Effects of fruit dimorphism on genetic structure and gene flow in the coastal shrub Scaevola taccada

Abstract

Background and Aims

Plant propagules often possess specialized morphologies that facilitate dispersal across specific landscapes. In the fruit dimorphism of a coastal shrub, Scaevola taccada, individual plants produce either cork-morph or pulp-morph fruits. The former is buoyant and common on sandy beaches, whereas the latter does not float, is bird-dispersed, and is common on elevated sites such as slopes on sea cliffs and behind rocky shores. We hypothesized that beach populations bridge the heterogeneous landscapes by serving as a source of both fruit types, while dispersal is biased for the pulp morph on elevated sites within the islands and for the cork morph between beaches of different islands. Based on this hypothesis, we predicted that populations in elevated sites would diverge genetically over time due to isolation by distance, whereas beach populations would maintain high genetic similarity via current gene flow.

Methods

The genetic structure and gene flow in S. taccada were evaluated by investigating genome-wide single nucleotide polymorphisms in plants from 17 sampling sites on six islands (belonging to the Ryukyu, Daito and Ogasawara Islands) in Japan.

Key Results

Geographical isolation was detected among the three distant island groups. Analyses within the Ryukyu Islands suggested that sandy beach populations were characterized by genetic admixture, whereas populations in elevated sites were relatively isolated between the islands. Pairwise F_ST values between islands were lowest between sandy beaches, intermediate between sandy beaches and elevated sites, and highest between elevated sites.

Conclusions

Dispersal across the ocean by cork morphs is sufficiently frequent to prevent genetic divergence between beaches of different islands. Stronger genetic isolation of elevated sites between islands suggests that bird dispersal by pulp morphs is restricted mainly within islands. These contrasting patterns of gene flow realized by fruit dimorphism provide evidence that fruit characteristics can strongly mediate genetic structure.

INTRODUCTION

Variation in seed dispersal phenotypes within a population is common across plant species and has critical ecological and evolutionary consequences (Howe and Smallwood, 1982; Rubio de Casas et al., 2012; Schupp et al., 2019). Generally, intraspecific variation in dispersal traits is continuous. However, in some cases, two or more types of fruits/seeds are produced with distinct dispersal characteristics within a plant population (Imbert, 2002). This phenomenon is called heteromorphism when distinct propagules are produced by a single individual plant (Imbert, 2002). However, when produced by different individual plants, it is called polyphenism or genetic polymorphism, caused by phenotypic plasticity or genetic variation, respectively (Gottlieb, 1984). In contrast to heteromorphic dispersal traits (Imbert, 2002), examples of genetic polymorphisms for dispersal traits are relatively sparse (Gottlieb, 1984; Traveset and Willson, 1998; Traveset et al., 2001; Whitney, 2005a, b; Emura et al., 2014). This is probably because of narrow ecological conditions favouring the maintenance of genetic polymorphism for dispersal traits within populations (Bonte et al., 2010; Fronhofer et al., 2011; Cenzer and M’Gonigle, 2019).

The morphology of plant propagules is often specialized for different dispersal methods via specific vectors, such as water, wind, ants, birds and mammals (Howe and Smallwood, 1982; Janson, 1983; Rubio de Casas et al., 2012). Therefore, to some extent, the morphology of a propagule determines the range of habitats it can reach (Sork et al., 1999; Ozinga et al., 2004). For example, water-dispersed fruits, characterized by corky and fibrous structures, are buoyant, which aids their dispersal through water bodies, such as rivers, lakes and oceans (Gunn and Dennis, 1999). Fruits dispersed by frugivorous birds have a pulpy mesocarp and can pass through dense vegetation and differences in elevation (Howe and Smallwood, 1982). Certain plants disperse their seeds employing two distinct dispersal vectors (Vander Wall and Longland, 2004).

In plants displaying genetic polymorphism in dispersal traits, different propagules may present distinct dispersal distances and differences in the types of landscapes they can travel. Consequently, they may form a specific genetic structure depending on the landscape. It has been reported that large-fruited plants of Frangula alnus in southern Iberia have stronger genetic structure at chloroplast DNA than small-fruited F. alnus in central Europe (Hampe and Bairlein, 2000; Hampe et al., 2003). Although this example clearly shows the consequences of dispersal traits on population genetic structures, those of genetic polymorphisms for fruit dispersal traits within populations have not yet been examined.

