• In plants, pollen donor composition can differ during the early phases of reproduction through various selection mechanisms favouring self, related or nonrelated pollen donors, but such differences have not been examined under natural conditions because paternity is difficult to analyse in a natural setting.
• Here, we performed paternity analyses based on microsatellite genotyping of individual pollen grains deposited on female flowers (n = 773) and seeds (n = 304) to evaluate pollen donor composition from three individuals of the insect-pollinated monoecious tree Castanea crenata in a natural forest. Spatial genetic structure was also investigated.
• A mean self-pollen rate of 90.2% was observed at the pollination stage, but a low selfing rate of 0.3% was observed at the seed stage. In outcross events, however, pairwise distance and relatedness between maternal and paternal parents were not different between pollination and seed stages. We also observed significant positive relatedness, based on clear fine-scale genetic structure of individual trees within 80 m of one another, and 71% of seeds were derived using pollen grains of related trees within 80 m.
• The results suggest that the mechanism of self-incompatibility strongly avoids self-pollen before seed production. However, the avoidance of biparental inbreeding was not obvious between pollination and seed stages.
Plants are immobile and consequently depend on abiotic and biotic vectors to transport their pollen for sexual reproduction. Under field conditions, however, the pollen grains deposited on female flowers are not always adequate for seed production (Wilcock & Neiland, 2002), such as when self-pollination or pollination from closely related neighbouring plants occurs, which can reduce the performance of the next generation through inbreeding depression (Heywood, 1993; Nason & Ellstrand, 1995; Herlihy & Eckert, 2004). In many angiosperms, pollen selection mechanisms that reduce inbreeding depression and promote outcrossing have evolved. These mechanisms include self-incompatibility, in which only outcross-pollen grains are used for seed production (Seavey & Bawa, 1986; de Nettancourt, 1997), and inbreeding avoidance, in which pollen grains from more distant and nonrelated parents are more likely to sire seeds (Waser & Price, 1993; Souto et al., 2002; Glaettli et al., 2006). In these systems, outcross-pollen grains typically have a much greater reproductive advantage relative to self- or related-pollen grains at pre-zygotic stages, such as pollen germination (de Nettancourt, 1997) or pollen tube growth (cryptic self-incompatibility, Bateman, 1956; inbreeding avoidance, Waser & Price, 1993; Souto et al., 2002; Glaettli et al., 2006). Ovarian inhibition of self-pollen tube or self-embryo abortion (late-acting self-incompatibility; Seavey & Bawa, 1986) and early-acting inbreeding depression also enhance the outcrossing rate. As a result, these traits may cause a difference in pollen donor composition between the pollination stage (i.e. reach stigmas) and seed stage (i.e. sire seeds). In addition, the difference may also be a result of mate quality factors, such as pollen tube competition (Mulcahy, 1979), and the different growth rates of fertilized ovules (Korbecka et al., 2002).
To demonstrate experimentally the effectiveness and evolution of pollen selection mechanisms, it is important to evaluate the extent of self- and related-pollen grain depositions in plant populations exhibiting fine-scale genetic structure. Under natural conditions, individual female flowers may receive a greater proportion of pollen grains from neighbouring plants relative to distant plants. If closely related individual plants (i.e. with fine-scale genetic structure) are distributed in a spatially aggregated manner, most of their seeds are likely to derive from the relatively unfavourable outcross-pollen grains from neighbouring plants. Furthermore, pollen limitation is frequently observed in natural populations (Burd, 1994; Wilcock & Neiland, 2002; Ashman et al., 2004), especially in self-incompatible tree species (Larson & Barrett, 2000) in which, as a result, outcross-pollen grains from related neighbouring plants may compensate for the limitation of more favourable pollen grains from more distant plants.
Therefore, in this study, we evaluated the process of pollen selection in the early phases of reproduction, and analysed the fine-scale genetic structure using microsatellite genotyping of pollen grains, seeds and potential paternal trees in the self-incompatible monoecious tree species Castanea crenata. We did not investigate inbreeding depression in C. crenata; however, Quercus crispula, belonging to the same family (Fagaceae), exhibits biparental inbreeding depression at both the seed germination and seedling stage (Ubukata et al., 1999). In addition, inbreeding depression is thought to be strong in long-lived perennial tree species (e.g. Koelewijn et al., 1999; Ishida, 2006), probably as a result of the high rate of genomic mutation per generation (Morgan, 2001).
