Introduction

In forest ecosystems around the world, many plant populations are naturally small, fragmented and sometimes isolated. However, owing to ongoing fragmentation of habitats resulting from anthropogenic processes, especially in tropical areas, an unprecedented number of populations have become smaller and more isolated, representing a potentially serious threat to animal and plant species and their genetic diversity (Saunders et al., 1991; Young et al., 1996). In Brazil, an alarming loss of forests has been reported because of current fragmentation and human-related disturbances (Laurance et al., 2012), resulting in hyperfragmented (Carvalho et al., 2009; Ribeiro et al., 2009) and highly endangered biomes (Ribeiro et al., 2009; Laurance et al., 2011; Gibson et al., 2013; Joly et al., 2014). More than 80% of the remaining Brazilian Atlantic Forest biome is composed of very small fragments (i.e., <50 ha) that are surrounded by open-habitat matrices and isolated by large distances (Ribeiro et al., 2009).

Plant populations that remain small and isolated for many generations after fragmentation suffer an increase in genetic drift and a subsequent loss of genetic diversity and elevated inbreeding, affecting population viability and limiting species’ evolutionary potential to respond to environmental change (Young et al., 1996; Lowe et al., 2005). As the effects of habitat fragmentation are dependent on the life history traits of species, such as lifespan and mating system (Jump and Peñuelas, 2006; Leimu et al., 2006; Bacles and Jump, 2011), the consequences of anthropogenic disturbance on the genetics of plant populations remain highly debated (Kramer et al., 2008; Bacles and Jump, 2011). However, a positive relationship between population size and genetic variation has been documented, implying that negative effects of habitat fragmentation on genetic variation of plants are common (Leimu et al., 2006). Moreover, even in highly fragmented landscapes, low levels of pollen or seed dispersal are sufficient to counteract the long-term detrimental effects of inbreeding and loss of genetic diversity resulting from genetic drift, founder effects and genetic erosion (Ellstrand and Elam, 1993; Young et al., 1996; Ellstrand, 2003, 2014). For example, a pollen-flow study involving Ficus arpazusa in a highly fragmented landscape in Brazil revealed high levels of genetic diversity and a lack of inbreeding due to pollen dispersal over long distances (Nazareno and Carvalho, 2009).

As population size reduction becomes more common because of habitat fragmentation and degradation, it is vital to measure the contemporary gene flow rates across generations to better understand the effects of habitat changes on the genetic structure of plant species in small and isolated populations. Mating parameters, such as outcrossing rates and the dispersal distances of gene flow, influence patterns of reproduction of tropical plant species in natural populations (Barrett, 2010). These parameters determine how genes are recombined and maintained by a species to perpetuate its natural genetic variability, which is the basis of its continued potential evolution (Brown, 1978; Ritland and Jain, 1981; Sork et al., 2002). Thus, assessing genetic diversity and mating patterns is essential to formulate relevant policies on biodiversity conservation and to design programs for effective management, particularly in fragmented habitats. In this context, there is considerable interest in understanding how mating patterns influence genetic structure of plant species in small and isolated populations.

However, limitations of habitat fragmentation surveys related to experimental design (e.g., sample size, independence of replicates and uncontrolled variation in confounding variables; Nazareno and Jump, 2012) have impaired our ability to understand the ecological mechanisms that underlie the genetic consequences of forest fragmentation. Furthermore, although several studies have focused on assessing the influence of fragmentation on genetic variation among populations, questions remain as to how fragmentation affects intrapopulation genetic diversity. To address the methodological issues that arise from comparisons between undisturbed and disturbed populations (Bacles and Jump, 2011; Nazareno and Jump, 2012), and to explore the genetic consequences of fragmentation, we assessed the temporal dynamics of genetic diversity and mating patterns across generations in one very small population of Cariniana estrellensis (Lecythidaceae) in a hyperfragmented landscape of Brazilian Atlantic Forest. We focused primarily on characterizing the mating system and contemporary gene flow for this insect-pollinated and wind-dispersed tropical tree species. Furthermore, we explore the genetic consequences of population fragmentation in C. estrellensis by analyzing patterns of genetic diversity in a sample of adults, seedlings and progeny arrays. Our specific objectives were to: (i) document how C. estrellensis reproduces; (ii) apply parentage analysis to quantify realized gene dispersal distances, identify immigration of propagules from outside populations and characterize the pollen and seed dispersal kernel for C. estrellensis; (iii) assess patterns of fine-scale spatial genetic structure (SGS) across generations; and (iv) infer effects of habitat fragmentation on genetic variability. Our main goal is to generate information about mating patterns and genetic structure dynamics for C. estrellensis, a useful starting point for conservation initiatives, management activities and policies for this plant species.

