Museo Nacional de Ciencias Naturales, Consejo Superior de Investigaciones Científicas, MNCN-CSIC, Serrano 115 dpdo, E Madrid, Spain

Transcription

1 bs_bs_banner Botanical Journal of the Linnean Society, 2014, 176, With 6 figures Pollinator-mediated phenotypic selection does not always modulate flower size and number in the large-flowered Mediterranean shrub Cistus ladanifer (Cistaceae) ALBERTO L. TEIXIDO 1, * and FERNANDO VALLADARES 2 1 Área de Biodiversidad y Conservación, Escuela Superior de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, Tulipán s/n, Móstoles, E Madrid, Spain 2 Museo Nacional de Ciencias Naturales, Consejo Superior de Investigaciones Científicas, MNCN-CSIC, Serrano 115 dpdo, E Madrid, Spain Received 20 January 2014; revised 12 August 2014; accepted for publication 16 August 2014 Flower size and number usually evolve under pollinator-mediated selection. However, hot, dry environments can also modulate display, counteracting pollinator attraction. Increased pollen deposition on larger flower displays may not involve higher female fitness. Consequently, stressful conditions may constrain flower size, favouring smaller-sized flowers. The large-flowered, self-incompatible Mediterranean shrub Cistus ladanifer was used to test that: (1) this species suffers pollen limitation; (2) pollinators are spatially temporally variable and differentially visit plants with more/larger flowers; (3) increased visits enhance reproduction under pollen limitation; (4) stressful conditions reduce female fitness of larger displays; and (5) phenotypic selection on floral display is not just pollinator-mediated. We evaluated pollen limitation, related floral display to pollinator visits and fruit and seed production and estimated phenotypic selection. Flower size was cm and varied spatially temporally. Visitation rates (total visits/50 min) ranged from 0.26 to 0.43 and increased with flower size. Fruit set averaged 80% and seed number averaged 855, but only fruit set varied between populations and years. Selection towards larger flowers was detected under conditions of pollen limitation. Otherwise, we detected stabilizing selection on flower size and negative selection on flower number. Our results suggest that selection on floral display is not only pollinator-dependent through female fitness in C. ladanifer The Linnean Society of London, Botanical Journal of the Linnean Society, 2014, 176, ADDITIONAL KEYWORDS: fruit set pollen limitation pollinator visitation rates resource limitation seed number stabilizing selection. INTRODUCTION Pollinators are considered to be a major factor driving evolution of floral phenotypes (Darwin, 1862; Stebbins, 1970; Fenster et al., 2004). Because flower size and number are involved in pollinator attraction, these floral traits have often been described to evolve under pollinator-mediated selection (e.g. Galen, 1989; Herrera, 1993; Conner & Rush, 1997; Hodgins & Barrett, 2008; Harder & Johnson, 2009). However, large and numerous flowers are associated with greater requirements of biomass, energy and water *Corresponding author. alberto.teixido@urjc.es for floral development (Galen, 1999; Halpern, Adler & Wink, 2010) and with higher maintenance costs due to high respiration and transpiration rates (Vemmos & Goldwin, 1994; Galen, 2000). Accordingly, several studies have documented that simultaneous constraints from pollen availability and other environmental factors influencing female reproductive output can affect attractiveness of floral traits (Galen, 1999, 2000; Totland, 2001; Caruso, 2006; see also review by Strauss & Whittall, 2006). Pollinator assemblage and visitation rates that determine pollen availability have been reported to show spatial temporal variation (Herrera, 1988, 1996; Fenster & Dudash, 2001; Aigner, 2005; 540

2 SELECTION FOR FLORAL DISPLAY IN CISTUS 541 Minckley & Roulston, 2006). It follows that the strength and direction of pollinator-mediated selection may differ in space and time, complicating our understanding of the evolution of floral traits. Only a few investigations have analysed the effects of the variation in pollination environment on the direction and magnitude of pollinator-mediated selection acting on intraspecific spatial temporal floral variation (Aigner, 2005; Sletvold & Ågren, 2010; Sletvold et al., 2012). However, these studies are essential to identify reliable estimates of selection (Conner, 2006; see also Herrera, 2009). From an evolutionary perspective, the high spatial temporal variability in the pollination environment has mostly explained the existence of generalized pollination systems, thus promoting efficient animalmediated cross-pollination (reviewed by Gómez & Zamora, 2006). Such generalization may reduce pollen limitation and, consequently, limit selection towards larger flowers through female fitness (Aigner, 2001; Gómez & Zamora, 2006; Sahli & Conner, 2011). However, even under pollen limitation, positive directional selection on flower size and number through fruit and seed production may be relaxed due to other mechanisms, such as non-pollinator agents of selection or resource allocation among reproductive traits (Ashman & Morgan, 2004). For example, greater pollen deposition on larger flowers may not involve a differential fruit and seed production due to resource limitation, and it can lead to stabilizing selection (reviewed by Strauss & Whittall, 2006). However, despite some evidence that stabilizing selection on floral traits engaged in pollinator attraction may be relatively common in plant populations (Cresswell, 1998; Kingsolver & Diamond, 2011), this form of selection has rarely been measured in natural populations visited by a large and variable assemblage of pollinators. Cistus L. (Cistaceae), a common genus in the Mediterranean area, has a generalized pollination system and low levels of pollen limitation (Bosch, 1992; Herrera, 1992; Talavera, Gibbs & Herrera, 1993; Talavera et al., 2001). High temperatures and water shortage in Mediterranean ecosystems strongly limit plant reproduction and can disrupt the normal performance of flowers, affecting fruit and seed production (Larcher, 2000; Galen, 2005; Thompson, 2005; Aragón, Escudero & Valladares, 