Subtle variation in size and shape of the whole forewing and the red band among co‐mimics revealed by geometric morphometric analysis in Heliconius butterflies

Abstract Heliconius are unpalatable butterflies that exhibit remarkable intra‐ and interspecific variation in wing color pattern, specifically warning coloration. Species that have converged on the same pattern are often clustered in Müllerian mimicry rings. Overall, wing color patterns are nearly identical among co‐mimics. However, fine‐scale differences exist, indicating that factors in addition to natural selection may underlie wing phenotype. Here, we investigate differences in shape and size of the forewing and the red band in the Heliconius postman mimicry ring (H. erato phyllis and the co‐mimics H. besckei, H. melpomene burchelli, and H. melpomene nanna) using a landmark‐based approach. If phenotypic evolution is driven entirely by predation pressure, we expect nonsignificant differences among co‐mimics in terms of wing shape. Also, a reinforcement of wing pattern (i.e., greater similarity) could occur when co‐mimics are in sympatry. We also examined variation in the red forewing band because this trait is critical for both mimicry and sexual communication. Morphometric results revealed significant but small differences among species, particularly in the shape of the forewing of co‐mimics. Although we did not observe greater similarity when co‐mimics were in sympatry, nearly identical patterns provided evidence of convergence for mimicry. In contrast, mimetic pairs could be distinguished based on the shape (but not the size) of the red band, suggesting an “advergence” process. In addition, sexual dimorphism in the red band shape (but not size) was found for all lineages. Thus, we infer that natural selection due to predation by birds might not be the only mechanism responsible for variation in color patterns, and sexual selection could be an important driver of wing phenotypic evolution in this mimicry ring.


