Spatial Scales of Genetic Structure in Free-Standing and Strangler Figs (Ficus, Moraceae) Inhabiting Neotropical Forests

Wind-borne pollinating wasps (Agaonidae) can transport fig (Ficus sp., Moraceae) pollen over enormous distances (> 100 km). Because of their extensive breeding areas, Neotropical figs are expected to exhibit weak patterns of genetic structure at local and regional scales. We evaluated genetic structure at the regional to continental scale (Panama, Costa Rica, and Peru) for the free-standing fig species Ficus insipida. Genetic differentiation was detected only at distances > 300 km (Jost´s Dest = 0.68 ± 0.07 & FST = 0.30 ± 0.03 between Mesoamerican and Amazonian sites) and evidence for phylogeographic structure (R ST>>permuted R ST) was only significant in comparisons between Central and South America. Further, we assessed local scale spatial genetic structure (SGS, d ≤ 8 km) in Panama and developed an agent-based model parameterized with data from F. insipida to estimate minimum pollination distances, which determine the contribution of pollen dispersal on SGS. The local scale data for F. insipida was compared to SGS data collected for an additional free-standing fig, F. yoponensis (subgenus Pharmacosycea), and two species of strangler figs, F. citrifolia and F. obtusifolia (subgenus Urostigma) sampled in Panama. All four species displayed significant SGS (mean Sp = 0.014 ± 0.012). Model simulations indicated that most pollination events likely occur at distances > > 1 km, largely ruling out spatially limited pollen dispersal as the determinant of SGS in F. insipida and, by extension, the other fig species. Our results are consistent with the view that Ficus develops fine-scale SGS primarily as a result of localized seed dispersal and/or clumped seedling establishment despite extensive long-distance pollen dispersal. We discuss several ecological and life history factors that could have species- or subgenus-specific impacts on the genetic structure of Neotropical figs.


Introduction
In contrast with high latitude forests, where wind is often the primary vector of pollen-and seed-mediated gene flow in trees, animals are the predominant pollinators and dispersers of tropical trees [1,2]. Biotic interactions thus play a vital role in gene flow and the maintenance of genetic diversity in tropical trees, and genetic based parentage analyses have shown that insects frequently transport pollen over hundreds to thousands of meters (reviewed in [2]). To date, the greatest insect-mediated pollen dispersal distance of 160 km [3] has been recorded for a fig tree (Ficus spp., Moraceae). Fig trees have specialized, often species-specific relationships with minute (1-2 mm) wasps in the family Agaonidae [4,5,6]). Although fig-pollinating wasps are weak flyers and live only 1-2 days, they can be carried long distances by wind currents [3,7]. Figs, particularly in the Neotropics, typically occur at low population densities (e.g. one adult/ha). Further, they flower asynchronously, which increases the already large distances between potential mates [4]. Thus, large pollen dispersal distances in figs are the rule rather than the exception and breeding areas are considered to be huge [7]. Some authors have even suggested that figs might be panmictic at regional spatial scales [8,9].
There is some debate about how patterns of gene dispersal distances, and therefore spatial genetic structure (SGS), are likely to differ among Ficus life forms and subgenera [10,11,12]. For example, functionally dioecious figs of Paleotropical forests are often small in stature, and have clumped distributions and locally dispersing wasp pollinators [10,11,13], traits which could impede long distance gene flow and enhance SGS relative to monoecious species (Table 1). Dev et al. [11] hypothesized that Neotropical species should exhibit weaker SGS than the Indian species in their study, and in dioecious Paleotropical Ficus more generally. Nazareno et al. [12] subsequently performed an SGS analysis of two fig species (subgenus Urostigma) in Brazil and indeed found evidence of significantly lower SGS as measured by the Sp statistic, which measures degree of SGS [14].
Within the Neotropical Ficus, the two subgenera exhibit contrasting life histories that could further influence genetic structure (see Table 1). For example, free-standing figs (subgenus Pharmacosycea) require high light intensities mostly encountered in large light gaps [15], which could lead to clumped pattern of seedling recruitment. In contrast, strangler figs (subgenus Urostigma) establish as hemi-epiphytes on host trees, rendering effective seed dispersal a more spatially random process. At the biogeographic scale, genetic structure has been shown for several fig species [16,17]. A recent study of the free-standing Neotropical species Ficus insipida subs. insipida showed significant cpDNA haplotype differentiation between Mesoamerican and western Amazon samples [18]. Using a molecular clock approach, these authors suggested a mid-Miocene (14.6 Ma, range 5.1 to 26.4 Ma) divergence time of the South and Central American haplotypes. Given the absence of differentiation at the nuclear ITS locus, the authors suggested that the phylogeographic structure was caused primarily by seed dispersal limitation combined with the geographically isolating effects of the northern Andes and Llanos region.
Here we present an analysis of genetic structure in Neotropical figs based on nuclear microsatellite data. Our study spans local, regional, and continental spatial scales for Ficus insipida, and compares local scale spatial genetic structure (SGS) in an additional free-standing fig, F. yoponensis (subgenus Pharmacosycea), and two species of strangler figs, F. citrifolia and F. obtusifolia (subgenus Urostigma) within Panama. Using data from F. insipida, we developed and parameterized an agent-based model to estimate minimum pollen dispersal distances. We then evaluated the extent to which pollen dispersal is expected to drive SGS in our study species. Finally, in order to determine whether our microsatellite data concurs with results from plastid and ITS sequences in showing population differentiation for F. insipida across its distribution range [18], we sampled the widespread species F. insipida (subgenus Pharmacosycea) broadly within the Panama Canal Watershed, and in Costa Rica and Peru and tested for phylogeographic structure at these nuclear loci.

