Variation in the functioning of autonomous self-pollination, pollinator services and floral traits in three Centaurium species

Abstract

Background and Aims

Reproductive assurance through autonomous selfing is thought to be one of the main advantages of self-fertilization in plants. Floral mechanisms that ensure autonomous seed set are therefore more likely to occur in species that grow in habitats where pollination is scarce and/or unpredictable.

Methods

Emasculation and pollen supplementation experiments were conducted under laboratory conditions to investigate the capacity for, and timing of autonomous selfing in three closely related Centaurium species (Centaurium erythraea, C. littorale and C. pulchellum). In addition, observations of flower visitors were combined with emasculation and pollen addition experiments in natural populations to investigate the degree of pollinator limitation and pollination failure and to assess the extent to which autonomous selfing conferred reproductive assurance.

Results

All three species were capable of autonomous selfing, although this capacity differed significantly between species (index of autonomous selfing 0·55 ± 0·06, 0·68 ± 0·09 and 0·92 ± 0·03 for C. erythraea, C. littorale and C. pulchellum, respectively). The efficiency and timing of autogamous selfing was primarily associated with differences in the degree of herkogamy and dichogamy. The number of floral visitors showed significant interspecific differences, with 1·6 ± 0·6, 5·4 ± 0·6 and 14·5 ± 2·1 floral visitors within a 2 × 2 m² plot per 20-min observation period, for C. pulchellum, C. littorale and C. erythraea, respectively. Concomitantly, pollinator failure was highest in C. pulchellum and lowest in C. erythraea. Nonetheless, all three study species showed very low levels of pollen limitation (index of pollen limitation 0·14 ± 0·03, 0·11 ± 0·03 and 0·09 ± 0·02 for C. erythraea, C. littorale and C. pulchellum, respectively), indicating that autonomous selfing may guarantee reproductive assurance.

Conclusions

These findings show that limited availability of pollinators may select for floral traits and plant mating strategies that lead to a system of reproductive assurance via autonomous selfing.

INTRODUCTION

In flowering plants, seed production is often limited by deposition of an insufficient number of pollen or low-quality pollen (Knight et al., 2005), which may be caused by limited pollinator visitation, inefficient pollen transfer, or lack of potential mates (Ashman et al., 2004). Reduced seed set due to pollen limitation may cause selection on floral traits and plant mating strategies that lead to a system in which selfing ensures seed production (i.e. reproductive assurance) (Lloyd, 1992; Ashman and Morgan, 2004; Morgan and Wilson, 2005; Eckert et al., 2006). Reproductive assurance may result from both autonomous selfing (spontaneous within-flower selfing) and, to a lesser extent, facilitated selfing (within-flower selfing facilitated by a pollinator) (Lloyd and Schoen, 1992; Vaughton and Ramsey, 2010). However, when pollinators are missing, only self-pollination that is autonomously realized may provide reproductive assurance to a plant (Lloyd, 1992). Facilitated selfing, on the other hand, only provides reproductive assurance when mates are scarce, but generally not when pollinators are limiting.

The timing of autonomous selfing during floral development is expected to be an important trait under selection, depending on the magnitude of pollination limitation and associated consequences of selfing, such as gamete and/or seed discounting (Kalisz et al., 1999; Herlihy and Eckert, 2002; Zhang and Li, 2008; Eckert et al., 2010). Depending on the timing during a flower's lifespan, three modes of autonomous selfing are generally defined (i.e. ‘prior’, ‘competing’ and ‘delayed’ selfing sensu Lloyd, 1979). In plants that are chronically exposed to low pollinator services and limited pollen deposition, autonomous selfing can confer reproductive assurance with little or no pollen and seed discounting, irrespective of the timing of occurrence (Schoen and Brown, 1991; Lloyd, 1992). Under such conditions, it can be expected that a mechanism of ‘competing’ or even ‘prior’ selfing may offer a selective advantage to a plant, by reducing the investment in traits that may increase pollinator attraction and visitation, such as floral longevity, floral size and nectar production (Ornduff, 1969; Elle and Carney, 2003). On the other hand, plants are expected to develop a mechanism of autonomous self-pollination later in floral life (i.e. ‘delayed’ selfing) when the fitness of selfed progeny is significantly lower than that of outcrossed seeds (Cruden and Lyon, 1989; Schoen et al., 1996), when the presence of pollinators or potential mates is unpredictable (Goodwillie et al., 2005), or when the plant is characterized by a short lifespan (e.g. Barrett, 2002). Under such conditions, cross-pollination can occur early in floral life, but delayed selfing guarantees successful pollination at the end of a flower's lifespan if pollinator-mediated pollen deposition fails (e.g. Lloyd, 1979; Kalisz and Vogler, 2003, Jacquemyn and Brys, 2008).

