Mating system variation along a successional gradient in the allogamous and colonizing plant Crepis sancta (Asteraceae)

Authors


 P.-O. Cheptou, Centre d'Ecologie Fonctionnelle et Evolutive (CNRS), 1919 Route de Mende, F-34293 Montpellier Cedex05,France. Tel.: 33 4 67 61 33 05; fax: 33 4 67 41 21 38; e-mail: cheptou@cefe.cnrs-mop.fr

Abstract

We analysed mating system in an annual and colonizing plant, Crepis sancta, that occupies different successional stages in the French Mediterranean region. Based on a previous experiment, we hypothesized that low inbreeding depression measured in young successional stages should select for selfing whereas higher inbreeding depression in old stages should select for outcrossing. Nine populations of C. sancta (Asteraceae) from contrasting successional stages were used to analyse (1) Seed set after autonomous and enforced selfing in controlled conditions and (2) outcrossing rates in natural conditions using allozymes (progeny array analysis). We found that C. sancta possesses a pseudo-self-incompatibility system and that mating system varies among populations. Allozymes revealed that the population multilocus outcrossing rates vary from 0.77 to 0.99. The lowest outcrossing rates occur in the youngest successional stages and complete outcrossing is found in old stages. The data partially agree with the predictions we made and the results are more generally discussed in the light of factors changing during succession. We did not find any evidence of reproductive assurance in the nine populations, contrary to what is often assumed as a major factor governing mating system evolution in colonizing species. We propose that mating system variation can be interpreted as the result of the balance between the cost of outcrossing and inbreeding depression in a metapopulation context.

Introduction

Mating systems in plants can be influenced by both ecological and genetic factors (Barrett & Harder, 1996). The study of how ecological factors influence mating systems was initiated by Darwin (1862) who proposed the reproductive assurance hypothesis for the evolution of selfing in plants. As plants are sessile, outcrossing requires an external agent such as wind or animal pollinators and therefore the probability of mating is directly related to the possibility of pollen transfer. In particular, colonizing species are expected to self in order to reproduce sexually with a single colonizing individual (‘Baker's rule' Stebbins, 1957). In contrast, the genetical approach considers that selfing possesses an intrinsic advantage (‘automatic advantage’sensuJain, 1976) in comparison with outcrossing because of the 3/2 transmission benefit of selfing genes (Fisher, 1941). However, this advantage may be cancelled by the deleterious effect of inbreeding on fitness (inbreeding depression). This is classically defined as one minus the relative fitness of selfed progeny compared with outcrossed progeny (Charlesworth & Charlesworth, 1987). Depending on the value of inbreeding depression (δ), either complete selfing (δ < 0.5) or complete outcrossing (δ > 0.5) have been predicted to evolve (Lloyd, 1979; Lande & Schemske, 1985). According to Lande & Schemske (1985), intermediate selfing rates (ISR) are not evolutionary stable, and therefore are considered as transient states although many empirical studies have documented ISR (see Vogler & Kalisz, 2001; for review).

Most studies on the evolution of selfing in plants consider inbreeding depression to be the key evolutionary parameter (Charlesworth & Charlesworth, 1987; Husband & Schemske, 1996). This contrasts with the reproductive assurance argument that is often put forward as a major factor governing mating system evolution in colonizing or invading organisms (Brown & Burdon, 1987). This latter view is partially consistent with the fact that many colonizing species, weeds for instance, are selfers (Price & Jain, 1981), although studies demonstrating the selective advantage of reproductive assurance are currently lacking (Holsinger, 2001). Moreover, many but not all pioneer species, follow ‘Baker's rule'(Carr et al., 1985; Abbott & Forbes, 1993; Sun & Ritland, 1998) indicating that reproductive assurance is not necessarily the main factor governing the evolution of selfing in such species.

