Multiscale spatial genetic structure within and between populations of wild cherry trees in nuclear genotypes and chloroplast haplotypes

Abstract Spatial genetic structure (SGS) of plants mainly depends on the effective population size and gene dispersal. Maternally inherited loci are expected to have higher genetic differentiation between populations and more intensive SGS within populations than biparentally inherited loci because of smaller effective population sizes and fewer opportunities of gene dispersal in the maternally inherited loci. We investigated biparentally inherited nuclear genotypes and maternally inherited chloroplast haplotypes of microsatellites in 17 tree populations of three wild cherry species under different conditions of tree distribution and seed dispersal. As expected, interpopulation genetic differentiation was 6–9 times higher in chloroplast haplotypes than in nuclear genotypes. This difference indicated that pollen flow 4–7 times exceeded seed flow between populations. However, no difference between nuclear and chloroplast loci was detected in within‐population SGS intensity due to their substantial variation among the populations. The SGS intensity tended to increase as trees became more aggregated, suggesting that tree aggregation biased pollen and seed dispersal distances toward shorter. The loss of effective seed dispersers, Asian black bears, did not affect the SGS intensity probably because of mitigation of the bear loss by other vertebrate dispersers and too few tree generations after the bear loss to alter SGS. The findings suggest that SGS is more variable in smaller spatial scales due to various ecological factors in local populations.

Genetic differentiation between populations is expected to differ between genetic markers under different modes of inheritance (Ennos, 1994;Petit et al., 2005). Maternally inherited loci are predicted to show higher interpopulation genetic differentiation than biparentally inherited loci due to smaller effective population sizes and fewer opportunities of gene flow in the former loci (Hamilton & Miller, 2002); (Appendix S1). The ratio of pollen and seed flow between populations can be estimated from the ratio of interpopulation genetic differentiation in maternally and biparentally inherited genetic markers (Ennos, 1994); (Appendix S1). The median of pollen-to-seed-flow ratio estimates was 17 in various coniferous and angiosperm taxa, indicating more extensive pollen flow than seed flow between plant populations (Petit et al., 2005).
In addition to genetic differentiation between populations, spatial genetic structure (SGS) within populations reflects gene dispersal in a local scale (Epperson, 2007). Fine-scale SGS is often described as declining genetic relatedness with increasing spatial distance between individuals in a local area (Vekemans & Hardy, 2004) and quantified as linear regression of the genetic relatedness on the logarithm of spatial distance (Furstenau & Cartwright, 2016).
The index Sp calculated from the regression slope and the relatedness between neighbors is often used to compare SGS intensity among populations (Vekemans & Hardy, 2004). The SGS intensity is expected to differ between genetic markers under different modes of inheritance (Chybicki et al., 2016). Maternally inherited loci are expected to show higher SGS intensity than biparentally inherited loci due to lower effective population densities and fewer opportunities of gene dispersal in the former loci (Hardy & Vekemans, 1999) (Appendix S1). The ratio of pollen and seed dispersal area can be inferred from the ratio of regression slopes in maternally and biparentally inherited loci (Chybicki et al., 2016); (Appendix S1). Because of the low polymorphism of maternally inherited loci within populations, fine-scale SGS has rarely been compared between genetic markers with maternal and biparental inheritance. Variants of a cytoplasmic male sterility gene (de Cauwer, Dufay, Cuguen, & Arnaud, 2010) and chloroplast haplotypes (Latouche-Hallé, Ramboer, Bandou, Caron, & Kremer, 2003;Silvestrini, McCauley, Zucchi, & Santos, 2015;Torroba-Balmori et al., 2017), which are maternally inherited, exhibited more intensive SGS than biparentally inherited nuclear genotypes.
