Gender variation and primary succession in a tropical woody plant, Antirhea borbonica (Rubiaceae)

Authors

  • ISABELLE LITRICO,

    Corresponding author
    1. UMR 5175 Centre d’Ecologie Fonctionnelle et Evolutive, CNRS, 1919 Route de Mende, 34293 Montpellier cedex 5, France
    2. UMR C53 Peuplements végétaux et bioagresseurs en milieu tropical, Université de La Réunion, 15 avenue René Cassin 97400 St Denis, La Réunion
      Isabelle Litrico, Centre d’Ecologie Fonctionnelle et Evolutive, CNRS, Montpellier, France (tel. +33 4 67 61 32 14; fax +33 4 67 41 21 38; e-mail isabelle.litrico@cefe.cnrs.fr).
    Search for more papers by this author
  • THIERRY PAILLER,

    1. UMR C53 Peuplements végétaux et bioagresseurs en milieu tropical, Université de La Réunion, 15 avenue René Cassin 97400 St Denis, La Réunion
    Search for more papers by this author
  • JOHN D. THOMPSON

    1. UMR 5175 Centre d’Ecologie Fonctionnelle et Evolutive, CNRS, 1919 Route de Mende, 34293 Montpellier cedex 5, France
    Search for more papers by this author

Isabelle Litrico, Centre d’Ecologie Fonctionnelle et Evolutive, CNRS, Montpellier, France (tel. +33 4 67 61 32 14; fax +33 4 67 41 21 38; e-mail isabelle.litrico@cefe.cnrs.fr).

Summary

  • 1Ecological context can significantly affect plant reproduction: in particular, differences in the relative allocation of resources to male and female function can contribute to gender variation.
  • 2We quantified variation in sex expression among populations of Antirhea borbonica, a sexually dimorphic woody pioneer species found both on young lava flows and in later stages of primary succession on the volcanic island of La Réunion. We also tested whether environmental conditions influence maternal fertility by experimentally manipulating resource levels in natural populations.
  • 3The polliniferous morph showed significant gender variation, from almost zero fruit production in pioneer populations to high fruit set, albeit of less than females, in late-succession. Repeated observations over 4 years showed that, even in pioneer populations, most polliniferous plants are capable of producing fruits.
  • 4A significant increase in fruit production by the polliniferous morph was observed after 2 years of resource supplementation in a pioneer population.
  • 5Sex ratios were close to 1 : 1 in all but one of 12 populations. Seeds and seedlings produced by the polliniferous morph in a late-succession population had significantly lower viability than those from females. These results indicate that there is strong selection against the maternal offspring produced by polliniferous plants and that, regardless of successional status, A. borbonica possesses a functionally dioecious sexual system.
  • 6The combination of multi-year surveys and experimental manipulation of resource availability provides evidence of phenotypic gender plasticity. This provides a novel illustration of how the heterogeneous environment of primary succession can influence trait variation.

Introduction

In plants, sex expression can be highly variable within and among populations. In several species, variation in sex expression among habitats is associated with a shift from hermaphrodism to unisexuality, i.e. gender specialization, in habitats which incur resource limitation and/or drought stress (Costich 1995; Wolfe & Shmida 1997; Ashman 1999; Humeau et al. 2000; Case & Barrett 2001). These patterns fit the predictions of theoretical models that emphasize the role of resource limitation for the optimal allocation of resources to male and female function and the evolution of dioecy (the occurrence of males and females) from hermaphrodism (Charnov 1982; Charlesworth 1999). In some cases, the effects of resource limitation on the occurrence of different gender strategies in contrasting habitats may also stem from differences in either survival or the expression of inbreeding depression (Case & Barrett 2004). In addition, gender variation may be closely related to plant size (and/or age); for example, in perennial hermaphroditic species relative allocation to female function may increase with size (e.g. Wright & Barrett 1999; Méndez & Traveset 2003).

