Ultraviolet and yellow reflectance but not fluorescence is important for visual discrimination of conspecifics by Heliconius erato

Color patches of animals are complex traits composed of multiple components (Grether et al., 2004). The pigment cells known as chromatophores in the skin of fishes, reptiles and amphibians, for example, are color-generating structures composed of distinct pigmentary and structural layers that vary in their ability to reflect light. The feather barbs or integument of birds or the wing scales of butterflies similarly have diverse nano-structure architectures, thin films and pigments, which produce a dazzling array of colors (Prum and Torres, 2003; Vukusic and Sambles, 2003; Shawkey and Hill, 2005; Stavenga et al., 2011, 2014). These pigmentary and structural components of color patches work in tandem to produce signals used in a variety of contexts (e.g. crypsis, mimicry, aposematism and mate choice). Because the biochemical and developmental mechanisms underlying the pigmentary and structural properties of color differ, each of these components may be subject to different selective pressures and hence independent evolutionary trajectories (Grether et al., 2004). Here, we looked specifically at how two components of a butterfly visual display – UV reflectance and human-visible yellow reflectance due to selective filtering by a specific wing pigment – may function as a signal in mate choice and predation. We also looked at what contribution fluorescence makes, if any, to the signal.

Many butterfly species have colorful wing patterns in both the human-visible (400–700 nm) and UV (300–400 nm) ranges (Silberglied and Taylor, 1978; Eguchi and Meyer-Rochow, 1983; Meyer-Rochow, 1991; Rutowski et al., 2005; Briscoe et al., 2010). While the idea that UV coloration – invisible to humans – may serve as a ‘private channel’ of communication has been challenged (Cronin and Bok, 2016; but see Cummings et al., 2003), there is ample evidence that UV signals are important in animal communication (Rutowski, 1977; Johnsen et al., 1998; Smith et al., 2002; Cummings et al., 2003; Robertson and Monteiro, 2005; Kemp, 2008; Obara et al., 2008; Detto and Blackwell, 2009; Painting et al., 2016). However, although many butterflies have UV-visible color patches, in the absence of behavioral evidence, it is unclear whether the UV reflectance functions as a signal or whether it is simply an epi-phenomenon of the scale structure overlaying pigment granules. The same question can of course be applied to the colors produced by the pigments.

Studies of several butterfly groups suggest in fact that for color patches with both UV and visible reflectance, only variation in the UV component of the signal affects mate choice. Pierid butterfly males, Colias eurytheme and Colias philodice, have forewing colors with both UV iridescence due to the structural scattering of light by the scale lamellae (Ghiradella, 1974) and yellow–orange reflectance due to pterin pigments (Watt, 1964). In behavioral experiments, female Colias were shown to use the UV-reflection difference between the two species as a mate and species recognition cue, but not the human-visible color difference (Silberglied and Taylor, 1978). Female Eurema hecuba (Coliadinae: Pieridae) were similarly shown to prefer males with the brightest UV iridescence overlaying a diffuse pigment-based yellow (Kemp, 2007a). Given that many other butterflies have color patches with UV reflectance, and that butterfly color vision systems are astonishingly diverse (Arikawa et al., 2005; Briscoe and Bernard, 2005; Stalleicken et al., 2006; Koshitaka et al., 2008; Sison-Mangus et al., 2008; Chen et al., 2013), it is worthwhile investigating in other species whether it is the UV or the human-visible part of the color patch reflectance spectrum, or both, that is being used for signaling. It is particularly interesting to investigate this question where there has been a phylogenetic transition from using one type of pigmentation to another, as for the yellow wing colors in the passion-vine butterflies of the genus Heliconius (Briscoe et al., 2010; Bybee et al., 2012) (see below).