As theoretically expected, empirical studies of plant species suggest that both genetic differentiation between populations and SGS intensity within populations are associated with pollen and seed dispersal systems (Duminil et al., 2007;Hamrick & Godt, 1996;Hardy et al., 2006;Nybom, 2004;Vekemans & Hardy, 2004). Dioecious or outcrossing species, which seem to have wider ranges of pollen dispersal, tend to have lower genetic differentiation (Duminil et al., 2007;Hamrick & Godt, 1996;Nybom, 2004) and lower SGS intensity (Vekemans & Hardy, 2004) in nuclear markers. Species with seed dispersal by gravity, which seem to have narrower ranges of seed dispersal, tend to have higher genetic differentiation in organelle markers (Duminil et al., 2007) and higher SGS intensity in nuclear markers (Hardy et al., 2006). The SGS intensity in nuclear markers increases as local plant density increases because higher density may result from shorter seed dispersal and may result in shorter pollen dispersal under the same effective population size (Epperson, 2007;Hardy et al., 2006). In addition to pollen and seed dispersal systems, various factors, such as perenniality (Duminil et al., 2007), geographic distributional range (Duminil et al., 2007), colonization history (Hamrick & Trapnell, 2011), affect the genetic differentiation or SGS intensity. Among populations within species, habitat conditions, such as disturbance (Rico & Wagner, 2016) and fragmentation (Yamagishi, Tomimatsu, & Ohara, 2007), affect the SGS intensity in nuclear markers. Although these effects on interpopulation genetic differentiation or within-population SGS intensity have often been evaluated using either nuclear or organelle markers, both markers have been rarely used together to evaluate those factors.
Wild cherry species of the genera Cerasus and Padus are common trees in mountainous regions on the Japanese mainland (Iwasaki, Aoki, Seo, & Murakami, 2012;Tsuda, Kimura, et al., 2009a). Many microsatellites in both nuclear and chloroplast genomes are available in the cherry species (Cho, Yoon, & Kim, 2018;Kato et al., 2014). As in most angiosperms, chloroplast genomes are inherited maternally in Cerasus species (Brettin, Karle, Crowe, & Lezzoni, 2000). Thus, SGS in cherry tree populations can be compared between maternally and biparentally inherited loci using chloroplast and nuclear microsatellites, respectively. We expected higher interpopulation genetic differentiation and more intensive within-population SGS in chloroplast haplotypes than in nuclear genotypes as mentioned above.
Thus, both species are expected to have similar ratios of pollen and seed flow between populations. Also within populations, both species are expected to have similar ratios of pollen and seed dispersal area.
Cerasus and Padus species have different life history traits, such as regeneration strategies. Padus species have more restricted regeneration habitats than Cerasus species (Masaki, 2002), which may result in different patterns of spatial tree distribution between the genera.
Populations of Asian black bears had declined until the 20th century due to human impacts (Oi & Yamazaki, 2006), and they have been locally extinct for about 150 years in some mountainous regions on the Japanese mainland (Hanai, 1980;Tsujino, Ishimaru, & Yumoto, 2010).
Thus, the seed disperser loss may reduce seed dispersal relatively to pollen dispersal, resulting in a difference in fine-scale SGS between nuclear genotypes and chloroplast haplotypes.
To study the expected higher interpopulation genetic differentiation and higher within-population SGS intensity in chloroplast haplotypes than in nuclear genotypes, we investigated 17 tree populations of three wild cherry species, Cerasus jamasakura (Koidz.) H.
K. Schneider. Next, we evaluated whether the spatial distribution of trees and the loss of seed dispersers affect fine-scale SGS in these populations.

| Study sites and sampling
We selected four sites where bears were present and four sites where bears were absent in mountainous regions on the Japanese mainland ( Figure 1) on the basis of the national survey of U. thibetanus distribution in 2003 (Ministry of Environment, 2004). At bearabsent sites, Asian black bears also had not been recorded in 1978 (Ministry of Environment, 2004) and had been rare since at least the middle of the 19th century (Hanai, 1980;Tsujino et al., 2010). Thus, most of cherry trees at bear-absent sites had been regenerated without seed dispersal by the bears.