Materials and methods

Study species and study site

C. estrellensis (Raddi) Kuntze (Lecythidaceae), commonly known as jequitibá-branco, is an emergent neotropical tree that prefers wet and deep soils. This climax forest tree can reach up to 45 m in height, with a diameter at breast height up to 120 cm. C. estrellensis is found mainly in Brazil and also occurs in Bolivia, Paraguay and Peru (Leite, 2007). Different densities of the species have been reported, ranging from 5 to 27 trees ha−1 (Toledo Filho et al., 2000; Nunes and Petrere, 2012). C. estrellensis flowers—creamy white and clustered in axillary racemes—are well distributed over the crown but not produced in large numbers, and physical barriers protect pollen and stigma from insects. During pollination, insects must enter the urceolus to touch the stigma. According to Prance and Mori (1979), Trigona and Melipona are the primary stingless bees involved in pollination of C. legalis, a species closely related to C. estrellensis. As allogamy predominates in the Lecythidaceae family (Mori et al., 2010), we expected to find some evidence of self-incompatibility in C. estrellensis. Fruits are small, cylindrical-oblong, woody, brown fibrous pyxis containing up to 35 alate seeds with membranous wings. Cariniana trees generally flower toward the end of the dry season, and the fruit mature at the beginning of the subsequent dry season (Prance and Mori, 1979). During the dry season, the wind-dispersed, winged seeds are released when the operculum is spontaneously dropped. Seed predation by black howler monkeys has been reported for the species (Oliveira-filho and Galetti, 1996). C. estrellensis wood is of high commercial value and the species is important for degraded area recovery in the tropics (Leite, 2007).

This study was conducted during the dry season of 2007 in a semideciduous, 8.0 ha forest fragment (21°17′47″S, 47°40′29″W; Figure 1a) located in Ribeirão Preto (São Paulo State, Southeastern Brazil). This internationally known agricultural region is one of the most anthropogenically devastated forest landscapes in Brazil. The removal of natural vegetation in this area began more than 100 years ago to make room for coffee production. The remaining natural vegetation is highly fragmented and exposed to continued processes of degradation due to urban settlement and sugarcane production. A rapid floristic survey conducted in 99 forest remnants (sizes varying from 1.5 to 247.0 ha) in the region identified C. estrellensis in 27 remnants (3.0–75.0 ha), although no information exists about their densities (Kotchetkoff-Henriques, 2003). The majority (n=23) of these fragments are isolated by long distances and are very small (<50 ha). Based on floristic surveys in the region, the nearest C. estrellensis individual to the studied population is located 1 km away.

Figure 1
figure 1

Satellite images showing (a) the location of the studied fragment of Atlantic Forest (dashed dark line) within the highly fragmented landscape in Ribeirão Preto, São Paulo State, Southeastern Brazil; the spatial distribution of adults (b) and seedlings (c) of C. estrellensis. The image was accessed via Google in January 2015.

Full size image

Sample collection, DNA extraction and SSR amplification

Samples of C. estrellensis were taken from leaf material of individuals from three ontogenetic stages (adults, seedlings and progeny arrays). In the studied population (density of 3.5 trees ha−1), all adult trees (28 plants reaching a height of 20 m with a trunk diameter ranging from 1 to 5 m; Figure 1b) and seedlings (39 regenerated juvenile trees with heights ranging from 0.15 to 1.5 m; Figure 1c) were sampled, tagged with a numbered aluminum plate and mapped using GPS with ±6 m accuracy (Garmin, Atchison, KS, USA). As C. estrellensis is a long-lived tropical tree species, and based on the size of the adult trees in the studied population, we assume that current adult individuals predate recent fragmentation.

From all adult trees that blossomed in 2007 (n=7), we obtained a total of 399 progeny arrays (65 progeny from five seed trees; 50 progeny from one seed tree and 24 progeny from another seed tree). Although the estimates of inbreeding and genetic diversity can be biased when seeds are collected directly from mother trees, we decided to use this sampling method because the botanical identification of seeds germinated on the forest floor is extremely difficult. Seeds were raised in a native tree nursery at the Ribeirão Preto Campus, University of São Paulo, and information was recorded for each seed tree. Leaves from all 466 individuals (28 adult trees, 39 seedlings and 399 progeny arrays) were stored at −20 °C until DNA extraction.

