UV bullseye contrast of Hemerocallis flowers attracts hawkmoths but not swallowtail butterflies

Abstract The color and patterns of animal‐pollinated flowers are known to have effects on pollinator attraction. In this study, the relative importance of flower color and color contrast patterns on pollinator attraction was examined in two pollinator groups, swallowtail butterflies and hawkmoths using two Hemerocallis species; butterfly‐pollinated H. fulva and hawkmoth‐pollinated H. citrina, having reddish and yellowish flowers in human vision, respectively. Flowers of both species have UV bullseye patterns, composed of UV‐absorbing centers and UV‐reflecting peripheries, known to function as a typical nectar guide, but UV reflectance was significantly more intense in the peripheries of H. citrina flowers than in those of H. fulva flowers. Comparison based on the visual systems of butterflies and hawkmoths showed that the color contrast of the bullseye pattern in H. citrina was more intense than that in H. fulva. To evaluate the relative importance of flower color and the color contrast of bullseye pattern on pollinator attraction, we performed a series of observations using experimental arrays consisting of Hemerocallis species and their hybrids. As a result, swallowtail butterflies and crepuscular/nocturnal hawkmoths showed contrasting preferences for flower color and patterns: butterflies preferred H. fulva‐like colored flower whereas the preference of hawkmoths was affected by the color contrast of the bullseye pattern rather than flower color. Both crepuscular and nocturnal hawkmoths consistently preferred flowers with stronger contrast of the UV bullseye pattern, whereas the preference of hawkmoths for flower color was incoherent. Our finding suggests that hawkmoths can use UV‐absorbing/reflecting bullseye patterns for foraging under light‐limited environments and that the intensified bullseye contrast of H. citrina evolved as an adaptation to hawkmoths. Our results also showed the difference of visual systems between pollinators, which may have promoted floral divergence.

The two Hemerocallis species, diurnal H. fulva and nocturnal H. citrina, both have flowers with a UV bullseye pattern. Hemerocallis fulva, a butterfly-pollinated species, has diurnal, reddish or orangecolored flowers in human vision without perceivable scent and H. citrina, a hawkmoth-pollinated species, has nocturnal flowers with yellow color in human vision and a sweet scent in human olfaction.
In an experimental array of Hemerocallis (Hirota et al., 2013), hawkmoths showed neither constant preference for human red or yellow color nor any preference for floral scent intensity, corolla orientation, and stem height. Thus, it is likely that hawkmoths use any other floral trait differing between the two species of Hemerocallis as a foraging cue for detecting the flowers. Between H. fulva and H. citrina, not only human visible flower color but also the floral UV reflectance is different: the intensity of UV reflectance in the peripheral part is stronger in H. citrina than in H. fulva (Hirota et al., 2012, Figures 1 and 2). Furthermore, there is evidence that pollinators use not only overall color but also color contrast as foraging cue (Chittka & Raine, 2006;Schmidt, Schaefer, & Winkler, 2004). This might imply that the color contrast between central and peripheral parts ("bullseye contrast") acts as an efficient advertisement for hawkmoths.
As opposed to the trichromatic color vision of hawkmoths, swallowtail butterflies of Papilio have a tetrachromatic color vision (UV, blue, green, red), which enables them to discriminate differences in color with higher accuracy (Koshitaka, Kinoshita, Vorobyev, & Arikawa, 2008). In addition, swallowtail butterflies can perceive the color of smaller targets than the trichromatic honeybees (Takeuchi, Arikawa, & Kinoshita, 2006). Thus, flower color may be a more reliable foraging cue for butterflies than for hawkmoths. In laboratory condition, naive individuals of swallowtail butterflies P. xuthus preferred more reddish colors when the background was green (Kinoshita, Shimada, & Arikawa, 1999). In the wild, swallowtail butterflies showed a significant preference for human reddish flowers over human yellowish ones while they also preferred weaker scent and taller stems (Hirota et al., 2012(Hirota et al., , 2013. In contrast, little is known about the preference of butterflies for the floral UV bullseye pattern.
Hemerocallis fulva and H. citrina provide an extraordinary opportunity to test independent effects of overall flower color and UV bullseye pattern on the attraction of diurnal and nocturnal pollinators. Artificial F2 hybrids show highly variable floral traits, ranging from the floral traits of H. fulva to those of H. citrina (Hirota et al., 2012(Hirota et al., , 2013Nitta, Yasumoto, & Yahara, 2010 (Hirota et al., 2012). The F1 interspecific hybrids show only weak sterility and can be backcrossed , and natural hybrid populations with various intermediates have been known in some localities (Hasegawa, Yahara, Yasumoto, & Hotta, 2006 (Hirota et al., 2012(Hirota et al., , 2013.

