Introduction

Attractiveness to pollinators plays a key role in the reproductive ecology of entomophilous plants. Larger flowers increase visits rates, favoring pollen dispersal and deposition and, consequently, increasing both male and female fitness (Bell 1985; Conner and Rush 1996; Aigner 2005; Nattero et al. 2011). Hence, pollinators are considered as one of the main factors causing evolution on flower size (Fenster et al. 2004; Willmer 2011). Phenotypic selection studies in natural populations have also confirmed significant selection towards larger flowers (reviewed in Harder and Johnson 2009). However, small flowers and spatial–temporal variation in flower size still persist in populations so, from an evolutionary perspective, a unilateral view of the role of pollinators is probably oversimplistic (Galen 1999). For example, larger flowers are associated with more visits of floral enemies (Shykoff et al. 1996; Galen 1999; McCall and Irwin 2006) and greater requirements of biomass, carbon and water for production and maintenance of floral structures (Galen et al. 1999; Halpern et al. 2010; Teixido and Valladares 2013). Consequently, flower size is not only influenced by pollen limitation but also by resource limitation due to the combined effect of biotic and abiotic environmental factors. Since this trait is genetically controlled and is heritable, its variability generates potential to evolve in natural plant populations (Weiss et al. 2005; Ashman and Majestic 2006).

Bateman (1948) proposed that the intensity of selection is sex-dependent being higher through male fitness, whereas female fitness is more resource-limited. Therefore, male function could benefit more from pollinator attraction and thus flower size would mainly evolve through selection on male reproductive success (“male function” hypothesis, see also Wade 1979; Burd and Callahan 2000; Jones 2008). This implies that female function is less dependent on pollinator attraction, since few visits could be enough to fertilize all the ovules, and more dependent on resource availability for fruit and seed production, whereas more visits would be required to disperse high amounts of pollen. In fact, this assumption is supported in dioecious zoophilous plants where male individuals have larger flowers than conspecific females, but it is a more complex process in hermaphroditic species since there could be differential and conflicting sexual selection where its strength and direction are usually context-dependent (reviewed in Ashman and Morgan 2004).

Reproductive success analyses and estimates of selection in hermaphroditic plants require simultaneous studies through both sexes together with studies of spatial variation in pollinator assemblage. Most studies on reproductive success and phenotypic selection on flower size in hermaphroditic plant species have exclusively focused on female function, mainly due to the complexity of measuring male function in natural populations (Herrera et al. 2006). Molecular analyses with genetic markers, including DNA extraction and genotyping are needed to record direct estimates of male reproductive success (reviewed in Conner 2006). Though indirect methods such as the amount of dispersed pollen do not necessarily imply paternity success, i.e., that a particular dispersed pollen grain from a certain plant’s flower confers paternity to one seed, pollen dispersal is a representative component of male success and is a useful measure to disentangle the mechanisms, the strength and the direction of phenotypic selection through this sex (Snow and Lewis 1993; Maad and Alexandersson 2004; Arista and Ortiz 2007). Many works have analysed the spatial variation in phenotypic selection on flower size and/or its variation through both sexes (e.g. Caruso et al. 2003; Maad and Alexandersson 2004; van Kleunen and Ritland 2004; Arista and Ortiz 2007; Hodgins and Barrett 2008; Nattero et al. 2010a), but less have considered the effects of spatial variation in pollinator environment and patterns of visit rates on the strength and direction of phenotypic selection (Aigner 2005; Sletvold and Ågren 2010; Sahli and Conner 2011; Sletvold et al. 2012). However, as far as we know, there is a lack of studies combining pollinator assemblage in different populations with selection analyses on flower size through male and female success, but they are essential to identify reliable estimates of phenotypic selection (Conner 2006; see also Herrera et al. 2006).

The extent to which current selection on flower size is mediated by pollinators has rarely been determined experimentally. Overall, reliable estimates of pollinator-mediated phenotypic selection on flower size require assessment of a relationship between trait and relative fitness and, ultimately, if this relationship is at least partly the result of interaction with pollinators. For female fitness, pollinator-mediated phenotypic selection on flower size can be detected by comparing the strength and direction of selection between open-pollinated and hand-pollinated flowers receiving supplemental pollination (reviewed in Ashman and Morgan 2004). Following this methodology, a growing body of studies has documented pollinator-mediated positive and directional phenotypic selection on flower size through female function (Galen 1996; Totland 2001; Fishman and Willis 2008; Parachnowitsch and Kessler 2010; Sletvold and Ågren 2010; Bartkowska and Johnston 2012). Other studies have compared selection with and without the selective agent by experimentally manipulating the presence of the most abundant or effective pollinators (e.g. Galen 1989; Sahli and Conner 2011). For male fitness, a relationship between pollinator visit rates and flower size variation can be used to test whether larger flowers are differentially visited by more pollinators. Then, analyzing pollen removal rates in relation to flower size variation and differential visitation rates can be used to estimate pollinator-mediated phenotypic selection on flower size through male function, at least in an indirect way (Snow and Lewis 1993; Herrera et al. 2006).

