Oecologia (1998) 117:342±352 Ó Springer-Verlag 1998

Luc Gigord á Claire Lavigne á Jacqui A. Shyko€ Partial self-incompatibility and inbreeding depression in a native tree of La ReÂunion (Indian Ocean)

Received: 2 March 1998 / Accepted: 3 August 1998

Abstract We investigated the reproductive system of that D. acutangula possesses an incompatibility system the threatened taxon acutangula ssp. acu- similar to that found in other species such tangula Cav. (Sterculiaceae), an endemic tree of the as Theobroma cacao L. Such an incompatibility system Mascarene archipelago (Indian Ocean). A controlled allows a certain amount of sel®ng, and di€erent indi- crossing experiment was performed in two natural viduals vary in their degree of self-incompatibility. The populations located in the remnants of the low-eleva- low success of crosses among close neighbours in one tion dry forest on the island of La Re union. Active population suggests that there was spatial structure for pollination, probably mainly by insects, was necessary incompatibility alleles in that population. This could for reproduction in this species. Individuals varied in partly explain the decline of the species in fragmented their degree of self-sterility from 0 to 100%. Out- and disturbed , since relatedness at incompati- crossing between nearby individuals produced lower bility loci may increase in small or isolated population seed set than did crosses between more distant indi- and thus reduce mate availability. viduals within one of the two tested populations. The variation in reproductive success on sel®ng and in the Key words Dombeya acutangula ssp. acutangula á di€erent types of crosses could result from inbreeding Island á fragmentation á Conservation á depression causing embryo death, and we provide evi- Crossing experiment dence that progenies from sel®ng have lower seed size and quality. However, for inbreeding depression to account for the dramatic variation in seed set found in Introduction our crossing experiment, the distribution of genetic load and number of lethal factors required appear A major source of environmental change and distur- unrealistic. We favour an alternative interpretation, bance involves habitat fragmentation, which disrupts normal biological and physical ecosystem processes (Saunders et al. 1991; Noss and Csuti 1994), and often L. Gigord leads to a decrease in population size thereby increasing Laboratoire de Biologie et Physiologie Ve ge tales, random genetic drift and inbreeding depression, de- Faculte des Sciences, Universite de La Re union, creasing interpopulation gene ¯ow and increasing the 15 Avenue Rene Cassin, F-97715 Sainte-Clotilde Cedex 09, Ile de La Re union, France probability of local of populations within a metapopulation (Gilpin 1991; Young et al. 1996). In- J.A. Shyko€ Laboratoire d'Evolution et Syste matique, trinsic population processes in small populations may Universite Paris-Sud [CNRS (URA 2154)], BaÃtiment 362, result in decreasing population viability (for an overview F-91405 Orsay Cedex, France see Soule 1987). While some characteristics may pre- C. Lavigne dispose particular species to persist or even spread under Institut de Biotechnologie des Plantes, changing environmental conditions, others expose some Laboratoire de Phytopathologie Mole culaire, Universite Paris-Sud, species to diculties (Schonewald-Cox 1985); in partic- BaÃtiment 630, F-91405 Orsay Cedex, France ular, in ¯owering , the reproductive and pollina- Present address: tion system may in¯uence the viability of small L. Gigord (&) populations, but these e€ects have not received sucient Laboratoire d'Evolution et Syste matique, Universite Paris XI, BaÃtiment 362, F-91405 Orsay Cedex, France, attention to date (Menges 1991). e-mail: [email protected], The rate of outcrossing versus self-pollination may Tel.: +33-1-69156115, Fax: +33-1-691 57353 be an important factor for population viability (Barrett 343 and Kohn 1991). Plants possess many di€erent mech- become extinct and currently approximately 50 are anisms that promote outbreeding, including genetic considered threatened (Bosser et al. 1976; Dupont et al. self-incompatibility systems that preclude the produc- 1989). tion of inbred zygotes after self-pollination and crosses Dombeya acutangula (sensu lato) is a widely distrib- among some related individuals. Self-incompatible uted tree in Africa (Seyani 1991), Madagascar (Are neÁ s species are expected to harbour higher genetic load 1959) and islands of the Indian Ocean (Friedmann 1987). than species that habitually inbreed (Barrett and On La Re union, it is only represented by the endemic Charlesworth 1991) and should su€er more from in- subspecies D. acutangula ssp. acutangula (Sterculiaceae) breeding depression than would autogamous species which is considered threatened (Dupont et al. 1989). On (Charlesworth et al. 1990). In fact, inbreeding depres- Mauritius and Rodrigues, the species is nearly extinct sion is found in the rare o€spring produced by sel®ng (Bosser et al. 1976). On La Re union, the species can be in largely self-incompatible species (Manasse and found throughout the low-elevation areas of human Pinney 1991; Ramsey and Vaughton 1996). In small habitation and only isolated fragments of natural pop- populations resulting from habitat fragmentation this ulations remain. A previous study (Gigord et al., in press) e€ect should be particularly problematic because small has shown that the reproductive output of the species is populations are more prone to the accumulations of signi®cantly a€ected by habitat fragmentation. Further- deleterious alleles than are large populations (Hauser more, anecdotal reports of isolated individuals of D. et al. 1994; Lynch et al. 1995). acutangula ssp. acutangula that produced no seeds de- There are two typical self-incompatibility systems: spite active pollinator visitation suggested that this spe- sporophytic systems, where matings between parent cies might exhibit a self-incompatibility system that could plants that share any alleles at the self-incompatibility contribute to its decline in natural habitats in locus are prevented, and gametophytic systems, where La Re union. We therefore investigated the reproductive pollen growth fails if it shares an allele with the re- system of this species using a series of controlled crosses. cipient (de Nettancourt 1977). Other incomplete The goal of this study was to determine whether this self-incompatibility systems, however, exist. These may species is self-compatible and whether it requires pollen be leaky, such that sel®ng or crosses between indi- vectors for fertilisation. We further tested if distances viduals bearing the same incompatibility alleles result between pollen donors and recipient plants in¯uenced in seed (Reinartz and Les 1994). Incomplete incom- seed production and components of progeny ®tness. patibility will therefore expose deleterious recessive This was done to investigate the potential relatedness alleles to selection more often than complete incom- among more or less geographically distant individuals, patibility systems, so the genetic load should be and to determine whether low mate availability and/or reduced by selection. Incompatibility systems a€ect inbreeding depression may indeed be contributing to the the availability of possible mates. Clearly rare incom- decline of this threatened species. patibility alleles should spread, and high allelic diver- sity occurs for loci governing self-incompatibility (see Richman and Kohn 1996). In small populations, Study species however, even selected alleles can be lost by drift (Kimura 1983), so genetic variability at the incompati- D. acutangula ssp. acutangula Cav. (Sterculiaceae) is a bility locus may decline, thus reducing mate availabi- small tree up to 8 m tall, sometimes ¯owering as a shrub lity. When sel®ng or pollinations between certain as young as 3 years old. Flowering lasts for 2±7 weeks individuals fail to produce seed, either inbreeding between March and May in La Re union. Mature adult depression leading to zygote death or incompatibility plants produce hundreds of large white ¯owers (diame- of the mating can be invoked. ter about 20±30 mm) grouped in cincinnate axillary Island plant species might be particularly prone to in¯orescences of approximately 20 ¯owers. Each ¯ower diculties from habitat change induced by human bears one ovary enclosing ten ovules shared among ®ve activity. They are often bad competitors, low dispersers locules. These ¯owers are protandrous with extrorse and obligatory outcrossers (Janzen 1971; Bawa 1982; anthers, and the pistil becomes receptive 1 day after Baker and Cox 1984; Mueller-Dombois and Loope stamen dehiscence. Pollination is not well documented, 1990), and islands are arenas for the evolution of rare but observations of insect visits in the ®eld and the reproductive systems (Baker 1967). The islands of the production of a large quantity of nectar suggest that Mascarene archipelago (Mauritius, La Re union and plants are mainly pollinated by butter¯ies and a large Rodrigues) are particularly interesting for investigating range of insects such as bees, ants, ladybirds, bumble- this problem because they were free from human in¯u- bees or beetles, and also occasionally by a small endemic ence until the middle of the 16th century, unlike the bird (Zosterops borbonicus Bodaert) known to pollinate Hawaiian archipelago that was colonised a few thou- species of the closely related genus Trochetia (Gill 1971). sand years ago (Schonewald-Cox 1985). On the island of The fruit is dry, indehiscent and dispersed by gravity La Re union, low-altitude habitats have su€ered greatly and wind. This plant can be found in various types of from human colonisation and the introduction of exotic habitats including woodland, bushland, thickets, and species. At least 10 endemic plant species have already sometimes on stream banks or in rocky places. 344