Here, we specifically addressed the following questions:
• Does pollen donor composition differ between the pollination and seed stages?
• If so, which pollen donors (i.e. self, related or nonrelated individuals) are more likely to sire seeds in a natural population of C. crenata?
• Is there genetic structure?
• If so, are the outcross-pollen grains of related individuals used for seed production in the self-incompatible tree species C. crenata?
Materials and Methods
Castanea crenata Sieb. et Zucc. (Japanese chestnut) is a hardwood tree common in the temperate forests of Japan (Horikawa, 1972; Seiwa et al., 2002; Tanaka et al., 2005). Castanea spp. are late-acting self-incompatible (McKay, 1942; Seavey & Bawa, 1986) with two types of inflorescence: unisexual staminate catkins at proximal positions of the shoot and bisexual catkins at terminal positions (Klinac et al., 1995). Multiple female flowers with three ovaries are present at the bases of the bisexual catkins and, on pollination, each flower develops into a single cupule within which one of the 16–22 ovules per ovary develops into a seed (Nakamura, 2003). Flowering occurs between late June and early August at our study site. On individual trees, the flowering periods of both male and female flowers generally overlap. The flowers typically require pollinators for seed production (de Oliveira et al., 2001) and are visited by various insects (e.g. honeybees, bumblebees, syrphid flies, butterflies, moths and beetles; Y. Hasegawa et al., unpublished). The seeds (three per cupule) mature in mid-October of the same year that flowering occurred.
The study site was located in a deciduous broad-leaved forest dominated by Quercus crispula, Fagus crenata and Castanea crenata in Ippitsu Forest Reserve, Miyagi, Japan (38°49′N, 140°45′E; 718 m above sea level). We established a 6-ha (240 × 250 m) plot in which all stems of C. crenata greater than 5 cm in diameter at breast height (DBH) were mapped (range, 8.1–99.4 cm; n = 281; Fig. 1). As all of these trees reach the canopy layer, we considered them to be potential pollen donors in this study. This sample plot was located in the middle of a large population of C. crenata, with numerous reproductive trees of C. crenata outside the plot edge.
Leaf tissues for DNA microsatellite genotyping were collected from each of the 281 trees representing the entire (100%) reproductive population in the plot. In late July 2004, staminate catkins and female flowers were collected from three trees (K458, K467 and L503) located in the northeast corner of the plot in order to acquire the longest possible measurement of the pollen dispersal distance (Fig. 1). Pollen samples (n = 34, 23, 33) isolated from a single staminate catkin of each tree were collected for a preliminary check of the DNA amplification technique. For paternity analysis at the pollination stage, pollen samples (n = 1219, 912, 1440) were isolated from 33, 17 and 20 female flowers from the three trees. For paternity analysis at the seed stage, individual seeds (n = 102, 101, 101) were isolated from 40, 35 and 43 cupules from the three trees (K458, K467 and L503, respectively) in mid-October 2004. Tissue samples were stored at –20°C before DNA analysis.
Treatment of pollen grains
Pollen grains were collected from anthers of staminate catkins and styles of female flowers. In C. crenata, pollen grains germinate only on the tip of the needle-shaped style, indicating that this is the stigma (Nakamura, 1992). If a pollen tube has started to grow on a stigma, DNA cannot be extracted from the pollen grain. Thus, we collected pollen grains at the pollination stage from the surface of styles. In this case, the pollen donor composition at the pollination stage is represented by the accumulated pollen grains deposited during the period from flower opening to the sampling time, because the style surfaces were usually moist and the pollen grains easily adhered to the surface during the flowering period. Furthermore, most female flowers bloomed synchronously at the sampling time within and among the three mother trees. These traits suggested that the sampling method effectively represented the pollen grain composition at the pollination stage.