Studied species
There are more than 750 Ficus species (family Moraceae) worldwide [19,20] from monoecious and functionally dioecious subgenera. In the Neotropics, only monoecious species of two of the four acknowledged fig subgenera are present [21]. Molecular and morphological evidence suggest that subgenus Pharmacosycea is phylogenetically distant from Americana [22]. Trees of the subgenus Pharmacosycea are pollinated by fig wasps of the genus Tetrapus (Agaonidae) and grow as free-standing trees that mostly establish in early successional stages due to their initial dependency on high light intensity [15]. Species belonging to the subgenus Urostigma (section Americana) grow mostly as hemi-epiphytes or strangler figs on host trees (hereafter we will refer to Urostigma as "strangler figs"). Americana figs are pollinated by fig wasps of the genus Pegoscapus [19].
In our study area in Panama, 13 of 16 fig species ripen medium to large, green, inconspicuous fruits that are mostly removed at night [23] suggesting that the seeds are mainly dispersed by fruit-eating bats of the Neotropical family of leaf-nosed bats (Phyllostomidae, Chiroptera). We concentrated our study on the most common and widespread species F. insipida. At the local scale we included a second free-standing fig species F. yoponensis, and two species of strangler figs, F. citrifolia and F. obtusifolia, all of which are mainly dispersed by fruit-eating bats. Ficus obtusifolia usually grows as a strangler fig or hemi-epiphyte, and seldom as freestanding tree. Large individuals are most frequently found in mature forests. In contrast, F. citrifolia mostly grows as a small hemi-epiphyte or bush and is frequently found on palm trees in the study area (see Table 1 for more details on the studied species).