Each seedling was repotted twice at 2-month intervals after Methods emergence and pots were maintained in the greenhouse. Seedling height at the second repotting was measured on all seedlings per Study sites mother plant and treatment when fewer than 30 seedlings were available. When more than 30 seedlings were available, approxi- The two study populations were chosen in the disturbed low dry mately two-thirds, but with a minimum number of 30, were chosen forest. These populations were located in the north-west of the island haphazardly and measured. The average number of seedlings and were approximately 5 km apart. The ®rst population, named measured per cross and mother was 15 (ranging from 3 to 52). ``Ravine aÁ Malheur'' (hereafter Malheur), contains 83 individuals of Seedling mortality was assessed over the experiment. D. acutangula. The second population, named ``Petite Ravine des Lataniers'' (hereafter Lataniers), contains 53 individuals. Ten adult individuals were chosen at random within the same Analysis of data size category (height about 3 m) in each population. These indi- viduals produced approximately the same number of in¯orescences All data analyses were performed using the GLM procedure of the (between 60 and 70 in the year of the experiment). SAS statistical package (SAS 1990). We tested for the e€ect of pollination treatment on all response variables with analyses of variance employing type III sums of squares. To meet the Crossing design normality and homoscedasticity assumptions of the analysis of variance, data for number of seeds per fruit, growth rate and All crosses were performed in situ between March and May 1995. We germination date were square-root-transformed, seed mass was log- performed ®ve types of crosses: simple bagging without hand-polli- transformed and all proportions (i.e. proportion of fruits per nation (`passive') and with hand-pollination (`handself'), crosses ¯ower, ¯attened seed, germinated seeds and dead individuals) were within populations between close neighbours separated by an aver- arcsine-square-root-transformed before performing any analyses age distance of 4 m (`within-population near') and between indi- (Sokal and Rohlf 1981). For seed mass, seed set per fruit, germi- viduals as far apart as possible, which represents a distance greater nation date, and seedling growth rate, where multiple measures per than 150 m (`within-population far'), and crosses among individuals in¯orescence were available, we included the level of the in¯ore- between populations separated by a distance of about 5 km (`be- scence, nested within the individual by treatment interaction, in the tween population'). All crossing treatments were performed on all analysis. ten individuals from both populations. Six in¯orescences per indi- ANOVAs for all response variables followed this form. Main vidual were subjected to the `between-population' treatment, three factors tested were the populations of origin (Lataniers versus in¯orescences per individual were used for each of the other treat- Malheur), the individual nested within population of origin, the ments. Each outcrossed in¯orescence received the pollen from a cross treatment, and in¯orescence, nested within the individual by di€erent donor plant. treatment interaction. Interactions were tested between treatment All in¯orescences that were to be used as pollen recipients were and (1) population, and (2) individual within population. In¯ore- bagged with large white ®ne-mesh bags before ¯owers opened. For scence, individual within population and its interaction with treat- the `passive' treatment, in¯orescences were simply kept bagged. ment were declared as random e€ects. Consequently, all expected Self- and cross-hand-pollinations were accomplished by rubbing mean squares were calculated as appropriate linear combinations of dehiscent anthers from a ¯ower bud of the appropriate pollen other mean-square terms, and denominator degrees of freedom donor that was ready to open onto the stigma of the seed parent. were calculated by Satterthwaite approximation (SAS 1990). We did not emasculate buds prior to hand-pollination because the In the ®rst analysis, we compared `handself' with `passive' species is protandrous, which reduced the probability of viable crosses within individuals to determine whether full seed set could pollen remaining within a ¯ower when the stigma became receptive, occur in the absence of pollinators. Next, we assessed the success of and the stamens are extrorse, which reduced the deposition of the selfed versus within-population outcrossed pollinations, comparing pollen of a ¯ower on its own stigma. All in¯orescences were bagged `handself' with the grouped `within-population near' and `within- again after pollination until the stigma dried out. Ripe fruits were population far' cross types. Lastly, we considered crosses between collected 13±14 weeks after pollination treatments. individuals and tested the e€ect of the distance (i.e. `within-popu- lation near', `within-population far' and `between population'). We chose a critical alpha value of signi®cance of 0.025. We used this Traits measured conservative level since the tests performed used overlapping sub- sets of data and the data from certain treatments were employed in Traits measured for each in¯orescence were: (1) number of fruits two tests. Furthermore, critical alpha levels were corrected for each per pollinated ¯ower, (2) number of seeds per fruit, (3) mass of each e€ect in the ANOVA models by the number of variables tested seed, (4) seed germination, (5) germination date, (6) proportion of separately in each of the three ANOVAs, using the sequential mortality among seedlings and (7) seedling growth rate. Because Bonferroni procedure of Rice (1989). Signi®cance levels reported D. acutangula is a long-lived tree species with a long juvenile phase, take into consideration the corrections for critical alpha values it was only possible to measure traits on juvenile progeny. Two (i.e. a single asterisk that normally de®nes P < 0.025 represents types of seeds were identi®ed, full and ¯attened. Flattened seeds this critical value divided by the number of tests performed for a were easily recognisable because they were lighter in colour than particular hypothesis test). the full ones and their morphology was always obviously di€erent. In addition, we compared ¯attened and full seeds for mass, For each seed, we noted whether or not it was ¯attened and also germination date, germination success, seedling growth rate and determined (8) the proportion of ¯attened seeds. Each seed was seedling mortality using t-tests on means per individual calculated weighed on a precision balance to an accuracy of 10)5 g. on the in¯orescence means.