In species which show some form of gender variation, it is common to observe more labile gender expression on pollen-producing plants than on females (reviewed by Lloyd & Bawa 1984; Webb 1999). In gynodioecious species, whereas females are usually constant in their sex expression and, in most species, show significantly enhanced female function (seed size and number, germination rates and seedling vigour) relative to hermaphrodites (reviewed by Webb 1999), hermaphrodites are often more labile, in terms of both fruit and seed production (Delph 1990; Delph & Lloyd 1991; Sakai & Weller 1991; Thompson & Tarayre 2000; Delph & Carroll 2001) and progeny performance (Thompson et al. 2004). Reduced viability of the maternal offspring produced by polliniferous plants may be caused by sexual specialization, inbreeding depression and/or features of the genetic control of sex expression (Delph & Mutikainen 2003; Thompson et al. 2004). Reduced female function, and thus specialization towards male function, in the polliniferous morph is predicted to be correlated with the population sex ratio (Lloyd 1976). As the polliniferous morph becomes more male, and the population moves towards functional dioecy, the sex ratio should tend towards 1 : 1.

Gender lability in polliniferous plants may be caused by genetic and/or environmental factors. Experiments in controlled conditions have shown evidence of genetically based differences in gender among hermaphrodites in some species, e.g. Thymus vulgaris (Thompson & Tarayre 2000; Thompson et al. 2004). Likewise, experimental manipulation of plants with contrasting gender expression has provided evidence of gender plasticity in relation to resource availability in several species (e.g. Delph 1990; Costich 1995; Vallius & Salonen 2000; Dorken & Barrett 2004). Such data are critical to our understanding of the role of ecological context in the evolution of gender dimorphism (Ashman 1999).

Gradients of primary succession provide an intriguing system for the study of how plants may alter sex expression in relation to environmental heterogeneity, in particular variation in resource availability. In the initial stages of succession, plants establish on barren substrates that lack a true soil cover: run-off and percolation are rapid and available organic and nitrogenous compounds occur only in very small amounts (del Moral 1993; Miles & Walton 1993; Kirman 2003). Hence, in early-succession, resource limitation and aridity may dramatically affect establishment, growth and reproduction. Although some authors (Gray 1993; Thompson et al. 1993) have argued that phenotypic plasticity plays an important role in trait variation during primary succession, the potential occurrence of gender plasticity in relation to resource variation during primary succession has not been studied.

In this paper we quantify variation in reproductive traits in early- and late-succession populations of the pioneer woody species Antirhea borbonica J.F. Gmel (Rubiaceae). In addition, we experimentally test whether resource limitation contributes to variation in maternal fertility. Although A. borbonica has been suggested to be dioecious (Verdcourt et al. 1989), preliminary observations during two years of flowering (1995–96) indicated that males often produce small numbers of fruits, although females do not produce pollen (J.D. Thompson and T. Pailler personal observations). We will thus refer to the two sexual phenotypes as ‘polliniferous’ and ‘female’, respectively. The first of our objectives is to quantify the phenotypic gender of plants in the different stages of succession by asking the following questions. (i) Is sex expression (flower size and number and fruit and seed production) variable among populations, particularly between those in early and late-succession? (ii) Is seed derived from the polliniferous morph similar in biomass and germination capacity compared with seed from females? (iii) Are sex ratios similar to 1 : 1 in natural populations as might be expected in a functionally dioecious species? Our second objective is to examine whether variation in sex expression can be associated, at least in part, with resource limitation. In this context we examine whether resource limitation in pioneer populations can contribute to gender specialization and whether gender variation is correlated with differences in tree size among populations. Our approach involved repeated observations of flowering and fruiting in natural populations during four consecutive years coupled with an experimental study of the effects of resource supplementation on growth and fruit set.

Methods

study species and sites

Antirhea borbonica is endemic to La Réunion and Mauritius. This pioneer tree species is common on young lava flows that have reached at least ∼25 years of age and also persists through different successional stages into climax rain forest (Strasberg 1996). Trees in older forests may be more than 100 years old. A. borbonica produces small, tubular, white flowers which are visited by a range of insects (Lepidoptera, Diptera and Hymenoptera) and fleshy fruits which are dispersed by birds, mostly native Zosterops spp. and introduced Pycnonotus jocosus (T. Pailler personal observations). Each fruit contains one seed that is enclosed in an endocarp.