Genomic DNA was extracted from leaves of each sample using a modified CTAB protocol as described in Alzate-Marin et al. (2009). Amplification protocols for nine microsatellite loci can be found in Guidugli et al. (2009). Amplicons (10 μl) were denatured and separated with 10% denaturing polyacrylamide gels (39:1 acrylamide to bisacrylamide) and stained with silver nitrate. Gels were run with 1 × TBE buffer (90 mm Tris, 92 mm boric acid and 2.5 mm ethylenediaminetetraacetic acid) using a vertical electrophoresis apparatus at constant voltage (550 V) and electric current (23 mA) for 4 h. Allele sizes were estimated by comparison with a 10 base-pair DNA ladder standard (Invitrogen, Carlsbad, CA, USA). To minimize possible genotyping error in size-calling alleles, we ran amplicons of progeny arrays together with their maternal seed tree. After calling alleles for each locus, we verified the sizes by ranking alleles, from lowest to highest, in a denaturing polyacrylamide gel, which was run using the same method described above.

Genetic diversity parameters

As genetic structure, parentage analysis and mating system presuppose that alleles at different loci segregate independently, we undertook Mendelian inheritance analyses for each locus, based on the mother tree and her progeny as proposed by Gillet and Hattemer (1989). Considering that gametic disequilibrium creates pseudoreplication for analyses in which the loci are assumed to be independent samples of the genome, we used the FSTAT 2.9.3.2 program (Goudet, 2002) to test all loci for linkage disequilibrium, applying the Bonferroni correction for multiple comparisons (Rice, 1989). We used the MICROCHECKER 2.2.3 program (Van Oosterhout et al., 2004) to detect null alleles and genotyping artifacts in the data. If null alleles were detected, we estimated their frequency following the method of maximum-likelihood estimator as described by Kalinowski and Taper (2006).

Allele frequencies and descriptive population genetic parameters (number of alleles, number of private alleles, observed heterozygosity, unbiased expected heterozygosity (Nei, 1978) and tests for Hardy–Weinberg equilibrium) for adults, seedlings and progeny arrays of C. estrellensis were obtained using GenAlEx (Peakall and Smouse, 2006). Allelic richness (AR) independent of the sample size was also obtained for adults, seedlings and progeny arrays. We followed the rarefaction procedure method proposed by Kalinowski (2005), which is used to standardize the allelic richness to the smallest sample size across generations. The 95% confidence interval (CI) of the standard error of this parameter was calculated using a jackknife procedure across all loci.

Determination of mating system

The progeny arrays (n=399) from seven C. estrellensis seed trees were analyzed to study mating system. Genotypes obtained were analyzed under mixed-mating and correlated mating models using the multilocus mating system program MLTR version 3.2 (Ritland, 2008). Our analysis took into account that all loci may contain null alleles even if there are none. The mating system parameters estimated were multilocus outcrossing rate (tm), single-locus outcrossing rate (ts), selfing rate (1−tm), biparental inbreeding (or mating among relatives, tm−ts) and multilocus paternity correlation (rp(m)). The average number of pollen donors per seed tree, Nep, was calculated as (Ritland, 1989). We also calculated the inbreeding coefficient of seed trees and progeny arrays. Mating system parameters were estimated for individual families (n=7 families) and at the population level (n=399). Analyses at the population level were carried out using the probabilities of expectation maximization method, and analyses at the individual level used the method of moments according to Ritland (2008). The standard error for each parameter was calculated from 1000 bootstrap replicates with resampling among families. We also quantified the coefficient of pollen pool structure Φft from two-generation analyses (TWOGENER; Smouse et al., 2001). This statistic is analogous to Wright’s FST and ranges from 0 to 1. The standard error for Φft was calculated using a jackknife procedure over loci. All calculations were carried out using the R language for TWOGENER analysis as described by Hirao (available at http://hosho.ees.hokudai.ac.jp/~hirao/TWOGENER/TwoGener.html).

Categorical parentage analysis

Estimates of gene flow from pollen and seeds were carried out by parentage assignment (paternity or maternity) using the maximum-likelihood-based method of Marshall et al. (1998), implemented in the program CERVUS version 3.0.3 (Kalinowski et al., 2007). Seedlings (n=39) and progeny arrays (n=399) from seven C. estrellensis seed trees were analyzed to study pollen dispersal, with all adult trees (n=28) as paternal candidates. To study seed dispersal, we analyzed the genotypes of all seedlings and adult trees, and all adult trees were considered as putative maternal and/or paternal candidates.

