Introduction

Understanding the processes regulating similarities and differences of fine scale spatial genetic patterns between populations is a central focus in evolutionary biology. Genetic isolation for selectively neutral loci generally decreases rapidly as distance of separation increases (Malécot, 1948), and so does spatial autocorrelation of allele frequencies (Barbujani, 1987). Generally as gene dispersal increases, the spatial structure becomes increasingly diffuse, ultimately becoming virtually nonidentifiable (Epperson, 2003). However, the distribution of genetic variation within a plant population depends not only on pollen and seed dispersal, but also on mating patterns, survivorship through life cycle stages and the population density.

In outcrossing, polycarpic plants, reproduction is influenced by the individual reproductive periodicity, which determines the number of mating partners and effective pollination. Genetic isolation by distance within plant populations may be more strongly influenced by seed dispersal than by pollen flow (Hamilton, 1999). Hence, among sympatric and congeneric species, contrasts in either pollination or seed dispersal traits can affect spatial differences in genetic variation. In other words, the relationship between alternative pollination mechanisms and spatial genetic structure among sympatric species also depends on contrasts in fruit attributes and seed dispersal modes. Small spatial-scale genetic structure should be lessened when the degree of spatial and temporal overlap of matrilineal family groups is greater (Hamrick et al, 1993; Vekemans and Hardy, 2004). Combining analyses of spatial genetic structure with observations of demography may allow greater insight into how the spatial distribution within populations is determined. In addition, standardized measures of dispersal such as Wright's (1943) neighborhood size may be accurately estimated from spatial autocorrelation (Epperson et al, 1999).

Studies of spatial genetic structure of non-woody plants from tropical rainforests are still scarce, particularly in climbing plants and palms, despite their substantial contribution to tropical rain forest diversity (Gentry, 1991; Henderson et al, 1995). Thus, an integrated understanding of microevolutionary processes that regulate and maintain spatial genetic structure of the most locally distinctive tropical species remains an outstanding topic in population evolutionary biology.

In this paper, we characterize the spatial genetic structure within populations of two Neotropical palms having contrasting life histories, Chamaedorea tepejilote and Chamaedorea elatior. Both species are long-lived perennial, polycarpic, dioecious and obligate outcrossers, showing annual reproduction and similar seed dispersal features. However, the species differ in pollination syndromes, growth forms, population density, spatial distribution, frequency of reproductive plants and individual reproductive periodicity. In this research, three objectives are addressed: (1) to study the influence of contrasting life histories on the spatial genetic structure within natural plant populations, (2) to estimate the extent of genetic isolation by distance within populations with high levels of genetic variability, and (3) to estimate the neighborhood size based on the total amount of gene dispersal. To achieve these objectives, weighted Moran's I-statistics and a significance test for differences between spatial autocorrelation coefficients are conducted.

Materials and methods

Study system

Chamaedorea, the largest genus of Neotropical palms, includes approximately 77 species (Henderson et al, 1995). These dioecious palms are distributed in tropical rain forests from Mexico to Brazil and Bolivia (Hodel, 1992). Two Chamaedorea species showing contrasting life history traits were selected. C. tepejilote is an arboreal palm with a single stem up to 5 m tall. It is common in the understory of Los Tuxtlas Reserve, forming dense patches in places where slight disturbances have occurred (Oyama, 1990). The small, greenish-yellow, wind-pollinated flowers appear from October to December (Otero-Arnaiz and Oyama, 2001). Seeds are mainly gravity-dispersed, although several species of birds and bats eat the fruits (Trejo-Pérez, 1989), and seed predation by rodents is common (Oyama, 1991, 1997). Demographic data from permanent plots followed for 4 years (Oyama, 1990) showed that the seed and seedling stages have the highest mortality rates. First reproduction was recorded in palms higher than 50 cm. Most mature individuals produce flowers or fruits during consecutive years, and about 12% of pistillate flowers produced fruits.

The second species, C. elatior, is the only truly vine like and climbing member of the genus, and hence is one of the most distinctive species (Hodel, 1992; Henderson et al, 1995), with a single stem that reaches up to 22 m in length and up to 2 cm in diameter. C. elatior is uncommon (Hodel, 1992). In Mexican rainforest, it may be found in shady sites and occasionally on the slopes of hills. Flowering takes place between February and June (Aguilar, 1986), when small greenish-yellow flowers are produced. Petals of male flowers are joined at the tips and at the top of the pistillode. These flowers are fragrant and have sticky pollen, which can be accessed through lateral slits between the petals (Hodel, 1992; Henderson et al, 1995). Accordingly, insect pollination has been suggested (Henderson, 1986). In one population of Los Tuxtlas, pooled demographic data of C. elatior collected over 4 years (Luna, 1999) showed that mortality was high at adult stages. Reproductive structures were produced by only 83 individuals out of 308 palms scored. Most of these reproduced only once during this period.