Seed germination Results Seeds were germinated in a greenhouse. Each seed was sown in an individual pot containing a 1/2 natural local soil, 1/4 compost and Handself versus passive treatment 1/4 peat mixture. All pots were watered ad libitum and identically for all seeds and seedlings. For each germinated seed, the time until germination was determined. The pots were checked over In both populations, in¯orescences that were selfed 10 months to estimate ®nal germination percentages. produced more fruit per ¯ower and more seeds per fruit 345 than those subjected to the passive treatment. No sig- individuals were entirely self-sterile, one produced very ni®cant di€erences, however, were found for the other few seeds and may be considered as mostly self-sterile, traits measured. The interaction between individual and ®ve individuals were partially or totally self-fertile within population and treatment indicated that the re- (Fig. 1b). Individuals were considered self-sterile if they sponse to pollination treatment also varied among in- produced no seeds, even if fruits were produced. More- dividuals for fruit set and seed set (Table 1, Fig. 1a,b). over, individuals di€ered signi®cantly for all measured In fact, of the ten individuals tested in the Lataniers traits except fruit set per ¯ower and seedling growth population, three were entirely self-sterile when selfed, rate, and the two populations di€ered signi®cantly for three produced very few seeds and may be considered as seed mass (Table 1). Within a treatment, in¯orescences mostly self-sterile, and four were partially or totally self- varied for germination date and seedling growth rate, as fertile but two only produced seeds with the passive shown by the signi®cant in¯orescence, nested in the treatment (Fig. 1a). In the Malheur population, four treatment by individual interaction.

Table 1 Mixed-model hierarchical analysis of variance on four and, when available, in¯orescences within the individual by treat- adult and four progeny traits in Dombeya acutangula. E€ects tested ment interaction. For details of hypothesis testing see statistical were the population, individual nested within population, cross methods in the text (dfn degrees of freedom of the numerator, dfd treatment (`handself' versus `passive'), the interactions between degrees of freedom of the denominator, MS mean square) cross treatment and population and individual within population

Source of variation Adult traits

Fruit set per ¯ower Seed set per fruit Proportion Seed mass of ¯attened seeds

Population MS 1.58 ´ 10)2 0.687 0.197 2.977 (dfn,dfd) (1,18) (1,18) (1,12) (1,13) F 0.027 0.172 0.494 29.67*** Individual MS 0.594 6.61 0.479 0.138 (Population) (dfn,dfd) (18,18) (18,9) (11,35) (10,13) F 2.13 4.83* 17.43*** 12.33*** Treatment MS 13.33 20.59 7.96 ´ 10)3 8.47 ´ 10)4 (dfn,dfd) (1,18) (1,9) (1,34) (1,27) F 48.02*** 18.05** 0.427 0.058 Population MS 2.46 ´ 10)5 2.32 ´ 10)4 4.03 ´ 10)4 9.67 ´ 10)4 ´ Treatment (dfn,dfd) (1,18) (1,9) (1,35) (1,30) F 8.8 ´ 10)5 2.02 ´ 10)4 0.020 0.065 Individual (Population) MS 0.279 1.302 4.02 ´ 10)3 8.22 ´ 10)3 ´ Treatment (dfn,dfd) (18,78) (9,93) (5,30) (5,88) F 5.31*** 9.74*** 0.027 0.299 In¯orescence [Individual MS 0.153 2.32 ´ 10)2 (Population)] ´ Treatment (dfn,dfd) (45,271) (30,136) F 1.454 0.664 Error MS 0.052 0.105 0.147 0.035 Progeny traits Seed germination Germination date Seedling mortality Seedling growth rate

