Mating systems in fragmented landscape: the case of a tropical Caesalpiniaceae species S Chauvet 1,2, C Dutech 3, N Raufaste 4 and BE Giles2 1

Muséum National d'Histoire Naturelle, Département Ecologie et Gestion

de la Biodiversité, UMR 8571 CNRS-MNHN, 4 av. du Petit Château, F91800 Brunoy, France 2

Department of Ecology and Environmental Science, Umeå University, SE-

90187 Umeå, Sweden. 3

Laboratoire de Génétique et d’Ecologie Moléculaire, Cirad-forêt, BP 701,

97387 Kourou cedex, France. and: INRA, UMR Biodiversité, Gènes et Ecosystèmes, Domaine de la Grande Ferrade, BP81, F-33883 Villenave d’Ornon, France. 4

Laboratoire Génômes, Populations et Interactions, UMR 5000, cc 63,

Université Montpellier II, Place Eugène Bataillon, 34095 Montpellier Cedex, France. Corresponding author: Stéphanie Chauvet, Department of Ecology and Environmental Science, Umeå University, SE-90187 Umeå, Sweden. Phone: +46 90 786 71 35; Fax: +46 90 786 67 05; [email protected] Keywords: forest fragmentation, mating system, microsatellites, neotropical rainforest, pollen dispersal, Vouacapoua americana. Running title: Mating and gene flow in a fragmented forest Main text: 4136 words.

1

Abstract

2

Tropical forest tree species are predominantly outcrossed. However,

3

changes in mating systems can occur in response to forest fragmentation, as

4

reductions in population size and restricted gene flow can increase rates of

5

selfing and mating between relatives. In this study, the mating system of

6

Vouacapoua americana (Caesalpiniaceae) was examined using five

7

polymorphic microsatellite markers in two populations from a recently

8

fragmented landscape: one located in a forest fragment, and another in

9

continuous forest. Both populations had very low selfing rates (< 5 %),

10

suggesting that mating systems were little affected by the recent

11

fragmentation. These populations were also genetically different in both

12

adult and offspring stages and in pollen clouds, and may therefore be

13

considered as independent reproductive units. We propose that this

14

restricted gene flow could make Vouacapoua americana strongly vulnerable

15

to long-term effects of fragmentation and to dramatic decrease in number of

16

reproductive individuals.

2

17

Introduction

18

In tropical forests, mating systems of tree species are predominantly

19

outcrossed, with selfing rate often lower than 20 percent in most studied

20

species (Hamrick and Murawski, 1990; Bawa, 1992). However, mating

21

systems can be influenced by several factors, including conspecific adult

22

density, composition and behavior of the pollinator community, and forest

23

fragmentation (Bawa, 1990; Murawski et al, 1994; Doligez and Joly, 1997;

24

Aldrich et al, 1998). In fragmented forests, smaller effective population

25

sizes and limited gene flow may dramatically increase rates of selfing and /

26

or matings between relatives, leading to an increase of consanguinity across

27

generations and consequently a loss of fitness in progeny (Charlesworth and

28

Charlesworth, 1987). Increased inbreeding can also be responsible for a

29

decrease in fruit set and / or a reduction in seed viability (Wiens, 1984;

30

Jennersten, 1988; Aizen and Feinsinger, 1994; Ghazoul et al, 1998).

31

Unfortunately, the effects of such environmental changes on mating systems

32

remain unknown for most tropical tree species, and further investigations

33

are needed. The expected increased spatial isolation and reduced effective size of

34 35

fragmented populations are expected to induce an erosion of genetic

36

variation within populations and an increase in genetic divergence among

37

populations, through founder effects, genetic drift, inbreeding and reduced

38

gene flow (Ellstrand and Elam, 1993; Young et al, 1996). Among the

3

39

proximal factors responsible for reduced gene flow, the alteration of pollen-

40

vector interactions in fragmented habitats is of particular importance

41

(Rathcke and Jules, 1993). Biotic pollination is ubiquitous in tropical

42

lowland rainforests, with over 98 % of forest tree species depending on

43

animals, especially insects, for pollen dispersal (Bawa, 1990). The

44

dominance of insect pollination suggests that tropical tree species can be

45

particularly vulnerable to disruption of pollen-vector associations in

46

fragmented habitats (Powell and Powell, 1987; Nason et al, 1997). Indeed,

47

research in Amazonian fragments indicated that clearings as little as 100

48

meters wide can diminish inter-fragment movements for some insects, and

49

that both abundance and diversity of insect pollinators decreases in

50

fragmented landscapes (Vinson et al, 1993; Didham et al, 1996). However,

51

for some tropical tree species, pollen flow via insects has been reported over

52

long-distances, such as for Dipterocarpaceae in Thailand (Konuma et al,

53

2000), and Meliaceae in Honduras (White et al, 2002). It is therefore of

54

primary importance to assess the frequency of pollen exchanges between

55

tree populations in fragmented landscapes for different species, in order to

56

evaluate the general effects of fragmentation on gene flow. In this study, we focused on the mating system of a neotropical