Study population and field methods

The study was conducted in a 640-ha lowland tropical rain forest at Los Tuxtlas Field Station, in southeast Mexico (long. 95°04′–95°09′W, lat. 18°34′–18°36′N, alt. 150–700 m a.s.l.). Annual rainfall is approximately 4700 mm, and annual mean temperature is 23°C. Full descriptions of the region have been detailed in González et al (1997).

For C. tepejilote, 188 palms were mapped in a lattice of 24 quadrats of 25 m2 each in a mature forest plot of 600 m2. This species had an aggregated spatial pattern and a density of 1097 individuals >0.5 m in a vegetation census of 1 ha (Bongers et al, 1988). C. elatior was found along a steep slope of ‘El Vigia’ mountain in Los Tuxtlas Reserve. A total of 309 palms were mapped in an area of 550 m long × 200 m wide: a population density of 28 individuals/ha, excluding seedlings. These palms showed a nonuniform spatial distribution, consisting of three main areas within the sampled population.

In both species, three pinnae from the youngest leaf were collected from each palm, excluding seedlings to minimize a demographic impact. Samples were frozen in liquid nitrogen and stored at −70°C until processed for enzyme analysis.

Laboratory and analytical methods

Leaf samples of 126 individuals of C. tepejilote and 183 of C. elatior palms were cut finely and ground in an extraction buffer that was prepared with three parts YO buffer (Yeh and O'Malley, 1980) and one part Veg II buffer (Cheliak and Pitel, 1984). Enzyme genotypes were obtained using standard methods for starch gel electrophoresis (Soltis and Soltis, 1989; Kephart, 1990). In all, 31 polymorphic loci of C. tepejilote and 27 of C. elatior were identified from 18 enzymes using four buffer systems: (1) C system, Lithium-borate (pH 8.3) (Stuber et al, 1988) was used to resolve diaphorase ‘Dia’ (Cheliak and Pitel, 1984), glucose 6-phosphate dehydrogenase ‘G6pd’ (Soltis et al, 1983), glutamate dehydrogenase ‘Gdh’ (Conkle et al, 1982), esterase ‘Est’ (Selander et al, 1986), leucine aminopeptidase ‘Lap’ (Werth, 1985), glutamate oxaloacetate transaminase ‘Got’ (Wyatt, 1989), cathodic ‘Cpx’ and anodic ‘Apx’ peroxidase (Hakim-Elahi, 1981), and rubisco ‘Rub’ (E.C. 4.1.1.39). (2) D system, histidine-citrate (pH 6.5) (Stuber et al, 1988) resolved malate dehydrogenase ‘Mdh’ (Conkle et al, 1982), phosphoglucomutase ‘Pgm’ (Soltis et al, 1983), malic enzyme ‘Me’ (Soltis et al, 1983) and phosphoglucose isomerase ‘Pgi’ (Conkle et al, 1982). (3) Morpholine-citrate pH 6.1 (Wendel and Weeden, 1989) resolved acid phosphatase ‘Acph’ (Hakim-Elahi, 1981), 6-phosphogluconate dehydrogenase ‘6pgd’ (Cheliak and Pitel, 1984), isocitrate dehydrogenase ‘Idh’ (Soltis et al, 1983) and shikimate dehydrogenase ‘Sdh’ (Werth, 1985). (4) PP system, a discontinuous sodium-borate/tris-citrate system (pH 7.5) (Mitton et al, 1979) was only used to resolve menadione reductase ‘Mnr’ (Conkle et al, 1982). Allozymes had banding patterns consistent with expected quaternary structures (Wendel and Weeden, 1989; Soltis and Soltis, 1989). The mode of inheritance for the isozyme patterns was not further tested by crossing. Putative loci and alleles were designated sequentially. The one with the most anodally migrating isozyme or alleles was designated as ‘1’, the next as ‘2’ and so on.

