Molecular Ecology (2006) doi: 10.1111/j.1365-294X.2006.03193.x BIlackwemll Publishing Ltdpact of selective logging on inbreeding and gene dispersal in an Amazonian tree population of Carapa guianensis Aubl. D. CLOUTIER,* M. KANASHIRO,† A. Y. CIAMPI‡ and D. J. SCHOEN* *Biology Department, McGill University, 1205 Ave. Docteur Penfield, Montréal, Québec, Canada, H3A 1B1, †Embrapa Amazônia Oriental, Trav. Dr Eneas Pinheiro, S/N Marco, CP 48, CEP 66095-100, Belem, Para, Brazil, ‡Embrapa Cenargen, PqEB, Final Av. W5 Norte, CP 2372, CEP 70770-900, Brazilia, DF, Brazil Abstract Selective logging may impact patterns of genetic diversity within populations of harvested forest tree species by increasing distances separating conspecific trees, and modifying physical and biotic features of the forest habitat. We measured levels of gene diversity, inbreeding, pollen dispersal and spatial genetic structure (SGS) of an Amazonian insect- pollinated Carapa guianensis population before and after commercial selective logging. Similar levels of gene diversity and allelic richness were found before and after logging in both the adult and the seed generations. Pre- and post-harvest outcrossing rates were high, and not significantly different from one another. We found no significant levels of bipa- rental inbreeding either before or after logging. Low levels of pollen pool differentiation were found, and the pre- vs. post-harvest difference was not significant. Pollen dispersal distance estimates averaged between 75 m and 265 m before logging, and between 76 m and 268 m after logging, depending on the value of tree density and the dispersal model used. There were weak and similar levels of differentiation of allele frequencies in the adults and in the pollen pool, before and after logging occurred, as well as weak and similar pre- and post-harvest levels of SGS among adult trees. The large neighbourhood sizes estimated suggest high historical levels of gene flow. Overall our results indicate that there is no clear short-term genetic impact of selective logging on this population of C. guianensis. Keywords: genetic diversity, mating system, microsatellite loci, pollen dispersal, selective logging, spatial genetic structure Received 9 July 2006; revision accepted 7 October 2006 a population genetics perspective, such impacts may Introduction induce changes in patterns of genetic diversity, inbreeding, The exploitation of a proportion of forest trees chosen for gene flow, and the effective size of populations. Two the economic value of their wood, i.e. selective logging, has expected genetic consequences of small population size a direct impact on the demography of harvested tree include increases in the level of genetic drift and in- populations, as it: (i) removes large individuals that likely breeding (Ellstrand & Elam 1993). The potential losses of contribute more to reproduction than smaller individuals, genetic variation associated with logging may decrease the and (ii) reduces the overall population density of repro- ability of populations to adapt to directional abiotic and biotic ductive individuals. Additionally, tree harvesting may environmental changes as well as reduce their flexibility to result in ‘collateral damage’ to the forest habitat (e.g. a respond to short-term challenges such as pathogens and shift in the abundance, diversity and behaviour of animal herbivores (Lowe et al. 2005). Thus, reductions in population pollinators) that in turn can have an impact on the size and increase in inbreeding resulting from selective reproductive biology of remaining tree populations. From logging may pose significant threats to the long-term viability of harvested tree populations. Correspondence: Dominic Cloutier, Fax: + (1) 514-398-5069; E- The directly measurable population genetic effects of mail: [email protected] selective logging could fall into several categories. First, © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd 2 D. CLOUTIER ET AL. selective harvesting could lead to an immediate loss of Materials and methods genetic diversity as a consequence of removal of adult trees; or in the seed progeny generation, as a consequence Study species of reduced availability of pollen donors. Such changes, if they occur, should be detectable with marker genes. A Carapa guianensis Aubl. (Meliaceae) is a widespread study of Carapa guianensis in Central America found that Neotropical tree found in the Caribbean islands, Central marker allelic richness in saplings was lower than in the America and northern South America. The species inhabits adult population in logged forests, but not in unlogged upland terra firme forests as well as floodplain varzea forests forests (Dayanandan et al. 1999; but see Hall et al. 1994). in the Amazon basin. In Brazil, it is an important timber Second, coupled with the reduction in density following tree species, and the oil extracted from the seed is exten- tree harvest, there may be a reduction in the number of sively traded and used for its medicinal properties. C. individuals available to donate pollen. This may restrict guianensis is a monoecious species, and the small white pollen movement and increase inbreeding, two factors that flowers (3–5 mm diameter) are visited by small insects (e.g. could lead to loss of genetic diversity in the long term. In the stingless bees, beetles and moths) (D. Cloutier, personal review by Lowe et al. (2005), six out of eight experimental observation). Individual trees above a size of ∼20 cm studies that assessed progeny arrays for levels of inbreeding diameter at breast height (d.b.h.) may be seen with flowers highlighted significant anthropogenic impacts, including a (Dendrogene project, unpublished data) and trees below study of Carapa procera in French Guiana where lower out- ∼30 cm d.b.h. typically