| Mapping floral spectral reflectance into insect color space
The spectral reflectance (
We translated the reflectance spectra into chromatic representations of P. xuthus and D. elpenor using the following equation (Chittka & Kevan, 2005).
where P i is the relative number of quanta absorbed by the photoreceptor class i of P. xuthus (UV, blue, green, and red) or D. elpenor (UV, blue, green), S i is the spectral sensitivity function of receptor i, I S and I B are the spectral reflectance functions of a flower part (peripheral or central) and a green foliage background, respectively, and D 65 (Wyszecki & Stiles, 1982) is the irradiance spectrum of CIE standard. The spectral sensitivity function S i was determined using the reported sensitivity function of photoreceptor classes in the retina of the swallowtail butterfly P. xuthus (Koshitaka et al., 2008, Supporting Information Figure S1a) or the Stavenga-Smits-Hoenders rhodopsin template (Stavenga, Smits, & Hoenders, 1993) with the sensitivity maxima (350 nm, 440 nm, and 525 nm) of crepuscular hawkmoth D. elpenor (Höglund et al., 1973;Schwemer & Paulsen, 1973, Supporting Information Figure S1b). The denominator in the calculation of P i standardizes the numerator by a green foliage background, considering that receptor sensitivity depends on a predominant background (i.e., green foliage) (Chittka & Kevan, 2005). This standardization by a green background is known to be less variable under the change of illumination spectrum from daytime to night (Johnsen et al., 2006). Thus, we only used the single illumination spectrum.
For butterflies with tetrachromatic vision (Koshitaka et al., 2008), flower color can be mapped into a three-dimensional butterfly color space (Ohashi, Makino, & Arikawa, 2015). For this mapping, P i was transformed into the degree of receptor excitation E i describing the physiological input to the brain varying from 0 to 1 and being 0.5 for the green foliage background (Chittka, 1992): Then, x, y, and z coordinates in a three-dimensional butterfly color space were calculated as follows (Ohashi et al., 2015): For hawkmoths with trichromatic vision (Höglund et al., 1973;Schwemer & Paulsen, 1973), flower color is mapped on a Maxwell color triangle, which is a projection of the three-dimensional color space with a plane of equal intensity (Kelber, Balkenius, & Warrant, 2003). The color coordinates q i in the Maxwell color triangle were calculated as.
To quantify the flower color difference contributing to pollinators' preference, we applied a linear discriminant analysis to the flower colors of H. fulva and H. citrina in pollinators' color space. To quantify the difference of color contrast contributing to pollinators' preference, we used the Euclidean distance between coordinates in the color spaces that can be viewed as the chromatic distance (Balkenius & Kelber, 2004;Chittka, 1992;Ohashi et al., 2015). We used the Euclidean distance between two color coordinates of the central and peripheral parts of an individual flower as a quantitative variable of color contrast of the bullseye pattern ("bullseye contrast"), and the Euclidean distance between color coordinates of the peripheral part of an individual flower and that of the green foliage background as a quantitative variable of the color contrast between a peripheral part of a tepal and the background ("background contrast"). Here, we define "flower colour" and "contrast" operationally as a variable measured by the above variables and test its correlation empirically.

