EFFECTS OF KIN-STRUCTURED SEED DISPERSAL ON THE GENETIC STRUCTURE OF THE CLONAL DIOECIOUS SHRUB ILEX LEUCOCLADA

Authors


Corresponding author: E-mail: tomaru@agr.nagoya-u.ac.jp

Abstract

Nonrandom patterns of gene dispersal have been identified as possible causes of genetic structuring within populations. Attempts to model these patterns have generally focused solely on the effects of isolation by distance, but the processes involved are more complex than such modeling suggests. Here, we extend considerations of gene dispersal processes beyond simple isolation by distance effects by directly evaluating the effects of kin-structured gene dispersal mediated by the group dispersal of related seeds within fruits (i.e., kin-structured seed dispersal) by birds on genetic structure in Ilex leucoclada, a clonal dioecious shrub. To examine the genetic structure patterns, we established two 30×30 m plots (one with immature soils in old-growth forest and one in secondary forest, designated IM and SC, respectively) with different I. leucoclada stem densities. In these two plots 145 and 510 stems were found, representing 78 and 85 genets, respectively, identified by analyzing their genotypes at eight microsatellite loci. The clonal structure was stronger in the SC plot than in the IM plot. Correlograms of coancestry for genets in both plots exhibited significant, positive, high values in the shortest distance class, indicating the presence of strong genetic structure. However, Sp statistics revealed that the pattern of the genetic structure differed between the plots. In addition, to estimate the family structure within fruits, we sampled forty fruits, in total, from 15 randomly selected plants in the area around the IM and SC plots, and found that 80% of the fruits were multiseeded and 42–100% of the multiseeded fruits contained at least one pair of full sibs. Simulations based on these estimates demonstrated that the group dispersal of related seeds produced through correlated mating both within and across fruits, but not unstructured half-sib dispersal, could generate the observed magnitude and trends of genetic structure found in the IM plot. Furthermore, in addition to kin-structured seed dispersal, isolation by distance processes is also likely to promote genetic substructuring in the SC plot. After discussing possible ecological factors that may have contributed to the observed genetic structure, we contrast our results with those predicted by general isolation by distance models, and propose that kin-structured seed dispersal should promote some evolutionary phenomena, and thus should be incorporated, where appropriate, in models of gene dispersal in natural plant populations.

Genetic structure, which can be defined as the nonrandom distribution of genetic variation within and among individuals grouped at a hierarchical spatial scale, is a consequence of evolutionary processes such as gene movement, genetic drift and selection. Since the formulation of Wright's (1943) model of isolation by distance, predictions of gene movement obtained from it have been tested against observed relationships between genetic similarities (e.g., kinship) and geographical distance (e.g., Malécot 1967; Sawyer 1977; Rousset 2000). A basic assumption of the model is that single offspring are dispersed from their mothers. Many organisms, however, produce multiple offspring simultaneously that move in a group, and such group dispersal of related individuals (i.e., kin-structured gene movement) has been reported to be a contributing mechanism of genetic structuring in humans (Fix 1975) and other animals (Furuya 1968; Nash 1976; Bygott et al. 1979; Chepoke-Sade and Sade 1979). Furthermore, theoretical studies have shown that the level of relatedness of offspring dispersed in groups has substantial effects on the homogeneity of the spatial distribution of genetic variation (Rogers 1987; Levin and Fix 1989) and the rate of spread of advantageous genes among populations (Fix 1981; Levin and Fix 1989). Because these two modes of gene movement are not expected to be mutually exclusive, gene movement models should be reconstructed to incorporate the possible occurrence of kin-structured gene movement and the levels of relatedness of the offspring involved, rather than considering solely isolation by distance processes, to improve our understanding of the microevolutionary processes underlying the origin of genetic structure.

Several studies have shown that kin-structured gene movement (via kin-structured seed dispersal) may also have substantial effects in populations of plant species (e.g., Hamrick and Nason 1996; Giles et al. 1998; Ingvarsson and Giles 1999) because in many plant species the unit of dispersal is a multiseeded fruit (van der Pijl 1972), in which significant proportions of the seeds may have strongly related (or identical) fathers, as well as sharing the same mother plant. Because kin-structured seed dispersal can generate spatial aggregations of related individuals, it is likely to lead to evolutionary phenomena in which the spatial distribution of related individuals is important, such as biparental inbreeding depression (e.g., Fenster 1991) and kin selection (e.g., Kelly 1996). However, despite its significant roles in evolutionary processes, few studies have quantitatively evaluated the effects of kin-structured seed dispersal on within-population genetic structure. This is probably because both isolation by distance and kin-structured seed dispersal can generate spatial aggregations of related individuals within populations, and it is difficult to distinguish quantitatively the effects of these processes. Because levels of relatedness of seeds produced by the same maternal plants (e.g., the level of correlated mating within fruits) varies among plant species (Hardy et al. 2004) and/or populations (Wells and Young 2002), estimates characterizing microevolutionary factors such as gene dispersal parameters may be inaccurate if a general isolation by distance model is applied without considering kin-structured seed dispersal. Thus, to distinguish these two modes of gene dispersal and to improve our understanding of gene dispersal processes in plant populations, more realistic models incorporating both kin-structured seed dispersal and the levels of relatedness of the offspring involved should be developed and tested against observed data.