do not produce fruits in closed crossing rates were found in logged plots compared to undisturbed forests (D. Cloutier, personal observation), undisturbed plots (Doligez & Joly 1997). Alternatively, one while trees above the size of 30 cm d.b.h. may produce no may hypothesize that pollinators are able to cope with the fruits or produce fruits once or several times per year. Both increased intertree distance following selective logging, and flowers and fruits may be observed simultaneously on that gene dispersal could be maintained (or even increased). the same tree (D. Cloutier, personal observation). At the This alternate hypothesis is supported by studies where population level, flowers and fruits are produced year high levels of pollen dispersal were discovered in an round, with several intrayear peaks of flowering intensity undisturbed population of C. guianensis (Cloutier et al. in and fruit release (Dendrogene project, unpublished data). press), while increased pollen flow has been reported During the study period, most fruits were released around following habitat fragmentation in Swietenia humilis (White the onset (January–February) and the end (May–June) of the et al. 2002). Third, population thinning due to logging could wet season, allowing us to sample fruits at each of those modify the spatial genetic structure of reproductive adults, times. The fruit of C. guianensis contains 4–20 seeds, weighing potentially leading to an increased proportion of matings 20–40 g each, which may be dispersed by water in flood- among unrelated individuals in the population. Support prone forests. However, in the upland terra firme populations, for this hypothesis comes from studies where self-thinning the seeds are dispersed by gravity, and secondarily by in tree populations resulted in a reduction of spatial medium-sized scatter-hoarding rodents over distances less genetic structure from younger to older cohorts (Hamrick than 25 m (Guariguata et al. 2002; Cloutier et al. 2005). As a et al. 1993). result, seeds and seedlings are typically found directly The goals of the present study were to assess the level of beneath the maternal parent tree at upland sites. genetic diversity, inbreeding, gene flow and spatial genetic structure in a natural population of the ‘small-insect’- Study site pollinated tree species C. guianensis managed for timber harvest. For the purposes of forest management in the The study site lies at the Tapajós National Forest (Flona Brazilian Amazon, C. guianensis is classified as a high-density Tapajós), Pará state, Brazil, at 2°51′S, 54°57′W, and 175 m and fast-growing climax species. Seed progenies were above sea level. The area is a flat plateau covered by a collected over a 22-month period, before and after selective dense terra firme forest. The climate is tropical with a mean logging occurred, allowing us to test the following broad annual rainfall of 2000 mm (peaks from February to May) genetic hypotheses: (i) selective logging may change levels and an annual dry season of 2–3 months (August to October). of genetic diversity; (ii) selective logging may change gene Before the creation of the Flona Tapajós in 1974, the only flow through pollen and change levels of inbreeding; and known recent disturbances were sporadic low-impact (iii) selective logging may change levels of spatial genetic activities such as hunting and selective logging of a few structure among adult trees. The results are discussed with tree species, but not of C. guianensis (Jennings 1997). Under respect to demographical and ecological aspects of C. the auspices of the Dendrogene project (Kanashiro et al. guianensis, the historical context of the study population, 2002), an intensive study plot (500 ha) is being monitored and the short-term consequences of logging for the genetic at Tapajós for the effects of selective logging (reduced resource base of C. guianensis. impact logging — RIL) on the ecology and genetics of this © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd GENETIC IMPACTS OF AMAZONIAN TREE HARVESTING 3 Fig. 1 (A) Spatial distribution of all Carapa guianensis adult trees above 30 cm d.b.h. in the central 400 hectares (2000 m × 2000 m) of the study site. Adult trees not sampled for genetic analyses (not genotyped) and those sampled (genotyped) are indicated, respectively, by light grey and dark grey open circles. Filled black circles correspond to the 21 trap trees from which fruits were collected. The size of the circles is proportional to the d.b.h. of the tree. (B) Same as (A) except that the spatial locations of the harvested adult trees at the end of 2003 are indicated by filled light grey circles. (C) Spatial distributions of the juvenile trees (black squares) and the seedlings (black triangles) sampled for genetic analyses. and other species. The study species were chosen to have was conducted in the study plot, and every tree larger than different characteristics such as growth response to light, 20 cm d.b.h. was identified and mapped. reproductive ecology, and population demography. This A mean density of 2.5 C. guianensis trees/ha ≥ 30 cm d.b.h. was done in order to place species into groups for manage- was found in the 400 ha of the central plot area (Fig. 1), ment purposes, and to predict (through genetic modelling) indicating that C. guianensis has the second highest average impacts on future generations with respect to genetic tree density at this site. At the end of 2003, the study plot diversity. Among the species included in the study, C. was selectively logged and at least 40 tree species were guianensis was classified as a fast-growing climax species, harvested, C. guianensis being one of them. The size below locally abundant relatively to other species, and with which C. guianensis trees were not harvested (cutting