Study sites and sampling
The study of local SGS was conducted in tropical moist forest on several islands and peninsulas within the Barro Colorado Nature Monument (BCNM, Fig 1 and S1 Fig) [26]. In contrast, the free-standing fig F. yoponensis (n = 37) is relatively rare along most shorelines and thus sampling was limited to the secondary forest on BCI. The strangler figs F. citrifolia (n = 62) and F. obtusifolia (n = 59) were mostly sampled along the shoreline, as few individuals are known from the core region of the main island (S1 Fig). Because exhaustive census data are restricted to the main island and to the shoreline of the surrounding islands and peninsulas, we were not able to determine the overall fig tree density but calculated Nearest Neighbor (NN) distances based on all censused individuals (Table 1).
Our focal species F. insipida was additionally sampled at several sites within the Panama Canal watershed. First, we collected samples along a 7 km stretch of Pipeline Road (PLR in the following, n = 155), within in Soberanía National Park, for more intensive studies of spatial genetic structure, including a comparison of SGS in adults (dbh ! 10 cm, n = 67) and saplings (< 3 m in height and with soft leaves, n = 73). While mostly situated within 20 m from the dirt road, some individuals were located up to 600 m in the forest interior (S1 Fig). At the regional scale, we sampled F. insipida in the eastern sector of Soberanía National Park, close to the town of Gamboa, and in Parque Nacional San Lorenzo on the Caribbean Coast. All sampling sites  were located close to the Panama Canal and spanned a total distance of about 43 km (Fig 1). To test for genetic differentiation over biogeographic scales, we sampled F. insipida from Costa Rica and northern Amazonian Peru. In Costa Rica, samples were collected in and around Tirimbina Ecological Reserve in Sarapiqui. In Peru, samples were obtained from sites in the Peruvian regions Loreto and San Martín (see S1 Table for geographic coordinates and numbers of samples for each site).
Leaves were collected either directly from the crown or in a few cases from the ground below the canopy if there were no conspecific trees nearby and if the leaves were still green. All leaf samples were immediately transferred to Ziploc bags containing silica gel. Due to the generally low tree densities and the concentration of most individuals along the shorelines or forest roads, we did not follow a particular sampling scheme [14,27], but rather sampled as many individuals as feasible. Genotyping DNA was extracted from ca. 10 mg of desiccated leaf tissue using a modified CTAB protocol [28]. We scored 7 to 10 nuclear microsatellite loci for each species (S2 Table) with markers that were developed from an EST library of F. citrifolia and F. popenoei [29] or were published previously [30,31]. Details on PCR conditions and genotyping are described in Heer et al. [29]. We used one fluorescently labeled primer that could be applied universally with all microsatellites via an attached M13 adaptor sequence [32]. For genotyping, we pooled 2-4 loci using 1 μl of each PCR product in 12 μl formamide and a ROX500 internal standard (Applied Biosystems, Carlsbad, CA, USA). Genotyping was performed in an ABI Prism 3730 Sequencer. Alleles were scored with GeneMarker V1.97 Demo Version (SoftGenetics, State College, PA, USA). Samples were only included in the final analysis when ! 60% of the loci could be scored. We also tested a number of maternally inherited chloroplast markers (cpSSR) in order to obtain additional information on seed dispersal distances but none of them was polymorphic within the Panama Canal watershed (see S1 Methods for details). Microsatellite data are within Supporting Information files (S1 Dataset).

Genetic Data Analysis
For each species and sampling site we calculated total numbers of alleles (N A ), expected heterozygosity (H e ), global estimates of F IS over alleles [33] and tested for deviations from the Hardy-Weinberg equilibrium (HWE) using Genepop 4.1 [34]. Presence of null alleles was determined with MicroChecker 2.2.3 [35]. If null alleles were present, we calculated the mean null allele frequency from the four estimates provided by the software.
For the local scale SGS analyses, we calculated the kinship coefficient F ij proposed by Loiselle et al. [36] that measures genetic similarity between individuals i and j relative to the mean similarity of any random pair of individuals in the sample. This statistic is robust even in the presence of alleles with low frequencies [14]. With the information on spatial and genetic distance between individuals we compared spatial genetic structure among cohorts, populations and species with the Sp statistic, which is a summary statistic introduced by Vekemans and Hardy [14]. This statistic largely circumvents the limitations of other methods (e.g. spatial autocorrelation) that are strongly dependent on sampling scheme and arbitrary choice of distance classes. The Sp statistic is defined as Sp ¼ Àb ð1ÀF 1 Þ with b as the slope of the regression of F ij on ln(d ij ), the natural logarithm of the spatial distance between individuals, and F 1 being the mean pairwise kinship coefficient of the first distance class that includes nearest neighbors (with d ij < 200 m for free-standing figs and d ij < 300 m for strangler figs taking account of the density of sampled individuals). To determine the significance of b we conducted 9999 permutations of the spatial positions of all individuals. Standard errors (SE) of b were approximated by jackknifing. We restricted the range for regression to d ij = 0-3.0 km, with 3.0 km being the maximum distance of the smallest sampling area (for F. yoponensis on BCI). For F. insipida, we calculated Sp for all samples collected in the BCNM and for samples exclusively from BCI to compare them to the results of F. yoponensis. We always refer to the Sp statistics when discussing SGS in this manuscript. When comparing SGS among sampling sites or species we classified differences of Sp as significant if standard errors (± 2 SE) of b did not overlap.
In order to visualize the data for each species, we plotted mean F ij for n = 5-14 distance classes such that the number of pairwise comparisons within each distance interval was approximately constant. This way, the values for the first distance class do not correspond to the F 1 values calculated for the Sp statistics. We obtained 95% confidence intervals (CI) by permuting F ij of all pairs of individuals per distance class 9999 times. Standard errors were assessed by jackknifing for each distance class. All calculations were performed with the program SPAGeDi version 1.3d [37]. In addition, we regressed F ij over ln(d ij ) in R [38] and generated plots with all pairwise comparisons and regression line (S6 Fig).
At larger spatial scales, we used Jost´s D est and F ST to determine whether there is genetic differentiation among sampling sites of F. insipida [39]. Under a model of isolation by distance (IBD) genetic differentiation among populations is expected to increase linearly with distance [40,41]. We used GenAlEx 6.5b3 [42] to calculate pairwise D est among the sampling sites and SPAGeDI to calculate F ST . In order to test whether D est increases significantly with distance we conducted a regression of D est over distance in R version 3.0.1. To this end, we calculated a mean value across all pairwise comparisons for the Panamanian sampling sites that were not genetically differentiated (one data point) and used the pairwise comparisons among Panama, Costa Rica, Eastern and Western Peru (six data points). Further, we performed a STRUCTURE analysis [43] to test whether individuals from different Panamanian sampling sites could be assigned to one or more clusters based on multi-locus genotype data. We selected the setting with correlated frequencies as it improves estimates of admixture and left all other parameters at their default values [44]. We used 250,000 MCMC repetitions after a burn-in period of 100,000. To determine the most likely number of clusters K, we tested K values from 1-10 and carried out 10 runs for each K. Afterwards, we followed the procedure suggested by Evanno et al. [45] to select the optimal K. With the same settings, we analyzed all samples from Panama, Costa Rica and Peru with 10 repetitive runs for each K = 1-7.
Phylogeographic structure can be inferred from the microsatellite DNA data set if allele sizes within a populations or geographic region are more similar than in the overall sample [14]. This was tested by comparing the allele-size based differentiation statistic R ST with its value after permuting allele sizes within loci ("permuted R ST "-similar to F ST ) using SPAGeDI with 10,000 permutations. A significant one-sided test establishes the alternative hypothesis of R ST > "permuted R ST ". Sequential Bonferroni corrections [46] were applied to significance levels of multiple tests.