| Flower color in human vision, scent intensity, and floral morphology
To compare with our previous studies (Hirota et al., 2012(Hirota et al., , 2013, the flower color in human vision was also assessed by recording the flower color of the central part (not the peripheral part) by a simple matching with the standard color chart (SCC) of the Royal Horticultural Society, London, England. Scent intensity was measured immediately after flower opening by a handheld odour meter (OMX-SR; Shinyei, Japan) (Hirota et al., 2012(Hirota et al., , 2013. The measurements were performed for at least three flowers per plant and then averaged. Two morphological traits significantly affecting pollination success (e.g., pollinator preference, Hirota et al., 2013) were measured before observations commenced: corolla orientation (the angle between a flower's main axis (from the floral base to the tip of pistil) and the horizontal) and stem height (from the ground to the top of an inflorescence).

| Design of experimental arrays and pollinator observations
In the foraging experiments, an experimental array was composed of 36 potted plants of Hemerocallis randomly arranged in a 6 × 6 square with a distance of 50 cm between each pot and placed inside a net cage or outside in the experimental field of the Department of Biology, Kyushu University where swallowtail butterflies (Papilio spp.) and crepuscular hawkmoths (Theretra spp.) were observed in previous studies (Hirota et al., 2012(Hirota et al., , 2013. Additionally, nocturnal hawkmoth, Agrius convolvuli, was common (Hirota personal observation). We randomly selected one flower and cut off all remaining ones just before the observation if a plant had two or more flowers.
We replaced some of the 36 plants with new ones day by day because the longevity of a flower is only half a day, and each individual plant did not flower everyday. The following Experiments 1, 2, and 3 were performed from 20 July-2 August 2010, from 11-29 July 2014, and from 18-29 July 2015, respectively. These dates were around the flowering peak of the two Hemerocallis species.

| Experiment 1: innate preference of swallowtail butterflies
In Experiment 1, we examined the innate preference of swallowtail butterflies for floral traits. A naive individual of Papilio xuthus was released to the experimental array in each observation because P. xuhus is one of the major pollinators of H. fulva in that field (Hirota et al., 2012(Hirota et al., , 2013 Table S1. The naive butterflies on the next day of emergence were used in the experiment after keeping them away from feeding. Only one butterfly was released at a time and allowed to fly freely in the net cage and then caught after 5 min of the latest visitation. One observer watched an experimental array and recorded pollinator visitation sequence. Simultaneously, we recorded pollinator behavior with High-Definition Video Camera Recorder (XL H1; Canon, Japan). The observation was performed from 10:00 to 18:34 that corresponded to the flower-visiting time of swallowtail butterflies (Hirota et al., 2012).

| Experiment 2: preference of crepuscular and nocturnal hawkmoths in the field
In Experiment 2, we examined the preference of crepuscular and  Table S1. Foraging behaviors of pollinators were recorded by an infrared video camera recorder (DVS A10FHDIR, Kenko, Japan) with two LED infrared illuminators (850 nm, IRSK02-BK, Fuloon, China).
The infrared video records enabled us to observe hawkmoth behavior through illumination but we could not identify species due to low resolution and monochrome images of the video records.
We performed video observations each day from 18:30 until 24:00.
The sunset time at Fukuoka for the experimental period was 19:21 -19:32.