| INTRODUCTION
Neotropical Heliconius butterflies represent a conspicuously variable group from a morphological perspective (Holzinger & Holzinger, 1994), generally considered as a good model of evolutionary studies.
The adaptive nature of warning color patterns in Heliconius butterflies has been demonstrated experimentally in the wild by Benson (1972), Mallet and Barton (1989), and Kapan (2001). According to Joron (2009), mimicry is advantageous to mimetic butterflies, first by lessening the risk of predation by birds, which associate their warning color pattern with unpalatability. Second, because variant individuals existing in a given population may not be recognized as unpalatable, they are likely under greater predation risk. However, there is little information about the threshold for variation in the size and shape of these signals under natural conditions, if any (Mérot, Poul, Therv, & Joron, 2016), as well as the effect of concurrent selective pressures.
The existence of phenotypic variation also may be advantageous to bird predators, because those individuals that better associate the unpalatability signal on the butterfly wing may have greater survivorship (Mallet & Barton, 1989). Learning in this case may be associated with several phenotypic, behavioral, and ecological factors associated with the butterflies, for example: (1) color (Svádová et al., 2009) and pattern (Ihalainen, Lindström, Mappes, & Puolakkainen, 2008) of the wings; (2) level of unpalatability (Ihalainen, 2006); and (3) relative frequency of the co-mimics (Mérot et al., 2016;Rowland, Wiley, Ruxton, Mappes, & Speed, 2010;Speed, 2001). Variation in the former factors has not been quantitatively examined in the wild for any Heliconius mimicry ring. Unless predators are unable to detect fine-scale differences, it is expected that natural selection alone would eventually lead to identical patterns of wing shape for each mimetic pair in Heliconius butterflies. In other words, "the ultimate prediction of Müllerian mimicry is that butterflies of a similar size should all ultimately converge on the same color pattern" (Brower, 1996;Mallet, Jiggins, & McMillan, 1996).
However, some degree of imperfection in mimicry is possible, especially if the cost to the mimetic butterfly for mate discrimination is greater than the benefit of mimicry protection. Phenotypic resemblance among co-mimics could also impose costs due to possible mistakes in species identity during courtship (Estrada & Jiggins, 2008).
Consequently, there could be a conflict between the outcomes of natural and sexual selection in mimetic species. Resemblance might also evolve under several selective pressures acting in a sex-specific manner (Su, Lim, & Krushnamegh, 2015). Thus, additional, nonmutually exclusive, hypotheses may explain the existence of nonidentical mimics in the wild, including the "eye-of-the-beholder," where imperfect mimicry might be attributable to differences between the dimensions of organisms that humans notice versus the ones their ecologically relevant signal receivers pay attention to (Kikuchi & Pfennig, 2013). Similarly, the "mimetic breakdown" (Brower, 1960) assumes that imprecise mimicry reflects a trade-off between gene flow and selection. Moreover, some wing traits could have evolved under concurrent pressures (see Rossato et al., 2018) with, for example, the shape of the red band being influenced by sexual selection in Heliconius (Emsley, 1970).
Finally, morphological similarities between Müllerian co-mimics might be influenced not only by natural selection favoring accurate mimicry, but also by the genetic architecture underlying variation in wing phenotypes (Mérot et al., 2016), including introgression (e.g., H. melpomene and H. timareta; Heliconius Genome Consortium 2012; Pardo-Diaz et al., 2012). Wing color pattern is controlled by few loci of major effect in Heliconius erato and Heliconius melpomene (Baxter F I G U R E 1 Phylogenetic tree of major clades within Heliconius. Illustrations on the right side depict wing pattern. The postman mimicry-ring pattern is represented by the three species marked in red: H. erato ("erato" clade), H. melpomene ("melpomene/ cydno" clade), and H. besckei ("silvaniform" clade). The tree was adapted from Kozak et al. (2015) including taxa of interest to this study et al., 2008). Thus, a common "tool kit" of genes, consisting of approximately five unlinked genetic loci that control almost all of the variation, has been used repeatedly by different species to produce both convergent and divergent wing patterns (Huber et al., 2015;Joron et al., 2006;Kronforst et al., 2006). Three of these loci that control size and shape of wing color patterns have been characterized at a molecular level: (1) the transcription factor optix, which controls the distribution of red color pattern across the wings (Reed et al., 2011), Kronforst & Papa, 2015), (2) the cell cycle regulator cortex, responsible for yellow patterning (Nadeau et al., 2016), and (3) WntA, a signaling ligand that controls melanin patterning across the wing (Martin et al., 2012).
In this study, we evaluate fine-scale morphological differences in the whole forewing and the red band trait among members of the postman mimicry ring of Heliconius butterflies. We explored the species H. erato phyllis, and its distantly related Müllerian co-mimics H. besckei, H. melpomene nanna, and H. melpomene burchelli (Figure 1) that display the same warningly colored wing patterns in local populations, yet exhibit pattern diversity between geographic regions. Landmark and contour analysis based on semilandmarks were used to characterize spatial variation in forewing size and shape, and in the red band in H. erato phyllis across its distributional range, including areas of sympatry with H. besckei, H. melpomene nanna, and H. melpomene burchelli. H. erato phyllis is known to vary substantially in space, in terms of overall wing size and shape, due in part to its use of several species of host plants (passion vines) that also differ in their distribution (Jorge, Estrela, Klaczko, Moreira, & Freitas, 2011;Rodrigues & Moreira, 2002).
However, there are no data about whether this wing pattern variation is spatially structured. There is also a lack of information about variation in size of the phenotypic signals that predators may use as cues, with the exception of those provided by Klein and Araújo (2013) for populations of H. erato phyllis and H. besckei.
If phenotypic resemblance among co-mimics resulted mainly by predation pressure (i.e., convergence for mimicry; Kapan (2001); Mallet and Barton (1989)), we expected (1) nonsignificant differences among species (i.e., nearly identical co-mimics) considering the size and shape of whole forewing and (2) reinforcement of wing patterning when co-mimics are in sympatry. Alternatively, significant differences might exist in the size and shape of the red band, as this trait is also involved in sexual communication (Estrada & Jiggins, 2008), and therefore could be influenced by sexual selection. Finally, we inferred evolutionary patterns using the mitochondrial gene Cytochrome oxidase I (CoI) and the wing patterning gene optix, to evaluate evidence for introgression, which is also involved in morphological resemblance. We chose optix, in particular, because it is responsible for red wing pattern variation across Heliconius species (Reed et al., 2011).

