#### study species

*Primula modesta* is a small rosette herb that grows in a range of montane to alpine habitats in Japan. Like many other *Primula* species, *P. modesta* is distylous, with self and intramorph incompatibility (Wedderburn & Richards 1990). It produces one flowering stem 7–15 cm tall and an umbel with 3–20 flowers. In the subalpine zone of central Japan, bud burst occurs in early May, and the flowers bloom from late May to early July. The fruits (capsules) mature in late August to mid-September, and the seeds are dispersed on sunny days from a small apical opening that closes hygromechanically on rainy days. The seeds are minute (5–10 µg) and have no special adaptations for dispersal. Seed dispersal continues through autumn and winter.

The seeds require at least a month of moist chilling to be released from their primary dormancy, after which more than 90% of seeds are capable of germination at temperatures of 12–20 °C in the light (germination remains negligible in the dark; Shimono & Washitani 2004). More than 70% of buried seeds are still viable after burial for a year (A. Shimono, unpublished data), although the longevity of seeds in the seed bank and average time spent as a dormant seed are unknown.

#### study site

The study site was located on an oligotrophic fen in the subalpine zone (2100 m a.s.l.) of Mt Asama (36°24′12″ N, 138°31′34″ E, 2568 m a.s.l.) in Nagano Prefecture, central Japan. The site is flat to gently sloping. *Juncus fauriensis* Buchen., *Carex doenitzii* Böckeler and *Drosera rotundifolia* L. form a sparse vegetation cover, with a layer of *Sphagnum* spp. covering the ground. The total number of flowering ramets of *P. modesta* was approximately 5000, and their density was 20 m^{−2}. The frequencies of the long- and short-styled morphs were roughly equal. Total ramet density, including non-flowering ramets, was approx. 400 m^{−2}.

The climate of the study site can be deduced from the records available from the Karuizawa meteorological station, about 7 km to the east (36°20′3″ N, 138°32′9″ E, 999 m a.s.l.). Assuming a standard temperature decline of 0.55 °C for every 100-m increase in altitude, we estimated the annual mean air temperature (30-year average for 1971–2000) at the study site to be about 3 °C. The mean annual precipitation at the meteorological station was 1198 mm (30-year average for 1971–2000).

#### seed bank

To assess the spatial genetic structures of both the soil seeds and flowering plants, we set a 2.5 × 5.5 m quadrat, with the long side (*y* axis) parallel to the slope (Fig. 1). We divided the quadrat into 55 subquadrats (0.5 × 0.5 m) and collected 40 soil cores (5 cm diameter, 5 cm deep) at the intersections of the lattice in late April of 2003, just before spring germination began. We also collected the surface litter layer, if present, with each core and divided each soil core into depths of 0–1 cm (surface) and 1–5 cm (deeper). *Primula modesta* seeds have a strict light requirement (Shimono & Washitani 2004), which prevents the germination of even slightly buried seeds, because light is strongly attenuated by penetration through even thin layers of soil (Bliss & Smith 1985; Tester & Morris 1987). As samples collected at soil depths of 2–5 cm hold a relatively small fraction of the seed bank (approximately 8% of total soil seeds), 5 cm depth is sufficient for sampling all soil seeds. Each soil sample was bagged and brought to the laboratory, then spread into a thin layer (about 0.5 cm deep) over a 10 cm depth of vermiculite in trays. Emerging *P. modesta* seedlings were identified every 5 days during the following 5 months. The seedlings were transplanted into pots and grown for later genetic analysis. We defined the seed bank size as the mean density of germinable seeds in the 40 cores. Seed density per unit area was calculated by dividing the total seeds in the 40 cores by the area of the soil surface covered by the cores (40 × π × 2.5 cm × 2.5 cm). Flowering ramets in the quadrat were mapped in 2003.