Partially self-incompatible colonizing plants in ruderal habitats provide the opportunity to study selection on reproductive traits during colonization (Sun & Ritland, 1998). The species studied here, Crepis sancta, has a self-incompatibility system. But, partial self-fertility has been found in a previous study (Cheptou et al., 2000a). It is a pioneer of open habitats in the Mediterranean region, particularly in the early stages of succession (Imbert et al., 1999). Populations of C. sancta are subject to changes in the environment because of successional processes, which involve changes in both species diversity and density (i.e. competition). Interestingly, inbreeding depression in C. sancta was found to be strongly associated with successional process (Cheptou et al., 2000a). Based on survival and fecundity estimates, inbreeding depression measured in natural environment without interspecific competition (to simulate a new colonizing population) was found to be nearly absent, which is paradoxical for an outcrossing species. In contrast, a significant inbreeding depression was found in the presence of interspecific competition (simulating an older successional stage). This suggests that the magnitude of inbreeding depression changes in time with interspecific competition in natural populations of C. sancta.

Thus, the hypothesis we address here is that natural selection may favour outcrossing in advanced successional stages because of an increase of the magnitude of inbreeding depression within a population. In contrast, selfing may be selected in young open fields where inbreeding depression is low. In this case, the shift from selfing to outcrossing would be driven by environmentally determined variation of inbreeding depression and could result in variation in the mating system at the level of the metapopulation because of the colonization/ extinction dynamic.

The aim of this paper was to analyse the mating system of C. sancta along a successional gradient in the south of France. In order to measure the potential for selfing, we first estimated seed set after autonomous and enforced selfing in controlled environment using populations from different successional stages. Secondly, we estimated outcrossing rates with allozymes using natural progeny (see Ritland, 1990) for the same populations in order to analyse the realized outcrossing rate in natural populations.

Materials and methods

Study system

Dynamic processes of secondary succession in Mediterranean old fields have been well studied by Escarréet al. (1983) and Debussche et al. (1996). Escarréet al. (1983) described five successional stages: (1) dominance of annuals (1-year old), (2) dominance of Lamiaceae such as Thymus vulgaris, Lavandula latifolia, Satureja montana and Calamintha nepeta (10–15 years old), (3) dominance of Graminae such as Brachypodium phoenicoides, Bromus erectus and Poa pratensis (20–50 years old), (4) colonization by trees and (5) closing of the forest canopy. The focal species C. sancta (Asteraceae) is an annual weed and widespread in cultivated vineyards. These vineyards can be considered as young successional stages, in which successional processes have been stopped by human activity. In vineyards, C. sancta grows alone without competitors, which provides a quasi-experimental situation to analyse the evolution in young successional stages. This species also occurs in recently abandoned old fields where it may persist more than 20 years after abandonment (Imbert et al., 1999), but perennial grasses such as B. phoenicoides and B. erectus ultimately exclude it. Thus, C. sancta is found mostly in stages 1 and 2 although some populations can be found more rarely in older stages (Escarre, pers.obs.).

The successional process affects the size and the biomass of C. sancta individuals. In 1999, the mean (±SE) number of capitula per plant was of 86 (±14) capitula and the individual plant biomass of 6.65 (±0.8) g in a vineyard under cultivation. In a 5-year-old abandoned field, the mean number of capitula was 8.5 (±1.5) and the individual plant biomes was 0.35 (±0.07) g in the same year. Although morphological differences may greatly vary over the years because of variation in moisture level, large differences between successional stages are fairly consistent (Cheptou et al., 2000a).

Crepis sancta is a diploid plant with 2 n =10 ( Dimitrova & Greilhuber, 2000 ). It produces two types of achenes, a few peripheral ones (3–10 per head), which are heavy and have no pappus, and a large number of central achenes (70–100 per head), with a pappus ( Imbert et al., 1996 ). Reproduction is strictly sexual, starting early in the spring (March) and lasting 5–8 weeks in a given population. Self-incompatible Asteraceae posses a sporophytic and homomorphic self-incompatibility system ( Hiscock, 2000 ) but the genetics of self-incompatibility in C. sancta has not been investigated. Pollination is mostly entomophilous with generalist pollinators such as bees, which are common visitors in all successional stages (Cheptou, pers. obs). Even if the density varies among populations (see sampling populations below), population sizes are large with a minimum of a few hundred individuals.

Sampling

Three populations were sampled in the region of Montpellier in each of three sites representative of the Mediterranean landscape around Montpellier. The maximum distance between sites was about 40 km. In each site, we selected three ecologically contrasting successional populations. The three sites were not equivalent in the sense that we did not find exactly the same stages in each of the three sites.