57 58

forest tree species, Vouacapoua americana (Caesalpiniaceae), in a

59

fragmented landscape in French Guiana. This species is a canopy tree with

60

hermaphroditic flowers pollinated by small insects and characterized by

4

61

limited effective pollen flow (Dutech et al, 2002). It is therefore likely to be

62

affected by disruption of its pollen-vector interactions in fragmented

63

habitats. We analyzed progeny arrays of V. americana for two populations:

64

one in a fragmented forest remnant, and another in continuous forest. Five

65

polymorphic microsatellite loci were analyzed in order to: (i) describe the

66

mating system of V. americana in both populations and determine whether

67

the selfing rate is enhanced in the isolated population; (ii) assess whether

68

pollen flow is more restricted in the isolated population than in continuous

69

forest due to alterations in pollen-vector interactions; and (iii) examine the

70

extent of pollen movement between the two populations.

71 72

Materials and methods

73

Species studied

74

Vouacapoua americana (Caesalpiniaceae) is a canopy tree species present

75

in mature tropical rainforest from Amazonian Brazil to French Guiana and

76

Suriname. The distribution of this species is strongly aggregated, and its

77

density within clusters is about ten individuals per hectare (diameter at

78

breast height [dbh] > 10 cm) (Forget et al, 1999). It is a hermaphroditic

79

species with small yellow fragrant flowers measuring a few millimeters

80

long, suggesting that insects such as Hymenoptera or Thysanoptera (e.g.

81

Thrips) are involved in pollination (Bawa, 1990; Dutech et al, 2000b).

82

Reproductive phenology is discontinuous and irregular across years, as is

5

83

characteristic for “mast-fruiting” species (sensu Janzen, 1974). In French

84

Guiana, blooming and fruiting occur respectively in February-March and

85

April-June, and is synchronous over the landscape. Minimum diameter

86

observed for reproducing individuals is 22 cm dbh (Chauvet, 2001). The

87

fruit is a pod containing one (rarely two) large seed (ca. 30 g), which is

88

exclusively dispersed by scatterhoarding rodents (Forget, 1990). Dispersal

89

distances are generally short (< 10 meters from seed sources) with

90

occasional dispersal up to 30 meters. However evidences for some long-

91

distance dispersal events (> 150 m) have been reported (S. Traissac, pers.

92

comm.).

93 94

Study site

95

Our study took place in French Guiana, at Saint-Eugène (4°51’ N, 53°04’

96

W; Claessens et al, 2002). The study sites are located on the Sinnamary

97

River, above a hydroelectric dam built in the early 1990’s and flooded

98

during 1994-95. This flooding caused the forest to be fragmented into a

99

large number of islands, of various shapes and sizes (from 0.1 to 67 ha), and

100

degrees of isolation (ca. 20 to 500 m). Because of the clustered distribution

101

of V. americana even before flooding, this species is absent from some

102

fragments, and when present, its population sizes vary. We studied two

103

populations, one where V. americana is widely extended over a peninsula of

104

continuous forest, and another on an island where the species is more

6

105

restricted (Figure 1). The island encompasses two populations, likely to be

106

the remnants of two former aggregates. The first population is located at the

107

north-eastern side (N = 25 trees), and the second one at the south-western

108

side of the island (N = 30 trees), resulting in a total of only 55 trees (with

109

dbh > 20 cm) for a 28-ha island. In this study, we focused on the north-

110

eastern population, distributed over 4 hectares with a local density of about

111

6 individuals per hectare. The experiment was conducted over an equivalent

112

4-ha area in the continuous forest that in contrast exhibits 40 trees (with dbh

113

> 20 cm), i.e. 10 individuals per hectare (Figure 1). In addition to this

114

contrasting local density, these two populations also differ in the diameter

115

distribution of the V. americana trees greater than 20 cm dbh, as 10 trees out

116

of 25 on the island have dbh ranging from 20 to 25 cm, while only one out

117

of 40 in the continuous forest falls in this range, suggesting that the island

118

population is a more recent stand than the one in continuous forest. The two populations under study are separated from each other by

119 120

about 500m of water (Figure 1), and are referred to as populations of forest

121

F and island I, respectively.

122 123

Sampling

124

In the island I population, trees of small diameter (i.e. 20 < dbh < 25 cm)

125

only produced a limited number of fruits (S. Chauvet, pers. obs.). In

126

consequence, we only studied the13 trees at island I that produced at least

7

127

100 fruits, whereas 20 trees were randomly chosen within the forest F

128

population. In order to assess the effect of fragmentation on mating system,

129

we collected seeds produced from mating that occurred four years after

130

fragmentation, namely in May 1998. For the 33 trees, 60 seeds per tree were

131

collected from the ground, and germinated in a greenhouse. Leaf samples

132

were collected from 10 randomly selected seedlings per tree, giving a total

133

sample size of 330. All leaves collected were stored fresh at –80°C for later

134

genetic analysis.