Population MS 0.041 0.128 0.132 9.53 ´ 10)4 (dfn,dfd) (1,9) (1,13) (1,12) (1,15) F 0.119 0.041 0.936 0.474 Individual MS 0.384 3.723 0.163 2.25 ´ 10)3 (Population) (dfn,dfd) (9,32) (10,23) (11,23) (10,7) F 27.32*** 4.09** 8.47*** 1.479 Treatment MS 4.74 ´ 10)3 1.015 0.006 1.58 ´ 10)3 (dfn,dfd) (1,4) (1,3) (1,3) (1,4) F 1.511 2.58 0.843 1.075 Population MS 2.96 ´ 10)2 9.21 ´ 10)3 6.07 ´ 10)3 3.14 ´ 10)5 ´ Treatment (dfn,dfd) (1,4) (1,2) (1,3) (1,4) F 0.943 0.027 3.986 0.0212 Individual (Population) MS 3.14 ´ 10)3 0.449 8.02 ´ 10)3 1.48 ´ 10)3 ´ Treatment (dfn,dfd) (4,36) (4,121) (4,29) (4,40) F 0.045 0.125 0.126 0.698 In¯orescence [Individual MS 4.412 2.43 ´ 10)3 (Population)] ´ Treatment (dfn,dfd) (29,121) (29,119) F 3.24** 1.985* Error MS 0.069 1.362 0.064 1.23 ´ 10)3

*P < 0.025; **P < 0.005; ***P < 0.0005; n.s. P > 0.025 346

Fig. 1 Seed set per pollinated ¯ower (a,b) and proportion of ¯attened Since, within an individual, di€erent in¯orescences were seeds (c,d)ofeachDombeya acutangula individual of the two studied pollinated by di€erent pollen donors, this e€ect suggests populations for the three types of cross treatment: `passive' (white areas), `handself' (light shaded areas) and `outcross' within populations that, on one maternal plant, some pollen donors en- (dark shaded areas) (`within-population near' grouped with `within- hanced seed production over that of sel®ng far more population far'). Bars represent mean values over in¯orescences and than did other pollen donors (Fig. 2). error bars represent SEs

Handself versus outcross within-population treatment

Outcrossed in¯orescences from both populations pro- duced signi®cantly superior fruit set per ¯ower and seed set per fruit and signi®cantly fewer ¯attened seeds than did the handselfed in¯orescences. The interaction between individuals within population and treatment e€ects indicated that the response to treatment varied among individuals in the two populations (Table 2, Fig. 1). This treatment e€ect on seed set remained sig- ni®cant when the analysis was performed only using data from the self-fertile individuals (see above) within the two populations [F(1,8) ˆ 7.72, P < 0.05] indicat- ing that there were di€erences in the level of self-sterility among self-fertile individuals. Moreover, individual- within-population e€ects were signi®cant for several reproductive traits measured, and there was a highly signi®cant population e€ect on seed weight (Table 2). Fig. 2 Seed set per fruit in the two populations of D. acutangula for three types of cross treatment: `within-population near' (white areas), Within a treatment, in¯orescences varied for number of `within-population far' (light shaded areas) and `between population' seeds produced, as shown by the signi®cant in¯ore- (dark shaded areas). Bars represent mean values over in¯orescences scence, nested in the treatment by individual interaction. and error bars represent SEs 347

Table 2 Mixed-model hierarchical analysis of variance on four and when available, in¯orescences within the individual by treat- adult and four progeny traits in D. acutangula. E€ects tested were ment interaction. For details of hypothesis testing see statistical the population, individual nested within population, cross treat- methods in the text (dfn degrees of freedom of the numerator, dfd ment (outcross versus handself), the interactions between cross degrees of freedom of the denominator, MS mean square) treatment and population and individual within population and,

Source of variation Adult traits

Fruit set per ¯ower Seed set per fruit Proportion Seed mass of ¯attened seeds

Population MS 0.012 3.217 0.276 8.361 (dfn,dfd) (1,18) (1,19) (1,12) (1,19) F 0.032 0.789 1.721 26.47*** Individual MS 0.420 6.332 0.207 0.552 (Population) (dfn,dfd) (18,18) (18,18) (18,10) (18,3) F 0.895 1.123 0.687 75.88** Treatment MS 6.212 95.98 7.828 5.15 ´ 10)2 (dfn,dfd) (1,18) (1,19) (1,12) (1,23) F 12.81* 27.64*** 33.28*** 0.501 Population ´ Treatment MS 1.09 ´ 10)3 1.34 ´ 10)2 4.22 ´ 10)3 8.43 ´ 10)5 (dfn,dfd) (1,18) (1,19) (1,12) (1,23) F 2.26 ´ 10)3 3.86 ´ 10)3 0.018 8.19 ´ 10)2 Individual (Population) MS 0.489 5.475 0.249 9.19 ´ 10)2 ´ Treatmemt (dfn,dfd) (18,134) (18,94) (11,111) (10,210) F 7.79*** 9.604*** 3.545*** 0.792 In¯orescence [Individual MS 0.921 0.010 (Population)] ´ Treatment (dfn,dfd) (69,872) (58,500) F 5.288*** 0.732 Error MS 0.063 0.174 0.070 1.37 ´ 10)2 Progeny traits Seed germination Germination date Seedling mortality Seedling growth rate

Population MS 0.037 4.41 ´ 10)5 4.73 ´ 10)3 0.001 (dfn,dfd) (1,19) (1,19) (1,21) (1,22) F 0.155 2.44 ´ 10)6 0.043 0.172 Individual MS 0.329 31.51 0.140 9.36 ´ 10)3 (Population) (dfn,dfd) (18,6) (18,5) (18,8) (18,1) F 14.74** 29.71*** 2.328 44.9 Treatment MS 0.061 0.269 8.99 ´ 10)4 1.57 ´ 10)4 (dfn,dfd) (1,10) (1,17) (1,12) (1,31) F 3.056 0.264 0.017 0.237 Population ´ Treatment MS 0.048 0.819 0.045 4.34 ´ 10)4 (dfn,dfd) (1,10) (1,17) (1,12) (1,31) F 2.402 0.803 0.838 0.656 Individual (Population) MS 0.020 1.036 0.055 4.96 ´ 10)4 ´ Treatment (dfn,dfd) (9,110) (10,172) (11,111) (10,185) F 1.441 1.09 1.53 0.463 In¯orescence [Individual MS 0.917 9.92 ´ 10)4 (Population)] ´ Treatment (dfn,dfd) (58,476) (58,469) F 0.933 0.841 Error MS 0.014 0.984 0.036 1.18 ´ 10)3

