trees counteracted the expected isolation by distance and contributed to the high average distance of pollen dispersal estimated for S. torminalis (Oddou-Muratorio et al., 2005
).
It was interesting to observe this correlation in this species with a relatively short flowering period (3 weeks) and where different flowering stages can usually be observed simultaneously on a single individual tree (from closed buds to faded flowers). Based on flowering records of 126 trees in 2000 (data not shown) and using the method of Keatley, Hudson, and Fletcher (2004)
, the flowering synchrony in our population was 79.3% (S. Oddou-Muratorio, unpublished data). This level is consistent with the high flowering synchrony observed within populations of various tree species (e.g., 80% in Pinus sylvestris, Gomory et al., 2003
; mean of 88% across three Eucalyptus species, Keatley et al., 2004
; mean of 74% across six neotropical shrubs, Augspurger, 1983
; mean of 61% across seven Rubiaceae species, SanMartin-Gajardo and Morellato, 2003
).
Flowering dates had an impact on the level of differentiation among pollen clouds, but not on the diversity within a pollen cloud: the latest or earliest bloomers did not have reduced Nep or increased Fs, probably because they could find enough simultaneously flowering mating partners. In the case of P. quinata, Fuchs et al. (2003)
failed to observe a significant impact of flowering schedules on patterns of relatedness among pollen clouds. Clearly, other experimental studies in various plant species are needed to elucidate the consequences of flowering schedules on mating patterns.
High interannual variations in mating neighborhood
Our results indicate that the genetic composition of the pollen cloud sampled by a single mother differed notably from one year to the next, as shown by the significantly lower correlation of paternity between than within years. These differences could be, in part, a consequence of the interannual variation in male fecundity, which is expected in natural populations where trees produce variable quantity of pollen among years and have variable flowering schedules (let alone stochastic processes). For instance, flowering surveys in 2001, a year of more limited flowering than 1999 (data not shown), showed that the precocity and length of the flowering period were not correlated from one year to the next.
This interannual variation in the fathers available to a given mother tree is probably a general feature in long-lived species (see also Rocha and Aguilar, 2001
). But it may be of particular importance in low-density species, where it allows the long life cycle to smooth the effects of a low number of effective pollen donors. Indeed, even if some seed trees contribute massively to the regeneration process, father contributions accumulated over years will be less uneven than within a single year, increasing the number of effective pollinators. Such cumulative effects will also decrease the spatial level of genetic structure in regeneration around isolated seed-trees.
Strong impact of spatial heterogeneity on mating patterns
Globally, we showed that the diversity and composition of pollen clouds received by maternal trees in S. torminalis is primarily determined by spatial patterns of pollen sources across the landscape: isolated individuals sample more diversity through more even paternal contributions, low relatedness among paternal genes and high rates of long-distance pollen dispersal within their progenies. Not only the density of conspecifics, but also the characteristics of the ecological neighborhood (notably silvicultural context) determined mating pattern variation in our population. Maternal phenotypic traits related to pollinator attractiveness also have an effect, but only in the context of strong competition, when larger mother trees with higher flowering intensity sample more diversity.
The silvicultural context was the main determinant of the variation in FS and Nep, with other significant ecological factors acting mainly in interaction with this variable. Mother trees located in zone A, a young stand with high density of both young S. torminalis and young beech and oak, had a significantly higher level of inbreeding within pollen clouds (FS 0.17) and lower rates of long-distance pollen flow (PF250 0.47) than mother trees located in mature stands/regeneration stands (zone B + C: FS 0.06; PF250 0.64). However, we cannot rule out that these effects of a silvicultural context resulted in part from patterns of covariation with the other explanatory variables, which were almost all strongly significantly dependent on silvicultural context (Table 3, Fig. 4).
Nevertheless, our results highlight the variation in mating patterns across heterogeneous landscapes. The lower fraction of long-distance pollen flow observed in zone A was consistent with theoretical predictions for large (zone A) vs. small (zone B&C) patches of resource for pollinators (Charnov, 1976
; Pyke, 1984
; Ellstrand and Elam, 1993
). In addition, in experimental observations large patches of animal-pollinated plants are less susceptible than small patches to incoming gene dispersal by cross-pollination because they result in longer pollinator residence (Cresswell and Osborne, 2004
). Moreover, the high correlation of paternity within pollen clouds in zone A suggests that pollinators tend to forage within small groups of neighbors, probably revisiting trees within these groups. This result is consistent with a similar study by García et al. (2005)
in another scattered, animal-pollinated forest tree (Prunus mahaleb), where dense vegetation cover around mother trees was shown to enhance locally restricted foraging. Finally, additional analyses presented in Appendix S2 (see Supplemental Data with online version of this article) also highlight the impact of silvicultural neighborhood, by showing a steeper selection gradient of distance on male mating success in zone A than in zone B&C (Table S2 in Appendix S2).
