Effects of human-mediated pollinator impoverishment on floral traits and mating patterns in a short-lived herb: an experimental approach


  • Rein Brys,

    Corresponding author
    1. Research Institute for Nature and Forest, Kliniekstraat 25, BE-1070 Brussels, Belgium
    2. Terrestrial Ecology Unit, Department of Biology, Ghent University, Ledeganckstraat 35, B-9000 Ghent, Belgium
    Search for more papers by this author
  • Hans Jacquemyn

    1. Division of Plant Ecology and Systematics, Biology Department, University of Leuven, Belgium
    Search for more papers by this author

Correspondence author. E-mail: rein.brys@inbo.be


1. Human-induced impoverishment of pollinator faunas may affect plant–pollinator interactions and limit pollen availability. Under these conditions, chronic outcross pollen limitation is expected to select for floral characters that maintain seed production, including autonomous selfing.

2. In this study, the impact of anthropogenic disturbances of the pollinator environment of the short-lived Centaurium erythraea on mating patterns was investigated. First floral traits and the capacity for autonomous selfing were compared between two contrasting pollinator environments. In addition, transplantation experiments were combined with hand-pollination and emasculation treatments to assess the extent of pollen limitation and the contribution of autonomous selfing to total seed production in these pollinator environments.

3. Under severe pollinator impoverishment, C. erythraea produced fewer and smaller flowers that showed no herkogamy and strongly reduced P/O ratios. The capacity for autonomous selfing was 36·1% higher in these pollinator-limited environments than in more natural, pollinator-rich environments, where plants developed more, larger and markedly herkogamous flowers.

4. When assigned to the pollinator-rich environments, plants from pollinator-limited populations showed significantly higher outcross pollen limitation compared with the original plants. In contrast, plants from pollinator-rich environments assigned to pollinator-poor populations did not experience higher pollinator-mediated seed production and showed lower total seed production than plants originally occurring in these pollinator-limited environments.

5. These results demonstrate that human-induced pollen limitation selects for selfing as a means of reproductive assurance, whereas in the pollinator-rich environments, traits that support outcrossing are favoured.


It is widely accepted that human-induced disturbances can have a major impact on the abundance and diversity of pollinators, which in turn may result in altered plant–pollinator interactions and increased pollen limitation, either through reduced pollen deposition or because of lower pollen quality (e.g. Ashman et al. 2004; Knight et al. 2005; Aguilar et al. 2006; Eckert et al. 2010). Moreover, because pollinator faunas have been shown to be declining world-wide (Ricketts et al. 2008; Winfree et al. 2009; Potts et al. 2010), such effects are expected to increase in the near future.

Whereas the ecological impact of anthropogenic disturbances on plant reproduction has been well studied (e.g. Ashman et al. 2004; Ghazoul 2005; Aguilar et al. 2006), surprisingly little attention has been given to the potential evolutionary consequences of such reductions in pollinator availability (reviewed by Eckert et al. 2010 and Harder & Aizen 2010). Although it can be expected that evolutionary effects will generally take a longer time to arise than ecological effects, they may be more insidious and thus fundamentally alter the phenotypic and genetic makeup of plant populations (Charlesworth & Wright 2001; Cheptou 2004; Eckert et al. 2010). Theory suggests that variable levels of pollinator availability can generate diverse selective forces on plant mating systems that are mainly driven by the resulting amount and variation in pollen limitation (e.g. Lloyd 1992; Morgan & Wilson 2005; Eckert et al. 2010). In case of chronic pollen limitation, it is expected that floral traits are selected that promote autonomous self-fertilization and thus provide reproductive assurance, such as reductions in the level of herkogamy or dichogamy (Cruden & Lyon 1989; Lloyd 1992; Morgan & Wilson 2005; Porcher & Lande 2005). At the same time, when autonomous selfing evolves, selection for reduced investment in pollinator attraction (such as reductions in flower size, corolla diameter, number of flowers, floral longevity, etc.) and rewards (such as reduced nectar or pollen production) can also be expected to occur (Ornduff 1969; Eckert et al. 2010).