On the island of La Réunion (Indian Ocean), vegetation colonization on different-aged lava flows on the slopes of the Piton de la Fournaise volcano has produced a chronosequence of primary succession. Currently, all lava flows occur on the east flank of the caldera in the zone known as the ‘Grand Brulé’, which therefore consists of fragmented pockets of natural lowland tropical rain forest in a landscape of recent lava flows. The lava flows are chemically similar, occur in the same climatic zone, do not contain a seed bank when they are created and their ages are well documented because of the high frequency of lava flows in the last 100 years (Strasberg 1994). The establishment of species on the lava flows thus requires dispersal and occurs in similar environmental conditions in different sites. The main environmental differences among sites are thus associated with vegetation succession.

We chose to work at 12 sites on La Réunion and one site on Mauritius (Table 1). On La Réunion, eight of these populations occur in the Grand Brûlé at < 300 m elevation, and therefore in a highly disturbed landscape where population extinction and re-colonization are frequent. Sites 1–6 contain populations in pioneer vegetation on lava flows which occurred in the last 100 years, but sites 7 and 8 represent fragments of forest which are, nevertheless, less than 300 years old. Four populations occur outside of the Grand Brûlé in less fragmented climax forest. One population (site 9) occurs in the only remaining piece of natural lowland rain forest outside of the Grand Brûlé on La Réunion in the Mare Longue nature reserve (∼300 m elevation). The three remaining populations occur at 1000–1500 m elevation outside of the active caldera at the altitudinal limits of the distribution of A. borbonica on La Réunion. They occur in semi-dry tropical forest (site 10), mid-elevation tropical forest (site 11) or tropical cloud forest (site 12). These sites cover the range of the distribution of this species across the island and in different forest types. In addition, we were able to sample one population in the Pétrin natural reserve on Mauritius.

Table 1.  Characteristics of the studied populations
SiteCodeLava flow ageSuccessional status and community type
1C611961Early-succession: very young population with few reproductive individuals
2C43a1943Early-succession: young population with many reproductive individuals
3C43b1943Early-succession: young population with many reproductive individuals
4C43c1943Early-succession: young population with many reproductive individuals
5C311931Early-succession: young population with many reproductive individuals
6C191900Early-succession: young population with many reproductive individuals
7KP1200–300 yearsMid-succession: forest fragment untouched by recent lava flows
8KP2200–300 yearsMid-succession: forest fragment untouched by recent lava flows
9MAL> 500 yearsLate-succession: climax forest in the Mare Longue nature reserve at ∼300 m elevation
10DOD> 500 yearsLate-succession (Piton des Neiges volcano): semi-dry climax tropical forest at ∼1000 m near Dos D’âne
11CIL> 500 yearsLate-succession (Piton des Neiges volcano): climax mid-altitude tropical forest at ∼1000 m near Cilaos
12BEB> 500 yearsLate-succession (Piton des Neiges volcano): climax tropical cloud forest (Bébour) at 1400 m elevation
13MAU> 500 yearsPétrin natural reserve on Mauritius

reproductive traits and plant size

In January–February 2001, three flowers per tree were randomly collected on 10–20 individuals of each morph in the C43b, C19, KP1, MAL, DOD, BEB and MAU populations to measure corolla length, style length and anther height. Measurements were made with digital callipers to 0.01 mm. Mean values were calculated for each individual plant. In October 2002, three flowers were randomly collected on 10 trees of each morph in the C43b and MAL populations to quantify the number of pollen grains (based on 2–3 replicate observations of a single dehisced anther per flower), the number of ovules per flower (counted under an optical microscope) and the proportion of open stigmas per plant.

Between October 2003 and April 2004 we collected data on vegetative size and flower and fruit production in all 12 study populations on La Réunion. We measured the diameter (1.30 m d.b.h.), height, width and number of branches on a total of 852 trees in the following populations: C61 (n = 82), C43a (n = 90), C43b (n = 136), C43c (n = 60), C31 (n = 30), C19 (n = 104), KP1 (n = 134), KP2 (n = 30), MAL (n = 84), CIL (n = 30), BEB (n = 67) and DOD (n = 65). On 15–20 individuals of each morph, we determined the mean number of inflorescences per node (10–20 nodes randomly sampled from throughout each tree) and the mean number of flowers and fruit per inflorescence (5–10 inflorescences randomly sampled from throughout the tree).