Of the eight sites, we sampled C. jamasakura at four sites, C. leveilleana at six sites, and P. grayana at seven sites because some species were rare at some sites (Table 1). Thus, we investigated 17 populations of the three species at the eight sites. Linear sampling, which is associated with a linear transect, such as a river, road, or trail, was performed as well as random sampling to estimate SGS (Oyler- McCance, Fedy, & Landguth, 2013). We sampled trees along one or two routes at each site ( Figure S1) because random sampling from an entire area of seed dispersal ranges would require an impractical effort. For C. jamasakura and C. leveilleana populations at Okutama site and C. leveilleana population at Ogawa site, we selected two routes because their habitats were fragmented ( Figure S1). The total length of routes at each site ranged from 1.3 to 12.8 km depending on the local tree density (Table 1). We sampled all trees with a diameter >5 cm at the breast height, which potentially reproduced, within widths 10 m from both sides of a route except for topographically inaccessible sides. We recorded the locations of sampled trees using GPSMAP 64 (Garmin) and confirmed them using a fine-scale topo-  (Table 1). We collected leaves from sampled trees and dried them with silica gel.
Because Cerasus species sometimes hybridize with each other, we identified C. jamasakura and C. leveilleana based on multilocus F I G U R E 1 Study site locations and cherry species on the Japanese mainland. Gray circles indicate study sites. Black pixels indicate bear distribution in 1978 genotypes of the nuclear microsatellites using the reference genotypes of identified specimens of these species (Kato et al., 2014). To confirm whether loci were independent with rare null alleles in the nuclear microsatellites, significant deviation from the Hardy-Weinberg equilibrium at every locus and significant linkage disequilibrium for every locus pair were tested in each population using GenePop 4.6.9 (Rousset, 2008) in R 3.3.2 (R Core Team, 2016). For diploid genotypes at the confirmed nuclear loci, the expected heterozygosity (H E ) in each population and the Wright fixation indices within populations (F IS ) and among populations within species (F ST ) were estimated using GenePop 4.6.9. Significant deviation from the Hardy-Weinberg equilibrium (significantly positive or negative F IS ) at all loci in each population was tested using GenePop 4.6.9. Statistical significance (p < .05) was adjusted in multiple tests for the loci, locus pairs, and populations in Bonferroni method using the function p.adjust in R 3.3.2.
Because of the complete linkage among loci in a chloroplast haploid genome, we discriminated chloroplast haplotypes based on allelic combinations of the microsatellite loci. For the discriminated chloroplast haplotypes, H E in each population and F ST among populations within species were also estimated using GenePop 4.6.9.

| Genetic differentiation between populations
Pairwise F ST of 6, 15, and 21 population pairs in C. jamasakura, C. leveilleana, and P. grayana, respectively, was calculated in nuclear genotypes F ST(b) and chloroplast haplotypes F ST(m) using GenePop 4.6.9. Spatial distance (m) of these population pairs was calculated from the longitudinal and latitudinal coordinates of the study sites. Interpopulation genetic differentiation parameters in nuclear genotypes for every population pair in each species (Appendix S1). Relationship between these parameters and the logarithmic spatial distance is thought to be linear on a two-dimensional ground surface (Rousset, 1997).
The ratio of the genetic differentiation parameters in nuclear genotypes and chloroplast haplotypes depends on the ratio of pollen flow to seed flow between populations in the assumptions of Wright island model (Ennos, 1994;Hamilton & Miller, 2002;Petit et al., 2005;Wright, 1951). The ratio of ) is expected to be 3 when pollen flow is equal to seed flow and to increase as the pollen-to-seed-flow ratio increases (Appendix S1). To verify these patterns, we performed Wilcoxon signed rank

| Spatial genetic structure within populations
The Loiselle kinship coefficients in nuclear genotypes and chloroplast haplotypes were calculated for every pair of sampled trees in each TA B L E 1 Properties of tree populations of wild cherry species population using SPAGeDi 1.5 (Hardy & Vekemans, 2002). For the tree pairs, three-dimensional Euclidean distance (m) was calculated from the longitudinal and latitudinal coordinates and the elevations of sampled trees. The spatial distance between trees was transformed to the natural logarithm because relationship between the kinship coefficient and the logarithmic spatial distance is thought to be linear on a two-dimensional ground surface (Vekemans & Hardy, 2004). In each population, an intercept and a slope of linear regression from the logarithmic spatial distance to the kinship coefficient were estimated from nuclear genotypes and chloroplast haplotypes using the function lm in R 3.3.2. In the regression, we assumed that the kinship coefficient followed a normal distribution, which seemed valid in our samples although a few tree pairs sharing rare haplotypes showed extremely high kinship coefficient in chloroplast haplotypes ( Figures S3 and S4). To compare the SGS intensity among populations, the index Sp was calculated as is the regression slope, and F 1 is the mean kinship coefficient in the first distance class, which was defined as a <10 percentile of spatial distance in each population (Vekemans & Hardy, 2004).