Heliconius erato has yellow scales on its hindwings that contain the pigment 3-hydroxy-dl-kynurenine (3-OHK) (Tokuyama et al., 1967; Reed et al., 2008). The yellow bars reflect UV light and have a step-like reflectance at longer wavelengths – a rapid rise then a plateau in reflectance in the visible (400–700 nm) range (Fig. 1A,B, yellow lines) (see also Stavenga et al., 2004). Either the UV or the human-visible part of 3-OHK wing reflectance, or both, may serve as a signal for inter- and intra-specific communication. Intriguingly, the appearance of 3-OHK in Heliconius co-occurred with the evolution of the butterflies' duplicated UV opsins, UV1 and UV2 (Briscoe et al., 2010; Yuan et al., 2010; Bybee et al., 2012). In some Heliconius species, UV1 and UV2 are found in both males and females (K. J. McCulloch and A.D.B., unpublished data). In H. erato, UV1 is a female-specific UV receptor with λ_max=355 nm, while UV2 is a violet receptor found in both sexes with λ_max=390 nm (McCulloch et al., 2016).

In addition to the components of the 3-OHK visual signal mentioned above, the yellow wing bars of Heliconius fluoresce under a hand-held blacklight (Movie 1). Fluorescence occurs when short-wavelength light is absorbed and then re-emitted as a longer wavelength, i.e. lower energy, light. Fluorescent pigments are widespread in nature (Vukusic and Hooper, 2005; Lagorio et al., 2015) and are typically identified using spectrally narrow-band light; however, terrestrial illumination has a broad spectrum so it is unclear whether a pigment's fluorescence contributes much to a potential signal under natural conditions. The emission spectra of the 3-OHK pigment overlap with the visible part of the reflectance spectrum of 3-OHK on Heliconius wings (see below) and so would be well suited to being detected by the blue-sensitive receptor of H. erato with λ=470 nm if it did (McCulloch et al., 2016).

Butterflies from the genus Eueides, which is a sister taxon to Heliconius, have mimetic wing patterns strikingly similar to those of some Heliconius species. These two genera co-occur in the same habitats, yet Eueides’ yellow wing pigments lack the step-like reflectance spectrum of 3-OHK (Fig. 1A,B, gray line) (Bybee et al., 2012), and they do not fluoresce (data not shown). The yellow pigments in the two butterflies appear similar to the human eye in natural light, but their spectra differ strongly (Fig. 1A,B, yellow and gray lines). Although modeling of wing colors suggests in principle that Heliconius can distinguish between Heliconius 3-OHK yellow and Eueides yellow (Bybee et al., 2012), it remains unknown whether Heliconius actually do so in nature. Previous work has shown that H. erato prefer chromatic over achromatic signals in the context of mate choice (Fig. S1) (Finkbeiner et al., 2014); but it is unclear whether the visible, the UV or both parts of the reflectance spectrum of 3-OHK and fluorescence contribute to signaling. Prior work has also shown that avian predators will differentially attack achromatic compared with chromatic butterfly paper models (Fig. S1) (Finkbeiner et al., 2014; Dell'Aglio et al., 2016), but it is unknown whether avian predators will differentially attack butterfly paper models that vary in yellow coloration resembling the differences between Heliconius and Eueides yellow. While Heliconius wing color patterns warn avian predators of their toxicity (Benson, 1972; Chai, 1986), 3-OHK may further serve as a conspecific signal especially in courtship (Bybee et al., 2012; Llaurens et al., 2014). Demonstrating that Heliconius species can in fact discriminate 3-OHK yellow from other yellows in nature is an important step in elucidating the adaptive significance of 3-OHK pigmentation.

To further investigate the contribution of 3-OHK to H. erato signaling, we carried out two sets of experiments. The first set of experiments tested responses of both male and female H. erato to four types of colored models with spectra that were intended to approximate those of either Heliconius species or their mimics, such as Eueides. The first pair of spectra, which are designated Y+ or Y−, resemble 3-OHK (Heliconius) yellow or Eueides yellow, respectively; the second set of reflectance spectra have identical yellow and red coloration in the visible range, but UV reflectance is either present (UV+) or absent (UV−). The second, complementary set of experiments tested the hypothesis that predatory birds will not differentially attack 3-OHK yellow from other yellows when presented with model butterflies as a result of the aposematic function of yellow in general.