On the other hand, when competition for pollinator services and pollen limitation tends to vary among plants, selection for traits that increase attractiveness (such as increasing floral size and flower production, higher nectar production, etc.) and ultimately outcross pollination is likely to occur (Ashman & Morgan 2004; Harder & Johnson 2009). Because such floral adaptations are expected to act in the opposite direction of traits that are associated with the promotion of autonomous selfing, it is difficult to predict which evolutionary trajectory a population or species will take when pollination environments are critically disturbed (e.g. Eckert et al. 2010). In addition, when pollinator abundance and pollinator limitation vary in space and time (Goodwillie, Kalisz & Eckert 2005; Eckert et al. 2010), plants may evolve to a mixed mating system. Under such a scenario, establishment of a delayed selfing mechanism will be most favourable, as it increases female fitness with little seed or pollen discounting (i.e. production of selfed progeny that pre-empts the production of outcrossed progeny; Kalisz, Vogler & Hanley 2004).

Although it is likely that significant evolutionary shifts in plant breeding systems are currently happening in a large number of plant species inhabiting human-disturbed habitats, it remains unclear whether such adaptive responses are detectable yet (Harder & Aizen 2010). This is especially true for long-lived iteroparous species, in which several generations may show an overlap in flowering and it may take many generations before significant floral adaptations become visible. In short-lived monocarpic species, on the other hand, floral evolutionary shifts can be expected to occur more rapidly, because of their faster generation time, the fact that each year another generation is flowering, and the more stringent need for reproductive assurance in such species (Barrett, Harder & Worley 1996; Schoen et al. 1997). Nonetheless, there are still very few studies focusing on the effects of human disturbances on breeding system evolution, even in short-lived species (Eckert et al. 2010; Harder & Aizen 2010).

In this study, we focused on the short-lived, monocarpic herb, Centaurium erythraea. Although flowers of C. erythraea are generally characterized by herkogamy (anther–stigma separation), significant differences in this floral trait have been observed between populations (Fig. 1), ranging from clear anther–stigma separation in populations located at the Belgian coast (see the study by Brys & Jacquemyn 2011) to populations that show hardly no herkogamy in the UK (Ubsdell 1979). It can thus be hypothesized that the level of herkogamy may be involved in the capacity and efficiency of autonomous selfing in this species (see the study by Brys & Jacquemyn 2011) and that chronic pollen limitation may cause selection against this floral trait. The major aim of this study was to investigate how human-mediated impoverishment of pollinator communities and associated pollen limitation affected several floral traits and pollination patterns. To do so, four large natural C. erythraea populations were studied that occurred in two highly contrasting pollinator environments: two populations growing in a coastal dune area that is naturally rich in pollinators (Brys et al. in press) and two populations located in an industrial area where pollinator faunas were only a subset of those encountered in populations occurring in more natural and undisturbed conditions (see Appendix S1a). More specifically, we first measured floral traits and determined the capacity to self-autonomously in each of the selected populations. Secondly, reciprocal translocation experiments combined with supplemental hand-pollinations and emasculations were conducted to test the hypothesis that pronounced differences in pollinator availability resulted in significant differences in pollen limitation and pollination patterns, and whether these effects were mediated by differences in floral traits.

Figure 1.

 Flowers of Centaurium erythraea originating from natural populations growing in two contrasting pollinator environments, with (a) larger and more herkogamous flowers in the pollinator-rich environment, and (b) smaller flowers lacking herkogamy in the pollinator-poor environment.

Materials and methods

Study Species

Centaurium erythraea (Rafn.) is a biennial monocarpic herb, belonging to the Gentianaceae. It is often found in species-rich, calcareous grasslands, but it also occurs in more transient habitats and anthropogenic-disturbed areas such as forest clearings, artificial sand suppletion areas, etc. C. erythraea produces showy pink flowers that are hermaphroditic and self-compatible (Ubsdell 1979). Flowering starts at the end of June and lasts until mid-August. Anthesis of individual flowers mostly takes about four to sporadically 5 days (Ubsdell 1979; Brys & Jacquemyn 2011). Centaurium erythraea does not produce any nectar and is primarily visited and pollinated by pollen gathering hoverflies (Diptera, Syrphidae), but occasionally some bees (Hymenoptera, Apidae), small flies (Empididae-Muscidae), moths and butterflies (Lepidoptera) have been documented as floral visitors (Müller 1883; Knuth 1909; Ubsdell 1979; Van Rossum 2009; Brys & Jacquemyn 2011). When successful pollination fails, flowers are able to produce a certain amount of seed production through a system of delayed autonomous selfing in which anthers curl at the end of anthesis (Brys & Jacquemyn 2011). In August, plants produce tiny seeds (<0·01 mg) in large quantities (on average 251 ± 90 seeds per fruit; Brys et al. in press).

