Since Carroll et al.  described the first gene expression patterns associated with butterfly wing pattern development, researchers have used the candidate gene approach to identify over a dozen developmental genes associated with eyespot and stripe pattern development in various butterfly species [49–56]. Additional studies focusing on ommochrome and melanin candidate genes [57–61] increased the total of number of wing-pattern related genes to around 20 and provided some insight into the identity of potential downstream genes involved in pattern realization. The recent positional cloning of the optix color pattern gene  demonstrates the potential of forward genetics for identifying further genes. In this study we sought to accelerate gene discovery by moving beyond the candidate gene paradigm. Our work is the first large-scale expression assay for butterfly wing pattern genes. We have identified over 200 genes associated with color patterning, including several potential regulators of optix and a host of structural and pigmentation genes that have expression patterns that are correlated with adaptive color pattern variation in natural populations. These new data allow us to begin to understand the structure of the broader network of patterning and pigmentation genes in Heliconius and bring us closer to understanding the developmental genetic architecture of color pattern evolution.
Optix and the color patterning gene network
Consistent with its role as a red color pattern switch gene , optix was the first transcript observed to show clear red-specific differential expression in our array data (Table 2, Figure 4). Furthermore, among the color-specific genes, optix had the longest persisting differential expression, lasting from Day 3 of the pupal stage through to ommochrome synthesis near the end of pupation (Figure 4, Table 2). Additional color-specific genes were expressed primarily during ommochrome and/or melanin pigment development. The long-term persistence of optix transcription accords well with its likely role as a selector gene , acting to reinforce positional fate through sustained expression rather than being part of a transient developmental cascade. Interestingly, out of the 51 genes that were expressed in association with optix there were no other obvious transcription factors or developmental signaling genes. Instead, the transcripts we identified largely represented genes involved in cuticle structure and pigment synthesis. Although our data are not exhaustive, they suggest that optix may play a relatively direct role in regulating scale and pigment development, as opposed to a more intermediate role in coordinating further downstream pattern formation processes.
The overall temporal sequence - optix expression followed by cuticle gene expression, which in turn was followed by pigment gene expression - is consistent with the progression of wing development, as scale maturation and sclerotization precedes the appearance of pigments (Figure 1B). Scales of different pigmentation also differ in cuticular fine structure, suggesting an interaction between cuticle formation and pigment synthesis . Cuticle proteins compose a highly diverse gene family consisting of hundreds of genes involved in an extensive, yet poorly characterized, functional diversification [63–65]. The enrichment of this class of genes throughout the differentially expressed gene sets further suggests the importance of these genes in color pattern differentiation.
Early prepatterning genes
Many of the genes differentially expressed across the proximal-distal wing axis were higher basally and involved structural genes, including extracellular matrix, cytoskeleton, and muscle genes. This may reflect potential differences in wing tissues at the basal hinge region versus the tip of the wing. Among these genes there was little evidence for expression differences being driven by developmental timing of scale development, as patterns involved consistent basal or apical differentiation across stages rather than evidence of delayed effects (as in Figure 3).
We found that early pupal genes expressed in association with proximal-distal wing pattern sections were enriched for morphogenesis and transcriptional regulation functions (Figures 2 and 3, Table 1), which was consistent with our expectation of finding genes involved in regulating development. Most of these genes have been previously recognized to play a role in D. melanogaster early wing axis formation, and several are worth highlighting as especially strong candidates for regulators of optix. In particular, orthologs of the transcription factors zfh2, homothorax, and araucan showed sustained proximal expression throughout pupal wing development, potentially suggesting selector gene roles for these molecules. tiptop also showed a strong association with the proximal section of the wing early in development, although its expression waned at Day 3 to Day 5, coincident with the onset of optix expression. As for the known functions of these candidates, in D. melanogaster homothorax is a homeodomain protein known to be involved in establishing proximal wing fate  and zfh2 proximal expression plays a role in preventing distal fates . araucan is a transcription factor primarily known for its role in D. melanogaster wing vein specification . This is a particularly interesting new candidate gene because the rayed hindwing pattern develops relative to wing venation, with many of the rays positioned parallel to and halfway between the wing veins. tiptop is a selector gene involved in specifying positional identity in various insect appendages  and is known to interact developmentally with homothorax and optix in D. melanogaster. four
vestigial, and distalless, which showed a specific association with the distal tip of the pupal wing before and during optix differential expression, are known as distal appendage and/or wing determinants in D. melanogaster[70, 71, 72, 73]. serrate
bowl, and wnt6 have less significant associations in our data, but are implicated in various aspects of positional specification in D. melanogaster wing development. Follow-up in situ hybridizations are needed to more rigorously assess the potential role of these genes in prepatterning.