The main characteristics of populations are shown in Table 1. The La Boissière (B) site is a mosaic of different successional ages and cultivated vineyards. The site near Claret (C) is dominated by cultivated vineyards. The site near St-Mathieu-de-Tréviers (SM) is a mosaic of different successional stages. No intermediate stage was found in this site and was replaced by a cultivated meadow of lucerne (Medicago sativa). A supplementary old population, Vmag, was used for the progeny analysis in order to replace C3 that did not flower and was extinct in 1998. It is a 10–12 year-old abandoned field. The site of Vmag was chosen because it was found to be ecologically equivalent to the site of Claret, i.e. dominated by vineyards. Density was measured at the flowering season using a 0.4 × 0.5 m quadrat. In each population, we used three transects with 20 quadrats per transect. Mean densities and standard errors are shown in Table 1.

Table 1.  Main characteristics of the sampled populations (see also text for explanations).
LocalityPopulationDescriptionDensity (SE) of C. s per square metresPredominant co-occurring plants
La BoissièreB1 (young)Vineyard in activity<0.1None
3°38′E
43°40′N
 B2 (medium)9-year-old   abandoned field28.5 (3.7)Annuals: Avena barbata, Crepis taraxacifolia,   Conyza canadensis, C. sumatrensis
 B3 (old)20-year-old   abandoned vineyards13.8 (3.7)Picris hieracioides , Aristolochia clematitis   shrub: Cornus sanguinea, Prunus spinosa Pyrus    sp. Quercus ilex, Q. pubescens
ClaretC1 (young)Vineyard in activity55 (9)None
3°54′E
43°53′N
 C2 (medium)5-year-old   abandoned field22 (4)Annuals: Avena barbata, Crepis taraxacifolia,   Conyza canadensis, C. sumatrensis
 C3 (old)15-year-old   abandoned field/Perennials: Brachypodium phoenicoides, Bromus erectus
St-Mathieu-de-TreviersSM1 (young)Vineyards in activity<0.1None
3°51′E
43°46′N
 SM2 (young/medium)Cultivated meadow45 (4.2)Medicago sativa
 SM3 (old)15 years-old   abandonned field4.8 (1.1)Perennials: Brachypodium phoenicoides, Bromus erectus
Villeneuve-les   MagueloneVmag (old)10/12-year-old   abandoned field4.8 (0.7)Annuals: Avena barbata  Perennials: Brachypodium phoenicoides,   Plantago lanceolata
3°50′E   Shrub: Ulmus minor
43°32′N

Self-incompatibility in an insect-proof greenhouse

In a previous experiment on C. sancta, Cheptou et al. (2000a) showed that self-incompatibility is not strict in this species but the seed set after selfing varies in a continuous way. They also found that the lower seed set after selfing was because of self-incompatibility and not because of inbreeding (early acting inbreeding depression) since inbred crosses caused by mating between close relatives (F=0.32) did not cause any reduction in seed set. This justifies the use of seed set after selfing as a measure of self-incompatibility.

In February 1998, an average of 20–25 plants were harvested at random in each of the nine natural populations described in Table 1, i.e. in every population except Vmag. The mean distance between sampled plants was about 5 m. Rosettes were transplanted to an insect-proof Glasshouse at the CEFE-CNRS experimental garden in Montpellier. Plants were grown in 1-L pots until the flowering season (April). The plants were protected with paper bags to avoid pollen transfer between plants. Hand-pollinations were performed when all the florets from the same capitulum were receptive (50–100 florets) as described in Cheptou et al. (2000a). The pollen was collected from several heads of the donor plant using a paintbrush, which was then brushed against recipient heads to achieve pollination (Kearns & Inouye, 1993). Three treatments were performed: (1) no pollination in order to test the possibility of autonomous self-fertilization, (2) self-fertilization by hand pollination and (3) outcross pollination for control with pollen from another genotype of the same population chosen at random. Each cross was repeated on two capitula. After fertilization, we measured the number of viable seeds and the total number of florets per head with a binocular microscope. Viable seeds are plump and dark and can be easily distinguished from unfertilized ovules. These data were used to estimate the seed/ovule ratio because each floret can potentially produce just one seed.