In this paper, we estimated phenotypic selection on flower size and analysed whether this process differs through both sexes by means of an experimental approach in two populations of Cistus ladanifer L. (Cistaceae), a pollinator generalist, self-incompatible, hermaphroditic and large-flowered Mediterranean shrub (Herrera 1992; Talavera et al. 1993; Guzmán et al. 2013). Though we did not compare selection between open- and hand-pollinated flowers, we know that C. ladanifer in our study populations may suffer some pollen limitation and, consequently, may be sensitive to pollinator-mediated phenotypic selection on flower size through female function (Teixido and Valladares, submitted elsewhere). We here experimentally assessed the relationship between flower size and relative fitness and, at least partly, its dependence on pollinators by conducting flower size manipulation and pollinator visit rates. This implies that flower size may be potentially mediated by pollinators and other selective agents, such as environmental factors. In this regard, high temperatures and water shortage of Mediterranean environment may affect flowering and limit plant reproduction (Larcher 2000; Thompson 2005; Aragón et al. 2008). Larger-flowered individuals of C. ladanifer involve higher indirect costs in terms of fruit and seed production (Teixido and Valladares 2013). As a consequence, Mediterranean stressful conditions may limit flower size, potentially favoring small-flowered plants and thus indicating that resource limitation may be greater than pollen limitation (Galen 2005; Teixido and Valladares 2013).

Studying a large-flowered species inhabiting a Mediterranean ecosystem represents a good model system to determine current phenotypic selection on flower size through male and female success, thus evaluating whether flower size is more resource-than pollen-limited through female function and it is male fitness the one differentially responding to flower size variation. Specifically, we hypothesized that (1) larger flowers receive more pollinator visits; (2) this relationship increases male reproductive success and, to a lesser extent, the female one; and (3) this pattern translates into sex- and context-dependent phenotypic selection on flower size, with differential selection towards larger flowers through male function.

Materials and methods

Species and study area

Cistus ladanifer (Cistaceae) is a shrub 100–250 cm tall that inhabits open, hot and dry areas with acid soils of the western Mediterranean. The flowering period spans March to June and each plant produces white flowers of ~7–10 cm in diameter, often exhibiting dark-coloured spots at their bases (Muñoz-Garmendía and Navarro 1993; Teixido et al. 2011). The flowers are the largest in the family with an average of more than 150 anthers and 1,000 ovules, are self-incompatible and hermaphroditic and secrete some nectar (Herrera 1992). Flower opening occurs synchronously each day within populations and flowers last only several hours when pollinated and/or under warm temperatures (Teixido and Valladares unpublished data). Fruits are globular woody capsules with a variable number of valves (5–12) and seeds (approx. range 300–1,200) 0.8 × 0.6 mm in size (Talavera et al. 1993; Narbona et al. 2010).

The study was conducted from April to July of 2013 in two populations in Madrid province, central Spain (39°53′–41°09′N, 3°03′–4°34′W). The two populations were merely chosen as replicates. Both populations had similar orientation (south), slope (0°–10°) and tree canopy cover (0–10 %). One population was located in Tres Cantos [732 m above sea level (a.s.l.); 40°34′N, 3°42′W], where individuals bloom between April and May. Substrate is predominantly clay and sand and vegetation is dehesa-like with scattered Quercus ilex L. (Fagaceae) and Pinus pinea L. (Pinaceae) interspersed in a shrub matrix. Mean annual temperature is 14 °C and mean annual precipitation is 544 mm (Ninyerola et al. 2005; N = 20 years). The other population was located in El Escorial (1,156 m a.s.l.; 40°35′N, 4°09′W) where plants bloom in June. Substrate is granite and shrubby vegetation is interspersed with scattered Pinus pinaster Aiton (Pinaceae) and Juniperus oxycedrus Sibth. and Sm. (Cupressaceae) trees. Mean annual temperature is 11 °C and mean annual precipitation is 899 mm (Ninyerola et al. 2005; N = 20 years).