135 136

DNA extraction and microsatellite analysis

137

Total DNA was extracted in liquid nitrogen and a CTAB buffer, following

138

the protocol outlined in Dutech et al (2000b). Individual genotypes of

139

progenies were assessed using five polymorphic microsatellite loci isolated

140

from the nuclear genome of V. americana (Wac1, Wac5, Wac7, Wac10, and

141

Wac13; Dutech et al, 2000a). Amplification of DNA was performed using a

142

Polymerase Chain Reaction (PCR) method following Dutech et al (2000a).

143

PCR products were separated by electrophoresis in denaturing

144

polyacrylamide sequencing gels and stained with silver nitrate (Streiff et al,

145

1998).

146 147

Data analysis

148

Genetic diversity within and among populations ⎯ The within-population

8

149

diversity was estimated using the GENETIX software package, version 4.01

150

(Belkhir et al, 2000). For both populations, we analyzed offspring diversity

151

using one offspring randomly chosen per family (N = 33), as well as

152

maternal tree diversity (arising from recruitment before fragmentation) (N =

153

33). Maternal genotypes were inferred using the MLTR software (version

154

0.9) and the genotype of the 10 offspring per family. According to Brown

155

and Allard (1970), the size of the progeny array utilized here is sufficient to

156

ascertain maternal genotypes to about 95 % certainty in mainly outcrossing

157

populations. Observed heterozygosity (Ho), expected heterozygosity (He),

158

and values of Fis were estimated following Weir and Cockerham (1984),

159

and their respective standard errors estimated by a bootstrap method that

160

resamples over individuals (1000 replicates). The absence of population

161

differentiation between island I and forest F was tested using the exact test

162

(Raymond and Rousset, 1995a) available in the GENEPOP package version

163

3.3 (Raymond and Rousset, 1995b). The latter analysis was conducted for

164

both adult and offspring populations.

165 166

Mating system ⎯ The progeny array (N = 330) was analyzed using a mixed

167

mating model (Ritland and Jain, 1981). Single locus (ts) and multilocus (tm)

168

maximum-likelihood estimates of outcrossing rate in progeny were

169

computed using MLTR, and the expectation-maximisation (EM) algorithm.

170

Variances were estimated using a bootstrap method (1000 bootstraps among

9

171

families). The MLTR program also provides two correlated-mating-system

172

parameters (rs and rp) describing the genetic correlations of sibs (Ritland,

173

1989). The first, rs, is the correlation of selfing between sib-pairs (if rs is

174

equal to 1, then sib-pairs are either both selfed or both outcrossed). The

175

second parameter, rp, is the correlation of paternity between two outcrossed

176

sibs, or the proportion of full-sibs among outcrossed sib-pairs (if rp is equal

177

to 1, then offspring of the same family are full-sibs, while rp equal to 0

178

indicates that all sibs have different fathers). MLTR is designed for a

179

maximum of eight alleles per locus. Therefore, the locus Wac1 (which

180

exhibited 10 alleles) was analyzed by pooling the three less frequent alleles

181

(with allelic frequencies from 0.15 to 1.2 %).

182 183

Pollen clouds ⎯ For each of the 330 offspring studied, the haplotype of

184

pollen grains involved in fertilization was deduced by a method comparing

185

offspring and maternal genotypes. For this purpose, we designed a simple

186

application (available from the authors), written in Delphi TM version 4.0

187

(Delphi, 1998), which assigns offspring alleles to maternal and paternal

188

gene pools as follows. When offspring genotype differs from the mother for

189

one allele, then this allele is assigned to the pollen haplotype with a

190

probability equal to 1. In contrast, when both alleles from the offspring are

191

present in the maternal genotype, then each of them is assigned to the pollen

192

haplotype with a probability ½, according to Mendelian segregation. Given

10

193

the exceptionally low rates of selfing observed in our populations (see

194

below), no distinction was made between outcrossed and selfed offspring.

195

No incompatibility between maternal and offspring genotypes was found in

196

the data set. The genetic structure of pollen clouds was studied using two

197 198

methods. We first tested whether the island I and forest F populations had

199

differentiated pollen clouds using likelihood ratio tests. We compared three

200

nested models: M1 (each tree received a differentiated pollen cloud), M2

201

(each population received a differentiated pollen cloud, but there was no

202

differentiation between trees within-populations), and M3 (the pollen cloud

203

was homogeneous for every tree and both populations). These models are

204

nested because they gradually constrain the allele frequencies in putative

205

pollen clouds. If n is the number of alleles over all loci in the whole sample,

206

and ni the number of alleles at locus i, then M1 has:

207

(20+13)*3 over loci (ni -1) = 33*(n-5) degrees of freedom (d.f.), M2 has 2*(n-

208

5) d.f. and M3 has only 1*(n-5) d.f. We computed the likelihood of each

209

model, as follow: ln (L) = 3 over loci [Nhetero * ln (2) + 3 over alleles [Nx * ln (f (x))]]

210 211

where ln (L) is the log-neperian of the likelihood,

212

Nhetero is the observed number of heterozygotes,

213

Nx is the observed number of alleles x,

11

214

f (x) is the frequency of allele x, parameter of the model (its

215

maximum likelihood estimation is the observed frequency of allele

216

x).