| Experiment 3: preference of swallowtail butterflies and crepuscular hawkmoths in the field
In our previous studies, hawkmoths showed a condition-dependent preference for flower color using a H. fulva-biased experimental array: Their preferences for flower color are possibly influenced by diurnal pollinators through the distribution of nectar source (Hirota et al., 2012(Hirota et al., , 2013. In Experiment 2, the influence of diurnal pollinators was excluded by using unvisited flowers. In

| Data analysis and statistics
We defined a trip of pollinator foraging as a process from the arrival of one pollinator at the experimental array to its departure from the array. Bayesian generalized linear mixed-effects models, including If the 95% Bayesian credible interval (CI) for a partial regression coefficient included zero, the corresponding explanatory variable had a non-significant effect and was classified into the no effect group. If the CI did not include zero, the corresponding explanatory variable had a significant effect and was classified into the negative or the positive effect group depending on the sign of the median of the posterior distribution of each regression coefficient. For better convergence in parameter estimation, all explanatory variables were standardized to mean = 0 and SD = 1.
All models were fitted in the R statistical environment ver- variables, to which we added the residuals from the model (Worley & Harder, 1996). Consequently, the data points in the figures cannot be directly comparable among figures. While Worley and Harder (1996) made adjustments using normal mixed models, we here adopted the adjustment using Poisson mixed-effect models (detailed in Supporting Information Methods).

| Difference of flower color and contrast between H. fulva and H. citrina
We quantified flower color variation using the discriminant scores.

| Correlations among floral traits
In F2 hybrids, the correlations among flower color and color con- Among the explanatory variables, the discriminant score of the peripheral part was strongly correlated with the bullseye contrast in hawkmoth color vision. Thus, the discriminant score of the peripheral part was excluded from the statistical models of hawkmoth's preference.  Figure 5a,b). The visitation rate also significantly increased with stem height but decreased with the scent intensity. Contrastingly, the visitation rate was not significantly affected by the bullseye contrast ( Figure 5c).  Figure S4). The visitation rates of both crepuscular and nocturnal hawkmoths (dataset 2c, 2n) significantly increased with the bullseye contrast (Tables 3   and 4; the upper CI was lower than zero), indicating that stronger bullseye contrast was preferred (Supporting Information Figure   S5b,d). Although the visitation rate did not clearly increase with the discriminant score of the central part (Supporting Information Figure S5a,c), the posterior distribution of the partial regression coefficient of the discriminant score was biased for positive in dataset 2c, 2n (but 95% CI included zero). This indicates that H. citrina-like colored flower was marginally preferred by crepuscular and nocturnal hawkmoths. Additionally, the visitation rate of nocturnal hawkmoths significantly decreased with corolla orientation. The visitation rates of butterflies significantly decreased with scent intensity (Table 5). The 95% CI of the partial regression coefficients of the other traits included zero, indicating that swallowtail butterflies did not show any significant preference for these traits including flower colors (discriminant scores) (Supporting Information Figure S6). The hawkmoth visitation rates significantly increased with bullseye contrast (Supporting Information Figure S7b) and decreased with scent intensity (Table 6). However, there was no obvious decrease in visitation rate with flower color (discriminant score) in Supporting Information Figure S7a.  (Figures 2 and 3c,d). Both crepuscular and nocturnal hawkmoths showed a significant preference for higher bullseye contrast whereas the hawkmoths showed weak or no preference for H. citrina-like central color (Tables 2-4, 6).