Study Sites

The study was conducted in four large (N > 5000) C. erythraea populations in Belgium that strongly differed in pollinator assemblages. Two populations were located in a coastal dune area at the Western part of the Belgian coast between Oostduinkerke and De Panne [Ter Yde (TY) and Doornpanne (DP)]. The other two populations were located within an industrial area (Waaslandhaven) close to the river Scheldt near Antwerp [Waaslandhaven 1 (WH1) and Waaslandhaven 2 (WH2)]. Initial assessment of the pollinator faunas in each of these populations revealed that C. erythraea is almost exclusively pollinated by hoverflies (with Episyrphus balteatus, Sphaerophoria spp. and Scaeva spp. being the most abundant (>96%) insect visitors. Pollinator availability was about three times lower in the disturbed populations (WH1 and WH2) compared with the more natural populations in dune habitats (for more details, see Appendix S1a, Supporting Information). In analogy, the number of floral visits per unit area also differs significantly between the types of populations, whereas on average 5·8 ± 1·0 floral visits were recorded per 20-min observation time in populations in the industrial zone, 15·9 ± 2·3 visitors were recorded in dune populations (see Appendix S1b, Supporting Information). Due to the pronounced differences in pollinator assemblages, in the following populations of dune, areas are denoted as pollinator-rich populations, where populations located in the industrial area are referred to as pollinator-poor populations.

Floral Measurements and Determination of the Capacity for Autonomous Self-Fertilization

In each study population, we randomly selected 40 plants during peak flowering (middle of July) and for each plant, we determined the total number of flowers. One recently opened flower was harvested per plant to determine flower size (i.e. corolla diameter) and the degree of herkogamy (measured as the nearest distance from the top of the stigma to the lowest part of the anther). Photographs were taken of the cross-section of each flower under a dissecting microscope, and both floral traits were measured using Image J (Rasband 2011). Because the degree of herkogamy does not change during floral development and is very constant within the same plant (R. Brys and H. Jacquemyn, unpublished results), harvesting a single flower is sufficient to characterize these traits at plant level. Another flower was harvested of a subset of 20 plants to determine mean pollen and ovule production per flower. Each of these flowers was sampled just before anthers dehisced to ensure that total pollen production could be accurately measured. Per flower one anther was excised in the laboratory and used for the determination of the total amount of pollen (for more details see the study by Brys & Jacquemyn 2011). Total ovule production per flower was determined by dissecting the ovary and counting the ovules under a binocular microscope after staining with methyl blue. Afterwards, the pollen/ovule (P/O) ratio was calculated for each flower.

To investigate the capacity for autonomous selfing, 20 individuals per population were transferred into pots and brought to a pollinator-free environment. Opened flowers were left intact and unmanipulated during the entire flowering period, allowing assessing spontaneous pollination through selfing. Once seeds were ripe, we harvested five fruits per plant and determined mean fruit-level seed production per plant by averaging the total amount of seeds of each of these fruits. Three plants were lost during the experiment.

Pollination Patterns and Seed Production in Contrasting Pollination Environments in the Field

To test for outcross and total pollen limitation and to tease apart the effects of pollinator environment and floral traits related to pollinator availability on seed production, a reciprocal translocation experiment was set up, comparing seed production between original and introduced plants in both pollinator environments. Two pairs of populations were used (WH1–TY and WH2–DP, respectively). In early June (2010), 3 weeks before flowering started, 60 reproductive individuals in each of the four aforementioned populations were excavated and planted into pots (8 × 8 × 15 cm), resulting in a total of 240 transplants. In each population, we selected similar sized plants representing the average size and floral display of that particular population. Per population, 30 of these transplants were put in the original population again, whereas the other subsample was introduced in the other population located in the opposite pollination environment.