Together, these experiments substantiate and elaborate our understanding of the function of 3-OHK yellow and UV coloration. We show also that fluorescence – although clearly visible in laboratory conditions, but with illumination restricted to the UV excitation wavelengths – is not likely to have any impact under the broadband and relatively low UV illumination found in nature.

Four paper model types of the Heliconius erato petiverana Doubleday 1847 butterfly were made as described in Finkbeiner et al. (2012) with their colors modified as follows: with (Y+) and without (Y−) 3-OHK yellow, and with (UV+) and without (UV−) ultraviolet reflectance. The Y+ treatment had 3-OHK on the yellow portion of the wing (0.010 and 0.015 mg μl⁻¹ 3-OHK in methanol applied to the ventral and dorsal sides, respectively). This provided the models with the same pigment as found in the butterfly yellow scales (Fig. 1A,B, orange lines). The yellow portion of the non-3-OHK yellow models (Y−) was covered with yellow Manila paper (Creatology^® Manila Drawing Paper, item no. 410590). Manila paper has a reflectance spectrum that resembles non-3-OHK yellow reflectance from the sister genus to Heliconius, Eueides, which is a Heliconius mimic (Bybee et al., 2012) (Fig. 1A,B, gray and black lines). A thin film UV filter (Edmund Optics, item no. 39-426) was placed over the Manila paper to create a closer match to Eueides yellow pigment. As a control, Mylar film was added to the yellow portions of models with 3-OHK for the Y+ treatment. Mylar film resembles the UV filter but acts as a neutral-density filter. The red portions of the wings were identical in Y+ and Y− treatments.

For the UV+ models, an odorless UV-reflective yellow paint (Fish Vision™) was added to the dorsal and ventral yellow band of the model wings to provide UV reflectance (Fig. 1A,B, purple line), and the red portions of the wings were printed as described in Finkbeiner et al. (2014). For UV models, a thin film UV filter was placed over both the yellow and red/pink UV-reflective portions on the wings. The UV filter prevents any light reflectance up to 400 nm (Fig. 1, blue line). Mylar film was added to the yellow and red/pink portions of models used for the UV+ treatment to function as a control.

Reflectance spectra of the paper models and individual Heliconius erato petiverana (n=15), Eueides isabella, E. surdus, E. thales (n=3 per species) and E. heliconoides (n=2) butterfly wings were measured by first aligning each measured wing in the same orientation as shown in appendix B of Bybee et al. (2012). If the viewer was looking directly from above at the oriented wings, the fixed probe holder (Ocean Optics RPH-1) was placed horizontally on top of the wing such that the axis of the illuminating and detecting bifurcating fiber (Ocean Optics R400-7-UV/VIS) was at an elevation of 45 deg to the plane of the wing and pointed left with respect to the body axis. Illumination was by a DH-2000 deuterium–halogen lamp, and reflectance spectra were measured with an Ocean Optics USB2000 spectrometer. A spectralon white standard (Ocean Optics WS-1) was used to calibrate the spectrometer. For the irradiance spectra measurements, the USB2000 spectrometer, a calibrated tungsten light source (Ocean Optics LS-1-CAL), a 100 or 400 µm diameter fiber (Ocean Optics P100- or P400-2-UV-Vis) and cosine corrector (Ocean Optics CC-3-UV), which produces vector irradiance measures, were used (Cronin et al., 2014). Five irradiance spectra measurements of down-dwelling light were taken and averaged per site.