The second main determinant of maternal pollen cloud diversity and composition was the distance to the nearest conspecific neighbor (DNCN). In zone A where this distance was always <50 m, correlated paternity was higher and the effective number of pollen donors was lower as the distance to the nearest conspecific neighbor decreased (Fig. 4A, B). These patterns within zone A were consistent with the higher correlated paternity observed in zone A (average DNCN = 16 m) as compared to zone B&C (average DNCN = 65 m). Thus, very close neighbors tended to saturate the maternal pollen cloud, which can be interpreted as a direct consequence of isolation by distance between active pollen sources and passive recipients. Interestingly, this trend has also been observed in other temperate (Robledo-Arnuncio et al., 2004a
; García et al., 2005
) and tropical (Ward et al., 2005
) tree species.
However, isolation by distance is clearly not the only factor shaping the diversity of maternal pollen clouds. Indeed, our results also highlight that the pollen cloud of the largest mother trees of those with the most flowers was less saturated by the pollen of their neighbors than the others, showing that mother trees as pollen recipients cannot be considered as completely passive. Large trees with massive flowering are indeed likely to attract more effective pollinators, in agreement with predictions and observations of bumblebee behavior made by Ohashi and Yahara (2002)
. These authors showed that pollinators tend to probe proportionally fewer flowers when large displays of nectar resource are available within a patch (i.e., a tree). Thus, at the tree level, the effective number of pollinators will increase with tree size and flowering intensity. Moreover, the increase in PF250 with tree DBH also suggested that large mother trees attract pollinators from a greater distance.
Classical statistical analyses in the frame of the linear model were shown to be an efficient tool for disentangling the various factors shaping the diversity and composition of maternal pollen clouds. We also used multivariate analyses tools to finely investigate joint patterns of variations between mating and ecological variables and in particular to analyze joint patterns of variation in the mating variables. We used canonical correlation analysis as described in García et al. (2005)
, which largely confirmed our results based on ANCOVA analyses. The interpretation of multivariate analyses, however, was judged to be trickier (see corresponding author for details).
Conclusion
This study highlights how mating patterns are affected by temporal (phenological variance) and spatial (distribution of pollen sources) heterogeneity in the neighborhood of mother trees, on one hand, and by the mothers' specific traits (fruit structure, mother size, and flowering intensity), on the other hand. In particular, isolated trees of S. torminalis represent sinks of genetic diversity, gaining even paternal contributions, low relatedness among paternal genes and high rates of long-distance pollen dispersal within their progenies. These results have important implications both for evolutionary and conservation points of view. Indeed, considering the large genetic load in forest trees (Ledig, 1986
), this increases the probability of generating highly fit pioneer seedlings in this colonizing species, especially because selfing remains limited, and that even rather highly isolated trees do not seem to suffer from pollen limitation. Thus, management practices should favor the presence of such isolated trees as relays between patches of related individuals, give them priority as seeds/seedlings sources for natural regeneration, and consider using them as seed sources for ex situ genetic conservation and restoration. Finally, this study also suggests that the genetic diversity of S. torminalis will be resilient in the face of habitat fragmentation and reduced species density.
FOOTNOTES
1 The authors thank J. Shykoff and S. Aitken for helpful comments on successive versions of this manuscript. This study was supported by the French National Forest Services (ONF) and by the Bureau des Ressources Génétiques. 
6 Author for correspondence (oddou{at}avignon.inra.fr
; present address: INRA–URFM. 20, Avenue Vivaldi F-84000 Avignon, France; phone: +33-432 722 904; Fax: +33-432 722 902
LITERATURE CITED
Augspurger C.. 1983. Phenology, flowering synchrony, and fruit set of six neotropical shrubs. Biotropica 15: 257-267.
Barrett S. C. H.. 2002. The evolution of plant sexual diversity. Nature Review Genetics 3: 274-284.[CrossRef][ISI][Medline]
Bonnet E. Van de Peer Y.. 2002. zt: a software tool for simple and partial Mantel tests. Journal of Statistical Software 7: 1-12.