| Species samples
Variation in overall forewing size and shape, and in the forewing red band, was analyzed in 229 field-collected dried specimens of H. erato phyllis, H. besckei, H. melpomene burchelli, and H. melpomene nanna (Table 1). Specimens were chosen randomly, according to availability (mainly from the insect collection at the Universidade Federal do Paraná-UFPR, Brazil), until at least 20 specimens for each species or subspecies had been examined, following the geographic boundaries established by Brown (1979) and Rosser, Phillimore, Huertas, F I G U R E 2 Distribution of species in the "postman" mimicry ring in Brazil. (a) Specimens from each co-mimic of the mimicry ring. Maps represent the geographic locations of samples used in this study, as follows: (b) H. erato phyllis, with samples located in the central, northern, and southern regions, represented by red, blue, and green circles, respectively; (c) H. besckei (green squares); and (d) H. melpomene burchelli and H. melpomene nanna (red and blue triangles, respectively). Gray areas show the overall distribution of each mimetic-ring member, according to Rosser et al. (2012). Biogeographical subregions are shown in brown (Amazon Forest), pale yellow (Chacoan), and pale green (Atlantic Rain Forest), following Morrone (2006)

| Morphometric data
Dorsal surfaces of individual forewings were photographed by the same person (DOR) using a Sony Cybershot H20 digital camera, 5-megapixel resolution, Iso200, one-shot, flash off, and macro function activated ( Figure 3a). The dorsal surface was chosen because it is likely to be subject to natural and sexual selection. In addition to the whole wing, we analyzed the size and shape of the red forewing band. Here, we focus on this wing color trait (and not the basal yellow stripe, for example) because it is known to be used as a visual cue in H. erato phyllis courtship and is likely also to be important for H.
melpomene (Emsley, 1970). We used a total of 19 landmarks (Jorge et al., 2011) for the entire wing ( Figure 3b) and eight landmarks plus 35 semilandmarks for the red band ( Figure 3c) (for a complete description of the landmarks, see Table S1). Landmarks and semilandmarks were digitized by the same person (DOR), using TPSDig 2.17 (Rohlf, 2013).
Coordinates from the entire forewing and the red band were superimposed using a generalized Procrustes analysis (GPA) (Dryden & Mardia, 1998)  sliding along the outline curve until they matched as closely as possible the positions of the corresponding point on a reference specimen (Bookstein, 1997). The consensus configuration (mean shape) was calculated, and the difference among mean landmarks and individual landmarks resulted in a residual matrix (Jolliffe, 1986). This matrix was used in a principal components analysis (PCA) as the new shape variables. This procedure allowed us to reduce the dimensionality of the dataset and to work with independent variables (Cordeiro-Estrela, Baylac, Denys, & Marinho-Filho, 2006).
Comparative analyses of shape and size of the forewing and red band were used to evaluate evidence for geographic convergence between co-mimics and differences between sexes. Samples were grouped in different ways for analyses focused on mimetic conver-

| Forewing and red band size
Images were scaled using software IMP-CoordGen6f (Sheets, 2001), in order to compare size among members of the mimetic ring. Size was estimated as the log-transformed centroid size, which represents the square root of the sum of squared distances of each landmark from the centroid of the configuration (Bookstein, 1991). We performed a one-way analysis of variance (ANOVA), followed by Tukey's pairwise comparison tests, to determine whether forewing and red band size differed among mimicryring members. ANOVA was also used to evaluate the effect of different pressures (mimicry convergence and sexual selection) on F I G U R E 5 First two axes of the principal components analysis (PCA) on shape residuals for the whole forewing (a) and red band (b)  Regression analyses were performed between log-centroid size of the whole forewing and red band to test for the existence of allometry, in relation to each member regarding isometry, and between genders within each mimicry-ring member. Slope lines and intercepts were compared using one-way analysis of covariance (ANCOVA). Allometric analyses and graphs were performed with GraphPad Prism 5.00 (Motulsky & Christopoulos, 2003) and edited in CorelDraw X4 (Corel Corporation).

