Variation in context‐dependent foraging behavior across pollinators

Abstract Pollinator foraging behavior has direct consequences for plant reproduction and has been implicated in driving floral trait evolution. Exploring the degree to which pollinators exhibit flexibility in foraging behavior will add to a mechanistic understanding of how pollinators can impose selection on plant traits. Although plants have evolved suites of floral traits to attract pollinators, flower color is a particularly important aspect of the floral display. Some pollinators show strong innate color preference, but many pollinators display flexibility in preference due to learning associations between rewards and color, or due to variable perception of color in different environments or plant communities. This study examines the flexibility in flower color preference of two groups of native butterfly pollinators under natural field conditions. We find that pipevine swallowtails (Battus philenor) and skippers (family Hesperiidae), the predominate pollinators of the two native Texas Phlox species, Phlox cuspidata and Phlox drummondii, display distinct patterns of color preferences across different contexts. Pipevine swallowtails exhibit highly flexible color preferences and likely utilize other floral traits to make foraging decisions. In contrast, skippers have consistent color preferences and likely use flower color as a primary cue for foraging. As a result of this variation in color preference flexibility, the two pollinator groups impose concordant selection on flower color in some contexts but discordant selection in other contexts. This variability could have profound implications for how flower traits respond to pollinator‐mediated selection. Our findings suggest that studying dynamics of behavior in natural field conditions is important for understanding plant–pollinator interactions.

While these associations between pollinator groups and flower color are seemingly ubiquitous, there are numerous exceptions (as described in reference Ollerton et al., 2009). Some of these exceptions are due to pollinators exhibiting flexibility in color preference. Understanding the degree to which pollinators are consistent or flexible in their preference patterns is key to determining the strength and direction of selection pollinators impose on plant populations.
Flexibility in flower color preference can arise through a variety of mechanisms. Bees, flies, and butterflies can alter their innate color preference by learning an association between nectar or pollen reward and a trait such as color (Goulson, Cruise, Sparrow, & Harris, 2007;Gumbert, 2000;Raine & Chittka, 2008;Weiss, 1997).
For example, while bumblebees often display an innate preference for blue flowers, they can readily learn to associate a reward with a novel flower color (Gumbert, 2000;Raine & Chittka, 2007).
Although rarely studied in the field, these laboratory studies suggest that learning could explain some of the observed variation in flower color preference we see in nature. Plant community characteristics can also contribute to variation in color preference. For example, variation in the degree to which flower color contrasts with the background can drive flexibility in color preference (Osorio & Vorobyev, 2008). In addition, the presence of other morphologically similar flowering species or presence of a dominant pollinator competitor can contribute to context-dependent flower color preferences (Brosi & Briggs, 2013;Fornoff et al., 2016). Finally, some studies have shown that innate preference and learning of a particular trait, such as color, can vary depending on other aspects of the complex floral display such as scent, size, and shape. For example, strength and direction of preference for a certain flower color can depend on the presence or absence of scent signals (Knauer & Schiestl, 2016;Leonard & Masek, 2014;Russell, Newman, & Papaj, 2016;Yoshida, Itoh, Ômura, Arikawa, & Kinoshita, 2015). For these reasons, it is likely that pollinators in nature display extensive flexibility in floral color preference across plant species and in different communities; and yet, the extent of this flexibility is largely unknown.
The flexibility of pollinator color preference can have important implications for the evolution of floral traits. Pollinator preference leads to increased floral visitation and thus selection for the preferred flower type (Aldridge & Campbell, 2007;Schemske & Bradshaw, 1999). Despite the recognized importance of pollinator preference on plant trait evolution, we are lacking studies that examine the consistency of these pollinator behaviors under field conditions. Furthermore, little attention has been given to how observed preference in one context might vary in other contexts.
Understanding flexibility in pollinator behavior can provide insights into the stability of selection on floral traits and the reliability of pollination services across changing environments.
Wild lepidopterans are well suited to investigating flexibility in color preference in a natural field setting because they are important but understudied pollinators (Rader et al., 2016). They exhibit variation in visual systems across families and even species, which could translate to differential selection pressures on flower color (and other traits) within a given community of co-occurring pollinators and plants (Briscoe, 2008;Stavenga & Arikawa, 2006). In addition, lepidopterans have been shown to display innate color preferences and yet can alter preference through learning (Blackiston, Briscoe, & Weiss, 2011;Kandori, Hirao, Matsunaga, & Kurosaki, 2009). Finally, while largely unexplored, there is some evidence that butterflies can have flexible color preference depending on the environmental context of the display (Kinoshita, Shimada, & Arikawa, 1999). Despite the evidence that butterflies can be flexible in their color preference, very few studies have explored the extent to which they are flexible in their color preference in natural systems.
We investigate the flexibility of flower color preference in two groups of butterfly pollinators that co-occur in natural communities. In particular, we examine whether color preference changes depending on the plant species present. We observed the foraging behavior of pipevine swallowtails (Battus philenor; hereafter pipevine swallowtails, Figure 1a) and a variety of skipper species (family Hesperiidae; hereafter skippers, Figure 1a of their range, but P. drummondii has three additional flower color morphologies-light-red, dark-red, and dark-blue ( Figure 1b). With this system we can ask how preference for light-blue flower color changes depending on which plant species is present. We investigate color preference in replicate contrasts between light-blue and the three other possible P. drummondii flower colors, which allows us to decouple flower color from plant species identity and explore the flexibility of color preference in two co-occurring groups of butterfly pollinators.
In this study, we ask specifically: (a) Is pollinator color preference flexible depending on floral context (i.e., plant species identity or the identity of co-occurring and morphologically similar plant species)?

