Introduction

Population genetic structure and the mating system are closely interrelated aspects of a species' ecological genetics. The amount of genetic variation in a species is affected by various ecological phenomena (Loveless & Hamrick, 1984; Hamrick & Godt, 1990), as well as by the mating system (Holtsford & Ellstrand, 1989). Furthermore, the way in which genetic variation is distributed among populations is closely tied to gene flow, both by pollen and seed movement (Levin & Kerster, 1974; Loveless & Hamrick, 1984). Because genetic structure is generated by the dispersal, survival and reproduction of individuals in a population over time, an appreciation of the evolutionary genetics of a species must take into account not only strictly genetic events, such as fertilization and the mating system, but also ecological and life history traits as they influence the evolutionary and ecological dynamics of the population.

Recent studies have examined the genetic make-up of tropical tree populations (reviewed in Loveless, 1992; Hall et al., 1994; Boshier et al., 1995a,b; Gibson & Wheelwright, 1995). In general, tropical plant species have levels of allozyme variability equal to those of their temperate counterparts (Hamrick & Godt, 1990; Loveless, 1992). Analyses of genetic differentiation at a local scale on Barro Colorado Island, Panama (hereafter BCI) show that populations separated by one to several kilometres have relatively little genetic differentiation, implying high levels of gene flow. Such patterns of population differentiation can be correlated with breeding biology, individual distribution, and pollen and seed dispersal (Hamrick & Loveless, 1989; Hamrick & Murawski, 1990). Evidence suggests that gene flow in tropical forests is often substantial, tying together populations of even relatively rare tropical trees into large demes or breeding units (Hamrick & Murawski, 1990; Chase et al., 1996; Nason et al., 1996; Stacy et al., 1996).

Information on the reproductive biology and mating systems of tropical trees provides an independent corroboration of high levels of gene flow. Initial studies of tropical plant mating systems demonstrated that dioecy and other sexual systems which enforce outcrossing were more common in tropical than in temperate species (Bawa & Opler, 1975). High levels of outcrossing have been confirmed by genetic analyses of tropical tree mating systems (reviewed in Loveless, 1992). Fewer data are available for seed dispersal; however, the preponderance of animal-mediated seed dispersal suggests that dispersal should have a high variance, permitting long-distance seed movement (Hamrick & Loveless, 1986; Hamrick et al., 1993).

In this paper, local patterns of genetic organization in populations of Tachigali versicolor (Fabaceae: Papilionoideae) on BCI, Panama, are described and the mating system is analysed using allozyme markers. These data allow calculation of indirect and direct estimates of gene flow in a natural population of this tropical tree.

Materials and methods

The study species

Tachigali versicolor (Fabaceae: Papilionoideae) is found from Costa Rica to western Colombia (Croat, 1978). It is a relatively common canopy tree in secondary and primary forests on BCI (Foster, 1977; Kitajima & Augspurger, 1989). Tachigali versicolor is monocarpic: adults die after flowering (Foster, 1977). Other, South American species of Tachigali are also apparently monocarpic (Gentry, 1993). Age at flowering is unknown, but growth rates vary widely, depending on the light environment of seedlings and saplings (Augspurger, 1984; Kitajima & Augspurger, 1989), and trees may flower in a wide range of size classes (Foster, pers. obs.; Loveless, pers. obs.). Flowering is pulsed at intervals of 4 to 6 years (Foster, 1977, pers. obs.). On BCI, flowering occurred in 1970, 1974, 1978, 1983, 1984, 1989 and 1994 (Foster, 1977, pers. obs.). Although individuals sometimes flower (and die) out of synchrony, in most off-years there is no flowering.

Flowering occurs from January to July. Individual trees flower for 6–12 weeks. The mature fruit is a large, wind-dispersed, single-seeded samara. Most fruits fall <100 m from the parent (Kitajima & Augspurger, 1989). Fruits suffer both pre- (bruchid beetles and parrots) and postdispersal (rodents, peccaries and fungal infection) mortality (Kitajima & Augspurger, 1989).

Seeds lack dormancy, and germination occurs at the start of the rainy season, in late April and early May. Germination of undamaged seeds is high, and seedling shadows may be dense beneath adults. Seedlings are shade-tolerant, persisting for many years in the understorey. However, seedlings in light-gaps grow rapidly, forming a cohort that is heterogeneous in size (Kitajima & Augspurger, 1989). To reach canopy size, several gap-mediated growth spurts are probably necessary.