2008). As a consequence, these stressful conditions may constrain flower size, making small flowers potentially advantageous for plants living in these environments (Galen, 2000, 2005; Herrera, 2005; Teixido & Valladares, 2013). However, most Cistus species have large flowers, and therefore pollinator-mediated selection on flower size could be relevant. In agreement with this, pollen limitation could impose stronger selective pressures than those imposed by resource limitation (Ashman & Morgan, 2004). Hence, experimental analyses of pollen availability by supplemental hand-pollination are necessary to determine pollen limitation levels in natural populations and the role of pollinators as selection agents (Galen, 1996; Knight et al., 2005; Fishman & Willis, 2008; Sandring & Ågren, 2009). Here, we test whether pollinators can act as agents of selection on flower size and number through two components of female fitness in natural populations of Cistus ladanifer L., a large-flowered, generalist, hermaphroditic and self-incompatible shrub inhabiting the Mediterranean area (Herrera, 1992). The species possesses an extremely stable self-incompatibility system, regardless of populations and years (Guzmán, Narbona & Vargas, 2013). A large-flowered, selfincompatible species occurring in the Mediterranean is a good study system to determine the potential role of pollinators on selection on floral display through female fitness in a stressful environment. Selection on flower size could also occur through male fitness by enhancing pollen dispersal and siring success, as reported for other Cistus species (Talavera et al., 2001; Arista & Ortiz, 2007). Sex allocation theory suggests that selection on flower size should be stronger through male fitness, whereas female fitness is resource-limited, especially in non-pollen-limited species (Bateman, 1948; Bell, 1985; Ashman & Morgan, 2004). However, there is a growing body of literature questioning the generality of the male function hypothesis (Wilson et al., 1994; Morgan & Conner, 2001; Wright & Meagher, 2004; Hodgins & Barrett, 2008) and sex allocation theory seems to be more context-dependent (Ashman & Morgan, 2004). Therefore, and despite the potential influence of male function, the conditions of our study system (i.e. a large-flowered species inhabiting a Mediterranean ecosystem) suggest a key role for the components of female fitness, which have been the target of our research. In this work we have first experimentally evaluated pollen limitation and then observationally estimated pollinator-mediated selection in two populations over two years to examine variation in the evolution of floral traits involved in pollinator attraction. Specifically, our study addressed the following hypotheses: (1) natural populations of C. ladanifer may suffer pollen limitation, at least to a certain degree; (2) pollinator visitation rates are spatially temporally variable (between sites and years) and are affected by floral display, with more and larger flowers attracting more visitors; (3) increased pollinator visitation rates enhance female fitness only under pollen limitation conditions; (4) under favourable conditions for pollination, stressful conditions of the Mediterranean environment reduce female fitness of

3 542 A. L. TEIXIDO and F. VALLADARES more and larger flowers; and (5) the direction and magnitude of phenotypic selection on floral display is not always dependent on pollinator visitation rates and may constrain the number and size of flowers under the stressful conditions in which C. ladanifer lives. MATERIAL AND METHODS STUDY SPECIES Cistus ladanifer is a shrub cm tall that inhabits acidic and dry soils in warm open areas in the western Mediterranean. The flowering period is from March to June, each plant producing many white flowers c cm in diameter, often exhibiting darkcoloured spots at their bases (Muñoz-Garmendía & Navarro, 1993). These flowers are the largest in the family and have on average > 150 anthers and 1000 ovules (Herrera, 1992). Flowers are solitary, hermaphroditic and self-incompatible and secrete some nectar (Herrera, 1992). Flower opening occurs synchronously each day within the populations and flowers last only several hours when pollinated and/or under high temperatures (A. L. Teixido and F. Valladares, unpubl. data). Fruits are globular woody capsules with a variable number of valves (5 12) and seeds (approx. range: ) mm in size (Talavera et al., 1993; Delgado et al., 2008; Narbona et al., 2011). STUDY SITES We examined pollen limitation of C. ladanifer in two populations differing in flower size and climatic conditions in 2010 in Madrid province, central Spain ( N, W) and recorded pollinator visitation rates and measured phenotypic selection on floral display (i.e. flower size and number) at those populations in two consecutive years (2010 and 2011). Our survey at two populations in two years was intended to include variations in floral display, pollinator environment and, consequently, phenotypic selection. Both populations had a similar orientation (south), low slope inclination (0 10 ) and low tree canopy cover (0 10%), and were 30 km apart. The first population is located in Tres Cantos (elevation 732 m; N, 3.42 W) where individuals bloom between April and May. The substrate is predominantly clay and sand, and the vegetation is dehesalike with scattered Quercus ilex L. (Fagaceae) and Pinus pinea L. (Pinaceae) interspersed in a shrub matrix. The second population is located in Puerto de Canencia (elevation 1300 m; N, 3.50 W; hereafter Canencia) where individuals flower from late May to late June. The substrate is granite, and shrubby vegetation is interspersed with scattered Quercus pyrenaica Willd. (Fagaceae) and Pinus sylvestris L. (Pinaceae). In Tres Cantos, the climate is dry, whereas in Canencia it is subhumid (Ninyerola, Pons & Roure, 2005; see Table S1 for climatic data). Air temperature ( C), relative humidity (%) and rainfall (mm) were recorded at each population during the whole study period (i.e. from 1 January 2010 to 31 December 2011). Air temperature and relative humidity were recorded every hour with data loggers (Pro H8032, Onset Hobo) located 1 m above the