We used a two-phased approach: first, we performed 10 000 simulated genotypes to obtain the critical value of Delta (Δcrit); second, paternity assignment was performed to obtain the value of Delta (Δ), which is the difference between the highest and the second highest logarithm of odds scores. Parentage analyses were performed using all loci (with a genotyping error rate of 0.01 per locus estimated in Microchecker) and a confidence level of 95%. Only the potential parent with Δ>Δcrit was considered to be the true parent of the analyzed progeny arrays or seedlings. Additionally, we used the set of assigned individuals to assess pollen and seed dispersal patterns in terms of distance. Pollen and seed dispersal (Euclidean) distances were calculated based on the position of the parent tree and the putative pollen or seed parent within the population. Pollen dispersal distance was calculated for progeny arrays and seedlings based on the position of the seed-tree and putative pollen parent within the small population. For seedlings, we adopted one of the following procedures: (a) when a seedling had only one putative parent within the population, we assumed that the assigned parent was the mother, and this parent was used to infer the distance of seed dispersal; (b) when a seedling had two parents within the population, the candidate nearest to the seedling was assumed as the maternal parent and the further candidate was the paternal parent. Although the later procedure can introduce bias in estimating distance of seed dispersal, it has been used when identifying the maternal parent is impossible, as is the case with monoecious and hermaphroditic species (Dow and Ashley, 1996).

A Kolmogorov–Smirnov test was applied to determine whether the success of pollination is dependent on the geographical distance between adult trees. For this, we compared the distributions of frequencies of pollen dispersal with the frequency distribution of the distances among all adult trees in the population. Additionally, pollen and seed immigration rates (mp and ms, respectively) were calculated as the percentage of genotypes not assigned to a candidate parent within the population (Smouse and Sork, 2004). We also estimated cryptic pollen flow, a value that represents the fraction of genotypes assigned to a putative father within the population when the true father is outside the population (Dow and Ashley, 1996). Assuming a circular area around each seed tree, the effective pollination neighborhood area (Aep=2πσ2; Levin, 1988) was calculated from the variance of pollen flow distance (σ2).

Pollen and seed dispersal kernels

We used a maximum-likelihood procedure to determine an individual pollen dispersal distance probability density function based on genotypes of seedlings (n=39) and progeny arrays (n=399). We estimated the probability of pollen and seeds traveling from their original locations to a given position (i.e., pollen and seed dispersal kernels) for C. estrellensis using a spatially explicit NEIGHBORHOOD model as proposed by Burczyk et al. (2002). Using a maximum-likelihood fractional parent assignment, the NEIGHBORHOOD model was calculated in the software NM+ version 1.1 (Chybicki and Burczyk, 2010). The neighborhood parameter was set to ‘infinite’ to include all sampled adults in our small population as the neighborhood size (Chybicki and Burczyk, 2010). Initial setting values used in the NEIGHBORHOOD model were those estimated from mating system (s, selfing rate) and categorical parentage analysis (dp, mean pollination distance; ds, mean seed dispersal; m, apparent pollen immigration; ms, seed immigration rate). Pollen and seed dispersal were modeled using the three most useful models (i.e., Exponential-power, Weibull and 2Dt) related to dispersal kernels in plants (Austerlitz et al., 2004; Chybicki and Burczyk, 2010; Ottewell et al., 2012; Côrtes et al., 2013). The shape of dispersal kernel tails (b), their scale (a), the average distance of pollen (dp) and seed dispersal (ds), and immigration rates (mp for pollen and ms for seeds) were estimated for all dispersal kernel models. If null alleles were present, the rate of genotype mistyping errors corresponding to the frequency of null alleles at each locus was taken into account in the analysis.

Fine-scale genetic structure

Wright’s inbreeding index, FIS, was calculated and significance tested using the software SPAGEDI v.1.2 (Hardy and Vekemans, 2002). To explore whether C. estrellensis has the ability to survive in a relictual population following habitat fragmentation, we compared the inbreeding index and allelic richness between adult trees (prefragmentation), seedlings and progeny arrays (postfragmentation). We used allelic richness in the comparisons as it is highly influenced by population reduction because of rapid elimination of rare alleles (Cornuet and Luikart, 1996) and because it may more accurately reflect current levels of genetic diversity within the fragmented populations (Jump and Peñuelas, 2006).

We assessed fine-scale SGS for adults and seedlings using SPAGEDI v.1.2 (Hardy and Vekemans, 2002). We calculated Nason’s estimator of kinship coefficient (Fij), as described in Loiselle et al. (1995), because it displays robust statistical properties (Vekemans and Hardy, 2004). At least seven distance classes were identified to reach a minimum of 70 and 100 pairs of individuals per distance class for adult trees and seedlings, respectively. Spatial locations were permuted 20 000 times to obtain error estimates for the null hypothesis (Fij=0). Error estimates for the observed kinship values among adult trees and seedlings were calculated by jackknifing loci 20 000 times. To visualize SGS, we plotted the kinship coefficient against geographical distance. The strength of SGS was assessed using the Sp statistic as described by Vekemans and Hardy (2004), calculated as −blog/(1−F1), where blog is the mean slope of regression of kinship coefficients on log 10 distance and F1 is the mean kinship coefficient of the first distance class. We tested SGS by assessing the significance of the regression slope using SPAGeDi v.1.2 (Hardy and Vekemans, 2002).