Pollen dispersal model
In order to explore the possible role of pollen dispersal limitation in shaping fine scale genetic structure, we designed a computational model using the R programming language [38] that allowed us to estimate the number and distance of possible pollen donors for F. insipida trees in the BCNM. Based on the available census data we included all F. insipida trees on BCI that were still alive in 2007 and all trees mapped along the shoreline of the islands and peninsulas (n = 342 trees, see S3 Fig).
In the model, each tree undergoes a reproductive cycle that starts at a randomly chosen day of the year. The cycle consists of a receptive phase of six days, a ripening phase of 28 days and a phase of six days during which fig wasps are released and search for fig trees in the receptive phase. The next reproductive cycle starts after a randomly varying period of 228 to 326 days. The model was run for two consecutive cycles for each tree and was repeated ten times. In addition, we repeated the runs with a receptive and release phase of five and seven days, respectively. Trees were classified as potential pollen donors when their release phase overlapped with the receptive phase of another tree, independently of the geographic distance. For each tree we calculated the number of potential pollinators and the Euclidian distance between pollen donor and receptor. Based on these results we estimated the mean number of pollen donors for a number of predefined distance classes. The phenological data for the model was derived from Milton et al. [38,47], Milton [48] and Morrison [24]. We used the R packages "calibrate", "maptools", "plotrix" and "sp" [49,50,51,52]. More detailed information on the model is presented as supplementary material (S3 Fig and S2 Methods).