Ommochrome pigments: Enzyme regulation and novel transporters
Genes with color-specific differential expression almost always showed higher expression in red pattern elements (Table 2, Additional file 3). Perhaps unsurprisingly this pattern of upregulation encompassed many genes implicated in the synthesis of ommochromes, the class of pigments that imparts the red coloration in these butterflies and whose precursor, 3-OH-kynurenine, imparts the yellow pigmentation. However, some specific gene expression patterns we observed were unexpected and suggest that a revision of the current model of ommochrome synthesis in butterfly wings is required.
Most of what is currently known about the genetic basis of ommochrome synthesis comes from work with D. melanogaster eye mutants, and we have previously relied on this work to propose a model of how ommochromes might be produced in butterfly wings . D. melanogaster ommochrome mutations tend to fall into three functional classes: transporters (e.g., white, scarlet, karmoisin), pigment synthesis enzymes (e.g., cinnabar, vermilion, kf), and granule formation proteins (e.g., garnet, claret, ruby). Previous work in H. erato and H. melpomene has shown that several of these ommochrome enzyme and transporter genes are expressed in Heliconius wings, and some of them, especially cinnabar, are strongly upregulated in red regions of the wing pattern. Beyond these gene expression associations, however, little is know about how similar ommochrome biosynthesis is between D. melanogaster eyes and butterfly wing scales. In particular, major questions remain regarding the specific precursors that are transported from the hemolymph into scale cells, whether there is anything analogous to pigment granules in scale cells, where precursor transporters are located in the scale-building cells, and what molecules might be active in later steps of ommochrome synthesis and stabilization.
In terms of the expression of enzyme genes, both kf and cinnabar were differentially expressed between color pattern morphs, but in different ways (Figure 5). cinnabar differential expression began after Day 5 and was higher in red and yellow regions in both ommochrome and melanin stages, with highest expression in yellow regions during the melanin stage. This expression pattern is consistent with the inferred role of cinnabar in the production of 3-OH-kynurenine, both as a precursor for red ommochromes and for deposition in the melanin stage as the yellow pigment. These results support local synthesis of 3-OH-kynurenine, in addition to the uptake of 3-OH-kynurenine from the hemolymph . In contrast to cinnabar
kf differential expression occurred during the ommochrome stage, where it was upregulated only in red patterns, with all tissues showing similarly high expression levels by the melanin stage. vermilion, which encodes the initial enzyme in the ommochrome pathway, has yielded inconsistent pattern of differential expression in previous studies, with evidence of higher expression within the red band in ommochrome stages by Reed et al.  and indication of differential expression only in late melanin stages by Ferguson & Jiggins . In the cross-developmental analyses here there were no significant patterns of differential expression in vermilion; instead it showed high, ubiquitous expression early in pupal development (before optix expression), progressively dropping lower over time to barely detectable levels during the ommochrome and melanin stages. None of these enzyme genes showed any obvious pattern of spatial or temporal co-regulation, supporting the previous hypothesis that they are independently regulated , and that the control of timing of pigment synthesis may depend on the regulation of transporters of precursor metabolites.
Accordingly, some of the most interesting findings from our study relate to the expression of ommochrome precursor transporter genes. The ommochrome pigment transporter genes observed in D. melanogaster eyes - white
scarlet, and karmoisin - have uncertain roles in ommochrome synthesis in Heliconius wings. Ferguson and Jiggins  did not find white or karmoisin to be expressed at appreciable levels in H. melpomene wings but did find scarlet to be differentially expressed in the red mid-forewing band in ommochrome- and melanin-stages . Our data showed that white and scarlet are expressed at low levels, with no significant color associations (Figure 4). Likewise, karmoisin was not even represented on our array because its expression levels were too low for its transcript to be identified through EST or 454 sequencing. In contrast to the results with white
scarlet, and karmoisin, we identified transcripts encoding two new ABC transporters potentially involved in pigmentation (i.e., in the same functional class, but not the same subclass, as white and scarlet) and a poorly known monocarboxylate transporter (i.e., in the same functional class as karmoisin) that were significantly upregulated at relatively high levels in red wing sections during ommochrome synthesis (Figure 5).