The capacity and timing of autonomous selfing in plants is often associated with adaptations in floral morphological traits. A reduction in the spatial (i.e. herkogamy) and/or temporal separation (i.e. dichogamy) of the male (anthers) and female parts (stigmas) within flowers may affect the efficiency and timing of autonomous self-pollen deposition, because it may increase the ability to self autonomously (Moeller, 2006) and reduce outcrossing rates (Bertini and Newman, 1993; Belaoussoff and Shore, 1995; Karron et al., 1997). In addition, an increasing capacity to self autonomously may also indirectly affect other morphological floral traits such as pollen : ovule (P/O) ratios, because such plants may evolve to allocate smaller amounts of resources to traits that contribute to male function relative to female function (Lloyd, 1987; Michalski and Durka, 2009).

It can thus be hypothesized that different levels and causes of pollination limitation may lead to different evolutionary outcomes in terms of the mode and timing of selfing and concomitant changes to floral biology (Eckert et al., 2010). In this study, the capacity for autonomous selfing and the extent to which it provides reproductive assurance is investigated in three closely related short-living Centaurium species (C. erythraea, C. littorale and C. pulchellum) that occur in different habitat types. Each of these study species occurs within coastal dune areas of western Europe, but typically establish in different habitats. Centaurium erythraea generally grows in well-developed, late successional species-rich dune grasslands, whereas C. littorale develops more under dynamic and ruderal conditions, such as dune slacks, which often have a much lower diversity of flowering plants (van Tooren et al., 1983; Schat et al., 1989). Finally, C. pulchellum typically occurs in harsh, often inter-tidal pioneer situations, characterized by a poorly developed, open vegetation (van Tooren et al., 1983; Schat et al., 1989). Within these different habitat types, it can be expected that pollinator availability may also differ (Ubsdell, 1979; van Tooren et al., 1983), with C. pulchellum generally growing in the most impoverished pollinator environment and C. erythraea in the richest pollinator environment. Because each of three above-mentioned study species, are rare and endangered in Flanders and in other coastal dune areas in western Europe, knowledge about the functioning of their breeding system may also contribute to the development of appropriate conservation and management strategies.

A combination of field observations, experimental pollinations, both under natural and laboratory conditions, and floral measurements were conducted for each species to compare and contrast (a) the capacity for and timing of autonomous selfing, (b) variation in floral traits, (c) floral visitors and visitation rates, (d) the extent of pollination failure and pollen limitation within natural populations, and (e) the relationship between pollination failure and the ability to self autonomously.

MATERIALS AND METHODS

Species description

The three study species (Centaurium erythraea, C. littorale and C. pulchellum) belong to the Gentianaceae, and are characterized by a west-European more-or-less Atlantic distribution. All three species are short-lived herbs, with C. pulchellum being an annual, and C. littorale and C. erythraea being biennials, that grow in different habitats, ranging from early-successional vegetations up to dry well-developed calcareous grasslands (van Tooren et al., 1983; Schat et al., 1989). The timing of flowering differs between the three study species, with C. erythraea and C. littorale flowering from the beginning of July until the middle of August, and C. pulchellum flowering from the end of July until the beginning of September. The three species are characterized by similar looking, showy pink flowers, that are hermaphroditic and self-compatible (Ubsdell, 1979) (see Fig. 1A). Flowers do not produce any nectar and the period during which individual flowers are open and functional usually lasts for only a few days, with C. pulchellum flowers being open for 2–3 d, and C. littorale and C. erythraea flowers being open for 4 to sporadically 5 d (Ubsdell, 1979; R. Brys and H. Jacquemyn, pers. obs.). In each of the three species pollen is shed via a system of anther curling during maturation (see Fig. 1B). Anthers of C. pulchellum curl and release pollen the first day of flowering, whereas those of C. littorale curl and dehisce at day 1 or 2 and those of C. erythraea on day 2 or 3 (see Fig. 1B).

Fig. 1.

(A) Flowers of Centaurium erythraea, C. littorale and C. pulchellum illustrating the variation in flower size and herkogamy. (B) Timing of different stages of anther maturation during anthesis in C. erythraea.