To obtain statistical errors of the indices F 1 and Sp and the intercept and slope of regression, Mantel tests were performed. We randomly permuted sampled trees in each population, calculated the pairwise spatial distance, and estimated an intercept and a slope of linear regression from the permuted spatial distance to the observed kinship coefficient. We repeated this procedure 10 4 times and ob- Padus = 1), the mean tree interval, CV in tree intervals, and the bear absence (present = 0 and absent = 1). We selected models with low Akaike Information Criterion (AIC), that is, delta AIC < 2, using the function glm(family = gaussian) and dredge(rank = "AIC") in R 3.3.2.

| RE SULTS
In 17 tree populations of three wild cherry species, the mean interval of neighboring trees along a sampling route ranged from 42 to 244 m, indicating that tree density varied among the populations (    lotypes was significantly higher and significantly >3 times higher than that in nuclear genotypes (p < .001), indicating that pollen flow exceeded seed flow between populations (Appendix S1). The ratio of the genetic differentiation parameter in chloroplast haplotypes to that in nuclear genotypes was significantly higher in Padus (mean = 9.35) than in Cerasus (5.92, p = .008; Figure 2a), indicating more extensive pollen flow relative to seed flow in Padus (pollen-toseed-flow ratio = 7.35) than in Cerasus (3.92).
In the 17 populations, linear regression from the logarithmic spatial distance to the kinship coefficient was estimated in nuclear genotypes ( Figure S3) and chloroplast haplotypes ( Figure S4). The regression slopes in nuclear genotypes b b and chloroplast haplotypes b m were obtained, and b b (1 + F IS ) and b m were examined to evaluate pollen and seed dispersal (Appendix S1). In nuclear genotypes, significantly negative slopes b b (1 + F IS ) and significantly positive regression intercepts and indices F 1 and Sp were detected in 10 populations (p < .05; Table 2). In chloroplast haplotypes, this pattern was detected in five populations although an inverse pattern with a significantly positive slope b m and significantly negative intercept and Sp was detected in P. grayana population at Hiruzen site (p < .05; Table 2). The slopes did not significantly differ between Cerasus and Padus populations in chloroplast haplotypes (p = .696) or nuclear genotypes (p = .143) although much steeper slopes were found only in Padus (Figure 2b). The slope in chloroplast haplotypes b m was neither significantly >3 times higher than that in nuclear gen-  (Table 2).
Three indices, Sp in nuclear genotypes, Sp in chloroplast haplotypes, and the slope ratio b m /b b (1 + F IS ), did not significantly F I G U R E 3 Comparison of SGS indices in populations of different genera (Cerasus and Padus, a-c), with local tree densities (mean tree interval, d-f) and tree aggregation levels (coefficient of variation in tree intervals, g-i), and at bear-absent and bear-present sites (j-l). The SGS indices are Sp in nuclear genotypes (a, d, g, j) and Sp in chloroplast haplotypes (b, e, h, k) and ratio of regression slope in chloroplast haplotypes to that in nuclear genotypes when the latter is negative (shown as b m /b b (1 + F IS ); c, f, i, l). Lines indicate medians, boxes indicate 25 and 75 percentiles, and whiskers indicate ranges (a-c, j-l). Open and filled symbols indicate Cerasus and Padus populations, respectively (d-i). Triangles and circles indicate bear-absent and bear-present sites, respectively (d-i) differ between Cerasus and Padus populations (p > .143) although Sp tended to be higher in Padus than in Cerasus (Figure 3a-c). These indices were not significantly correlated with the mean tree interval (p > .825; Figure 3d-f). The correlations between CV in tree intervals and Sp were significant in nuclear genotypes (Kendall τ = .397, p = .027) and marginally significant in chloroplast haplotypes (Pearson r = .472, p = .055), indicating more intensive SGS as trees became more aggregated (Figure 3g,h). The slope ratio was not significantly correlated with CV in tree intervals (p = 1.000; Figure 2i).