Experimental design

Flower size manipulation

During the flowering peak (when all the individuals bloomed more than 20 flowers per day). 30 plants per population without spots on their corollas were randomly selected and tagged. We selected this phenotype to avoid possible effects of these spots on pollinator visit rates. In other species, dark petal spots have been shown to act as visual signals for insect pollinators (Johnson and Midgley 1997; Thomas et al. 2009). Additionally, flowers would turn into dark flowers when reduced. Flower size was experimentally manipulated to evaluate its effect on pollinator visit rates and three reproductive success components, one for the male function (pollen dispersal) and two for the female one (fruit and seed production). We conducted the experiment under sunny conditions, suitable for pollinator activity. At each population, for 10–15 days, we cut petals at sepal height of the half of the flowers of each plant on 2–5 plants per day at predawn before the opening of the flowers. Thus, we artificially reduced flower diameter and we also divided each plant in two different treatments having a similar number of flowers to avoid the possible effect of flower number on pollinator attraction: (1) unmanipulated control flowers and (2) flowers with reduced size by cutting petals (hereafter “reduced flowers”) (Fig. 1).

Fig. 1
figure 1

Control (left) and reduced (right) flowers of Cistus ladanifer

Full size image

In each treatment, flowers (except those collected to estimate male success; see “Male and female reproductive success”) were tagged with threads differing in colour and left to natural pollination until afternoon. Reduced flowers were randomly chosen to avoid possible differences in the number of flowers at each treatment with a particular orientation. Petals were cut with large kitchen scissors which allowed a simple and single cut without mechanical damage to corollas. Reduced flowers also maintained the original floral shape of C. ladanifer, resembling natural small flowers (Fig. 1). Corolla diameter (cm) of five flowers per treatment and plant was recorded using a caliper (to the nearest mm) and then averaged per treatment and plant as a proxy for flower size. The diameter of reduced flowers varied between 3.0 and 6.1 cm (mean ± SD 4.77 ± 0.54 cm), about the half of the natural diameter (mean ± SD 8.24 ± 0.88 cm).

Pollinators

We evaluated the relationship between the number and identity of pollinators with flower size and subsequent effects on reproductive male and female success. The insect observations were conducted on sunny days with little wind during the flowering peak at each population. Between 2 and 3 plants used in the experimental design were daily observed. Overall, pollinator visit rates were recorded in 20 plants per population and, on each plant, we observed during four 10-min periods on five flowers per treatment, covering up to ~27 h of sampling per population. In Tres Cantos, observations were conducted between 10:00 a.m. and 14:00 p.m., and in El Escorial between 9:00 a.m. and 13:00 p.m., corresponding to the peak of pollinator activity, respectively.

During each observation period we noted the number and identity of visitors to flowers and number of visits per each visitor. A visit was defined to have occurred when the visitor’s body contacted stigma and/or the anthers. At each treatment and plant, we calculated visit rate as total number of visits per 40 min. We categorised each visitor into seven pollinator functional groups or clusters of pollinator species that behave in a similar way in the flowers (Fenster et al. 2004). The functional groups were bumblebees (Bombus spp.), solitary bees (Andrenidae, Colletidae and Halictidae), honeybees (Apis mellifera), wasps (Ichneumonidae), hover flies (Syrphidae), muscoid flies (Muscidae and Anthomyiidae) and beetles (Coleoptera). Then we recorded the frequency of visits of each pollinator functional group to each plant. In the absence of data on a visitor’s efficiency, the frequency of visits can be used as a surrogate of their relative potential importance for the plant species (Fenster et al. 2004).

Male and female reproductive success

To evaluate the male fitness based on differences in flower size we carried out an indirect estimate by means of the amount of dispersed pollen in the flowers of each treatment and plant after natural pollination occurred. Hence, we collected four flowers on each plant at predawn, before the anthesis, and four flowers per treatment and plant between 14:00 a.m. and 15:00 p.m., after the peak of pollinator activity. Though occasionally the flowers remained open after those hours, floral longevity in C. ladanifer is short and lasts only few hours (Teixido et al. 2011), so after noon petals drop off and calyxes close up. All flowers were individually frozen at −10 °C. In the laboratory, the anthers of each flower were collected and oven-dried for 24 h at 60 °C and then weighed to the nearest 0.1 mg with a microbalance (MX5; Mettler-Toledo International, Greifensee, Switzerland). Thus, we recorded the dry mass of each flower’s pollen (mg) and assessed the mean dry mass of both undispersed pollen per plant from those flowers collected at predawn before anthesis and undispersed pollen for control and reduced flowers. To obtain an estimate of pollen dispersal, the individual dry mass of undispersed pollen per treatment and plant was subtracted from the mean dry mass of undispersed pollen per plant at each population. This difference is a good estimate of pollen dispersal in C. ladanifer, where pollen dry mass is significantly correlated with number of anthers (r p = 0.91, p = 0.004, N = 40) and number of anthers with flower size (Herrera 1992; Talavera et al. 1993).