Spatial structure analyses were conducted separately for each allele at each polymorphic locus in both species, using Moran's I spatial autocorrelation statistics (Sokal and Oden, 1978). Alleles with three or fewer copies in the sample were excluded as noninformative for the spatial analysis. For diallelic loci, the autocorrelation coefficients for each allele are identical, and thus only one value was considered. At multialleleic loci, the values of the most common allele at each locus were excluded when the I-averages were calculated. This procedure results in statistical near-independence among the remaining values (Epperson, 2003), and allows the standard errors of average I-statistics to be estimated accurately. Average I-values use the same information as most estimates of spatial covariance or conditional kinship, R, and indeed if the fixation index is known and certain conditions obtain, the results for I can be extended to R estimates, which had previously unknown distributions (Epperson, 2003).

In the calculation of allele-specific I-statistics, each pair of individuals was assigned to one distance class according to the Euclidean distance separating the two palms. As small-scale measures represent the spatial structure more accurately, the upper limit of the first distance class was designed based on an estimate of the average distance that separates the nearest neighboring individuals. Thus, the majority of the neighboring pairs were included in this class. The average number of comparisons over loci for each distance class is shown in Tables 1 and 2. The distance classes constructed for C. tepejilote were: (1) 0<2.5, (2) 2.5<5, (3) 5<7.5, (4) 7.5<10, (5) 10<15, (6) 15<20, (7) 20<25 and (8) 25–32 m. Those for C. elatior were: (1) 0<10, (2) 10<25, (3) 25<50, (4) 50<100, (5) 100<200, (6) 200<300 and (7) 300<600 m.

Table 1 Summary of Moran's I spatial autocorrelation coefficients of 59 alleles in one population of C. tepejilote for eight distance classes
Full size table
Table 2 Summary of Moran's I spatial autocorrelation coefficients of 53 alleles in one population of C. elatior for seven distance classes
Full size table

Unweighted Moran's I-statistics (Iu) were calculated for each allele and distance class (Sokal and Oden, 1978). Each Iu-value was tested for significant deviation from the expected value, E(Iu)=−1/(ni−1) under the null hypothesis of random spatial distribution (Cliff and Ord, 1981). Arithmetic averages of Iu and E(Iu) were calculated separately across alleles for all loci, for each sample. In addition, the variance of each average Iu value was estimated by σ2 Iu=(∑var(Iu)i) (k2)−1 (Epperson, 2003). Furthermore, significance tests for the average values of each distance class were developed by estimating the standard errors of Īu under the null hypothesis as SE(Iu)=(σ2Iu)1/2. These values were used directly to build 95% confidence intervals for the average values, because the theoretical results further allow the assumption that the standard errors on the I-averages under the null hypothesis are scarcely affected by the spatial structure (Epperson et al, 1999; Epperson, 2003).

Significant positive I-values indicate that pairs of individuals in the distance class considered have more alleles in common than would be expected by chance, whereas a significant negative I-value indicates that such individuals have fewer alleles in common. Statistical significance of each set of I-statistics for mutually exclusive distance classes, or I correlogram, was tested using Bonferroni's criteria (Sakai and Oden, 1983). All Moran's I spatial analyses were performed using Wartenberg's SAAP computer program (1989). Finally, the null hypothesis that Moran's I did not differ significantly from the expected value, among all individuals in the population, was tested using SNDu=((Iu−E(Iu)) SE(Iu)−1), under the assumption that samples are drawn at random from a normal distribution (Sokal and Oden, 1978). Values of the SND ≥1.96 or ≤−1.96 indicate that I-value differs significantly from the null expectation with a two-tailed probability of 0.05.

In addition, weighted average Moran's I-statistics were estimated to obtain a more accurate measure of the I-average over alleles for all loci, because the inspection of the estimated standard errors of I-values indicated dissimilar stochastic and statistical variances among alleles as a result of differences in final sample size among loci. It has been noted that to calculate weighted I-statistics requires a priori information on the weights (Epperson, 2003). Accordingly, standard errors of I-values were used to calculate weighted averages of Moran's I-statistics (Iw) for each distance class and to provide more appropriate consensus correlograms, using the formula:

where wi is the inverse of the standard error of Iu-value for ith allele, so wi=1/SE((Iu)i); and (Iu)i is the unweighted Moran's I-coefficient for ith allele. Thus, the variance of Iw was estimated by means of σ2Iw=(∑wi2) (var ((Iu)i)) ((∑wi)2)−1. Hence SE(Iw)=(σ2Iw)1/2, which was used directly to construct 95% confidence intervals for the average values. The weighted average expected value, under the null hypothesis of a random spatial distribution, was estimated for each distance class as E(Iw)=(∑wi (E(Iu)i)) (∑wi)−1. As before, a standard normal deviate SNDw was developed for the testing of significance within distance classes.