Genetic polymorphisms for dispersal traits have been reported in fruit colours (Whitney, 2005a), winged and wingless fruits (Ganders et al., 1977), and epizoochorous and myrmecochorous fruits (de Enrech and Mathez, 1998). In these cases, genetic dimorphism probably modifies rather than completely changes the dispersal kernels. For example, fruit colour polymorphism affects the preferences of frugivorous animal vectors and, consequently, the dispersal patterns (Traveset and Willson, 1998; Whitney, 2005b). A system with genetic dispersal polymorphism, wherein two morphs utilize distinct vectors, can be used to determine the contribution of seed dispersal across different landscapes in the formation of specific genetic structures.

We previously reported inter-individual fruit dimorphism of a dispersal-related morphological trait in the common subtropical and tropical coastal shrub Scaevola taccada (Emura et al., 2014). Two distinct fruit morphs exist within the populations of the species – individuals produce fruits either with a corky mesocarp or with a pulpy mesocarp, and are referred to as cork and pulp morphs, respectively (Emura et al., 2014). Scaevola taccada populations are largely polymorphic, where the cork morph is abundant in populations in the sandy beach habitat, and pulp morphs are more common in elevated sites, such as those established on sea cliffs or behind rocky shores (Emura et al., 2014). The dimorphism is presumably genetic as an examination of 4467 individuals over a broad geographical region revealed that two types are distinct, with only a 1.5 % detection rate for intermediate forms (Emura et al., 2014). The cork morph can float in water for at least 6 months, while the pulp morph is not buoyant (Emura et al., 2014). Both morphs possess white surfaces and a pulpy epicarp and are eaten by birds (Emura et al., 2012, 2014). The corky mesocarp of the cork morph remains intact even after being digested by birds. The appearance of the two morphs is similar, and the reflectance spectra of fruits suggest that they cannot be distinguished visually by frugivorous birds (Tanaka et al., 2015). However, bird droppings contain pulp morphs more frequently than expected by chance (N. Emura et al., unpubl. data). This is probably because the pulpy mesocarp of a pulp-morph fruit serves as a greater reward to frugivores and modifies foraging behaviour; thus, birds tend to stay longer on plants with pulp-morph fruits or may learn the locations of the plants. The study system is a rare case of fruit dimorphism showing a considerable difference in the relative efficiency of the dispersal of morphs across distinct landscapes, i.e. across the water and the land, and in the likelihood of post-dispersal establishment in different habitats, i.e. sandy beaches and elevated sites. The genetic structure of the species has been studied at a global scale, and strong genetic structures are observed between geographically distinct regional groups, including Africa, India and the west side of the Malay Peninsula, South-east Asia and Australia, and South China (Banerjee et al., 2021). Analysing genetic structures on a smaller geographical scale using hyper-polymorphic markers may elucidate the roles of distinct fruit morphs in the current gene flow.

Here, we conducted a genetic analysis of cork- and pulp-morph individuals sampled from different island habitats in the north-western Pacific Ocean using restriction site-associated DNA sequencing (RAD-seq). The RAD-seq method allows for the analysis of multiple genome-wide single-nucleotide polymorphisms (SNPs) (Davey and Blaxter, 2011; Davey et al., 2013).

The objectives of this study were three-fold. The first was to elucidate the overall geographical genetic structure of S. taccada in the north-western Pacific Ocean region. In this region, the study species is distributed across three island groups, the Ryukyu, Daito and Ogasawara Islands. The Ryukyu Islands are continental islands that belong to the Ryukyu archipelago. Daito and Ogasawara Islands are oceanic islands separated from the Ryukyu Islands by the ocean. By sampling wide-ranging areas, we examined the genetic structure of the populations of the study species between three nested geographical hierarchies: island groups, between islands within an island group and between populations within islands. The second objective was to examine whether within-population genetic structuring is associated with fruit morphs. The presence or absence of strong assortative mating that can cause genetic isolation between morphs is critical for understanding the mechanisms responsible for maintaining dimorphism. However, no morphological or phenological distinctions were observed between the flowers of trees that bear the two types of morphs.

The third objective was to evaluate the hypothesis that beach populations serve as a source of both fruit types. It was hypothesized that there exists a bias for the dispersal of pulp-morph fruits by birds to elevated sites within the islands and of cork-morph fruits between beaches of different islands within the range of regular long-distance sea-current dispersals. Based on this hypothesis, we predicted that populations established in elevated sites would diverge genetically over time due to isolation by distance while beach populations would maintain higher genetic similarity within the geographical ranges of regular sea-current dispersals. The application of RAD-seq allowed us to estimate pairwise indices, such as F_ST^R (Reich et al., 2009), Gʹ_ST (Hedrick, 2005) and Jost’s D (Jost, 2008), representing gene flow and genetic differentiation between populations (Weir and Cockerham, 1984; Alcala et al., 2014). We evaluated the contribution of the types of landscape between plant populations in determining the level of gene flow.