The species are most often visited by pollen-gathering hoverflies (Diptera, Syrphidae), but occasionally some bees (Hymenoptera, Apidae), small flies (Empididae-Muscidae), moths and butterflies (Lepidoptera) are also documented as sporadic visitors on flowers of each of the three species (Müller, 1883; Knuth, 1909; Ubsdell, 1979; Van Rossum, 2009; R. Brys and H. Jacquemyn, pers. obs.). For all three species, reproduction can only take place by means of seeds, which are very small (<0·01 mg) and produced in large quantities.

Plant performance and floral traits

The study was conducted during the flowering season of 2009. Study sites were located along the western part of the Belgian coast between Oostduinkerke and De Panne, where each of the three species can frequently be found. Two populations (>1000 individuals) per species were chosen, and in each population 30 plants were randomly selected. These focal plants occurred at comparable and representative densities and under relatively homogenous environmental conditions. All plant and floral measurements, as well as the emasculation and supplemental pollination experiments in the field (see below), were conducted on each of these focal plants.

The total number of inflorescences and total flower production were recorded per plant. Additionally, one freshly opened flower per plant was harvested during peak flowering to determine the degree of herkogamy (the minimum distance between the stigmatic surface and the anthers), stigma height and width, petal length and width, and corolla diameter. The degree of herkogamy did not change during floral development (R. Brys and H. Jacquemyn, pers. obs.). All floral measurements were done with Image J using digital photographs taken from a cross-section of the flowers under a dissecting microscope. To determine mean pollen and ovule production rates in each of the populations studied, one flower per plant was harvested from 15 individuals per population. Flowers were sampled just before the onset of anthesis to ensure that total pollen production could be accurately measured. For each flower, one anther was excised and placed in an open, labelled 1·5-mL microcentrifuge tube to allow air drying and dehiscence during 48 h at room temperature. One drop of ethanol was added to each tube prior to storage. For pollen counting, 500 µL HCl was added per tube to digest all tissue except the exine, and the tubes were placed in an oven at 40 °C for 2 h. The tubes were then centrifuged at 5000 rpm (198 g) for 10 min. After removal of the supernatant, the pellet was resuspended in 500 µL of a 1 : 3 solution of lactic acid and glycerol. Three 10-μL subsamples were taken from each sample and placed on glass slides. The total number of pollen grains per subsample was counted under a dissecting microscope. To obtain an estimate of total pollen production per flower, the mean number of pollen grains within these three 10-μL subsamples per flower was multiplied by 250 (50 × 5 anthers per flower). Total ovule production per flower was determined by dissecting the ovarium, and counting the ovules under a binocular microscope after staining with methyl blue. The P/O ratio was calculated for each flower.

Capacity and timing of autonomous self-fertilization

To investigate the capacity for autonomous selfing, 20 individuals for each species were transferred into pots (n = 10 per population) and brought to a pollinator-free environment. Similar sized plants were selected to prevent confounding effects of resource availability on seed production (see Wesselingh, 2007). Three pollination treatments were assigned to two different flowers per plant at peak flowering: (1) emasculation prior to flowering in combination with supplemental self-pollination at the first day of flowering; (2) flowers that were left intact and unmanipulated; (3) emasculation of flowers prior to flower opening that were left unmanipulated. The final treatment (3) was included to test whether the species are able to produce any seeds through apomixis. Each treatment was applied on flowers within the same inflorescence and at the same developmental stage, and was replicated once within the same individual but on another inflorescence, so that in total six flowers were used per individual. Fruits were harvested once seeds were ripe and seed set was determined as the number of seeds divided by the total number of ovules of that flower. Following Lloyd and Schoen (1992), the index of autonomous selfing [i.e. autofertility (AF)] was determined as [(mean autonomous seed set per plant)/(mean seed set of supplemental pollinated flowers per plant)].

The timing of autonomous selfing was investigated on the same individuals, but on another subset of flowers. The following treatments were applied on two flowers per treatment and per plant: anther removal (1) prior to flower opening at 0 h, (2) at 12 h, (3) at 36 h, (4) at 60 h and (5) at 84 h, and (6) a control treatment in which anthers were left intact. Treatment 5 was not applied on C. pulchellum because flowers of this species only opened for a maximum of 3 d. Once fruits were mature, for each fruit, seed set was determined as the number of seeds/the total number of ovules.

Flower visitors and visitation rates

For each species, standardized surveys of flower visitors in each of the study populations were conducted at peak flowering times. During these surveys, all floral visitors were identified and quantified during 24 observation intervals of 20 min within 2 × 2 m² plots in each of the study populations. This was done on six sunny days between 1000 and 1400 h.