Charnov E. L.. 1976. Optimal foraging, the marginal value theorem. Theoretical Population Biology 9: 129-136.[CrossRef][ISI][Medline]
Cresswell J. E. Osborne J. L.. 2004. The effect of patch size and separation on bumblebee foraging in oilseed rape: implications for gene flow. Journal of Applied Ecology 41: 539-546.[CrossRef][ISI]
Dick C. W.. 2001. Genetic rescue of remnant tropical trees by an alien pollinator. Proceedings of the Royal Society of London, B, Biological Sciences 268: 2391-2396.[CrossRef]
Dick C. W. Etchelecu G. Austerlitz F.. 2003. Pollen dispersal of Neotropical trees (Dinizia excelsa: Fabaceae) by native insects and African honeybees in pristine and fragmented Amazonian rainforest. Molecular Ecology 12: 753-764.[CrossRef][Medline]
Ellstrand N. C. Elam D. R.. 1993. Population genetic consequences of small population size: implication for plant conservation. Annual Review of Ecology and Systematics 24: 217-242.[CrossRef][ISI]
Fuchs E. J. Lobo J. A. Quesada M.. 2003. Effects of forest fragmentation and flowering phenology on the reproductive success and mating patterns of the tropical dry forest tree Pachira quinata. Conservation Biology 17: 149-157.[CrossRef][ISI]
García C. Arroyo J. M. Godoy J. A. Jordano P.. 2005. Mating patterns, pollen dispersal, and the ecological maternal neighbourhood in a Prunus mahaleb (L.) population. Molecular Ecology 14: 1821-1830.[CrossRef][Medline]
Gomory D. Bruchanik R. Longauer R.. 2003. Fertility variation and flowering asynchrony in Pinus sylvestris: consequences for the genetic structure of progeny in seed orchards. Forest Ecology and Management 174: 117-126.[CrossRef][ISI]
Goodell K. Elam D. R. Nason J. D. Ellstrand N. C.. 1997. Gene flow among small populations of a self-incompatible plant: an interaction between demography and genetics. American Journal of Botany 84: 1362-1371.[Abstract]
Griffin A. R. Sedgley M.. 1989. Sexual reproduction of tree crops Academic Press, London, UK.
Hardy O. Vekemans X.. 2002. SPAGeDi: a versatile computer program to analyse spatial genetic structure at the individual or population levels. Molecular Ecology Notes 2: 618.[CrossRef][ISI]
Hardy O. J. Gonzalez-Martinez S. C. Colas B. Freville H. Mignot A. Olivieri I.. 2004. Fine-scale genetic structure and gene dispersal in Centaurea corymbosa (Asteraceae). II. Correlated paternity within and among sibships. Genetics 168: 1601-1614.
Hodgins K. A. Barrett S. C. H.. 2006. Mating patterns and demography in the tristylous daffodil Narcissus triandrus. Heredity 96: 262-270.[CrossRef][ISI][Medline]
Keatley M. Hudson I. Fletcher T.. 2004. Long-term flowering synchrony of box-ironbark eucalypts. Australian Journal of Botany 52: 47-54.[CrossRef]
Klein E. K. Lavigne C. Gouyon P.-H.. 2006. Mixing of propagules from discrete sources at long distance: comparing a dispersal tail to an exponential. BioMed Central Ecology 6: 3 [online journal], http://www.biomedcentral.com/1472-6785/6/3.
Ledig F.. 1986. Heterozygosity, heterosis, and fitness in outbreeding plants. In M. Soulé [ed.] Conservation biology: the science of scarcity and diversity 77-104 Sinauer, Sunderland, Massachusetts, USA.