MATERIALS AND METHODS

Plant materials

For RAD-seq analyses, we selected 17 sampling sites on six islands within the subtropical region of Japan: Tokunoshima Island, Okinawa-honto Island, Miyakojima Island, Ishigakijima Island, Minami-daitojima Island and Hahajima Island (Fig. 1A, Table 1). The first four islands belong to the Ryukyu Islands group. Minami-daitojima Island belongs to Daito Islands group and is separated from the nearest island in the Ryukyu Islands by ~300 km. Hahajima Island is an oceanic island within the Ogasawara Islands, separated from the other investigated islands by 1100–1800 km. The study sites were categorized into two habitat types: sandy beaches and elevated habitats. The latter was represented by slopes on the sea cliffs and included other habitats elevated from the sea level, i.e. slopes behind rocky shores and an inland open site (Table 1). Most elevated sites are located higher than the level of high tide in the normal tidal cycle. They can be inundated during occasional storm surges, but we judged low sand accumulations represent the lower probability of deposition for small floating particles. The distance between sites on the same islands ranged from 2 to 40 km (Supplementary Data Fig. S1A).

Fig. 1.

Sampling locations and the results of population genetic structure analyses. (A) Map showing the locations of six islands of the Ryukyu, Daito and Ogasawara Islands where sampling was performed for RAD-seq. The northern limits of Scaevola taccada distribution in the Ryukyu and Ogasawara Islands are marked by arrowheads. (B) CV error as a function of the number of genetic clusters (K). (C) ADMIXTURE genetic clustering analysis results (for K = 2, 3 and 10) for 95 plants from 17 sites in the Ryukyu, Daito and Ogasawara Islands. Fruit morphs are indicated by white, black and grey circles for cork, pulp and intermediate morphs, respectively, and schematic diagrams of fruit types are shown on the right. Sampling sites are denoted by abbreviations of island names followed by habitat type (s, sandy beach; ns, sea cliff including rocky shore and inland; see also Supplementary Data Fig. S1). (D) The results of ADMIXTURE analysis for K = 10 arranged by habitat types within the Ryukyu Islands.

Table 1.

Site description [island, habitat type, site code and pulp-morph frequency by Emura et al. (2014); population size and elevation], number of samples (C, P and I for cork, pulp and intermediate morphs, respectively), and genetic diversity within sampling sites expressed by the unbiased expected heterozygosity (uH_E) and fixation index (F_IS) calculated based on 7278 SNPs in RAD-seq.

Site description							Number of samples			uH  E	F  IS
Islands		Habitat type	Site code	Pulp-morph frequency	Population size α	Elevation (m)	C	P	I
Ryukyu Islands
	TO (Tokunoshima Island)	Elevated (slope on the sea cliff)	TO-e1	0.80	1000 < α ≲ 5000	35	3	3	0	0.16	−0.02
		Elevated (slope behind rocky shore)	TO-e2	0.54	100 < α ≲ 1000	7	3	3	0	0.17	−0.04
		Sandy beach	TO-s	0.02	100 < α ≲ 1000	4	3	0	0	0.16	−0.13
	OK (Okinawa-honto Island)	Elevated (slope on the sea cliff)	OK-e1	0.75	1000 < α ≲ 5000	15	3	3	0	0.16	−0.04
		Elevated (slope on the sea cliff)	OK-e2	0.63	1000 < α ≲ 5000	3	3	3	0	0.16	−0.03
		Sandy beach	OK-s1	0.16	100 < α ≲ 1000	2	3	2	0	0.17	−0.07
		Sandy beach	OK-s2	0.02	100 < α ≲ 1000	2	3	2	0	0.17	−0.05
	MY (Miyakojima Island)	Elevated (slope on the sea cliff)	MY-e	0.86	1000 < α ≲ 5000	22	3	3	0	0.17	0.01
		Sandy beach	MY-s	0.04	100 < α ≲ 1000	2	3	2	0	0.18	−0.06
	IS (Ishigakijima Island)	Elevated (slope on the sea cliff)	IS-e1	0.30	100 < α ≲ 1000	6	3	3	1	0.18	0.02
		Elevated (slope behind rocky shore)	IS-e2	0.02	100 < α ≲ 1000	30	3	3	3	0.18	0.07
		Sandy beach	IS-s	0.04	100 < α ≲ 1000	3	3	2	1	0.19	0.02
Daito Islands
	MD (Minami-daitojima Island)	Elevated (slope on the sea cliff)	MD-e1	0.96	1000 < α ≲ 5000	25	3	3	0	0.16	−0.01
		Elevated (slope on the sea cliff)	MD-e2	0.57	1000 < α ≲ 5000	15	3	2*	0	0.13	0.03
Ogasawara Islands
	HA (Hahajima Island)	Elevated (inland)	HA-e1	0.05	100 < α ≲ 1000	80	3	3	0	0.16	−0.01
		Elevated (slope on the sea cliff)	HA-e2	0.04	100 < α ≲ 1000	80	3	2	0	0.15	−0.14
		Elevated (slope behind rocky shore)	HA-e3	0	100 < α ≲ 1000	8	3	0	0	0.15	−0.23
Total							51	39	5

^*One sample (MD-e2-6) was excluded because of an exceptionally low number of reads (Supplementary Data Table S1).