In previous controlled pollinations, selfed progeny have been screened by enzyme electrophoresis (unpublished data) and no foreign alleles have been found so that accidental outcrossing events in the greenhouse caused by insects can be dismissed.

Mating system analysis

Allozymes were used for the progeny array analyses. In 1998, when the seeds were mature, but prior to dispersal, open pollinated progenies were harvested in every population, except C3 that did not flower. One capitula per plant, chosen at random, was used for the analysis. The seeds collected in natural populations were sown in the greenhouse. The greenhouse conditions allow inbreeding depression at the germination stage to be minimized (<5%; Cheptou et al., 2000a) and thus reduced the potential bias in outcrossing rate estimates. In this study, we used the central achenes of capitulum to estimate mating system parameters. Eight individuals per family (capitulum) were used, as suggested by Ritland (1986) for an outcrossing species.

When the seedlings formed rosettes of 7 or 8 cm in diameter, proteins were extracted from fresh leaves with sand (Fontainebleau) in 0.1 m Tris–HCl buffer (pH: 7.5), 5% sucrose (w/w) and 0.6% Mercaptoethanol. Extracts were either used for immediate electrophoresis or conserved at −70 °C. An early survey revealed nine polymorphic enzymes systems. Gels at 12% starch were used for electrophoresis. PGI-1 (E.C. 5.3.1.9), PGD-1, PGD-2 (E.C. 1.1.1.44) and SKD (E.C. 1.1.1.25) were resolved on Histidine pH 6.5 (Soltis & Soltis, 1989). PGM-1, PGM-2 (E.C. 5.4.2.2), PGI-2, LAP (E.C. 3.4.11.1) and ADH (E.C. 1.1.1.1) were resolved on lithium borate pH 8.3 (Soltis & Soltis, 1989). Enzyme stain recipes were used as described by Soltis & Soltis (1989).

Data analysis

Self-incompatibility variation

Data from experimental cross were analysed using general linear model using Proc GLM (SAS, 1990). The response variable was the seed/ovule ratio. Because the age of each type of populations (young, medium and old) varies between site (see Table 1), the analysis was performed in each site independently. We performed a mixed-model analysis of variance (anova) considering stage (three levels) and cross treatment (three levels) as fixed effects. The genotype effect (nested in population) and its interaction with cross treatment were specified as random effects. The denominator degrees of freedom were calculated using Satterthwaite's approximation. Type III sums of squares were used to calculate F-ratios. The seed/ovule ratio was transformed using angular transformation in order to satisfy model assumptions (Sokal & Rolf, 1995). Multiple comparison of mean values was performed using Tukey test.

Progeny array analysis

Genetic data were analysed using the multilocus mating system program (MLTR) (available from K.Ritland). In this analysis, we used genotypes of progenies to estimate mating system parameters of the maternal parent harvested in the field using maximum likelihood estimates (Ritland, 1990). Outcrossing rate estimates were calculated using multilocus and average single locus estimates (respectively tm and ts) based on Ritland's mixed-mating model (Ritland & Jain, 1981). The inbreeding coefficient (F) of the parental generation is also estimated. The comparison between the two outcrossing estimates (tm–ts) provides an estimation of biparental inbreeding, i.e. inbreeding as a result of crossing with related individuals (Ritland, 1986): In this analysis tm is considered as the nonbiased outcrossing rate estimate. MLTR also estimates the correlated mating parameters; namely the correlation of selfing (rs) and the correlation of outcrossed paternity (rp) within progeny arrays (Ritland, 1989). A lack of correlation of selfing indicates that the selfing rate does not vary among families within a population. In contrast, a high correlation of selfing shows that some families are high selfers and others high outcrossers. The correlation of outcrossed paternity can be viewed as the proportion of full sibs among outcrossed sibs (Ritland, 1989) and 1/rp gives an estimation of the number of different paternal parents within progeny (Sun & Ritland, 1998). The MLTR program can estimate allelic frequencies in the pollen pool and in the ovule pool separately. When pollen and ovule population allele frequencies did not differ significantly, we constrained the equality of frequencies to increase the statistical power for other parameters. When the inbreeding coefficient of maternal parents (F) did not significantly differ from zero, we constrained it to zero. Ifthe frequency of alleles was too low (<5%) such alleles were pooled (Ritland, pers. com.). Standard errors on estimates were calculated using 100 bootstraps, the unit of resampling being the progeny array. Assuming a Gaussian distribution of estimates, Z-tests were used to compare estimates with a reference value, for instance comparison with the 100% value of outcrossing (Ritland, pers. com.).