Pollinator failure, pollen limitation and reproductive assurance in the field

Simultaneously with the identification of floral visitors, three pollination treatments were applied to assess the degree of pollinator failure and reproductive assurance: (1) emasculation prior to flowering; (2) supplemental outcross-pollination on the first day of flowering; and (3) flowers left intact and unmanipulated. For each fruit, seed set was determined in the laboratory as the number of seeds/the total number of ovules. With these data the following indices per plant were calculated: (a) total pollen limitation as [1 − (seed set of intact flowers/seed set of supplemental pollinated flowers)] and (b) pollinator failure as [1 − (seed set of emasculated flowers/seed set of supplemental pollinated flowers)] (see Lloyd and Schoen, 1992; Kalisz and Vogler, 2003; Eckert et al., 2010).

Statistical analyses

The total number of flowers and inflorescences per plant, the number of flowers per inflorescence, petal width and length, flower and stigma size, herkogamy, the number of pollen (P) and ovules (O) per flower and the P/O ratio were compared between species using one-level ANOVA with species as a fixed factor and population as a random (categorical) variable nested within species to correct for random population effects (SAS Institute, 2005).

To investigate whether the capacity for autonomous selfing differed among the three study species, a one-level ANOVA was used with species included as the main effect and population as a random effect nested within species. Additionally, a two-level ANOVA was used to test whether the timing of autonomous seed set during floral development differed significantly among species. To do so, seed production proportional to the maximum number of seeds produced via autonomous selfing in that particular species was calculated, and this was entered as the dependent variable into the model. The main effects in this analysis are species, time of emasculation and their interaction, and the design further included two levels of nesting, both involving random effects. Because each treatment (i.e. timing of emasculation) was applied on each of the transplants used in this experiment, plants were the basic unit of replication that are nested within populations, whereas populations are nested within species.

To test for significant ‘intra-specific’ differences in seed set following the three pollination treatments applied in the field (open pollination, emasculation and supplemental pollination), a one-level ANOVA was used. For each model that was run per species, pollination treatment was included as a fixed factor and population as a random (categorical) variable. When significantly different, a post-hoc Tukey–Kramer adjustment was used to compare treatment means.

To test for significant ‘inter-specific’ differences in floral visitation rates, seed set following the different pollination treatments, indices quantifying different aspects of pollination success (i.e. pollinator failure (PF) and total pollen limitation (PL)), and the capacity to self autonomously (AF), a one-level ANOVA was again used. In these analyses, species was included as the fixed factor and population as a random factor nested within species. Again post-hoc Tukey–Kramer adjustment was used to compare species means.

In each of the above-mentioned analyses, the MIXED procedure in SAS 9·1 was used to analyse normal continuously distributed data (Littell et al., 2002). For the dependent variables with a binomial distribution (values of proportional seed set and pollination indices) the GLIMMIX procedure with logit link function was used (SAS Institute, 2005).

RESULTS

The three Centaurium species studied differed significantly in the total number of flowers that they produced per plant (F_2,177 = 18·39; P < 0·0001), the number of inflorescences per plant (F_2,177 = 60·21; P < 0·0001) and the number of flowers developed per inflorescence (F_2,177 = 270·62; P < 0·0001). Centaurium erythraea showed the largest number of flowers per plant and per inflorescence, whereas C. pulchellum had the fewest flowers per plant and per inflorescence (see Table 1). The Centaurium species studied also differed significantly in the size of their petals, both in width and length (P < 0·0001), and in total flower size, measured as the corolla diameter (F_2,169 = 309·61; P < 0·0001; see Table 1 and Fig. 1A), with C. erythraea producing the largest flowers, C. littorale being intermediate, and C. pulchellum being the smallest. Stigmas of C. erythraea were also significantly larger than those of C. littorale and C. pulchellum (F_2,169 = 60·27; P < 0·0001; see Table 1).

Table 1.

Flowering performance and floral characteristics of the related Centaurium erythraea, C. littorale and C. pulchellum obtained from two large (>1000 individuals) populations per species