The styles and anthers were placed in 0.01% sodium dodecyl sulphate (SDS) solution on slides with a water-repellent finish. Thereafter, pollen grains that were morphologically identified as C. crenata and bore no structural damage were collected with a micropipette under a stereodissecting microscope.
DNA extraction and genotyping
One pollen grain and 0.5 µl of 0.01% SDS solution were placed in a PCR tube that contained 1.0 µl of reaction buffer (10 mM Tris-HCl, pH 8.3 at 20°C; 1.5 mM MgCl2; 50 mM KCl; 0.01% Proteinase K), incubated for 60 min at 54°C and heated for 10 min at 95°C. The extract was used directly as a PCR template. Total DNA from leaf and seed tissues was isolated using a DNeasy 96 Plant Mini Kit (Qiagen K.K., Tokyo, Japan) according to the manufacturer's protocol.
PCR was performed in a GeneAmp PCR System 9600 (Applied Biosystems, Foster City, CA, USA). The forward primers (Table 1) were labelled with fluorescent dye (G5 dye set: 6-FAM, VIC, NED or PET; Applied Biosystems) to simultaneously analyse 11 microsatellite loci of similar allelic size and to avoid overlaps among loci with the same dye. Multiplex PCR amplification was carried out using a Multiplex PCR Kit (Qiagen K.K.) in a 6.0 µl volume containing 1 × Qiagen Multiplex PCR Master Mix, 0.2 µM of each primer and 1.5 µl of template extract from a pollen grain or 0.5 µl of template from leaves or seeds. We used the thermal cycler under the following cycle conditions: 94°C for 15 min (hotstart), 40 (pollen templates) or 35 (leaf and seed templates) cycles at 94°C for 30 s, 57°C for 90 s and 72°C for 1 min, and a final step at 60°C for 30 min. PCR products were electrophoresed on an ABI PRISM 3100-Avant Genetic Analyzer (Applied Biosystems), and allele sizes were determined with the fragment analysis software packages GeneScan 3.0 and Genotyper 2.1 (Applied Biosystems).
Table 1. Characteristics of the microsatellite loci, including the observed number of alleles (NA), observed (HO) and expected (HE) heterozygosity, probability of exclusion when one parent is known (PEX) and estimated null allele frequency by locus, for the 281 individual Castanea crenata trees in the Ippitsu Forest Reserve, Miyagi, Japan
Pollen grains are haploid but, in some pollen samples, two or three alleles were found at one locus, indicating that two or three pollen grains had been placed in one PCR tube and then amplified at the same time. Therefore, we did not use these samples in the following analyses.
For paternity analysis, we used 11 loci for pollen grains and 10 loci for seeds. Apparent null alleles were detected as genotype mismatches between the mother tree and seed for the KT004a locus. This method produced a higher estimate of null allele frequency for this locus than for the other loci (Table 1). We eliminated the KT004a locus from the overall paternity analysis of seeds as a heterozygote seed bearing a null allele shows the same pattern as a homozygous seed for the other allele. We still used this locus in the analysis of pollen grains, because null alleles can be recognized by the absence of an allele.
As several loci were not amplified by pollen genotyping for several pollen grains, we added a locus (KT004a) in the analysis of pollen grains to increase the exclusion probability. For the following analysis, we used pollen samples that had more than eight genotyped loci (mean exclusion probability, 0.989). All sampled seeds were used for analysis because they had all 10 genotyped loci (exclusion probability, 0.995).