217

Nhetero, Nx and f (x) were calculated within pollen clouds, namely for each

218

tree taken separately in M1, for each population taken separately in M2, and

219

for both populations in M3.

220

The likelihood of each model were then compared by a likelihood ratio test

221

(Edwards, 1972), in order to assess whether the increase in likelihood

222

(L(M1)>L(M2)>L(M3)) outbalanced the increase in parameters

223

(d.f.1>d.f.2>d.f.3). For instance, if the simplest model was true (i.e. M3),

224

then 2*ln(L(M2)/L(M3)) should follow a chi-square with (d.f.2-d.f.3)

225

degrees of freedom. Secondly, we tested whether within-population pollen flow was

226 227

spatially restricted, such that the similarity between individual pollen clouds

228

decreased with the distance separating the pairs of pollen clouds (position is

229

given by the maternal tree receiving the pollen cloud). We therefore

230

calculated the Reynold genetic distance and the geographic distance

231

between each possible pair of individual pollen clouds within populations.

232

The correlation between these two matrices of distance was analyzed using

233

Mantel’s test (1967), available in GENETIX version 4.01 (Belkhir et al,

234

2000).

235

12

236

Results

237

Genetic diversity and population genetic structure

238

In both the island and continuous forests, offspring V. americana

239

populations exhibited values of He (i.e. expected heterozygosity) greater

240

than 0.5, while parental values were 0.39 and 0.53 for the island I and forest

241

F populations respectively (Table 1). Both adult and offspring populations

242

were characterized by mean number of alleles higher than 2.6 and a large

243

number of observed heterozygotes (range 0.43 to 0.51). Fis values did not

244

significantly depart from Hardy-Weinberg expectations, which is consistent

245

with the observed mainly outcrossed mating system (see below) (Tables 1

246

and 2). Fst values calculated between island I and forest F over all nuclear

247

loci were 0.146 and 0.085, between the two adult and offspring populations,

248

respectively. Exact tests of genic differentiation indicate that populations

249

were significantly differentiated for both adult and seedling stages (Fisher’s

250

exact test overall loci: P < 0.0001 and P < 0.001, for adult and offspring

251

populations, respectively). In contrast, Fst values calculated between

252

offspring and adults within populations were very low: –0.014 and 0.015 for

253

forest F and island I respectively. This indicates an absence of

254

differentiation between parents and offsprings within both populations.

255 256

Mating system

257

Estimated outcrossing rates at the population level were high, 0.98 (± SE

13

258

0.02) and 0.96 (± 0.03) at island I and forest F respectively, suggesting that

259

selfing rarely occurred (Table 3). Single locus (ts) and multilocus (tm)

260

estimates of outcrossing rates gave similar values. The difference between

261

tm and ts, which give an indication of levels of inbreeding occurring from

262

processes other than selfing, such as biparental inbreeding (Ritland and Jain,

263

1981), was therefore negligible for both populations (Table 3). The

264

inbreeding coefficient for maternal parents obtained with MLTR was less

265

than 0.001 at island I, and equal to 0.024 (± 0.050) at forest F, which is

266

consistent with estimation of Fis values previously obtained for parent tree

267

populations. The correlation of selfing was similarly low at both

268

populations, being 0.019 (± 0.007) and 0.062 (± 0.047) at island I and forest

269

F respectively. In contrast, the correlation of outcrossed paternity was high,

270

averaging 0.167 (± 0.048) and 0.319 (± 0.070) at island I and forest F

271

respectively. The difference between populations for these outcrossed

272

paternity estimates was marginally non-significant (ts = 1.79; d.f. = 1998; P

273

< 0.10).

274 275

Genetic structure in pollen clouds

276

The comparison between the nested models of genetic structure in pollen

277

clouds was significant when comparing M1 and M2 (χ2 = 0.912; d.f. = 558;

278

P < 0.05), meaning that M2 was accepted in regard to M1 (Table 4). Neither

279

the comparison between M2 and M3 nor between M1 and M3 gave

14

280

significant results, meaning that M3 was rejected. This result indicates that

281

island I and forest F had different pollen clouds, and that these pollen clouds

282

were homogeneous within each population.

283

The Reynold’s genetic distances between pairs of groups of pollen grains

284

were weakly correlated with their geographic distance at both island I

285

(Mantel test: rp = 0.137, P = 0.129) and forest F (Mantel test: rp = 0.138, P =

286

0.086) (Figure 2). As shown with the nested models, this last result suggests

287

that there is no conspicuous genetic structure of pollen clouds within

288

populations, at the spatial scale we considered (4 hectares per population).