To determine the female fitness, all ripe fruits from previously tagged flowers per treatment and plant were picked before seed dispersal in July to evaluate fruit set and mean seed number per fruit, treatment and plant (hereafter “seed number”). Fruit set estimates pollination intensity as a proportion of pollinated flowers, whereas seed number estimates the quality of mating (Fenster et al. 2004). Fruit set per treatment and plant was obtained by dividing the number of mature fruits set by all flowers tagged per treatment and plant, respectively. To determine seed number, 5–10 mature fruits were randomly selected per treatment and plant and then the seeds per fruit were counted using a four digit manual hand tally counter. Subsequently, the number of seeds recorded per fruit was added up at each treatment and plant and then averaged (±SD) by dividing by the number of fruits utilized, thus recording mean seed number per fruit, treatment and plant (i.e., seed number).

Statistical analysis

To test differences between populations and years in the frequency of visits of each pollinator functional group to each plant we used PerMANOVA. PerMANOVA is a permutation-based version of the multivariate analysis of variance (Anderson 2001). It uses the distances between samples to partition variance and randomizations or permutations of the data to produce the p value for the hypothesis test. It is non-parametric (or semi-parametric for multi-factor models) and, therefore, robust to the assumption of multivariate normality making it less prone to Type I errors. Count data of visits of each pollinator functional group were square root transformed to improve normality. Bray–Curtis similarity index was calculated before performing the analysis (Anderson 2001). All PerMANOVA analyses were performed in Primer 6.0 (Clarke and Gorley 2006).

To determine whether pollinator visit rates increase with flower size and significantly differ between populations, we conducted an ANOVA including population and treatment (fixed factors), plant within population (random factor) and the interaction between population and treatment. A significant interaction indicates a differential effect of the treatment on visit rates depending on the population. Assumptions of normality and homogeneity of variance were tested using Shapiro–Wilk’s test and Levene’s test, respectively. ANOVA was performed using the lme4 statistical package in R v2.12.1 (R Development Core Team 2010).

To determine the effects of the treatment on reproductive success we fitted three Generalized Linear Mixed Models (GLMMs) with population and treatment (fixed factors), plant nested within population (random factor) and the interaction between population and treatment on every fitness component, i.e., pollen dispersal, fruit set and seed number. A significant interaction indicates a differential effect of the treatment on components of reproductive success between populations. For pollen dispersal and seed number we assumed a normal error distribution with an identity link function. For fruit set we assumed a binomial error distribution with a logit link function. For all models we used the restricted maximum likelihood (REML) and, because our data were unbalanced, we used Satterthwaite’s method to determine the approximate denominator degrees of freedom of residuals (Littell et al. 1996; Quinn and Keough 2002). Additionally, since the treatment affected both pollinator visit rates and components of reproductive success (see “Results”), we subsequently tested the effect of visit rates on pollen dispersal, fruit set and seed number in each population, respectively, by means of linear regression. All the GLMMs were performed using the GLIMMIX Macro of SAS (SAS Statistical Package 1990; SAS Institute, Cary, NC, USA) and the regressions were analysed with R (R Development Core Team 2010).

To estimate phenotypic selection on flower size through every component of reproductive success we assessed standardized selection differentials (s) using linear regression analyses with relative pollen dispersal, fruit set and seed number (individual fitness/population mean fitness, w) as the response variable, respectively, and standardized flower size (with a mean of 0 and a variance of 1) as explanatory variable (Lande and Arnold 1983). We used floral diameter of both reduced and control flowers as an estimate of flower size of each population. Likewise, we used pollen dispersal, fruit set and seed number of both reduced and control flowers as fitness components. Each fitness component was correlated with flower size in a linear regression where the slope was the estimate of the strength and direction of linear selection (Lande and Arnold 1983; Kingsolver et al. 2001). Additionally, we calculated nonlinear selection gradients (γ) to estimate stabilizing/disruptive selection by obtaining quadratic deviations from the mean for both single and quadratic terms of flower size (Lande and Arnold 1983). Therefore, we used flower size and its quadratic component in the regression model. Quadratic regression coefficients were doubled to estimate properly stabilizing/disruptive selection gradients (Lande and Arnold 1983; Stinchcombe et al. 2008). All the regression models were performed in R (R Development Core Team 2010).