	C. erythraea		C. littorale		C. pulchellum
Character	Pop. A	Pop. B	Pop. A	Pop. B	Pop. A	Pop. B
Total number of flowers*	71·2 ± 7·2	52·3 ± 5·2	50·9 ± 5·4	32·3 ± 4·22	27·4 ± 3·7	36·2 ± 2·9
No. inflorescences*	1·7 ± 0·2	1·6 ± 0·2	12·6 ± 1·2	6·7 ± 0·8	8·5 ± 0·8	13·1 ± 1·1
No. flowers/inflorescence*	45·9 ± 5·1	36·5 ± 3·1	4·3 ± 0·3	5·2 ± 0·4	3·0 ± 0·2	2·8 ± 0·1
Herkogamy (mm)*	0·96 ± 0·11	0·92 ± 0·08	0·27 ± 0·03	0·28 ± 0·04	0·04 ± 0·01	0·08 ± 0·01
Stigma height (mm)*	0·91 ± 0·01	0·85 ± 0·02	0·62 ± 0·01	0·60 ± 0·02	0·69 ± 0·01	0. 70 ± 0·01
Stigma width (mm)*	1·20 ± 0·02	1·05 ± 0·02	1·32 ± 0·02	1·03 ± 0·03	1·05 ± 0·03	1·04 ± 0·03
Petal width (mm)*	3·19 ± 0·06	3·14 ± 0·07	2·86 ± 0·07	3·18 ± 0·05	1·83 ± 0·05	1·88 ± 0·04
Petal length (mm)*	6·86 ± 0·08	6·52 ± 0·08	5·50 ± 0·13	5·99 ± 0·07	4·24 ± 0·10	4·46 ± 0·06
Corolla diameter (mm)*	14·36 ± 0·17	14·72 ± 0·19	13·18 ± 0·14	12·05 ± 0·26	9·54 ± 0·22	9·95 ± 0·14
Pollen grains per flower†	93575 ± 6429	110325 ± 7928	35175 ± 5995	39575 ± 5676	18750 ± 4236	21302 ± 2799
Ovules per flower†	289·3 ± 12·2	322·8 ± 13·3	235·7 ± 9·2	231·3 ± 8·2	254·4 ± 6·5	288·6 ± 12·5
P/O ratio†	328·9 ± 26·8	345·2 ± 25·9	156·3 ± 30·9	170·2 ± 23·8	74·8 ± 17·0	72·0 ± 7·8

*n = 30 per population, ^†n = 15 per population.

The degree of herkogamy varied significantly between the three species (F_2,169 = 286·3; P < 0·0001; Table 1). Centaurium erythraea showed the largest level of herkogamy, with anthers and stigma protruding well above the corolla tube, but bending in opposite directions (see Fig. 1A). Centaurium littorale showed an intermediate level of herkogamy, but with both reproductive organs oriented in a vertical fashion (Fig. 1A). Flowers of C. pulchellum lacked substantial spatial separation of its anthers and stigma, with anthers positioned closely around the stigmatic surface (Fig. 1A).

The three species also differed significantly in the number of ovules (F_1,57 = 20·6; P < 0·001) and pollen grains (F_1,57 = 109·7; P < 0·001) that are produced per flower. Centaurium erythraea produced the largest number of ovules and pollen grains per flower, whereas ovule and pollen grain production rates of C. littorale and P. pulchellum were much lower (see Table 1). The P/O ratio also varied significantly between the three species (F_1,27 = 69·2; P < 0·001), with C. erythraea showing the largest, C. littorale an intermediate and C. pulchellum the lowest ratio.

Flowers of each of the three study species that were emasculated just before flower opening did not produce any seeds in the pollinator-free environment, indicating that they were not able to develop any seeds through apomixis. The three Centaurium species showed significant variation (F_2,58 = 16·03; P = 0·025) in their capacity for autonomous selfing (mean seed set = 0·81 ± 0·03, 0·56 ± 0·06 and 0·49 ± 0·05, in C. pulchellum, C. littorale and C. erythraea, respectively), resulting in AF values of 0·92 ± 0·03, 0·68 ± 0·09 and 0·55 ± 0·06, in the aforementioned species. In each of the three study species, the amount of seed set resulting from autonomous selfing increased significantly the longer the duration of anthesis, but the rate of increase significantly differed among the species as indicated by the significant interaction between timing and species (see Table 2). Centaurium pulchellum realized half of its total autonomous seed set at the first day of flowering, whereas the rate of increase in autonomous selfing occurred much later in C. erythraea and was intermediate in C. littorale (Fig. 2).

Table 2.

Results of a two-way ANOVA (using the PROC GLIMMIX procedure in SAS) for the effect of species (C. erythraea, C. littorale and C. pulchellum) and timing of emasculation since flowers started anthesis (0, 12, 36, 60 h and control) in a pollinator-free environment, on autonomous seed production rates proportional to the maximum number of seeds that is produced via this mode of selfing

Source	d.f.	F	P
Proportional seed set
Species	2, 3	18·70	0·0202
Timing	4, 282	162·16	<0·0001
Timing × species*	7, 282	11·22	<0·0001

*Note that the numerator degrees of freedom of the interaction term are only seven instead of eight. This is due to the fact that the final treatment (emasculation after day 4) is missing in C. pulchellum, because flowers of this species only flower for 3 d.