For plant species with animal-mediated seed dispersal, Hamrick and Loveless (1986) provided evidence that genetic structure may depend on the type of fruits involved, and the behavior of individual dispersers. For example, no clear genetic structure has been found within populations of the woody species Neolitsea sericea (Chung et al. 2000b) and Cinnamomum insularimontanum (Chung et al. 2003). In these cases, the evidence suggests that their individual seeds may move more or less independently because their single-seeded fruits are large (about 12 mm in diameter) and birds can harvest only a few at a time (Chung et al. 2003). In contrast, clear fine-scale genetic structure has been observed among seedlings and/or adult plants in cases where birds may move seeds in groups of half- or full-siblings (i.e., kin-structured seed dispersal), through processes such as seed-caching (Pinus albicaulis[Furnier et al. 1987]; P. pumila[Tani et al. 1998]) or harvesting multiseeded fruits (Swartzia simplex var. ochnacea[Hamrick et al. 1993]; Cecropia obtusifolia[Epperson and Alvarez-Buylla 1997]; Eurya emarginata[Chung and Epperson 2000]). Furthermore, repeated feeding on single-seeded fruits from the same plants is also likely to produce situations where related individuals will be dispersed in groups. These examples and considerations suggest that it is important to distinguish between within-population genetic structures caused by kin-structured seed dispersal and the effects of more general isolation by distance processes, because even if animals do act as random vectors of seeds genetic structure may be generated through kin-structured seed dispersal.

Ilex leucoclada (Aquifoliaceae) is a dioecous shrub that produces up to four seeds within its fruits (Yamazaki 1989). The reproductive characteristics of I. leucoclada, and other members of the Aquifoliaceae, include insect-mediated pollination and avian seed dispersal (Watanabe 1994). In addition, in previous studies it was found, using random amplified polymorphic DNA (RAPD) markers, that the species forms distinct patches by layering of the stems when they are pressed to the ground by heavy snow (Torimaru et al. 2003; Torimaru and Tomaru 2005). However, due to the dominant expression of RAPD markers, the data obtained gave no indications of genetic structure at the allelic level. We hypothesized that if microsatellite markers were used instead, the production of fruits with multiple seeds and bird-mediated seed dispersal of I. leucoclada would make the species an attractive subject for studying processes of genetic substructuring mediated by kin-structured seed dispersal within plant populations.

The objective of the study presented here was to test this hypothesis by examining the effects of kin-structured gene dispersal, especially the group dispersal of related multiple seeds within fruits, on the genetic structure of I. leucoclada populations. Although both isolation by distance and kin-structured seed dispersal can generate spatial aggregations of related individuals within plant populations, maternal plants should theoretically be the focal points of such aggregations generated by the former process, but not the latter. Thus, one way in which the effects of kin-structured seed dispersal on genetic structure may be distinguished from those of other processes is to investigate populations in which the time that has elapsed since they were established is relatively short. In such cases, there may have been too little time for isolation by distance relationships between individuals to develop via the cumulative effects of variations in the reproductive success of maternal plants. Following this rationale, we focused on the patch dynamics of I. leucoclada within populations, and inferred whether patch establishment and development were associated with kin-structured seed dispersal and/or isolation by distance. For this purpose, we examined two populations of I. leucoclada with different stem densities, because the extent of the clonal structure provides a useful indication of the time that has elapsed since they were established (see, for instance, Pornon et al. 2000). Then, we used microsatellite markers to examine their clonal structure, spatial distribution of genets, genetic variation, and family structure within fruits. The observed genetic structures were then compared with expected structures generated by simulations assuming kin-structured seed dispersal. This allowed rigorous interpretation of the factors potentially responsible for the genetic substructuring within populations of the species. Finally, we contrasted the results with predictions based on a general isolation by distance model, aiming to provide further insight into processes of gene dispersal via isolation by distance and kin-structured seed dispersal in plant populations.

Materials and Methods

STUDY ORGANISM

Ilex leucoclada (Aquifoliaceae) is an evergreen broad-leaved dioecious shrub, reaching heights up to two meters. It is found in mountainous regions where there is heavy snowfall in Honshu and the southern part of Hokkaido, Japan. This species is relatively common in the deciduous broad-leaved forests in the cool temperate zone along the Sea of Japan, which are dominated by Japanese beech, Fagus crenata (Ishizuka 1974). The characteristics of flowers and fruits of this species are as follows (Torimaru and Tomaru, 2006). The plants bloom in May and inflorescences are produced at the axils of current-year shoots. Each inflorescence produces approximately eight flowers in males, and two flowers in females. Each flower has four white petals in both sexes, and the diameters of individual flowers are, on average, 7 mm and 10 mm in males and females, respectively.