Results

Pollinators

Five and seven functional groups were identified in Tres Cantos and in El Escorial, respectively (Fig. 2). In Tres Cantos, functional groups were similar for both treatments but, in control flowers, honeybees accounted for nearly 50 % of visits, whereas the variability of pollinators on reduced flowers was higher, dominated mostly by bees and flies (muscoid and hoverflies) (approx. 95 %). In El Escorial, beetles were the dominant functional group in both treatments, especially in reduced flowers (approx. 40 %), along with muscoid and hoverflies. Together, these three groups (beetles, muscoid flies and hoverflies) accounted for 80–90 % of visits in this population. Visits by bees were scarce and so were visits by bumblebees and wasps (these only in reduced flowers). The latter differences in type of pollinators were significant between populations and treatments (pseudo-F 1, 44 = 8.45, p < 0.001; pseudo-F 1, 44 = 23.54, p < 0.001, respectively), but no statistical differences were found between the interaction of treatment and population (pseudo-F 1, 44 = 0.98, p = 0.389).

Fig. 2
figure 2

Frequency (% ± SE) of visits of each functional group to flowers of Cistus ladanifer depending on the treatment in a Tres Cantos and b Canencia

Full size image

Whereas flowers of C. ladanifer were significantly larger in El Escorial (F 1, 95 = 6.30, p = 0.013, see Table 1), visit rates were constant between populations (Tables 1, 2). However, visit rates differed significantly between treatments (Table 2). This entailed a positive effect of flower size since in control flowers there was a twofold increase in visit rates in both populations (Fig. 3, mean range 0.30 ± 0.22–0.61 ± 0.29; see also Table 2, Population × Treatment not significant). Pollinator visit rates also showed a high variation between plants and treatments (range 0.03–1.08), with minimum values corresponding to reduced flowers (≤0.78) and the highest ones to control flowers (≥0.23).

Table 1 Measurements of size and total pollen dry mass of flowers, pollinator visit rates, and male and female fitness (mean ± SD) in Cistus ladanifer in the two study populations
Full size table
Table 2 ANOVA for differences in pollinator visit rates to flowers of Cistus ladanifer between populations, plant (population), treatment and the interaction population × treatment
Full size table
Fig. 3
figure 3

Differences in mean pollinator visit rates (±SE) between control and reduced flowers at each study population in Cistus ladanifer

Full size image

Male and female reproductive success

The amount of pollen dispersed per plant in C. ladanifer was highly variable (mean range 16.7–66.6 %), but the mean dry weight of dispersed pollen was similar between populations, despite differences in mean total pollen per flower (Tables 1, 3). Thus, overall, flowers of El Escorial dispersed the same amount of pollen as flowers of Tres Cantos, despite having more pollen available (Table 3). Control flowers dispersed significantly more pollen than the reduced ones (Table 3; Fig. 4a). Overall, control flowers dispersed about 8–48 mg of pollen, whereas reduced flowers dispersed between 0 and 40 mg. In Tres Cantos, control flowers dispersed up to 44 % of pollen, whereas reduced flowers dispersed up to 37 %. In El Escorial, these values were 38 and 29 %, respectively. Concerning the relationship between visit rates and male reproductive success, significant effects were detected only in El Escorial (Table 4). In fact, this relationship was also detected in this population when analyzing control flowers separately, i.e., under natural conditions of flower size variation (R 2 = 0.58, F 1, 19 = 12.33, p = 0.007).

Table 3 General linear mixed models (GLMMs) for differences in the three components of fitness (dispersed pollen, fruit set and seed number) of Cistus ladanifer between populations, treatments (fixed factors), plant (random factor nested within population) and the interaction population × treatment (fixed factor)
Full size table
Table 4 Simple linear regressions for pollinator visit rates against pollen dispersed, fruit set and seed number in each of the study populations
Full size table

Fruit set in C. ladanifer ranged between 31 and 100 % but was, on average, relatively high (mean range ± SD 77.05 ± 13.43 and 85.30 ± 12.25 % for reduced and control flowers, respectively, Fig. 4b). Although we did not find any significant difference in fruit set between populations, this component did significantly vary between treatments and plants (Table 3). However, the effect of treatment was due to differences found in El Escorial since in Tres Cantos fruit set was similar between control and reduced flowers (Population × Treatment marginally significant, Table 3; see also Fig. 4b). Visit rates did not affect fruit set in any population (Table 4).