Pollen grains and seeds were assigned as self if they did not contain nonmaternal alleles. The paternity of each outcross-pollen grain and outcross-seed was assigned by a simple exclusion approach based on the multilocus genotypes of the 281 trees. If a pollen grain or seed lacked any potential pollen donor genotypes among the 281 candidate trees, we assumed that the pollen donor came from outside the study plot. If a pollen grain or seed had two or more possible pollen donor candidates, we inferred paternity based on the maximum likelihood paternity assignment using CERVUS 2.0 (Marshall et al., 1998). The natural logarithm of the likelihood ratio of the loci was termed the logarithm of odds (LOD) score (Meagher, 1986). Marshall et al. (1998) then defined a statistic Δ as the difference in LOD scores between the most likely male and the second most likely male. The program CERVUS conducts the simulation to find critical values of Δ for a strict and relaxed confidence level (95% and 80% by default, respectively). The simulation parameters for CERVUS were as follows: 10 000 tests; the number (n = 281) of all individuals in the study plot; the proportion of candidate parents sampled (93% for pollen grains, 88% for seeds); the proportion of the loci typed (88% for pollen grains, 100% for seeds); a typing error rate of 0%; and a confidence level of 80%. When more than one individual shared the same LOD score, or when the significance of the paternity analysis was less than 80%, paternity was assigned to the spatially nearest individual with the highest positive LOD score.
Allele segregation ratio of pollen from anthers
Pollen samples from anthers (i.e. self-pollen grains) were used to check the DNA amplification technique from single pollen grains. A chi-squared test was used to determine whether alleles had segregated in a 1 : 1 ratio in loci with two alleles in mother trees.
Self-pollen rate at the pollination stage and selfing rate at the seed stage
Self-pollen rates at the pollination stage and selfing rates at the seed stage were determined directly by paternity assignment. The mean self-pollen rate, or the ratio of self-pollen grains to total pollen grains (including the outcross-pollen grains) at the pollination stage, and the mean selfing rate, or the ratio of self-seeds to total seeds (including outcross-seeds) at the seed stage, were calculated at whole-plant levels. We compared the self-pollen rates at the pollination stage and the selfing rates at the seed stage by applying a generalized linear mixed model (GLMM) with binomial errors and a logistic link, using the glmmML package in R (R Development Core Team, 2006). Here, individual trees were treated as a random effect, stage (pollination or seed) as a fixed effect and the pollen donor (self or outcross) as the response variable.
Relatedness between maternal and paternal parents
For each outcross-pollen grain and outcross-seed, relatedness between the maternal and paternal parents was estimated using MER 3.0 (Wang, 2002). To determine whether mean relatedness at the seed stage was lower than at the pollination stage, a one-tailed bootstrap test was used according to Hogg et al. (2004). Statistical significance was determined by 10 000 bootstraps.
Pollen dispersal distances
We also compared the frequency distributions of the pollen dispersal distances of outcross-pollen grains and outcross-seeds between the two stages using a Kolmogorov–Smirnov test. These analyses were conducted after pooling the data from the three mother trees. All statistical analyses were performed using R 2.3.1 (R Development Core Team, 2006).
Fine-scale genetic structure
The analysis of fine-scale genetic structure was carried out at the individual level using a relationship coefficient computed as Moran's I (Moran, 1950; Sokal & Oden, 1978). To guarantee that the first distance class includes the majority of near-neighbour pairs, its upper limit was chosen as 1.5 times the square root of the inverse of the sampling density, according to Epperson & Chung (2001). We adopted the equal frequency method (Escudero et al., 2003), that is, uneven lags that comprise a constant number of individuals. We used 15 distance classes (upper limits for each distance class: 22, 35, 50, 66, 80, 93, 104, 114, 124, 134, 145, 162, 199, 225 and 250 m) to ensure that each distance class included an approximately similar number of pairs (n = 2457–2687). To test its significance, each Moran's I was compared with the distribution of the statistics under the null hypothesis for no spatial structure that was generated using 10 000 resamplings of the data, permuting spatial locations among the distance groups. These analyses were conducted using SPAGeDi 1.2b (Hardy & Vekemans, 2002).
DNA amplification from pollen grains of anthers and segregation ratios of their alleles
In 90 (59%) of 152 individual pollen grains of C. crenata isolated from stamens, we successfully amplified and genotyped DNA fragments at more than eight loci. In two (2.2%) of the 90 samples, two alleles were obtained at one locus, and nonmaternal alleles were found in only one pollen grain (i.e. outcross-pollen grain). For the other 87 samples, the expected 1 : 1 ratios of allele segregation were observed in 15 (88%) of the 17 loci (Table 2).
Table 2. Chi-squared test for allele segregation ratios of pollen grains from anthers in each tree
Pollen grains per allele
A locus in which the allele segregation ratio differed significantly from 1 : 1 (P < 0.05).