Results
For each Ficus species, we were able to score 7-10 markers with a mean allele number (per locus) of 7.0-10.9 and total allele numbers of 56-109 alleles per species (S2 Table). The number of loci differs between individuals because some of the SSRs did not amplify across all species, had a peak pattern that we could not score reliably, or an excess of null alleles. Null alleles were present in two and three markers for F. citrifolia and F. obtusifolia, respectively (S2 Table) and markers with null allele frequencies > 0.2 were omitted from the data analysis.
Each sampled individual had a unique multi-locus genotype. All of the species exhibited high levels of heterozygosity (mean H O = 0.58-0.79) and low inbreeding coefficients (mean F IS = -0.056-0.054), except for F. obtusifolia, which exhibited a relatively high F IS of 0.146. Ficus obtusifolia had an excess of homozygosity in 4 (of 7) loci. Because F IS is not uniformly high across all loci, we interpret deviations from HWE in the 4 loci as the result of allelic drop-out rather than inbreeding, which would affect all loci. Allele dropout is commonly encountered in microsatellite markers that were not developed for the focal species.
For the other three fig species we found significant SGS in the BCNM as well. The Sp statistics revealed strongest SGS for the strangler fig F. obtusifolia (Sp = 0.031 ± 0.023, p < 0.001) compared to F. citrifolia (Sp = 0.013 ± 0.008, p < 0.001), F. yoponensis (Sp = 0.008 ± 0.012, p = 0.032) and F. insipida. When comparing Sp based on the overlap of 2SE, only F. obtusifolia and F. insipida showed significant differences with much stronger SGS in F. obtusifolia. When we restricted analysis for F. insipida to trees growing on BCI to better compare it to F. yoponensis, we found no significant SGS (Sp = 0.001 ± 0.003, ns). To exclude the possible of null alleles affecting these results, we repeated the analysis by omitting markers with null alleles (S2 Table). This did not alter the results significantly.
The autocorrelation analysis largely supported the finding of significant SGS and indicated that pairwise kinship was higher than expected under conditions of random pollen and seed dispersal (reflected by the 95% CI) for distance classes up to 1 km in F. insipida (Fig 2). For both strangler figs the pairwise kinship coefficient of the first distance class (mean pairwise distance < 500 m) was significant. In F. yoponensis, the autocorrelations was not significant, but F ij decreases with increasing distance as well except for the first distance class.

Pollen dispersal model
When the receptive and release phase in our model were set to six days, we found that 48 out of 342 F. insipida trees had no potential mate within a radius of 500 m (mean number of mating partners < 500 m: 1.7 ± 1.0). On average, 4.0 ± 2.1 potential pollen donors could be encountered within a radius of 1 km in our simulations (Fig 3). In total, each receptive tree could potentially receive pollen from 25.8 ± 2.3 trees in our simulation. Mean distance between potential mating partners was 3.0 ± 0.5 km. If the receptive and release phase lasted for five or seven days, mean number of pollen donors decreased by 13% and increased by 14%, respectively (S4 Fig). Genetic differentiation of F. insipida at regional and biogeographic spatial scales At the regional scale, genetic differentiation among sampling sites located along the Panama Canal was low (D est = 0.009 ± 0.006, F ST = 0.006 ± 0.004) and did not increase with distance. The STRUCTURE analysis detected only one cluster in this entire area. In contrast, at the biogeographic scale (Costa Rica to Peru) we found a strong linear correlation of D est over linear Euclidian distance (adjusted R 2 = 0.962, p < 0.001, Fig 4, corresponding F ST in Table 3). The most likely number of clusters of the corresponding STRUCTURE analysis was K = 2 separating the samples from Peru and Central America. However, the Evanno method has a strong bias to identify K = 2 as the most likely number due to the low variance of K = 1 [53]. Thus, we also assessed the second most likely number of clusters, which was K = 5. In that case, the The phylogeographic analysis based on microsatellite size variation showed significant difference between R ST and permuted R ST only in comparisons between Central America (Costa Rica or Panama) and Peru (P < 0.01) ( Table 3). There was no significant contribution of allele size mutation to population differentiation within Central America or within Peru.