STUDY SITE AND FIELD METHODS

The study site is part of the forest reserve on Mt. Daisen in the north-central Chugoku Mountains in southwestern Japan. A 4-ha permanent plot (200 × 200 m) was established at about 1100 m above sea level on a southeast-facing slope in a continuous cover old-growth beech forest in the reserve in the period 1987–1988. Tree censuses have been performed for all stems with a diameter at breast height (d.b.h.) ≥ 4 cm, and the characteristics of the community structure of the plot have been described in detail by Yamamoto et al. (1995). To investigate community structure and dynamics in the secondary beech forests, a 1-ha plot (100 × 100 m) was also established at about 950 m above sea level on the same slope and a similar tree census was undertaken in 2003 (N. Nishimura et al., unpubl. data).

In 2003, a subplot (0.09 ha, 30 × 30 m) was delineated on an area with immature soils within the old-growth plot (hereafter referred to as the “IM plot”). The surface of the immature soils in the area is unstable as a result of slight movements of gravel and sand due to surface flow, or occasional small landslides (Yamamoto et al. 1995). Such instability should prevent the formation of I. leucoclada patches by clonal growth through layering (Torimaru et al. 2003). In 2003 we also delineated a subplot of the same size in the secondary beech forest plot (designated the “SC plot”). Here there was a denser distribution of I. leucoclada stems. The spatial coordinates (x, y), stem length, absence/presence of flowers, and sex of each stem were recorded. Leaves were collected from all stems in both subplots for microsatellite analysis, and were stored at −30°C until the DNA was extracted.

In 2003, 15 stems were randomly selected from the area around the IM and SC plots, and one to five mature fruits were sampled per stem (mean ± SD = 2.7 ± 1.1 fruits per stem). The seeds in the fruits were dissected and we determined whether embryos were present in them. The embryos in the seeds were used for DNA extraction.

DNA EXTRACTION AND MICROSATELLITE ANALYSIS

The techniques used for extracting DNA from the leaves and embryos, and microsatellite amplification by polymerase chain reaction (PCR), are described in detail by Torimaru et al. (2003, 2004). In this study, we used eight primer pairs developed by Torimaru et al. (2004). The genotypes of the stems and the seeds at the eight microsatellite loci (ILE03-01, ILE03-38, ILE03-65, ILE04-04, ILE04-06, ILE04-17, ILE04-18, and ILE05-83) were determined using an ABI 3100 Genetic analyzer in conjunction with GeneScan and Genotyper software (Applied Biosystems, Foster City, CA).

DATA ANALYSIS

Genet determination

Multiple stems with identical multilocus genotypes were considered to be members of the same distinct genets. To ensure that the resolution power of the microsatellite markers used was sufficient to distinguish genets in the present study, we calculated two types of probabilities: Psex and Psex-sib. The Psex for each multilocus genotype was defined as the probability of observing at least n stems of that genotype by chance in a sample of N stems in each plot given panmictic sexual reproduction with free recombination. The Psex values were calculated using a binominal probability function (Park and Werth 1993):

image

where Pgen= (ΠLi=1piqi)2h is the expected frequency of the multilocus genotype under sexual recombination and pi, qi= the estimated frequency of the first and second alleles at loci i among genets in a plot, h= the number of loci that are heterozygous, and L= the number of loci used. The second probability, Psex-sib, for each multilocus genotype was defined as the probability of observing at least n stems of that genotype in a sample of N full-sibs in each plot (modified from Schuster and Mitton [1991] for applying a binominal probability function):

image

where inline image is the probability that two stems from the same full-sib would have an identical multilocus genotype by chance and inline image, inline image, k, m= the number of possible maternal and paternal genotypes at locus i, respectively, P(MPjl) = the probability of maternal genotype gj and paternal genotype gl producing offspring with genotype gi, f(gj) = the frequency of genotype gj in a population, and Nj= the number of alleles in common between offspring with genotype gi and parental genotype gj. Sequential Bonferroni correction with the Dunn-Šidák method (Sokal and Rohlf 1995) was used to determine the significance of the Psex and Psex-sib values.

To assess the spatial association among multilocus genotypes in each plot, we determined the probability of genet identity (Fr) as a function of spatial distance (e.g., Harada and Iwasa 1996; Alberto et al. 2005; Suzuki et al. 2006). For that purpose we computed the proportions of pairs of stems in 2-m distance intervals sharing the same multilocus genotypes. For each plot, spatial distances were permuted randomly among pairs of stems 1000 times to test, for each spatial distance class, whether the observed Fr values were different from those expected under a random distribution of genotypes.