ground. Rainfall data come from the El Goloso weather station (40.33 N, 3.42 W; elevation 725 m) for Tres Cantos and the Rascafría-El Paular weather station (40.53 N, 3.53 W; elevation 1269 m) for Canencia. Hence, we recorded both annual and flowering period means of the climatic variables measured for 2010 and Overall, the climate was hotter and drier during 2011, when precipitation was half that in 2010 during the flowering season (Table S1). POLLEN LIMITATION During the flowering peak (when all individuals had more than five flowers per day) in 2010, ten plants per population were randomly selected and tagged. Under mild weather conditions, suitable for pollinator activity, three randomly picked flowers were handpollinated (hand-pollinated flowers) and another three randomly picked flowers were left to natural pollination (control flowers) daily at each plant. Both handpollinated and control flowers were tagged with tags specifying the treatment on the label. This was done for 7 days, so that per plant a total of 21 flowers were hand-pollinated and 21 served as control. Handpollination was carried out with a fresh pollen mixture collected from five different individuals within a 5-m radius of the recipient flower. Outcross pollen was deposited on the stigma with a paintbrush 3 h after anthesis to ensure stigmatic receptivity (Herrera, 1992). Flower diameter of hand-pollinated and control flowers was recorded using a caliper (to the nearest mm) on each plant each day. In early July, all ripe fruits from 21 tagged flowers per treatment were picked before seed dispersal, and fruit set (%) was calculated for hand-pollinated and control flowers and per plant as the number of fruits set over the total number of tagged flowers. Ten fruits of both handpollinated and control flowers were randomly selected per plant to record the mean number of seeds per fruit and plant (hereafter seed number). We focused on these components of female fitness as fruit set is the most commonly measured response variable whereas seed number is the most appropriate variable for most questions related to the study of pollen limitation (Ashman et al., 2004). Additionally, both variables are strongly correlated with and determine well the magnitude of pollen limitation (Knight et al., 2005).

4 SELECTION FOR FLORAL DISPLAY IN CISTUS 543 FLORAL DISPLAY AND COMPONENTS OF FEMALE FITNESS During the flowering peak another set of 39 plants in Tres Cantos and 34 in Canencia was selected and tagged in plots of m. The same set of plants at each population was surveyed in 2010 and We recorded two plant traits related to floral display (flower size and number) to evaluate their effects on two components of female fitness (fruit set and seed number). Under mild weather conditions, suitable for pollinator activity, the number of open flowers every day was noted and four to six flowers at random were tagged daily at each plant. Overall, flowers were tagged per plant. This gave a total sample size of 1765 and 1170 flowers in Tres Cantos and 1029 and 1003 in Canencia, in 2010 and 2011, respectively (total N = 4967). Diameter (cm) of tagged flowers per plant was measured using a caliper (to the nearest mm) and then flower size and number of open flowers every day (hereafter flower number) at each plant were averaged. Mean flower size and number per plant were utilized for averaging these traits at each population and year. Although flowers secrete some nectar and its concentration and/or volume is usually correlated to pollinator visitation rates (reviewed by Fenster et al., 2004), nectar variation is also correlated to flower size variation (Herrera, 2009) so we do not consider this trait. In July, all ripe fruits from previously tagged flowers per plant were picked before seed dispersal to evaluate fruit set and 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; Knight et al., 2005). Fruit set per plant was obtained by dividing the number of mature fruits set between all flowers tagged per plant. To determine seed number, 8 15 mature fruits were randomly selected per plant. In the laboratory, the seeds of each fruit were collected and compiled in an envelope (one envelope per fruit, 8 15 envelopes per plant). Then all seeds of each envelope were carefully scattered on a white sheet and photographed to be counted by image processing in ImageJ v1.43 (ImageJ, US National Institutes of Health; Lastly, the number of seeds recorded per envelope (i.e. per fruit) was added up for each plant and then averaged (± SD) by dividing by the number of fruits utilized, thus recording the mean number of seeds per fruit for each plant. POLLINATORS We evaluated the relationship between number and identity of flower visitors with floral display and subsequent effects on fruit set and seed number. The insect observations were conducted on four warm sunny days with little wind during the flowering peak in each population and year. Fourteen plants per population and year differing in mean flower size were chosen and utilized for fruit set and seed number measurements. On each day, we tagged and observed five flowers on each of 14 plants during 5-min periods between 12:00 and 17:00 h in Tres Cantos and 09:00 and 14:00 h in Canencia. Overall, we observed 20 flowers per plant for 50 min. All plants were observed every day to avoid possible differences in individual visitation rates due to climatic variability. We observed every plant at each of the 5 h twice on two different days, to give a total of ten observation periods per plant and 140 periods during 700 min (c. 12 h) at each population and year. The order of observation of plants was random and we never observed the same plant two consecutive times or more than three times per day. The observations were always conducted by ourselves by means of direct observations. Each of us observed the same number of plants every day. We also recorded flower size of every observed flower and flower number per plant. Then both traits were averaged at each plant. 