For the mate-choice experiments, the von Kries-tranformed quantum catches for stimuli (Kelber et al., 2003) were first calculated for H. erato males and females separately using high light intensity and sunny cage irradiance spectra. Pairwise discriminabilities between artificial models and natural wing reflectance spectra were determined using a trichromatic vision model for H. erato males and tetrachromatic vision models for H. erato females (Vorobyev and Osorio, 1998). Parameters for the butterfly visual models were as follows: Weber fraction=0.05 (Koshitaka et al., 2008); photoreceptor peak sensitivity, λ_max, of 355 nm (female only), 390 nm, 470 nm and 555 nm; and relative abundance of photoreceptors, violet-sensitive (VS)=0.13, B=0.2, G=1 (male) or UV=0.09, VS=0.07, B=0.17, G=1 (female) (McCulloch et al., 2016). For the predation experiments, von Kries-transformed quantum catches for only ventral wing stimuli (as the butterflies were presented with their wings folded) were calculated using high light intensity and irradiance spectra from two of the four habitats where the models were placed: forest cover and forest edge. (The other two habitats, Pipeline Road and paved road, were found to have normalized spectra that were identical to forest cover.) Discriminabilities between stimuli were determined using tetrachromatic models of bird vision representing two types of avian visual system, the UV-type (blue tit, Cyanistes caeruleus) and violet-type (chicken, Gallus gallus) systems (reviewed in Frentiu and Briscoe, 2008). For chicken, we used ocular media of Lind and Kelber (2009) and Toomey et al. (2016) and behaviorally determined parameters of Olsson et al. (2015); namely, a Weber fraction of 0.06 for the L cone, and relative abundance of cones: VS=0.25, S=0.5, M=1, L=1. For the blue tit, we followed the work of Hart et al. (2000) including the effects of blue tit ocular media and used a Weber fraction of 0.05 for the L cone, and relative abundance of cones: UV=0.37, S=0.7, M=0.99, L=1.

To test whether Heliconius 3-OHK yellow and UV serve as visual signals for conspecifics, mate preference experiments were carried out using insectary facilities in Gamboa, Panama, from September 2013 to February 2014. Data were collected from 80 wild-caught H. erato petiverana butterflies: 40 males and 40 females. Each butterfly was introduced individually into experimental cages (2 m×2 m×2 m) and presented with one of two pairs of the artificial butterfly models: Y+ versus Y−, or UV+ versus UV−. The models were separated by 1 m and attached to an apparatus used to simulate flight (see Finkbeiner et al., 2014). Movies 2 and 3 show an example of female butterfly trials with Y (Movie 2) and UV (Movie 3) models. Individual butterflies experienced six 5 min trials – three 5 min trials with each of the two pairs. During trials, two variables were recorded: (1) approaches, which consisted of flight unequivocally directed toward the model, and in which the butterfly came within 20 cm of the model, and (2) courtship events, which were classified as sustained hovering or circling behavior around the model (for examples, see videos 2 and 3 in Finkbeiner et al., 2014). Mate preference data were analyzed using a two-way ANOVA in R to examine the effects of model type and sex. Measurements of spectral irradiance (see above) were taken to provide quantitative information about the illumination conditions during the trials (Fig. S2).

Previously, we have shown (Finkbeiner et al., 2014) that avian predators differentially attack achromatic local-form butterfly models compared with chromatic models as well as models that display non-local or color-switched patterns (Fig. S1). Here, we tested whether avian predators would differentially attack local wing color form paper models where UV or yellow is manipulated. Predation experiments were completed in Panama at the Smithsonian Tropical Research Institute Gamboa field station and at selected forest sites in Soberanía National Park (including Pipeline Road), from June to September in 2013. Models were fitted with Plasticine abdomens and tied to branches with thread to represent natural resting postures in the following habitat types: forest cover (15 sites), forest edge (17 sites), Pipeline Road (unpaved road with partial forest cover, 55 sites) and paved road with partial forest cover (13 sites). Examples of foliage cover in each of these habitat types, along with corresponding spectral irradiance measurements, are presented in Fig. S3. For the 3-OHK yellow pigment study, five artificial models of each treatment (Y+ and Y−) were randomly placed in 100 forest sites (Finkbeiner et al., 2014). The sites were separated by ∼250 m to account for avian predator home range (home ranges described in Finkbeiner et al., 2012). There were 500 Y+ models and 500 Y− models for a total of 1000 models. The same methods were used for the UV study, using 500 UV+ models and 500 UV− models in non-overlapping sites from the Y± models, and with the same number of habitat types.