Genetic variation

For each locus in each plot, allele frequencies, observed heterozygosity (HO), unbiased expected heterozygosity (HE; Nei 1987), and fixation index (FIS= 1 −HO/HE; Nei 1987) were calculated. The significance of any departure from Hardy–Weinberg genotypic proportions was assessed by the Markov-chain method (Gao and Thompson 1992) using GENEPOP software (Raymond and Rousset 1995).

Genetic structure among genets

When genetic structure is analyzed within populations of clonal plants, it is important to distinguish genets (genetic individuals established via seeds) in samples that include clonal ramets because inferences of microevolutionary processes from the pattern and magnitude of genetic structure must be based on genets rather than ramets (Chung et al. 1999; Chung and Epperson 1999; Hämmerli and Reusch 2003; Setsuko et al. 2004; Alberto et al. 2005; Chung et al. 2006). In the present study, spatial autocorrelation was used to evaluate the genetic structure among distinguishable genets within each plot. This was done by estimating the coancestry (Cockerham 1969) between all possible pairs of genets in different distance classes. The centroid coordinates (x, y) were used as the spatial locations of the genets with multiple stems. To estimate coancestry, genotypes of genets were first expressed as allele frequencies of 1.0, 0.5, or 0.0, depending on whether the genotypes included 2, 1, or 0 copies, respectively, of the target alleles (Heywood 1991). Estimates of coancestry (fij) were then calculated using the formula (Loiselle et al. 1995):

image

where pi and pj are the frequencies of the same alleles at a locus for genets i and j, inline image= the mean frequency for that allele, k=n (n− 1)/2 gives the number of possible pairs between n genets located in each distance class, and N= the total number of genets in the plot. The second term adjusts for any bias resulting from a finite sample size, which causes the estimate of coancestry to be zero for a population in Hardy–Weinberg equilibrium. The contribution of each allele is weighted by its respective polymorphism, plm (1 - plm), with plm being the frequency of the mth allele at locus l. The average fij over pairs of genets was computed for distance intervals of 2 m to ensure there was a minimum of 100 pairs of genets per distance class. The genetic structure of each distance class was determined using the permutation method (1000 permutations); spatial distances were permutated randomly among pairs of genets, and the estimated average coancestry value was compared to the distribution after permutations. These calculations were performed using SPAGeDi 1.1 software (Hardy and Vekemans 2002). Furthermore, to compare correlograms between plots we calculated Sp statistics (Vekemans and Hardy 2004) for each plot, which are functions of the genetic relatedness in the plots and can be affected by several factors (e.g., actual level of relatedness, overlap of seed shadow), and are obtained from the slope of the regression of a kinship coefficient ( fij) against the logarithm of the distance:

image

where bF= the regression slope, and F1 (the coefficient of coancestry for the shortest distance class. The bF and F1 values for each plot were obtained using SPAGeDi software. Standard errors were estimated by means of jackknife tests over loci. To determine whether the Sp values significantly differed between the plots, t-tests were used, assuming that the distribution of the Sp values for each locus was approximately normal.

Family structure within fruits

For the sampled fruits, the number of fertilized seeds per fruit was determined, and the frequencies of fruits with each number of seeds were recorded. The sibling relationships among seeds within fruits were inferred by the likelihood method based on the genotypes of maternal parents and their seeds, using COLONY version1.3 software (Wang 2004). The likelihood of a given assignment of seeds into half- and full-sib families was calculated, and in each case the assignment with the maximum likelihood was identified using the Markov-chain Monte Carlo method with the simulated annealing technique, which allows null alleles and all types of stochastic errors to be taken into account (Wang 2004). In the present study, to estimate the frequencies of each family structure (i.e., each combination of half- and full-sibs) among seeds within fruits containing the same number of seeds, using the COLONY software, we first determined both null allele and stochastic error rates. This preliminary study indicated that COLONY exactly assigned seeds into half-sib family when seeds from the same stems were used and both of the error rates were set to 0.05. We then performed COLONY analyses with those error rates under the condition that seeds within the same fruits were considered to be half-siblings. We conducted five independent runs with the same dataset to confirm that all runs converged to the same family structure with the same maximum likelihood.

Simulation of genetic structure

Because I. leucoclada plants produce fruits with multiple seeds (up to four; Yamazaki 1989), related seeds are likely to be dispersed in groups by birds. Therefore, simulations were conducted to test whether the observed patterns of genetic structure were consistent with the group dispersal of related seeds.