#### dna extraction and microsatellite genotyping

To determine the multilocus genotypes of the plants, we sampled one young leaf from every mapped flowering ramet in 2003 and from each of the seedlings that originated from the soil cores. Total DNA was extracted from leaves using a modified hexadecyltrimethyl ammonium bromide (CTAB) method (Murray & Thompson 1980). Leaves were stored at −30 °C until DNA extraction and ground to a fine powder. They were mixed with 750 µL of CTAB extraction buffer and incubated at 60 °C for 10 minutes. One volume of chloroform–octanol (24 : 1) was added and emulsified by shaking. The mixture was centrifuged at 15 000 r.p.m. for 10 minutes, and the aqueous phase was collected. DNA was then precipitated by adding two-thirds of a volume of isopropanol and washed once with 70% ethanol. The DNA was then vacuum-dried and resuspended in an appropriate volume (200–500 µL) of TE (Tris EDTA). For microsatellite amplification, 5–10 µg (1 µL) of the DNA was used in 10 µL of polymerase chain reaction (PCR) mixture. PCR was performed for 10 loci (*PM175*, *PM179*, *PM324*, *PM668*, *PM769*, *PM770*, *PM772*, *PM801*, *PM850* and *PM901*) and genotypes analysed as described in detail by Shimono *et al*. (2004).

#### gene diversity

Genetic diversity parameters were estimated for each component (surface or deeper seed bank, flowering genet) by using FSTAT 2.9.3.2 software (Goudet 1995). Observed heterozygosity (*H*_{O}), expected heterozygosity (*H*_{E}), and allelic richness (*A*) were calculated. Deviation from Hardy–Weinberg equilibrium (HWE) within each component was tested with fixation index (*F*) (Weir & Cockerham 1984). The statistical significance of these values was determined on the basis of 20 000 randomizations of alleles among individuals within components.

Nei's (1973) gene diversity formula (*G*_{ST}) was used to evaluate the distribution of genetic diversity between the soil seed bank and above-ground populations.

#### fine-scale genetic structure

Fine-scale genetic structure was assessed by spatial autocorrelation analysis of genetic relatedness based on pairwise kinship coefficients (Loiselle *et al*. 1995), calculated as the frequency of allele sharing between two individuals relative to the average frequency of allele sharing based on the frequencies of the alleles in the sample population (the ‘reference population’). *F*_{ij}, the average multiallelic and multilocus kinship coefficient between individuals *i* and *j*, is calculated as follows:

where *p*_{ila} and *p*_{jla} are the frequencies of allele *a* at locus *l* in individuals *i* and *j*, respectively, *p*_{la} is the frequency of allele *a* at locus *l* in the reference population, and *n* is the number of genes defined in the sample per locus. The second term adjusts for the bias attributable to finite sample size.

To analyse the relationship between pairwise physical distance and pairwise kinship coefficients, we calculated the *F*_{ij} for seven distance classes: 0–30, 30–60, 60–90, 90–120, 120–150, 150–180 and > 180 cm. As soil seeds were sampled at 50-cm intervals in the quadrat, setting the distance classes at much lower intervals would not have helped elucidate details of the spatial structure. The coordinates of the soil seeds were assumed to be randomly distributed within the area of each soil core.

The *F*_{ij} value was tested against the null hypothesis by randomization. The observed *F*_{ij} for a given distance class was then compared with the randomized empirical distribution. The randomizations were conducted by randomly permuting multilocus genotypes, whilst keeping their locations in the stand constant. These permutations were generated 1000 times, and the *F*_{ij} was calculated for each permutation. The overall significance of the trend shown in the correlograms was tested according to Bonferroni criteria.

The overall presence/absence of isolation by distance can be assessed by the slope (*b*) of a correlogram, i.e. by regressing pairwise *F*_{ij} coefficients against the logarithm of the pairwise geographical distances (Vekemans & Hardy 2004). The significance of the linear regression slope was tested by Mantel tests (Manly 1997) with 1000 random permutations. Jackknife standard errors for *b* were obtained by jackknifing over loci.

In cases where isolation by distance was detected, we also evaluated its strength using *Sp* statistics, calculated as –*b*/(1−*F*_{(1)}), where *F*_{(1)} is the *F*_{ij} for the first distance class. *F*_{(1)} can be considered an approximation of the kinship between pairs of neighbours, provided the first distance class contains enough pairs of individuals to obtain a reasonably precise *F*_{(1)} value (Vekemans & Hardy 2004).

Spatial autocorrelation analysis was also used to test whether the spatial structures of two components (surface seed bank vs. flowering genet, deeper seed bank vs. flowering genet, or surface seed bank vs. deeper seed bank) were dependent on each other, with *i* and *j* representing individuals from each of the two components.