The absence of Asian black bears did not significantly affect any indices (p > .149; Figure 3j-l). The model selection was consistent with these comparisons. Models selected to predict nuclear Sp frequently included a positive effect of the genus Padus, and models selected to predict chloroplast Sp frequently included positive effects of the genus Padus and CV in tree intervals (Table S2). A model selected to predict the slope ratio with the lowest AIC included no effects (Table S2).

| D ISCUSS I ON
Maternally inherited loci are predicted to have higher interpopulation genetic differentiation and more intensive within-population SGS than biparentally inherited loci because of smaller effective population sizes and fewer opportunities of gene dispersal in the former loci (Chybicki et al., 2016;Ennos, 1994;Hamilton & Miller, 2002;Hardy & Vekemans, 1999;Petit et al., 2005). To compare these loci, we examined chloroplast haplotypes in maternally inherited loci and nuclear genotypes in biparentally inherited loci. As predicted, the estimated genetic differentiation between populations of the studied cherry species was higher in chloroplast haplotypes than in nuclear genotypes. The ratio of interpopulation genetic differentiation in maternally and biparentally inherited loci is thought to depend on the ratio of pollen and seed flow between populations (Ennos, 1994).
The estimated pollen-to-seed-flow ratio (3.92 in Cerasus and 7.35 in Padus) indicates that pollen flow exceeds seed flow. This ratio is less than the median ratio (17) in various plant taxa (Petit et al., 2005).
In contrast to the interpopulation genetic differentiation, the estimated within-population SGS intensity did not significantly differ between nuclear genotypes and chloroplast haplotypes due to a itat. The SGS intensity in the colonized habitat seems to increase as descendants from the founders regenerate around them. In contrast, kin-structured seed dispersal, which leads to aggregated colonization from the same maternal source, seems to create intensive SGS in a founder population (García & Grivet, 2011;Torimaru, Tani, Tsumura, Nishimura, & Tomaru, 2007). Regeneration of the founders with overlapping seed shadows may reduce the SGS intensity.
Subsequent immigration to the colonized habitat would alter the SGS intensity (Hamrick & Trapnell, 2011). Thus, the SGS intensity varies depending on the stages of colonization history, which seems specific to individual populations. A significantly negative Sp index, which was not expected from the equilibrium of spatial population dynamics (Furstenau & Cartwright, 2016), was observed in a population. This result may reflect a nonequilibrium state in population development. Therefore, the variation in fine-scale SGS in the studied cherry populations may result from population-specific states in colonization history.

Consistent differences in SGS within and between populations
were found between Cerasus and Padus. In both spatial scales, more intensive SGS was observed in Padus than in Cerasus. These findings suggest lower effective population size and/or more limited gene flow in Padus than in Cerasus (Ennos, 1994;Hardy & Vekemans, 1999). Furthermore, the estimated pollen-to-seed-flow ratio was higher in Padus than in Cerasus, suggesting that the intensive SGS mainly results from more limited seed flow. Because the breeding systems and the pollen and seed dispersal systems are similar between these genera, their differences in SGS are related to other life history traits. A candidate trait is regeneration strategy, which is associated with different patterns of spatial tree distribution. At the study sites, local tree density was higher in Padus than in Cerasus.
Also in a sparse pine forest, the germination rate of sound seeds was higher in P. grayana than in C. leveilleana, resulting in higher juvenile density (Shirota, Miyauchi, Saito, Maruyama, & Okano, 2015a, 2015b. In a mature broadleaf forest, P. grayana occurred in only young stands, whereas C. leveilleana occurred in both young and old stands (Masaki, 2002). These features indicate denser and more aggregated trees in more restricted regenerating habitats in Padus than in Cerasus. In tropical forests, more intensive SGS was observed in tree species, of which local density was higher (Hardy et al., 2006).