Loiselle B. A. Sork V. L. Nason J. Graham C.. 1995. Spatial genetic structure of a tropical understory shrub, Psychotria officinalis (Rubiaceae). American Journal of Botany 82: 1420-1425.[CrossRef][ISI]
Marshall T. C. Slate J. Kruuk L. E. B. Pemberton J. M.. 1998. Statistical confidence for likelihood-based paternity inference in natural populations. Molecular Ecology 7: 639-655.[CrossRef][Medline]
Motten A. F. Antonovics J.. 1992. Determinants of outcrossing rate in a predominantly self-fertilizing weed, Datura stramonium (Solanaceae). American Journal of Botany 79: 419-427.[CrossRef][ISI]
Nielsen R. Tarpy D. R. Reeve H. K.. 2003. Estimating effective paternity number in social insects and the effective number of alleles in a population. Molecular Ecology 12: 3157-3164.[CrossRef][Medline]
O'Connell L. M. Russell J. Ritland K.. 2004. Fine-scale estimation of outcrossing in western red cedar with microsatellite assay of bulked DNA. Heredity 93: 443-449.[CrossRef][ISI][Medline]
Oddou-Muratorio S. Demesure-Musch B. Pélissier R. Gouyon P.-H.. 2004. The roles of colonization and history in the local genetic structure of a scattered temperate tree species. Molecular Ecology 13: 3689-3702.[CrossRef][Medline]
Oddou-Muratorio S. Houot M.-L. Demesure-Musch B. Austerlitz F.. 2003. Pollen flow in the wildservice tree, Sorbus torminalis (L.) Crantz. I. Evaluating the paternity analysis procedure in continuous populations. Molecular Ecology 12: 3427-3439.[CrossRef][Medline]
Oddou-Muratorio S. Klein E. K. Austerlitz F.. 2005. Pollen flow in the wildservice tree, Sorbus torminalis (L.) Crantz. II. Pollen dispersal and heterogeneity in mating success inferred from parent-offspring analysis. Molecular Ecology 14: 4441-4452.[Medline]
Ohashi K. Yahara T.. 2002. Visit larger displays but probe proportionally fewer flowers: counterintuitive behaviour of nectar-collecting bumble bees achieves an ideal free distribution. Functional Ecology 16: 492-503.[CrossRef][ISI]
Pyke G.. 1984. Optimal foraging theory: a critical review. Annual Review of Ecology and Systematics 15: 523-575.[CrossRef][ISI]
Quesada M. Fuchs E. J. Lobo J. A.. 2001. Pollen load size, reproductive success, and progeny kinship of naturally pollinated flowers of the tropical dry forest tree Pachira quinata (Bombacaceae). American Journal of Botany 88: 2113-2118.
Robledo-Arnuncio J. J. Alia R. Gil L.. 2004a. Increased selfing and correlated paternity in a small population of a predominantly outcrossing conifer, Pinus sylvestris. Molecular Ecology 13: 2567-2577.[CrossRef][Medline]
Robledo-Arnuncio J. J. Smouse P. E. Gil L. Alia R.. 2004b. Pollen movement under alternative silvicultural practices in native populations of Scots pine (Pinus sylvestris L.) in central Spain. Forest Ecology and Management 197: 245-255.[CrossRef][ISI]
Rocha O. J. Aguilar G.. 2001. Variation in the breeding behavior of the dry forest tree Enterolobium cyclocarpum (guanacaste) in Costa Rica. American Journal of Botany 88: 1600-1606.
Sanmartin-Gajardo I. Morellato P.. 2003. Inter and intraspecific variation on reproductive phenology of the Brazilian Atlantic forest Rubiaceae: ecology and phylogenetic constraints. Revista de Biología Tropical. 51: 691-698.[ISI][Medline]
SAS Institute.. 1998. SAS/STAT users guide, release 6.03 SAS Institute, Cary, North Carolina, USA.
Smouse P. E. Robledo-Arnuncio J. J.. 2005. Measuring the genetic structure of the pollen pool as the probability of paternal identity. Heredity 94: 640-649.[CrossRef][ISI][Medline]
Ward M. Dick C. W. Gribel R. Lowe A. J.. 2005. To self, or not to self. A review of outcrossing and pollen-mediated gene flow in neotropical trees. Heredity 95: 246-254.[CrossRef][ISI][Medline]
Weis A. E. Kossler T. M.. 2004. Genetic variation in flowering time induces phenological assortative mating: quantitative genetic methods applied to Brassica rapa. American Journal of Botany 91: 825-836.
White G. M. Boshier D. H. Powell W.. 2002. Increased pollen flow counteracts fragmentation in a tropical dry forest: an example from Swietenia humilis Zuccarini. Proceedings of the National Academy of Sciences, USA 99: 2038-2042.
| ||||||||||||||||||||||||||||||||||||
) and 2000 (
). Other adults: 
. The three different silvicultural zones are indicated (see Materials and Methods, section Study site and field sampling for details)
