Experimental field tests of Batesian mimicry in the swallowtail butterfly Papilio polytes

Abstract The swallowtail butterfly Papilio polytes is known for its striking resemblance in wing pattern to the toxic butterfly Pachliopta aristolochiae and is a focal system for the study of mimicry evolution. Papilio polytes females are polymorphic in wing pattern, with mimetic and nonmimetic forms, while males are monomorphic and nonmimetic. Past work invokes selection for mimicry as the driving force behind wing pattern evolution in P. polytes. However, the mimetic relationship between P. polytes and P. aristolochiae is not well understood. In order to test the mimicry hypothesis, we constructed paper replicas of mimetic and nonmimetic P. polytes and P. aristolochiae, placed them in their natural habitat, and measured bird predation on replicas. In initial trials with stationary replicas and plasticine bodies, overall predation was low and we found no differences in predation between replica types. In later trials with replicas mounted on springs and with live mealworms standing in for the butterfly's body, we found less predation on mimetic P. polytes replicas compared to nonmimetic P. polytes replicas, consistent with the predator avoidance benefits of mimicry. While our results are mixed, they generally lend support to the mimicry hypothesis as well as the idea that behavioral differences between the sexes contributed to the evolution of sexually dimorphic mimicry.

How predators perceive and respond to warning phenotypes are consequential questions for understanding how natural selection shapes mimetic adaptations. Batesian mimicry is known to occur across animals and plants through different sensory modalities, but visual signals are the best known and understood (Cott, 1940). Visual mimicry signals include changes in coloration, body structure, and behavior, and these often function in concert to achieve integrated mimetic phenotypes (Wickler, 1968). Birds are a major predator class across invertebrate and vertebrate Batesian mimicry systems (Cott, 1940;Wickler, 1968) and have become models for studying predator psychology and visual physiology. Depending on the mimicry system, birds can learn to avoid mimetic phenotypes (Endler, 1991;Gittleman & Harvey, 1980;Guilford, 1990;Roper & Redston, 1987), or may evolve innate avoidance if models are harmful enough (Schuler & Hesse, 1985;Smith, 1975Smith, , 1977.
Papilio polytes is a female-limited polymorphic mimic distributed widely throughout Southeast Asia; males have a single nonmimetic phenotype and females have either a male-like nonmimetic form or one of three mimetic forms that each resembles a distinct toxic Pachliopta swallowtail. A rich body of theory and experimental work addressing the evolutionary genetics of mimicry and mate choice preferences has emerged from this system (Clarke & Sheppard, 1972;Fryer, 1914Charlesworth & Charlesworth, 1975Kunte 2009;Kitamura & Imafuku, 2010Low & Monteiro, 2018;Nishikawa et al., 2013;Ohsaki, 1995;Uésugi, 1996) including the discovery of doublesex as the genetic locus controlling mimetic polymorphism in P. polytes (Kunte et al., 2014;Nishikawa et al., 2015). While this work has clarified the molecular underpinnings of P. polytes and Pachliopta mimicry, less has been done experimentally to examine the ecological consequences associated with the mimetic relationship itself. Mimicry in P. polytes was first reported by naturalists who noted the similarity in wing patterns between morphs of P. polytes and two Pachliopta species (Fryer, 1913;Wallace, 1865). Clarke and Sheppard (1972) later found correlations between the ranges of different mimetic morphs and their putative corresponding models across Southeast Asia. Ohsaki (1995) examined beak marks on the wings of wild-caught P. polytes and Pachliopta aristolochiae as a proxy for bird predation. His analysis reported comparable beak mark percentages (28%-29%) on the toxic P. aristolochiae and the mimetic form of P. polytes that matches the P. aristolochiae wing pattern. In contrast, Ohsaki found an elevated percentage of beak marks (53%) on nonmimetic P. polytes, suggesting that nonmimetic P. polytes experienced more predation than mimetic P. polytes females. Uésugi (1996) conducted feeding trials with seven wild-caught birds (Hypsipites amaurotis) in a laboratory setting and found that after an initial training session where birds were fed P. aristolochiae the birds reduced their consumption of mimetic P. polytes, consistent with a learned aversion to the wing pattern. In addition, analyses of flight kinematics suggest that mimicry extends beyond wing patterning to behavioral similarity between P. polytes and P. aristolochiae in wing movements and flight path (Kitamura & Imafuku, 2010. Taken together, these studies are highly suggestive of an adaptive resemblance between mimetic P. polytes and P. aristolochiae resulting in predator avoidance, but there exist no direct tests of Batesian mimicry using natural populations of these species and free-ranging bird predators. Replicas of naturally occurring prey have been used to measure predator-mediated natural selection in diverse taxa, including insects (Lövei & Ferrante, 2017), fish (Caley & Schluter, 2003), frogs (Saporito, Zuercher, Roberts, Gerow, & Donnelly, 2007), salamanders (Kuchta, 2005), lizards (Stuart-Fox, Moussalli, Marshall, & Owens, 2003), snakes (Pfennig, Harcombe, & Pfennig, 2001), turtles (Marchand, Litvaitis, Maier, & DeGraaf, 2002), birds (Ibáñez-Álamo et al., 2015), and mice (Vignieri, Larson, & Hoekstra, 2010). These experiments allow for precise manipulation of artificial models in order to test specific hypotheses about how mimicry phenotypes, or parts thereof, may experience differential predation. The artificial prey method has been implemented in diverse butterfly systems to address the relationship between wing patterning and predation (Dell'Aglio, Stevens, & Jiggins, 2016;Finkbeiner, Briscoe, & Mullen, 2017;Finkbeiner, Briscoe, & Reed, 2012Finkbeiner, Fishman, Osorio, & Briscoe, 2017;Ho, Schachat, Piel, & Monteiro, 2016;Merrill et al., 2012;Seymoure & Aiello, 2015;Wee & Monteiro, 2017). In this study, we applied the artificial prey method to study how female-limited Batesian mimicry operates in wild populations. We constructed replicas of P. polytes morphs and P. aristolochiae, with realistic color patterns and reflectance, and exposed them to natural predators in the field. We directly assayed bird predation rates based on attack marks or the loss of a bait and tested for differential predation among the sexes and morphs of P. polytes in comparison with P. aristolochiae in order to analyze the selective advantage of mimicry.

| MATERIAL S AND ME THODS
We conducted two complementary predation experiments to address different aspects of mimicry in the field. In the first phase, we deployed a large number of stationary artificial prey of four different phenotypes to assay the role of wing patterning alone. Phase I replicas were deployed in an open wing basking position. In the second phase, we used fewer artificial prey types but incorporated biologically meaningful features like butterfly replica movement and a live bait providing body movement and odor to make the trials more realistic. Phase II replicas used a closed wing resting position.

| Paper replica construction
We constructed four artificial prey types: mimetic Papilio polytes female, nonmimetic P. polytes female, P. polytes male (which are always nonmimetic), and Pachliopta aristolochiae ( Figure 1). P. aristolochiae represents the toxic, unpalatable model in this mimicry system whereas the Papilio morphs are nontoxic and palatable. Although P. polytes has multiple mimetic forms across its range, we built our replicas based on the local mimetic morph found in Singapore, where all experimental trials were conducted. Similarly, we based our P. aristolochiae replicas on the local morph that P. polytes mimics. Artificial butterfly replicas were designed following methods described in Finkbeiner et al. (2012), where natural butterfly wings were referenced to create spectrally matched paper wings. We first generated digital images of each replica type's wing pattern in Adobe Illustrator and then printed the wings on low-reflective Whatman qualitative filter paper (No. 1001-917), using an Epson Stylus Pro 4880 printer with UltraChrome K3 ink. We applied Crayola ® crayon (Gel FX yellow) over the white/yellow hindwing patches of the printed wings where reflectance properties were difficult to reproduce with printed colors alone. We cut out the printed wings with a laser cutter and affixed them to black cardstock backings. To increase the replicas' durability in the field we dipped the nonchromatic (black/gray) portions of the replicas in molten paraffin wax. Finally, we attached plasticine abdomens (Newplast ® ) to capture the imprints left by predators. The three Papilio replica types were given black abdomens and the Pachliopta replicas had half black, half red abdomens to emulate the abdomens of real Pachliopta.
To evaluate the similarity between real butterflies and our constructed replicas we measured the reflectance spectra of real butterfly wings and abdomens, and artificial butterfly wings (after the molten wax treatment) and abdomens using an Ocean Optics USB2000 fiber optic spectrometer, with a bifurcating fiber cable (R400-7-UV-vis, Ocean Optics) connected to a deuterium-halogen tungsten lamp (DH-2000, Ocean Optics). We used a white spectralon standard (WS-1-SL, Labsphere) during calibration, and placed the detecting fiber in a probe holder at a 45° angle to the plane of the butterfly wing. We designed butterfly replicas to resemble butterflies basking with wings open, and therefore only utilized the dorsal side.

| Predation tests
All experimental data were collected in Singapore between June and   (Ho et al., 2016) and placed 10 replicas in 2 m intervals at each site. A site was considered to experience active predation if 50% of the replicas were attacked within 2 days. All four new sites (JEG, KTPH, NBVR, UAR) fulfilled this condition and were used in subsequent trials.

| Paper replica construction
We constructed two new artificial prey types to resemble mimetic and nonmimetic P. polytes with their wings closed, showing only their ventral wing surfaces. We generated digital images of these wings using a Leica DMS1000 and Adobe Photoshop CS6. The wing images were printed on Daler-Rowney 95gsm A4 Sketch Paper using a Samsung Laser Jet 500 Colour M551 printer. The printed white patches were colored using a Derwent Studio Chinese White 72 color pencil to more closely match the real specimens. We treated the nonprinted side of the wings with colorless paraffin wax to waterproof them and capture potential beak imprints without altering the printed colors. Reflectance spectra were taken from ventral wing surfaces of real P. polytes and from the printed replicas using an Ocean Optics USB2000 spectrometer and using a light probe placed at 90° to the sample. One live mealworm (Tenebrio molitor larva) was secured to each pair of wings using Faber-Castell Tack-It and these were mounted on wooden sticks using green floral wire (Figure 2d-e). We coiled the mounting wire to enable winddriven movement of the replicas. The base of each wooden stick was coated with a water-based insect repellent to deter unwanted (nonavian) predators.

| Predation tests
At each of the sites (JEG, KTPH, NBVR, UAR) we set out three groups of 10 P. polytes replicas (five mimetic and five nonmimetic). Within each group, alternating mimics and nonmimics were planted in the ground at a height of 25-30 cm, approximately 2 m apart from one another.
Replicas remained at the site for 2 days for each trial. These were inspected daily and predation was recorded when the mealworm was removed or damaged or when bite marks were present on the wings.
We used a GZLM with a binomial distribution, and replica type and site identity as fixed effects to analyze differences in predation.

| Discriminability tests
We calculated discriminability using a bird vision model to verify that our paper replicas accurately resembled their real counterparts through the eyes of avian predators. A number of insectivorous birds inhabit Singapore (Castelletta, Thiollay, & Sodhi, 2005), but the specific predators of P. polytes in this area are unknown. For each replica type, we compared color patches (white, black, and red if present) between real specimens and paper replicas and calculated their similarity in units of just noticeable differences (JNDs) using the receptor-noise model of Vorobyev and Osorio (1998). The comparisons were made using the blue tit (Cyanistes caeruleus) cone sensitivities, which represent the UV-type avian visual system. We followed the work of Hart, Partridge, Cuthill, and Bennett (2000) including the effects of blue tit ocular media and used a Weber fraction = 0.05 and relative abundances of cones (UV = 0.37, S = 0.7, M = 0.99, L = 1). We analyzed replicas from Phase I using low light intensity and Endler's (1993) forest shade irradiance spectra because they were placed in the forest understory. For Phase II replicas we used low light intensity and Endler's (1993) daylight irradiance spectra because these replicas were placed in open areas. Low light intensity conditions were selected for both Phase I and Phase II as a more ecologically relevant prediction: avian predators are most active during morning hours when natural light is less bright, and previous work in the tropics has shown experimental evidence that avian attacks on artificial butterfly models occur almost exclusively during the early morning (Finkbeiner et al., 2012).

| Discriminability tests
The spectral comparisons between color patches of real specimens and paper replicas are presented in Figure 3  F I G U R E 3 Reflectance spectra of real and replica color patches, and results of discriminability modeling of the UV-type avian visual system. Percent reflectance for each color patch is shown from 300 to 700 nm, which is the avian visual range. Avian vision modeling results for each color patch are in units of just noticeable differences (JNDs)  Figure 2a-c), arthropod mandibles, and from one small reptile or mammal. Across the four types, we calculated an overall avian predation rate of 1.9%. Cases where the paper replicas could not be found at the end of the test period were excluded from the data set.
Avian predation results for Phase I are summarized in Table 1.
While we did find a larger number of attacks on nonmimetic female P. polytes compared to mimetic P. polytes, consistent with the expectations of Batesian mimicry, we also found that P. aristolochiae replicas experienced the most attacks and nonmimetic males the fewest attacks. Results from our GZLM showed that replica type was not a significant predictor of attack number, but site identity was a significant predictor (

| Phase II
A total of 120 individual paper replicas (mimetic and nonmimetic P. polytes) were tested among the four sites (15 mimetic and 15 nonmimetic at each site). Predation results for Phase II are summarized in Table 3. We again tested for differential predation between mimetic and nonmimetic types. At each site mimics experienced fewer predation events than nonmimics, and all attacks combined showed approximately 30% less predation on mimics than nonmimics. However, the p-value for replica type was slightly above the statistical significance cutoff of 0.05 (Table 4). Predation was not significantly different between sites in Phase II (Table 4).

| D ISCUSS I ON
We generated realistic artificial prey and tested for mimetic protection from predation in P. polytes. In Phase I, we observed generally low attack rates and did not find significant differences in predation between mimetic and nonmimetic P. polytes. In Phase II, we observed the expected lower levels of predation on mimetic replicas compared to nonmimetic replicas at individual sites, but these were not statistically significant overall and fell slightly above χ 2 = 497.01, df = 1, p < 0.001). Phase I predation rates were also lower than those reported in other butterfly predation studies using similar methods (6.38% in Finkbeiner et al., 2012;54.4% in Ho et al., 2016;51% in Wee & Monteiro, 2017). Replicas in Phase I were fixed to leaves and remained stationary while those in Phase II were mounted on springs that allowed for movement and also incorporated a live bait. It is possible that the immobile conditions of Phase I made the artificial prey difficult for predators to visually detect.
Other possible explanations for differences in predation rates between field sites might include the availability of alternative prey, frequency of models to mimics, predator abundance, and/or predator/prey seasonality .
In Phase II, we observed slightly increased attacks on nonmimetic morphs compared to mimetic morphs at each individual site that just missed the p-value significance cutoff of 0.05, when analyzed using a GZLM. This result suggests a modest benefit to mimetic individuals of this species, and with higher power and a larger sample size, a significant difference in predation is expected between mimetic and nonmimetic individuals. More trials need to be conducted to explore this question. The black reflectance values in Phase II replicas were higher than in Phase I, possibly making the Phase II replicas more detectable to predators and thus more likely to be attacked overall.
Although Phase I replicas more closely resembled their real counterparts than Phase II replicas (Phase I had lower JNDs overall), Phase II replicas elicited predation differences in the direction expected under Batesian mimicry.
Behavioral differences between the sexes have long been hypothesized to drive the evolution of female-limited mimicry. Wallace (1865) proposed that female butterflies experience more predation than their male counterparts when searching for host plants while laden with eggs and while hovering over plants during oviposition.
The evolution of flight mimicry in P. polytes supports this idea; wing beat and flight path are significantly different between mimetic and nonmimetic individuals, and mimetic individuals fly like the toxic Pachliopta aristolochiae model (Kitamura & Imafuku, 2010. Ohsaki (1995) proposed that females disproportionately benefit from evolving mimicry and indeed found that wild-caught nonmimetic P. polytes females had nearly double the proportion of beak marks on their wings compared to P. polytes males (which are always nonmimetic). More recently, Su, Lim, and Kunte (2015) tested the mimetic resemblance of several sexually monomorphic and female-limited Batesian mimics, including P. polytes, using bird vision models. Their results show that females are better mimics than males of sexually monomorphic taxa, and that female-limited mimics are as good as sexually monomorphic mimics, supporting the idea that females disproportionately benefit from mimicry. If the benefit to mimicry derives in part from behavioral differences between the sexes, our Phase I results may reflect baseline predation that P. polytes experience due to visual detection alone and in the absence of any behavior, including flight behavior and smaller movement such as body/leg/antennal movement as well as in the absence of body odor.
The continuum between Batesian and Müllerian mimicry is determined by the relative unpalatability of the taxa involved (Huheey, 1976). In Batesian systems, where the mimic is palatable, both the abundance and unpalatability of the model are expected to strongly influence the mimic's benefit Lindström, Alatalo, & Mappes, 1997). Pachliopta aristolochiae are known to occur at some of the experimental sites, and their relative abundances in comparison with P. polytes across those sites are variable (Jain et al., 2017). The degree of unpalatability of P. aristolochiae to bird predators is not well known and could influence how strongly birds avoid its phenotype (Speed, 2000). Feeding experiments using captive bulbuls (Hypsipetes amaurotis) found that birds experienced a negative physical reaction immediately after ingesting P. aristolochiae, and subsequently decreased their feeding on mimetic P. polytes (Uésugi, 1996). It is interesting that, however, birds in Uésugi's experiment also showed a small decrease in feeding on nonmimetic P. polytes over the course of the feeding trials. This could reflect a generalization of avoidance beyond the mimicry phenotype to, for example, butterflies with black wings (Johki, 1983;Johki, Kon, & Hidaka, 1986;Speed, 2000). It could also imply that P. polytes itself may be somewhat unpalatable. Smetacek (2006) presented freshly caught P. polytes to free-ranging predators in Uttaranchal, India and observed that only 44.4% were eaten by birds in comparison with 77%-100% of palatable control species eaten. If reclassified, this would not be the first case in which a Batesian mimic was actually found to be an unpalatable Müllerian comimic (Ritland & Brower, 1991).
Further experiments are needed that integrate phenotypic and behavioral qualities of Batesian mimics with predator psychology and that assess the importance of model frequency dependence in Batesian systems. Papilio swallowtails are likely to be a key system for these studies, owing to the frequent occurrence of mimicry in this group. With these further studies, the growing literature concerning the genetics of mimicry will also benefit from a better understanding of how natural selection on mimicry proceeds in natural populations.

ACK N OWLED G M ENTS
We thank the Singapore National Parks Board for permission to con-