Sampling of Scaevola taccada (Gaertn.) Roxb. [Goodeniaceae] was performed during the summer (fruiting) season of 2012. At most sites, six trees were sampled, including three of each morph type, i.e. cork and pulp morphs, except for those sites having less than three pulp-morph trees (Fig. 1A, Table 1). At sites with an intermediate morph, we sampled up to three trees to examine whether they had a specific genetic background. In total, we sampled the leaves of 96 individuals across all 17 study sites, i.e. 51, 40 and five trees with cork, pulp, and intermediate morphs, respectively, with one pulp-morph sample excluded later during the genotyping process (Table 1). Although the sample sizes per site were from three to nine (mostly six), SNP numbers used in this study were >7000. Willing et al. (2012) showed that for high SNP numbers (>1000), RAD-seq provides accurate estimates of population genetic differentiation with small sample sizes of two to six per site. From each sampled shrub, ~4-cm² pieces of healthy leaf blades were harvested. Each sample was placed in a teabag made of thin paper and immediately desiccated using silica gel in a plastic bag. Samples were transported to the laboratory and stored at room temperature (10–25 °C) until DNA extraction.

DNA extraction, RAD-seq and genotyping

DNA was extracted using a DNeasy Plant Mini Kit (Qiagen, Hilden, Germany). A library for double-digest RAD-seq (Peterson et al., 2012) was created using a modified protocol, as previously described (Sakaguchi et al., 2015), and the combination of restriction enzymes, BglII and NdeI, was used to digest genomic DNA. PCR conditions were modified as follows: 94 °C for 30 s; followed by 30 cycles of 98 °C for 10 s, 65 °C for 30 s and 72 °C for 30 s; and a final extension at 72 °C for 5 min. The prepared RAD-seq library was sequenced by BGI (BGI-HongKong Co., Ltd, Hong Kong, China) with 49-bp single-end reads in one lane using the Illumina HiSeq2000 sequencing system (Illumina, San Diego, CA, USA). Overall, 78 351 718 raw reads (median 812 194 reads per sample) were obtained from 96 samples (Supplementary Data Table S1). One sample with an exceedingly low number of reads (MD-s2) was excluded from further analyses. Sequence data are available from the DNA Data Bank of Japan (DDBJ, DRA008523; https://ddbj.nig.ac.jp/DRASearch/submission?acc=DRA008523).

Quality filtering and adapter trimming were performed using the software fastp (ver. 0.20.0; Chen et al., 2018) with parameters ‘-f 3 -l 46’. The first three nucleotides, which were of relatively low quality compared to the others, were removed and aligned to 46 bp in length. SNPs were identified using the denovo_map.pl pipeline in STACKS (ver. 2.61; Catchen et al., 2011). The pipeline executes each of the STACKS components individually. Options of ustacks were set as m = 3 and M = 2, and that of cstacks was set as n = 1. After creating stacks, SNPs were exported by the ‘populations’ command with the following settings: minimum percentage of individuals across populations required to process a locus, R = 0.6; maximum observed heterozygosity, --max-obs-het = 0.6; and minimum minor allele frequency, --min-maf = 0.01. Only the first SNP of each locus was retained using the ‘--write-single-snp’ option. Then, an additional filtering step was performed for the obtained SNPs to extract only those SNPs that were genotyped in more than 60 % of individuals in the two groups, i.e. from the Ryukyu Islands and Minami-daitojima, and the Ryukyu Islands only, respectively. After filtering, the genotyping rate and observed heterozygosity (H_O) of each sample were calculated using Geno Summary in TASSEL (ver. 5.0; Bradbury et al., 2007). The three SNP datasets obtained from the above process for the three population sets, i.e. overall, from the Ryukyu and Daitojima Islands, and the Ryukyu Islands only, were used for subsequent analyses.