Within each population, transplants were placed in 10 subgroups of three transplants that were spatially arranged in a triangle. Within these subgroups, transplants originated from the same population and were positioned at 50 cm from each other to assure an equal density of coflowering congeners and similarity in other local environmental conditions. This design allowed calculating the index for total and outcross pollen limitation for each subgroup later on (see following paragraph). When flowering started, each of these three transplants per subgroup was allocated randomly to one of the following treatments: (i) natural open pollination of intact flowers (i.e. control treatment), (ii) supplemental outcross-pollination on the first day of flowering and (iii) natural open pollination of flowers that were emasculated prior to flowering. For each transplant, all flowers received the same treatment. Previous investigation showed that floral life span, the capacity for seed production and floral visitation rates did not differ significantly between emasculated and intact open-pollinated flowers (> 0·05; R. Brys and H. Jacquemyn, unpublished results). To minimize variation in pollen delivery in the supplemental pollination treatment, all flowers were supplementary pollinated by the same person using forceps to transfer pollen. Pollen for supplemental pollinations was obtained from at least five donor plants in each population. At the end of July, when all flowers of the treated transplants withered and stigmas were no longer receptive, all transplants were brought to the greenhouse to allow maturing of fruits under similar environmental conditions and to prevent any kind of herbivory or seed loss. Once seeds were ripe, five fruits per plant were harvested and for each fruit, the number of seeds was counted. Average fruit-level seed production per plant was determined by calculating the average number of seeds.

For each treatment (pollinator environment and population of origin), indices of total and outcross pollen limitation were calculated. The index of outcross pollen limitation was calculated as [1−(mean seed production of emasculated plants/mean seed production of supplemental pollinated plants)], whereas the index of total pollen limitation was calculated as [1−(mean seed production of intact plants/mean seed production of supplemental pollinated plants)] (see the studies by Lloyd & Schoen 1992; Kalisz & Vogler 2003; Eckert et al. 2010).

Data Analysis

The impact of pollination environment on floral display, corolla size and herkogamy, the number of pollen (P) and ovules (O) per flower, P/O ratio and the capacity for autonomous selfing was tested using a generalized linear mixed model (GLMM). To incorporate the nested structure of the design, population nested within origin was included as a random factor in the model to correct for random population effects.

A linear mixed model was used to investigate whether average seed production obtained from the reciprocal translocation experiment in the field was significantly affected by pollination treatment, pollinator environment and population of origin, and all the 3- and 2-way interactions between these predictor variables. Because the experiment was replicated, a blocking factor was incorporated in the model as a random factor to account for the dependence of the observations within blocks. When significantly different, a post hoc Tukey’s HSD adjustment was used to compare the means of pollination environment and population of origin.

Finally, a GLMM was used to test whether both total and outcross pollen limitation differed significantly among pollination environment, the origin of the transplants and their interaction. A blocking factor was again included as a random factor to incorporate the replicated structure of the translocation experiments. When significantly different, a post hoc Tukey’s HSD adjustment was run using least square means derived from the model to compare the means of each pollination environment vs. origin combination. Each of the aforementioned analyses were conducted with sas, version 9.2 (SAS Institute Inc.©), using the Mixed procedure (Littell, Stroup & Freund 2002).


Floral Variation and Capacity for Autonomous Selfing

Plants from the pollinator-rich environments produced significantly more flowers per plant (F1,158 = 45·68; < 0·0001; Fig. 2a) and larger flowers (F1,158 = 405·48; = 0·003; Fig. 2b) than plants in pollinator-poor environments. Flowers differed significantly in the level of herkogamy (F1,158 = 40·64; < 0·0001), with anthers clearly exposed above the top of the stigma surface in the pollinator-rich populations, whereas anthers were positioned around the stigma in populations located in pollinator-poor environments (Figs 1 and 2c). Populations from both pollination environments did not show significant (F1,78 = 0·38; = 0·541) differences in the total number of ovules produced per flower (mean number of ovules per flower (±SE) = 322·0 ± 25·2 and 317·5 ± 39·3 in pollinator-rich and pollinator-poor populations, respectively). The total number of pollen grains per flower was, however, significantly (F1,78 = 103·90; = 0·009) smaller in pollinator-poor than in pollinator-rich populations (on average 44250 ± 2645 and 103806 ± 3486 pollen per flower in the pollinator-poor and pollinator-rich populations, respectively). As a result, P/O ratios were significantly larger (F1,78 = 122·04; = 0·008) in pollinator-rich than in pollinator-poor populations (Fig. 2d).

Figure 2.

 Differences in (a) total flower production per plant, (b) flower size (i.e. corolla diameter), (c) degree of herkogamy and (d) pollen/ovule ratio (mean ± SE) in each of the studied Centaurium erythraea populations (N = 40 per population) occurring under contrasting pollinator environments.