Fig. 2.

Interspecific differences in seed set proportional to the total amount of seeds that is produced via autonomous selfing at different times of anther removal during the period of anthesis (0, 12, 36, 60, 84 h and control) between Centaurium erythraea, C. littorale and C. pulchellum (mean ± s.e.). Data from each of the two study populations are pooled per species (n = 10 per population).

The capacity for autonomous selfing was related to the degree of herkogamy and P/O ratios (Fig. 3A, B). C. erythraea showed the lowest index of autonomous selfing and the largest degree of herkogamy and P/O ratio, whereas C. pulchellum showed the highest index of autonomous selfing and the lowest degree of herkogamy and P/O ratio.

Fig. 3.

The association between (A) the index of autonomous selfing (AF) and herkogamy, and (B) pollen/ovule ratio and the index of autonomous selfing in Centaurium erythraea, C. littorale and C. pulchellum (mean ± s.e.). Data from each of the two study populations are pooled per species (n = 30 per population).

In total, 514 insect visitors were observed on flowers of the three Centaurium species, 347 on flowers of C. erythraea, 129 on C. littorale and 38 on C. pulchellum, with each of the three species receiving visits by largely the same insect species (see Appendix 1). The majority of visiting insects were pollen-gathering hoverflies (Diptera, Syrphidae), most often including Episyrphus balteatus, Sphaerophoria spp., and Scaeva spp. (see Appendix 1). Other flower visitors occurred at much lower frequencies and were small flies (Empididae–Muscidae; <15 %), some bees and bumblebees (Hymenoptera, Apidae; <6 %), and occasionally a butterfly (Lepidoptera; <4 %). Mean floral visitation rates within the 2 × 2 m² plots per 20-min observation time differed significantly between the three species studied (F_2,69 = 27·20; P < 0·0001), with C. erythraea receiving 14·5 ± 2·1 visitors, C. littorale receiving 5·4 ± 0·6 visitors and C. pulchellum receiving 1·6 ± 0·6 visitors.

Seed set of intact naturally pollinated flowers did not vary significantly between the three Centaurium species (F_2,117 = 1·34; P = 0·266; Fig. 4). In contrast, seed set of emasculated flowers exposed to open pollination in the field was significantly lower compared with that of unmanipulated flowers in each of the species studied (P < 0·0001; Fig. 4; see Appendix 2) and differed significantly between the three species (F_2,117 = 96·69; P < 0·0001). Seed set after supplemental cross-pollination was very similar between the three study species (F_2,117 = 1·20; P = 0·304) and was significantly higher than that of open pollinated flowers in C. erythraea and C. littorale (Fig. 4 and Appendix 2).

Fig. 4.

Interspecific differences in mean proportional seed set per fruit after emasculation, non-manipulation, and supplemental pollination between Centaurium erythraea, C. littorale and C. pulchellum in the field (mean ± s.e.). Data from each of the two study populations are pooled per species (n = 30 per population).

Pollen limitation did not differ significantly between the three Centaurium species (F_2,117 = 1·01; P = 0·367) and was low overall, ranging through 0·14 ± 0·03, 0·11 ± 0·03 and 0·09 ± 0·02 in C. erythraea, C. littorale and C. pulchellum, respectively. The index of pollinator failure, however, differed significantly between the three species studied (F_2,117 = 88·56; P < 0·001), and was 0·28 ± 0·02, 0·57 ± 0·04 and 0·91 ± 0·03 in C. erythraea, C. littorale and C. pulchellum, respectively (see Appendix 2). Pollinator failure was significantly related to floral visitation rates (Fig. 5A) and the proportion of seeds that was autonomously fertilized (Fig. 5B).

Fig. 5.

The association between pollinator failure (PF) and (A) number of floral visitors within 2 × 2 m² plots per 20-min periods of observation (n = 24 per species), and (B) the index of autonomous selfing (AF) (n = 30 per population), in Centaurium erythraea, C. littorale and C. pulchellum. Data from each of the two study populations are pooled per species (mean ± s.e.).