In January/February 2001, 2002, 2003 and 2004 we estimated mean seed mass on 15–20 females and 8–14 plants of the polliniferous morph (which set fruit) in the MAL population (seeds were randomly sampled from fruits collected throughout the tree). For each individual, mean seed mass was determined based on 5–10 seeds. Using seeds produced in the MAL population in 2002, seed germination was quantified in greenhouse conditions at the CEFE-CNRS experimental gardens in Montpellier (southern France). In March 2002, 20 seeds from each of 15 female trees and 8–20 seeds from eight trees of the polliniferous morph (the mean seed mass of five seeds was determined for each individual) were sown, 10 seeds per pot (15 × 8 cm), in a mixture of one-third sand and two-thirds garden soil. Pots were arranged in a single randomized block in a temperature-controlled glasshouse (25–27 °C and 90% humidity). Seed germination and seedling mortality were recorded weekly for 10 weeks (after which no further seed germination was observed). In July, three seedlings from each maternal plant were individually potted in the same potting compost and their height measured monthly until February 2003.

Sex ratios were estimated on a sample of roughly 60–100 reproductive trees in 2004 in each of the 12 La Réunion populations. In the very young C61 population we could only find 16 reproductive individuals in 2004 (the previous year we saw no flowering trees at this site).

experimental resource supplementation

A resource supplementation experiment was carried out from September 2001 to March 2004 in two populations, C43b and MAL. In the C43b population, 10 trees of each sexual phenotype were randomly allocated to one of three treatments, resource supplementation, control and defoliation, to give a total of n = 60 trees. To assess how to perform defoliation we quantified vegetative growth of fertilized and control individuals from October 2001 to May 2002, randomly marking a single node (with colour-coded tape) on each of three branches per tree and noting monthly increases in the number of nodes per branch. In October 2002, we removed a number of leaves per branch equivalent to the increase in leaf number observed in the resource supplementation treatment (relative to controls) in year 1. In the MAL population we were unable to carry out this defoliation treatment because of the size of trees, and hence n = 40 for this site. In the resource supplementation treatment, 100 g of NPK fertilizer (solid granules) was sprinkled on a single date each month under the canopy of treated trees, over three growing seasons: November 2001 to March 2002, November 2002 to March 2003 and November 2003 to March 2004. We quantified the number of inflorescences per node on 10 randomly sampled nodes and the number of flowers and fruits per inflorescence on 10 randomly sampled inflorescences monthly.

data analyses

All data analyses were conducted using SAS (2001). Vegetative traits, floral traits, pollen and ovule number per flower, inflorescence number per node and flower number per inflorescence were analysed in anova using PROC GLM. Population, sex and their interaction were treated as fixed effects. Seed mass and seedling height were analysed in anova using PROC GLM and sex was treated as fixed effect. In these analyses data were log or square root transformed where necessary to satisfy anova assumptions. To evaluate differences among populations, pair-wise means comparisons (Lsmeans) were made with the Bonfferoni significance level adjustment. In the analyses of inflorescence number per node, an estimation of vegetative size (diameter * height * width * number of branches/4) was included as a covariate. We analysed deviation of sex-ratios from 1 : 1 using heterogeneity G-tests.

The proportion of open stigmas, individuals bearing fruit and fruit number per flower were analysed using PROC GENMOD, with population, sex and their interaction as fixed effects. To analyse seed germination, PROC GENMOD was used to test for variation between sexual phenotypes. In the analyses of seed germination, mean seed mass of the maternal parent was included as a covariate.

To illustrate any variation in sex expression within and among populations and morphs we calculated the standardized phenotypic gender (G) following Lloyd & Bawa (1984). This provides an estimate of the relative allocation of individuals in a population to male (number of polliniferous flowers) and female (ratio of fruit number to flower number) function. Phenotypic gender was calculated for a given individual (i) in each population as:

Gi = Fni/[Fni + (PFni × E)]

where Fni = fruit number on tree i, PFni = polliniferous flower number on tree i, E = ΣFnpop/ΣPFnpop, Fnpop = fruit number on all individuals in a population and PFnpop = polliniferous flower number on all individuals in a population.

To test the effect of resource supplementation on vegetative growth, inflorescence number per node and the number of flowers per inflorescence, we used PROC MIXED with the variable ‘year’ incorporated using a heterogeneous covariance matrix (CSH). In this analysis, individual was treated as a random effect in anova and sex, treatment, year and their interaction were treated as fixed effects. Data for fruit set and the proportion of individuals with at least one fruit were analysed with PROC GENMOD and Generalized Estimating Equations (GEE) models with a ‘repeated’ statement. Pairwise means comparisons (Lsmeans) were made with Bonfferoni adjustment. In these analyses resource accumulation and time effects are grouped together within the variable ‘year’.