Denser and more aggregated distribution may result from shorter seed dispersal and may result in shorter pollen dispersal, leading to more intensive SGS (Doligez, Baril, & Joly, 1998). However, higher local density may also result in larger effective population size, which leads to less intensive SGS (Hardy & Vekemans, 1999). Thus, because effects of spatial distribution on SGS are complicated (Doligez et al., 1998), it is difficult to determine the factors responsible for the differences in SGS between Cerasus and Padus.
In the studied cherry populations, the SGS intensity increased as trees were more aggregated but did not depend on local tree density. As discussed above, aggregation is expected to increase the SGS intensity (Doligez et al., 1998), which is consistent with the former result. However, relatives are not always aggregated, and clumping of seedlings from various maternal sources can reduce the SGS intensity (Sagnard, Oddou-Muratorio, Pichot, Vendramin, & Fady, 2011).
Therefore, the aggregation observed in the studied cherry trees often consists of siblings originating from the same maternal source.
As mentioned previously, local density has the opposite effects on the SGS intensity, which increases due to the bias of gene dispersal toward shorter (Hardy et al., 2006) or decreases because the effective population density increases (Hardy & Vekemans, 1999). In an herb species in segregated pastures, the SGS intensity decreased as the population size in individual pastures increased (Rico & Wagner, 2016). In the studied cherry populations, the opposite effects may counterbalance each other, resulting in the unclear effect of local tree density on the SGS intensity.
The loss of effective seed dispersers, Asian black bears, is expected to reduce seed dispersal distance and result in more intensive SGS in chloroplast haplotypes than in nuclear genotypes (Calvino-Cancela et al., 2012;Gelmi-Candusso, Heymann, & Heer, 2017;Silvestrini et al., 2015). However, the ratio of regression slopes in chloroplast haplotypes and nuclear genotypes did not differ between populations at bear-absent and bear-present sites. Not only bears but also other animals, such as birds, macaques, and carnivores disperse seeds of the studied cherry species (Fujitsu et al., 2016;Koike, Kasai, et al., 2008;Koike, Morimoto, et al., 2008;Masaki et al., 2012). These animals, which are common in both bear-absent and bear-present sites, contribute to seed dispersal to some extent. This contribution may reduce the impacts of bear loss on fine-scale SGS.
In addition to the mitigation of bear loss, the duration of bear loss at bear-absent sites seems insufficient to increase the SGS intensity because the formation of fine-scale SGS requires several generations and subsequent replacement of trees (Hamilton & Miller, 2002), which may spend some hundred years in the studied cherry species.
Comparison of SGS in multiple spatial scales using genetic markers with different modes of inheritance is a feasible indirect method to evaluate pollen and seed dispersal (Chybicki et al., 2016;Hamrick & Trapnell, 2011;Petit et al., 2005). The findings from replicated populations of wild cherry trees imply that SGS is more variable in smaller spatial scales. Thus, the precise evaluation of gene dispersal seems to require a proper spatial scale, at which signals of the focal dispersal process exceed noises in SGS. Furthermore, the findings suggest that various ecological factors in local populations affect fine-scale SGS. Because of these potential factors, it is difficult to detect the impacts of a specific factor on fine-scale SGS.
In spite of these difficulties, fine-scale SGS gives insight into the characteristics of gene dispersal and the maintenance mechanisms of genetic diversity in local plant populations.

ACK N OWLED G M ENTS
This study was financially supported by the Japan Society for the Promotion of Science (JSPS KAKENHI Grant Numbers 25241026 and 17H00797 to TM). We thank the staff of the Kanto, Chubu, and Chugoku Regional Forest Offices, as well as the Hiruzen Experimental Forest of Tottori University, for their permission for our research; Yasuko Kawamata, Shoko Hisamatsu, and Akiko Takazawa for their help with the laboratory work; Hiroki Itoh for his advice on the statistical analyses; and Mitsue Shibata for her suggestions to improve the manuscript.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
SKo, TM, and TN designed this study. KS, SKi, SN, and TN conducted the field work. KS, SKi, and TN arranged the genetic data. TN analyzed the data and wrote the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Locations, tree sizes, nuclear multilocus genotypes, and chloroplast haplotypes of sampled trees in populations of cherry species in study sites: Dryad https ://doi.org/10.5061/dryad.gv5qp70.