size reproduction through monoecy and nonsynchronous flower- limit) was about 53 cm d.b.h. A total of 257 C. guianensis ing. Prior to this study, a commercial pre-logging inventory trees were harvested and sent to the sawmill (from the 474 © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd 4 D. CLOUTIER ET AL. between 53 and 111 cm d.b.h. present prior to logging), i.e. over the 22-month period from each of the 21 trap trees, more than 50% of C. guianensis trees of harvestable size were varying between 12 and 37 (total of 391) for the pre-harvest removed, leaving approximately 2.0 trees/ha ≥ 30 cm d.b.h. period, and between 10 and 33 (total of 390) for the post- after logging. The distribution of the harvested and non- harvest period. The seed were sent to the laboratory in harvested C. guianensis trees at the study site is shown in closed plastic bags where the small living embryos were Fig. 1. Trees left standing often had a hollow or poorly extracted for DNA analyses. formed trunk. It is worth noting that a substantial number Leaf and cambium tissue were sampled in the popula- of additional trees were likely damaged or killed directly tion to assess levels of genetic diversity and spatial genetic during the logging operations, or indirectly (trees that fell structure in three different cohorts (seedling, juvenile and following increased exposure to wind after the logging), adult), defined as follows. The adults (n = 199) were repro- but no estimate of this additional damage is available, and ductive individuals, defined here as trees of size above therefore the harvested trees in Fig. 1 should be considered 30 cm d.b.h. The juveniles (n = 82) were trees between 10 as a conservative estimate of the loss of C. guianensis and 30 cm d.b.h., mostly nonreproductive. The seedlings conspecifics. (n = 84) were plants less than 2 years old and less than 1 m in height, and were entirely nonreproductive. The distri- bution of sampled adults, juveniles and seedlings is shown Field sampling in Fig. 1. When leaves were accessible, a piece of leaf was An important feature of this study is that seed progeny taken in the field and dried in silica. Otherwise, a piece of were collected from the same C. guianensis trees monitored cambium (∼150 mg) was extracted and stored in a buffer of during 22 months from 2003 to 2005, within which time 70% ethanol and 0.3% β-mercaptoethanol before being sent interval controlled commercial selective logging was to the laboratory. The leaf and cambium samples analysed conducted at the end of 2003. The samples taken in the field were randomly selected in the laboratory from the set of during that time interval can be divided into two types: (i) field samples available, but since some areas were more seed sampled from 21 reproductive trees monitored before intensively sampled than others in the field, the distribution and after logging, and (ii) leaf and cambium sampled before of analysed samples is not uniform (Fig. 1) (e.g. seedlings logging from three cohorts of the existing population (e.g. were sampled along three transects). seedling, juvenile and adult). Seeds that were not previously handled by animals were Laboratory analyses sampled directly beneath selected seed parents both before and after logging. The seed parents can be described as bio- Samples were brought into the plant genetic laboratory of logical ‘pollen traps’, and are referred to hereafter as ‘trap EMBRAPA-CENARGEN (Brazilian Agricultural Research trees’. We began by regularly visiting a number of poten- Corporation — Centre for Genetic Resources and Bio- tial trap trees, separated by a range of different intertree technology) in Brazilia, Federal District, Brazil. The six distances, ranging from trap trees that are close enough to microsatellite (SSR) loci used in this study are described by receive pollen from the same source trees, to those far apart Dayanandan et al. (1999) for loci cg05 and cg07, and by and likely to receive pollen from different source trees. Vinson et al. (2005) for loci cg01, cg06, cg16 and cg17. New Trees that produced at least 10 fruits in both the pre-logging reverse primers were designed for cg05 (5′-GAGGAT- and the post-logging period were selected as trap trees CTTGTACGTTGGC-3′) and cg07 (5′-CTGTTCGTTGA- (Fig. 1). The distance among the 21 trap trees varied between AGAACTTGG-3′) in order to obtain shorter polymerase 16 m and 1883 m, averaging at 832 m. We sampled seeds chain reaction (PCR) products. DNA extraction, PCR con- from the same 21 unlogged trap trees (before and after ditions and electrophoresis were carried out as described logging) aiming to minimize the effects of local spatial in Cloutier et al. in press). To rule out incorrect assignment variation so as to better examine temporal variation in of progeny to the mother trees, we compared trap tree the population. A separate genetic analysis conducted on and seed genotypes to detect and remove mismatched C. guianensis multiseeded fruits collected in 2004 at the individuals from the data set (thus 0.9% of seed genotypes study site revealed that seeds within individual fruits are were removed). significantly more likely to share the same father than are seeds taken from different fruits of the same mother tree Genetic diversity analyses (D. Cloutier, unpublished data), suggesting pollen codis- persal from the same male parent(s) to single flowers. Standard genetic diversity parameters were estimated for Thus, to sample as many independent pollination events the seeds collected before and after logging; for the seedling, as possible (a requirement for obtaining good estimates juvenile and adult cohort; and for the harvested and of pollen dispersal; Hardy et al. 2004), we analysed a single nonharvested adult trees larger than the size harvest limit. seed per fruit