Field collections

Genetic diversity on BCI was sampled at two spatial scales. Four collection areas located about 200 m apart were chosen within a 50 ha mapped forest plot established in 1980 (the Forest Dynamics Plot, or FDP; Hubbell & Foster, 1983). Within each area, 72 individuals were sampled. Two additional off-plot sites, each located at least 1 km from the FDP, were also identified. Fifty individuals were sampled in these sites. In all locations, individuals of different size classes were sought over an area of 6–8 ha. Leaf collections were kept cool in the field, vacuum-dried for at least 48 h, packaged with dessicant, and returned to the U.S.A. Dried samples were kept at −70°C until used for electrophoresis.

Seed collections were made in February and March 1985, when mature fruits from the 1984 flowering episode (Fig. 1) were being dispersed. Seeds were collected directly beneath 25 adult trees. In those cases where seed shadows of different adults might have overlapped, collections were made in a direction away from other seed sources. Seeds were germinated in a growing house on BCI, and leaf tissue was vacuum-dried.

Fig. 1
figure 1

Map showing the location of flowering individuals and sampled trees (squares) of Tachigali versicolor on Barro Colorado Island, Panama, in 1984. The forest dynamics plot is shown by the box in the centre of the island. The five trees indicated by the star are those used in the direct estimate of pollen movement.

Full size image

Laboratory procedures

Horizontal starch gel electrophoresis was performed on the samples. The dried leaves were crushed under liquid nitrogen, mixed with an extraction buffer and adsorbed onto filter paper wicks. A total of 31 allozyme loci in 17 enzyme systems were resolved in the population samples. Isozymes and buffers used in this study are given in Table 1, and are modified from Soltis et al. (1983).

Table 1 Enzyme systems resolved in electrophoretic analyses of Tachigali versicolor. Buffer systems are taken from Soltis et al. (1983)
Full size table

Four loci (Idh, Dia, Fe1 and Fe3) were used to assess mating systems. In all cases, progeny banding patterns were consistent with a simple Mendelian interpretation of the allozyme loci.

Statistical analysis

Standard statistics of genetic diversity were calculated for individual population samples and for BCI as a whole by pooling over all populations. For each population, we determined the proportion of polymorphic loci (P), the mean number of alleles per locus (A) and per polymorphic locus (AP), and the mean genetic diversity (He=1−Σ p2i where pi is the frequency of the ith allele; Nei, 1973). Statistics were calculated for each locus and then averaged over all loci. Observed and expected heterozygosities for each polymorphic locus were compared by calculating Wright's fixation index (where F=1−[ Ho/He]) to determine deviations from random-mating expectations. Deviations of F from zero were tested using χ2 (Li & Horovitz, 1953).

Differences in allele frequencies among populations were examined using a χ2 analysis (Workman & Niswander, 1970). For each polymorphic locus, statistics of gene diversity were calculated (Nei, 1973, 1977), including total gene diversity (HT), mean gene diversity within populations (HS), gene diversity among populations (DST) and GST, the proportion of the total gene diversity found among populations (GST=DST/HT). Overall means were calculated by averaging over all polymorphic loci. Nei's coefficients of genetic distance (Nei, 1972) were also calculated between each pair of population samples.

The mating system was analysed according to the methods of Ritland & Jain (1981), using a program from Ritland (1990). The procedure estimates multilocus outcrossing (tm), mean single-locus outcrossing (ts) and maternal genotypes, and calculates pollen allele frequencies and individual outcrossing rates for each maternal family. Variances for individual family outcrossing rates were based on 100 bootstraps; variances for population outcrossing estimates were based on 500 bootstraps. Differences in pollen allele frequencies among families were tested using a heterogeneity χ2 (Workman & Niswander, 1970).