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 the stigma of the flower, as we only consider components of female fitness. On the large flowers of our study species it was easily possible to distinguish visits contacting the stigma and those that did not. However, some large beetle species did contact the stigma but they were not considered as pollinators but rather as florivores as they consume petals and nectar and when visiting another open flower they damage its stigma, thus avoiding fruit ripening (Teixido, Méndez & Valladares, 2011). Organisms not touching the stigma were generally ants, beetles and spiders. Ants usually behave as florivores by picking the stamens (Teixido et al., 2011). Some small beetle species use the corolla as a mating environment, whereas some spider species hide between the corolla and use a sit-and-wait strategy to capture prey (A. L. Teixido, pers. observ.). At each plant, we calculated visitation rate as the total number of visits per 50 min, thus having pollinator visitation rates per minute. We categorized 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). We did not identify any visitor to species level. The functional groups were: bumblebees (Bombus spp.), solitary bees (Andrenidae, Colletidae and Halictidae), honeybees (Apis mellifera), wasps (Icneumonidae), hover flies (Syrphidae), muscoid flies (Muscidae and Anthomyiidae) and beetles (Coleoptera). Then we recorded the

5 544 A. L. TEIXIDO and F. VALLADARES frequency of visits of each pollinator functional group to each plant. In the absence of data on visitor efficiency, the frequency of visits can be used as a surrogate of their relative potential importance for the plant species (Fenster et al., 2004). Lastly, we evaluated fruit set and seed number as discussed above (see Floral display and components of female fitness ). DATA ANALYSIS Pollen limitation In our experimental study to evaluate pollen limitation levels in C. ladanifer, we tested for differences in fruit set and seed number between populations and treatments (fixed factors), plant nested within population (random factor) and the interaction between population and treatment, by fitting generalized linear mixed models (GLMMs) for every response variable, i.e. fruit set and seed number. For fruit set we assumed a binomial error distribution with a logit link function. For seed number, we assumed a normal error distribution with an identity link function. For all models we used the restricted maximum likelihood (REML). All the computations were performed using the GLIMMIX macro of SAS (SAS Statistical Package, 1990; SAS Institute). Floral display and components of female fitness To estimate pollinator-mediated selection, we tested for differences in flower size and number, and in fruit set and seed number, between populations and years (fixed factors), plant nested within population (random factor) and the interaction between population and year, by fitting GLMMs for every response variable, i.e. flower number, flower size, fruit set and seed number. In all cases, except fruit set, 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 REML. All the computations were performed using the GLIMMIX macro of SAS. Pollinators To test differences between populations and years in the frequency of visits of each pollinator functional group to each plant we used PerMANOVA, 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 by each pollinator functional group were square root transformed to improve normality. The Bray Curtis similarity index was calculated before performing the analysis (Anderson, 2001). All Per- MANOVA analyses were performed in Primer 6.0 (Clarke & Gorley, 2006). We analysed the relationship between visitation rates and flower size and number and whether that relationship differed between sites and years. We conducted an ANCOVA including population and year (fixed factors), flower number and flower size (covariates) and all the possible interactions. Significant interactions show spatio-temporal differences in pollinator visitation rates and their correlation with flower number and/or flower size. We subsequently analysed whether any spatio-temporal variation in visitation rates had effects on fruit set and seed number. We conducted two new ANCOVAs, where the dependent variables were fruit set and seed number, respectively. Fruit set was arcsin transformed to achieve normality. We included population and year (fixed factor), visitation rates (covariate) and all the possible interactions. Phenotypic selection We estimated selection through fruit set and seed number in both populations in both years following Lande & Arnold (1983). We used multiple regression analyses with relative fruit set and relative seed number (individual fitness/population mean fitness, w) as the response variable and standardized trait values (with a mean of 0 and a variance of 1) as explanatory variables. We also estimated correlational selection between flower size and number as the product of both floral traits and so a measure of total floral display. Additionally, we calculated non-linear selection gradients (γ) to estimate stabilizing/ disruptive selection, by obtaining quadratic deviations from the mean for both single and quadratic terms of characters (Lande & Arnold, 1983). Therefore, we used flower number, flower size and the product flower number flower size and the quadratic components flower number 2, flower size 2 and (flower number flower size) 2 in the regression model. Quadratic regression coefficients were doubled to estimate properly stabilizing/disruptive selection gradients (Lande & Arnold, 1983; Stinchcombe et al., 2008). All the regression models, univariate and multivariate, were performed using the GLM procedure of SAS. RESULTS POLLEN LIMITATION Fruit set differed between populations, plants within populations and treatments, significantly increasing