DISCUSSION

The spatial and/or temporal separation of the female and male functions within flowers of self-compatible plants is generally assumed to have a direct impact on the degree of self-pollination and the capacity to self autonomously (e.g. Barrett, 2002; Moeller, 2006). The present data from the pollinator-exclusion experiment indicated that each of the three Centaurium species has the capacity to self autonomously, but that the degree of autonomous selfing differed significantly between the study species. The association between the overall degree of herkogamy and the capacity for autonomous selfing suggests that anther–stigma separation plays an important role in the efficiency of this mode of selfing. Furthermore, there were also clear inter-specific differences in the timing of autonomous selfing that could be attributed to variable levels of dichogamy. In C. erythraea the occurrence of autonomous selfing fell well within the period of ‘delayed selfing’ (sensu Lloyd, 1992). This is realized by a system of protogyny in which the stigma becomes receptive during the first day of anthesis, when anther dehiscence has not yet begun. This is indicated by the high seed production obtained when flowers are experimentally pollinated the first day of flowering. By the second, but especially the third day onwards anthers start gradually curling and dehiscing, allowing self pollination (see Fig. 1B). Apart from the spatial segregation of anthers and stigmas in C. erythraea, this temporal segregation of the female and male phase appears to be sufficient to prevent self-pollination at the beginning of flowering, but provides reproductive assurance at the end of a flower's lifespan when pollinator-mediated pollen deposition has failed. Similar mechanisms of delayed selfing have been reported in Kalmia latifolia (Rathcke and Real, 1993) and Sanguinaria canadensis (Lyon, 1992).

In C. littorale and especially in C. pulchellum a large number of ovules are autonomously fertilized early in the lifespan of the flowers, suggesting that the timing of autonomous selfing in these species falls more within the category of ‘competing selfing’ (Lloyd, 1992). In these species, anthers start curling and release pollen much earlier during anthesis. This in combination with the close juxtaposition of anthers and stigma, especially in C. pulchellum (Fig. 1A), may explain the high indices of autonomous selfing in this species. Competing selfing can be expected to confer little or no pollen and seed discounting when outcross pollen deposition is very limited (Schoen and Brown, 1991). These observations thus suggest that the degree of herkogamy is mainly responsible for the efficiency of autonomous selfing in the three sister taxa, whereas the timing of anther curling, and thus degree of protogyny, may determine when autonomous selfing is realized during anthesis.

The present findings are, however, in contrast with those of Ubsdell (1979) who studied seed-set patterns following autonomous selfing and open pollination in natural populations of C. erythraea and C. littorale in the UK. He documented the highest autonomous selfing levels (which was calculated as: autonomous seed set/seed set of open pollinated flowers) in C. erythraea (on average 0·86) and a much lower capacity for autonomous selfing in C. littorale (on average 0·47). Although no exact measurements of the level of herkogamy were given in that study, he described C. erythraea flowers as having little to no herkogamy, whereas stigmas of C. littorale were mostly exposed above the anthers. These findings thus suggest that even within the same species the level of herkogamy may show substantial variation, and that such differences may in turn contribute to variable amounts of autonomous selfing to total seed set.

In this study, flowers of C. pulchellum were significantly smaller and had lower P/O ratios than those of C. littorale and especially those of C. erythraea. These data are consistent with the hypothesis that autogamously self-fertilizing plants may allocate a smaller proportion of resources to traits that contribute to male function relative to female function compared with closely related more outcrossing taxa (Lloyd, 1987). Compared with the classification given by Cruden (1977) and that of Michalski and Durka (2009), the P/O ratios of C. pulchellum fell within the group of obligate autogamous species, whereas those of C. erythraea show closer association with species that behave more like facultative out-crossers.

Observations of flower visitation and pollination success in the field, in combination with emasculation experiments, indicate that in C. pulchellum pollinator failure was 1·6 times higher compared with that of C. littorale and even 3·1 times higher compared with that of C. erythraea. These trends correlate with flower visitation rates and suggest that pollinator failure is more likely to result from differences in floral visitation rates than from differences in pollinator efficiency. In addition, previous investigation also revealed that floral lifespan, the capacity for seed production and floral visitation rates did not differ significantly between emasculated and intact control flowers in each of the three species studied (P < 0·05; R. Brys and H. Jacquemyn, unpubl. results), which leads to the conclusion that the observed patterns are not caused by treatment effects.