To estimate the expected values of coancestry under the genetic structure generated by the assumption of group dispersal of related seeds, simulations were conducted as follows. For each plot, based on the observed frequencies, fruits with various numbers of seeds were modeled; the total number of seeds was equivalent to the actual number of genets in each plot. The family structure among seeds within the fruits was then assigned and three types of simulation were run for each plot. In the first, seeds within fruits were considered to be half-siblings with no full-sibling relationships, implying that there was no correlated mating. The second type of simulation was based on the estimated frequencies of each family structure among seeds within fruits containing the same number of seeds. In the first and second types of simulation, genotypes of the seeds were assigned stochastically (i.e., assuming random mating), according to the allele frequencies estimated by Torimaru et al. (2003) in a reference population comprising 27 patches of I. leucoclada in the old-growth beech forest. However, the third type of simulation (which was identical to the second type in other respects) incorporated the effects of nonrandom mating, as described in the following paragraph.

Because fruit set of female stems of I. leucoclada was significantly higher in patches including both male and female genets (called mixed patches) than in those consisting solely of female genets (female patches) (Torimaru and Tomaru, 2006), we concluded that within-patch mating significantly contributes to nonrandom mating in the I. leucoclada population. Our previous study (Torimaru et al. 2003) indicated that the kinship coefficient between sexes within patches in the examined population was 0.106 (SE, 0.017) and not significantly different from 0.125 (i.e., half-siblings) (test for the mean, P= 0.255). Thus, we adjusted the model in the third type of simulation to predict the effects of a certain proportion of mating occurring between half-siblings. Fruit set of female stems in the mixed patches was 2.96 times higher than in the female patches in a year with normal climate (Torimaru and Tomaru, 2006), so assuming that the increment of fruit set in the mixed patches was due to the mating within patches, we estimated the proportion of mating with half-sibs in the mixed patches to be 0.662 (=[2.96 − 1/2.96). Then, because the number of the mixed patches was 4.25 times higher than that of the female patches in the population examined by Torimaru et al. (2003), we set the fraction of mating with half-sibs at 0.613 (= 0.662 × 4.25 × 2.96/[4.25 × 2.96 + 1 × 1]). In each 30 × 30 m plot, related seeds within fruits were considered to be dispersed within a circle, the radius of which was determined as follows. We counted the number of genets within a circle of a given radius (including the focal genet) from each focal genet, and calculated its average over focal genets. We then determined the radius of the circle with an average observed number of genets equivalent to the average observed number of seeds within fruits. Coancestry values were subsequently estimated for each 2-m distance class. The procedure was repeated 1000 times, and we compared each coancestry value in each distance class estimated from the actual dataset with the 95% confidence limits (two-tailed) of the distribution generated by the simulations. Rejection of the hypothesis requires the actual values of coancestry to fall outside the confidence interval for neighbor distances or for greater distances in some systematic way.

Results

CLONAL STRUCTURE AND GENETIC VARIATION

We found a total of 146 and 546 stems in the IM and SC plots, respectively, and were able to determine 78 and 85 different multilocus genotypes amongst the 145 and 510 analyzed stems, respectively, using the eight-microsatellite loci. We detected 17 (21.8 %) and 37 (43.5 %) multilocus genotypes in two or more stems; each of the remaining 61 and 48 genotypes appeared in only a single stem in the IM and SC plots, respectively. For the 54 genotypes represented by multiple stems, the validity of considering them as different genets was assessed by calculating Psex and Psex-sib values. Both of these probabilities for every genotype were under the values obtained after sequential Bonferroni correction by the Dunn-Šidák method. Furthermore, both plots showed significantly high Fr values in the short distance classes, indicating that neighbors were most likely to have the same genotypes (Fig. 1A, B). Thus, we can be confident that we detected stems derived from identical genets and distinguished related genets in each plot. Assuming that distinct genotypes correspond to different genets, the genets in the SC plot were represented by approximately three times as many stems as those in the IM plot (mean and maximum numbers of stems per genet: IM plot, 1.9 and 14; SC plot, 6.0 and 49, respectively). The probability of genet identity in the shortest distance class was higher in the SC plot (0.556) than in the IM plot (0.324) (Fig. 1B). Therefore, the stems of identical genets were more closely aggregated in the SC plot than in the IM plot (see also Fig. 1A).

Figure 1.

(A) Spatial distribution of genets and stems of Ilex leucoclada in the IM plot (top, 78 genets) and SC plot (bottom, 85 genets). Plus signs (+) represent genets whose stems accounted for < 5.0% of the total number of stems within each plot and crosses (×) represent stems for which multilocus genotypes could not be determined due to the low quality of the template DNA. The other symbols represent different genets whose stems accounted for ≥ 5.0% of the total number of stems within each plot. (B) Correlograms of the estimated values of probabilities of genet identity for Ilex leucoclada stems in the IM plot (top) and SC plot (bottom). Distance classes were defined at 2-m intervals from 0–2 m to 14–16 m. Thick lines represent the sample statistics Fr and dashed lines represent 95% (two-tailed) confidence intervals based on the null hypothesis of no spatial autocorrelation, obtained after 1000 permutations of spatial distances between pairs of stems.