Fig. 4
figure 4

Differences between control and reduced flowers in a mean dispersed pollen (mg ± SE), b mean fruit set (% ± SE) and c mean seed number (±SE) at each study population in Cistus ladanifer

Full size image

Variability in seed number was very high (mean ± SD 127.51 ± 145.32 and 1533.33 ± 234.87) and all the analysed variables had a significant effect (Table 3). Overall, between treatments, seed number was 13 % higher in control flowers (mean range ± SD 796.56 ± 206.70 and 703.31 ± 108.51; control vs. reduced flowers, respectively). Between populations, seed number was 14 % higher in Tres Cantos (Table 1). The effect of the interaction between population and treatment was also significant, as evidenced by a larger decline in seed number in reduced flowers compared to control flowers in El Escorial compared to Tres Cantos (Fig. 4c). Although the number of seeds decreased a 4 % in reduced flowers in Tres Cantos, this decrease reached to a 25 % in El Escorial. Visit rates also significantly influenced seed number in El Escorial (Table 4).

Phenotypic selection

The results revealed sex-dependent effects in the phenotypic selection on flower size. Thus, phenotypic selection had a differential effect through male success. In both populations, a direct positive selection was detected on flower size through pollen dispersal (Table 5). Relative to female function, selection towards larger flowers was only detected in El Escorial through seed number, but the strength of selection was lower than for pollen dispersal (Table 5).

Table 5 Standardized selection coefficients for flower size in Cistus ladanifer on the three components of fitness (dispersed pollen, fruit set and seed number) in each of the study populations
Full size table

Discussion

This study provides insights into spatial variation in the patterns of pollinator visits as well as current phenotypic selection on flower size through both sexual functions in a hermaphroditic and large-flowered species in a Mediterranean environment. Overall, we verified that larger flowers in C. ladanifer attract more pollinators within populations and this relationship differentially affects sexual functions, especially favoring male fitness by significantly increasing pollen dispersal. The female function benefits to a lesser extent and may depend on the strength of pollen and/or resource limitation on fruit set and seed number. In Tres Cantos, where the flowers were smaller, pollinator assemblage might reduce pollen limitation. Otherwise, in El Escorial, pollination seems to be scarce to fertilize all the ovules of these larger flowers. In agreement with this, interestingly, we detected sex-dependent phenotypic selection on flower size, with stronger selective pressures towards larger flowers through male fitness.

Variation in pollinator visit rates

The interactions between C. ladanifer and pollinating agents recorded in our study supported the generalist character of this species. Pollinators were numerous and diverse and also significantly varied between populations. In fact, pollinator assemblage was also different in other populations of this species in SE Spain (79 % of Diptera and 4 % of bees: Talavera et al. 1993). C. ladanifer has large flowers with an unrestrictive morphology that favors attraction to a high diversity of insects and also increases individual fitness in generalist plants (reviewed in Willmer 2011). This pollination system is shared with most of Cistaceae (Bosch 1992; Herrera 1992; Talavera et al. 1993, 2001) and may have worked as an adaptation to spatial–temporal variation in the relative abundances of most effective pollinators (Herrera 1996). Therefore, generalization is common in C. ladanifer as a relevant factor towards the adaptation to the high variability in pollination environment, favoring pollen transfer among individuals and, thus, a high percentage of fruit and seed production, which ultimately may favor the display of these large flowers in the Mediterranean ecosystem.

Sex differences and selection patterns

Larger flowers of C. ladanifer attracted higher visit rates and a different pollinator assemblage that, overall, favored pollen dispersal. Flowers of this species produce a high pollen amount and this pattern is also related to flower size (Herrera 1992; Teixido and Valladares 2013). This favors pollen dispersal in larger flowers not only by increasing pollinator visit rates and efficiency, but also by having more available pollen. Although in our study reduced flowers contained on average the same amount of pollen than control flowers since only size was experimentally manipulated, this process of pollen dispersal may not be uncommon in natural populations of this species. As a consequence, selection would tend to operate on a correlation between flower size and pollen amount.