Results

Genetic diversity parameters

The microsatellite set for C. estrellensis had a Mendelian inheritance (Supplementary Table 1) and we found no significant linkage disequilibrium after Bonferroni correction (P<0.001) between loci for all life stages. MICROCHECKER 2.2.3 revealed that locus CES02 may contain null alleles at moderate frequencies ranging from 0.031 (adult trees) to 0.132 (seedlings). However, we chose to include it in this study and accounted for null alleles in our mating system and parentage analyses.

DNA analysis of nine microsatellite markers in the total sample of 466 genotypes revealed 68 alleles (Supplementary Table 2). The number of alleles per locus ranged from 2 (Ces14) to 14 (Ces04), with averages ranging from 6.11 (seedlings) to 6.89 (adults) (Supplementary Table 3). The allelic richness differed significantly between life stages according to the 95% CI (Table 1). Of the nine private alleles observed, five (55.5%) were found in C. estrellensis adults (Table 1). Some alleles were also found exclusively in seedlings or progeny arrays (Table 1). The number of alleles and the observed and expected heterozygosity are presented in Table 1. Allele frequencies for all life stages are provided as Supplementary Information (Supplementary Table 2).

Table 1 Population genetic estimates for one small C. estrellensis population, in São Paulo State, Southeast Brazil
Full size table

Across generations of C. estrellensis, a test for Hardy–Weinberg equilibrium found that of 27 locus generations, 2, 1 and 4 (7.4%, 3.7% and 14.8%) showed significant deviation at P<0.05, 0.01 and 0.001, respectively. However, these deviations were observed in only seedlings and progeny arrays. In relation to the dynamics of inbreeding between generations, we observed values of FIS ranging from negative to positive. The mean FIS between loci was not significantly different from zero for all generations (Table 1).

Mating system

C. estrellensis was highly outbred with a multilocus outcrossing rate not significantly different from unity (Table 2). The difference between the multilocus and single-locus outcrossing rate was negative but not significantly different from 0 (Table 2). As the multilocus outcrossing rates were not significantly different from unity, the selfing rate is considered nil. Furthermore, inbreeding was not detected in either adult trees (F=−0.057; 95% CI=−0.121 to 0.017) or progeny arrays (F=−0.043; 95% CI=−0.183 to 0.061). Family-level estimates of the multilocus outcrossing rates were high but not significantly different from unity (Supplementary Table 4). These results indicate that C. estrellensis is an effectively outcrossing species and it seems to be self-incompatible. A moderate degree of correlated paternity within seeds (rp=0.121) was observed. This paternal identity rate within progeny arrays translated into an estimated average of 8.2 effective pollen donors per seed tree (Table 2). The coefficient of pollen pool structure showed low but significant global differentiation between pollen pools fertilizing the seed trees (Φft=0.065, P<0.05). Considering no correlated paternity among seed-tree pairs, this translates into an estimate of 1/2 Φft=7.7 effective pollen donors per seed tree (see Equation (27) in Austerlitz and Smouse, 2001).

Table 2 Population-level estimates of mating system and paternity analysis for a small Cariniana estrellensis (Raddi) Kuntze population in Ribeirão Preto, Sao Paulo State, Southeast Brazil
Full size table

Paternity analysis

We assigned a paternal parent within the fragment to 305 (76%) of 399 progeny arrays with at least 80% confidence. With 95% confidence, we assigned paternity for 237 of the 305 sampled progeny arrays (78%) to C. estrellensis adult trees located within the study population. In addition, 26 of the 28 adult trees contributed paternity to progeny arrays from the analyzed seed trees. Male reproductive success was highly skewed towards a few individuals. For instance, the majority of pollen donor trees contributed single progeny arrays, whereas seven of the 26 pollen donors contributed more than 63.9% of progeny arrays. The percentage of progeny arrays that do not have a parent tree assigned within the population (i.e., pollen immigration) was high (mp=23.5%; Table 2). With a probability of cryptic gene flow of 13% (Table 2), the total (actual) pollen immigration for progeny arrays (i.e., pollen immigration rate plus cryptic gene flow) was 36.5%. Given the high exclusion probability (EP=0.993; Supplementary Table S3) and sampling of all adult trees within the C. estrellensis population, the high pollen immigration rate indicates that the small forest fragment studied herein is not genetically isolated.