Genetic variation at the investigated microsatellite loci was estimated using data obtained for the 78 and 85 genets recognized in the IM and SC plots, respectively. For the IM plot, the expected heterozygosity for each locus ranged from 0.606 to 0.907, with an average of 0.804 (Table 1), whereas for the SC plot, the expected heterozygosity for each locus ranged from 0.575 to 0.922, with an average of 0.797 (Table 1). These figures were not significantly different between plots (Wilcoxon test). Thirteen out of the 16 values for the fixation index indicated significant deviations from Hardy–Weinberg genotypic proportions for both plots. To obtain evidence for the presence of null alleles, the deviations from the genotypic proportions at Hardy–Weinberg equilibrium were compared between the plots. If null alleles exist at these loci, heterozygote deficiencies can be expected in both plots. Heterozygote deficiencies were consistent in both plots for ILE 04-06 and ILE 04-18 (Table 1), indicating that null alleles may be present at these two loci. Therefore, although all of the eight loci were used to distinguish different genets, the ILE 04-06 and ILE 04-18 loci were excluded from the other analyses. Although a certain fraction of the seeds was likely to be derived from mating with half-sibs (see the following sections) in both plots, there appeared to be little evidence of heterozygote deficiency even when those loci were excluded.

Table 1.  Genetic variation at the eight examined microsatellite loci in Ilex leucoclada genets in the IM and SC plots. A, number of alleles detected; HO, observed heterozygosity; HE, expected heterozygosity; and FIS, fixation index. Deviations from Hardy–Weinberg genotypic proportions were assessed using the Markov-chain method.
LocusIM plotSC plot
AHOHEFISAHOHEFIS
  1. *, P < .05.

ILE03-01 8.808.693−.167 7.741.690−.074*
ILE03-3820.821.907 .096*23.953.922−.034*
ILE03-6510.846.787−.076*11.847.746−.135*
ILE04-0415.808.898 .10115.894.880−.016*
ILE04-0619.808.891 .094*18.812.885 .083*
ILE04-1710.859.846−.015*10.906.818−.108*
ILE04-1814.564.606 .070*11.365.575 .366*
ILE05-8313.756.802 .057 9.918.860−.066*
Mean/total14.784.804 .025*13.804.797−.009*
(Excluding ILE04-06 and 04-18) (.007*)   (−.024*)

FAMILY STRUCTURE WITHIN FRUITS

Forty fruits were sampled from the 15 randomly selected stems and in total 94 seeds were obtained. Varying numbers of fertilized seeds were detected among the fruits and 80% of the fruits sampled contained multiple fertilized seeds (Table 2). The average number of fertilized seeds ± SD per fruit was 2.4 ± 1.0. Different runs using COLONY software produced very similar, although not exactly identical, assignments so we estimated the frequencies of each family structure among seeds within fruits with the same numbers of seeds by averaging the data from the five runs. The proportions of multiseeded fruits with two, three and four seeds that produced seeds with full-sib relationships amounted to 42%, 85%, and 100%, respectively (Table 2).

Table 2.  Estimated frequencies of family structure (i.e., combinations of half- and full-sibs) among seeds within fruits containing the same number of seeds.
No. of seeds per fruitFrequency of fruits with each seed numberNo. of full-sibling seeds within fruitsFrequency of family structure among seeds within fruits
  1. 1Four seeds all with the same parents.

  2. 2Two pairs of seeds, each pair being full-sibs.

1.2001.000
2.3250 .585
 2 .415
3.3500 .154
 2 .277
 3 .569
4.1250 .000
 2 .200
 3 .000
 41 .120
 42 .680

GENETIC STRUCTURE AMONG GENETS

The coancestry values were significantly positive in the shortest distance class (IM plot = 0.188; SC plot = 0.202) and in the two next shortest classes, that is, 2–4 and 4–6 m, for both plots (Fig. 2). However, while the coancestry values fluctuated in the larger distance classes for the IM plot, with both positive and negative values, the values were consistently and significantly negative from the fifth distance class (i.e., 8–10 m) upward for the SC plot. The Sp values (SE) were 0.059 (0.008) and 0.080 (0.006) for the IM and SC plots, respectively, and the patterns of the correlograms for the two plots were significantly different from each other.

Figure 2.

Correlograms of the estimated values of coancestry for Ilex leucoclada genets in the IM plot (top) and SC plot (bottom). Distance classes were defined at 2-m intervals from 0–2 m to 14–16 m. Thick lines represent the sample statistics fij and dashed lines represent 95% (two-tailed) confidence intervals based on the null hypothesis of no spatial autocorrelation, obtained after 1000 permutations of spatial distances between pairs of genets.