Results

Genetic variation within and among populations

Of 31 loci surveyed for T. versicolor on BCI, 13 (41.9%) were polymorphic in at least one population. On average, however, only 29.6% of the loci were variable in any given population (Table 2). Despite the relative proximity of the populations on BCI, there were six private alleles (Slatkin, 1985) present. Populations were similar in levels of polymorphism, number of alleles per locus and per polymorphic locus, and observed and expected heterozygosities (Table 2). Only three of the 13 polymorphic loci had more than two alleles (BCI mean AP=2.23), and these alleles were rare where they were found (frequencies of <0.02). Mean expected heterozygosity was 0.073, and the pooled He-value for all BCI sites was only slightly higher (0.080). Observed heterozygosities were consistently smaller than expected, indicating a slight deficiency of heterozygotes (Table 2). When allele frequencies at each locus and in each population were tested against Hardy–Weinberg expectations, 11 of 55 fixation indices (20%) were significantly different from zero. In every instance, the deviation from zero was positive, indicating heterozygote deficiencies at these loci.

Table 2 Estimates of genetic variation for six populations of Tachigali versicolor on Barro Colorado Island (BCI), Panama (see footnote for definition of abbreviations)
Full size table

Allele frequencies at 10 of the 13 polymorphic loci differed significantly among the six populations (Table 3). However, DST values overall were relatively low, resulting in small GST values (mean= 0.069). Values for GST are additive and can be subdivided hierarchically within BCI to compare the four populations within the FDP with the two samples from outside the FDP. Hierarchical GST statistics (Table 3) indicate, however, that population differentiation does not conform to expectations for these spatial patterns. The four neighbouring populations within the FDP have more than twice the among-population differentiation than they have with populations at the level of BCI. Calculations of genetic identity among the six populations also reflect this; FDP1 is relatively different from all the other locations (mean I=0.982 vs. I=0.995 for all other pairwise comparisons).

Table 3 Heterogeneity χ2-values and Nei's statistics of genetic diversity for 13 polymorphic loci in Tachigali versicolor (Nei, 1973)
Full size table

Mating system analysis

Analyses of progeny arrays from 25 adult trees that flowered in 1984 indicate that T. versicolor is completely outcrossed. The multilocus outcrossing rate was tm=0.998 and the mean single-locus outcrossing rate was ts=1.001 (Table 4). For two of the loci (Fe1 and Fe3), the frequency of the second allele was very low, and the iteration algorithm did not permit variance estimates. Such individual ts rates are probably not very dependable. Bootstrap variances were calculated for Idh and Dia, and indicate complete outcrossing (Table 4). If single-locus outcrossing rates are lower than multilocus rates, this difference can be attributed to biparental inbreeding (Shaw et al., 1980; Ritland & Jain, 1981). In T. versicolor, therefore, there is no evidence for biparental inbreeding.

Table 4 Estimates of population outcrossing rates in Tachigali versicolor. Standard errors are shown in parentheses
Full size table

Chi-squared values were calculated to test the fit of progeny arrays to the mixed-mating model. Both Idh and Dia showed slight deviations from the mixed-mating model (χ21=5.79, 0.01<P<0.025 and χ21=3.93, 0.025<P<0.05, respectively), indicating an excess of heterozygous progeny from homozygous parents. Progeny arrays at Fe1 and Fe3 were not significantly different from those predicted by the mixed-mating model.

Adult genotypes for these 25 trees were inferred from their progeny arrays, permitting separate estimations of allele frequencies in the ovule and pollen pools. Our population samples for BCI, using trees of all size classes, provided a measure of allele frequencies in the population as a whole. Comparisons between ovule and pollen allele frequencies were not significant. However, significant differences occurred between allele frequencies in the reproductive gene pool (pollen and ovules) and the vegetative population in five of eight comparisons (Table 5). Pollen and allele frequencies at all four loci also differed significantly among the 25 adults (Table 6).

Table 5 Comparison of allele frequencies among different source groups in Tachigali versicolor on Barro Colorado Island (BCI). ‘Adults’ refers to the 25 trees sampled from the 1984 flowering cohort. ‘Pollen’ is the frequency of alleles in the total pollen pool sampled (1017 gametes). ‘BCI’ represents the overall allele frequencies generated by pooling the six populations shown in Table 2. Standard errors are shown in parentheses for those values which were estimated using maximum likelihood methods. χ2-values test for differences among different sources using a contingency χ2
Full size table
Table 6 Range in pollen allele frequencies and in multilocus outcrossing rates among 25 progeny families of Tachigali versicolor. Allele frequencies of the rare allele at each locus are shown
Full size table