In addition, to interpret the effect of secondary dispersal on the spatial association between the seed bank and a flowering genet, we classified pairs of individuals *i *(seed bank) and *j* (flowering genet) according to the angle (θ) between the *X* axis of the quadrat (perpendicular to the slope) and a vector from the flowering genet to the soil seed as: (i) right, −45° < θ < 45°; (ii) upward, 45° ≤ θ ≤ 135°; (iii) left, 135° < θ and θ < −135°; or (iv) downward, −135° ≤ θ ≤ −45°. It is likely that secondary dispersal occurs by rainfall running down the gentle slope (i.e seeds located downwards of a flowering plant could be in this category). In the directional analysis, the slope (*b*) of a correlogram was calculated in the 0–200 cm distance range that contained a roughly equal number of pairs in each direction. The number of pairs in the upward and downward directions increased with distance, while those in the right and left directions were saturated at distances of roughly 200–300 cm because the *x* axis (250 cm) is shorter than the *y* axis (550 cm).

#### spatiotemporal pattern of seedling survival

Seedling emergence was monitored in a 0.5 × 4.5 m subquadrat established within the main quadrat, because there were too many seedlings to monitor in the entire quadrat. In the subquadrat, the locations of newly emerged seedlings of *P. modesta* were mapped, and seedlings were marked with coloured toothpicks at 2- to 4-week intervals during the growing season from May to September in 2002. Seedling survival was monitored until May 2003. Flowering ramets had been mapped in 2001. Spatial relationships between seedlings (live + dead, live or dead) and flowering ramets in the previous year were analysed by examining neighbourhoods around individuals (Condit *et al*. 2000):

where *d*_{ij} is the distance of *i*th seedling (*i* = 1 …*n*_{1}) and *j*th flowering ramet (*j* = 1 …*n*_{2}), respectively; *A*_{x} is the annulus area between radius *x* and *x*+ Δ*x*; *I*_{x} is an indicator function that equals 1 when *x* < *d*_{ij} ≤ *x* + Δ*x* and is 0 otherwise; and *w*_{ij} is a weighting factor correcting for edge effects. The weighting factor is the proportion of the circumference of the circle centred at *i*, passing through *j*, which is inside the plot. It is closely related to Ripley's *K*-function (Ripley 1977), but *K*(*x*) refers to neighbourhoods of ≤*x* from the focal individual, whereas *D*(*x*) refers to an annulus between *x* and *x*+ Δ*x*.

*D*(*x*) was standardized by dividing by the mean density of given plants across the whole plot. This standardized index is called the relative neighbourhood density function (*NDF*) (Perry 2004), and simplifies display and interpretation of results as, for all values of *t*, the reference value under spatial randomness is 1. Thus *NDF* > 1 indicates spatial aggregation, and *NDF* < 1 indicates a pattern that is more regular than expected by random.

*NDF* was calculated at 1-cm intervals up to 25 cm by using SpPack (Perry 2004). Values of observed *NDF* were tested against the null hypothesis that distributions of dead seedlings were spatially independent of flowering ramets. The 95% confidence envelopes of *NDF* were estimated from 499 simulations of a random point process.

#### safe site for establishment

To evaluate the distribution of safe sites for recruitment, we monitored seedling emergence and survival in a series of 10 permanent quadrats (0.5 × 0.5 m) at 0.5-m intervals along a 10-m transect, taking considerable care not to cause any disturbance that might bias future censuses of surface plants. The locations of newly emerged seedlings of *P. modesta* were mapped, and seedlings were marked with coloured toothpicks at 2- to 4-week intervals during the growing season from May to September in 2001, 2002 and 2003. Seedling survival was monitored until May 2004.

We recorded ground surface conditions to characterize the heterogeneity of microenvironments within each quadrat. We subdivided each quadrat into four 0.25 × 0.25 m subplots and rated the ground surface conditions in each subplot as mossy hammock with sparse vegetation cover, mossy hammock with dense vegetation cover or hollow bare ground. To determine whether seedling survival differed among the microenvironments, we compared the survival curves of seedling cohorts that emerged in 2001, 2002 and 2003 until spring 2004 among categories, using the Kaplan–Meier procedure for survival analysis (Kaplan & Meier 1958). This procedure estimates survival functions from the survival durations of the seedlings, defined as the number of days between seedling emergence and death. We performed a Mantel–Cox test for homogeneity of survival across the sites.