DNA amplification from pollen grains deposited on female flowers
For 946 (26.5%) of the 3571 pollen samples isolated from female flowers, DNA fragments were successfully amplified and genotyped at more than eight loci. In 169 (17.9%) of the 946 samples, we obtained two alleles at one locus, and in four (0.4%) of the 946 samples, we found three alleles at one locus. Because we did not use the data from the 173 multi-allele samples, we obtained microsatellite genotypes for a total of 773 pollen grains (n = 270, 153 and 350 from K458, K467 and L503, respectively).
Comparison of pollen donor composition between pollination and seed stages
In four of the 57 outcross-pollen grains and 36 of the 303 outcross-seeds, we assumed that the pollen donor came from outside the study plot. For 12 outcross-pollen grains and 39 outcross-seeds, we found more than one paternal match. We conservatively assigned the spatially nearest of these as the paternal parents.
The self-pollen rate at the pollination stage (mean ± SE, 90.2 ± 6.7%, n = 3) was significantly higher than the selfing rate at the seed stage (0.3 ± 0.3%, n = 3; GLMM, P < 0.01; Fig. 2).
The genetic relatedness of the outcross-pollen donors and the maternal trees was not significantly different between the pollination stage (0.035 ± 0.038) and the seed stage (0.076 ± 0.032) (one-tailed bootstrap test, P = 0.53; Fig. 3). The frequency distributions of the distance between the outcross-pollen donors and the maternal trees (i.e. pollen dispersal distances) were not significantly different between the pollination and seed stages (Kolmogorov–Smirnov test, D = 0.11, P = 0.62; Fig. 4).
Fine-scale genetic structure and distribution of paternal trees
Individual trees of C. crenata were preferentially distributed on ridges and west-facing slopes, resulting in three main patches of trees within the study plot (Fig. 1). The diameter of individual patches was c. 90 m. Significantly positive kinship coefficients were observed within a range of 80 m, and significantly negative kinship coefficients were found more than 104 m apart (randomization test, P < 0.05). Therefore, individual trees within each patch were closely related to each other (Fig. 5).
At the seed stage, the crossing rate between individual trees in the same patch was much higher (71%) than that for trees in two different patches and outside of the study plot (29%), most probably because the pollen dispersal distance was much shorter within patches (range of dispersal distance, 2.9–56.5 m) than between patches (92.4–242.6 m; see Figs 1 and 4).
We investigated the pollen donor composition of early phases of reproduction in a natural population of C. crenata. Using microsatellite analysis, we showed a drastic change in self-pollen rates between the pollination and seed stages. In crossing events, however, the relatedness between the pollen donor and maternal tree was no different between pollination and seed stages. This is probably the result of pollen quality limitation (Ramsey & Vaughton, 2000; Ashman et al., 2004; Aizen & Harder, 2007) caused by limited movements of pollinators. To our knowledge, this is the first study documenting the pollen grain selection for seed production in a natural population of a plant species based on microsatellite genotyping of individual pollen grains and seeds.
DNA amplification from a single pollen grain
In this study, the pollen donor composition at the pollination stage of a self-incompatible tree, C. crenata, was identified under natural conditions using paternity analysis of single pollen grains. The applicability of paternity analysis was demonstrated using a preliminary check of the DNA amplification technique. Most pollen samples isolated from staminate catkins exhibited one allele per locus, and most alleles segregated at a 1 : 1 ratio. However, in this study, the percentage of successful DNA genotyping from a pollen grain was relatively low (59% of pollen grains from stamens and 26.5% of pollen grains from female flowers were genotyped at more than eight loci) compared with a previous study (Matsuki et al., 2008), in which 83% of the pollen grains were successfully genotyped at more than five loci. To increase the percentage of successful genotyping, sufficient cytoplasm of the pollen grain should be released into the PCR solution by effectively disrupting the pollen cell wall to facilitate germination (Ziegenhagen et al., 1996; Aziz & Sauve, 2008), crushing with a pipette tip (Suyama et al., 1996; Parducci et al., 2005; Matsuki et al., 2007, 2008; Paffetti et al., 2007; Ito et al., 2008), pressing between two glass slides (Zhou et al., 2007) or freezing and re-lysing in a pollen suspension buffer (Paffetti et al., 2007).