The models remained at their sites for 4 days, and each model was examined for evidence of predation. A butterfly was considered attacked if damage to the abdomen and wings appeared in the form of beak marks and/or large indentations in the abdomen (for examples of attacked models, see Finkbeiner et al., 2012, 2014). The attack response was modeled as a binomial variable (yes or no) dependent upon butterfly model type using a zero-inflated Poisson regression model, including sites as a random effect, in R with the ‘pscl’ package (Zeileis et al., 2008; R Development Core Team, 2010; Jackman, 2011). To examine whether forest light environment affected predator behavior, the same analysis was used to compare predation between model types in four main habitat types: forest cover, forest edge, Pipeline Road (unpaved road with partial forest cover) and paved road with partial forest cover.

To determine the possible contribution of 3-OHK fluorescence to its yellow coloration, we measured the absorption, excitation and emission spectra of 1.5 mg 3-hydroxy-dl-kynurenine (3-OHK; Sigma-Aldrich, catalog no. H1771) in 3 ml methanol (Fisher Chemicals, Optima LC/MS grade, catalog no. A456-1). The resultant solution was diluted to an optical density (OD)=0.3 to get it within the linear range for fluorescence measurement (Dhami et al., 1995). The absorption spectrum of the pigment was measured with a Cary-50 spectrometer (Varian), while the emission and excitation spectra were acquired with a Cary Eclipse fluorimeter (Varian).

We determined the quantum yield of 3-OHK pigment using the comparative method of Williams et al. (1983). The method makes use of a well-characterized standard with a known quantum yield and an absorbance spectrum that is similar to the absorbance spectrum of the sample of interest, in this case 3-OHK. When the reference and the sample of interest have a similar absorbance at the fluorescence excitation wavelength, the amount of photons being absorbed by the reference and test solutions can be assumed to be the same. In this case, a simple ratio of integrated fluorescence is equal to the ratio of the quantum yields of the reference and sample of interest. For greater accuracy, six additional experiments were performed using solutions of various absorbances (ODs). The integrated fluorescence intensity was then plotted against the absorbance of each solution and if this represented a linear function, where no reabsorption occurred, then the measurement was retained; otherwise, the experiment was discarded. The ratio of the slopes of these functions for the reference and sample of interest is equal to the quantum yield ratio. For this particular experiment, Coumarin 500 (Exciton, catalog no. 05000) was chosen as a reference as its emission and absorption spectrum are extremely similar to those of 3-OHK.

The reflectance spectrum measurements of H. erato wings were made using an Ocean Optics USB2000 spectrometer, a UV-cut off filter (Edmund Optics no. 39-426), a 150 W Xenon Arc lamp (which resembles daylight illumination) and a Spectralon white standard.

To test the hypothesis that our Y+ and UV+ paper models resembled real H. erato yellow wing colors, and that our Y− and UV− paper models resembled real Eueides yellow wing colors, we calculated pairwise discriminabilities between real wings and model spectra. We did this for the male and female H. erato visual system, and then for the UV-type and VS-type avian visual systems. We found that for both male and female H. erato eyes, Y+ was an excellent match to H. erato dorsal and ventral yellows, and that Y− and UV− were excellent matches to Eueides dorsal and ventral yellows under high light illumination [Table 1; 66.7–100% of pairwise comparisons fell below 1 just-noticeable difference (JND) and 100% fell below 2 JNDs]. This means that under lower light levels, model spectra would be an even better match to real wings. For the UV+ treatment, only ventral yellow was an excellent match to the H. erato ventral yellow for either H. erato sex. From this, we conclude that the Y+ paper model bears a strong resemblance to real H. erato yellow wings and the Y− paper model bears a strong resemblance to real Eueides yellow wings for H. erato butterflies under the experimental illuminant conditions in which they were tested.

For the UV-type and VS-type avian visual systems, the match between Y+ and UV+ and H. erato ventral yellow and between Y− and UV− Eueides ventral yellow was less good than if these same stimuli were viewed by the butterflies (Table 2). These results indicate that for birds at least, under forest shade or edge illumination, no pair of stimuli fully captured the spectral differences between Heliconius or Eueides yellow wing colors. All pairs of model spectra used in behavioral experiments, however, differed by >1 JND for both birds and butterflies (except for Y+ versus Y− for ventral yellow viewed through the male eye; Table 3). This indicates that for both birds and butterflies, there was sufficient difference between the four model types to potentially elicit a behavioral response in the experiments described below.

To determine how Heliconius yellow and UV affect conspecific recognition, we presented wild-caught H. erato butterflies with artificial butterfly models that had manipulated yellow and UV coloration. Preference toward models was measured in the form of approaches and courtship events. We found a strong model type effect on the number of butterfly approaches toward 3-OHK yellow and UV models. There were significantly more approaches toward Y+ than Y− models (two-way ANOVA, F=16.287, P<0.0001, n=80), and toward UV+ than UV− models (F=10.469, P=0.002, n=80; Fig. 2A, black lines). There was no apparent effect of sex on butterfly approach behavior (F=2.738, P=0.099, n=80 for Y; F=0.049, P=0.952, n=80 for UV), suggesting that males and females approach the models at equal rates. Specific male and female behaviors for all comparisons are illustrated in Fig. S4.

Regarding courtship behavior, we found a strong model type effect where Y+ models were courted much more than Y− models (F=11.731, P=0.0008, n=80; Fig. 2A, red lines). The test for the main effect of sex shows that males court Y models at a significantly higher rate than females (F=9.211, P=0.0002, n=80). However, we found no significant model type effect on the number of courtship events directed toward UV+ and UV− models (F=2.304, P=0.131, n=80). There was also no effect of sex on butterfly courtship behavior toward the UV models (F=0.701, P=0.498, n=80).

Previously, we showed that birds preferentially attack achromatic H. erato models over Y+ chromatic models (Fig. S1) (Finkbeiner et al., 2014), as expected if chromatic cues serve as aposematic signals to avian predators. To test whether birds differentially attack yellow- or UV-manipulated models, predation was measured as the frequency of avian attacks on models in the forest. A total of 110 avian attacks were recorded (over 4 days of predator exposure for 500 models of each type): 27 and 24 attacks on Y+ and Y− models, and 27 and 32 attacks on UV+ and UV− models, respectively. Using a zero-inflated Poisson regression model, we detected no difference in predation between Y+ and Y− models: (z-value=−0.014, P=0.989, n=1000; Fig. 2B), and no difference in predation between UV+ and UV− models: (z-value=−0.536, P=0.592, n=1000; Fig. 2B). A test of whether forest type affected predator behavior found no difference in predation between the model types in forest cover, forest edge, Pipeline Road (unpaved road with partial forest cover) and paved road with partial forest cover (all P>0.10). Although our prior experiments indicated that avian predators differentially attack H. erato paper models that differ in both red and yellow color and pattern (Finkbeiner et al., 2014), the results presented here indicate that avian predators do not differentially attack 3-OHK yellow and other yellow or UV+ and UV− models in field trials.

The absorption spectrum of 3-OHK has a distinctive peak (λ_max) at 380 nm (Fig. 3C), so this wavelength was chosen as the excitation wavelength for fluorescence measurements (10 nm bandwidth). The excitation spectrum of the pigment (Fig. 3D, black line) is in full agreement with absorption measurements demonstrating that 380 nm is the peak excitation wavelength. The fluorescence of the pigment has a broad spectrum, with the peak of the emission around 508 nm (Fig. 3D, green line). Notably, the emission spectrum of 3-OHK overlaps well with the visible portion of Heliconius yellow, suggesting the fluorescence of 3-OHK might in principle contribute to the signal in the visible range.

In order to measure the efficiency of this emission, and hence understand whether the fluorescence might contribute significantly to the signal, we determined the fluorescence quantum yield of 3-OHK. Quantum yield is characterized as the ratio of the number of photons emitted to the number of photons absorbed (Williams et al., 1983; Nad and Pal, 2003). Quantum yield of 3-OHK was obtained by comparing 3-OHK with a standard and well-characterized fluorescent molecule, Coumarin 500 (Dhami et al., 1995), which has similar absorbance and fluorescence peaks to 3-OHK (Fig. S5). We were therefore surprised that the quantum yield of 3-OHK in methanol indicated that the emission is unlikely to be visible under normal illumination (quantum yield=5.1×10⁻⁴). By contrast, the quantum yield of our standard Coumarin 500 was 0.46 (Nad and Pal, 2003) or nearly a thousand times brighter than 3-OHK under similar conditions.

To be certain that these conclusions for 3-OHK in solution would also apply to 3-OHK on real wings in daylight illumination, additional experiments were carried out. Reflectance spectra of H. erato wings with and without a neutral-density filter (Mylar film) or a 400 nm cut-off filter (UV film), using a 150 W xenon arc lamp as a light source (which has a spectrum that resembles daylight illumination), were measured. If 3-OHK fluorescence does not contribute to the Heliconius yellow signal in broad-spectrum light, then measurements of H. erato wing reflectance spectra using a UV-cut off filter, which blocks excitation, should have no effect on the measured spectra in the visible range. That is indeed what we observed (Fig. 4). This series of experiments leads us to conclude that fluorescence does not contribute to the 3-OHK visual signal under broad-spectrum illumination.

Butterflies are astonishingly diverse in their coloration, but the phylogenetic origins of new pigmentary coloration and the evolutionary forces that may have governed the adoption of a new pigment have rarely been investigated. Previously, we showed that 3-OHK pigmentation is a synapomorphy of the genus Heliconius, being an ancestral character for the genus, but absent for sister genera such as Eueides (Briscoe et al., 2010). Here, we have attempted to investigate how 3-OHK pigmentation functions as a signal for H. erato mate choice and defense. Heliconius yellow coloration has a spectrum that includes reflectance maxima in the UV and human-visible range as well as fluorescence (Fig. 1A,B, Fig. 3A–D). Evidence here indicates that both the UV and long-wavelength components of the reflectance spectrum contribute to the visual signal H. erato butterflies use for conspecific recognition, but qualitatively that the UV part may be less important for H. erato courtship than it is for approach behavior. Specifically, the butterflies demonstrated clear preferences under all circumstances for Y+ over Y− (Fig. 2A). It is notable that our discriminability modeling of male and female H. erato vision indicates that for the butterflies at least the Y+ yellows are a good match to real H. erato yellow wing colors and Y− yellows are a good match to real Eueides yellow wing colors (Table 1). These results provide the first empirical evidence that H. erato butterflies prefer 3-OHK yellows to yellows found on the wings of their sister-genera, Eueides, and the first empirical evidence that the evolution of 3-OHK pigmentation in Heliconius may have been driven by sexual selection.

The interpretation of the UV+ and UV− treatments is a little less clear. The UV+ and UV− models had the same long-wavelength reflectance, but differed in the UV. UV+ models were approached by both sexes more frequently than UV− models, but while there was a trend towards preferring UV+ models during mating attempts, this difference was non-significant. This observation is perhaps surprising in view of the idea that at least for birds UV may be a short-range signal (Stevens and Cuthill, 2007). However, our discriminability calculations indicate that the UV+ dorsal yellow model color was not a good match to real H. erato dorsal yellow (Table 1). Neither the long-wavelength nor the UV reflectance for dorsal yellow UV treatments was as similar to natural H. erato dorsal yellow as was the Y+ treatment (Fig. 1A, Table 1). It may be that a closer match to the natural H. erato spectrum – including in the UV – is needed to elicit a stronger courtship response.

Many prior studies of butterfly mate choice have examined the preferences of one sex or the other but not both (Knüttel and Fiedler, 2001; Fordyce et al., 2002; Ellers and Boggs, 2003; Sweeney et al., 2003; Kemp, 2007b). We note that our mate preference results indicate equal responses to models by males and females with respect to approach behavior. This shows that females are ‘active’ during such preference studies (see Movies S2 and S3), and that females and males may share similar preferences for Heliconius yellow and UV in conspecifics. In nature, females may use approach behavior in non-mating-related interactions (Crane, 1955, 1957), such as following between pollen resources or to new roosting locations (Waller and Gilbert, 1982; Finkbeiner, 2014).

Our field study results show that 3-OHK yellow and UV do not alter avian predation rates in themselves, despite studies showing that birds use UV for mate recognition and foraging (Bennett et al., 1996; Siitari et al., 1999; Lyytinen et al., 2004). Recent work has shown that birds have even lower than expected UV sensitivity when looking at stimuli against a UV-poor background (Chavez et al., 2014) and understory-dwelling birds may have lower UV opsin expression than canopy-dwelling birds (Bloch, 2015). Our results resemble those of Lyytinen et al. (2000), who also found no support for UV as an aposematic signal for bird predators. Moreover, we provide experimental evidence that, in natural conditions, the mimicry between Heliconius yellow/UV coloration and non-Heliconius yellow/non-UV coloration in butterflies is successful for deterring birds. Given that we found no indication that Heliconius yellow and UV enhance aposematic signaling toward avian predators, this reinforces the notion that the phylogenetic switch from using other yellow pigments to 3-OHK as a signal on Heliconius wings is significant exclusively in relation to intraspecific communication.

Several studies have concluded that fluorescence is an important component of complex signals in aquatic animals because of the contrast between narrow-band down-welling blue light and long-wavelength fluorescence (Mazel et al., 2004; Gerlach et al., 2014). However, the evidence that fluorescence contributes to signaling in terrestrial animals, where the illumination spectrum is broadband, is much more limited and somewhat mixed. For instance, one laboratory study of fluorescence in budgerigars (Melopsittacus undulatus) suggested that fluorescence contributed to sexual signaling (Arnold et al., 2002) while two other studies of the same species did not (Pearn et al., 2001, 2003). In spiders, lab studies indicate that fluorescence plays a role in male mate choice while UV plays a role in female mate choice (Lim et al., 2007). A paper investigating UV and fluorescence in damselfly signaling (Guillermo-Ferreira et al., 2014) concluded that there might be a possible contribution of fluorescence to the signal; however, important controls necessary to confirm this were absent.

To our knowledge, we report here for the first time that the yellow wing coloration of Heliconius is fluorescent (Fig. 3), although Rawson (1968) mentions anecdotally that H. erato and H. charithonia wings are fluorescent but without specifying that it is the yellow portion of the wings, and without identifying the fluorescent chemical. However, by measuring the absorption, excitation and emission spectra and quantum yield of 3-OHK, together with wing reflectance spectra using daylight-simulating illumination, we found no evidence that 3-OHK fluorescence enhances the reflectance spectrum of Heliconius yellow under broad-band illumination. Although the spectral sensitivity of the H. erato blue receptor (470 nm) is well-suited to detecting 3-OHK fluorescence (McCulloch et al., 2016) we found no evidence that under natural illumination, fluorescence contributes to the 3-OHK signal in the visible range. Our result highlights the importance of quantifying fluorescence using several methods, and specifically under broad-band daylight-simulating illumination, before concluding that it contributes to a signal under terrestrial environments (e.g. Andrews et al., 2007).

In summary, we demonstrate that Heliconius butterflies prefer 3-OHK yellow pigments in the context of conspecific signaling, these pigments have likely been selected for their reflectance properties in the visible range, and that fluorescence does not contribute to the visual signal. These results advance our understanding of the selective forces driving the transition from using other yellow pigments to using 3-OHK pigmentation in the genus Heliconius. We provide strong evidence that 3-OHK pigmentation is maintained because it allows Heliconius species to recognize conspecifics for interspecific communication and sexual selection, whilst retaining the potential benefits of Müllerian mimicry with genera such as Eueides.

We thank Robert Reed, Owen McMillan, Kailen Mooney and Nancy Burley for advice and aid in project design; Ella Yuen, Angela Oh, Nina Chiu, Santiago Meneses and David Carter for field assistance; Adriana Tapia, Oscar Paneso and Elizabeth Evans for insectaries assistance; Brett Seymoure, Kyle McCulloch and Johannes Spaethe for assistance with taking irradiance and reflectance spectra; Darrell Kemp, Gil Smith, Kyle McCulloch, Jennifer Briner, Ana Catalán and Aide Macias-Muñoz for manuscript feedback; Dave Krueger and UCI ImageWorks for printing models; the Smithsonian Tropical Research Institute (STRI) for use of field sites; and La Autoridad Nacional del Ambiente (ANAM, Panama) for research permit approval.