6 SELECTION FOR FLORAL DISPLAY IN CISTUS 545 Table 1. Generalized linear mixed models for differences in fruit set and seed number (i.e. mean number of seeds per fruit and plant; see Pollen limitation in Material and methods ) of C. ladanifer between populations, hand-pollination, plants within populations and the interaction population hand-pollination Fruit set Seed number Effect d.f. Estimate ± SE Test value P d.f. Estimate ± SE Test value P Random Plant (Population) 19, ± , ± < Fixed Population 1, ± < , ± < Hand-pollination 1, ± , ± Population Hand-pollination 1, ± , ± Plant (Population), as random factor, was tested with Wald Z-test, and fixed factors with Type III F-tests. Significant P values are marked in bold. Table 2. Generalized linear mixed models for differences in flower size and flower number (i.e. mean number of open flowers per plant; see Floral display and components of female fitness in Material and methods ) of C. ladanifer between populations, years, plants within populations and the interaction population year Flower size Flower number Effect d.f. Estimate ± SE Test value P d.f. Estimate ± SE Test value P Random Plant (Population) 71, ± < , ± < Fixed Population 1, ± < , ± < Year 1, ± , ± < Population Year 1, ± < , ± < Plant (Population), as random factor, was tested with Wald Z-test, and fixed factors with Type III F-tests. Significant P values are marked in bold. in hand-pollinated flowers (mean: 81.69%, 95% CI: 69.61, 93.71%; and 87.88%, 95% CI: 78.23, 97.53%, for control and hand-pollinated flowers, respectively; Table 1). Relative to seed number we did not find any evidence of pollen limitation. Seed number differed significantly between populations and plants within populations but was similar between control and hand-pollinated flowers (mean ± SD: 878 ± 294 vs. 891 ± 280, respectively; Table 1). FLORAL DISPLAY AND COMPONENTS OF FEMALE FITNESS We did not generally detect any correlation between flower size and number, except in Canencia in 2010 (Tres Cantos: r = 0.05, P = 0.784, N = 39 and r = 0.11, P = 0.421, N = 39, for 2010 and 2011, respectively; Canencia: r = 0.45, P < 0.01, N = 34 and r = 0.28, P = 0.098, N = 34, for 2010 and 2011, respectively). Overall, flower size and number differed between populations and plants within populations, but only flower number showed temporal variation (Table 2). Individual mean flower size ranged from 7.2 ± 0.4 to 10.5 ± 0.5 cm and was significantly larger in Tres Cantos (Fig. 1A). Over time, variation in flower size was opposite-signed between populations (Population Year significant, Table 2). In Tres Cantos, flower size decreased by up to 2% and in Canencia increased by 1.5% from 2010 to 2011 (Fig. 1A). Individual mean flower number showed a large variation (range: 5.8 ± ± 36.4) and also was significantly higher in Tres Cantos across the two study years (Fig. 1B). Between years, flower number increased significantly, especially in Canencia (Population Year significant, Table 2; see also Fig. 1B).

7 546 A. L. TEIXIDO and F. VALLADARES Relative to components of female fitness, fruit set averaged 80% (95% CI: 65.48, 94.52%; range: %) and seed number averaged 855 ± 233 (range: ). Fruit set differed significantly between populations and plants within populations but remained invariable over time (Table 3). Fruit set was also constantly higher in Canencia and this difference increased significantly between years (Population Year significant, Table 3; see also Fig. 2A). Seed number differed significantly between populations and plants within populations (Table 3). In contrast to fruit set, seed number was consistently higher in Tres Figure 1. A, flower size ± SD (cm) and B, flower number ± SD at each population for each year of study in Cistus ladanifer. Interaction Population Year was significant for both traits (Table 1). Figure 2. A, fruit set (% with 95% CI) and B, seed number ± SD at each population for each year of study in Cistus ladanifer. Interaction Population Year was only significant for fruit set (Table 2). Table 3. Generalized linear mixed models for differences in fruit set and seed number of C. ladanifer between populations, years, plants within populations and the interaction population year Fruit set Seed number Effect d.f. Estimate ± SE Test value P d.f. Estimate ± SE Test value P Random Plant (Population) 71, ± < , ± < Fixed Population 1, ± < , ± Year 1, ± , ± Population Year 1, ± < , ± Plant (Population), as random factor, was tested with Wald Z-test, and fixed factors with Type III F-tests. Significant P values are marked in bold.

8 SELECTION FOR FLORAL DISPLAY IN CISTUS 547 Cantos than in Canencia (Fig. 2B). Between years, seed number similarly increased in the two study populations (Population Year not significant, Table 3; see also Fig. 2B). Figure 3. Frequency ± SD (%) of each pollinator functional group in Cistus ladanifer between years in (A) Tres Cantos and (B) Canencia. POLLINATORS Six and five functional groups were identified in Tres Cantos and Canencia, respectively (Fig. 3). In Tres Cantos, solitary bees were dominant and accounted for c. 45% out of 268 and 258 visits reported for 2010 and 2011, respectively. With honeybees and flies (Syrphidae, Muscidae and Anthomyiidae), solitary bees accounted for 90% of visits, whereas beetles and wasps accounted for the remaining 10% (Fig. 3A). In Canencia, solitary bees were also abundant, but hover flies were the dominant group. Overall, both groups accounted for most visits over time (Fig. 3B). Muscoid flies and wasps were absent, but bumblebees occurred, although at low frequency (< 5% out of 158 and 257 visits for 2010 and 2011, respectively; Fig. 3B). These differences in pollinator assemblage were significant between populations and years (Per- MANOVA analyses: Pseudo-F 1,52 = 47.39, P < 0.001; Pseudo-F 1,52 = 5.27, P = 0.001, respectively). However, pollinator assemblage varied uniformly between populations over time (Population Year not significant; Pseudo-F 1,52 = 2.09, P = 0.102). Pollinator visitation rates showed spatial temporal variation (range: 0.26 ± ± 0.15; Population Year significant, Table 4; see also Fig. 4). Although they were similar between sites, they increased significantly by c. 19% in 2011 (Table 4; Fig. 4). This difference was mainly due to the large increase (> 38%) in Canencia from 2010 to Table 4. ANCOVA for differences in pollinator visitation rates of C. ladanifer between populations, years, flower size, flower number and all the interactions Effect d.f. MS F P Population Year Flower size < Flower number Population Year Population Flower size Population Flower number Year Flower size Year Flower number Population Year Flower size Population Year Flower number Error Significant P values are marked in bold.