*P < 0.025; **P < 0.005; ***P < 0.0005; n.s. P > 0.025

E€ect of distance pollen donors, showed very variable seed production per fruit (Table 3, Fig. 2). Particularly in the Lataniers pop- The signi®cant interaction between distance treatment ulation, ®ve combinations of pollen recipients and donors e€ect and population e€ect on seed set per fruit indicates produced virtually no seeds, while other donors produced that the response of this trait to distance varied with the good seed set on the same plants. In the Malheur popu- population of origin (Fig. 3). Separate analyses showed lation we found no pollen recipient-donor combinations that the distance treatment did not in¯uence seed set per that gave no seeds, though one produced about half the fruit in the Malheur population [F(2,19) ˆ 2.16, average seed set (Fig. 2). P ˆ 0.14], while in the Lataniers population, seed set per Similarly, the interaction between distance and popu- fruit was signi®cantly lower when individuals were cros- lation e€ects on the proportion of ¯attened seeds indicates sed with a near neighbour [F(2,19) ˆ 12.76, P < 0.001] that its response to distance depended on the population. (Fig. 3). Di€erent in¯orescences, pollinated by di€erent In the Lataniers population, the proportion of ¯attened 348

Fig. 3a,b Seed set per fruit for each D. acutangula individual in the tors. The extrorse anthers and protandry also promote two studied populations for the cross treatment `within-population outcrossing, though geitonogamous pollination is not near', showing the three in¯orescences, each pollinated by a di€erent pollen donor. Bars represent mean values over individual ¯owers precluded. We found that pollinators were required to within in¯orescences and error bars represent SEs improve both fruit set and seed set in this species (Fig. 1). Only a few individuals were able to produce seeds when in¯orescences were simply bagged but even seeds was signi®cantly greater when individuals were they always produced fewer seeds per fruit compared to crossed with a near-neighbour [F(2,22) ˆ 14.38, P < active self-pollination (Fig. 1). Since we did not emas- 0.001]. This was not true in the Malheur population culate the ¯owers in the crossing treatments, we cannot [F(2,19) ˆ 2.15, P ˆ 0.14] (Fig. 4). rule out that some of the fruits or seeds produced in the As in the previous analysis, individual e€ects were other pollination treatments resulted from passive self- highly signi®cant for all measured traits of reproductive pollination. However, since in all cases where there was success except on fruit set per ¯ower, seed set per fruit a di€erence between sel®ng and outcrossing, the success and the proportion of ¯attened seeds, and the popula- of outcrossing was higher, noise generated by mixtures tion e€ect was highly signi®cant for seed weight of autogamous sel®ng in outcrossed treatments would (Table 3). Furthermore, a signi®cant in¯orescence e€ect only serve to make our comparisons more conservative. was found for seed set, with di€erent in¯orescences Overall, reproductive success of the species appeared to subjected to the same pollination treatment, but polli- be dependent on pollinators. nated by di€erent donors, reacting di€erently. Our results furthermore showed that some individu- als of D. acutangula are apparently fully self-sterile. When handselfed, these produced no seeds though some Seed and progeny quality empty fruits were produced. Other individuals were partly self-fertile, producing fewer seeds per fruit on Flattened seeds were lighter than full seeds sel®ng than on outcrossing. Still others were able to (2.6 ‹ 0.20 mg, n ˆ 20 and 4.2 ‹ 0.24 mg, n ˆ 20, produce as many seeds per pollinated ¯ower when selfed respectively; t ˆ 5.4, df ˆ 38, P < 0.0001) in both as when outcrossed (Fig. 1). The high variability among populations (Lataniers t ˆ 5.13, df ˆ 18, P < 0.001; individuals in the level of self-sterility suggests that Malheur t ˆ 5.13, df ˆ 18, P < 0.001). Flattened D. acutangula possesses neither a classical sporophytic seeds also had lower germination success than full seeds nor gametophytic self-incompatibility system as de®ned in both populations (0.16 ‹ 0.05, n ˆ 20 and by de Nettancourt (1977). Such incompatibility systems 0.78 ‹ 0.03, n ˆ 20, respectively; t ˆ 11.1, df ˆ 38, completely prevent sel®ng either by preventing the ger- P < 0.001). In contrast, there was no di€erence between mination of the pollen grain or by arresting pollen tube ¯attened and full seeds for germination date, seedling growth high in the style. Cases of partial or incomplete growth rate and seedling mortality in the two popula- self-incompatibility under such incompatibility systems tions (P ˆ 0.49, P ˆ 0.82 and P ˆ 0.37, respectively). are generally rare and involve only few individuals in natural populations of plant species (Reinartz and Les 1994). If self-incompatibility in D. acutangula were Discussion governed by such a system, this high frequency of partly compatible individuals would be unusual. D. acutangula exhibits several characters, such as Inbreeding depression could generate variable seed abundant nectar production and large attractive ¯owers production upon sel®ng if di€erent individuals vary in grouped in dense in¯orescences, that attract insect visi- the number of lethal equivalents they carry. Inbreeding 349

Table 3 Mixed-model hierarchical analysis of variance on four and population and individual within population and, when adult and four progeny traits in D. acutangula. E€ects tested were available, in¯orescences within the individual by treatment inter- the population, individual nested within population, cross treat- action. For details of hypothesis testing see statistical methods in ment (distance, i.e. within-population near, within-population far the text (dfn degrees of freedom of the numerator, dfd degrees of and between population), the interactions between cross treatment freedom of the denominator, MS mean square)