Differential pollen dispersal in larger flowers in C. ladanifer provides some suggestion that pollinators play a role in selection on flower size through male fitness in the study populations. We are aware that pollen dispersal is only a component of male fitness and has been found to have little relationship to siring seeds in other systems (Conner 2006; Herrera et al. 2006). However, as we stated above, pollen dispersal is often measured as a representative component affecting male reproductive success and its measurement is valuable for elucidating the mechanisms of selection (Snow and Lewis 1993; Arista and Ortiz 2007). Although not all dispersed pollen will contribute male fitness, pollen collection and export should be proportional to pollen donation (Holland et al. 2004).

Following our data, visitation rate only explains male fitness estimates in El Escorial, even only for natural variation in flower size as measured in control flowers. Therefore, it is difficult to see how pollinators are driving selection in Tres Cantos, where there is no relationship between visitation and pollen dispersal. The absence of this relationship could imply that in some control flowers a few visits disperse higher pollen amount than more visits do in reduced flowers, which would be evidence for the importance of pollinator efficiency on pollen dispersal. In a generalist species as C. ladanifer it is likely that a high percentage of visitors are inefficient pollinators, as showed in the sister species, C. libanotis (Talavera et al. 2001). However, pollinator visit rates as well as dispersed pollen responded similarly to the treatment in both populations, so efficiency-dependent pollen dispersal appears not to be relevant. Either way, data suggest that pollinators are not the agents driving selection in this population and, likely, we failed in detecting pollinator-mediated phenotypic selection on flower size through pollen dispersal. Otherwise, in El Escorial, flowers were larger and had a higher amount of pollen, and pollinator visit rates were the same as in Tres Cantos. Therefore, there was a higher pollen dispersal limitation and, consequently, higher selection opportunity through male fitness. Most interestingly, this opportunity is also sizable taking into consideration only natural variation in flower size as recorded in control flowers, thus supporting the importance of this trait for increasing male reproductive success. Overall, our results entail that pollen limitation through male function may also be important and that the strength of selection on flower size through this sex may likewise vary among populations.

Several works have documented current phenotypic selection towards larger flowers through male fitness in different plant species (Stanton et al. 1986; Galen 1989; Maad and Alexandersson 2004; Arista and Ortiz 2007; Hodgins and Barrett 2008). Compared to the sister species C. salviifolius, the strength of selection through pollen dispersal in our study is somewhat low (see Arista and Ortiz 2007). We recorded that pollen dispersal was lower than 50 % (against 80 % in Arista and Ortiz’ work), which is far from satisfying male function. This suggests that pollen dispersal does not become saturated and plants may produce more pollen than that is potentially available, always favoring male function under benign pollination conditions. However, our data of pollen dispersal could be due to differences in the methodology carried out. Arista and Ortiz (2007) recorded pollen dispersal on flowers picked at sunset, in contrast with flowers picked at afternoon in our study. Other plausible explanations may also be due to differences in visit rates and/or efficiency of pollinators, but they did not record pollinator data, whereas we did not record their efficiency, so interpretations should be made with caution.

Flower size had less effect on female fitness in our study species. We only detected selection towards larger flowers through seed number in one population. Female function is usually dependent on ecological context, thus being pollen or resource-limited (Ashman and Morgan 2004). A general absence of a positive relationship between pollinator visit rates and female fitness suggests low levels of pollen limitation in C. ladanifer, a common pattern in other Cistus (Bosch 1992; Herrera 1992; Talavera et al. 1993, 2001; Arista and Ortiz 2007). Following this assumption, selection opportunity on flower size through female sex is potentially low, even for large differences in pollinator visit rates among individuals with contrasting flower size (Johnson 1996; Totland 2004). Absence of selection on flower size through female fitness components has been detected in other plant species, especially depending on ecological context (Totland 2001; Ashman and Morgan 2004) and also in other Cistaceae (C. salviifolius: Arista and Ortiz 2007; Halimium atriplicifolium: Teixido 2013).

However, we did detect a trend to produce more fruits and seeds in larger flowers in El Escorial. In fact, we also detected that differences in pollinator visit rates affected seed number in this population. This suggests that smaller flowers may be limited by pollen receipt, thus reducing the number of seeds per fruit. Most interestingly, phenotypic selection on flower size through seed number in this population was almost as strong as through pollen dispersal. This may be due to flowers in El Escorial being larger than in Tres Cantos. However, they received a similar number of visit rates and different composition of pollinators. Hence, both the number and the efficiency of pollinators were inadequate to fertilize all the ovules, thus increasing pollen limitation in smaller flowers and favoring directional selection on flower size. Previous studies have documented current phenotypic selection towards larger flowers through seed number associated with pollen limitation (Totland 2001; van Kleunen and Ritland 2004; Hodgins and Barrett 2008; Nattero et al. 2010b).