sampled. An average of 37 fruits were sampled The number of alleles per locus (N ), the allelic richness all © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd GENETIC IMPACTS OF AMAZONIAN TREE HARVESTING 5 (R) (i.e. the number of alleles independent of sample we analysed pollen dispersal with the twogener algorithm size, allowing comparisons between samples of different developed by Smouse et al. (2001), using an implementation sample sizes; El Mousadik & Petit 1996), the unbiased gene of the algorithm programmed by F. Austerlitz (Université diversity (H ), the observed heterozygosity (H ), and the Paris-Sud, France). The statistic Φ , which measures E O FT fixation index estimates (F ) were calculated with the the differentiation in allelic frequencies among the pollen IS program fstat version 2.9.3 (Goudet 1995). Exact tests for clouds sampled by different seed parents (i.e. trap trees, fixation index estimates (F ) (Weir & Cockerham 1984) whose spatial coordinates are known), was computed to IS and for linkage disequilibrium among loci were conducted estimate the pollen dispersal parameters. A formal relation- with the program genepop (Raymond & Rousset 1995). ship between Φ and pollen dispersal parameters has FT been derived for different pollen flow models (Austerlitz & Smouse 2001). To estimate pollen dispersal parameters Mating system analyses either a global Φ , based on average pollen pool differen- FT Maximum-likelihood estimates of single-locus (t) and tiation and average physical distance over all sampled s multilocus (t ) outcrossing rates were calculated following mothers, or alternatively, a pairwise Φ , based on pollen m FT the mixed-mating model (Ritland 2002) using the seed pool differentiation and physical distance between each together with their known maternal genotypes. If mating pair of sampled mother can be used — both are employed among relatives (biparental inbreeding) occurs, t should below, and referred to, respectively, as ‘global’ and m be higher than t, the difference giving a minimum ‘pairwise’ estimators. s estimate of apparent selfing due to biparental inbreeding. We estimated the effective number of pollen donors con- Multilocus correlation of paternity (r ) was calculated to tributing to the progeny of the average mother tree (N ), p ep estimate the probability that two outcrossed progenies which is derived directly from the global estimate of Φ as FT drawn at random from the same seed family are full-sibs. N = 1/2Φ (Smouse et al. 2001), as well as the parameter ep FT Mating system analyses were conducted using the program δ (i.e. the average distance of realized pollen dispersal, mltr version 3.1 (Ritland 2002). The default settings were which equals σ(π/2)0.5 for the normal distribution) used, except that the parental fixation index was con- (Austerlitz & Smouse 2001; Austerlitz & Smouse 2002). The strained to 0.029 (i.e. the value observed in the adult global Φ (Austerlitz & Smouse 2001), and the pairwise FT cohort). We assumed no apomixes and no seed mortality Φ (Austerlitz & Smouse 2002) were used to estimate δ by FT due to inbreeding prior to the assay. Standard errors for t, assuming an underlying density of reproducing trees (d) s t and r were estimated from 1000 bootstraps at the level and a pollen dispersal model (i.e. normal, exponential or m p of the seed families. power exponential). Because N/N ratios in plant popula- e Two different mating system analyses were performed. tions may vary from values close to one, to as low as 1/10 First, we contrasted the seed progenies sampled before (e.g. due to varying fertility and nonsynchronous flowering) logging (n = 391) vs. those sampled after logging (n = 390), (Frankham 1995), we estimated δ from global Φ assum- FT to assess the impact of logging on mating system. Second, ing an upper bound estimate of the density of reproducing we divided the seed progenies into four temporal groups trees (d ) (i.e. the global density of conspecific trees above max in order to assess whether there was seasonal variation in 30 cm d.b.h. in the field), and a lower bound estimate of d mating system. The four temporal groups were composed (d = d /10). The pairwise Φ was used also to jointly min max FT of: (i) seed collected in April–June 2003 (n = 243) and January– infer δ and d. To assess the 95% confidence intervals of the February 2004 (n = 148), thus resulting from pre-logging pollen dispersal parameters, we created bootstrap data sets pollination; and (ii) seed collected in May–July 2004 (n = by resampling the 21 trap tree families. Because of the 158) and January–February 2005 (n = 232), thus resulting prohibitive amount of computer processing time required from post-logging pollination. Assuming that about 4 months for the calculations, we used 200 bootstrap samples (this separate pollination from fruit release, it is possible that required 52 Apple Macintosh G4 CPUs over 2 months). seeds collected from a few trap trees in January–February While the ‘power exponential’ model has been suggested 2004 were actually resulting from pollination that occurred to approximate the shape of the dispersal curve (Austerlitz during the logging because a portion of the study site et al. 2004), this model repeatedly failed to converge to began to be logged at that period. realistic values using the original and bootstrapped data sets and therefore only results based on the ‘normal’ and ‘exponential’ dispersal model are reported here. Pollen dispersal distance analyses and effective number of pollen donors Detection of bottlenecks in pollen pool allele frequencies Genotyping of every potential pollen donor, as in a paternity analysis, was not practical in this study, since trees were To assess the magnitude of logging-mediated paternal continuously distributed over a very large