9 548 A. L. TEIXIDO and F. VALLADARES Figure 4. Interaction between population and year in pollinator visitation rates in Cistus ladanifer (Fig. 4). Pollinator visitation rates were also positively correlated with flower size, but were independent of flower number (Table 4). The positive correlation with flower size was higher in Canencia over the study period (Population Flower size significant, Table 4; see also Fig. 5). However, increased pollinator visitation rates did not affect fruit set or seed number (F 1,48 = 2.72, P = for fruit set; F 1,48 = 2.68, P = for seed number). PHENOTYPIC SELECTION The results revealed positive directional selection on flower size through seed number in Canencia in 2010 (Table 5). In Tres Cantos, phenotypic selection tended to decrease flower number through fruit set and seed number during In addition, we detected constant stabilizing selection on flower size through fruit set in this population over the study period (significant negative quadratic terms, Table 5; Fig. 6). DISCUSSION Our study reports spatial temporal variation in patterns of pollinator visits and analysis of phenotypic selection on traits involved in floral attraction. We experimentally verified that natural populations of C. ladanifer can show pollen limitation, at least in terms of fruit set. We also found evidence of a positive relationship between flower size and pollinator visitation rates, but this was not translated into differential fruit set or seed number. Accordingly, we did Figure 5. Interaction between population and flower size in pollinator visitation rates in Cistus ladanifer. not generally detect a significant positive directional phenotypic selection on flower size through the two components of female fitness. Specifically, we found evidence for stabilizing selection through fruit set in Tres Cantos. However, we did detect significant directional selection through seed number for increased flower size in Canencia in 2010, indicating that ecological context dependence (i.e. variation in flower size, pollination environment and/or resource availability among populations) might have the potential to affect patterns of sex-specific selection. We also detected significant negative directional selection on flower number through fruit set and seed number in Tres Cantos in Overall, our study indicates that in natural populations of the large-flowered Mediterranean shrub C. ladanifer variation in flower size is wide, dynamic and only partially dependent on pollination environment. Hence, selection for increasing flower size and number is not always pollinator-mediated through fruit set and seed number. Although the two study populations differ in climatic conditions (at least in mean annual temperature and precipitation), we ignore differences in other possible ecological parameters (e.g. soil moisture and composition, relative humidity, resource availability). Therefore, our study does not provide any statistical power to infer differences in selection patterns because of climatic conditions.

10 SELECTION FOR FLORAL DISPLAY IN CISTUS 549 Table 5. Standardized selection gradients for flower size, flower number and their correlation in C. ladanifer on fruit set and seed number at each population for each year of study Tres Cantos Canencia Fruit set Seed number Fruit set Seed number Year/Trait β γ β γ β γ β γ 2010 Flower size ± ± 0.058* ± ± ± ± ± 0.042** ± Flower number ± ± ± ± ± ± ± ± Flower size ± ± ± ± ± ± ± ± Flower number 2011 Flower size ± ± 0.040* ± ± ± ± ± ± Flower number ± 0.035*** ± ± 0.130* ± ± ± ± ± Flower size ± ± ± ± ± ± ± ± Flower number Linear (β ) and quadratic (γ ) coefficients ± SE are shown. Selection gradients with significant P values are marked in bold. *P < 0.05; **P < 0.01; ***P =

11 550 A. L. TEIXIDO and F. VALLADARES Figure 6. Magnitude and shape of stabilizing selection on flower size in Cistus ladanifer through fruit set in Tres Cantos in (A) 2010 and (B) 2011 (see Table 5). EFFECTS OF FLORAL DISPLAY ON VISITATION RATES AND COMPONENTS OF FEMALE FITNESS The interactions between C. ladanifer and pollinating agents recorded in our study were generalist in nature; the species was visited by an array of pollinators that in fact varied in space and in time. Variety of plant-pollinator systems seems to be a rule for large-flowered Cistus species, in which the relative proportions of bees, beetles and flies are highly variable among species and sites (Bosch, 1992; Talavera et al., 1993, 2001). Specifically, our results contrast with those of Talavera et al. (1993) who recorded a high proportion of flies (79%) and a low one of bees (4%) in pollinator spectra of a population of C. ladanifer in south-western Spain. Hence, generalization is common in this species and the high variability in pollination environment among populations might entail some adaptation in floral display to local pollinating conditions. Flower size and number can have contrasted or additive effects on visitation rates (Conner & Rush, 1996; Thompson, 2001). Pollinators are attracted by reliable visual signals such as flower number due to their greater resource content (reviewed by Jones, 2001). However, conversely to flower size, a higher flower number did not involve a higher visitation rate or a significant increase in fruit set and seed number in our study system, as reported in other works (Bell, 1985; Brody & Mitchell, 1997; Thompson, 2001; Brys et al., 2004; Liao et al., 2009; Brys & Jacquemyn, 2010; but see Mitchell, 1994; Grindeland, Sletvold & Ims, 2005). Our data suggest that many-flowered plants do not have a higher percentage of their flowers visited and the balance between size and number leads towards larger and less numerous flowers. Thus, flower size seems to be more important for individual fitness in our study system, as generally reported for generalist plants (reviewed by Willmer, 2011). We reported that increased pollinator visitation rates did not enhance fruit set or seed number, as reported in other species (Young & Stanton, 1990; Kudoh & Whigham, 1998; Talavera et al., 2001; but see Johnson, Delph & Elderkin, 1995). The absence of any correlation between pollinator visitation rates and fruit set or seed number could be due to the generally low levels of pollen limitation experienced in Cistus (Bosch, 1992; Herrera, 1992; Talavera et al., 1993, 2001; Arista & Ortiz, 2007). However, we did find pollen limitation in terms of fruit set, so natural populations of C. ladanifer, as we reported here, may show different levels of pollen availability between populations and years according to pollinator environment. Additionally, feasible variability in effectiveness of individual pollinator functional groups could account for these levels, as reported for the congeneric C. libanotis L. (Talavera et al., 2001). ASSESSMENT OF POLLINATOR-MEDIATED PHENOTYPIC SELECTION The patterns of selection reported in our study help to disentangle some of the evolutionary forces acting on flower size and to understand its dynamic variation in this species. However, we are aware that any interpretation of patterns in relation to habitat (i.e. Tres Cantos and Canencia, or drier and wetter sites, respectively) should be clearly labelled as speculation requiring further tests. We detected positive directional selection in Canencia in 2010 for flower size through seed number. We did not detect pollen limitation in terms of seed number, but our results

12 SELECTION FOR FLORAL DISPLAY IN CISTUS 551 indicate that an effect of pollinators on fruit set and/or seed number in C. ladanifer in our study populations is plausible and that the correlation between visitation rates and flower size may limit seed production in smallest-flowered plants. Our contrasting results, showing positive directional selection but absence of pollen limitation, may be due to the difference in sample sizes between the experimental and the observational study. Although our study supports pollinator-mediated selection on flower size through seed number associated with pollen limitation in agreement with previous studies (Totland, 2001; van Kleunen & Ritland, 2004; Gómez, Perfectti & Camacho, 2006; Hodgins & Barrett, 2008; Nattero, Cocucci & Medel, 2010) our findings have to be taken with caution. We also detected negative directional selection for flower number through both fruit set and seed number in Tres Cantos in To our knowledge, this is the first study detecting negative selection on flower number in natural populations. Other researchers have identified selection for increased flower number (Conner & Rush, 1997; Caruso, Peterson & Ridley, 2003; Sletvold, Grindeland & Ågren, 2010), no selection (Nattero et al., 2010) or stabilizing selection (Parachnowitsch & Kessler, 2010). A higher flower number may involve an increase in ovule discounting in hermaphrodite plants with different forms of self-incompatibility, thus decreasing fruit and seed set (de Jong et al., 1992; Duffy & Johnson, 2011). Additionally, the proportion of flowers visited decreases with flower number and cross-pollen is only deposited on the first few flowers (see Brys et al., 2004, and references therein), However, because we found no compelling evidence of any effect of flower number on visitation rates and a subsequent differential fruit set and seed number, this pattern is more difficult to explain. Alternatively, a high flower number could be related to attraction of other visitors such as pre-dispersal fruit predators (Brys et al., 2004) and to significant costs of floral allocation translated into negative effects on reproductive output (Montalvo & Ackerman, 1987; Mitchell, 1994; see also Teixido & Valladares, 2013 for the study species). Apart from these exceptions we generally identified absence of selection through fruit set and seed number on flower size and number. No selection through components of female fitness on traits involved in floral attraction has also been reported in other plant species, particularly under resourcelimited conditions (reviewed by Ashman & Morgan, 2004) and, specifically, in the congeneric C. salviifolius L. (Arista & Ortiz, 2007). We identified stabilizing selection on flower size through fruit set in Tres Cantos during the study years. Several studies of female selection gradients on flower size have found evidence for this form of selection (Herrera, 1993; Wright & Meagher, 2004; Nattero, Sérsic & Cocucci, 2010; Sahli & Conner, 2011). Stabilizing selection on flower size through fruit set cannot be explained based on the patterns of pollinator attraction or pollen limitation detected in our study. Cresswell (1998) suggested several possible mechanisms that might underlie stabilizing selection on floral traits, including a structural match between pollinators and flowers and resource allocation among fitness-related sinks. Generalist plants in a multiple-pollinator environment may have less capacity to respond to directional selection on floral traits (Cresswell, 1998; Johnson & Steiner, 2000; Gómez & Zamora, 2006). In fact, these plants may be under stabilizing selection as mediumsized flowers may result in the optimal values in their pollination environment (reviewed by Kingsolver & Diamond, 2011). In a recent study, Sahli & Conner (2011) identified floral adaptation to multiple pollinators by reporting strong selection on flower size through seed number with only one pollinator but stabilizing selection when pollinators acted altogether in Raphanus raphanistrum L. Additionally, the presence of floral herbivores (florivores) could help to explain the existence of stabilizing selection on fruit set in our study populations. The incidence of florivory was significantly influenced by flower size in C. ladanifer and was > 30% on the largest flowers in Tres Cantos (Teixido et al., 2011). Florivores may reduce fruit production by degrading the attractive properties of flowers for pollinator service or by direct consumption of viable gametes (Schemske & Horvitz, 1988; Krupnick, Weis & Campbell, 1999; Cardel & Koptur, 2010). In this way, florivores can exert negative selective pressure on the same floral traits positively selected for pollinators (Galen, 1999; Irwin, 2006). The most plausible mechanism influencing this form of selection could be based on a combined action of pollen limitation with resource limitation and subsequent floral costs. In this context, small increases in flower size would increase fruit set by reducing pollen limitation until flower size becomes too large and, then, elevated floral costs counteract the benefits of high rates of pollen deposition (Teixido, 2014). From a resource economy perspective, direct allocation to floral structures entails indirect costs (Chapin, 1989), i.e. negative effects on other floral functions such as reproductive output. Therefore, increased allocation to flower size may divert resources away from fruits and reduce fitness on larger flowers (Cresswell, 1998; Teixido, 2014). This process could be far more relevant under Mediterranean hot and dry conditions, under which large flowers are expected to entail