Population genetics analyses

Unbiased expected heterozygosity (uH_E) and fixation index (F_IS) were calculated at all study sites using GenAlex (ver. 6.4; Peakall and Smouse, 2006) for all 95 samples with 7278 SNPs to estimate the levels of genetic diversity and mating patterns. The samples were classified into two types of three hierarchical levels: island/site/morph and habitat type/site/morph. To evaluate which level of the grouping explained a significant part of the total genetic variation, analysis of molecular variance (AMOVA) was performed using the poppr.amova function in the poppr package (Kamvar et al., 2014) in R (ver. 3.6.2; R Core Team, 2016). Random permutation tests of AMOVA with 1000 permutations were conducted with the randtest function in the ade4 package (Dray and Dufour, 2007) in R. Additionally, AMOVA was applied for 90 samples, excluding intermediate morphs with 7135 SNPs, to evaluate whether the inclusion of intermediate morph affected the results.

Genetic differentiation among sites was estimated using the model-based program ADMIXTURE (ver. 1.3.0; Alexander et al., 2009) and principal component analysis (PCA) using the dudi.pca function of the adegenet package within R (ver. 4.0.3; R Core Team, 2020) using all 95 samples with 7278 SNPs. Individuals were assigned to several genetic groups, i.e. clusters, with membership probabilities using the ADMIXTURE program. Meaningful K values were chosen from the lowest cross-validation (CV) error for each instance of K = 1–20.

Phylogenetic network analysis was performed using the neighbour-net method in the SplitsTree program (ver. 4.14.5; Huson and Bryant, 2006), taking the average sequence divergence, i.e. uncorrected p-distance, as the genetic distance. In addition to an analysis of overall samples, a PCA was conducted for groups containing S. taccada samples from the Ryukyu and Daito Islands as well as the Ryukyu Islands only (7060 and 6408 SNPs, respectively, with > 60 % of individuals).

Reich-Patterson F_ST^R (Reich et al., 2009) was calculated for all population pairs using the reich.fst R script (Junker et al., 2020, available at GitHub, https://github.com/jessicarick/reich-fst) using samples of the Ryukyu Islands. Gʹ_ST and Jost’s D were also estimated using the fastDivPart function of the diveRsity package (Keenan et al., 2013) in R. The estimated pairwise F_ST^R, Gʹ_ST and Jost’s D values were classified into four habitat type combinations: ‘within the same island’, ‘between sandy beaches’, ‘between elevated and sandy beach sites’, and ‘between elevated sites’. The latter three included combinations between islands only. The relative magnitudes of directional gene flow between populations were estimated based on Jost’s D using the divMigrate function of the diveRsity package (Keenan et al., 2013, Sundqvist et al., 2016). The divMigrate function of the diveRsity package produces a migration network graph with relative values for gene flow among populations scaled to 1 at the highest magnitude estimated. In the divMigrate analysis, an additional filtering step was performed on the SNP data set to remove SNPs with a minor allele frequency of < 0.03 to reduce noise due to minor alleles. Multiple pairwise comparisons based on Welch’s two-sample t-test with Holm correction among the four or five categories were performed using the pairwise.t.test function of R. Then, correlations between the values of F_ST^R/(1 − F_ST^R) and geological distances were evaluated using the Mantel test with 999 permutations and Spearman’s rank correlation using the mantel function in the vegan package (Oksanen et al., 2022) of R for each of the habitat type combinations described above.

RESULTS

The genotyping rate (percentage loci) of 95 samples was a minimum of 52 % and a maximum of 91 %, with an average of 79 %, with no notable bias by island or habitat environment (Supplementary Data Table S1). Genetic diversity (uH_E) and fixation indices ranged from 0.13 to 0.19 and −0.23 to 0.07, respectively, in the study populations (Table 1). In the ADMIXTURE analysis, the CV error values were low with K = 1, 2 and 3 (Fig. 1B). The overall genetic structure was dependent on geographical isolation between Ryukyu and Ogasawara (1100–1800 km) and between Ryukyu and Daito Islands (300–700 km) as detected in structures with K = 2 and K = 3 (Fig. 1C). At K = 2, plants on Ryukyu and Ogasawara Islands were grouped into distinct clusters, and plants on Daito Islands were a mix of the two gene clusters (Fig. 1C). In the analysis with K = 3, all plants from the Ryukyu Islands and some plants from the Daito Islands, the other plants from the Daito Islands, and all plants from the Ogasawara Islands were grouped into three distinct clusters, respectively (Fig. 1C). In addition, these patterns were supported by the phylogenetic network (Fig. S1B). No further clear genetic structures were detected with K = 10 in the ADMIXTURE analysis (Fig. 1D). However, when populations within the Ryukyu Islands were rearranged based on habitat environments, the analyses indicated that sandy beach populations were characterized by genetic admixture. On the other hand, elevated populations were relatively distinct from one another (Fig. 1D).