Jointly, resource limitation may also be important in this population and plants might allocate more resources to control, more visited flowers, than to reduced, less visited flowers. Under resource limitation conditions, a selective abortion in fruit and seed production in pollen-limited flowers may improve both the quantity and quality of offspring in pollen-saturated flowers on the same plant (Lloyd 1980; Haig and Westoby 1988). In fact, the pattern of resource investment between flowers and fruit and seed production is relevant in the reproductive function of C. ladanifer (Teixido and Valladares 2013). Nevertheless, this assumption should be taken with caution since we lack data of resource availability and allocation in the present study. Flower size is significantly correlated with ovule number in this species (Herrera 1992) and, in El Escorial, flowers were larger than in Tres Cantos, but they received the same pollinator visit rates and produced less fruits and seeds. Based on these results, pollen rather than resource limitation seems to be more important in this population.

Differences in pollinator assemblage, pollen limitation and their effects on seed number between populations could also respond to the effect of other biotic factors such as coflowering species (nurse plants: Ghazoul 2006; competitor plants: Caruso 2001), the effect of flower number, which may or may not increase pollinator visit rates (Thompson 2001; Harder and Johnson 2009; Brys and Jacquemyn 2010) and/or to selective pressures imposed by antagonists. For example, the incidence of florivores was significantly influenced by flower size in C. ladanifer (Teixido et al. 2011). Florivores may reduce fitness of both sexes by degrading the attractive properties of flowers and/or by direct consumption of available gametes (Krupnick et al. 1999; Cardel and Koptur 2010). In this context, florivores may exert negative selective pressures on the same floral traits positively selected by pollinators (McCall and Irwin 2006). Other plausible mechanisms could also be related to the effect of abiotic conditions giving rise to spatial variation in the patterns of selection through female fitness (Herrera 1995; Totland 2001, 2004). In this regard, floral costs and differential resource allocation between functions involves indirect effects on female function (Cresswell 1998). In C. ladanifer, flower size is significantly related to greater floral resource allocation (Herrera 1992; Teixido and Valladares 2014). This differential resource allocation to larger flowers has shown that flower size entails higher indirect costs in terms of fruit and seed production (Teixido and Valladares 2013). Overall, these patterns support the idea that selection on flower size is not only dependent on pollinators, but rather that this trait responds to adjustments between costs and benefits and that this balance is closely linked to sex.

When both sexes have different optima for those traits related to reproductive success, there may be a sexual conflict giving rise to traits in equilibrium (Chapman 2006). In C. ladanifer, flower size affected both sexes in El Escorial but only the male function in Tres Cantos. The result in the latter population implies differences in the evolutionary pressures of both sexes that could lead to sexual selection conflicts on this trait in this population. Taken together, our results show that differences in selective pressures between sexes are essential to understand the variation in flower size and the evolution of large flowers in Mediterranean environments. In order to improve our understanding about evolution of flower size in this large-flowered Mediterranean species, it would also be necessary to carry out further among-population variation and temporal studies to determine which sex differentially influences plasticity on this trait and more strongly affects phenotypic selection.

Conclusions

This work shows a positive relationship between flower size and pollinator visit rates and its effects on reproductive success in C. ladanifer. In addition, we detected sex- and context-dependent (i.e., spatial variation between populations) phenotypic selection on flower size. Overall, our results suggest that pollinators may play a role in selection on flower size in natural populations of this species, at least to a certain degree. The experimental manipulation of this trait and the use of pollen dispersal as an indirect estimate of male fitness showed adequate methods to successfully evaluate our objectives. The different patterns of selection on flower size verify the importance of estimating phenotypic selection through both sexes and different components of fitness together with differences in environmental conditions between populations. Taken together, these patterns support the “male function” hypothesis in our study species but also suggest that inherent conditions to each site have the potential to create among-population differences in flower size by local adaptations to climatic and pollinator environment (i.e. to resource and pollen limitation, respectively). As a consequence, optimal flower size through male fitness appears to be larger than through female one, in such a way that whether flower size were halfway between both optimal values, an actual sexual conflict would also act as a stabilizing mechanism of this trait. Therefore, flower size variation will ultimately be a sexual conflict modulated by the environment and mainly constrained by female fitness in this large-flowered Mediterranean species.