Discussion
We collected nuclear microsatellite data to estimate genetic structure at different spatial scales in the common free-standing fig F. insipida. Significant genetic differentiation among Mesoamerican and Amazonian sites indicates that pollen dispersal is restricted over biogeographic scales. At the local scale we detected weak but significant SGS in F. insipida as well as in three other sympatric fig species from both New World subgenera Pharmacosycea (free-standing figs) and Urostigma section Americana (strangler figs). Our model results were consistent with previous genetic studies [7] that showed that pollen dispersal routinely exceeds the area within which we estimated SGS. Thus, our results suggest that local scale SGS results from relatively limited seed dispersal in Neotropical figs. In the following, we explore key processes that influence the formation of SGS in Neotropical figs and relate SGS at the local scale to patterns of geographic differentiation.
Figs exhibit the highest pollination distances known in insect-pollinated plants [3]. In Panama, fig wasps often need to travel hundreds to thousands of meters to encounter a receptive conspecific host plant. Individual figs routinely receive pollen from multiple genetically distinct pollen sources that can be dozens of kilometers apart [7]. Asynchronous mating further reduces the frequency of nearest neighbor pollination. In accordance with this, we observed high H O and low F IS values in our study which indicate high levels of random mating. Our model results using data from F. insipida are consistent with Nason et al.´s [7] findings with stranglers: only very few pollen donors are routinely located among the nearest neighbors of a receptive fig tree: For 44% of the modeled trees 0-1 potential pollen donors were located within a radius of 500 m, while a much larger number of pollen donors was located beyond 1 km (84.5% of all potential fathers in our model), far exceeding the spatial scale of the fine scale SGS in our study. Our model likely underestimates the number of pollen donors at larger distances because it is restricted to the area of BCNM. As we do not integrate wind speed and direction, we probably overestimate the number of pollen donors at short distances. Nevertheless, our estimate demonstrates that close neighbor mating plays a minor role in pollen mediated gene flow and thus, is unlikely to generate local scale SGS. Our pollination model is likely to apply While long distance pollen dispersal is a common feature in Neotropical figs, SGS patterns may vary predictably among Neotropical Ficus species for other reasons, including life history differences associated with fruit size, and differences in habitat requirements for seedling establishment, both of which we explore in the following paragraphs. As any measure of SGS will only provide a temporally and spatially restricted snapshot, we also discuss the potential influences of colonization history and fluctuations in population density on SGS of the studied fig populations.  Table 3. Pairwise measures of genetic differentiation between F. insipida populations. F ST (above diagonal) and R ST (below diagonal). If R ST > permuted R ST , it is indicative for the contribution of mutations to population differentiation (for p < 0.05 after correction for multiple testing indicated by bold numbers).

Panama
Costa

Regeneration ecology, colonization history and SGS
While fig species mostly occur at low densities, free-standing figs might occur-temporally and/ or locally restricted-at higher densities, especially in former disturbed areas and light gaps [26]. The intensity of SGS can depend on the degree to which large light gaps are colonized by seeds from multiple sources [54]. The relatively low SGS observed in F. insipida in our study, coupled to the fact that SGS is not significant even in saplings, suggests that light gaps are being seeded by multiple maternal sources. As free-standing figs are most dominant in early successional stages of forests, colonization history has to be considered as well. The northwestern part of BCI experienced a long history of agricultural use and the forest only began to regenerate about 150 years ago. Almost all F. insipida and F. yoponensis trees on BCI are restricted to this secondary forest [26]. Most free-standing Ficus trees established in early successional stages and appear to be the same age on BCI. As the forest matured, further recruitment was impeded by low light conditions under the canopy cover and, currently, sapling Ficus trees are extremely rare in BCÍ s closed canopy forest [26,55]. almost simultaneously from multiple seed sources that were likely growing at the edge of the agricultural fields. Initial patterns of SGS would then be weakened by demographic thinning [56]. In contrast, seeds of strangler figs mostly establish in the canopies of other host trees, and they should encounter evenly scattered germination sites within intact forest. A relatively random distribution of fig trees at low density, such as is observed in the stranglers, should increase SGS because seed shadow overlap is expected to be relatively low [14]. We expect that future studies will find similar differences in the intensity of SGS between strangler and free-standing figs due to their fundamentally different requirements for seedling establishment.