A statistical test was created to evaluate whether or not there was a difference between C. elatior and C. tepejilote in the amount of spatial autocorrelation in the shortest distance classes, which in both species contained primarily nearest neighbors. The test was formed by subtracting the excess of average Moran's I for C. tepejilote, It over its expected value under the null hypothesis of random distribution E(It), from the equivalent excess for C. elatior. This difference, denoted Y, was explicitly calculated as Y= ((Ie−E(Ie))−(It−E(It))). The value of Y is then approximately distributed as normal with mean zero and variance σY2=SE(Ie)2+SE(It)2, under the null hypothesis that the averages are the same. The standard error of the difference Y would then be SE(Y)=(σY2)1/2. Consequently the coefficent Y/SE(Y) was tested for statistical significance using a one-tailed probability of 0.05. Thus, the excess of average Ie over E(Ie) is regarded as statistically different from the excess of average It over E(It) at the P<0.05 significance level if Y is greater than 1.645. In addition, the standard error of Y was used directly to construct 90% confidence intervals using ±1.645 times the SE(Y). Y-statistics were calculated separately for unweighted and weighted I-averages.

Neighborhood sizes for C. elatior and C. tepejilote were estimated by using the averaged (over alleles and loci) observed values minus the averaged expected values of Moran's I-statistics for the first distance class, and comparing these to theoretical values in Table 4 of Epperson et al (1999).

Results

A total of 776 Moran's I spatial autocorrelation coefficients for 97 alleles of C. tepejilote and 637 I-coefficients for 91 alleles of C. elatior were calculated including alleles with less than four copies (seven and 11 alleles for each species, respectively). After the most common allele of each locus was excluded, 59 alleles of 31 polymorphic loci of C. tepejilote and 53 alleles of 27 polymorphic loci of C. elatior were used to estimate the unweighted and weighted values of the I-average (Tables 1 and 2).

The spatial structure of the C. tepejilote population was examined using a total of 472 Moran's I spatial autocorrelation coefficients from the eight distance classes. Among these, 74 (15.7%) I-values were significantly different (27 at P<0.01, and 47 at P<0.05) from the expected value, under the null hypothesis of random distribution. For each distance class, the overall percentages of significant I-statistics, and the total of positive (+) and negative (−) significant values were: (1) 15.25 (+7, −2); (2) 13.56 (+5, −3); (3) 16.95 (+7, −3); (4) 22.03 (+7, −6); (5) 22.03 (+6, −7); (6) 13.56 (+2, −6); (7) 10.17 (+0, −6) and (8) 11.86 (+0, −7). For example, for the particularly important first distance class, nine out of 59 (15.25%) values were statistically significant; seven of these were positive and two were negative. Correlograms, entire sets of I-statistics, for eight distance classes showed significant values at the 5% level for 15 (25.42%) out of a total of 59 cases.

For C. elatior, the spatial pattern of genetic structure was examined using a total of 371 Moran's I spatial autocorrelation coefficients for the seven distance classes. In this case, 46 (12.4%) I-values were significantly different (26 at P<0.01, and 20 at P<0.05) from the expected value, E(Iu)=−0.00925, for randomly distributed alleles. For each distance class, the overall percentages of significant I-statistics, and the total positive (+) and negative (−) significant values were: (1) 16.98 (+9, −0); (2) 13.21 (+5, −2); (3) 7.55 (+2, −2); (4) 11.32 (+5, −1); (5) 13.21 (+1, −6); (6) 15.09 (+5, −3) and (7) 9.43 (+0, −5). For example, for the particularly important first distance class (0 to <10 m), nine significantly positive cases were observed, whereas no significantly negative cases were detected. Correlograms were statistically significant at the 5% level in 14 (26.42%) out of a total of 53 cases.

The average values for Moran's I over the 59 alleles of C. tepejilote and the 53 alleles of C. elatior at all allozyme loci for each distance class are listed in Tables 1 and 2. Also, the estimated standard errors under the null hypothesis, the corresponding standard normal deviate Z and the P-values for two-tailed tests are shown. Overall, the average correlogram for the C. elatior population indicates short distance autocorrelations, whereas that for C. tepejilote shows a nearly random distribution of genotypes over the sampled area (Figure 1). For the C. tepejilote population, the unweighted and weighted Moran's I-averages for the first distance class were equal to 0.0085 and 0.0093, respectively, which were statistically significant for a two-tailed test (P=0.046 and 0.028, respectively). Moreover, there was a significant negative average autocorrelation at the largest distance class (eight). For the C. elatior population, consensus correlograms indicate that the highest relative levels of genetic similarity occur among individuals separated by <10 m, and unweighted and weighted average values for the first distance class were 0.0387 and 0.0335, respectively (two-tailed test, P<0.00001 in both estimations).

