Our results show that different groups of cells on the developing wing epidermis, which correspond to different aspects of the color pattern on adult female wings, have characteristic sensitivities to changes in temperature during pre-adult development (Figure 3), as well as to changes in ecdysone levels during the pupal stage (Figure 4). We could identify not only which traits are, and are not, responsive to manipulations of the external cue and internal signal, but also identify groups of sensitive traits that display distinct patterns of coordinated responses (Figure 5). Finally, we show that the spatial compartmentalization of hormone sensitivities is not due to the spatial or temporal compartmentalization of the hormone receptor protein (Figure 6).
Response of wing traits to developmental temperature
To assess how different groups of cells on the developing wings respond to external environmental cues, we measured wing patterns of butterflies reared at three temperatures, representing typical wet- and dry-inducing extremes (27°C and 19°C, respectively) and an intermediate temperature (23°C). We then compared phenotypes between temperatures. Figure 3 shows the thermal reaction norms for the 19 target traits in adult females. For the first time, this involved considering separately and simultaneously the distinct color rings (white, black, and gold) of multiple eyespots on different parts (anterior and posterior) of the same wing surface and on different wing surfaces (ventral and dorsal) (Figure 2).
This extensive analysis of wing pattern traits revealed that, in contrast to what had been described, some aspects of the dorsal wing pattern are plastic in relation to developmental temperature (Figure 3A). Previous studies of plasticity on dorsal forewing color pattern had investigated the most posterior eyespot (our trait 2) and found it to be largely non-plastic across seasonal environments [22],[25]. Our results confirm this but, by also analyzing other pattern elements on the same wing surface, show that the lack of temperature-sensitivity is not a property of the whole dorsal wing surface. The more anterior eyespot on the dorsal forewing (trait 1) did increase significantly with temperature (Figure 3A). As expected from previous studies, wing pattern components on the ventral surface of the wings showed clear thermal plasticity (Figure 3B, C, E, F, G; see Additional file 1).
Only the wing pattern element implicated in mate choice does not respond to temperature
Previous work largely focused on ventral wing patterns because this is the surface exposed to predators in butterflies at rest and, thus, the surface under predator-driven natural selection for plasticity [20]. Seasonal variation in ventral wing patterns is associated with seasonal variation in the resting background and to alternative strategies for butterflies to avoid predation. In the cooler dry season, duller brown wing patterns with no striking color elements are cryptic in relation to the resting background of dry brown leaves. In the warmer wet season, more conspicuous color elements along wing margins can function as targets for predator attacks away from the more fragile body [1],[32].
The dorsal patterns, on the other hand, are typically not exposed in the butterfly at rest and presumably not under selection by predators. Instead, those patterns are exposed during courtship and thought to evolve under sexual selection [25],[26],[33]. In particular, some of the UV-reflecting white pupils of dorsal eyespots have been shown to influence mate choice [25],[34]. In our study of female butterflies, the only eyespot that showed no significant response to temperature (Figure 3D; trait 2) was the one that is sexually selected in males [25]. The white center of this eyespot had been found to be plastic in males; being larger and more UV-reflecting in wet season courting individuals [25]. Even though it has been proposed that dry season females do courtship [25], in a case of seasonally-plastic sexual selection, we found that the corresponding trait is not plastic in females (Figure 3D; trait 2w). Instead, a recent study proposed that male choice among potential dry-season mating partners depends on the number of white pupils found on the dorsal surface of the female hindwing [34]. The number of such pupils was shown to vary between females reared at 17°C versus 27°C [34]. In our study, we found that the mean (but not the median) number of white pupils on the ventral surface of the hindwing of non-injected females decreases with increasing temperature, but not significantly so (Figure 3I).
Response of wing traits to hormone manipulations
To examine how different groups of cells on the wings respond to changes in hormone levels, we measured the effect of hormone manipulations during the early pupal stage when the signaling from eyespot organizers and the response of the surrounding cells to the ring-determining morphogen are known to take place [19]. We manipulated the levels of active ecdysone in the hemolymph by injecting female pupae with 20-hydroxyecdysone (20E) [22]-[24] at two developmental time points (Figure 1). For each temperature and injection time point, we then compared adult wings between control-injected and hormone-injected individuals. Figure 4 shows the magnitude and statistical significance of the difference between control and hormone treatments for each of the target traits, injection time points, and rearing temperatures [see also Additional file 2].