The radii of the circles within which the average observed numbers of genets were equal to the average observed numbers of seeds within fruits (i.e., the radii of the circles including, on average, 2.4 genets) were 0.65 and 0.80 m in the IM and SC plots, respectively. Using these values, the tests for significant differences in genetic structure between the actual patterns and the patterns obtained assuming group dispersal of only half-sib seeds indicated that the coancestry value in the shortest distance class was higher than the simulation 95% confidence interval for both plots (Fig. 3A). However, when the observed degrees of kin-structured seeds within sampled fruits (Table 2) were included in the model the coancestry value in the shortest distance class appeared to be within the simulation confidence interval for the IM plot (Fig. 3B). Although a few estimates were significantly higher than the confidence limits, there was little to indicate any departure from the simulation confidence intervals based on the estimates obtained from fruits sampled in the IM plot. In contrast, the coancestry values in the shortest distance class were still significantly higher than the simulation confidence intervals based on the estimates obtained from the sampled fruits combined with random mating patterns, but were not significant when assuming nonrandom mating patterns (Fig. 3C).

Figure 3.

Comparisons of the coancestry values estimated from the empirical datasets (thick lines) with 95% confidence intervals (dashed lines) generated by simulations assuming: (A) the group dispersal of half-sib seeds within fruits; (B) group dispersal of related seeds within fruits, with the family structure estimated from the sampled fruits; and (C) nonrandom mating for Ilex leucoclada genets in the IM plot (top) and SC plot (bottom).

Discussion

POPULATION CLONAL AND GENETIC STRUCTURE

The clonal structure of I. leucoclada was stronger in the SC plot than in the IM plot. In the latter, the immature soil surface was not stable (Yamamoto et al. 1995) and probably hindered continuous seedling recruitment and patch-formation by clonal growth. Thus, the IM plot can be considered a good representative for populations in which relatively little time has elapsed since they were established. In contrast, in the SC plot, the population could develop by both seedling recruitment and clonal growth because there was no evidence of severe disturbance.

The correlograms obtained in the coancestry analysis demonstrated the presence of clear genetic structure among the I. leucoclada genets in both plots. Compared with figures presented in reviews of Sp statistics by Vekemans and Hardy (2004), the observed levels were higher than reported levels for both species that produce seeds dispersed by animals and outcrossing species (0.017 and 0.013, respectively). Furthermore, the fact that the coancestry values in the shortest distance class were comparable with those obtained in simulations of the effects of the group dispersal of full-sibling seeds within fruits (our additional simulations, data not shown) suggests the presence of strong genetic structure in the I. leucoclada populations of both plots.

THE CAUSE OF THE GENETIC STRUCTURE

The clear genetic structure in the IM plot could be explained using data from the simulations based on the assumption that related seeds within fruits are dispersed in groups. This is likely because the population in IM plot is representative of populations that have been established for relatively short times, and the initial template for spatial genetic variation generated through seed dispersal has probably been preserved. Furthermore, the simulation results were close to the observed genetic structure in the IM plot when the observed levels of correlated mating within fruits were incorporated, but not when the seeds were all assumed to be half-sibs. In addition, when multiple fruits were sampled from the same stems, 65.7% of the seeds were found to have full-sib relationships to seed from different fruits. Thus, although kin-structured seeds are expected to originate primarily from correlated mating within fruits, correlated mating among fruits borne by the same plants may also contribute to kin-structured seed dispersal in I. leucoclada. Evidence of correlated mating (at varying frequencies) has been found in genetic analyses of seeds within individual fruits of several plant species (Hardy et al. 2004) and hierarchical patterns of correlated mating within individuals (Muona et al. 1991). These findings and our simulations suggest that not only the occurrence of kin-structured seed dispersal but also the degrees of relatedness of the offspring involved do indeed have substantial effects on genetic substructuring in plant species.

In the SC plot, the coancestry values shifted from significantly positive to negative within a relatively short spatial scale (i.e., within 8–10 m). This distinct genetic structure seems to be consistent with the genetic similarity found among genets within the same patches (for which the mean length ± SD was 2.5 ± 1.1 m) and genetic differentiation between those in different patches (amongst which mean distances between the nearest neighboring patches ± SD was 9.6 ± 3.2 m) of I. leucoclada growing in the old-growth beech forest (Torimaru et al. 2003). The simulation indicated that nonrandom mating patterns have substantial effects that could generate the observed level of coancestry in the shortest distance class in the SC plot. Because there were more spatial aggregations of stems and genets of I. leucoclada in the SC plot, such a distribution is likely to affect pollinator behavior, leading to various effects including a preponderance of nearest neighbor pollination (e.g., Levin 1984; Fenster 1991; Fenster et al. 2003; García et al. 2005). Furthermore, theoretical models indicate that repeated generations of mating by proximity result in the development of genetic structure (Epperson 1990, 1993); a hypothesis supported by studies that have found the genetic structure among genets to be greater in mature populations than in younger populations (Chung et al. 1999, 2000a). These processes may have been involved in the formation of the genetic structure observed in the SC plots.