Influence of frugivore size and dispersal on SGS
While howler monkeys and frugivorous birds visit fruiting figs trees during daytime [57,58], in BCNM, fruit-eating bats of the family Phyllostomidae disperse more than 80% of the seeds of the studied fig species [23,59]. Bat-mediated seed dispersal distances are unknown due to the observational challenges posed by the high mobility and nocturnal lifestyle of bats. Although fruit-eating phyllostomid bats are generally regarded as potential long distance seed dispersers [60,61], several factors may limit their effective dispersal distances. After picking a fruit, these bats usually fly to temporary feeding roosts located within 50-200 m of the fruiting tree [62]. This behavior appears to be a response to predators that also congregate at ripe fig trees [62]. Ingested seeds are not likely to travel far as bats have short gut passage time for seeds (< 30 min) [63] which leaves germination rate of seeds largely unaffected [60]. Based on this, we expect that most fig seeds will be dropped below or near these temporary feeding roosts close to the maternal fig tree. Larger fig fruits are generally consumed by bat species with larger body size [64,65,66] that are more likely to disperse seeds over longer distances due to their higher mobility. According to that rationale, F. insipida and F. obtusifolia should be dispersed over larger distances and exhibit lower SGS (see Table 1 for fruit sizes). However, we found the lowest SGS in F. insipida and the highest in F. obtusifolia (Table 2). We conclude that measurable differences in seed dispersal distances are either not significant among fig species or that their effects on local SGS are overwhelmed by other factors. While encountering seeds dispersed by fruit-eating bats is not an easy task, genetic parentage analyses of fig seeds deposited under feeding roosts could provide the first empirical data on seed dispersal distance by fruit-eating bats.
Our results of low but significant SGS are also consistent with those of Nazareno et al. [12] who studied SGS in populations of F. citrifolia and F. eximia. It is also interesting to note that despite the extraordinary pollen dispersal distances documented in figs, other tropical species that are pollinated by insects and dispersed by animals show similar levels of SGS based on the Sp statistics as the figs from our study (compare [2,67]). We suggest that the observed SGS values in our study are primarily shaped by differences in tree density, and the interplay between regeneration strategies, colonization history, and average seed dispersal distances. However, while SGS measured with the Sp statistics is generally considered robust across different sampling schemes [14], we recognize that SGS can also be influenced by population dynamics and idiosyncratic population histories [68] that may obscure more deterministic differences in SGS caused by life history.
Regional and phylogeographic genetic structure of F. insipida At the regional scale within the Panama Canal watershed we detected no population substructure for F. insipida despite the spatial limitations of seed dispersal that likely generated SGS at the local scale. This is consistent with prior estimates of breeding area based on pollen dispersal studies for strangler figs [7]. At the biogeographic scale, we found that populations were genetically distinct and genetic differentiation increased with distance as predicted under conditions of isolation by distance (IBD) and/or isolation by geographic barriers (IBB) [69] ( Table 3). The geographic barriers separating our population samples from Costa Rica to Peru include the northern Andes, the dry Llanos region of northern South America, and various xeric and montane barriers between the Panamanian and Costa Rican sites. While the summary statistics F ST , R ST and D est only permit a post hoc interpretation, the phylogeographic structure of F. insipida detected between our Central and South American samples is consistent with the previously published cpDNA phylogeography [18] of the same species. Because the phylogeographic structure was stronger in the maternally transmitted cpDNA than in ITS, Honorio Coronado and colleagues [18] suggested that the geographic barriers associated with the northern Andes posed a greater barrier for seed than pollen dispersal. This pattern may have other explanations. The haploid, uniparentally transmitted chloroplast genome has a lower N e (1/4 the N e of nuclear markers) and genetic drift therefore plays a much stronger role in cpDNA markers. Chloroplast capture from congeneric species is also possible, and has been demonstrated in Ficus and other tree taxa [70,71]. The absence of ITS variation between Panama and Amazonian Ficus [18] is among the lowest levels of divergence found in lowland trees sampled on either side of the Andes [72,73,74,75] and is most similar to species that have relatively recently dispersed around the Andes [76]. This suggests fig seed dispersal and/or colonization ability (especially in drier habitats) has allowed F. insipida, and perhaps other fig species, to overcome the geographic barriers that have caused vicariance in many other tree species.

Conclusions
Neotropical fig species exhibit local spatial genetic structure despite frequent long pollen dispersal distances. The most likely underlying cause is limited seed dispersal, with fruit-eating Phyllostomid bat species being the most likely seed dispersers of the studied Ficus species in BCNM. In addition, different regeneration strategies of free-standing vs. strangler figs may cause differences in the degree of SGS. Nevertheless, gene flow over large distances is maintained via pollen dispersal and our study of meso-scale genetic structure in F. insipida confirmed that breeding populations cover large areas such as the entire watershed of the Panama Canal. Yet, at biogeographic spatial scales, genetic differentiation among regions becomes apparent, and is likely caused by a combination of IBD and IBB. Further studies are needed to test the emerging pattern of greater gene flow in monoecious than in dioecious species (this study and [11,12,77]). Importantly, species-specific traits (e.g. requirements for seedling recruitment) as well as the behavior of both pollen and seed dispersal agents have to be considered when predicting the spatial extent of gene flow in figs and other tree taxa.
Supporting Information S1 Dataset. SSR raw data for all individuals.  Table. Sampling sites of F. inspida in Panama, Costa Rica and Peru with the number of samples per plot (n) and the spatial coordinates of the locality. (DOCX) S2 Table. Characterization of the microsatellite markers for Ficus samples collected in the Barro Colorado Nature Monument. For each marker, we provide the percentage of missing data, the number of alleles (N a ), the expected heterozygosity (H e ) and the inbreeding coefficient (F IS ). (DOCX)