Results

Variation of self-incompatibility

A total of 1000 capitula were analysed. For all populations, the distribution of seed/ovule ratio in autonomous selfing and selfing by hand pollination varied continuously from zero (full self-sterility) to close to one (full self-fertility). However, the distribution was strongly skewed to the right because of ratios close to zero. For the three sites, the anova detected a highly significant effect of cross treatment (Table 2) as shown in Fig. 1. In each site, the three treatments significantly differ from each other (Tukey test, P < 0.05). The effect of genotype within stage was significant (P < 0.05) for La Boissiere and Claret sites but not for St Mathieu. For La Boissiere and St-Mathieu the effect of successional stage was not significant whereas its interaction with cross treatment was highly significant. For those two sites, this significant interaction was because of the differential response of seed/ovule ratio with hand self-pollination and spontaneous self-pollination among stages (see Fig. 1). In particular, the seed/ovule ratio for hand self-pollination for B2 was significantly higher than B1 and B3 (Tukey test, P < 0.001 and P < 0.01, respectively). No significant differences were found in spontaneous selfing. For St-Mathieu, seed/ovule ratio for SM1 in spontaneous selfing was significantly higher than for SM2 (P < 0.0001) and SM3 (P < 0.001) but no significant differences were found in hand self-pollination. In contrast, for the site of Claret, the effect of stage was significant (P < 0.05) but not the interaction with treatment. This absence of interaction is because of similar response of the two selfing treatments with stage. Comparisons of mean values within treatment between stages revealed that C3 has a significantly lower seed/ovule ratio than C2 (P < 0.001) and C1 (P < 0.001) for hand self-pollination. The same pattern were found for spontaneous selfing with P < 0.01 and P < 0.05, respectively.

Table 2.  Mixed model analysis of variance for seed/ovule ratio from experimental crosses in Crepis sancta . Analysis was carried out on each site separately and the ratio was Arcsin(√) transformed.
 La Boissiere (r2=0.89)Claret (r2=0.9)St-Mathieu (r2=0.9)
d.f.Mean square d.f.Mean square d.f.Mean square 
  • *** 

    P  < 0.001;

  • ** 

    P  < 0.01;

  • P  < 0,05; ns, nonsignificant.

Stage  2 0.15ns 20.38*  20.04ns
Genotype (stage) 61 0.09*600.1* 600.09ns
Cross  216.03*** 27.29***  25.53***
Stage × cross  4 0.21** 40.03ns  40.14*
Genotype (stage) × cross108 0.06**730.06** 790.02***
Residual195 0.03 990.04 1610.025 
Figure 1.

Mean seed/ovule ratio (SE) for the three stages in each of the three sites. Note that the scale is different in each treatment: outcross pollination (white), hand self-pollination (black) and spontaneous self-pollination (grey). Shared letters indicate nonsignificant difference between stage for each cross treatment separately at the level of P  < 0.05 (exact probabilities are given in the text).

We also found a significant genotype × cross treatment interaction for the three sites indicating that some genotypes were more self-compatible than others.

Mating systems parameters in natural populations

Nine enzyme loci were polymorphic in this study of nine populations. The number of alleles per locus varied from two to six and four loci had more than four alleles. ADH could only be interpreted when plants were young because of very weak enzyme staining in older plants and thus was not screened for all populations. The mean expected heterozygosity across loci (where the expected heterozygosity for one locus is He=1–pi2, where pi is the frequency of the ith allele) was nearly equivalent and varied from 0.51 to 0.56). No trend was found with successional stages.