Only traits that responded to changes in temperature during development responded to changes in hormone titers during early pupal life. That is, all traits for which differences between control-injected and hormone-injected individuals were significant (that is, any red circles in Figure 4) are traits for which the differences between temperatures for non-injected individuals were also significant (that is, reaction norms marked with stars in Figure 3). However, not all wing pattern traits that responded to temperature were affected by the hormone treatment. We found no significant effect of hormone manipulations for any of the traits in the dorsal wing surface (Figure 4A). In contrast, many traits on the temperature-plastic ventral wing surfaces significantly increased in area in response to hormone injections. In some cases, lack of effect of our hormone injections on temperature-responsive traits can be explained by the fact that trait determination occurred before the hormone treatment. This is the case for the white eyespot centers (traits 4w, 6w in Figure 2) and for hindwing area (trait 10). The establishment of the eyespot organizing centers [35] and most of wing growth [36] are known to take place during larval life, before our hormonal injections were done. However, for other non-responsive traits, notably eyespot color rings, that is not the case (see below).
Only pattern elements on the wing surface exposed to predators respond to changes in pupal ecdysone levels
For all dorsal (traits 1 and 2) and some ventral thermally-responsive color pattern elements (traits 4 and 7) that did not respond to hormone treatment, it seems unlikely that our treatment missed the relevant windows of trait determination. Certainly for eyespot rings, we know that it is during early pupal development that signaling from eyespot centers establishes concentric rings of cells fated to produce different color pigments [30],[31]. The lack of response of those traits to our hormone manipulations could be due to lower sensitivities to hormone titers and due to them requiring hormone concentrations higher than those we produced artificially. This, too, at least alone, seems unlikely because our post-injection hormone levels at 19°C surpassed the control levels at 27°C, a temperature difference for which the traits did change (see below and Figure 1B). The lack of response to hormonal manipulations suggests that thermal plasticity for these traits is not mediated (exclusively) by ecdysone.
It is curious to note that the color traits established in early pupae which we found to be thermally-sensitive but ecdysone-resistant are presumably under no, or weaker, selection by predators. As discussed before, this is the case for color patterns on the dorsal surface of the wing which is not exposed in the butterflies resting against the seasonally color-variable background foliage. Also, unlike other ventral pattern elements, the wing region containing the hormone-unresponsive traits 4 and 7 is typically covered by the hindwing in the resting butterfly. Therefore, these traits too are presumably less exposed to the predators that drive selection for seasonally plastic ventral wing patterns. A weaker selection pressure by natural enemies could explain why these particular traits evolved different levels of plasticity.
Levels and time windows of sensitivity to hormone manipulations
All traits that responded to hormone injection treatment (Figure 4) were larger in hormone-treated relative to control-treated butterflies. The hormone-induced increase in size is consistent with the temperature plasticity: development at warmer temperatures, associated with an earlier increase in natural 20E titers [21]-[24] (see Figure 1A), leads to the production of more conspicuous wing patterns with larger areas of non-background color (Figure 3). By artificially increasing hormone levels at the lower temperatures, we induced the production of the same type of phenotypic effect that higher temperatures have on wing patterns (Figure 4B; see also Additional file 3). The fact that the artificial increase in hormone levels phenocopied the temperature effect confirms a role for ecdysteroids at this early-pupal developmental stage in mediating thermal plasticity in wing patterns.
Strikingly, we detected the strongest responses to hormone manipulations for injections done at the early developmental time point, when the natural levels of pupal ecdysone are very low and differences between temperatures were previously undetectable [21], and not for injections at the later time point when hormone titer differences between temperatures are clear (Figure 1). This suggests a window of sensitivity to the hormone between our two injection time points, that is, between 3% and 16% of pupal life. For only one of the target traits (trait 5 g), did we see an effect of later hormone manipulation. This indicates some level of heterochrony in the development of this trait, which appears to have a later window of sensitivity to the hormone. Heterochrony, differences in the developmental times and/or rates, is an important contributor to phenotypic diversification, including for butterfly wing patterns [37],[38]. We have shown previously that hormone manipulations at later time points do affect a number of life-history traits [39].