Figure 1
figure 1

Spatial autocorrelation consensus correlograms (Moran's I) for C. tepejilote (○, •) and C. elatior (⋄, ♦). The solid symbols indicate the average expected values under the null hypothesis of a random distribution. The dotted lines represent weighted average values and the solid lines represent the unweighted average values. The upper bounds on the distance classes are shown in bold types for C. tepejilote and in italics for C. elatior. Standard errors are also shown.

Full size image

The statistical significance test of the null hypothesis, that the amount of autocorrelation for the shortest distance classes (which for each species contains primarily nearest neighbors) is the same for C. elatior and C. tepejilote, was rejected at a one-tailed probability of 0.05 (90% of the distribution), when it was evaluated using either unweighted or weighted I-averages. The amount of spatial genetic structure of the C. elatior population was shown to be significantly greater than that of the C. tepejilote population, with unweighted (SND=2.129, P=0.0166) and weighted (SND=1.885, P=0.0297) summary statistics. In addition, unweighted Moran's I-averages estimated for the C. elatior population and the C. tepejilote population were statistically different at a 95% confidence two-tailed test (SND=2.129, P=0.0333), although the overlap of those distributions was not rejected when weighted autocorrelation averages were tested at the same significance level.

Neighborhood size was approximately 100 individuals in the case of C. elatior and about 300 for C. tepejilote.

Discussion

Nonrandom spatial genetic patterns within natural populations of C. tepejilote and of C. elatior, and differences in the extent of genetic isolation by distance between them, were precisely measured in this study. Spatial genetic structure was significantly higher in C. elatior than in C. tepejilote. For the latter population, Moran's I spatial autocorrelation values were less than 0.01, yet were statistically distinguishable from a random distribution. For the population of C. elatior, mean autocorrelation values for the shortest distances were comparable to I-values of 0.075 for Psychotria nervosa (Dewey and Heywood, 1988), ∼0.05 for Quercus laevis (Berg and Hamrick, 1995) and 0.05 for an old growth population of Pinus strobus (Epperson and Chung, 2001). Weak spatial structure at the C. elatior (Iu=0.034, Iw=0.039) and C. tepejilote (Iu=0.008, Iw=0.009) populations is in broad terms consistent with those of some predominantly outcrossing, woody perennial plant species reported previously.

Precise estimates of effective population size are critical for interpreting many microevolutionary processes in population biology. The degree of spatial genetic structure at quasi-stationarity stage under an isolation-by-distance process is primarily a function of Wright's neighborhood size (Sokal and Wartenberg, 1983). Neighborhood size can be influenced by overlapping generations, spatial–temporal variations in number of reproductive individuals, variance in reproductive success and biased sex ratios (Álvarez-Buylla et al, 1996). The dioecious species C. tepejilote and C. elatior did not show significant departures from the theoretically expected sexual proportions (Oyama, 1990; Luna, 1999). In such cases, the value of neighborhood size represents the total number of males and females. For C. tepejilote, 1097 individuals with stem >0.5 m were recorded within 1 ha, where this species exhibited an aggregated spatial pattern (Bongers et al, 1988). Stable structure both of size and life stages indicated that reproductive plants constituted about one-third (109 out of 341) of the sample (Oyama, 1990). Then, assuming that the sex ratio of C. tepejilote remains equal to 1:1, the estimated neighborhood size of 300 individuals (∼150 males plus ∼150 females) would occupy an approximate area of 1 ha. For C. elatior, the estimated neighborhood size of 100 reproductive individuals (∼50 males plus ∼50 females) would be contained in 13.2 ha (1.2-fold the total sampled area), assuming similar patterns of plant distribution and reproduction than in the study site.