To elucidate how the distance between islands and between populations within islands and habitat types interacts with the genetic structure, PCA was applied hierarchically; that is, for all data (Fig. 2A), for the Ryukyu and Daito Islands (Fig. 2B), and for the Ryukyu Islands (Fig. 2C, D). We identified a distinct separation in PC1 vs. PC2 between the Ryukyu Islands, Ogasawara Islands (HA) and one of the two populations from the Daito Islands (MD) (Fig. 2A, B). Interestingly, in PCA, PC1 vs. PC2, for the Ryukyu Islands, the sandy beach populations on different islands tended to overlap, whereas those on elevated sites on different islands tended to be separated (Fig. 2C). A similar trend was seen in the plot of PC3 vs. PC4, although separations of elevated sites were stronger for distinct island pairs from those showing stronger separations in PC1 vs. PC2 (Supplementary Data Fig. S2). The patterns explain the low contribution rate of each PC, e.g. 4.2, 3.9, 2.8 and 2.5 % for PC1, PC2, PC3 and PC4, respectively (Fig. 2C; Fig. S2C), and different sets of SNPs contributed to separations of the elevated sites along the series of PCs, corresponding to different combinations of island pairs.

Fig. 2.

Principal component analysis (PCA) of PC1 and PC2 on RAD-seq data. PCAs for all data, including the Ryukyu, Daito and Ogasawara Islands (A), the Ryukyu and Daito Islands (B), and the Ryukyu Islands (C, D). Data sets from 95, 81 and 70 plants with 7278, 7060, and 6408 SNPs were used, respectively. In A, B and C, purple and orange circles represent sandy beaches and elevated sites, respectively. In D, blue, red and grey circles represent cork pulp, and intermediate morphs, respectively. Data points from the same islands (TO, OK, MD, MY, IS and HA, see Supplementary Data Fig. S1) are indicated using grey lines and those from the same populations are indicated using grey-shaded areas. The first two principal components are shown with contribution rates in parentheses.

We detected no genetic differentiation between fruit morphs within populations in the ADMIXTURE analysis (Fig. 1C, D) and PCA (Fig. 2D; Supplementary Data Fig. S2D). AMOVA of the geographical hierarchy, i.e. island–site–morph–individual, indicated that the genetic variance mainly originated from variation between individuals, ~75 %, and the differences among morphs explained only ~3 % (Table S2A). Statistically significant population differentiation was not evident among habitat types based on AMOVA of the habitat type hierarchy, i.e. habitat types–site–morph–individual (Table S2A). Additional AMOVAs confirmed that the exclusion of five intermediate morphs resulted in similar results (Table S2B).

ADMIXTURE (K = 2 and 3) and PCA results indicated that genetic separation between Ryukyu, Daito and Ogasawara Islands could be attributed to historical gene flow or past long-distance dispersals. Therefore, patterns in pairwise F_ST^R, Gʹ_ST and Jost’s D were examined using data from the Ryukyu Islands, assuming that these pairwise values represent the level of current gene flow. The frequencies of the occurrence of cork and pulp morphs were highest in the sandy beach and elevated habitats, respectively (Table 1). Thus, we assumed fruit dispersal morphs facilitate migration to different landscapes and dispersal within the corresponding habitat types. F_ST^R values of ‘between elevated sites’ were significantly higher compared to those of other habitat combinations, and those of ‘between sandy beaches’ were significantly lower compared to the rest of the habitat combinations between islands (P < 0.05; Fig. 3). Pairwise F_ST^R within islands showed intermediate values. (Fig. 3). Gʹ_ST and Jost’s D values also showed similar patterns to F_ST^R values (Supplementary Data Fig. S3). The relative magnitudes of directional gene flow among populations were not significantly different between the five categories of habitat type combinations, but the beach–beach habitat combination showed a tendency to have higher gene flows (Fig. S4). A Mantel test within the Ryukyu Islands identified significant positive correlations between genetic and geographical distances for combinations of habitat types involving elevated sites. However, isolation by distance was not evident among sandy beach populations (Fig. S5).

Fig. 3.

F  _ST  ^R between populations within islands and for habitat-type combinations between islands of the Ryukyu Islands. The box and lines denote the median, first and third quartile values. Different letters indicate significant differences between the habitat-type combinations according to Welch’s two-sample t-test with Holm correction (P < 0.05).