Our results demonstrate pollen movement beyond the boundaries of the small fragment over distances possibly >1 km. However, effective pollination distances within the small population were relatively short, with 50% of pollination events occurring at distances up to 72 m (Figure 2). The mean, minimum and maximum pollen dispersal distances to the nearest C. estrellensis pollen donor were 69.9±7.9 (s.d.), 4 and 355 m, respectively (Table 2). Nevertheless, the frequency distribution of observed pollination distances for progeny arrays and the frequency distribution of intertree distances of all adult trees relative to the sampled seed trees were significantly different (D=0.391, P<0.01; Supplementary Figure 1). Overall, the number of pollen donors for each seed tree ranged from 3.8 to 24.6, with an average of 12.9 paternal parents contributing to progeny arrays (Table 1). Effective pollination neighborhood (Aep) for C. estrellensis ranged from 1.0 to 4.4 ha between seed trees, with an average of 2.3 ha (Table 2), equivalent to a circle with a radius of 86 m around a seed tree.

Figure 2
figure 2

Histogram of observed pollination distances for progeny estimated from paternity analyses (gray bars) and intertree distances of all C. estrellensis trees relative to the sampled seed trees (dark gray bars).

Full size image

Parentage analysis

We assigned a paternal and maternal parent from within the fragment to 18 (46%) of the 39 seedlings with 95% confidence. For the remaining 21 seedlings, a putative mother tree was assigned within the population. Overall, the assignment of paternity and maternity to seedlings indicates that seed immigration was more restricted than pollen immigration (Table 3); 62% of the seedlings (53% of apparent pollen immigration along with 9% cryptic gene flow) were fathered by trees located outside of the population. Nevertheless, the actual seed immigration rate may be close to 10% based on the estimate of cryptic seed flow (9.3%).

Table 3 Parentage estimates for Cariniana estrellensis (Raddi) Kuntze seedlings in a small population in Southeast Brazil
Full size table

Compared with the paternity results, pollen immigration and mean pollen dispersal distance for seedlings was greater than that observed for progeny arrays (Table 2). In addition, the effective pollination distance was also greater than the effective seed migration distance, with 50% of pollination events occurring at distances of up to 85 m. The seed dispersal distance varied from 13 to 345 m, with an average of 119.64±8.2 m (s.d.) (Table 3). Approximately 51% of seedlings were dispersed up to 100 m, although 13% of them were dispersed as much as 180 m from the putative mother tree (Supplementary Figure 2).

Pollen and seed dispersal kernel

The best-fitting dispersal distributions were fat-tailed exponential-power curves (b<1; Tables 4 and 5; see Austerlitz et al., 2004) for both progeny arrays and seedlings. The exponential-power dispersal kernel had the highest model likelihood for both progeny arrays and seedlings (Tables 4 and 5), indicating a higher rate of pollen immigration (mp(progeny)=0.26 vs 0.23, mp(seedlings)=0.69 vs 0.53) and seed immigration (ms=0.16 vs 0.00) than expected from the categorical parentage analysis.

Table 4 Model LogL, mp, dp, a and b from Cariniana estrellensis (Raddi) Kuntze progeny arrays (n=399)
Full size table
Table 5 Model LogL, mp, ms, dp and ds, ap and as, and bp and bs from Cariniana estrellensis (Raddi) Kuntze seedlings (n=39)
Full size table

Fine-scale genetic structure

We observed limited SGS between adults and seedlings overall (Spadults=0.016 and Spseedlings=0.006), with a weak kinship coefficient between adjacent C. estrellensis individuals in the first distance class (F1 adults=0.0351, 226.4 m; and F1 seedlings=0.0226, 125.9 m; Figure 3). In addition, slopes (blog) of the correlograms for adult and seedling C. estrellensis generations were not significantly different from the null hypothesis of no SGS (blog adults=−0.0156 with associated determination coefficient (R2) equal to 0.012; blog seedlings=−0.0061 with R2=0.012).

Figure 3
figure 3

Average Loiselle’s kinship coefficient plotted against geographical distance between individuals (solid lines) for adults and seedlings of a smaller C. estrellensis population, São Paulo State, Southeast Brazil. When the kinship coefficient lies above its 95% CIs (dotted lines), individuals are significantly more similar than would be expected through random sampling.