Pollen donor composition at the pollination stage
In C. crenata, we observed that the self-pollen rates at the pollination stage were very high (mean, 90.2%; n = 3). The rates were almost identical to the self-fertilization rate of an insect-pollinated tree, Magnoria obovata (82–86%; Ishida, 2006), but were greater than those of several herbaceous plants (0–70%; reviewed by de Jong et al., 1993). In C. crenata, individual trees usually have large floral displays (Y. Hasegawa, pers. obs.), which could promote a high frequency of pollinator movements within an individual tree (Frankie et al., 1976; Karron et al., 2004). Also, we observed that most pollinators (e.g. bumblebees, syrphid flies and beetles) moved frequently within individual trees, but only rarely between trees (Y. Hasegawa et al., unpublished). Furthermore, we documented significant overlap of the flowering periods for both male and female flowers within individual trees (Y. Hasegawa et al., unpublished). These traits may in part account for the high rate of geitonogamous self-pollination.
In C. crenata, the selfing rate at the seed stage was very low (0.3%), although the self-pollen rate at the pollination stage was very high (90.2%). Such a drastic decrease in the selfing rate indicates self-incompatibility, which is very important for the rejection of selfing in C. crenata.
If outcross-pollen grains from related individuals are under-utilized for seed production in C. crenata, it is expected that the pollen dispersal distance will be increased in the seed stage relative to the pollination stage, simply because individual trees are more closely related to their immediate neighbours (i.e. fine-scale genetic structure), as demonstrated for the patches at our study site. However, there was little difference in the pollen dispersal distance between the pollination and seed stages. We also observed no difference in the relatedness of paternal and maternal parents in the outcross events between the pollination and seed stages. These observations strongly indicate that related outcross-pollen grains are used for seed production during these periods, and therefore suggest that biparental inbreeding avoidance was not operating in C. crenata. In addition, the crossing rate between individual trees at the seed stage was much higher within (71%) than between (29%) patches. Such distance-dependent crossing may promote biparental inbreeding depression after seed production (Heschel & Paige, 1995; Koelewijn et al., 1999; Ubukata et al., 1999; Richards, 2000; Ishida, 2006; Isagi et al., 2007). Further studies are needed to clarify the effect of biparental inbreeding on seed germination and seedling establishment in C. crenata.
In several hand-pollination experiments, the pollen germination and pollen tube growth rates were usually lower for outcross-pollen grains from related individuals than for those from nonrelated ones (Waser & Price, 1993; Souto et al., 2002; Glaettli et al., 2006). For the observation of such a reproductive bias, even under natural conditions, the proportion of outcross-pollen grains from related individuals would be expected to decrease from the pollination to the seed stage. As noted here, however, little difference was observed in the pollen donor compositions between the pollination and seed stages in C. crenata under natural conditions.
Tree species tend to exhibit pollen limitation (Larson & Barrett, 2000). If pollen limitation occurred in C. crenata, a mother tree would have few chances to select nonrelated individuals. Thus, related outcross-pollen grains may compensate for the limitation of nonrelated outcross-pollen grains in C. crenata. However, it remains unclear whether related pollen grains contribute to seed production even under conditions for which pollen is not limited. To further clarify the pollen selection mechanisms of C. crenata, additional studies, including hand-pollination experiments, are needed.
We thank K. Kanou, M. Terabaru and M. Yamazaki for help with the setting up of the study site, N. Numano for field assistance, Dr A. S. Hirao for statistical advice and Dr N. Ueno, Dr T. Yagi, M. Tomita, Editor Dr Ruth G. Shaw and anonymous reviewers for comments on the manuscript. The primer sequences for locus KT030a Ch9a were donated by Dr T. Yamamoto. This research was financially supported by the Japan Society for the Promotion of Science (No. 17380095) and a Sasakawa Scientific Research Grant from the Japan Science Society.