Results

sex ratios and gender

Sex ratios were not significantly different from 1 : 1, both globally (d.f. = 11, G2 = 5.69, P > 0.5 in 2004) and within each population (Table 2), with the exception of a biased sex ratio in the BEB population.

Table 2.  Mean diameter (cm), sex ratio values and G-test results, and percentage of fruit per flower on the polliniferous morph in 12 populations of Antirhea borbonica on La Réunion
PopulationSample sizeMean trunk diameter (± SE)Proportion of female morphsG2% fruit per flower on the polliniferous morph
  1. ns, P > 0.05; *P < 0.05.

C61161.28 (± 0.328)0.430.25 ns 0.26
C43a801.70 (± 0.101)0.520.20 ns 0.00
C43b811.34 (± 0.089)0.480.11 ns 0.09
C43c742.51 (± 0.136)0.460.49 ns 0.66
C31616.27 (± 0.799)0.460.41 ns 2.51
C19965.08 (± 0.347)0.432.05 ns 3.72
KP1817.13 (± 0.366)0.441.50 ns13.23
KP2788.80 (± 0.834)0.480.21 ns12.10
MAL957.80 (± 0.393)0.500.01 ns50.22
CIL715.61 (± 0.433)0.510.01 ns 6.86
DOD7510.7 (± 0.570)0.460.65 ns 4.80
BEB879.51 (± 0.778)0.385.12*22.61

A highly contrasting pattern of fertility variation was observed for the two sexual phenotypes. The standardized phenotypic gender (G) illustrates that female function of the polliniferous morph was greater in late-succession populations than in early-succession populations (Fig. 1). In 2004, the proportion of the polliniferous morph bearing fruit varied significantly among populations (d.f. = 11, χ2 = 78.08, P < 0.001) owing to late-succession populations having higher values than early-succession populations (Fig. 2a). Whereas the percentage of trees setting fruit was always < 20% in the early-succession populations, 50% of polliniferous plants set fruit in the late-succession populations. The percentage of females setting fruit was consistently close to 100% in all populations. Although this pattern was observed in each of the four years of study, the proportion of trees of the polliniferous morph setting fruit in the early-succession populations was equivalent (roughly 70%) to that in the late-succession populations when all four years were grouped together (Fig. 3).

Figure 1.

Illustration of phenotypic gender ‘G’ in (a) the early-succession C43a population and (b)–(d) three late-succession populations [(b) KP2, (c) DOD, (d) BEB] of Antirhea borbonica on La Réunion. In each figure the polliniferous morph is shown by closed circles and females by open circles.

Figure 2.

Maternal fertility of the polliniferous morph (closed squares) and females (open squares) in 12 populations of Antirhea borbonica on La Réunion in 2004: (a) proportion of individuals bearing fruit and (b) the mean (± SE) proportion of flowers with fruit. The value of some points is noted on the figure because of extremely low values. These populations differ in successional stage from early (left of figure) to late (right of figure) succession (see also Table 1).

Figure 3.

Cumulative proportion of individuals of the polliniferous morph (closed squares) and females (open squares) bearing fruit in three consecutive years (2002–04) in six populations of Antirhea borbonica on La Réunion. These populations differ in successional stage from early (left of figure) to late (right of figure) succession (see also Table 1).

Fruit/flower number (fruit set) of the polliniferous morph was also significantly less than for females and increased in late-succession (Fig. 2b). Fruit set varied significantly among populations (d.f. = 11, χ2 = 3092, P < 0.001) and sexes (d.f. = 1, χ2 = 12 208, P < 0.001) and showed that a population-by-sex interaction was significant (d.f. = 11, χ2 = 2913, P < 0.001).

variation in floral traits, flower production and plant size

The two sexual phenotypes showed clear-cut differences in floral morphology (Fig. 4). Flowers of the polliniferous morph had significantly longer corolla tubes (F1,287 = 1650, P < 0.001), with anthers positioned higher in the flower (F1,287 = 1864, P < 0.001) and shorter styles (F1,287 = 310, P < 0.001) than female flowers. All three traits showed significant variation among populations (F6,287 = 4.15, P < 0.001 for corolla length; F6,287 = 5.61, P < 0.001 for anther height; F6,287 = 6.26, P < 0.001 for style length), but no significant population-by-sex interaction.