On the other hand, none of our simulations could explain the genetic differentiation between distant genets observed in the SC plot, indicating that spatial limitations of seed dispersal also influence the genetic structures of the populations. A probable cause of the limited dispersal of seeds is due to the behavior of frugivorous birds in the SC plots. The I. leucoclada fruit rarely fall under gravity, because frugivorous birds tend to remove them efficiently before they fall (T. Torimaru et al., unpubl. data). Several studies have shown that the amounts of seeds dispersed by frugivorous birds around individual fruiting plants decline with increasing distance from the source plants (Hoppes 1988; Godoy and Jordano 2001). The cited studies provide evidence supporting the development of genetic patches, because this behavior is likely to lead to many seeds derived from the same female parent being dispersed in close proximity to each other. Thus, this process may enhance genetic variation among different patches of I. leucoclada.

MICROEVOLUTIONARY IMPLICATIONS OF KIN-STRUCTURED GENE DISPERSAL

Our simulations indicate that the causes of genetic substructuring within populations are more complex than general isolation by distance models predict (e.g., Crawford 1984, Vekemans and Hardy 2004). Thus, such models should be applied with caution in attempts to elucidate gene dispersal processes and their effects on microevolution for two main reasons. First, even in populations of plant species producing fruits that are dispersed by birds and other animals randomly and over long distance (with, therefore, no substantially overlapping seed shadows of different mothers), kin-structured seed dispersal could generate the observed levels of genetic structure in the I. leucoclada population. If standard genetic models are applied to populations of such plant species, the magnitude of gene dispersal may be underestimated. Second, the presence of isolation by distance relationships between genets was strongly supported in the older, but not the younger, population of I. leucoclada investigated here (see also the study of Silene dioica by Giles and Goudet [1997]). Because several studies have reported that the behavior of birds and other animals is affected by the type of vegetation present (Herrera and Jordano 1981; Hoppes 1988), changes in vegetation structure that occur as populations proceed through time will affect seed dispersal by birds and other animals. Thus, we should pay much more attention to possible temporal changes in seed dispersal patterns when considering the effects of isolation by distance on observed genetic structures.

Two lines of evidence in our results suggest that kin-structured seed dispersal may promote evolutionary phenomena in plant populations in which the spatial distribution of related individuals is important (e.g., biparental inbreeding depression and kin selection). First, our study has shown that kin-structured seed dispersal could generate strong genetic structure following the establishment of I. leucoclada populations, but no such indications have been found in several studies of plant species that produce single-seeded fruits (Chung et al. 1999, 2000a). This difference is probably due to isolation by distance processes requiring the establishment of maternal plants before offspring can be dispersed, and thus at least twice as much time as kin-structured seed dispersal to generate spatial aggregations of related individuals. Thus, kin-structured seed dispersal may substantially promote interactions between related individuals at early stages following the establishment of populations. Second, there are likely to be substantial interactive effects of the dispersal of kin-structured seeds and isolation by distance processes, which reinforce the spatial aggregations of related individuals and in turn promote mating between them in the older population of I. leucoclada. These interactive effects, in addition to the selection effects demonstrated by Epperson (1990), may help to stabilize genetic structure in populations influenced by them earlier than those generated solely by general isolation by distance processes. Development of models that incorporate the two accelerative factors should facilitate attempts to elucidate their relative importance in evolutionary processes affecting plant populations.

In summary, the empirical and simulation results we obtained demonstrate that the causes of genetic substructuring within populations of I. leucoclada, a species that produces bird-dispersed seeds, are more complex than simple isolation by distance models indicate. Because kin-structured seed dispersal and the levels of relatedness of the offspring can promote genetic substructuring in plant populations, as demonstrated in the present study, the nature, strength and scale of microevolutionary processes underlying observed patterns of genetic structure may be misinterpreted if isolation by distance processes are assumed to be the sole causes of genetic substructuring via gene dispersal.

Associate Editor: R. Mauricio

ACKNOWLEDGMENTS

The authors are grateful to S. Yamamoto and other members of the Laboratory of Forest Ecology and Physiology, Nagoya University for useful discussion and assistance in the field and laboratory. We thank J. Suzuki, C. Fenster, R. Mauricio, and three anonymous reviewers, for their valuable comments on the study, and a previous draft of the manuscript. We also thank the Tottori Distinct Forest Office for allowing us to conduct this study. The work was supported by Grants-in-Aid for Young Scientists (No. 15000944) and Scientific Research (No. 14206017) from the Japan Society for the Promotion of Science.

Ancillary