Despite high pollinator failure indices in C. pulchellum and C. littorale, flowers of both species showed little sign of pollen limitation in the field. This can be attributed to the fact that seed production under natural conditions is mainly the result of autonomous selfing, especially in C. pulchellum. It would seem reasonable to hypothesize that the latter annual species which is most often found in unstable, early successional habitats and has ephemeral populations with poor pollinator service (e.g. van Tooren et al., 1983), has developed a breeding system characterized by a high capacity for autonomous selfing. In contrast, the delayed selfing system of C. erythraea suggests that this species can be more characterized as having a mixed mating system, allowing opportunities for open pollination but ensuring seed production by autonomous selfing when pollinator-mediated pollination fails. Van Rossum (2009) indeed reported that when C. erythraea grows in suboptimal, peri-urban forests, these fragmented populations showed relatively low genetic diversity, high inbreeding coefficients (F_IS) and high genetic differentiation (F_ST), which are all indicators for a predominantly selfing behaviour. These observations thus illustrate that even within the same species, differences in habitat quality or temporal variation in weather conditions may cause variation in the pollinator environment, in the level pollinator-mediated seed set and ultimately in the amount of delayed autonomous selfing.

Conclusions

This study demonstrates that autonomous selfing in three closely related Centaurium species provides variable levels of reproductive assurance. The efficiency and timing of this mode of selfing is mediated by several inter-specific adaptations in floral morphology, such as the degree of herkogamy and protogyny. Patterns of pollinator services, pollination failure and pollen limitation under field conditions support the idea that the mode of autonomous selfing is strongly associated with the pollination environment and associated pollinator-mediated pollen deposition rates. These findings therefore also support growing empirical evidence that autonomous selfing can elevate seed production in natural plant populations, and that selfing may evolve differentially when plants experience variable pollinator services in closely related sister taxa. The present findings not only revealed that closely related species increased their capacity to self autonomously in response to pollinator failure, but also showed a reduced investment in floral traits that are involved with pollinator attraction, such as pollen production rates and corolla size. Once autonomous selfing begins to evolve, the reduced attractiveness of flowers to pollen-gathering insects, such as hoverflies, will most likely reinforce the further development of morphological adaptations to selfing.

APPENDIX 1

Frequencies of floral visitors observed on Centaurium erythraea, C. littorale and C. pulchellum, during 24 intervals of 20-min observation time within 2 × 2 m² plots per species.

	C. erythraea		C. littorale		C. pulchellum
	%	n	%	n	%	n
Hoverflies	91·6	319	80·6	104	81·6	31
Episyrphus balteatus		120		32		4
Sphaerophoria spp.		62		52		13
Scaeva spp.		58		9		2
Syrphus spp.		52		0		4
Eupeodes		9		9		6
Melanostoma		10		0		0
Helophilus trivittatus		8		0		2
Eristalis tenax		0		2		0
Small flies	4·6	16	13·4	13	10·5	4
Empididae		7		7		1
Muscidae		9		6		3
Bees	0·9	3	3·1	3	5·3	2
Lasioglossum		1		1		2
Megachile		0		2		0
Anthophora		1		0		0
Apis mellifera		1		0		0
Bumblebees	1·2	4	5·2	5	0·0	0
Bombus terrestrisch		2		2		0
B. ruderatus		1		1		0
B. praetorum		1		1		0
B. pascuorum		0		1		0
Butterflies	1·4	5	3·1	4	0·0	1
Zygaenidae		1		0		0
Arctiidae		2		2		0
Pieridae		2		2		1
Total		347		129		38

APPENDIX 2

Mean (± s.e.) proportional seed production after different pollination treatments and the obtained indices characterizing several aspects of the pollination success and breeding system in three Centaurium species in each of the study populations. EF = seed set in emasculated flowers, OF = open unmanipulated flowers, SF = supplemental cross-pollinated flowers, PF = index of pollinator failure and PL = index of total pollen limitation (n = 30 per population).

Species/population	EF (%)	OF (%)	SF (%)	PF	PL
C. erythraea
Pop. A	64·6 ± 4·4a	75·5 ± 2·8b	89·3 ± 1·2c	0·28 ± 0·05	0·16 ± 0·04
Pop. B	61·4 ± 4·2a	75·7 ± 3·4b	87·3 ± 1·3c	0·29 ± 0·05	0·12 ± 0·07
C. littorale
Pop. A	38·2 ± 4·2a	78·0 ± 3·2b	84·5 ± 2·6c	0·54 ± 0·05	0·10 ± 0·04
Pop. B	34·0 ± 4·5a	73·5 ± 3·1b	85·6 ± 1·8c	0·60 ± 0·05	0·13 ± 0·03
C. pulchellum
Pop. A	5·3 ± 3·1a	81·1 ± 2·4b	87·6 ± 2·7c	0·94 ± 0·04	0·05 ± 0·03
Pop. B	11·0 ± 3·3a	79·0 ± 3·1b	86·3 ± 2·1c	0·87 ± 0·04	0·08 ± 0·03

Different letters indicate significant differences between means for pollination treatments (ANOVA, P < 0·05).