| Experiment 3: preference of swallowtail butterfly and crepuscular hawkmoth in the field
This finding supports our hypothesis that hawkmoths use a higher bullseye contrast as a foraging cue more consistently than the other floral traits. In contrast, the preferences of swallowtail butterflies were affected not by the bullseye contrast but by the central and peripheral flower color (Table 1).
This is the first demonstration that hawkmoths are attracted by the floral bullseye pattern in the field. Hawkmoths showed consistent preference to stronger bullseye pattern independent from the composition of the experimental array and the influence of diurnal pollinators. Thus, we suggest that individuals with intensified bullseye contrast were advantageous to be pollinated by hawkmoths during the process of evolution from H. fulva-like ancestor to H. citrina. In laboratories, it has been known that hawkmoths can recognize floral patterns. First, Macroglossum stellatarum preferred artificial flowers with the ring (not striped) pattern to those with uniform pattern (Kelber, 2002). Second, hawkmoths probed on the yellow-colored and brighter area of a striped or crossed pattern (Goyret, 2010;Goyret & Kelber, 2012). However, the presence and function of the floral bullseye pattern in natural nocturnal flowers have not been demonstrated probably because our understanding about the visual system of nocturnal pollinators had been limited and we human lack UV perception. Recently, it has been reported that some nocturnal animals have highly sensitive eyes (reviewed in Warrant, 2008;Kelber & Lind, 2010) and can discriminate color under dim light conditions (Kelber, Balkenius, & Warrant, 2002;Roth & Kelber, 2004;Somanathan, Borges, Warrant, & Kelber, 2008).
Thus, we need to clarify how widely color vision is used in nocturnal animals and those studies will deepen our understanding of the evolution of nocturnal flowers.
In the hawkmoth color vision, the bullseye contrast was strongly correlated with the discriminant score of the peripheral part, but not with the discriminant score of the central part. This result implies that the intensity of the bullseye contrast of Hemerocallis is largely determined by the peripheral part having the UV reflectance.
Although many flowers pollinated by nocturnal hawkmoths lack UV reflectance uniformly (Raguso et al., 2003;White et al., 1994), some yellow flowers pollinated by nocturnal hawkmoths have a UV bullseye pattern composed of UV-absorbing and UV-reflecting parts (Hirota et al., 2012;Kawano et al., 1995). This UV bullseye pattern may be disadvantageous in white flowers because UV-absorbing white flowers show more reliable higher contrast against the background of green leaves under the various nocturnal conditions than the other colors (Johnsen et al., 2006). On the other hand, the UV bullseye pattern composed of UV-absorbing and UV-reflecting parts may be advantageous in yellow-flowered species to attract hawkmoths by compensating the relatively less reliable lower contrast of yellow color to the green background. Additionally, the stronger bullseye contrast with intense UV reflectance is possibly advantageous at night. Koski and Ashman (2015) showed the deleterious effect of UV irradiance on pollen grain viability in UV bullseye flowers in the daytime, for the peripheral parts of curving petals can directly reflect UV to anthers. At night, UV irradiance is much weaker than in daytime so that nocturnal flowers could not suffer from reduced pollen viability by floral UV reflection relative to diurnal flowers. Three asterisks indicate the difference is significant (p < 0.001) The preferences of hawkmoths for flower color were incoherent between Experiments 2 and 3: hawkmoths marginally preferred H. citrina-like central flower color in Experiment 2 (Table 2) but they preferred H. fulva-like central flower color in Experiment 3 (Table 6).
In Experiment 2, the absolute value of the partial regression coefficient of the discriminant score was lower than the coefficient of the bullseye contrast, indicating that flower color has a smaller effect on hawkmoth attraction than bullseye contrast. Under our experimental settings, the nectar availability at crepuscule was influenced by the abundance of the diurnal pollinators and the composition of the experimental array. In Experiment 3, H. fulva-like colored flowers were dominant in the experimental array and the visitation rate of butterflies to the array was low. This situation corresponds to Hirota et al. (2013) that showed that hawkmoths preferred human reddish flowers over yellowish flowers using a H. fulva-biased experimental array consisting of unvisited flowers. We suggest that this low visitation rate resulted because H. fulva-like colored flowers kept nectar until crepuscule and hawkmoths learned the association between H. fulva-like color and nectar avalability. It has been documented that hawkmoths can be trained to switch their color preference by learning an association of a certain color with a nectar reward (Balkenius & Kelber, 2006;Goyret, Pfaff, Raguso, & Kelber, 2008;Kelber et al., 2003). In the field, hawkmoth preferences for flower color, which are easily learned, should be largely influenced by the abundance of competitive pollinators, and the distribution of the remaining nectar source. More careful studies are needed to assess the magnitudes of selective pressures on attractive traits by considering the community level interaction and its annual fluctuation.
Available evidence supports that swallowtail butterflies can recognize the bullseye contrasts in two Hemerocallis species. First, using TA B L E 1 Medians, standard deviations, and 95% credible intervals of the posterior distribution of the partial regression coefficients in the Bayesian generalized linear mixed effects model for innate butterfly preference (Experiment 1)   artificial flowers, Kandori and Ohsaki (1998) demonstrated that the bullseye pattern enhanced foraging efficiency and flower constancy of butterflies, Pieris rapae. Second, the wavelength discrimination ability of Papilio is the highest among the animals tested so far (Koshitaka et al., 2008). The bullseye contrast intensities of H. fulva and H. citrina were 0.443 ± 0.016 and 0.677 ± 0.052, respectively. This is significantly larger than 0.03, a criterion for perceptible color discrimination in butterfly vision (Ohashi et al., 2015). Thus, the presence of perceptible bullseye pattern of Hemerocallis is expected to be used by butterflies. However, its intensity did not show any significant effect on butterfly attraction. This discrepancy can be explained by assuming that swallowtail butterflies have a threshold of response and non-response to the bullseye pattern, depending on the contrast intensity.  Figure S2). In Experiment 3, although most butterflies visited flowers of H. fulva, we did not detect the preference of wild butterflies for flower color.
It may be caused by the small visitation rate and the H. fulva-biased experimental array. The background contrast is stronger in H. fulva than in H. citrina (Figure 3a) and negatively correlated with the discriminant score of the peripheral part (Supporting Information Figure S3). Papilio xuthus uses the target-background intensity contrast when landing (Koshitaka, Arikawa, & Kinoshita, 2011 Figure 2). Most insects, such as bees and hawkmoths, lack a red receptor (Lunau & Maier, 1995), but swallowtail butterflies have a red receptor and can perceive longer wavelength (Kinoshita et al., 1999). It is more costly for at least bees to feed on the flowers that reflect only longer wavelength, such as human red, than on other flowers with more conspicuous colors for them, like pink or yellow, in terms of searching time for flowers (Spaethe, Tautz, & Chittka, 2001) although long wavelength light (up to 650 nm) can stimulate a "green" (540 nm) receptor for the majority of bees only if the light is very strong (Chittka & Waser, 1997). This is probably the reason why many bees tend not to feed on red flowers (Rodríguez-Gironés & Santamaría, 2004). In contrast, for butterflies, it is a better strategy to forage on flowers that reflect only longer wavelength which are seldom visited by bees.
Our study has uncovered the different effect of bullseye contrast on the attraction of diurnal and nocturnal pollinators by controlling the effect of background flower color. We revealed striking differences in the responses to flower color and the bullseye contrast between swallowtail butterflies and hawkmoths. This result indicates that the difference of visual systems between pollinators may have promoted floral divergence. There is increasing physiological evidence that pollinators use not only visual cue but also a variety of sensory information to find, feed on, and learn about flowers (e.g., von Arx, Goyret, Davidowitz, & Raguso, 2012;Clarke, Whitney, Sutton, & Robert, 2013). Further field observations based on knowledge about the varieties and differences of pollinator sensory systems will provide profitable clues to understand floral evolution mediated by pollinators.

ACK N OWLED G M ENTS
We are grateful to L. Kawaguchi for helpful comments for the measurement of spectral reflectance and Toshiyuki Satoh and Yudai Nishide for collecting larvae of the butterfly. We also thank the anonymous reviewers for valuable comments on the manuscript.
This work was partially supported by Grant-in-Aid for JSPS Fellows (223799, 17J01165) and Grant-in-Aid for Scientific Research A (24247005) from the Japan Society for Promotion of Science.

AUTH O R S CO NTR I B UTI O N S
SKH, NM, AAY, and TY: conceived and designed the study. SKH and NM: collected the data and analyzed it. SKH: wrote the manuscript, with the contribution of all authors. Bold values indicate that the parameters have significant positive or negative effects on the visitation rate.