| Forewing and red band shape
Variation in the shape of the whole forewing and the red band was explored using principal components analyses. Analyses of shape variation, based on the first two principal components of the wing and red band, were conducted with TPSRelw 1.49 (Rohlf, 2010a)

| Evolutionary patterns
Resemblances between Heliconius co-mimetic species could be partially driven by genetic similarities due to shared evolutionary history,  (Reed et al., 2011). Primers and PCR conditions used were as described by Beltrán, Jiggins, Brower, Bermingham, and Mallet (2007) and Hines et al. (2011) for CoI and optix, respectively. Aliquots of PCR products were treated with Exonuclease I and FastAP Thermosensitive Alkaline Phosphatase T A B L E 3 MANOVA results for shape variation of the forewing and red band in Heliconiu erato phyllis, taking into account geographic group and sex (total = three groups, 60 individuals, 29 female and 31 male) (2) Mahalanobis distance based on the whole forewing and the red band. Positive association (r > 0) between the whole forewing and red band distances was inferred through a Mantel test using Pearson's correlation coefficient in the software XLSTAT (Addinsoft).

| Variation in the size of whole forewing and red band among co-mimics
The ANOVA based on whole forewing centroid size showed significant differences among groups (F 5,223 = 5.777, r 2 = .09, p < .001).
There was no difference in the forewing size between the three T A B L E 5 Comparison of allometric coefficients in wing versus red band centroid size (regression analysis) within and between sexes in Heliconius mimicry-ring members

| Variation in the shape of whole forewing and red band among co-mimics
The results of the MANOVA based on shape of the forewing and red band showed significant differences among all members of the mimicry ring, between sexes (discussed in details below), and also for the interaction between these two factors ( Table 2). The first ten PCs for the whole wing shape explained 85.12% of the variation.
There was no clear separation among mimicry-ring members in this case for the first two PCs that together explained ~48% of variation subspecies is shown in Figure 6. Thus, although the MANOVA test indicated statistically significant difference, the differences in the shape of the whole forewing were relatively small. The posterior probability of classification to each co-mimic species and subspecies,

| Effect of sex and geographic groups on forewing in Heliconius mimetic-ring members
The ANOVA results comparing the effect of sex and geographic group on red band size suggest that both mimicry convergence and sexual dimorphism act in H. erato phyllis (F 5,54 = 2.60, r 2 = .12, p = .03) and that there is an interaction between these factors. The red band was smallest in the H. erato phyllis group that was sympatric with H. melpomene nanna. We did not find significant difference in whole forewing size (F 5,54 = 0.82, r 2 = −.01, p = .54). The MANOVA results for shape variation in H. erato phyllis indicate that both selective factors affect these wing traits, but they act on different structures (Table 3).
For instance, while geographic group was important to forewing shape, the red band shape varied according to sex.

| Differences between males and females
There was no significant difference between males and females in forewing size for all species and subspecies (Table 4). However, the red band size showed a significant difference between sexes for all species and subspecies, being relatively bigger in males in relation to females (Figure 7), except H. erato phyllis, in which no difference was found (Table 4). The size relationship between the wing and the red band showed negative allometry only in H. erato phyllis (y = 0.64x + 1.03, r 2 = .45, p < .0001, Figure 7a). For the other mimicry-ring members, the relationship was isometric, with the corresponding linear regression equations as follows: H. besckei (y = 0.94x + 0.99, r 2 = .71, p = .48, Figure 7b), H. melpomene burchelli (y = 0.98x + 0.01, r 2 = .75, p = .75, Figure 7c), and H. melpomene nanna (y = 0.98x + 0.006, r 2 = .781, p = .80, Figure 7d). For all mimicry-ring members, males and females did not differ from each other in terms of allometric coefficient.
However, comparing them to the isometric line, males and females of H. e. phyllis and males of H. besckei showed negative allometry. For the H. melpomene subspecies, males and females did not differ from isometry (Table 5; Figure 7). The MANOVAs on the whole forewing and red band shape for males and females showed significant differences for all mimicry-ring members (Table 6). In general, the main differences between sexes were concentrated in the first PC, and therefore, we show only the shape variation for PC1 in the PCA (Figure 8).

| Optix and COI phylogeny and genetic distances
We sequenced a 1.6-kb region of CoI and optix to infer phylogenetic relationships among H. erato phyllis and the co-mimics and identify potential patterns of introgression. Gene trees reconstructed based on CoI and optix resulted in similar topologies (H. erato phyllis (H. besckei and H. melpomene)), but with subtle differences in branch lengths ( Figure 9). Thus, no obvious discordance was evident, although H.
besckei has previously been shown to have acquired wing patterning mimicry, and the genomic segment around optix, from H. melpomene via introgression (Zhang, Dasmahapatra, Mallet, Moreira, & Kronforst, 2016). On the other hand, neighbor-joining (NJ) trees of Mahalanobis distance based on the whole wing and the red band indicated that H.
erato phyllis was more similar to H. melpomene subspecies (Figure 10).
Mantel tests indicated that genetic distance was positively correlated with morphology (Mahalanobis) distance for both the whole forewing (r = .90; p = .20) and red band (r = −.99; p = .92) shapes, although these were not statistically significant.

| Convergence and "advergence" in wing pattern among co-mimics
The morphological diversity of animals is generally assumed to be adaptive and shaped by natural selection. Nevertheless, traits can be influenced by multiple selective pressures, some of which may act in conflicting directions (Petrović et al., 2015). By refining the morphology of wing traits, we found that natural and sexual selection might be acting in opposing directions (as firstly pointed by Estrada and Jiggins (2008)) in members of the postman mimicry ring. Our results demonstrate that variation in size and shape is quite distinct between co-mimics considering the whole forewing and the red band.
First, the lack of differentiation in size and shape of whole forewing indicates that co-mimics are nearly identical, that is, they likely converged for mimicry as a response to predation pressure. This is particularly clear for the shape of whole forewing of the two broadly codistributed species, H. erato phylis and the subspecies of H. melpomene (Figure 10), which are phenotypically similar, despite being more distantly related. The evolutionary history of these species should be taken into account to explain such resemblance (Brown, Sheppard, & Turner, 1974;Eltringham, 1916;Flanagan et al., 2004;Quek et al., 2010;Turner & Mallet, 1996). However, the color pattern similarity observed between H. erato and H. melpomene might also result from an interaction between environmental changes, host plant distribution, and phenotypic plasticity (Rossato et al., 2018).
Additionally, a phylogenetic constraint could not be ruled out as we found a relevant signal for correlation between the wing pattern and genetic distance, although this was not statistically significant due to the low sample size. Convergence was also observed for the size of erato phyllis, in which the wing size and shape depend on passion vines that also differ in distribution (Jorge et al., 2011;Rodrigues & Moreira, 2002).
Second, our results revealed that differences in the shape of the red band were diagnosable and consistent among members of the postman mimicry ring, including the subspecies of co-mimics. The correct classification, based on shape, of each species (100%) or subspecies (>80%) showed the dissimilarity between geographic group of H. erato phyllis and sympatric co-mimics. Accordingly, co-mimics are not identical in the way that humans view their shape.
Thus, natural selection due to predation by birds may be only partly responsible for shaping forewing traits. It is possible that when the butterflies are actively flying, predators cue on the red band on a broader scale and do not recognize these phenotypic variants as different entities in the wild. Co-mimics may not be identical because they display a more generalized signal in which subtle differences are not perceived or are ignored by predators (Rowe, Lindström, & Lyytinen, 2004). Additionally, nonidentical convergence does not mean absence of mimicry, but could be result of other selective pressures (Srygley, 1994). In the MANOVA results (Table 2), the differences in shape are related to the interaction between sex and groups, suggesting that sexes and geographic groups result in different shape (Table 2). Co-mimetic species may, in fact, benefit from the presence of other co-mimics, even when these are slightly dissimilar (Rowe et al., 2004;Rowland, Ihalainen, Lindström, Mappes, & Speed, 2007;Merót et al., 2016;Finkbeiner, Briscoe, & Mullen, 2017). In the area of Brazil where we conducted this study, there are other members of the postman mimicry ring that occur at low densities, for example, Eresia lansdorfi (Nymphalidae). While this species displays a red forewing band similar to others in the postman mimicry ring, it may be a Batesian mimic and was not considered a Müllerian co-mimic. At least initially during the learning phase, chemical differences among co-mimics may be more important than visual similarities (Darst & Cumming, 2006;Lindström, Alatalo, & Mappes, 1997).
Heliconius butterflies have conspicuous exocrine glands on the last abdominal segments, which were originally presumed to be associated with defense in both sexes (Müller, 1912;Ross et al., 2001), but have not been explored in detail in the context of Müllerian mimicry. Lately, these structures have been associated with the production (males) and storage and dispersal (females) of antiaphrodisiacs (Gilbert, 1976;Schulz, Estrada, Yildizhan, Boppré, & Gilbert, 2008).
However, these functions may be not mutually exclusive. The possibility that chemicals produced by these structures have a role in species recognition also remains to be explored.  (Cespedes, Penz, & DeVries, 2015) and abiotic factors such as host plant usage (Jorge et al., 2011). The spatial distribution of each plant could vary substantially among areas, which could result in several phenotypically distinct co-mimics, defined as races or subspecies according to host plant occurrence (Brown, 1979;Hines et al., 2011;Rosser et al., 2012;Turner, 1981).

| Species identity and sexual dimorphism
We found evidence for the existence of forewing variation between all co-mimics (including the subspecies of H. melpomene) and between males and females of each species. Differences between the subspecies cannot be explained only by mimicry with H. erato. Thus, the differences in shape are consistent to each co-mimic (high percent of the correct classification in the same species). The differences in shape are bigger considering the red band of each co-mimic, suggesting a maintenance of this trait for sexual communication. It is also possible that they are used in courtship, because the MANOVA results demonstrate the red band shape is strongly influenced by sex (Table 4). As sexual dimorphism in the shape of the red band was found in all lineages explored in this study, the results also suggest that sexual selection may be involved in the evolution of this trait. The use of color and wing shape as visual stimuli during courtship in both sexes of H. erato was first demonstrated by Crane (1955). According to Emsley (1970), the red color is in fact used as a visual cue in H. erato phyllis courtship and is likely also to be important for H. melpomene melpomene.
However, preliminary tests conducted by Emsley suggested that the yellow color present on the hind wings of the mimicry-ring members, and which was not taken into account here, is important for H. besckei.
It is unclear whether such subtle differences in shape could be efficiently used as visual cues.
Similarly to Ramos and Freitas (1999) The negative allometry found for the forewing red band in H. erato phyllis was interpreted as evidence for the existence of a size threshold as an effective visual cue used in the context of either predation and/ or courtship. In other words, with individuals varying up to 50% in size in the wild (Rodrigues & Moreira, 2002), those with small forewings compensate by having a proportionally larger red band. This appears not be the case for H. melpomene nanna, perhaps because on average, they vary less and are naturally larger than H. erato phyllis.
Finally, the evolutionary relationships revealed by phylogenies inferred from CoI and optix were identical, consistent with the known phylogeny and a history of introgression between H. besckei and H.
melpomene. Probably, distinct alleles are involved in wing color shape and the type of variation that we observed in the postman mimicry ring is either controlled by a different allele or region than that studied.
Moreover, epistatic interaction between optix and the modifier locus N results in a narrow forewing red band (Martin et al., 2014), pointing to a more complex genetic architecture underlying the red wing pattern.

| CONCLUSION
In this study, we demonstrated the existence of distinct patterns of variation in shape and size of the whole forewing and the red band among co-mimics of the Heliconius postman ring, which suggests mimicry convergence and sexual selection acting in opposing directions.
The two most widely distributed species, and distantly related, H. erato phyllis and H. melpomene, are the most similar regarding whole forewing and red band. Also, we found consistent differences in the red band that enables us to distinguish among co-mimics, including subspecies, even when in sympatry. Sexual dimorphism not in size but in shape was found in relation to the red band suggesting that sexual selection might play a role in the evolution of this trait. Thus, we inferred that natural selection due to predation by birds, which in theory would lead to nearly identical color patterns, is not the only mechanism responsible for the variation in these phenotypic patterns. Whether the mimicry-ring members tested here use these differences as visual cues to identify, compete, and/or choose their partners remains to be tested. As postulated by Mallet et al. (1996), mimicry theory is still open to discussion, and "poses more questions than it answers."

ACKNOWLEDGMENTS
Thanks are due to Olaf Mielke (UFPR) for providing museum specimens used in the geometric morphometric analysis and for assisting in museum work. We are grateful to Amabilio Camargo (