The capacity for autonomous self-fertilization differed significantly between plants from the two pollinator environments (F1,75 = 38·35; < 0·0001). On average 248·3 ± 47·8 (ranging from 28 up to 314) seeds per fruit were produced via autonomous self-fertilization in individuals occurring in pollinator-poor populations, whereas on average 158·5 ± 75·8 (ranging from 113 up to 345) seeds per fruit were produced in individuals from pollinator-rich populations (mean difference of 36·1%).

Pollination Patterns and Pollen Limitation in the Field

Average seed production per fruit was significantly affected by pollinator environment, site of origin and pollination treatment. Moreover, interactions between these factors were also significant (Table 1). Supplemental hand-pollination significantly increased seed production compared with open pollination of control (overall increase of 14·2%) and emasculated (overall increase of 72·9%) plants (Fig. 3a). Seed production following supplemental hand-pollination, however, did not significantly differ between plants originating from pollinator-poor and pollinator-rich environments, nor was it significantly affected by the environment to which transplants were assigned (Fig. 3a). Emasculation significantly reduced seed production compared with open-pollinated control plants, both in pollinator-poor and in pollinator-rich environments (mean reduction = 68·4%, ranging from 51·1% to 79·9%). However, mean seed production of emasculated plants was significantly larger when they were assigned to pollinator-rich environments (on average 108 seeds per fruit), particularly if they originated from pollinator-rich populations, and seed production markedly reduced when plants were assigned to the pollinator-poor populations (mean reduction = 51·9%; Fig. 3a). On the other hand, when emasculated plants were transplanted into pollinator-poor environments, no significant difference in mean seed production was observed between plants originating from pollinator-poor and pollinator-rich environments (Fig. 3a). Open pollination of intact flowers in individuals originating from the pollinator-rich populations resulted in significantly lower seed production per fruit when they were transplanted into pollinator-poor populations (mean reduction of 32·5% in comparison with plants from the other populations (Fig. 3a)).

Table 1.   Results of the linear mixed model (using the Proc Mixed procedure in sas) for the effect of pollinator environment, origin of the transplants and pollination treatment (supplemental cross-pollination, open pollination after emasculation and open pollination of intact flowers) on mean fruit-level seed production per plant (N = 240) in four natural Centaurium erythraea populations (two in a pollinator-rich environment and two in a pollinator-poor environment)
Fixed effectsd.f. F P
Pollinator environment1, 71·625·69<0·0001
Origin1, 71·67·910·0063
Poll. env. × Origin1, 71·64·520·0370
Pollination treatment2, 145500·02<0·0001
Poll. env. × Poll. treat.2, 14510·18<0·0001
Origin × Poll. treat.2, 14517·589<0·0001
Poll. env. × Origin × Poll. treat.2, 1452·250·109
Figure 3.

 Differences in (a) seed production at fruit-level (mean ± SE) following different pollination treatments (‘control’ = natural open pollination of intact flowers, ‘emasculated’ = natural open pollination of flowers that are emasculated prior to flowering and ‘suppl. poll.’ = supplemental outcross-pollination on the first day of flowering) and (b) total and outcross pollen limitation (mean ± SE) in flowering Centaurium erythraea individuals assigned to each of the four pollinator environment vs. origin combinations (N = 20 per treatment; ‘origin’→‘pollinator environment’ to which transplants were assigned). Different letters indicate significant differences (< 0·05) by Tukey’s HSD contrast of LSMs.

Outcross pollen limitation was significantly larger in the pollinator-poor environment than in the pollinator-rich environment (mean PLoutcross = 0·83 and 0·62, respectively; Table 2, Fig. 3b), and this effect was the most pronounced when individuals from the pollinator-rich environment were transplanted into the pollinator-poor populations (Fig. 3b). Total pollen limitation was significantly larger in the pollinator-poor environment and in transplants originating from the pollinator-rich environment (Table 2, Fig. 3b). In particular, individuals originating from the pollinator-rich environment showed significantly higher total pollen limitation when they were transplanted into pollinator-poor environments (Table 2, Fig. 3b).