| Study system
Phlox is a butterfly-pollinated genus (Levin & Berube, 1972). P. cuspidata, P. drummondii has evolved dark-red flowers (Hopkins & Rausher, 2011. In the geographic region where lightblue and dark-red P. drummondii meet, all four flower colors can be found (Hopkins & Rausher, 2014  Our three floral contexts were as follows: (1)

| Pollinator observations
We assessed color preference over 3 years (2010, 2012, and 2015) in the month of May at the Brackenridge Field Laboratory. This site is located in the allopatric range of P. drummondii, and wild populations of light-blue P. drummondii exist nearby. The pollinator observations from 2011 and 2012 are included in previous publications investigating selection on Phlox (Hopkins & Rausher, 2012

| Data analysis
Only plant visits from pipevine swallowtails and skippers were in-  were targeted to test specific a priori hypotheses (Benjamini, 2010).
Implementing a correction does not change the interpretation of our results. Our contrasts were targeted to assess (a) significant differences in flower color preference within a given color type across all three foraging contexts for a given pollinator; and (b) differences in color preference between the two pollinator types in a particular foraging context (see Tables 1 and 2 for pairwise comparisons).
Due to the complicated nature of displaying significant results from a three-way interaction, we display the data in two separate figures (Figures 3 and 4).

| RE SULTS
Over 3 years, we observed a total of 2,441 visits from 312 pipevine swallowtail butterflies and 962 visits from 308 skipper butterflies foraging on our experimental arrays (Supporting Information Table   S1). The model that best predicted color preference of the two butterflies was the full model that included a three-way interaction between the fixed effects (foraging context, pollinator type, and flower color type). This model revealed a highly significant three-way interaction (see Tables 1 and 2 for pairwise comparisons). For the purpose of this study, we were interested in understanding how color preference and floral context interact to shape flexibility in pollinator color preference. Furthermore, we wanted to know whether the two main pollinators of Phlox show different or similar color preference within each foraging context. As such, we report post hoc tests relevant for answering those specific questions below. in the array (CC). We observed qualitatively similar patterns of color preference flexibility for the dark-red and dark-blue color arrays as well (see Figure 3 and Table 1 for contrast results).

| Context-dependent preference
In contrast, skippers do not exhibit significant flexibility in color preference and are generally consistent with color preference regardless of the floral context ( Figure 3, Notes. See Figure 3 for details on which color was preferred in each context. Bolded text in columns indicate significant differences in color preferences between contexts for a given pollinator group.
between light-red and light-blue, skippers display no color preference regardless of the Phlox species present contexts. Skippers exhibit a strong preference for light-blue flowers over dark-red flowers in all three floral contexts, regardless of the flower species identity.
Skippers in the dark-blue arrays exhibit preference for light-blue flowers in two of the three contexts and weak to no preference when light-blue P. cuspidata is paired with dark-blue P. drummondii (SI).

| Pollinator contrasts
Second, we were interested in determining whether the two primary

| D ISCUSS I ON
We found that two groups of generalist pollinators, pipevine swallowtails and skippers, vary in the consistency of their color preference while foraging in a natural field experiment. This variation in flexibility across the pollinators means that pollinator-driven selection on flower color is inconsistent across floral contexts. Our results are based on observing wild butterflies, which have unknown foraging experience, foraging on arrays of native plants in their natural habitat. This experiment was performed using two Phlox wildflower species that depend on these pollinators for as much as 95% of their Notes. See Figure 4 for more details about which color was preferred in each context. Bolded text indicates significant differences in color preferences between swallowtail and skippers within a given context. TA B L E 2 Results from post hoc pairwise comparisons testing whether two pollinator groups differ in their color preference across foraging contexts using generalized linear mixed-effects models with binomial errors F I G U R E 3 Context-dependent flower color preferences vary by pollinator species: Mean proportion of visits to light-blue flowers versus other color flowers across the three contexts (see Figure 2 for full description of floral contexts) for pipevine swallowtails (in black) and skipper butterflies (in gray). 95% bootstrap CIs are plotted around the mean. Letters indicate significant differences in color preferences between the three contexts for each butterfly group and color type. Model results from the contrasts comparing the preference across contexts are displayed in Table 1 0 Previous studies in this system demonstrated that flower color variation across the range of P. drummondii is maintained by pollinator-mediated selection (Hopkins & Rausher, 2012. Dark-red flower color is favored in populations sympatric with P. cuspidata because pollinator behavior decreases costly hybridization between the two species when they have different flower colors (Hopkins & Rausher, 2012). In addition, pollinator behavior in allopatry favors light-blue flower color and thus maintains the ancestral phenotype in western P. drummondii populations (Hopkins & Rausher, 2014). In this system, understanding the flexibility of pollinator behavior across plant species and community contexts is important to determine the stability of selection on flower color across geographic space and time. Our study suggests that much research is needed to understand whether flexible color preference in pipevine swallowtails leads to spatially or temporally varying selection on flower color. For example, do pipevine swallowtails in the sympatric range actually discriminate against light-blue P. drummondii plants because P. cuspidata is in the community? While P. drummondii and P. cuspidata have similar light-blue flowers, the flowers differ both in size (P. cuspidata has smaller flowers) and nectar amount (P. cuspidata has lower nectar volume and sugar concentration compared to P. drummondii) (R. Hopkins, unpublished data). Variation in traits other than flower color could lead to context-dependent preferences in pipevine swallowtails. This would suggest an additional mechanism through which pollinator-mediated selection acts on flower color.
Much of what we have learned about color preference in butterflies comes from laboratory studies, often explored through the use of artificial flowers (Kelber & Pfaff, 1997;Kinoshita et al., 1999;Weiss, 1997;Weiss & Papaj, 2003). Therefore, despite the wealth of information we have about butterfly color preference, we know little about how these behaviors translate to natural systems. From these laboratory studies, it is evident that many butterflies display innate color preferences as well as learned associations between colors and nectar rewards. For example, pipevine swallowtails have an innate preference for blue flowers over yellow in the laboratory (Weiss, 1997). In our study, we found that while pipevine swallowtails preferred blue flowers of one species (P. drummondii), they were strongly deterred by the blue flowers of another co-occurring species (P. cuspidata) and ultimately displayed preference for blue flowers based on species identity. These results suggest that our understanding of pipevine swallowtail flower color preference from the laboratory does not necessarily translate to behavior we observed in the field and that these butterflies are likely using other cues in addition to color (such as shape or scent) to guide their foraging preferences. In contrast, we found that skippers more consistently base their foraging decisions on flower color, as color preference did not appear to be influenced by floral context. The indiscriminate colorbased preference that skippers exhibit could have important implications for co-occurring plants in the community. Skippers may be more likely than swallowtails to move between plant species with the same flower color, transferring heterospecific pollen in the process. The few field studies that examine butterfly color preference suggest that context-dependent color preference may be common (Clements, 1923;Pohl, Van Wyk, & Campbell, 2011) making the case for future studies that explore color preference in natural contexts.