We did not observe significant effects of hormone manipulations at higher temperatures (Figure 4), even if our manipulations did significantly change hormone titers. We measured 20E concentration in the hemolymph of pupae at 3.5% and 8.5% of pupal development time for the two extreme experimental temperatures after early injection of hormone and of control solutions (Figure 1B). Hormone levels are significantly higher for hormone-injected versus control-injected pupae at both rearing temperatures [see Additional file 4]. Control pupae show higher 20E levels when reared at 27°C relative to 19°C, consistent with the relatively faster increase in natural hormone titers that occurs at higher temperatures (Figure 1A). After hormone injection we can no longer detect differences in internal levels between temperatures (Figure 1B).
Differences in trait associations in response to external and internal cues
Focusing on the eyespot traits that are plastic in relation to temperature and/or to hormone titers, we can identify different categories of response [see Additional file 5 summarized in Figure 5]. The principal component analyses [see Additional file 6], a standard approach for analyses of multidimensional datasets such as ours, identified traits with similar and contrasted responses but not with the same resolution as our analyses of individual traits (compare Figures 3 and 4).
The groups identified based on the response to temperature largely contrast eyespots on the forewing versus hindwing (Figures 3 and 5A). All forewing eyespot traits are significantly smaller at 19°C and do not differ between 23°C and 27°C, while all hindwing eyespot traits significantly increase in size with temperature. In summary, for the effects of temperature on wing patterning, we observed looser integration across autonomously-developing wings and tighter coordination of traits on the same wing. The single hindwing trait (trait 5 g) that responds to temperature in the same manner as all forewing traits (Figure 3 and Additional file 5: Figure S2A) is also the only trait significantly affected by late hormone manipulations (Figure 4). It is unclear what, developmentally or ecologically, might be the uniqueness of this trait.
For the traits that we found to be sensitive to early manipulations of pupal hormone levels, we found a different pattern of coordinated responses. Because 1) color rings of each eyespot are specified by the same morphogen gradient established from each eyespot’s center [19],[40], 2) each eyespot center produces morphogen independently of other eyespots [30], and 3) eyespot centers have been shown to have higher levels of ecdysone receptor protein [41], we had hypothesized that all rings of a single eyespot would respond to hormone manipulations in concert and relatively independently from those of other eyespots [19],[28]. However, rings of the same color, and not rings of the same eyespot, responded in a similar manner (Figures 4 and 5B). All plastic black rings showed hormone-related changes only at 19°C while all golden rings showed hormone-related changes both at 19°C and 23°C (Figure 4). Among the golden rings, we can further distinguish between those from the anterior versus the posterior-half of the wings. They differ in relation to how much hormone-related change we saw at 19°C versus 23°C (Figure 4, Additional file 5: Figure S2B). This is consistent with studies showing coupling of anterior (and of posterior) portions across wing surfaces [27] and uncoupling of anterior versus posterior eyespots within the same wing surface [28],[42].
Compartmentalization of hormone effects is not explained by hormone receptor localization
As a mechanism for local sensitivities to systemic levels of 20E, we hypothesized that groups of cells that responded differently to 20E manipulations would differ in expression of ecdysone receptor (EcR). To test this hypothesis, we investigated the localization of EcR protein in wings from pupae reared at different temperatures using an antibody against B. anynana’s EcR [43]. We found EcR in cells on the entire pupal wing epidermis at all temperatures and throughout the whole early pupal life, extending well after the 16% of developmental pupal time used as our last injection time point (Figure 6). The density of EcR-positive cells was higher in circular regions corresponding to the eyespot organizing centers [41]. These regions were smaller for pupae reared at 19°C relative to 27°C (Figure 6B, C versus 6F, G; [44]), and for smaller versus larger eyespots (Figure 6B, F versus 6C, G).
Surprisingly, however, this pattern of EcR expression was detected both for the highly plastic ventral and the hormone-unresponsive dorsal eyespots. This shows that the non-responsiveness of the dorsal color traits to hormone manipulations cannot be due to the corresponding cells not having the receptor for the systemic signal, as had been previously proposed [22]. Our data also did not reveal visible differences in EcR levels between the regions of the presumptive black versus golden eyespot rings (Figure 6B-D and 6F-G) that showed different sensitivities to the hormone injections (Figure 5). This indicates that differences in the way they respond to hormone manipulations (Figure 5B) must be determined either upstream of the binding of 20E to its receptor in the cell nucleus (for example, cell permeability to hormone) or downstream of that (for example, factors interacting with the activated EcR (compare with [14]).