Table 2.   Results of the generalized linear mixed model (using the Proc mixed procedure in sas) for the effect of pollinator environment and origin of the transplants on the level of total and outcross pollen limitation in four natural Centaurium erythraea populations (two in a pollinator-rich environment and two in a pollinator-poor environment)
Pollination treatmentd.f. F P
Outcross pollen limitation
 Pollinator environment1, 7342·43<0·0001
 Origin1, 733·310·0729
 Poll. env. × Origin1, 737·150·0093
Total pollen limitation
 Pollinator environment1, 7414·420·0003
 Origin1, 7424·34<0·0001
 Poll. env. × Origin1, 743·810·0448


When pollinator faunas become impoverished and result in chronic limitation of pollinator-mediated pollen deposition, one might expect that plants select for mechanisms that allow autonomous selfing and provide reproductive assurance (Kalisz et al. 1999; Kalisz, Vogler & Hanley 2004; Knight et al. 2005; Morgan & Wilson 2005; Fishman & Willis 2008). In case of the studied C. erythraea populations, our data showed that when pollinator availability was severely limited, populations showed significantly higher levels of outcross pollen limitation compared with sites where pollinator availability was high. Moreover, plants in pollinator-limited environments showed a complete lack of herkogamy (see Fig. 1), and a significantly higher capacity to self-autonomously under pollinator-free conditions (mean increase of fruit-level seed production of 36·2%), thus suggesting selection for autonomous selfing in human-disturbed, pauperized pollinator environments. In the herkogamous Gentianella germanica, Luijten et al. (1999) documented that the mean level of herkogamy became reduced when populations experienced a significant reduction in size, which they attributed to associated reductions in pollinator-mediated pollen deposition. According to Luijten et al. (1999), the observed reduction in herkogamy could lead to increased levels of autonomous selfing in this species. Similar findings have been reported by Eckert & Herlihy (2004) and Herlihy & Eckert (2007), who observed a significant increase in the autogamous component of total selfing with decreasing herkogamy in Aquilegia canadensis. Similarly, Takebayashi & Delph (2000) and Moeller (2006) also showed that variation in the level of herkogamy in the annual plant species Gilia achilleifolia and Clarkia xantiana was correlated with the capacity to self-autonomously.

Although our results indicate that the level of herkogamy was involved in the regulation of autonomous selfing, to be an important target for selection, it must, however, show genetically based phenotypic variation (Van Kleunen & Ritland 2004; Herlihy & Eckert 2007). Estimates of broad-sense heritability of herkogamy based on the regression between C. erythraea parents measured in the field and progeny measured in the greenhouse indeed suggest a heritable genetic component to the observed phenotypic variation in this floral trait (heritability index H2 = 0·623, determined on progeny derived from selfing, R. Brys and H. Jacquemyn, unpublished data). Similar levels of heritable variation in the level of herkogamy have been documented in Mimulus guttatus (mean H2 = 0·577; Carr & Fenster 1994), Gentianella campestris (mean H= 0·855; Lennartsson et al. 2000), Datura stramonium (mean H2 = 0·3; Motten & Stone 2000) and Aquilegia canadensis (mean H2 = 0·422; Herlihy & Eckert 2007).

Centaurium erythraea individuals also produced significantly fewer and smaller flowers, and less pollen per flower in pollinator-poor environments. Because ovule production did not differ among both environments, the lower P/O ratios in the pollinator-limited populations were mainly the result of lower pollen production rates. These observations support the hypothesis that in comparison with their more outcrossing congeners, plants that are characterized by a higher capacity of autonomous selfing allocate a smaller amount of resources to traits that contribute to the male function relative to the female function (Lloyd 1987; Goodwillie et al. 2010). Based on the classification given by Cruden (1977) and by Michalski & Durka (2009), P/O ratios of C. erythraea populations growing in the pollinator-poor environments showed a stronger association with species that behave like facultative autogamous species, whereas those growing in the pollinator-rich environments fell more within the group of facultative outcrossers. In addition, our findings also agree with a study by Harder & Aizen (2010), who showed that species growing in pollinator-poor habitats have smaller flowers and have evolved towards an autonomous selfing breeding system about three to four times more often than related species growing in pollinator-rich environments.