Spatial patterns of genetic variation within populations depend primarily on the patterns and distance of pollen and seed dispersal. Contrasting pollination mechanisms influence pollen movement distances, which, together with spatial distribution of the flowering palms and the individual reproductive periodicity, are the primary determinants of mating by proximity in these populations. For C. tepejilote, wind mediates pollination between reproductive palms distributed at random among the dense understory vegetation at the sampled site (Oyama, 1990). Its copious pollen release is triggered by strong winds during the local tropical storm season. Unpredictable fluctuations of pollen movement may enhance the genetic variation of this wind-pollinated species by elevating the mating-pair heterogeneity. In this population, 32% of the total palms reproduced during the 4 years study period, most of them (60%) having three or four annual flowering events. Paternity analysis suggested that almost one-third of seedlings had sires located outside the sampled area (Otero-Arnaiz and Oyama, unpublished data). This level of pollen flow is consistent with that in other wind-pollinated species (Muona, 1990; Schnabel et al, 1991). For C. elatior, mating between spatially nonuniform reproductive palms through pollinators such as thrips occurs in the subcanopy. Reproductive individuals constituted 27% of the total palms, and 68% of those flowered once or twice during the 4-year study period. Thus, limited pollen flow may restrict genetic recombination by decreasing the number of mating partners. Seeds in the same inflorescence could, as a result, be more genetically related than half-sibs. The population of C. elatior has high levels of isozyme genetic variation, although less than C. tepejilote (Luna, 1999; Luna and Oyama, unpublished data). Accordingly, mating patterns may influence the differences of genetic similarity among the nearest neighbors between the studied palms.

Fruit dispersal in both species is predominantly by gravity. However, secondary dispersal could occasionally cause seed movements distant from the maternal palms, which was evident in offspring recruitment observed during some periods at sites without local fruit production. In the C. elatior site, all fruits in 1 year were produced from just one female palm. The following year, there were many new juvenile plants established in the study site. Since mature seeds germinate in approximately 9 months (M Sinaca-Colín, personal communication), and the seedling–seed rate has been estimated at 3%, it is plausible to think that this recruitment is coming from seed dispersal of palms located in the vicinity of the population studied. Gravity-dispersed seeds from maternal palms in the surrounding area cannot explain such recruitment, given that the study site is flanked by down-slopes and is located on one of the highest mountains at the Los Tuxtlas Reserve. Accordingly, the offspring have likely arisen from animal-dispersed seeds from outside of the sampled area. In the case of C. tepejilote, understory vegetation between two permanent plots was cut down 50 years ago, but undisturbed during the last 15 years. In 2002, most of the palms (∼90%) sampled in this area were seedlings and juveniles <0.5 m in height. There were six (out of 212) plants with fruits. This regenerative local population may be attributable to extensive seed-mediated gene flow. Independent and long-distance dispersal of single-seeded fruits may account for extensive gene flow and little spatial genetic structuring, in contrast to the effects of correlated seed dispersal, for example, in intact multiseeded fruits (Epperson and Alvarez-Buylla, 1997; Chung et al, 2000; Chung et al, 2003b).

Factors that regulate population dynamics also influence spatial distribution of individuals, and thus the small amount of spatial genetic structuring detected within the study populations. The initial genetic correlation established in the gravity-dispersed progeny may decrease due to thinning processes and overlapping of survivors from different age cohorts. Seeds of C. elatior and of C. tepejilote had high mortality rates. Seedlings were produced by 2.6 and 7.2% of seeds, respectively. For C. tepejilote, about one-third of the viable seeds that reached the soil were ingested by a heteromyid rodent, Heteromys desmarestianus (Oyama, 1997). In both species, only a small proportion of seedlings survive to replace adults. Thus, the genetic correlations among young individuals may have diminished as they matured and were thinned out (Hamrick et al, 1993; Epperson and Álvarez-Buylla, 1997; Chung et al, 2003a).

The density of reproductive individuals is influenced by resource allocation to growth, survivorship and reproduction within populations (Stearns, 1992). Strong differences in these factors were noted between the study species. C. elatior showed high adult mortality, infrequent individual reproduction and peak fruiting years. In contrast, C. tepejilote showed much higher adult survivorship, annual individual reproduction and more regular fruit production. These differences may have resulted in differential densities, and hence weaker spatial genetic structure in the site of the common, wind-pollinated and understory species. For long-lived species, especially polycarpic plants, overlapping generations and seed shadows may greatly mask the genetic relatedness among individuals of the same age (Chung et al, 2003a; Walter and Epperson, 2004). The lower spatial genetic structure of C. tepejilote could thus be the result of a more diverse mixture of family patches, each containing a larger number of individuals, longer generations and higher genetic variation within populations, as compared to C. elatior. Finally, it is apparent that a better understanding of the dynamics of spatial genetic structure within populations will require explicit research into the processes that regulate the degree and time scale both of transient and long-term spatial genetic structure.