13 552 A. L. TEIXIDO and F. VALLADARES increased maintenance costs (Galen, 2000, 2005; Elle & Hare, 2002). In fact, we have shown that larger flowers in C. ladanifer increase indirect costs of corollas, especially under the most stressful conditions (Teixido & Valladares, 2013). This might also explain the lower fruit set in Tres Cantos, where flowers were larger, but pollinator visitation rates were higher, at least during the first study year (i.e. 2010). Likewise, strong heat and drought reported in 2011 (Table S1) could also explain the reproductive output found for this year in Canencia, when plants had even lower fruit set than in 2010, but pollinator visitation rate was higher. Although we lack data to infer differences in patterns of phenotypic selection based on floral costs, differences in pollinator environment alone do not modulate selection on flower size across populations of C. ladanifer. Overall, pollinator-mediated selection on the largest flowers with a modest sample size seems to be somewhat limited in C. ladanifer through fruit set and seed number. However, selection for increasing flower size through components of female fitness is feasible, as detected in Canencia in 2010, keeping large flowers in this Mediterranean species. Therefore, the patterns of selection detected in our study support the idea that the relative strength and direction of phenotypic selection on flower size through fruit set and seed number is not only pollinatordependent (Galen, 1999; Ashman & Morgan, 2004). Likewise, selection on this trait could occur through male fitness and larger flowers to increase pollen removal and siring success (Arista & Ortiz, 2007). Flower size showed temporal variation in each population in response to phenotypic selection on fruit set or seed number detected in the first year. Hence, total selection exerted on flower size in C. ladanifer may depend strongly on whether plants are more resource or pollen limited in their female reproduction. CONCLUSIONS The occurrence of stabilizing selection and the different patterns of selection on flower size and number through fruit set and seed number demonstrate the importance of assessing non-linear selection gradients and quantifying selection on different traits related to pollinator attractiveness through different female functions across populations and years. We identified a positive relationship between flower size and pollinator visitation rates, which was not translated into a differential fruit set or seed number and, consequently, into a directional selection towards larger flowers. However, we detected a positive selection on flower size through seed number in one population during the first study year. Additionally, we found evidence for stabilizing selection on flower size through fruit set and negative selection on flower number through fruit set and seed number in the other population across time and the second study year, respectively. Therefore, in natural populations of C. ladanifer pollinator attraction alone does not always drive selection on floral display through fruit set or seed number, but rather it is influenced by the ecological context. In fact, evidence of stabilizing selection on flower size through fruit set suggests that the evolution of this trait may be constrained by non-pollinator agents, thus creating conflicting selective pressures on components of female fitness in this large-flowered Mediterranean species. Lastly, we propose that experimental designs that would allow analysis of components of male fitness (e.g. pollen dispersal and/or siring success) that have been unexplored in this work are critical to evaluate realistic estimates of phenotypic selection on flower size and number in our study system. ACKNOWLEDGEMENTS Luis Giménez-Benavides, Brezo Martínez and Nina Sletvold provided valuable help with data analysis. Jakub Těšitel, Sophie Karrenberg, Anna L. Parachnowitsch and Adrián Escudero provided constructive comments on earlier versions of the manuscript. We also thank David Aparicio, Alejandro Aparicio- Valenciano, Elisa Berganzo, Francisco Castellanos, Tatiana Izquierdo, Ana Lázaro-Nogal and Silvia Santamaría for fieldwork support and Rubén Milla for image processing for counting seeds. This study was supported by CONSOLIDER MONTES project (CSD ) of the Spanish Ministerio de Ciencia e Innovación and project 014/2009 of the Spanish Organismo Autónomo de Parques Nacionales. A.L.T. held a PDI fellowship at Rey Juan Carlos University, Spain. REFERENCES Aigner PA Optimality modeling and fitness trade-offs: when should plants become pollinator specialists? Oikos 95: Aigner PA Variation in pollination performance gradients in a Dudleya species complex: can generalization promote floral divergence? Functional Ecology 19: Anderson MJ A new method for non-parametric multivariate analysis of variance. Austral Ecology 26: Aragón CF, Escudero A, Valladares F Stressinduced dynamic adjustments of reproduction differentially affect fitness components of a semi-arid plant. Journal of Ecology 96: Arista M, Ortiz PL Differential gender selection on floral size: an experimental approach using Cistus salviifolius. Journal of Ecology 95: