Phenotypic selection on flowering phenology and pollination efficiency traits between Primula populations with different pollinator assemblages

Abstract Floral traits have largely been attributed to phenotypic selection in plant–pollinator interactions. However, the strength of this link has rarely been ascertained with real pollinators. We conducted pollinator observations and estimated selection through female fitness on flowering phenology and floral traits between two Primula secundiflora populations. We quantified pollinator‐mediated selection by subtracting estimates of selection gradients of plants receiving supplemental hand pollination from those of plants receiving open pollination. There was net directional selection for an earlier flowering start date at populations where the dominant pollinators were syrphid flies, and flowering phenology was also subjected to stabilized quadratic selection. However, a later flowering start date was significantly selected at populations where the dominant pollinators were legitimate (normal pollination through the corolla tube entrance) and illegitimate bumblebees (abnormal pollination through nectar robbing hole which located at the corolla tube), and flowering phenology was subjected to disruptive quadratic selection. Wider corolla tube entrance diameter was selected at both populations. Furthermore, the strength of net directional selection on flowering start date and corolla tube entrance diameter was stronger at the population where the dominant pollinators were syrphid flies. Pollinator‐mediated selection explained most of the between‐population variations in the net directional selection on flowering phenology and corolla tube entrance diameter. Our results suggested the important influence of pollinator‐mediated selection on floral evolution. Variations in pollinator assemblages not only resulted in variation in the direction of selection but also the strength of selection on floral traits.

Studies in path analysis have suggested that variation in selection on floral shape among populations is linked to variation in the composition of local pollinator assemblages (Gómez, Perfectti, Bosch, & Camacho, 2009). In Aquilegia coerulea, populations visited by Sphinx vashti had longer spurs than populations visited by Hyles lineate (Sphingidae). In addition, flowers in populations of Aquilegia coerulea with a greater percentage of nectar-collecting pollinators were not whiter, larger, or with longer spurs than populations visited by few percentages of nectarcollecting pollinators (Brunet, 2009). Compared with diurnal pollinators in the orchid Gymnadenia conopsea population, only nocturnal pollinators selected for longer spurs and mediated stronger selection on number of flowers (Chapurlat et al., 2015). In recent years, there have been an increasing number of studies that use supplemental hand pollination to quantify pollinator-mediated selection on floral traits (Schiestl & Johnson, 2013;Sletvold, Moritz, & Ågren, 2015;. However, it is rare to experimentally quantify pollinator-mediated selection on floral traits together with supplemental hand pollination and pollinator observations. Using these methods, we can point out the contributions of different pollinators to the direction and strength of selection on floral traits. Pollinator-mediated selection can be calculated by comparing the intensity of selection in open-pollinated and supplemental handpollinated plants (Fishman & Willis, 2008;Sletvold & Ågren, , 2011. If pollinators are exerting selection on floral traits through male and/or female fitness, then selection should be stronger in the naturally pollinated treatment (Bartkowska & Johnston, 2012).
In this study, we experimentally quantified pollinator-mediated selection on flowering phenology, floral display, and pollination efficiency traits through female fitness and also pointed out the dominant pollinators at two Primula secundiflora populations. P. secundiflora has a generalized pollination system and is mostly visited by bumblebees and syrphid flies (Zhu, Jiang, Li, Zhang, & Li, 2015). After initial observations, we found that there were differences in the pollinator assemblages between our two studied populations. Because P. secundiflora is a distylous, self-and intra-incompatible herb, reproductive success of this species is dependent on pollinators. Different pollinator assemblages are likely to result in variations in the pollinator-mediated selection on flowering phenology and floral traits. Here, we tested variation in the net directional selection and pollinator-mediated selection on flowering phenology and floral traits between two P. secundiflora populations.
We specifically ask: (1) whether selection on flowering phenology, floral display, and pollination efficiency traits varies between populations with different pollinator assemblages; (2) whether this variation can be explained by variation in pollinator-mediated selection.

| Study species and sites
Primula secundiflora is a distylous [long style and short anther phenotype (L-morph), short style and long anther phenotype (S-morph)], self-and intramorph-incompatible perennial herb that is widely distributed throughout the alpine regions of southwest China. It produces leaves in a basal rosette and normally has 3-43 flowers in a single umbel. Its flowering period is from May to August, and the fruit-

| Pollinators
In 2016, we conducted pollinator observations at the two populations.
At each population, we set two or three 2 × 2 m plots. There were between 120 and 150 P. secundiflora individuals in each plot. Pollinators were recorded during a series of 30-min sessions from 0830 to 1,830 over three sunny days at each population. We recorded a plant visitor as a pollinator if it touched the sexual organs of the flower during its visit. For each observation, we recorded the types, numbers, and behaviors of pollinators. We used the mean numbers of pollinators/ plot/hour/individual as a proxy in our analysis.

| Field experiment
To quantify pollinator-mediated selection on flowering phenology, floral display, and pollination efficiency traits, we conducted experimental hand pollination at both populations during May-August 2016. Before flowering, plants with flower buds were randomly chosen (including both S-and L-morph plants) and individually tagged in each population. We randomly marked 137 and 120 plants at BGTC and PNP populations, respectively. The study populations were visited twice a week throughout the flowering period, and during each visit, all open flowers on plants in the supplemental hand pollination treatment were pollinated by hand with cross-pollen from other plants which were located at least 10 meters away from the target.
All flowers received supplemental hand pollination at least once. For the pollination of the S-morph flowers, we followed the methods of Zhu et al. (2015) and punctured the corolla tube near the stigma and brushed dehiscing anthers across receptive stigmas through the hole using tweezers. This method did not influence the seed production.

| Statistical analyses
Basically, there were no significant differences (p > .05) in the floral traits and reproductive performance between morphs (L and S) at either population (Table S1). Based on this information, we combined the L-and S-morph data in the statistical analyses in this study. Effects of pollination treatments (control vs. supplemental hand pollination) and populations on flowering phenology, floral traits, and reproductive performance were tested with two-way ANOVA. In order to improve normality of data analysis, all data were log 10 -transformed before the analyses.
Following the methods of Lande and Arnold (1983), we used multiple regression analysis to estimate the direction and strength of net directional selection and pollinator-mediated selection. In these regression models, we used the relative female fitness (individual female fitness/mean female fitness) and the standardized traits (with a mean of 0 and a variance of 1) to be as the response variable and explanatory variables, respectively. In addition, the relative female fitness and standardized traits were estimated separately for each treatment and each population. We estimated selection gradients separately for each treatment and each population. Using the multiple linear regression models, we quantified the net directional selection gradients (β i ). We estimated the nonlinear selection gradients (γ ii ) using the quadratic terms of multiple nonlinear regression models. In order to limit model complexity, we did not include the cross-product terms in the regression models. In order to obtain the real selection pressure, we doubled the nonlinear selection gradient coefficients (γ ii ) from the quadratic regression model (Stinchcombe, Agrawal, Hohenlohe, Arnold, & Blows, 2008). To test multicollinearity in the regression models, we calculated variance inflation factors (VIFs) for the linear terms and quadratic terms. All VIFs were <3.2, indicating no problem of multicollinearity.
We used ANCOVA to determine whether net directional selection gradients varied between populations using the data from the open-pollinated control treatment (the first model). In addition, we used ANCOVA to determine whether pollinator-mediated selection gradients varied between populations using the data from both open-pollinated control and supplemental hand pollination treatments (the second model). In the first model, we used the relative female fitness to be as response variable, and used five standardized traits (flowering start date, plant height, number of flowers, corolla tube length, and corolla tube entrance diameter), population and trait × population interactions to be as the explanatory variables.
In this model, a significant trait × population interaction demonstrated that net directional selection varied between populations.
In the second model, we used the relative female fitness to be as the response variable and used five standardized traits (the same as interaction demonstrated that pollinator-mediated selection varied between populations. Significant three-way interactions were detected in these models; we further examined the effect of pollination treatment on selection gradients for each population and used the trait × pollination interaction term to determine whether pollinatormediated selection on floral traits was significant. To quantify pollinator-mediated selection, we subtracted for each trait the estimated selection gradient for plants receiving supplemental hand pollination (β HP or γ HP ) from the estimate obtained for open-pollinated controls (β C or γ C ), Δβ poll = β C − β HP , or Δγ poll = γ C − γ HP (Chapurlat et al., 2015). Pollinator-mediated linear (Δβ poll ) and quadratic (Δγ poll ) selections on each floral trait were quantified separately for each population. All analyses were performed with R 3.2.3. We used Excel (2007) to generate graphs.

| Pollinator assemblages
We carried out a total of 16.5 hr of observations in each population. At

| Floral traits and phenotypic correlations
Flowering start date, plant height, number of flowers, and corolla tube entrance diameter varied between populations (Tables 1 and S2 (Table S3).