More and larger flowers per plant, with larger amounts of pollen per flower, may also affect pollinator behaviour and offer a selective advantage if these traits positively affect pollinator attraction and ultimately successful outcross pollination (Holtsford & Ellstrand 1992; Ashman & Morgan 2004). A recent review of pollinator-driven phenotypic selection on floral traits indeed suggests that selection, through female fecundity, for both larger and more flowers can occur under pollinator-rich conditions (Harder & Johnson 2009; Harder & Aizen 2010). If this would be the case in the studied C. erythraea populations, one should expect that in pollinator-rich environments, individuals with more and larger flowers are more attractive to pollinators and experience higher amounts of pollinator-mediated pollen deposition, resulting in higher seed production than more autonomous-selfing plants originating from the pollinator-poor populations. The seed set data of emasculated transplants indeed suggest that when small-flowered plants from the pollinator-poor environment were introduced into large-flowered pollinator-rich populations, these individuals produced significantly fewer seeds per fruit compared with the emasculated local plants. The observation that supplemental hand-pollination on plants from both origins resulted in similar amounts of seed production suggests pollinator-mediated seed production patterns in the emasculated plants thus mirror variable levels of pollinator attraction. Although it is generally assumed that increased attractiveness returns more benefit in terms of pollinator-mediated seed production when pollinators are rare (Harder & Aizen 2010), our observations indicate the opposite in the pollinator-poor populations. In the latter, large-flowered plants originating from the pollinator-rich populations did not show significantly higher seed production per fruit nor did they show marked differences in the level of outcross pollen limitation. A possible explanation might be that pollinators in these disturbed environments behave more like accidental passengers without an efficient foraging strategy in comparison with more resident pollinator faunas in the pollinator-rich environments. In addition, depletion and thus competition for floral rewards (i.e. pollen in this case) might also be much smaller under severe pollinator limitation, which might have lowered the advantage of exposing large and multiple flowers under these pollinator-poor conditions. These findings thus suggest that under pollinator-rich conditions, there seems to be a selective advantage via female fitness for plants with many and larger flowers, whereas such an advantage was absent under severe pollinator limitation. In the latter conditions, investment for multiple and showy flowers may thus only yield extra costs without a substantial profit in terms of pollinator-mediated seed production.

Because plants in the pollinator-poor sites produced on average less flowers than plants in pollinator-rich sites, this may lead to smaller amounts of seeds that are autonomously produced at plant level. This observation seems to be in contradiction with the above-mentioned selection towards a higher capacity of autonomous seed set at fruit-level under chronic pollinator limitation but can be explained by the fact that among-population variation in flower number can be affected by other environmental differences (e.g. Pélabon, Armbruster & Hansen 2011). It has, for instance, been documented that plant populations, species or races from extreme or more ephemeral habitats often grow to a smaller final size and consequently produce fewer and smaller flowers, most likely resulting from limited soil moister content or nutrient availability (Frazee & Marquis 1994; Carroll, Palladry & Galen 2001; Herrera 2005; Caruso 2006). In case of the studied C. erythraea populations, plants in the pollinator-poor sites were significantly (> 0·0001) smaller than in the pollinator-rich environments (average height: 19·6 ± 3·6 and 35·1 ± 4·9 cm, respectively; R. Brys, unpublished results). Although no abiotic data are available, it might be plausible that such environmental differences may have strengthened selection towards smaller and fewer flowers per plant, despite the associated reduction in plant-level reproductive assurance.

Since the quality of pollen arriving on stigmas may also change drastically with changes in pollinator availability and associated mating patterns, the level of inbreeding depression may have an important impact on breeding system differentiation as well. Future research should therefore focus more on selection via total lifetime fitness than at the level of seed set alone (e.g. Aizen & Harder 2007). It can be expected that high levels of total cumulative inbreeding depression (i.e. δ > 0·5) may discourage selection towards a more obligate selfing mechanism (Goodwillie, Kalisz & Eckert 2005; Harder, Richards & Routley 2008) and sustain significant anther–stigma separation in populations where pollinators are sufficiently available (Motten & Stone 2000; Van Kleunen & Ritland 2004). Accordingly, a better understanding of the variable causes of anthropogenic pollen limitation and their evolutionary consequences is required to predict, manage and mitigate such human impacts on the evolution of plant breeding systems (Mitchell & Ashman 2008).


We are grateful to Ivo Brys for practical help during the fieldwork. We also thank F. Hendrickx for statistical help and two anonymous referees for their constructive comments on an earlier draft of this manuscript. RB was supported by a postdoctoral grant of the Flemish Fund for Scientific Research (FWO) and HJ by the European Research Council (ERC starting grant 260601 – MYCASOR).