Based on the multilocus outcrossing estimate, the equilibrium inbreeding coefficient [Feq=(1−t)/(1+t); Wright, 1965] for this population is Feq=0.001, reflecting the absence of inbreeding among the progeny arrays. The mean fixation index in the sample of adults flowering in 1984 [calculated as F=1−(Ho/He) and averaged over the four loci] was negative (F=−0.106), indicating a slight excess of heterozygotes. A value of the inbreeding coefficient for the population samples was also obtained from FIS, which represents within-population inbreeding. The average FIS for the 13 polymorphic loci in the six multi-aged population samples was positive (FIS=0.187), and significantly different from the fixation index in the flowering adult sample (Student's t-test, t15=−2.434, P<0.05). There is a deficiency of heterozygotes in the population samples, and a moderate excess of heterozygotes among the flowering adults.

Measures of gene flow

Two indirect methods were used to estimate Nm, the number of migrants per generation from population structure. Wright's (1951) equation based on FST (GST) gave a value for Nm=3.32. The Slatkin method (Slatkin, 1985) gave Nm=6.65. These estimates are rather different in magnitude, but both suggest relatively large amounts of migration among T. versicolor populations on BCI.

A more direct method of measuring gene flow by pollen was possible because of the spatial distribution of genotypes in the 1984 flowering adults. Five flowering trees located in a cluster were homozygous for the common allele at the Idh locus. However, 10 of the 240 progeny were heterozygous for the rare allele, which must have originated from beyond these five trees. The value of m, the rate of gene flow, can be calculated from the equation:

where q0 is the frequency of the rare allele in the five adults (0.00), q1 is the frequency of the allele in their progeny (0.021), and Q is the frequency of the allele in the flowering population as a whole (0.200). This gives a value of m=0.105. Thus, 21% of the effective pollen migrated from beyond this cluster of five trees. In 1984, the nearest flowering tree to this group was ≲500 m distant (Fig. 1)

Discussion

Genetic variation within and among populations

Compared to 15 other common tree species on BCI, T. versicolor has low levels of genetic variation (Tables 2 and 7). Values of P in T. versicolor are lower than those for 16 rarer species on BCI (Hamrick & Murawski, 1991) and those found in a broad survey of tropical woody plants (Loveless, 1992). Mean heterozygosity is also lower in T. versicolor than in other tropical tree species (Table 7) Heterozygosity in this species is similar to values found in some of the rare species studied on BCI but is considerably below the mean for the rare species (Hamrick & Murawski, 1991).

Table 7 A comparison of estimates of genetic parameters in Tachigali versicolor and other tropical trees. Because samples for this species came from a limited geographical area (Barro Colorado Island, BCI), we compare the data to within-population values from other studies. Standard errors are shown in parentheses
Full size table

In fact, although T. versicolor is ecologically common on BCI (61.9 individuals≲1.0 cm d.b.h. ha−1; Condit et al., 1996), its unique life history makes it more similar, genetically, to rare tropical tree species. Because only a fraction of the population reproduces in any flowering year, the number of reproductive individuals is much less than the actual numerical population. Flowering cohorts on BCI in the last six flowering cycles ranged from 12 trees (in 1983, out of synchrony) to around 75 (in 1974). In effect, T. versicolor has reproductive densities of 0.05 individuals ha−1 or less, with seedling cohorts of different years being generated in genetic isolation from one another. Such small flowering populations are likely to experience genetic drift relative to the population as a whole, with consequent loss of genetic variation. The loss of genetic diversity would be enhanced if whole cohorts subsequently failed to reproduce. This may explain the relatively low level of genetic variation in T. versicolor on BCI. Because possibilities for recombination and the spread of alleles between cohorts are limited, with long lag times, genetic diversity can be lost to future generations by drift in every reproductive episode. Past evolutionary events (such as bottlenecks or speciation events) could also contribute to low levels of genetic diversity in this species.

The six local populations on BCI all had a deficiency of heterozygotes, which is usually explained by inbreeding, but T. versicolor is completely outcrossed. Because each of the population samples included individuals of different sizes and ages, and because flowering behaviour in T. versicolor strongly structures gene flow by pollen, population samples lumped together individuals from different, genetically isolated cohorts (i.e. a temporal Wahlund effect). Even within a single reproductive cohort, the seed are grouped around the maternal tree, which should to lead to fine-scale genetic structure in the seedlings (Hamrick & Loveless, 1986; Hamrick et al., 1993). Where such existing population subdivision is not recognized, the result is a Wahlund effect: a deficiency of heterozygous genotypes relative to the expectations of the composite population.

GST calculations (Table 3) indicated only small amounts of differentiation among these six populations. This differentiation did not, however, reflect spatial separation of the populations, suggesting that some process other than simple isolation by distance is acting to structure local populations.

Tachigali versicolor is similar to other tropical trees in having limited population differentiation (Table 7). In fact, it has more differentiation within the four sites on the 50-ha FDP than most of the other species studied (Hamrick & Loveless, 1989). This may reflect limited seed dispersal and seedling clumping in T. versicolor (Foster, 1977; Kitajima & Augspurger, 1989; Hamrick et al., 1993).

Mating systems

Tachigali versicolor is completely outcrossed. In this respect, the species resembles most tropical tree species for which mating systems have been described. Of 13 tropical woody angiosperm species (including T. versicolor) summarized by Loveless (1992), only two [Cavanillesia platanifolia (Murawski et al., 1990), and Ceiba pentandra (Murawski & Hamrick, 1992)] had mixed-mating systems. Even among species with mass canopy flowering, where geitonogamous pollen transfer must occur frequently, outcrossing is the rule.

There was also no evidence for biparental inbreeding in T. versicolor. Although some flowering trees are spatially close to one another, their progeny do not show apparent inbreeding that would indicate genetic relatedness among nearby adults. This is consistent with the assumption that individuals flowering in a given year are not from a common seedling cohort, but represent a genetically unrelated mixture of adults that happen to flower at a particular time.

There were no significant differences between ovule and pollen allele frequencies at any of the four loci (Table 5), suggesting that flowering trees contributed more or less equally to reproduction through pollen and ovules. However, there were significant differences between allele frequencies in the reproductive gene pool (pollen and ovules) and the vegetative population on BCI in five of eight comparisons. If the reproductive gene pool represented only the 25 sampled maternal trees, such differences could result from sampling error. However, given the likelihood of long-distance pollen movement, the pollen pool probably sampled a large fraction of the 61 individuals flowering in 1984 (Fig. 1). This suggests that there are real differences between trees flowering in 1984 and the overall population, a difference that reflects the genetic bottleneck through which this species passes during each flowering episode. Because local populations of T. versicolor are slightly different in allele frequencies, as GST statistics indicate, adults flowering in any year (Fig. 1) would combine individuals from many subpopulations. There are two levels of differentiation that could be involved: spatial differentiation among different parts of the island, and temporal differentiation among individuals derived from different cohorts. From this perspective, the reproductive and the vegetative samples represent different genetic phases of the population of T. versicolor on BCI. The 1984 flowering adults represent the allele frequencies in that reproductive population as it passes through a temporal bottleneck that will constrain the genetic make-up of the 1984 seed cohort. The vegetative samples, on the other hand, represent individuals of various sizes and ages, as seed cohorts from different flowering years are dispersed and grow to adult size in different parts of the island.

Pollen allele frequencies showed highly significant differences in the pollen pools sampled by different mothers (Table 6). Homogeneity of the pollen pool is an assumption of the mixed-mating model (Clegg, 1980; Ritland & Jain, 1981), and the violation of this assumption in T. versicolor may explain why, for Idh and Dia, there was an excess of heterozygous progeny among homozygous mothers. Where they have been looked for, differences in pollen allele frequencies among maternal individuals have been found in many plant populations. In tropical trees, significant variations in the pollen pool were described for Cavanillesia platanifolia (Murawski et al., 1990). Stacy et al. (1996) showed that local densities of flowering individuals significantly affect breeding structure, with clumped individuals being more likely to mate with their nearest neighbours. Although such spatial patterns would be a logical explanation for local variations in the pollen pool, little is known about pollinator movement among the canopies of mass-flowering tropical trees. As we have genotypes for only 41% of the trees that flowered in 1984, we cannot completely describe local pollen pools. In addition, there is phenological variation among T. versicolor trees flowering in the same year (M. D. Loveless, pers. obs.), so trees that are clustered could experience different pollen pools as nearby trees initiated or ended their flowering.