| Reproductive success and pollen limitation
Plants produced more fruits, more viable seeds per fruit, and more female fitness (total viable seeds per individual) of open pollination at the PNP population than those at the BGTC population (Table 1).
Supplemental hand pollination significantly increased fruit production, viable seeds per fruit, and female fitness (Tables 1 and S2 Table 2); however, later flowering start date was significant selected at PNP population (β C = 0.139 ± 0.069; Figure 3a,

| Pollinator-mediated selection
Pollinator mediated significant selection on flowering phenology, number of flowers, and corolla tube entrance diameter ( Table 2).
Much of the between-population variations in net directional selection on these three traits could be explained by the variations in pollinator-mediated selection (Figure 3b;

| DISCUSSION
Ascertaining In this study, variation in the net directional selection on flowering phenology and pollination efficiency (corolla tube entrance diameter) traits between two P. secundiflora populations was experimentally demonstrated. Corresponding to the variation in net directional selection, variation in pollinator-mediated selection on these two traits between populations was also detected. In addition, observed variations in net directional selection on these two traits between populations could be mostly attributed to mediations by different pollinators (bumblebees and syrphid flies).
In generalized pollination systems, pollinator assemblages often vary across the distributional range of plant species, and this can potentially cause variation in selection on flowering phenology (Ehrlén, 2015;Gómez et al., 2009). This hypothesis has been supported in the orchid species, Dactylorhiza lapponica (Sletvold, Grindeland, & Ågren, 2010). In our study, we observed different pollinator assemblages be-   Ågren, & Ehrlén, 2010). In this study, we also tested significant net  Ågren, 2016). Besides studies in one orchid species, researchers also test similar results among 12 other orchid species (Trunschke, Sletvold, & Ågren, 2017). Our results also support this hypothesis on several floral traits. In the present study, there was stronger pollen limitation of P. secundiflora at BGTC population than that at PNP population. Consequently, the strength (the absolute selection gradient) of net directional selection on flowering start date and corolla tube entrance diameter at BGTC population was stronger than that at the PNP population. These results may suggest the functional relationship between these two traits and the degree of pollen limitation. In order to clarify this relationship, manipulative experiments are needed in future studies.
As previous studies have suggested, an increase in the number of flowers is always selected because of the upper limit to seed production and attractiveness to pollinators (Grindeland, Sletvold, & Ims, 2005;Mitchell, Karron, Holmquist, & Bell, 2004;Sandring & Ågren, 2009). Selection on corolla tube or spur length can be attributed to interactions with local pollinators (Huang, Wang, & Sun, 2016), for example, coevolutionary elaboration between corolla tube length of Roscoea purpurea and its local pollinators (Paudel et al., 2015(Paudel et al., , 2016. However, our results support these hypotheses at only one of our studied populations. More flower production was selected at PNP population, and pollinator-mediated selection mostly accounted for this. In contrast, there was no net directional but stabilized selection on number of flowers at the BGTC population. These results may suggest that more flower production is attractive to bumblebees, whereas it is not sensitive to syrphid flies. Furthermore, these may also suggest that flower production of P. secundiflora in the BGTC population is optimal to attract pollinators and ensure reproductive success. Although significant net directional selection for longer corolla tube length was detected at BGTC population, pollinator-mediated selection on this trait was not detected. This may suggest that selection on corolla tube length has been selected by other agents, such as herbivores (Parachnowitsch & Caruso, 2008;Sletvold et al., 2015). Furthermore, flower production (floral display trait) and corolla tube length (floral pollination efficiency trait) may be correlationally selected by dominant pollinators, because of an additive effect of these two traits to reproductive success (Chapurlat et al., 2015).
Field experiments with P. farinosa suggest pollinator-mediated selection for taller plant height (Ågren, Fortunel, & Ehrlén, 2006;Ehrlén, Käck, & Ågren, 2002). However, our results do not support this hypothesis. There was no significant selection on plant height at either population. These results indicate that bumblebees and syrphid flies are not sensitive to this trait at our studied populations. In addition, it is possible that plant height may be selected by other agents, such as resource limitation, herbivores, or vegetation context (Ågren, Hellström, Toräng, & Ehrlén, 2013;Sletvold, Grindeland, & Ågren, 2013;Sletvold, Tye, & Ågren, 2017).  Cocucci, 2006;Hodgins & Barrett, 2008). In order to comprehensively explore plant-pollinator interactions, it is advisable to pay more attention to both male and female functions in future studies.