Full size image

Discussion

Mating system

To date, there have been no published descriptions of C. estrellensis mating system. The high multilocus outcrossing rate detected in this study indicates that it is an effectively outcrossing tree species displaying mechanisms of self-incompatibility. In other species belonging to the same family (Lecythidaceae), such as Bertholletia excelsa (O’Malley et al., 1988), Cariniana legalis (Tambarussi, 2013) and Eschweilera ovata (Gusson et al., 2006), the estimated rates of reproduction indicate a mixed system with a predominance of outcrossing and a probable absence of mechanisms for self-incompatibility in these species. Outcrossing rates for hermaphroditic species such as C. estrellensis depend on factors such as the presence and intensity of a self-incompatibility system, the degree of protandry and protogeny, pollinator foraging behavior between and within reproductive trees and flowering tree density in the population (Murawski and Hamrick, 1991). Nevertheless, according to Bawa et al. (1985), most hermaphroditic tropical species have strong barriers to selfing; for example, the floral structure of C. estrellensis likely prevents selfing in this species. According to Mori et al. (2010), representatives of the Lecythidaceae family with actinomorphic flowers (such as C. estrellensis) tend to have their stigmas (i.e., female receptive region) oriented in a specific position that ensures that the pollen is deposited on the receptive field as the pollinator enters the flower but not as it is leaving. Thus, pollen is not deposited on the stigma of the same flower from which it was collected. Specific studies of the genetic control of self-incompatibility in C. estrellensis are necessary to better understand this mechanism.

The results suggest that the reproductive system of C. estrellensis is resilient to a reduction in population size, as the estimates of biparental inbreeding indicate no mating between relatives. A lack of biparental inbreeding—or any inbreeding (Table 2)—in the studied population can be explained by the absence of SGS in the adult generation and by the occurrence of pollen flow over long distances. This is in contrast to several other fragmented tree species that have shown elevated levels of mating between related individuals (Ismail et al., 2012; Tambarussi, 2013), genetic structuring (Dubreuil et al., 2010; Ismail et al., 2012; Saro et al., 2014) and inbreeding (Fuchs et al., 2003; Jump and Peñuelas, 2006; Kettle et al., 2007; Dick et al., 2008; Vranckx et al., 2011; Zhang et al., 2012; Tambarussi, 2013; Finger et al., 2014).

The elevated outcrossing rates, accompanied by moderate values for the estimated paternity correlations observed in this study, strongly corroborate the high number of effective pollen donors involved in seed production in C. estrellensis. The low density of reproductive adult trees in the studied population (3.5 trees ha−1) may have contributed to pollination vectors of C. estrellensis visiting a larger number of individuals. This behavior ensures the maintenance of high rates of outcrossing and moderate correlated paternity rate in this population.

Gene flow

In the small C. estrellensis population, we found that gene dispersal is enhanced, in contrast to the prediction that a reduction in population size following habitat fragmentation disrupts gene dispersal. However, gene flow for C. estrellensis was somewhat lower compared with the values reported for wind-dispersed tropical trees and species with different dispersal patterns (Nakanishi et al., 2009; Berens et al., 2013) even in small populations (Ottewell et al., 2012; Tambarissu, 2013).

Realized gene dispersal distances in this small population were higher for pollen than seeds. This finding indicates that gene flow for C. estrellensis is maintained primarily by pollen dispersal, which is consistent with other studies (Ndiade-Bourobou et al., 2010; Berens et al., 2013). As a matter of course, plants like C. estrellensis, which are effectively outcrossing and insect-pollinated, should show pollen dispersal over long distances, as has been observed for other tropical tree species (Nason et al., 1998; Gaiotto et al., 2003; Nazareno and Carvalho, 2009). Indeed, even in a highly fragmented landscape, we observed a moderate to high frequency (23.5–53%) of gene immigration, indicating that pollen movement beyond the boundaries of the small fragment can reach distances >1 km (i.e., the distance between the small studied population and the nearest pollen source). Gene flow by pollen dispersal beyond the edges of seemingly isolated forest fragments has been reported for several plant species, including species that are animal-pollinated (Nason and Hamrick, 1997; Dow and Ashley, 1998; Dick et al., 2003; Sato et al., 2006; Nazareno and Carvalho, 2009; Buschbom et al., 2011; Ottewell et al., 2012; Côrtes et al., 2013; Tambarussi, 2013; Saro et al., 2014).

The fractional paternity approach revealed an exponential-power distribution with a fat-tailed dispersal curve (b<1) for seedlings and progeny arrays. This finding is in line with the trend towards fat-tailed distributions of gene dispersal in plants (Dick et al., 2003; Austerlitz et al., 2004; Oddou-Muratorio et al., 2005; Klein et al., 2006). The fat-tailed dispersal distribution observed for C. estrellensis implies that long-distance dispersal is more frequent in this species than in plant species exhibiting thin-tailed distributions. However, immigration rates, most notably for seeds, and mean gene dispersal distances, principally for pollen, were inconsistent between fractional paternity analysis and categorical paternity assignments (Tables 2, 3, 4, 5). A discrepancy between paternity approaches has been reported in other mating studies for tree species in small populations (Ottewell et al., 2012; Saro et al., 2014), and this difference likely arises owing to the nature of each analytical approach (Burczyk et al., 2002; Kalinowski et al., 2007).