Source of variation Adult traits

Fruit set per ¯ower Seed set per fruit Proportion Seed mass of ¯attened seeds

Population MS 0.022 4.414 0.253 33.16 (dfn,dfd) (1,18) (1,19) (1,18) (1,18) F 0.275 4.11 5.65 25.63*** Individual MS 0.081 1.082 0.045 1.372 (Population) (dfn,dfd) (18,42) (18,34) (18,40) (18,41) F 1.335 1.94 1.286 117.1*** Treatment MS 0.044 4.201 0.249 0.029 (dfn,dfd) (2,37) (2,36) (2,37) (2,46) F 0.746 7.534** 7.163* 2.402 Population ´ Treatment MS 0.023 7.516 0.206 1.04 ´ 10)2 (dfn,dfd) (2,37) (2,35) (2,37) (2,46) F 0.391 13.5*** 5.837* 0.872 Individual (Population) MS 0.059 0.589 0.035 1.15 10)2 ´ Treatment (dfn,dfd) (36,172) (36,168) (36,165) (36,192) F 0.608 0.579 0.877 0.703 In¯orescence [Individual MS 1.061 0.016 (Population)] ´ Treatment (dfn,dfd) (167,1147) (166,1094) F 12.17*** 0.96 Error MS 0.097 8.72 ´ 10)2 0.039 1.69 ´ 10)2 Progeny traits Seed germination Germination date Seedling mortality Seedling growth rate

Population MS 1.22 ´ 10)3 3.217 0.039 0.014 (dfn,dfd) (1,18) (1,18) (1,18) (1,18) F 2.88 ´ 10)3 0.043 0.272 5.32 ´ 10)5 Individual MS 0.428 81.03 0.143 1.54 ´ 10)2 (Population) (dfn,dfd) (18,58) (18,41) (18,77) (18,42) F 169.3*** 98.55*** 34.8*** 20.45*** Treatment MS 7.05 ´ 10)4 0.683 0.010 3.55 ´ 10)5 (dfn,dfd) (2,43) (2,48) (2,48) (2,49) F 0.317 0.81 3.074 0.046 Population ´ Treatment MS 2.14 ´ 10)3 0.202 2.99 ´ 10)3 1.27 ´ 10)3 (dfn,dfd) (2,43) (2,48) (2,48) (2,49) F 0.962 0.239 0.914 1.642 Individual (Population) MS 2.07 ´ 10)3 0.807 2.87 ´ 10)3 7.27 ´ 10)3 ´ Treatment (dfn,dfd) (36,165) (36,188) (36,165) (36,190) F 0.167 0.518 0.093 0.531 In¯orescence [Individual MS 1.584 1.38 ´ 10)3 (Population)] ´ Treatment (dfn,dfd) (166,1053) (166,1025) F 1.274 1.074 Error MS 0.012 1.244 0.031 1.28 ´ 10)3

*P < 0.025; **P < 0.005; ***P < 0.0005; n.s. P > 0.025 depression was clearly detectable in this study because an incompatibility mechanism because a proportion of experimental sel®ng resulted in a larger proportion of ¯attened seeds germinated and were clearly not aborted. ¯attened seeds. It is unlikely that seed ¯atness resulted We therefore suggest that seed ¯atness, with low seed from a lack of resources available to the mother plants germination, might re¯ect early acting inbreeding de- since very few ¯attened seeds were obtained in outcross pression, with the expression of recessive deleterious treatments on the same plants (Fig. 1c,d). Seed ¯atness alleles in homozygous condition. Inbreeding depression might result from either late-acting self-incompatibility has also been found in Theobroma cacao by Warren et or early acting inbreeding depression. Charlesworth al. (1995) within progenies of self-compatible individuals (1985) and Sage et al. (1994) pointed out that it is gen- in natural populations. On the other hand, we found no erally dicult to assess the relative role of these two evidence of inbreeding depression in seedling growth phenomena in explaining the failure of sel®ng. It is un- rate or seedling mortality rate, possibly because it was likely that seed ¯atness resulted from late expression of less likely to be expressed under greenhouse conditions 350