For the progeny array analysis, the biallelic PGM-2 locus was removed from the analysis because one allele was rare. LAP and PGD-2 gave aberrant parameter estimations, because of an abnormal excess of heterozygotes (especially LAP). As the six remaining loci were sufficiently polymorphic LAP and PGD-2 were also removed from the analysis. All the populations showed predominant outcrossing with multilocus outcrossing rates varying from 0.77 to 0.99 with an average of 0.91. However, significant selfing (i.e. t < 1:Z one-tailed test) was detected in five populations (Table 3). The multilocus outcrossing rates for old populations (B3, SM3 and Vmag) were close to one (complete outcrossing). In contrast, young populations exhibit an outcrossing rate significantly different from one. Similar to self-incompatibility measurements, populations of the Claret site showed the lowest outcrossing rates, especially C1 (77%). SM2 showed a 83% multilocus outcrossing rate. The single locus outcrossing estimates were generally lower than multilocus estimates, except for SM2 and Vmag. However, biparental inbreeding was not detected because the difference (tm–ts) was never significantly different from zero. This result indicates that mating between relatives is uncommon. All populations showed an inbreeding coefficient in maternal plant (F) nonsignificantly different from zero, which agrees with the high outcrossing rate observed as the relation F=(1 − t)/(1 + t) is expected at equilibrium (Crow & Kimura, 1970). The parameter F was then constrained to zero to increase statistical power for other parameter estimation.

Table 3.  Mating systems estimates for the nine populations (number of families studied in italics) of Crepis sancta (SEs in parentheses; 2 SE equals 95% confidence intervals). tm and ts define the multilocus and single locus outcrossing rate estimate, respectively. The outcrossing rates that significantly differ from one (i.e. complete outcrossing) are noted by a star (Z one-tailed test, P  < 0.05).
 B1 16B2 12B3 12C1 14C2 14SM1 19SM2 15SM3 16Vmag 20
F000000000
tm0.92 (0.04) *0.99 (0.02)0.99 (0.01)0.77 (0.08) *0.84 (0.05) *0.91 (0.04) *0.83 (0.08) *0.98 (0.02)0.97 (0.02)
ts0.87 (0.04) *0.97 (0.02)0.92 (0.05)0.71 (0.1) *0.81 (0.05) *0.88 (0.04) *0.86 (0.06)*0.93 (0.04)0.99 (0.01)
tmts0.05 (0.03)0.024 (0.02)0.07 (0.06)0.08 (0.05)0.03 (0.03)0.034 (0.03)−0.027 (0.07)0.05 (0.04)−0.023 (0.02)
rs0.09 (0.03)0.26 (0.01)0.24 (0.03)0.17 (0.04)0.14 (0.03)−0.58 (0.36)0.13 (0.06)0.2 (0.02)0.17 (0.03)
rp0.34 (0.1)0.62 (0.17)0.30 (0.1)0.38 (0.08)0.30 (0.08)0.41 (0.14)0.5 (0.1)0.16 (0.08)0.16 (0.06)
Paternal pool (1/rp)2.941.613.332.633.332.4326.256.25

Except for SM1, the correlation of selfing was low but differed significantly from zero for all populations (Z-test, P < 0.05). This indicates that selfing rate slightly varies among families within populations. However, no individual family estimates were performed because there were too few individuals per family. The correlation of outcrossed paternity ranged from 0.16 to 0.62 with a mean of 0.35. This parameter can be linked to the mean paternal pool (1/rp) which indicates that the number of paternal parent varied from 1.6 to 6.25. The highest numbers of paternal parents were found in the oldest populations, whereas the lowest numbers were found in SM2 and B2. Intermediate numbers of paternal parents were found in the youngest populations.

Discussion

This study shows that C. sancta is an outcrossing colonizer with high outcrossing rates, in accordance with the observation that most species of Crepis are outcrossers (Babcock, 1947). However, the interesting result of this study is the mating system variation among populations. Others colonizing Asteraceae, e.g. Senecio squalidus (Abbott & Forbes, 1993) and Centaurea solstitialis (Sun & Ritland, 1998) have been found to be self-incompatible with high outcrossing rates. In the Papaveraceae, Papaver rhoeas is also a successful colonizer and has been found to be strongly outcrossing (Campbell & Lawrence, 1981).