Pollinator preference for particular floral traits exerts selective pressures on plants. It is therefore of primary interest to understand the extent to which co-occurring pollinators exert either similar or F I G U R E 4 Butterfly groups differ in flower color preference across floral contexts. Mean proportion of visits to light-blue flowers versus other color flowers across the three contexts (see Figure 2 for full description of contexts) for pipevine swallowtails (in black) and skipper butterflies (in gray). 95% bootstrap CIs are plotted around the mean. Asterisks indicate significant differences in color preference between the two pollinator groups within a given context. Model results from the contrasts comparing the two butterfly groups are displayed in Table 2 0 The two butterfly groups in our study showed different degrees of flexibility in their color preference. This variation in color preference can be due to a number of factors including differences in visual systems and/or differential learning abilities. Unlike most other groups of pollinators, visual pigments of butterfly eyes vary across families and even species. This means that butterfly individuals of different species can both collect and perceive spectral information in different ways. It is not surprising that this variation in color perception can lead to differences in innate color preferences, the ability to learn new colors associated with rewards, and the degree to which color preference will be context-dependent and influenced by the environment (Blackiston et al., 2011;Briscoe, 2008). While some butterflies have red visual receptors, it appears that skippers do not, likely leading to passive discrimination against red colored flowers (Briscoe & Chittka, 2001). At present, there are no studies examining the pipevine swallowtail visual system, but closely related species exhibit exceptional long-wavelength visual abilities (Arikawa, 2003;Takemura, Kinoshita, & Arikawa, 2005). Future studies that link flexibility in preference to variation in visual systems will add an invaluable mechanistic understanding of how butterflies can impact selection on plant traits.
Conducting behavioral trials in natural systems is complex and challenging and as with most studies, our design involved some trade-offs. Because this study was part of a larger, project aimed at characterizing pollinator-mediated selection, we collected foraging data on one floral context per year. While this design could lead to differences in abiotic conditions, with the potential to impact the number and species composition of floral visitors, we saw a comparable number of each butterfly group across each year and throughout the sampling period Supplemental Table 1. Furthermore, we expect that if there was a "year" effect in our experiment, we would be unlikely to see such strong and consistent patterns within the two butterfly groups. This study clearly does not represent the breadth of possible plant contexts that a pollinator might encounter in the wild. Rather, it highlights the potential for two co-occurring butterfly groups to vary in their behavioral flexibility, with interesting implications for plant trait evolution. In addition, because we chose not to destructively sample the pollinators visiting our arrays, we were not able to identify the skippers in this study to species. As such, we are not able to rule out the possibility that there could be species-specific preferences that were masked by our data pooling.
Finally, other pollinator behaviors such as constancy are important for plant trait evolution. While we acknowledge that constancy is very important, for simplicity we decided to focus on color preference for this study. Future studies that include constancy will undoubtedly add important insight into how flexible foraging behavior impacts plant trait evolution in the field.
The frequency and pattern of pollinator foraging, as well as the composition of both pollinator and plant communities, can have a direct impact on plant reproductive success and the evolution of plant traits. Many laboratory studies suggest that pollinators display innate and learned color preferences and that these preferences can be flexible, but our study is one of few that explores the flexibility of color preference in the field. Therefore, it remains unclear how results from laboratory studies translate to behavior in natural systems. Our study reveals that two butterfly groups that provide the majority of pollination visitation to two native wildflowers display different flexibility in color preference and, in the case of the pipevine swallowtail, behave in ways that might be difficult to predict from laboratory studies. This study enhances our understanding of whether and how pollinators display flexible foraging preferences in the wild. Future studies that combine descriptions of visual systems with critical behavioral assays in the laboratory and in natural environments will allow us to understand the prevalence and mechanisms underlying flexibility in pollinator foraging behavior.

ACK N OWLED G M ENTS
We thank the Brackenridge field laboratory for field experimental support. We acknowledge thanks to University of Texas at Austin Fellowship (NDSEG). We also thank D. Papaj and three anonymous reviewers for their helpful comments on the manuscript.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
RH and SG designed the study. RH, SG, and HB collected the field data. HB, CS, and RH analyzed and interpreted data. HB and RH wrote the manuscript.

DATA D E P O S I T I O N
Data available from the Dryad Digital Repository: https://doi. org/10.5061/dryad.12t657f.