area. Instead, bottlenecks, we estimated the extent of differentiation (or © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd 6 D. CLOUTIER ET AL. divergence) between the estimates of pollen allele frequencies dispersal distance and D is the effective density of individuals. and of adult tree allele frequencies, using the estimator F The product 4πDσ2 is related to Wright’s neighbourhood k (Pollak 1983; Waples 1989). This statistic is related to the size, allowing us to estimate Nb = −−−−(1 − F[1])/b (apply- 1 ij k ‘variance effective population size’ of the pollen donor ing the correction proposed by Vekemans & Hardy 2004), pool. We estimated, separately, pollen pool allele frequencies where F[1] is the average kinship coefficient between ij before and after logging using mltr version 3.1 (Ritland neighbouring individuals. For the second approach, we 2002), while the actual allele frequencies of the 199 adults followed Epperson (2005), who described a nearly linear sampled were used for the adult allele frequencies. A 95% relationship between the natural logarithm of Nb and confidence interval on F was derived from the chi-square the finest scale spatial autocorrelation of individual allele k distribution following Waples (1989). frequencies, as measured by Moran’s I relationship co- efficient (R ) between neighbours. This allows one to ij estimate Nb = exp[(0.544 − R[1])/0.102], where R[1] is the Spatial genetic structure analyses 2 ij ij average Moran’s I coefficient between neighbour individuals. To take advantage of different approaches for assessing the Both approaches require one to define what ‘neighbour’ consequences of historical gene dispersal, spatial genetic implies, i.e. what is the maximal distance between two structure (SGS) was estimated using two measures of genetic ‘neighbouring’ individuals. Following Epperson (2003), similarity. First, we used a relative kinship coefficient (F ) we defined this distance relative to the average distance ij described in Loiselle et al. (1995). F = (Q − Q )/1 − Q , between contiguous individuals, as if they were uniformly ij ij m m where Q is the probability of identity in state for random distributed on a lattice model, including pairs of individuals ij genes from individuals i and j, and Q is the probability of separated by rook and bishop moves, and using 1.5 times m identity in state for genes from random individuals in the the square root of the inverse of adult density, which for population. Second, we used the relationship coefficient this C. guianensis population gave as neighbours, indi- R , computed as the correlation between individual allele viduals separated by less than 95 m. F[1], R[1], and b values ij ij ij k frequencies (i.e. Moran’s I autocorrelation statistic), as were tested against the null hypothesis of random distri- described in Hardy & Vekemans (1999). We used the pro- bution of genotypes by randomizing the spatial positions gram spagedi version 1.2 (Hardy & Vekemans 2002) for 1000 times. Standard errors for F[1], R[1] and b were ij ij k all SGS analyses. obtained by jackknifing genetic loci. The 95% confidence To visualize SGS, F values were plotted between interval for Nb was estimated as −(1 − F[1])/(b ± 2SE ), ij 1 ij k bk individuals according to the physical distance separating and for Nb as exp[(0.544 − (R[1] ± 2SER[1]))/0.102]. 2 ij ij them, but since using all pairwise F values on a graph To estimate historical gene dispersal distance (σ) from ij would obscure the overall trend of the SGS, we presented SGS, we first estimated σ = (b/4 πD)(0.5), using the census k average pairwise F for pairs of individuals falling into a density of adults as an estimate of D. But the linear relation- ij set of specific distance classes. The upper limit of each dis- ship of F over ln(distance) holds best within a restricted ij tance class was 3.6, 7.6, 14.4, 24, 38, 67, 109, 191, 362, 612, range between 1 and 20 times σ (Rousset 2000), therefore 1014, 1536 m for the seedlings; 30, 50, 70, 91, 118, 156, 211, we applied an iterative procedure to estimate σ using 303, 450, 722, 1213, 2745 m for the juveniles; and 30, 46, 61, restricted b implemented in spagedi (Hardy & Vekemans k 80, 102, 135, 175, 222, 273, 332, 397, 464, 542, 635, 740, 864, 2002). 1020, 1220, 1579, 2945 m for the adults. The distance classes were chosen to facilitate the visualization of fine-scale Results SGS on a logarithmic graph, while ensuring a minimum of 30 pairs in each distance class. To test for significant Genetic diversity and heterozygosity deviations from random SGS, observed values for each distance class were compared to the 95% confidence interval Allele frequencies for both the adults and the seed sampled derived from 1000 permutations of individuals among at the six microsatellite loci are available in the Supplement- locations. ary material. The total number of alleles for all samples was To estimate historical gene dispersal from SGS in terms 4, 12, 13, 9, 17 and 12, respectively, at the loci cg01, cg06, that are equivalent to Wright’s neighbourhood size (Nb), cg05, cg07, cg16 and cg17. Gene diversity (H ) was high at E we used two approaches. For the first approach, we fol- ∼0.8 for cg06, cg05, cg16 and cg17, while lower H was E lowed Rousset (2000), who showed that under isolation by found for cg07 (∼0.6) and cg01 (∼0.4). No linkage disequili- distance in a two- dimensional space, F is expected to be brium was detected among the loci. ij approximately linearly related to the logarithm of distance Genetic diversity statistics averaged over loci for the between individuals, and therefore the slope of the regres- different samples are shown in Table 1. There were minor sion (b) of F values over ln(distance) estimates 1/4πDσ2, differences in the mean number of alleles per locus (N ) k ij all where