ibility in T. cacao involves at least one S locus with several alleles ± at least ®ve according to Knight and Rogers (1955) and six according to Cope (1962). These alleles show dominance and/or an independence rela- tionship. The diploid constitution of the male and the female parent determine the success or failure of the cross. Indeed, the dominant allele at one locus inhibits the incompatibility power of the recessive allele at the same locus. In such a system, homozygous genotypes are those which exhibit the complete failure of self-pollina- tion while all heterozygotes are able to produce some seeds if self-pollinated (Cope 1962). The variation in self-fertility found among the D. acutangula individuals in this study could be explained by this dominance-independence type of incompatibility system. Such a system is extremely Fig. 4. Proportion of ¯attened seeds in the two populations of D. wasteful of ovules, because incompatible pollinations acutangula for three types of cross treatment: 'within-population near' preempt ovules and will cause a marked decrease in seed (white areas), 'within-population far' (light shaded areas) and 'between production on sel®ng or when identical alleles are shared population' (dark shaded areas). Bars represent mean values over by pollen donor and recipient. The reproductive output in¯orescences and error bars represent SEs of individuals would thus be greatly reduced if only few incompatibility alleles exist in a population or if there is in the absence of competition or stress (Dudash 1990). strong spatial structure for incompatibility alleles, and Furthermore, habitually outbreeding species generally crosses occur mainly between near neighbours. In the express inbreeding depression either very early or late in Lataniers population, but not in the Malheur popula- their life cycle (Husband and Schemske 1996). tion, we found a spatial pattern for the fertility of Although variation in genetic load among individuals crosses. Crosses between two spatially distant individu- can lead to variation in successful seed production upon als at Lataniers yielded higher seed production per fruit sel®ng, the highly deleterious alleles involved in early and a lower proportion of ¯attened seeds than did embryo failure should be rare. In our study, fully half of crosses between individuals growing close to each other, the individuals failed to produce any seed, even though though no such e€ect was found at Malheur. This sug- all produced some fruits, upon sel®ng. This would re- gests that near neighbours shared incompatibility alleles quire a unrealistically high number of lethal equivalents in the Lataniers population, since the reduction in mean or highly deleterious alleles. The genetic characteristics seed set (Fig. 3) resulted from a few particular crosses necessary to generate such a high level of inbreeding that yielded no seeds (Fig. 2). This high variation in the depression are unlikely to occur in these populations. fertility of pollen recipient-donor combinations appears The populations investigated have su€ered from frag- more likely to arise from an incompatibility system than mentation and size reduction for approximately from inbreeding depression. 150 years, which is equivalent to about ten generations This di€erence in spatial pattern for the fertility of of these plants. In that time it is likely that both sel®ng crosses between the two populations may be partly and biparental inbreeding have exposed deleterious al- explained by ecological di€erences. In contrast to the leles, particularly highly deleterious ones, to selection, so population of Malheur, the Lataniers population is lo- their frequency should be low. cated near a zone of human activity and has probably An alternative explanation for variable seed produc- been subjected to frequent disturbances such as bush ®res tion among individuals after self-pollination could be a and browsing cattle since the beginning of human pres- system of incompatibility known from other species of ence, resulting in fragmentation of the habitat. This his- Sterculiaceae such as Sterculia chicha (Taroda and tory could have created genetic bottlenecks reducing the Gibbs 1982) and Cola nitida (Jacob 1973). This incom- number of incompatibility alleles and therefore the pro- patibility system leads to failure of syngamy when in- portion of compatible matings (Wright 1939; Schierup compatible pollen enters the embryo sac, but ovules are et al. 1997). Loss in allelic diversity in fragmented habitat irretrievably lost to this fusion with incompatible pollen. has been described in several other species (Byers and Such a reaction to self-pollination was ®rst found in Meagher 1992; Reinartz and Les 1994; Godt and Ham- Sterculiaceae species, and Taroda and Gibbs (1982) rick 1995). Habitat fragmentation has moreover been suggested that it might be common in this family. More shown to a€ect native insect populations (Harris 1984; recently, such late-acting self-incompatibility systems Powell and Powell 1987; Turner 1996) and reduce polli- have also been observed in other plant families (Sage nation eciency (Aizen and Feinsinger 1994). A previous et al. 1994). D. acutangula may therefore possess an study (Gigord et al., in press) suggests that pollination incompatibility system similar to that found in other limitation is higher in Lataniers. This could have in- species of Sterculiaceae. The genetic basis of incompat- creased the sel®ng rate of self-compatible individuals 351 leading to an increase of homozygous individuals both Bawa KS (1982) Outcrossing and the incidence of dioecism in self-incompatible and incompatible with one another. island ¯oras. Am Nat 119:866±871 Bosser J, Cadet T, Gue ho J, Marais W (1976) Flore des Since dispersal of fruits of D. acutangula occurs mainly by Mascareignes: La Re union, Maurice, Rodrigues. MSIRI, gravity, and seedlings and juveniles are usually observed Mauritius; ORSTOM, Paris; Royal Botanical Garden, Kew under an adult plant (personal observation), neighbour- Byers DL, Meagher TR (1992) Mate availability in small popula- ing individuals might be closely related. If pollen ¯ow is tions of plant species with homomorphic sporophytic self- limited, this would favour the complete failure of crosses incompatibility. Heredity 68:353±359 Charlesworth D (1985) Distribution of dioecy and self-incompati- among some neighbours and the expression of inbreeding bility in angiosperms. In: Greenwood JJ, Harvey PH, Slatkin M depression and thus explain the observed distance e€ect (eds) Essay's in honour of J Maynard Smith. Cambridge on seed set per fruit and on the proportion of ¯attened University Press, Cambridge, UK, pp 237±268 seeds. This was not observed in Malheur, possibly be- Charlesworth D, Morgan MT, Charlesworth B (1990) Inbreeding depression, genetic load, and the evolution of outcrossing cause pollen ¯ow was less restricted in this unfragmented rates in a multilocus system with no linkage. Evolution population and more incompatibility alleles were present 44:1469±1489 within an individual progeny. Cope FW (1962) The mecanism of pollen incompatibility in Theobroma cacao L. New Phytol 72:157±182 Dudash MR (1990) Relative ®tness of selfed and outcrossed progeny in a self-compatible, protandrous species, Sabatia an- Conclusion gularis L. (Gentianaceae): a comparison in three environments. Evolution 44:1129±1256 D. acutangula possesses characteristics that might ex- Dupont J, Girard JC, Guinet M (1989) Flore en de tresse. Le livre plain its decline in fragmented habitats. It appears to be rouge des plantes indigeÁ nes menace es aÁ La Re union. SREPEN, a poor seed disperser and its maintenance within small Saint-Denis, La Re union Friedmann F (1987) Sterculiace es. In: Flore des Mascareignes, vol populations is impared by its dependence on pollinators 53. MSIRI, Mauritius; ORSTOM Paris; Royal Botanical for reproductive success. A further problem is its highly Garden, Kew variable self-fertility, probably resulting from an Gigord L, Picot F, Shyko€ JA (in press) E€ects of habitat frag- incompatibility system which allows some sel®ng but mentation on Dombeya acutangula on the island of La Re union (Indian Ocean). Biol Conserv involves loss of ovules to incompatible pollinations, Gill F (1971) Ecology and evolution of the sympatric Mascarene and thus results both in the expression of inbreeding white-eyes, Zosterops borbonica and Zosterops olivacea. Auk depression and in a decrease in seed production. We 88:35±60 suggest reinforcement of particularly threatened popu- Gilpin M (1991) The genetic e€ective size of a metapopulation. Biol J Linn Soc 42:165±175 lations through hand-pollination and introduction of Godt MJ, Hamrick JL (1995) The mating system of Liatris helleri individuals with appropriate incompatibility alleles to (Asteraceae), a threatened plant species. Heredity 75:398±404 slow the decline of this species in its natural habitat. Harris LD (1984) The fragmented forest: island biogeography theory and the preservation of biotic diversity. University of Acknowledgements The authors thank the Gigord Foundation and Chicago Press, Chicago Professor Jacques Figier for providing ®nancial and logistical Hauser TP, Damgaard C, Loeschcke V (1994) E€ects of inbreeding support for the study, Yann Jolivet, Fre de ric Picot, Fre de rique in small populations: expectations and implications for con- Paniandy and Rene Billot for their help in the ®eld, Herve Fossy, servation. In: Loeschcke V, Tomiuk J, Jain SK (eds) Conser- DanieÁ le Gigord, Yann Jolivet, Fre de rique Paniandy and Fre de ric vation genetics. BirkhaÈ useur, Basel, pp 115±129 Picot with setting up the germination plot and the seed weight Husband BC, Schemske DW (1996) Evolution of the magnitude evaluation, Daniel Lucas and Dominique Strasberg for logistical and timing of inbreeding depression in plants. Evolution 50:54± support in the Conservatoire National Botanique de Mascarin, le 70 MinisteÁ re de l'Environnement contrat SOFT no. DGAD/SRAE/ Jacob VJ (1973) Self-incompatibility mechanism in Cola nitida. 95050, the Universite de La Re union and the Conseil Re gional de Incompat Newsl 3: 60±61 l'õÃle de La Re union for ®nancial support. Janzen DH (1971) Seed predation by animals. Annu Rev Ecol Syst 2:465±492 Kimura M (1983) The neutral theory of molecular evolution. Cambridge University Press, Cambridge References Knight R, Rogers HH (1955) Incompatibility in Theobroma cacao. Heredity 9:69±77 Aizen MA, Feinsinger P (1994) Forest fragmentation, pollination, Lynch M, Conery J, BuÈ rger R (1995) Mutation accumulation and and plant reproduction in a Chaco dry forest, Argentina. the extinction of small populations. Am Nat 146:489±518 Ecology 75:330±351 Manasse RS, Pinney K (1991) Limits to reproductive success in a Are neÁ s J (1959) Flore de Madagascar et des Comores. Sterculia- partially self-incompatible herb: fecundity depression at serial ce es. Museum National d'Histoire Naturelle, Paris life-cycle stages. Evolution 45:712±720 Baker HG (1967) Support for Baher's law as a rule. Evolution Menges ES (1991) The application of minimum viable population 21:853±856 theory to plants. In: Falk DA, Holsinger KE (eds) Genetics and Baker HG, Cox PA (1984) Further thoughts on dioecism and conservation of rare plants. Oxford University Press, Oxford, islands. Ann Miss Bot Gard 71:244±253 pp 45±61 Barrett SCH, Charlesworth D (1991) E€ects of a change in the level Mueller-Dombois D, Loope LL (1990) Some unique ecological of inbreeding on the genetic load. Nature 352:522±524 aspects of oceanic island ecosystems. In: Lawesson JE, Hamann Barrett SCH, Kohn JR (1991) Genetic and evolutionary conse- O, Rogers G, Reck G, Ochoa H (eds) Botanical research and quences of small population size in plants: implications for management in Galapagos. Monogr Syt Bot Miss Bot Gard conservation. In: Falk DA, Holsinger KE (eds) Genetics and 32:21±28 conservation of rare plants. Oxford University Press, Oxford, Nettancourt D de (1977) Incompatibility in angiosperms. Springer, pp 3±30 Berlin Heidelberg New York 352