Figure 4.

Mean (± SE) values of (a) style length, (b) anther height and (c) corolla length in the polliniferous morph (closed squares) and females (open squares) in six populations of Antirhea borbonica on La Réunion and one population on Mauritius. These populations differ in successional stage from early (left of figure) to late (right of figure) succession (see also Table 1).

Inflorescence number per node showed significant variation among populations (F11,335 = 4.94, P < 0.001), but no significant variation due to sexual phenotype (F1,335 = 0.04, P > 0.5) and no population-by-sex interaction (F11,335 = 1.73, P > 0.05). The covariable (vegetative size) had a significant effect on inflorescence number (F1,335 = 15.21, P < 0.001). Flower number per inflorescence varied significantly among populations (F11,322 = 80.35, P < 0.001) because of a dramatic decline in flower number in the three populations in climax forest (i.e. distant from the Grand Brûlé). The polliniferous morph generally produced more flowers per inflorescence than females (F11,322 = 26.19, P < 0.001). There was no significant population-by-sex interaction for this trait.

As there are significant positive correlations among vegetative traits in the study populations (Litrico 2004), we used a single trait, trunk diameter, to illustrate variation. Trunk diameter showed significant variation among populations (F11,709 = 126, P < 0.001), due to a smaller diameter of trees in early-succession populations (Table 2). There was no significant variation among sexual phenotypes (F1,709 = 0.02, P > 0.5), and no significant sex-by-population interaction (F11,709 = 0.76, P > 0.5).

pollen and ovule production

Negligible amounts of pollen were produced by female trees in the two studied populations (on average five pollen grains per flower in the C43b population and three in the MAL population). In both populations, average pollen number per flower of the polliniferous morph (C43b: 2760 ± 178, MAL: 3795 ± 323) was significantly greater than on females (F1,37 = 577.92, P < 0.001). The population and interaction effects were not significant. Trees in the MAL population had significantly (F1,36 = 4.23, P < 0.05) more ovules per flower (polliniferous morph: 2 ± 0.04, female: 2.2 ± 0.05) than those in the C43b population (polliniferous morph: 1.90 ± 0.06, female: 1.96 ± 0.02). Ovule number per flower showed no difference between sexual phenotypes and no significant population-by-sex interaction. The percentage of open stigmas varied from 87% to 95% and showed no significant variation among populations or sexual phenotypes.

seed size and seedling production

In the MAL population, seeds produced by polliniferous plants were significantly lighter than those produced by females in 2001 (F1,33 = 10.7, P < 0.01), 2003 (F1,28 = 12.9, P < 0.01) and 2004 (F1,25 = 3.8, P < 0.05) but not in 2002 (F1,23 = 0.03, P > 0.5). Seeds from females varied significantly in biomass among populations in all four years of the study (2001: F5,107 = 35.1, P < 0.001; 2002: F5,111 = 55.9, P < 0.001; 2003: F7,100 = 35.5, P < 0.001; 2004: F8,132 = 16.3, P < 0.001), due to a significant (Scheffé means comparison at P < 0.05) increase in seed mass in the late-succession populations outside of the Grand-Brûlé.

Germination of seeds (d.f. = 1, χ2 = 31.9, P < 0.001) and seedling survival (d.f. = 1, χ2 = 12.2, P < 0.001) were significantly greater in the offspring of females compared with those of the polliniferous morph (46% vs. 88% and 70% vs. 95%, respectively). There was no significant difference (F1,18 = 1.13, P > 0.05) in seedling height among progenies of females and polliniferous morphs. The inclusion of seed germination and seedling survival in the estimate of the relative female function of each morph (i.e. proportion of individuals bearing fruits * inflorescence per node * flower per inflorescence * fruit number/flower * seed germination * seedling survival) showed that the functional female gender of the polliniferous morph was only 16% that of the female morph in the MAL population.