Although we have not measured functional traits of C. estrellensis, such as the seed terminal velocity and the height of seed release above the vegetation cover, or quantified wind speed and vertical turbulence around the edges of the studied population, it is known that these traits can affect the immigration rates for wind-dispersed plant species (Heydel et al., 2014). In this context, we hypothesize that the influence of these factors—besides the small number of seedlings in the studied fragment and the nature of the method used—could limit the amount of seed immigration into the C. estrellensis population, potentially leading to an underestimation of realized seed dispersal in this species using parentage analysis. Furthermore, the pattern of seed dispersal for C. estrellensis may be influenced by the degree of seed predation by primates, such as Alouatta caraya, that eat the seeds at the base of the wing (Oliveira-Filho and Galetti, 1996).

Overall, our findings indicate that this small population of C. estrellensis is not genetically isolated. Therefore, in the fragmented landscape, there are isolated trees or relictual populations that may contribute to the genetic pool of this specific population. Further studies linking the functional traits of the species and meteorological variables to the estimates of effective gene dispersal, in both fragmented and continuous populations, can strengthen our understanding of the mating pattern of this tropical species.

Diversity and fine-scale genetic structure

Significant reduction in allelic richness but no inbreeding was detected in both postfragmentation generations of C. estrellensis, pointing to a loss of genetic diversity between life stages in this small population. Nonetheless, these results do not fit well with the theory that inbreeding is observed immediately following fragmentation, but genetic diversity is lost slowly over generations (Lowe et al., 2005; Bacles and Jump, 2011); for long-lived plant species, this process may take decades.

Although a significant reduction in allelic richness following fragmentation was observed for C. estrellensis, the extensive gene flow detected in the small population could be attributed to a lack of inbreeding and SGS between the life stages of this species. As has been documented for other outcrossing species in small populations (Balloux, 2004; Saro et al., 2014), we observed an excess of heterozygotes in all life stages of C. estrellensis. Owing to the relatively large range of pollen dispersal, one could expect an effective mixing of genes by pollen flow or even seed dispersal within the studied population, hampering SGS. Moreover, because the number of C. estrellensis individuals within the population was too small to assess SGS with statistical rigor, our SGS results should be viewed with caution.

Another explanation for these findings is that the current adult and seedlings generations have developed in different landscape fragmentation contexts. Thus, it is expected that the structure of genetic variation in the adult generation is composed of overlapping successive generations from different gene pools with different histories of establishment, reflecting a historic situation where habitat fragmentation was less extensive. The seedling life stage, in turn, resulted from the germination of seeds in the current forest fragment. One must consider that this generation was subjected to a selection process and this process excluded inbred or less vigorous individuals. The same could be hypothesized for the progeny arrays, which were germinated in unnatural conditions. Indeed, we observed a low to moderate germination rate (at least 50%, data not shown) for progeny arrays. Given these considerations, the establishment and permanence of seedlings within this population of C. estrellensis can be considered a key factor in the replacement of genetic diversity for the species. Nonetheless, within the studied fragment, the small population of C. estrellensis showed a senescent demographic structure (data not shown) with a greater proportion of adults of older generations. We also observed a lack of regenerating individuals taller than 2 m, which may be a consequence of habitat fragmentation affecting establishment and/or C. estrellensis continuation within the area. This finding, along with the loss of genetic diversity across life stages of C. estrellensis, suggests that even recent habitat fragmentation is adversely affecting C. estrellensis and could compromise the long-term survival of this small population.

Based on our results, it is necessary to take urgent action to increase gene flow for C. estrellensis coupled with management strategies to promote more favorable conditions for the establishment and retention of new generations of seedlings in the habitat. In addition, to develop an efficient ex situ conservation strategy for C. estrellensis—considering also that rare and unique alleles were found in all life stages—seeds should be collected from all reproductive individuals in this small population because they were not genetically related. In parallel, to ensure that even rare and unique alleles present in non-reproductive trees can also be preserved, the use of cutting techniques such as those developed recently for C. estrellensis (Hernandes et al., 2013) is also feasible. Furthermore, collecting information about the number and size of populations throughout the species distribution area and carrying out long-term ecological and genetic studies should be a priority as this information is necessary in designing effective conservation strategies.

Data archiving

Data available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.4b18b.