DISCUSSION

Coastal plants, including S. taccada, often have propagules specialized for sea drift dispersal. The effects of geographical isolation were evident in overall geographical genetic structures. The greatest isolation was between Ogasawara Islands and Ryukyu and Daito Islands, and the second-largest isolation was between Ryukyu Islands and Daito Islands. In this region, the Ryukyu Current flowing north-eastward along the Ryukyu archipelago may facilitate gene flow between islands within Ryukyu Islands but restrict gene flow to Daito Islands (Yuan et al., 1998; Zhao et al., 2020). Furthermore, the distance between the islands within Ryukyu Island group is reduced when sea level is lower (Oshiro and Nohara, 2000), and beaches of consecutive islands may serve as stepping stones for gene flow between distant islands. However, the distance to oceanic islands, including Daito and Ogasawara, does not change significantly based on sea level.

All eight sister species of S. taccada produce dark-purple pulpy fruits, which attract frugivorous birds (Wheelwright and Janson, 1985; Howarth et al., 2003), and have narrow distribution ranges, often endemic to islands, e.g. New Caledonia, Fiji and Samoa (Howarth et al., 2003). Therefore, we speculate that S. taccada has expanded its geographical range throughout the Pacific and Indian Oceans through the drift dispersal of cork-morph fruits. Genetic structures reflecting geographical isolation have been reported in the study species (Banerjee et al., 2021) and in other coastal plants that disperse seeds by ocean currents (Takayama et al., 2006; Hanaoka et al., 2014; Miryeganeh et al., 2014; Gallaher et al., 2017; Yang et al., 2017).

The results of ADMIXTURE analyses clearly showed that the morph type does not dictate the genetic structure, but rather the habitat type and its interaction with morph types explain the genetic structure at the geographical ranges within the Ryukyu Islands. Our finding that the highest gene flow occurred between sandy beach populations supports the hypothesis that the cork morph is necessary for dispersal between islands. The genetic admixture of beach populations within the Ryukyu Islands is likely to be a consequence of current gene flow via the long-distance dispersal of cork-morph fruits, as they can potentially float for longer than 6 months (Emura et al., 2014). We speculate that S. taccada populations expanded from sandy beaches to elevated sites on sea cliffs and behind rocky shores within each island through the dispersal of pulp-morph fruits by frugivorous birds, resulting in genetic isolation between populations at the elevated sites of distinct islands. Generally, seed dispersal by frugivorous birds is confined to a range of several hundred metres because seeds only remain in the digestive tracts of birds for short periods (Jordano, 2017; Rehm et al., 2019). Furthermore, previous studies showed that plants dispersed by birds can have strong genetic structuring (Hampe et al., 2003; Worth et al., 2010).

An alternative explanation is possible because exceptional events that disperse propagules beyond their regular dispersal kernel can be crucial for long-distance dispersals (Higgins et al., 2003; Vellend et al., 2003), and evidence of long-distance dispersals across the sea by migratory birds exists (Viana et al., 2013, 2016). However, this is probably not the case for S. taccada because fruiting occurs in summer, which is out of season for migratory birds (McWhirter et al., 1996). We would expect equivalent (or higher) levels of gene flow between elevated sites if birds were equally (or more) important as drift for transporting fruits between islands. One may consider the role of Holocene glacial retreat in the origins of elevated populations when the sites were closer to sea level. However, the higher frequency of the pulp morph at most elevated sites and moderate levels of F_ST^R within islands support the idea that bird dispersal within islands is the primary determinant of the genetic compositions of elevated sites.

F  _ST  ^R values between different habitat types support the hypothesis that S. taccada fruit dimorphism allows the species to cross the ocean using cork-morph fruits and climb cliffs using pulp-morph fruits in successive generations. We treated the higher frequency of cork and pulp morphs in the sandy beach and elevated populations, respectively, as a consequence of morph-specific dispersal across ocean and inland elevational differences for cork and pulp morphs, respectively, and on-site superiority in short-distance dispersal within habitat types. We did not observe an apparent dependency of population sizes on habitat types (Table 1). Therefore, the differences in F_ST^R between habitat combinations are explained by neither population size nor population growth rate. The pattern identified here can be regarded as a specific form of isolation in environments, with the rates of gene flow higher among similar habitat environments (Sexton et al., 2014). Based on the original concept, the rates of gene flow are expected to be higher among similar habitat environments. In the present study, beach populations of different islands were observed to be tightly connected, whereas S. taccada populations in the elevated sites of these islands did not reveal a direct connection. Although our hypothesis attempts to explain habitat-type dependency of morph frequencies, the influence of stochastic forces cannot be ignored, as variations in morph frequencies were observed between populations with shared habitat types. In addition, our data also suggested the importance of the founder effect in remote oceanic islands, such as the Ogasawara Islands, where the cork morph dominates in sea-cliff habitats. We also need to note here the role of the gravity dispersal of unconsumed fruits ((Mora and Smith-Ramirez, 2016). We observed that visitation frequency by frugivorous birds to S. taccada fruits is not necessarily high in some populations (N. Emura et al., unpubl. data). In such cases, poor dispersal due to gravity may enhance the founder effects in determining morph frequencies.