289 290

Discussion

291

Outcrossing vs. selfing

292

Despite being a hermaphroditic species and despite forest fragmentation, V.

293

americana exhibited high outcrossing rates in both populations, as observed

294

generally for tropical tree species (Hamrick and Murawski, 1990; Bawa,

295

1992; Doligez and Joly, 1997). Furthermore, contrary to the observations of

296

Murawski et al, (1994), both populations showed equivalent outcrossing

297

rates (0.98 vs. 0.96) despite differences in tree density (6 vs. 10 trees per

298

hectare). Moreover, low correlation values for selfing indicated that selfing

299

rates did not vary among families. These results suggest two likely and non-

300

exclusive scenarios. First, efficient pollen dispersal on both sites may have

301

insured that the number of outcrossed pollen grains was not affected by

15

302

fragmentation or reduced tree density. Second, these results also suggest

303

that outcrossed pollens may have been more efficient in producing pollen

304

tubes than selfed pollens, resulting in a high probability of allofecundation,

305

independent of the selfed to outcrossed pollen grain ratio (Aizen and

306

Feinsinger, 1994).

307 308

Pollen flow within populations

309

Estimates of correlated paternity (rp) between seedlings sampled within the

310

same family were 17 and 32% for the island I and forest F populations

311

respectively. As (rp)-1 provides an estimate of the number of effective father

312

plants (sensu Ritland, 1989), these results suggest that mother plants were

313

randomly outcrossed to a pool of about six neighbors in the island I

314

population, but to a pool of only three neighbors in the forest F population.

315

Thus, contrary to what was expected, pollen flow seemed to be less

316

restricted in the more isolated population. Nason (1997) argued that

317

reproductive systems in tropical trees can be influenced by the local density

318

of flowering conspecifics. Thus, whereas the lower V. americana population

319

density at island I may be associated with expanded pollinator foraging

320

ranges and dispersal distances, the higher tree density at forest F may be

321

associated with lower pollinator movements and lower pollen dispersal

322

distance, thus resulting in pollination occurring predominately from nearest

323

neighbors. This hypothesis is consistent with observations made by

16

324

Ellstrand and Ellam (1993) in a non-fragmented habitat, where pollen

325

dispersal distances were greater in small populations than in larger ones. In

326

contrast, our results seem opposite to those of Ghazoul et al (1998), who

327

found that inter-tree movements of pollinators declined with increasing

328

distance between flowering Shorea siamensis trees. Furthermore, we found

329

no evidence either from analyses of correlations between Reynold’s

330

distances and spatial distances or from the nested models that pollen grains

331

engendering the progeny arrays were genetically structured within

332

populations. This suggests either that the genetic structure of pollen clouds

333

was not a function of geographic distance, or that the spatial scale we

334

considered in this study (4 hectares) was too small to show a clear

335

relationship between genetic and geographic distances. The lack of

336

relationship between differentiation of pollen pools and spatial distance

337

could also be attributable to the large variances of the estimates of pollen

338

pools and a lack of statistical power in the tests (Austerlitz and Smouse,

339

2002), and/or to the predominance of few common fathers in each

340

population responsible for the pollination of most maternal trees. This latter

341

scenario could result from differences in fertility between trees (eg, some

342

trees may produce more pollen). Future studies involving paternity analyses

343

would help clarify pollen movement in this species.

344 345

Gene flow among populations

17

346

Analysis of Fst values indicate that both parent tree (0.146) and offspring

347

populations (0.085) were differentiated between island I and forest F,

348

suggesting that gene flow between these two populations was restricted,

349

both before (for parent trees) and four years after (for offspring) forest

350

fragmentation. This result is similar to what Dutech et al (unpublished data)

351

observed for another site in French Guiana: a strong genetic differentiation

352

among clusters of individuals separated by 500 to 2500 m in a natural forest

353

stand. Furthermore, analysis of genetic structure among the pollen grains

354

that engendered our progeny arrays showed that pollen clouds were also

355

differentiated between island I and forest F. Pollen dispersal between

356

populations, if any, was therefore restricted. All of these results suggest that

357

the island I and forest F populations were independent units of reproduction,

358

even before forest fragmentation occurred. This is consistent with the

359

observation that V. americana trees have a distribution that is strongly

360

aggregative (Forget et al, 1999), and suggests that the island I and forest F

361

populations may belong to two different aggregates, with genetic

362

differentiation resulting both from the absence of seed dispersal (Forget,

363

1990; Dutech et al, 2002) and inadequate pollen dispersal between them.

364

This hypothesis is reinforced by field observations made prior to forest

365

fragmentation, according to which the actual island I and Forest F were

366

formerly separated by an area of swampy forest and Euterpe olivacea stands

367

(more narrow than the present distribution of water)(O. Claessens, pers.