σ2 is the second moment of the parent–progeny among samples, apparently depending on the sample size. © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd GENETIC IMPACTS OF AMAZONIAN TREE HARVESTING 7 Table 1 Genetic diversity averaged over six microsatellite loci in cohorts of Carapa guianensis Cohort N N R (SE) H (SE) H (SE) F (SE) all O E IS Pre-harvest seeds 390 10.0 7.7 (2.8)§ 0.68 (0.21) 0.69 (0.19) 0.025 (0.022)* Post-harvest seeds 391 10.6 7.6 (2.8) 0.67 (0.18) 0.70 (0.17) 0.049 (0.013)* Seedlings 84 8.8 7.6 (2.9) 0.65 (0.19) 0.71 (0.19) 0.079 (0.034)* Juveniles 82 9.0 7.9 (3.3) 0.68 (0.17) 0.72 (0.15) 0.048 (0.023)* Adults 199 10.3 8.1 (2.9) 0.69 (0.19) 0.71 (0.18) 0.029 (0.020)* Harvested† 46 8.3 8.0 (3.2) 0.67 (0.24) 0.71 (0.18) 0.056 (0.042)NS Unharvested‡ 49 8.3 7.7 (2.9) 0.65 (0.18) 0.70 (0.18) 0.059 (0.033)* N, sample size; N , number of alleles; R, allelic richness based on a minimum sample size of 33 diploid individuals; H , observed all O heterozygosity; H , gene diversity; F , fixation index (test of significance of the fixation index *: P < 0.05 and NS: P > 0.05). E IS †Harvested adults > 53 cm d.b.h. ‡Adults > 53 cm d.b.h. left unharvested. §Standard error over loci in parentheses. Table 2 Mating system of Carapa guianensis as estimated from presence of a small amount of selfing in the population. seeds collected before and after selective logging The level of mating between relatives (biparental inbreeding) was not significant, as measured by the lack of difference Mating system parameter Before logging After logging between single and multilocus outcrossing rates (t − t). m s The correlation of outcrossed paternity (r ) was low, but t (SE) 0.939 (0.027) 0.927 (0.033) p m significantly different than 0, suggesting that most seed t (SE) 0.924 (0.028) 0.899 (0.041) s t – t (SE) 0.015 (0.015) 0.028 (0.018) within families had a different father (i.e. half-sibs). Overall, m s r (SE) 0.050 (0.014) 0.054 (0.013) no significant differences in mating system parameters were p detected between pre- and post-harvest seed progenies. tm, multilocus outcrossing rate; ts, single-locus outcrossing rate; Seasonal variation in mating system is summarized in rp, multilocus correlation of outcrossed paternity. Fig. 2. All parameters t , t − t and r showed little variation m m s p over the 22-month period. However, slight nonsignificant changes were measured in seeds collected around January Using the measure of allelic richness (R), there is a modest 2004 for r [0.112 (0.139)] and May–June 2004 for t [0.859 p m but nonsignificant trend toward higher R values in oldest (0.122)]; both samples corresponding to periods where cohorts (adults and juveniles), and lower R values in the fruit production in the population was relatively low. youngest cohorts (seeds and seedlings). Trees above harvest- Overall, no significant temporal variation was detected size limit that were left unlogged had a slightly lower R over the time period investigated despite potential for than logged trees, and consequently, the post-harvest seeds variation in population phenology, pollen production, and appeared to have a lower R than pre-harvest seeds; how- pollinator composition. ever, in all cases the differences were not significant. Gene diversity (H ) was similar in all samples indicating that E Pollen dispersal common alleles remain at about the same frequencies regardless of logging or age effects. The observed hetero- Global pollen pool differentiation (Φ ) was estimated at FT zygosities (H ) were slightly lower than H in all samples, 0.044 before logging, with a 95% confidence interval (95% O E and fixation indexes (F ) were all low, but significantly CI) of 0.024–0.063, and at 0.054 after logging, with a 95% CI IS greater than zero at the 5% level, except for the harvested of 0.033–0.070. These values translate into an effective trees. There were no significant differences in F values number of pollen donors (N ) of 11.3 (95% CI: 7.9–20.8) IS ep between the samples when the 95% confidence intervals before logging, and of 9.3 (95% CI: 7.1–15.2) after logging. derived from a bootstrap among loci were compared. These differences before vs. after logging are not significant. Pollen dispersal distance analyses are summarized in Table 3. The minimum (δ ) and maximum (δ ) esti- Mating system min max mates of average distance of pollen dispersal using global Mating system estimates are shown in Table 2. Outcrossing Φ were 75 m and 265 m before logging, when both the FT rates (t ) were high both before (0.94) and after logging normal and the exponential model were considered, while m (0.93), but were significantly less than 1.00, suggesting the the estimates were similar after logging, with a minimum © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd 8 D. CLOUTIER ET AL. Fig. 2 Seasonal variation in the mating system of Carapa guianensis. Circles, squares and lozenges symbols represent, respec- tively, the outcrossing rate (t ), the level of m biparental inbreeding (t − t) and the m s correlation of paternity (r). Filled symbols p indicate values estimated from trees that flowered after harvesting. Bars correspond to the standard error of the estimates. Table 3 Pollen dispersal analyses based on estimates of pollen pool differentiation before and after selective logging in Carapa guianensis Dispersal Parameter Before logging* After logging† Type of estimate model estimated (95% CI) (95% CI) Estimation of δ using global Φ Normal δ § 75 (63–102)‡ 76 (67–98) FT min δ ¶ 237 (206–329) 240 (214–311) max Exponential δ 85 (71–115) 86 (75–110) min δ 265 (230–369) 268 (240–348) max Joint estimation of d and δ using pairwise Φ Normal d 0.30 (0.06–3.01) 0.75 (0.19–2.08) FT δ 204 (2–358) 123 (5–230) Exponential d 0.25 (0.04–9.36) 0.55 (0.10–581) δ 252 (18–534) 159 (5–306) Φ , pollen pool differentiation; δ, average distance of pollen dispersal (m); d, adult effective density (trees/ha). FT *Adult field density > 30 cm d.b.h. before logging, 2.5 trees/ha. †Adult field density > 30 cm d.b.h. after logging, 2.0 trees/ha. ‡Estimated 95% confidence intervals based on bootstrapping the seed families. §Assuming adult field density (d ). max ¶Assuming adult field density × 1/10 (d ) (see methods for details). min of 76 m and a maximum of 268 m. Before logging, the joint Detection of bottlenecks in pollen pool allele frequencies estimation of d and δ using pairwise Φ gave an effective FT density of pollen donors (d) of 0.30 and 0.25 trees/ha, and The differentiation in allele frequencies (F) between the k an average pollen dispersal distance (δ) of 204 and 252 m, adult and the pre-harvest seeds was 0.0084, with a 95% for the normal and exponential model, respectively. After confidence interval of (0.0056–0.0118), while the different- logging, the estimates were slightly different, as d increased iation between adult and post-harvest seeds was 0.0066 to 0.75 and 0.55 trees/ha, and δ decreased to 123 m and with a 95% confidence interval of (0.0044–0.0093). Therefore, 159 m, for the normal and exponential model, respectively. no significant differences between before and after logging The bootstrap 95% confidence intervals on the parameters were detected. were large, however, and no statistically significant differ- ences were detected between pre- and post-harvest periods. Spatial genetic structure Minimum and maximum estimates derived from the pair- wise Φ are not shown since they were very similar to the The SGS in the different cohorts is shown in Fig. 3. The FT estimates from the global Φ reported in Table 3. estimates of average F decreased with the logarithm of FT ij © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd GENETIC IMPACTS OF AMAZONIAN TREE HARVESTING 9 Fig. 3 Mean pairwise kinship coefficient (F) between individuals vs. interindividual ij distance for the seedling and juvenile cohorts (upper), and the pre- and post- harvest adult population (lower). Filled symbols show values departing significantly (α = 5%) from a random spatial distribution of genotypes. Table 4 Spatial genetic structure averaged over loci and Wright’s neighbourhood size in cohorts of Carapa guianensis Cohort N b (SE) F[1] (SE) Nb (95% CI) R[1] (SE) Nb (95% CI) k ij 1 ij 2 Seedling 3486 −0.0179* 0.0480* 53 0.0698* 104 (0.0025) (0.0078) (42–74) (0.0110) (84–130) Juvenile 3321 −0.0047 0.0156 209 0.0258* 161 (0.0031) (0.0104) (90–∞) (0.0160) (118–∞) Adult 19701 −0.0045* 0.0070 221 0.0175 174 Pre-harvest (0.0024) (0.0080) (107–∞) (0.0134) (134–∞) Adult 9591 −0.0063* 0.0041 0.0154 — Post-harvest (0.0037) (0.0128) —† (0.0243) N, total number of pairs; b, slope of the regression of F values over the natural logarithm of the spatial distance; F[1], average kinship k ij ij coefficient between neighbour individuals; R[1], average Moran’s I index between neighbour individuals; Nb, neighbourhood size ij 1 estimated as −(1 − F[1])/b; Nb, neighbourhood size estimated as exp[(0.544 − R[1])/0.102]; *, type I error rate < 0.05 associated with the ij k 2 hypothesis of no isolation by distance (see methods for details). †Not estimated since the pre-harvest estimates are accurate regarding historical dispersal-drift balance. distance in all cohorts, as expected under a model of different according to the approach used, but agree in indi- isolation by distance. For the seedling cohort, average F cating large values for Nb. ij values at distances from 2 to 60 m departed significantly The historical gene dispersal distance estimate was from the hypothesis of absence of SGS, while few or no σ = 238 m, assuming D = 2.5 trees/ha and unrestricted b. k values exhibited significant SGS for the adult and juvenile Using the iterative procedure that estimates σ within a cohorts. Moreover, the average F values for the seedling restricted range, we obtained σ = 273 m. This latter value ij cohort at short distances were higher than observed in the should be considered as the lower bound estimate for his- juvenile and adult cohorts. The overall pattern of SGS for torical gene dispersal distance because D is likely to be pre-harvest and post-harvest adult population was nearly lower than the census density of 2.5 individuals/ha (i.e. the indistinguishable, indicating that adult tree SGS is not iterative procedure applied with D < 0.8 led to an estimate significantly impacted by selective logging. of gene dispersal that is infinite). Estimates of SGS and neighbourhood sizes (Nb) are presented in Table 4. The global regression slope (b) of the k Discussion adult and the seedling cohorts is significantly different from a random spatial structure, confirming the presence Impact of tree harvesting on genetic diversity and of isolation by distance. Significant positive genetic simi- heterozygosity larity between neighbours separated by less than 95 m was detected in the adult and seedling cohorts, but not in the Following logging in this population, no evidence is seen juveniles. The seedling cohort gave estimates of Nb that are for immediate losses of genetic variation in either the adult significantly smaller than the adult and juvenile cohorts trees or their progeny, or for increases in fixation indexes. (Table 4). The estimates of neighbourhood size are slightly This may be partially explained by the fact that Carapa © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd 10 D. CLOUTIER ET AL. guianensis is relatively common in the forest, that its calculated with twogener using seeds obtained before reproductive phenology is asynchronous, that the population logging was 0.044 (0.010), while after logging increased has never been previously logged, and that the proportion (nonsignificantly) to 0.054 (0.010) (Table 3). Multilocus r p of trees harvested during the logging operation was not estimates, when t is constrained to 1.0 in mltr, were 0.067 m sufficient to cause a detectable decline in genetic diversity (0.094) before logging and increased to 0.104 (0.161) after or increase in inbreeding. Perhaps more importantly, we logging. Both approaches yield values that point toward a have measured genetic variation using a small number of slight, though nonsignificant decrease in the effective neutral SSR markers. It is possible that selective logging, number of pollen donors (N ) following logging. On the ep because it targets the largest and healthiest trees in the other hand, using pairwise Φ with twogener to jointly FT population, also reduced the diversity of genes underlying estimate d and δ, we obtained (for the normal model) these characteristics, and is therefore dysgenic, leading to d = 0.3 trees/ha and δ = 204 m before logging, and d = 0.75 erosion of the species’ genetic resources (e.g. quantitative trees/ha and δ = 123 m after logging, a difference that trait genes for rapid growth, or qualitative trait genes for suggests a (nonsignificant) increase in the number of pollen disease- or drought-resistance, etc.). As has been pointed donors in a more restricted area following logging. The out elsewhere, even when surveys of neutral genetic reduced differentiation in allele frequencies between adult diversity indicate the presence of substantial polymorphism, and progeny (F) observed after logging also points to an k genetic diversity for loci important to long-term survival increase in the number of pollen donors. A recent study and reproduction may still be in decline (Lowe et al. 2005). on another insect-pollinated tree species suggested that From earlier studies, we might have expected to detect a selective logging may not negatively affect pollen movement, decrease in F with age, indicating the presence of inbreeding and has fewer genetic consequences relative to clear-cut IS depression. Doligez & Joly (1997) found significant excess logging (Sork et al. 2005). Overall, it appears reasonable to homozygosity in seeds and excess heterozygosity in adults conclude that our results do not provide evidence for in Carapa procera, which they explained as due to selection strong genetic consequences of selective logging on C. in favour of heterozygotes. There is no such strong trend in guianensis. our study, however, as the estimated multilocus F was 0.025 There are a number of potential sources of error when IS for the pre-logging seed and 0.049 for the post-logging estimating pollen dispersal parameters. It is possible that seed, values that are similar to the seedling (0.079), juvenile we could have slightly underestimated average pollen (0.048) and adult (0.029) F . Lack of evidence for signifi- dispersal distances, due to the limited number of trap trees IS cant inbreeding depression in this population could be available, and the resulting restricted number of pollen explained by a low level of inbreeding, potentially due to dispersal models that could be applied to the data (Auster- the presence of a partial prezygotic incompatibility system litz et al. 2004). Another potential source of error is the (M. Maues, personal communication). value of tree density assumed. In this study, we defined reproductive individuals as trees larger than 30 cm d.b.h. However, it is likely that some trees above this size were Impact of tree harvesting on inbreeding, pollen dispersal also not reproductive, and that some trees below this size distance and effective number of pollen donors did, in fact, contribute to reproduction. Moreover, some A reduction in density of adult trees following logging reproductive trees are likely to be more important con- could, in theory, change levels of inbreeding and gene tributors than others due to interplant variation in flower flow, but we found no statistically significant differences in production. The reproductive phenology of C. guianensis the genetic parameters assessed before vs. after the tree further complicates the matter as trees can be seen flowering harvest. Estimates of the mating system show that there are intermittently. In this study, we tried to take into account low levels of inbreeding in this population, and that these factors by estimating minimum and maximum levels inbreeding levels remained unchanged after the harvest. of pollen dispersal, based on maximum and minimum esti- The lack of a logging effect on the mating system contrasts mates of effective tree density, but it is nevertheless possible with the results of Doligez & Joly (1997) on C. procera, but that the actual pollen dispersal distance is higher than the could be explained by geographical or biological differences estimated intervals. among the two populations studied. We found no biparental At least two ecological factors may help account for the inbreeding in this population (Table 1), a result that is in stability of gene dispersal and inbreeding in the face of accord with the weak spatial genetic structure observed selective logging. First, C. guianensis trees are relatively among adult trees (Table 4). Estimates of global pollen pool abundant at the study site and they are reproductively differentiation revealed that pollen dispersal is extensive, mature at a relatively small d.b.h. compared to other tree that there are a large effective number of pollen donors, species. Thus, as long as the size harvest limit (53 cm d.b.h. and that pollen dispersal parameters did not appear to be in this logging experiment) is substantially higher than the strongly influenced by logging. For instance, the global Φ size at which trees become reproductive, the impact on FT © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd