Eyespots deflect predator attack increasing fitness and promoting the evolution of phenotypic plasticity

Some eyespots are thought to deflect attack away from the vulnerable body, yet there is limited empirical evidence for this function and its adaptive advantage. Here, we demonstrate the conspicuous ventral hindwing eyespots found on Bicyclus anynana butterflies protect against invertebrate predators, specifically praying mantids. Wet season (WS) butterflies with larger, brighter eyespots were easier for mantids to detect, but more difficult to capture compared to dry season (DS) butterflies with small, dull eyespots. Mantids attacked the wing eyespots of WS butterflies more frequently resulting in greater butterfly survival and reproductive success. With a reciprocal eyespot transplant, we demonstrated the fitness benefits of eyespots were independent of butterfly behaviour. Regardless of whether the butterfly was WS or DS, large marginal eyespots pasted on the hindwings increased butterfly survival and successful oviposition during predation encounters. In previous studies, DS B. anynana experienced delayed detection by vertebrate predators, but both forms suffered low survival once detected. Our results suggest predator abundance, identity and phenology may all be important selective forces for B. anynana. Thus, reciprocal selection between invertebrate and vertebrate predators across seasons may contribute to the evolution of the B. anynana polyphenism.

Documenting the deflective function of eyespots, specifically ones on the margins of the animal, has proved difficult. Numerous studies suggest eyespots do not re-direct predator attack to the body margin or increase prey survival (e.g. [3][4][5][6][7][8]). There are a few experiments indirectly suggesting eyespots may be deflective [9,11]. For example, Olofsson et al. [9] demonstrated, under very specific lighting conditions of high ultraviolet (UV) and low visible light intensity, birds attack marginal wing eyespots on dead butterflies. The results of these few experiments beg the questions: can eyespots deflect predator attack of live prey? and does this deflection impact prey survival and reproduction?
Here, we investigated the deflection hypothesis and its impacts on prey survival and reproduction by assessing the behavioural response of an invertebrate predator to a butterfly species which exhibits a seasonal polyphenism in eyespot size (figure 1a). We presented live butterflies (Bicyclus anynana) to hand-reared Chinese mantids (Tenodera sinensis) in a series of laboratory experiments. In the first experiment, we observed the response of mantids to butterflies with small eyespots and large eyespots, noting the position of their attacks on the butterfly wing and the percentage of prey escapes. Second, we conducted a microcosm experiment with a single predator and multiple butterflies of one seasonal form evaluating both the length of butterfly survival and number of eggs laid. Finally, we controlled for the effect of butterfly behaviour across seasonal forms by transplanting eyespots from one form to the other and repeating the microcosm experiment.
The seasonal eyespot polyphenism of B. anynana is determined by early developmental temperature conditions [12]. If the immature experiences ambient temperatures above 238C, then the adult has large, bright and conspicuous marginal eyespots on its ventral hindwing. If the immature experiences ambient temperatures below 198C, then the adult has small, dull and cryptic marginal eyespots on its ventral hindwing. This polyphenism is referred to as wet season (WS) and dry season (DS) forms, respectively (figure 1a). Previous research demonstrated the smaller, duller DS eyespots are more advantageous against vertebrate predators relative to WS eyespots, by delaying prey detection and increasing latency to attack [8]. However, once the prey is discovered, both DS and WS have a low probability of escape and survival [7,8,10], calling into question the functional benefits and adaptive importance of the WS form.
Tenodera sinensis, the Chinese mantid, has a widespread distribution, is neurologically and behaviourally similar to other mantid species, and has been used as a model for investigating insect predator behaviour (e.g. [13,14]). Bicyclus anynana inhabits geographical areas with some of the highest global mantid diversity and abundance [15]. Mantids and other invertebrate predators have been observed as active predators in the field with B. anynana, especially during the wet season [16] making mantids good candidates as selective agents for butterflies with large eyespots. Here we investigated whether the large marginal eyespots of B. anynana WS form deflect mantid attack to the wing margin thereby increasing butterfly survival and reproduction when compared with the reduced margin eyespots of the WS form.

Material and methods (a) Experimental animals
Eggs of B. anynana were collected from females from a laboratory colony. Larvae were raised on young maize plants (Zea mays) in a climate room at either 278C (to produce the WS form) or 178C (to produce the DS form) with a 12 L : 12 D cycle and 80% relative humidity. Males and females were separated on the day of pupal eclosion so they were virgins at the onset of the experiments. All adults were kept at 228C and fed a maintenance diet of banana.
Egg cases of T. sinensis were purchased from Carolina Biological Supply Company and reared to adults in individual cages on a successive diet of fruit flies, houseflies and crickets. They were raised to the ultimate instar at 278C (wet season temperatures) with a 12 L : 12 D cycle and 80% relative humidity. Mantids were not exposed to either butterflies or eyespots before any trial.   The arena was designed to ritualize the encounter between prey and predator across trials; it consisted of three components: a rectangular ramp, a square floor and a cylindrical wall (electronic supplementary material, figure S1a). The ramp began outside the cylindrical wall, continued through a port in the wall, and continued upwards inside the wall at an 188 angle measuring a total of 35 cm long and 7 cm wide. Both ramp and floor were constructed of wood and covered in poster board paper, while the cylindrical wall was constructed of paper. Ramp, floor and wall were painted a uniform green (paint reflectance spectra in the electronic supplementary material, figure S1b). The arena was illuminated by two full-spectrum halogen lamps (Solux-Eiko 18003, 50W, 47008K, CRI 91, 368 field of illumination; see irradiance spectra in the electronic supplementary material, figure S1c). Each lamp was positioned 23 cm above the highest point of the ramp and 20 cm from the other lamp to fully illuminate the arena. A single butterfly was starved for 24 h then placed at the top of the ramp inside the arena wall. The posterior/anterior axis of the butterfly was perpendicular to the ramp. Butterflies were trained to remain in this position by a small food reward, a small piece of banana placed on a piece of removable parafilm. At the start of each trial, the butterfly acclimated for 5 min before the predator was introduced. After 5 min, a single mantid was placed at the ramp base outside the arena wall, out of view of the butterfly. Mantids are negatively geotatic, walking upwards from a lower position and were previously trained via a cricket food reward to walk directly up the ramp at a steady pace without stopping. Once a mantid demonstrated the desired walking behaviour consistently three times in a row, experimental trials with butterflies began. All mantids learned to walk up the ramp, without straying or stopping, over the course of three to five experiences until they completed the three trial series.
We used a repeated-measures design where each mantid (n ¼ 20) encountered both WS and DS butterflies in random order. Each mantid experienced each butterfly form four times, a new butterfly was used for each trial. We measured latency to orient on prey, latency to attack prey and prey survival after attack. Almost all trials resulted in a mantid attack sequence, but in two trials the butterfly flew away before an attack was initiated (n ¼ 2 out of 160 trials, 1% of trials resulted in no attack). These two butterflies were then removed from the experiment and the mantids were given another trial with a different butterfly to have even numbers of trials across mantids. Data were either square root (days, eggs) or arcsine transformed ( per cent wing damage), evaluated for homoscedasticity and sphericity then analysed using one-way repeated measures ANOVA (R v. 2.13.0) with mantid as a random effect. We present p-values from two-tailed tests with a ¼ 0.05. Means and 95% CI are reported.
We also noted the attack location on the butterfly. Because mantids remove the wings of butterflies before consuming them, we were able to record eyespot damage for both killed and escaped butterflies. WS wing damage data were evaluated with t-tests. We present p-values from two-tailed tests with a ¼ 0.05.

(c) Butterfly polyphenism and predation microcosm experiment
We used a 2 Â 2 factorial design (mantid presence Â butterfly form) to evaluate the effect of seasonal form on butterfly survival and reproduction in the presence of a mantid predator. There were four treatments: without mantid Â DS butterflies, with mantid Â DS butterflies, without mantid Â WS butterflies and with mantid Â WS butterflies. There were 12 trials per treatment (n ¼ 48), each conducted under full-spectrum lights at 228C. Each trial was performed in a microcosm cylindrical cage (50 cm diameter Â 100 cm height) with a full-spectrum light source, adult food source and oviposition substrate (Zea mays). We used 10 virgin butterflies (five males and five females) and 0-1 mantid in each trial. All experimental animals were naive, i.e. did not experience each other before the trial. Trials began at 12.00 and lasted until all the butterflies were consumed. We counted the number of half days until all the B. anynana were consumed by the mantid. We also recorded the amount of wing damage on the remaining wings left by the mantids and the number of eggs laid by the five females in each cage. Data were transformed, evaluated for homoscedasticity and sphericity then analysed using linear models with an interaction (R v. 2.13.0). We present p-values from two-tailed tests with a ¼ 0.05. Means and 95% CI are reported.

(d) Eyespot manipulation and predation microcosm experiment
The developmental polyphenism of B. anynana alters multiple traits other than eyespot size and brightness which may influence predator -prey dynamics [17,18]. WS butterflies may be able to anticipate and escape a predator attack better than DS butterflies, independently of their wing pattern because they are more active than DS at the same ambient temperature and light environment [19]. Additionally, WS and DS butterflies have different eye sizes and ommatidia dimensions, suggesting visual plasticity between forms [20]. We used a 2 Â 2 factorial design (butterfly form Â eyespot form) to evaluate the effect of ventral wing eyespots on butterfly survival and reproduction in the presence of a mantid predator. We transplanted a small strip of the ventral hindwing margin with all the eyespots to each butterfly securing the wing piece with Superglue. The transverse white band of the two seasonal forms was not transferred. There were four treatments, all with mantids: DS butterfly Â DS eyespots, DS butterfly Â WS eyespots, WS butterfly Â DS eyespots and WS butterfly Â WS eyespots. All other experimental and analyses protocols were identical to the polyphenism and predation microcosm experiment above.

Results and discussion
(a) Eyespots on wet season butterflies increase prey detectability, influence attack location and increase survival In the arena experiment, we found WS butterflies were easier for predators to detect, but harder to capture and more likely to escape with eyespot damage (figure 1    (figures 2a,b and 3a,b).
Past research indicated the large eyespots of WS B. anynana were not effective at deflecting vertebrate predator attack [7,8,10]. WS B. anynana were easier for vertebrate predators to detect; however, predators did not re-direct their attacks to the hindwing margin and WS B. anynana did not seem to accrue any fitness benefits for having large eyespots [7,8,10]. In these experiments, only approximately 4% of the numerous encounters between B. anynana and various vertebrate predators resulted in WS escape [7,8,10]. The discrepancy between vertebrate and invertebrate predator response to eyespots may be related to differences in the visual and nervous systems of the various experimental predators, or to the varying distances and angles from which distinct predators initiate their attack. Whatever the causes, mantids are misled by eyespot patterns more readily than vertebrate predators and alter their attack behaviour in ways that benefit survival and reproduction in WS B. anynana.
Across butterflies, eyespots have limited deflective effect on vertebrate predation under very specific light conditions. Low light conditions with accentuated UV promoted more avian attack to the wing margins of Lopinga achine [9]; however, these same butterflies suffered increased attacks under normal daylight conditions. Low light conditions are thought to increase the UV reflection of the white eyespot centre making the marginal eyespots more conspicuous to avian predators [9]. These findings are not transferable to mantids since mantids have monochromatic vision and are unable to see in the UV range [21], but the low light conditions may be relevant for other invertebrate predators, such as wasps, robber flies and spiders, which perceive signals in the UV range. More research is needed to understand the importance of predator identity, light conditions and their interaction on butterfly reproduction and survival.
Field surveys and mark recapture studies suggest butterfly and moth eyespot patterns deflect predator attack and are under selection in the wild. Individuals with more or larger eyespots have a higher recapture frequency and more damaged eyespots when recovered [22 -24]. Also, the tensile strength of the wing regions where hindwing eyespots are found is often weaker than other areas of the wing and more easily torn [25,26]. These indirect observations have been attributed to selection by avian predators [22][23][24][25][26], but they are also consistent with pressure by invertebrate predators such as mantids. Mantids make similar damage patterns to beak marks and their attack has been noted for its speed and strength [13,27] (electronic supplementary material, video). Eyespot patterns on lepidopteran wings are probably under selection by a diverse community of both vertebrate and invertebrate predators.
Eyespots are found in a variety of other animals suggesting this colour pattern is of general adaptive importance. Investigations in marine and aquatic environments indicate eyespot patterns on fishes are used for predator intimidation or sexual attraction [21,28,29]. There remains at least some evidence eyespot may serve a deflective function. Fully reproductive butterfly fish are found in wild populations with up to 10% of their posterior body missing where eyespots are normally found [30]. Experiments using model prey found eyespots can attract attack by fish predators [31]. The deflection function may be working in conjunction with intimidation and sexual attraction in these systems. We hope our approach provides a useful experimental

Conclusion
We leveraged a naturally occuring polyphenism in eyespot size to demonstrate larger, brighter eyespots deflected predator attack away from the more vulnerable body thereby increasing prey fitness. These marginal hindwing eyespots of WS B. anynana butterflies were more conspicuous to mantid predators; yet, these eyespots conferred greater survival and reproductive success compared to individuals with reduced, duller eyespots. Our results shed new light on the evolution of seasonal polyphenisms in B. anynana and other butterflies. Since ventral hindwing eyespots are beneficial to butterflies in the presence of invertebrate predators, these increases in reproduction and survival may offset the detection costs incurred against vertebrate predators [8]. Differences in phenology and longevity between invertebrate and vertebrate predators may explain the evolution of seasonal polyphenisms in eyespot size and colour.
Seasonal variation in eyespot size and other protective coloration has been primarily attributed to seasonal variation in avian predator intensity (reviewed in [1,2]). In B. anynana, the WS form is thought to have evolved as a result of relaxed avian predation during the rainy season coupled with increased avian predation in the dry season to produce the DS form [23]. Our results indicate increased invertebrate predation pressure may select for large, bright eyespots during the wet season while vertebrate predation pressure may select for small, dull eyespots in the dry season. This dynamic role of predator identity, their differences in vision, neurobiology and behaviour, in addition to predator abundance, may play an important role in the evolution of eyespot phenotypic plasticity.