18

368

comm.), habitats known to be unfavorable for V. americana (Forget et al,

369

1999).

370 371

Forest fragmentation

372

This study did not provide evidence for a possible influence of forest

373

fragmentation on the mating system of V. americana trees four years after

374

fragmentation, as both isolated and non-isolated populations showed

375

comparable breeding systems. Moreover, our results suggest that pollen-

376

vector interactions were not disrupted in the isolated patch, and that forest

377

fragmentation has merely further separated two areas that were previously

378

independent with respect to reproduction. The maintenance of genetic

379

variation and normal breeding systems within the isolated population has

380

three possible explanations: (i) the size of the island (28 hectares) may be

381

too large to observe conspicuous changes due to fragmentation effects, (ii)

382

the population at island I may not have fallen below some threshold size

383

below which normal breeding associations are impaired (as suggested by

384

Young et al (1993) for Acer saccharum), or (iii) there may have been

385

insufficient time since the fragmentation for significant modifications to

386

breeding systems and patterns of genetic variation to have occurred.

387

Therefore, despite the absence of short-term effects of fragmentation,

388

caution is needed in extrapolating these results to the longer term, since

389

perturbations to important processes may take longer to become manifest.

19

390

Possible longer-term changes that may yet affect V. americana could

391

include changes in the pollinator assemblages, or increased genetic drift in a

392

fragmented landscape, a consequence of the observed restricted gene flow

393

of V. americana.

394 395

Acknowledgements

396

We thank L. Maggia and B. Godelle and F. Bonhomme for welcoming S.C.

397

at the laboratory of CIRAD-INRA at Kourou (French Guiana) and in

398

University Montpellier II (France), respectively. Many thanks also to P.-M.

399

Forget, G. Dubost and C. Erard for financial support; E. Bandou and C.

400

Bidal for technical support; K. Belkhir for statistical advices; and A. Leitao,

401

C. Bidal, A. Dalecky and S. Ratiarison for help during sampling. Comments

402

by P.-O. Cheptou, P.-M. Forget, N. Galtier, P. Jarne, D. McKey, B. McKie,

403

E. Schupp and V. Sork improved earlier versions of this manuscript. This

404

study was financed by a PhD scholarship from the French Ministry for

405

Education and Research to S.C., by E.D.F. (Convention Museum / E.D.F.

406

GP7531) and MNHN-UMR 8571.

20

407

References

408

Aizen MA, Feinsinger P (1994). Forest fragmentation, pollination, and plant

409

reproduction in a Chaco dry forest, Argentina. Ecology 75: 330–351.

410

Aldrich PR, Hamrick JL, Chavarriaga P, Kochert G (1998). Microsatellite

411

analysis of demographic genetic structure in fragmented populations

412

of the tropical tree Symphonia globulifera. Mol Ecol 7: 933–944.

413

Austerlitz F, Smouse PE (2002). Two-generation analysis of pollen flow

414

across a landscape. IV. Estimating the dispersal parameter. Genetics

415

161: 355–363.

416

Bawa KS (1990). Plant-pollinator interactions in tropical rain forests. Annu Rev Ecol Syst 21: 399–422.

417 418

Bawa KS (1992). Mating systems, genetic differentiation and speciation in tropical rain forest plants. Biotropica 24: 250–255.

419 420

Belkhir K, Borsa P, Chikhi L, Raufaste N, Bonhomme F (2000). GENETIX

421

4.02, logiciel sous Windows TM pour la génétique des populations.

422

Laboratoire Génome, Populations et Interactions, CNRS UPR 9060,

423

Université de Montpellier II, Montpellier, France.

424

Brown AHD, Allard RW (1970). Estimation of the mating system in open-

425

pollinated maize populations using isozyme polymorphisms.

426

Genetics 66: 133–145.

427

Charlesworth D, Charlesworth B (1987). Inbreeding depression and its evolutionary consequences. Annu Rev Ecol Syst 18: 237–268.

428

21

429

Chauvet S (2001). Effets de la fragmentation forestière sur les interactions

430

plantes-animaux: conséquences pour la régénération végétale. Thèse

431

de l'Université Paris 6.

432

Claessens O, Granjon L, de Massary J-C, Ringuet S (2002). La station de

433

recherche de Saint-Eugène: situation, environnement et présentation

434

générale. Rev Ecol 57: 21–37.

435

Delphi (1998). Delphi TM version 4.0. Inprise Corporation.

436

Didham RK, Ghazoul J, Stork NE, Davis AJ (1996). Insects in fragmented forests: a functional approach. Trends Ecol Evol 11: 255–260.

437 438

Doligez A, Joly HI (1997). Mating system of Carapa procera (Meliaceae)

439

in the French Guiana tropical forest. Am J Bot 84: 461–470.

440

Dutech C, Amsellem L, Billotte N (2000a). Characterization of (GA)n

441

microsatellite loci using an enrichment protocol in the neotropical

442

tree species Vouacapoua americana. Mol Ecol 9: 1433–1449.