Pseudo self-incompatibility and its among population variation

In C. sancta self-incompatibility is not complete because seeds can be produced after selfing (either hand-selfing or spontaneous) in all populations. We confirm the partial self-incompatibility (Cheptou et al., 2000a) for nine populations unlike previous studies that have described C. sancta as a fully self-incompatible species (Imbert et al., 1996). The incompatibility system can be considered as pseudo self-fertility in that it corresponds to the definition of Levin (1996), i.e. ‘higher seed production with cross-pollen than with self-pollen’. Pseudo self-fertility has been documented in many plant families, and in particular many species of the Asteraceae (Levin, 1996). In a recent study, pseudo self-compatibility has been found in Senecio squalidus (Hiscock, 2000) although it is less self-compatible than C. sancta. Pseudo self-fertility has been proposed to be the result of modifier loci, unlinked to the self-incompatibility locus, that affect the activity of genes present at the SI-locus (Levin, 1996). The presence of the genotype by cross treatment interaction suggests polymorphism at SI-modifier loci within populations of C. sancta. Moreover we found variations in self-incompatibility levels among populations. Old populations were significantly more self-incompatible than medium populations for Claret and la Boissiere, but this trend was not significant for the site of St-Mathieu. The pattern of self-incompatibility in young populations is less clear (see below).

Natural outcrossing rate and its among population variation

Natural outcrossing rates based on multilocus estimates showed two main results. First, there were differences among sites, with the two populations from Claret having the highest selfing rates. Secondly, the populations from oldest stages of the succession were always nearly complete outcrossers and populations from young stages differed significantly from complete outcrossing. Moreover, the populations from Claret, particularly the young one, have the highest self-compatibility and the highest multilocus selfing rates. Progeny array analysis indicates that biparental inbreeding is not an important source of inbreeding and provides evidence that mating between related plants is negligible in natural populations. The correlation of outcrossed paternity (see ‘Materials and methods’) showed a mean value of 0.35 and corresponds to a mean number of fathers of 2.8 (1/rp), which is closed to values reported for annual populations of Mimulus guttatus, a bee-pollinated plant practicing partial selfing (0.24 < rp < 0.43, Dudash & Ritland, 1991). Very few studies have estimated this parameter in Asteraceae. Sun & Ritland (1998) found a much higher number of paternal parents per progeny (mean 8.8) in Centarea soltitialis using the same protocol. Our results for the nine populations show that old populations have low rp values (high number of paternal parents) whereas younger stages have higher rp values (lower number of paternal parents) than old stages. The average number of capitula of the plants and its variation among successional stages can explain this pattern. In young stages, pollinators are likely to gather pollen from a single plant because of the large number of capitula whereas in oldest stages of succession, pollinators have to gather pollen from a larger number of plants because of the small number of capitula per individual plant. Moreover, models by Harder & Barrett (1996) predict that the correlation of outcrossing should be maximum for intermediate numbers of flowers per plant, and produced experimental evidence for this prediction in Eichhornia paniculata. As the number of capitula decreases with succession, the high correlation of outcrossing in B2 and SM2 agrees with their predictions.

Factors influencing mating system evolution

Our study shows that mating system parameters vary among populations of C. sancta at the local scale (within the metapopulation). Empirical data revealing outcrossing rate variation among populations have been reported (Schemske & Lande, 1985). Mechanistic explanations for such variation have been proposed (see Schoen, 1982; Husband & Barrett, 1992) but adaptive explanations remain uncertain.

Self-fertility has been shown to vary within a population and between populations. Also, progeny array analysis revealed a heterogeneous selfing rate among families. Reproductive systems are known to respond rapidly to selection (Henny & Ascher, 1976), Petunia integrifolia (Dana & Ascher, 1985) and Phlox drummondii (Levin, 1995; Bixby & Levin, 1996). Furthermore, Good-Avila & Stephenson (2002) report inheritance of modifiers conferring self-fertility in the partially self-incompatible perennial, Camapanula rapunculoides L. (Campanulaceae). This indicates that selection on modifiers of self-fertility can respond to selection for selfing.