resource supplementation

In the C43b population, resource supplementation was associated with a significant (F1,36 = 14.2, P < 0.001) increase in mean node number per branch of both sexual phenotypes in year 1 (from 6.86 to 9.36 in the polliniferous morph and from 5.36 to 11.03 in females). In subsequent years the treatment by year interaction was significant (F3,88 = 3.27, P < 0.05) due to a significant (F2,54 = 16.39, P < 0.001; Fig. 5a) increase in inflorescence number per node on both sexual phenotypes after resource supplementation. For flower number per inflorescence, a significant treatment effect (F2,54 = 5.19, P < 0.01) was primarily due to a decrease in flower number in the defoliation treatment (Fig. 5b). Resource supplementation increased and defoliation decreased (d.f. = 2, χ2 = 9.03, P < 0.05) the number of polliniferous plants bearing fruit relative to controls in the second and third years (Fig. 5c). Resource supplementation was associated with a significant increase in fruit set in the third year of the experiment, causing a significant treatment-by-year interaction (d.f. = 5, χ2 = 49.40, P < 0.001) and a significant effect of treatment (d.f. = 2, χ2 = 18.23, P < 0.001). Fruit set thus increased after resource supplementation and decreased after defoliation (Fig. 5d). Over the three years of study, the cumulative percentage of polliniferous plants setting fruit varied from 50% in the defoliation treatment, to 80% of controls and 100% of resource-supplemented individuals. In the MAL population, no significant effects of resource supplementation were detected for any of the above traits.

Figure 5.

Mean values (± SE) for reproductive traits and maternal fertility in the resource supplementation experiment in the C43b population of Antirhea borbonica on la Réunion: (a) inflorescence number, (b) flower number per inflorescence, (c) proportion of individuals bearing fruit and (d) mean proportion of flowers bearing fruit per inflorescence. Treatments are: control (open squares), fertilized (closed squares) and defoliated (grey squares).

Discussion

Studies of gender variation in several herbaceous plants have reported variation in relation to resource availability, with unisexuality frequently favoured in stressful conditions (e.g. Costich 1995; Ashman 1999; Barrett et al. 1999; Case & Barrett 2004). In woody species, gender variation is a difficult topic to study as modifications may occur over a long period. The long-lived nature of woody species nevertheless makes them ideal candidates for the occurrence of gender plasticity, particularly for species that occur in temporally heterogeneous habitats. This issue is of primary importance for tropical floras which contain a large number of dioecious species, because these are commonly long-lived shrubs or trees (Bawa 1980; Renner & Ricklefs 1995; Sakai et al. 1995). Our study of sites that differ in age and which contain plants very different in size, combined with a field experiment in which we manipulated resource levels, provides important data for our understanding of gender variation in tropical woody species.

Our results show that Antirhea borbonica has a cryptic sexual polymorphism and shows marked gender variation during primary succession on La Réunion. In early-succession, individuals of the polliniferous morph only occasionally produce a small number of fruits, whereas, in late-succession, the fruit set of the polliniferous morph is relatively high. The response to resource supplementation strongly suggests that this variation has an environmental component.

The female morph of A. borbonica shows strict sexual specialization in female function in all studied populations. In contrast, phenotypic gender of the polliniferous morph is highly variable among populations. In a given year, the polliniferous morph in pioneer populations is composed predominantly of non-fruiting plants, compared with a majority of individuals bearing fruit in late-succession populations. These morph-specific patterns are consistent with those reported in other species, where a consistent pattern of greater gender lability of the polliniferous morph relative to females has been reported (Delph 1990; Delph & Lloyd 1991; Sakai & Weller 1991; Ashman 1994; Wolfe & Shmida 1997; Humeau et al. 1999, 2000; Thompson & Tarayre 2000; Case & Barrett 2004; Thompson et al. 2004). The patterns of gender variation we detected indicate that the evolution towards dioecy in A. borbonica may have involved transition via a gynodioecious breeding system (Webb 1999). An interesting feature of our results in this context concerns the floral morphology of A. borbonica, where females have stigmas positioned above the anthers and males have stigmas placed below the anthers. This variation points to an ancestral distylous condition, a polymorphism common in tropical Rubiaceae (Barrett & Richards 1990), which has, in some species of this family, evolved to dioecy (e.g. Pailler et al. 1998). A comparative study of closely related species would provide more precise information on the pathway leading to dioecy in A. borbonica.