443

Dutech C, Maggia L, Joly HI (2000b). Chloroplast diversity in Vouacapoua

444

americana (Caesalpiniaceae), a neotropical forest tree. Mol Ecol 9:

445

1427–1432.

446

Dutech C, Seiter J, Petronelli P, Joly HI, Jarne P (2002). Evidence of low

447

gene flow in a neotropical clustered tree species in two rainforest

448

stands of French Guiana. Mol Ecol 11: 725–738.

449

Edwards AWF (1972). Likelihood. Cambridge University Press: Cambridge.

450

Ellstrand NC, Elam DR (1993). Population genetic consequences of small

22

451

population size: implications for plant conservation. Annu Rev Ecol

452

Syst 24: 217–242.

453

Forget P-M (1990). Seed-dispersal of Vouacapoua americana

454

(Caesalpiniaceae) by caviomorph rodents in French Guiana. J Trop

455

Ecol 6: 459–468.

456

Forget P-M, Mercier F, Collinet F (1999). Spatial patterns of two rodent-

457

dispersed rain forest trees Carapa procera (Meliaceae) and

458

Vouacapoua americana (Caesalpiniaceae) at Paracou, French

459

Guiana. J Trop Ecol 15: 301–313.

460

Ghazoul J, Liston KA, Boyle TJB (1998). Disturbance-induced density-

461

dependent seed set in Shorea siamensis (Dipterocarpaceae), a

462

tropical forest tree. J Ecol 86: 462–473.

463

Hamrick JL, Murawski DA (1990). The breeding structure of tropical tree forest populations. Plant Species Biol 5: 157–165.

464 465

Janzen DH (1974). Tropical blackwater rivers, animals, and mast fruiting by the Dipterocarpaceae. Biotropica 6: 69–103.

466 467

Jennersten O (1988). Pollination in Dianthus deltoides (Caryophyllaceae):

468

effects of habitat fragmentation on visitation and seed fate. Conserv

469

Biol 2: 359–366.

470

Konuma A, Tsumura Y, Lee CT, Lee SL, Okuda T (2000). Estimation of

471

gene flow in the tropical-rainforest tree Neobalanocarpus heimii

472

(Dipterocarpaceae), inferred from paternity analysis. Mol Ecol 9:

23

1843–1852.

473 474

Mantel N (1967). The detection of disease clustering and generalized regression approach. Cancer Res 27: 209–220.

475 476

Murawski DA, Gunatilleke IAUN, Bawa KS (1994). The effects of selective

477

logging on inbreeding in Shorea megistophylla (Dipterocarpaceae)

478

from Sri Lanka. Conserv Biol 8: 997–1002.

479

Nason JD, Aldrich PR, Hamrick JL (1997). Dispersal and the dynamics of

480

genetic structure in fragmented tropical tree populations. In:

481

Laurance WF, Bierregaard RO (eds) Tropical forest remnants.

482

Ecology, management, and conservation of fragmented communities,

483

The University of Chicago Press: Chicago & London. Pp 25–41.

484

Powell AH, Powell GVN (1987). Population dynamics of male Euglossine

485

bees in Amazonian forest fragments. Biotropica 19: 176–179.

486

Rathcke BJ, Jules ES (1993). Habitat fragmentation and plant-pollinator interactions. Curr Sci 65: 273–277.

487 488

Raymond M, Rousset F (1995a). An exact test for population differentiation. Evolution 49: 1280–1283.

489 490

Raymond M, Rousset F (1995b). Genepop (version 1.2): population

491

genetics software for exact tests and ecumenicism. J Heredity 86:

492

248–249.

493

Ritland K (1989). Correlated matings in the partial selfer Mimulus Guttatus. Evolution 43: 848–859.

494

24

495

Ritland K, Jain S (1981). A model for the estimation of outcrossing rate and gene frequencies using n independent loci. Heredity 47: 35–52.

496 497

Streiff R, Labbé T, Bacilieri R, Steinkellner H, Glössl J, Kremer A (1998).

498

Within-population genetic structure in Quercus robur L. & Quercus

499

petraea (Matt.) Liebl. assessed with isozymes and microsatellites.

500

Mol Ecol 7: 317–328.

501

Vinson SB, Frankie GW, Barthell J (1993). Threats to the diversity of

502

solitary bees in a neotropical dry forest in Central America. In:

503

LaSalle J, Gauld ID (eds) Hymenoptera and Biodiversity, CAB

504

International: Wallingford. Pp. 53–81.

505

Weir BS, Cockerham CC (1984). Estimating F-statistics for the analysis of population structure. Evolution 38: 1358–1370.

506 507

White GM, Boshier DH, Powell W (2002). Increased pollen flow

508

counteracts fragmentation in a tropical dry forest: An example from

509

Swietenia humilis Zuccarini. Proc Natl Acad Sci 99: 2038–2042.