Based on the cost of outcrossing and inbreeding depression balance, we suggested the hypothesis that the low inbreeding depression observed in young successional stage should select for selfing whereas high inbreeding depression observed in old successional stage should select for outcrossing. Our empirical results partially validate this verbal model. In particular, the increase in self-incompatibility after hand self-pollination between middle and old populations (significant for Claret) agrees with our hypothesis. The less clear pattern in vineyards could be because of the fact that we did not know the date of the foundation of those populations, whereas, the age for more advanced stage is deduced from the community composition. Thus, it is possible that SM1 and B1 are new founded populations (as suggested by their low density) where selection for selfing would not have taken place yet. We also have some evidence that the genetic load does not vary among populations in the south. In a previous experiment, Cheptou et al. (2000b) have found nearly equivalent levels of inbreeding depression when it is measured in the same environment for SM1 and C2. Other studies concentrated on the genetic basis of inbreeding depression, assuming that its reduction is because of the purge of deleterious mutations with selfing (Holtsford & Ellstrand, 1990 and see review by Husband & Schemske, 1996). Here, we emphasize the importance of the population context in the expression of inbreeding depression.

However, as the number of capitula per plant greatly decreases with successional process, we have to envisage variation in pollination mechanisms, specifically geitonogamy as a factor that is influencing selfing. A high number of capitula per plant in new populations could increase pollen transfer within plants (among capitula) and then increase the realized or apparent selfing rate (Barrett et al., 1994). If geitonogamy could explain an increase in apparent selfing (revealed by progeny analysis), it could not explain a genetic selection for selfing in new populations. Indeed the automatic advantage of selfing depends on the ability for a selfing genotype to contribute as a pollen donor foroutcrossing in the population (Schoen et al., 1996). As a consequence, the advantage is cancelled if pollendevoted to outcrossing is used for selfing (pollendiscounting), which is precisely the case when geitonogamy occurs (Lloyd, 1992). Thus, geitonogamy cannot in itself create the favourable condition for the evolution of selfing.

We also have no evidence that reproductive assurance plays a role in the nine studied populations. As noted by Eckert & Schaffer (1998), autonomous selfing must contribute more to reproductive assurance than facilitated selfing because it does not require an external agent for pollination. Thus, the fact that seed set after autonomous selfing is much lower than seed set after hand selfing pollination suggests that reproductive assurance does not play an important role in increasing seed set in natural populations of C. sancta. Moreover, selfing is not negatively correlated with density in our nine populations. On the contrary, the higher selfing rate (≈25%) was found at a high density (C1) and populations with the lowest densities exhibit less than 10% of selfing. If low densities led to low mate availability, we would expect higher selfing in such situations because only selfed seeds should be produced. Van Treuren et al. (1994) have shown in experimental populations of the outcrossing plant Scabiosa columbaria that selfing can increase because of low mate availability. However, it is only in cases of extremely low density (distances of 100 m between plants). Such low densities do not occur in populations of C. sancta in the south of France and could explain why we did not find any evidence for reproductive assurance in our populations. At the same time, our study corroborates previous studies that failed to demonstrate a fertility advantage with selfing (Leclerc-Potvin & Ritland, 1994; Eckert & Schaffer, 1998).

In summary, we propose a scenario for the evolution of selfing in colonizing species based on the cost of outcrossing and inbreeding depression in ecological context. In this model, reproductive assurance is not required to explain the evolution of selfing, in contrast to the classical view of mating systems in colonizing plants. Moreover, our verbal model allows variation in selfing rate even when the evolution is governed by the automatic advantage of selfing and inbreeding depression. Finally, this corroborates Cheptou & Mathias′s (2001) theoretical study which showed that intermediate selfing rate or polymorphic selfing rates can be evolutionary stable under spatial and temporal fluctuations in inbreeding depression.

Acknowledgments

We thank John Thompson, Jacqui Shykoff and Kent Holsinger for helpful comments on the manuscript. We thank Gray Stirling for clarifying the text. We also thank Sandrine Garcia for helping us to perform isozymes electrophoresis This research was supported by the Centre National de la Recherche Scientifique, in addition to a grant to P.O. Cheptou from the Ministère de l'Enseignement Supérieur et de la Recherche.

Ancillary