The pattern/transcript associations described above were made possible due to our ability to section forewing tissues along color pattern boundaries. In the hindwings, we lacked similar tissue-specific controls for comparisons between morphs, thus conclusions on hindwing-specific regulation are more tentative. Two differentially expressed genes in the hindwing of interest, light and lightoid, are thought to play a role in ommochrome pigment granule transport [45–47]. Other regulatory genes with potential color pattern function were differentially expressed only in hindwings (e.g., ovo
seven in absentia), however further work is needed to determine to what extent their expression differences are related to color pattern development.
Overall our ommochrome gene expression data call into question the applicability of the D. melanogaster eye model (Figure 5) to butterfly wing scales. The low or undetectable expression levels of white, scarlet, and karmoisin suggest that these transporters may play little or no role in pigment synthesis in butterfly wings. Conversely, our discovery of three novel color pattern-associated transporters implies that a significant portion of the ommochrome biosynthesis regulatory mechanism in butterfly wings may be quite different from that found in D. melanogaster eyes.
Melanins: Color patterning by repression of pigment synthesis
Our results suggest that several melanin genes drive differences in pigmentation across color elements and forms. Among the known melanin genes represented on our array, ebony
Dat1, and yellow-d were significantly differentially expressed between color pattern phenotypes (Figure 6). Drosophila has shown similar variation in genetic modifiers of melanism, with yellow
ebony, and tan variably implicated in driving melanic variation across species and natural populations [74–78]. Furthermore, our results suggest that variation of black Heliconius wing patterns may be achieved largely through the upregulation of melanin pigment repressors rather than reduced expression of melanin synthesis genes.
ebony differential expression was confined to the melanin-stage and showed upregulation in red tissues relative to other tissues, an expression pattern noted by Ferguson et al.  in several Heliconius species. This expression pattern is consistent with the known function of ebony in D. melanogaster where the ebony protein shunts dopa that would be used for making black or brown melanin to N-β-alanyl dopamine, thus resulting in a lighter yellow sclerotization . The lack of dark pigmentation would thus enable more brilliant red coloration. While this is a simple and appealing model for red phenotypes, it does not apply to yellow phenotypes, which do not show ebony upregulation in those regions of the wing that are fated to be yellow. Thus, the question arises as to how yellow scale cells manage to repress dark melanization. In this regard, it was interesting to find that yellow-barred hindwings showed significantly higher expression of Dat1 during the melanization stage. Dat1 shunts dopamine, the precursor of black dopamine melanin, to N-acetyl dopamine, resulting in clear cuticle using NADA sclerotization . This differential expression is thus an alternative way to eliminate the expression of dark melanins, and makes it possible for 3-OH-kynurenine to produce a vibrant yellow color. In addition to ebony, Ferguson et al.  found that tan is more highly expressed in black parts of the wing during mid to late melanin stages. Both tan and pale approach the upper end of gene expression across tissues, where it is difficult to detect differences using microarray technology. Although not differentially expressed, there is some indication in the early melanin stage that tan expression may be higher in black tissues.
yellow-d was the other known melanin gene we found differentially expressed between color patterns. It showed an expression profile very similar to ebony, with significant upregulation in red pattern elements during pigment synthesis stages. The yellow gene family has diversified into nine major lineages within insects . The specific molecular functions of these yellow proteins are unknown, however several of the paralogs are known to play a role in pigmentation. yellow appears to have a conserved role in pigmentation, promoting melanization in D. melanogaster[74, 80], Coleoptera , and Lepidoptera [79, 82]. In B. mori
yellow-d is also implicated in increasing melanization, while yellow-e is associated with white coloration in larvae . Our finding that yellow-d is associated with a lack of melanization is therefore in contrast to previous results from B. mori. Our results are in accord, however, with recent work in H. melpomene where upregulation of yellow-d and, to a lesser extent yellow-h, was observed in red tissues . These results support the evolutionary labile functions of members of this gene family in pigmentation.