510

Wiens D (1984). Ovule survivorship, brood size, life history, breeding systems, and reproductive success in plants. Oecologia 64: 47–53.

511 512

Young AG, Merriam HG, Warwick SI (1993). The effects of forest

513

fragmentation on genetic variation in Acer saccharum Marsh. (sugar

514

maple) populations. Heredity 71: 277–289.

515

Young A, Boyle T, Brown T (1996). The population genetic consequences of habitat fragmentation for plants. Trends Ecol Evol 11: 413–418.

516

25

Figure legends Figure 1 Location of the Vouacapoua americana populations studied in Saint-Eugène. Populations F and I respectively belong to one continuous forest and one island (so-called CF3 and island 2 in Claessens et al 2002). The trees included in the genetic analysis are indicated as open circles, while other conspecifics are indicated as black circles.

Figure 2 Absence of within-population structure for pollen clouds. Reynolds genetic distance are represented in function of geographic distance, for all pairs of individual pollen clouds per population (each individual pollen cloud corresponding to one family).

Forest F

Island I

1km

100 m

0.20

Island I (rp = 0.137)

0.15

0.10

Reynolds genetic distance

0.05

0.00 0

50

100

150

200

250

100

150

200

250

0.20

Forest F (rp = 0.138) 0.15

0.10

0.05

0.00 0

50

Geographic distance (m)

Table 1 Genetic diversity among parent trees and offspring populations. For each population, the sample size (N), mean number of alleles per locus (nmd), observed heterozygosity (Ho), gene diversity (He) and inbreeding coefficient over all loci (Fis) are given. For offspring populations, one individual was randomly chosen among the progeny array for each family. Populations

Island I

Forest F

Parents

Offspring

Parents

Offspring

N

13

13

20

20

nmd

2.6

2.8

4.0

4.0

Ho

0.43

0.51

0.47

0.50

He

0.39

0.52

0.53

0.54

Fis

- 0.10

0.02

0.11

0.08

Table 2 Alleles frequencies at all microsatellite loci, for parent trees and offspring populations, at both the island and continuous forest sites. Island I Wac1

Wac5

Wac7

Wac10

Forest F

Alleles

Parents

Offsprings

Parents

Offsprings

1

0.038

0.077

0.025

0.025

2

0.308

0.308

0.175

0.275

3

0.077

0.038

0.125

0.175

4

0

0.115

0.075

0.050

5

0.538

0.462

0.450

0.300

6

0.038

0

0.075

0.125

7

0

0

0.025

0.025

8

0

0

0.050

0.025

1

0

0.192

0.400

0.300

2

0.154

0.115

0.100

0.125

3

0.846

0.692

0.475

0.550

4

0

0

0.025

0.025

1

0.308

0.462

0.750

0.700

2

0.692

0.538

0.250

0.300

Alleles

Alleles

Alleles

1

0

0

0.025

0.025

2

0.885

0.654

0.775

0.775

3

0

0

0.100

0.025

4

0.115

0.346

0.100

0.175

1

0.731

0.731

0.400

0.400

2

0.269

0.269

0.600

0.600

Mean Number of Allele

2.6

2.8

4

4

Standard Deviation

1.3

1.3

2.4

2.4

Wac13

Alleles

ALL LOCI

Table 3 Estimates of population-level outcrossing rates and correlated mating. The island I and Forest F populations are represented by 13 and 20 families, i.e. 130 and 200 offspring, respectively. Estimates of parental inbreeding coefficient were obtained using MLTR procedure (F), multilocus (tm) and single locus (ts) outcrossing rates, differences between multilocus and single locus estimates (tm - ts), mean outcrossing rates per family (ti), correlated selfing (rs) and correlated paternity (rp). Values are given with their standard errors in brackets, which were obtained from 1000 bootstraps. F

tm

ts

tm - ts

ti

rs

rp

Island I

< 0.001

0.983 (0.018)

0.988 (0.022)

- 0.005 (0.018)

0.901 (0.031)

0.019 (0.007)

0.167 (0.048)

Forest F

0.024 ( 0.050)

0.956 ( 0.029)

0.971 (0.029)

- 0.015 (0.025)

0.865 (0.036)

0.062 (0.047)

0.319 (0.070)

Table 4 Analysis of the genetic structure of pollen clouds, using likelihood ratio-tests for the comparison of three nested models: M1 (each tree received a differentiated pollen cloud), M2 (each population received a differentiated pollen cloud, but there was no differentiation between trees within-populations), and M3 (the pollen cloud was homogeneous for every tree and both populations). Models

M1

M2

M3

-1317.0

-1573.7

-1643.6

594

36

18

Model comparisons

M1 / M2

M2 / M3

M1 / M3

2 * [ln(Li) - ln(Lj)]

513.3

139.9

653.2

558

18

576

0.912

6.7*10-21

0.014

ln (Likelihood) d.f.

d.f. i - d.f. j χ2 probability