The inbreeding coefficient (F) among different groups of T. versicolor showed considerable variation. The F among flowering adults in 1984 was negative, indicating an excess of heterozygotes. The equilibrium Feq calculated from the mating system estimate of t was essentially zero, indicating panmixia among progeny produced in 1984. However, the mean FIS value calculated among the six population samples was positive, indicating a moderate deficiency of heterozygotes. This is often interpreted as the result of partial selfing within a mixed-mating system (Brown, 1979), but T. versicolor is completely outcrossed. Another explanation may be the effect of the life history of T. versicolor in generating temporal and spatial genetic structure. Separate cohorts, established in different years but each derived from an essentially island-wide mating episode, would produce nonoverlapping (but genetically correlated) subpopulations. Within each cohort, inbreeding might be zero, but different seed cohorts would come from different reproductive gene pools, each resulting from a temporal bottleneck created by the subset of flowering adults. A composite vegetative sample, combining individuals of different sizes and ages, would generate a temporal Wahlund effect analogous to that described for spatial subdivision.

Gene flow

The indirect estimates of Nm in the present study suggest substantial gene flow among populations. As seed dispersal is relatively limited (Kitajima & Augspurger, 1989), this most probably represents gene movement by pollen. The direct estimate of pollen movement indicates that pollinators do, in fact, move pollen relatively long distances (≲500 m) among flowering trees. The fact that even relatively isolated flowering individuals also usually show essentially complete outcrossing provides additional evidence that, during fertilization, gene flow is substantial. Pollination is not, however, panmictic, because there were significant differences among individuals in the genetic make-up of their pollen pools.

High rates of gene flow among individuals and between populations have been reported for other tropical tree species. Webb & Bawa (1983) demonstrated that pollen may move 225 m between populations of the hummingbird-pollinated shrub Malvaviscus arboreus. Using paternity analysis, Hamrick & Murawski (1990) showed that more than 25% of the pollinations of the bee-pollinated canopy tree Platypodium elegans on the FDP involved moves of more than 750 m. A recent study by Chase et al. (1996) using microsatellite markers showed pollen movement of up to 350 m in Pithecellobium elegans. Thus T. versicolor is not unusual in the degree of moderate-distance pollen dispersal it experiences.

A life history interpretation of genetic structure

In T. versicolor, genetic structure is unlikely to arise from classical isolation by distance during pollination, as has been suggested for many plant species (Levin & Kerster, 1974). Although neighbouring reproductive adults may exchange pollen at higher frequencies than more distant individuals, the asynchrony of eventual flowering within a cohort, frequent long-distance pollen movement, and the outcrossed mating system should generate considerable genetic mixing in the seeds produced in a single flowering season.

In contrast, seed dispersal in T. versicolor is much more local. Most fruits fall within 150 m of the adult tree (Kitajima & Augspurger, 1989). Growth rates among seedlings in a cohort vary widely (R. B. Foster, pers. obs.), masking their similar genetic origins. As a result, different spatial populations are composed of mixtures of seed cohorts, each genetically constrained by the genotypes of adults flowering in the year it was produced. Those trees that eventually flower then reconstitute a new, temporally constrained gene pool, within which pollen movement occurs over a spatially much larger scale. In this flowering cohort, a set of new seed genotypes is generated, but they again undergo limited dispersal. Spatial genetic structure, such as it is in this species, is thus most probably the result of limited seed movement and temporal genetic stratification among seed and seedling cohorts.

Given the amount of pollen movement among flowering individuals, are the six local samples in this study really populations? The answer seems to be, yes, sort of. They are overlapping dispersal shadows, winnowed progressively by mortality, but they are not genetic demes because the mixing of genes is limited in T. versicolor by its monocarpic flowering phenology. There is a window — that of flowering in the same year — for the individuals in a local area to pass through in order to be part of a deme, in the sense of sharing genes. The actual demic effective population size is determined by flowering phenology, by flowering tree density, and by patterns of pollinator movement among trees. Population samples are a composite of many seed cohorts, and of other, genetically different future reproductive cohorts. From a genetic perspective, being in the same seedling cohort is less important than being in the same flowering cohort. Thus, the life history dynamic of T. versicolor has a corresponding genetic dynamic; the population is repeatedly stratified and homogenized by pollen movement, and then undergoes spatial differentiation because of limited seed dispersal.