We hypothesize that S. taccada dimorphism is controlled by a few Mendelian loci based on our observation that fruit morphs are fixed stably within individual plants but are distinctive between plants, with a very rare intermediate morph (Emura et al., 2014). Experimental crosses between the two morphs produced viable seeds (N. Emura et al., unpubl. data). Segregation patterns of fruit morphs in successive generations will allow us to estimate the number of loci involved in determining the fruit dimorphism of S. taccada; however, we need to wait until seedlings grow to a sufficient size to produce fruit before the analyses.

As the fruit pericarp is a maternal tissue, the genotypes that determine the dispersal characteristics of propagules may differ from those of the seeds within (Roach and Wulff, 1987). We observed pulp-morph fruits produced by plants grown from seeds collected from cork-morph mothers (N. Emura et al., unpubl. data). The natural crossing of two morphs can allow the seeds carried by one fruit morph to produce the other fruit morph in the next generation. Potentially, this is a reason why both sandy beaches and elevated sites have individuals with both morphs. Furthermore, we hypothesize that this characteristic of the inheritance of fruit dimorphism may enable S. taccada to be dispersed across heterogeneous landscapes over multiple generations. The shift of fruit/seed morphology from being specialized for long- to short-distance seed dispersal (e.g. from ocean currents to other vectors in inland habitats) has been recognized as an evolutionary trend in island plants (Carlquist, 1974; Takayama et al., 2005; Kudoh et al., 2006). In the case of fruit dimorphism in S. taccada, the shift between vectors is reversible if fruit dimorphism is maintained within populations.

No evidence for morph-associated strong assortative mating that causes genetic isolation between cork- and pulp-morph plants was obtained. No morph-associated genetic structures could be seen within the population in this study. The two morphs coexist within most populations and flower simultaneously within these populations (Emura et al., 2014). No differences in flower morphology were observed between the morphs (Emura et al., 2014). The flower is either self-incompatible or self-compatible with higher abortion for self-pollinated flowers (Howell, 1995; Raju et al., 2019), which enhances outcrossing within populations. In addition, the results of our analyses suggested a low level of inbreeding with nearly-zero F_IS values in all S. taccada populations. These two morphs probably cross freely.

The selective force that maintains fruit dimorphism in S. taccada remains unknown. It is unclear whether multi-generational travel across heterogeneous landscapes is the primary factor in maintaining fruit dimorphism in metapopulation dynamics, or whether other mechanisms, such as heterozygote advantage (Hedrick, 2012) and disruptive selection in contrasting habitats connected by moderate gene flow, are involved. In either case, identifying a genetic system that determines the fruit dimorphism of S. taccada will help to answer these questions. The control of genetic polymorphism by a few Mendelian loci has been reported for fruit-colour dimorphism (Aalders and Hall, 1962; Hall and Aalders, 1963; Whitney, 2005a) and other fruit polymorphisms (Ganders et al., 1977; de Enrech and Mathez, 1998). Furthermore, future studies need to determine the origin of the cork morph, i.e. whether it originated from a mutation in a pulp-morph lineage, or whether it existed in ancestral plant lineages before the speciation of S. taccada and was lost in all other sister species.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: list of samples used in the RAD-seq analysis of this study. Table S2: summary of hierarchical analysis of molecular variance. Figure S1: locations of 17 sampling sites in Ryukyu, Daito and Ogasawara Islands used for RAD-seq on six islands, and neighbour-net analysis for 95 trees from 17 sites. Figure S2: principal component analysis of PC3 and PC4 on RAD-seq data. Figure S3: Gʹ_ST and Jost’s D between populations within islands and for habitat-type combinations between various islands comprising the Ryukyu Islands. Figure S4: directional relative migration based on Jost’s D between populations within islands and for habitat-type combinations between islands of the Ryukyu Islands. Figure S5: Mantel test between genetic and geographical distances with 6408 SNPs of the Ryukyu Islands.

FUNDING

This study was supported by Japan Society for the Promotion of Science (JSPS), KAKENHI Grant Numbers JP12J00375 and JP18K14779 to N.E., JP21H04977 to H.K. and Japan Science and Technology Agency (JST) through the program Core Research for Evolutional Science and Technology (CREST) no. JPMJCR15O1 to H.K., and by the Environment Research and Technology Development Fund, Ministry of the Environment (grant nos. 4-1605 and 4-1902) to Y.I. The study was also supported by the Joint Usage/Research Grant of the Center for Ecological Research, Kyoto University.