Several features of our results suggest that the gender variation we observed is due to plasticity in sex expression of individual trees. The significant increase in vegetative growth and flowering of both sexual phenotypes and a subsequent increased propensity for fruit production in the polliniferous morph following resource supplementation in the early-succession population illustrate that gender variation can be caused by resource limitation in these pioneer populations. The absence of any increase in fruit production in response to resource supplementation in the late-succession MAL population may be due to the short time period of the experiment at this site and/or greater competition at this site where A. borbonica occurs in denser and taller vegetation. The nature of the substrate in early-succession implicates resource limitation and stress, typical of early primary succession habitats (del Moral 1993; Miles & Walton 1993; Kirman 2003), as responsible for the effect there. Gender variation among populations in our study may also be due to differences in the age and/or size of the trees (tree size, which is greater in late-succession populations than in early-succession populations, is confounded with differences in population age). The effect of resources on the female function of the polliniferous morph actually occurred subsequent to an increase in vegetative size, and our experimental demonstration of an effect of resource supplementation in an early-succession population does not therefore rule out potential effects of increased size and age on gender variation among populations. Indeed, plant size is known to vary positively with increased allocation to female function in several perennial plants (e.g. Wright & Barrett 1999; Méndez & Traveset 2003). As with our study species, size can be correlated with age, and hence variation in resource availability, tree size and age may jointly contribute to gender plasticity.

Another indication that variation in sex expression results from gender plasticity rather than genetic differentiation among populations comes with the finding that when polliniferous trees in pioneer populations were observed over four years ∼70% of individuals were found to set some fruit in at least one year. These results also illustrate the importance of long-term studies for the detection and understanding of gender variation in perennial plants.

All study populations except one had a sex ratio close to 1 : 1, strongly suggesting that the female contribution of the polliniferous morph is almost always negligible. The threefold reduction in fruit (seed) per flower, the twofold reduction in seed germination and the fivefold reduction in seedling survival in the progeny of the polliniferous morph reduces its female function to only 16% of that of females, even in the late-succession MAL population where the female function of the polliniferous morph was greatest. These results were obtained in benign conditions for germination and seedling growth and the female advantage may be even greater in natural conditions. Plants of the polliniferous morph thus obtain most of their fitness via male function, even in late-succession populations with relatively high fruit set. The only population with a biased sex ratio is an old population where females may suffer a cost of reproduction that may reduce their longevity or flowering frequency, as observed by Pailler et al. (1998) for dioecious Chassalia in similar habitats on La Réunion.

There is therefore strong evidence that A. borbonica is functionally dioecious (Lloyd 1976), with the polliniferous morph specialized towards male function. Bagging of inflorescences has shown that polliniferous plants of A. borbonica can produce seed by selfing (Litrico 2004), and their reduced viability may therefore be caused by inbreeding depression. Resource compensation and the genetic control of sex expression may also contribute to the reduced female function in polliniferous plants (Delph & Mutikainen 2003). The sexual system of A. borbonica thus resembles the ‘leaky dioecy’ of other dioecious trees on La Réunion (Humeau et al. 1999) and elsewhere on oceanic islands (Baker & Cox 1984).

In A. borbonica, the evolution of dioecy may be relatively recent, and leakiness, a consequence of male sex inconstancy, may have been favoured during colonization of the islands of Mauritius and La Réunion. In the non-competitive conditions associated with island colonization, maternal progeny of polliniferous plants may have had increased prospects for survival. In contrast, in late-succession these offspring appear to make little contribution to future generations and populations are functionally dioecious. Ecological context may thus determine offspring fitness and hence the functional gender of populations in this species. Our results thus widen the literature containing experimental evidence of gender plasticity (e.g. Delph 1990; Delph & Lloyd 1991; Dorken & Barrett 2004; Ehlers & Thompson 2004) and provide a novel example of trait plasticity during primary succession.

Acknowledgements

We thank the CNRS, the Université de La Réunion and the Région de La Reunion for financial support, Marie Maistre, Anabelle Dos Santos, G. Debussche, Christian Collin and Gabriel D’argent for practical help, and Sandrine Maurice, Spencer Barrett and two anonymous referees for useful comments on a preliminary version of the manuscript.

Ancillary