Noss RF, Csuti B (1994) Habitat fragmentation. In: Me€e GK, Schierup MH, Vekemans X, Christiansen FB (1997) Evolutionary Caroll CR (eds) Principles of conservation biology. Sinauer, dynamics of sporophytic self-incompatibility alleles in plants. Sunderland, pp 237±264 Genetics 147:835±846 Powell AH, Powell GVN (1987) Population dynamics of male Schonewald-Cox C (1985) Genetics, minimum population size, and euglossine bees in Amazonian forest fragments. Biotropica the island preserve. In: Stone CP, Scott JM (eds) Hawaii's 19:176±179 terrestrial ecosystems: preservation and managment. University Ramsey M, Vaughton G (1996) Inbreeding depression and polli- of Hawaii, Honolulu, pp 432±458 nator availability in a partially self-fertile perennial herb Seyani JH (1991) The genus Dombeya (Sterculiaceae) in continental Blandfordia grandi¯ora (Liliaceae). Oikos 76:465±474 Africa. Maise, National Botanical Garden of Belgium Reinartz JA, Les DH (1994) Bottleneck-induced dissolution of self- Sokal RR, Rohlf FJ (1981) Biometry. Freeman, San Francisco incompatibility and breeding system consequences in Aster Soule ME (1987) Viable populations for conservation. Cambridge furcatus (Asteraceae). Am J Bot 81:446±455 University Press, Cambridge Rice WR (1989) Analyzing tables of statistical tests. Evolution Taroda N, Gibbs PE (1982) Floral biology and breeding system 43:223±225 of Sterculia chicha St. Hil. (Sterculiaceae). New Phytol 90:735± Richman AD, Kohn JR (1996) Learning from rejection: the 743 evolutionary biology of single-locus incompatibility. Trends Turner IM (1996) Species loss in fragments of tropical rain forest: Ecol Evol 11:497±502 a review of the evidence. J Appl Ecol 33:200±209 Sage TL, Bertin RI, Williams EG (1994) Ovarian and other late- Warren JM, Kalai, Misir S (1995) An unnatural breeding system acting self-incompatibility systems. In: Williams EG, Clarke polymorphism in cacao (Theobroma cacao, Sterculiaceae) in AE, Knox BR (eds) Genetic control of self-incompatibility and Trinidad. Am J Bot 82:1126±1130 reproductive development in ¯owering plants. Kluwer, Wright S (1939) The distribution of self sterility alleles in popula- Dordrecht, pp 1116±1140 tions. Genetics 24:538±552 SAS (1990) SAS user's guide, version 6, 4th edn. SAS Institute, Young A, Boyle T, Brown T (1996) The population genetic con- Cary NC sequences of habitat fragmentation for plants. Trends Ecol Evol Saunders DA, Hobbs RJ, Margules CR (1991) Biological conse- 11:413±418 quences of ecosystem fragmentation: a review. Conserv Biol 5:18±32