ABSTRACT: Heliconius butterfly wing pattern diversity offers a unique opportunity to investigate how natural genetic variation can drive the evolution of complex adaptive phenotypes. Here we took a large-scale transcriptomic approach to identify the network of genes involved in Heliconius wing pattern development and variation. This included applying 147 microarrays representing the Heliconius transcriptome to assay shifts in gene expression across pupal development among several wing pattern morphs of Heliconius erato. We focused in particular on genes differentially expressed relative to the gene optix, which controls red pattern elements in wings. We combined expression results from three hindwing morphs from Peru and from dissected basal to apical wing elements in two forewing morphs to uncover two main classes of genes. First we looked for candidate upstream regulators of optix by determining transcripts expressed differently across basal to apical sections of the forewing prior to optix expression. Second, we assessed how optix regulates downstream gene expression by targeting transcripts with differential expression similar to optix, where expression differs among red wing pattern elements of both the forewing and hindwing. This study is an analysis of two distinct datasets generated using the same microarray platform. One dataset involved comparative analysis of forewing sections of different color morphs, while the other compared whole hindwings with different color patterns. For the forewing analysis we compared proximal, medial, and distal wing sections of two color pattern morphs: H. erato petiverana and a hybrid H. himera x H. erato etylus. The proximal section in H. erato petiverana is black and the hybrid form orange/red, the medial section is red in H. erato petiverana and pale yellow in the hybrid form, and the distal section is black in both races. For the hindwing analysis, we compared hindwing color pattern gene expression in three races that meet in a hybrid zone in Peru. H. erato emma has a rayed hindwing, H. erato favorinus has a yellow-barred hindwing, and H. erato amphritrite has a black hindwing. Wings were dissected at five time intervals: 1, 3, and 5 days after pupation, when orange/red ommochrome pigments were beginning to be expressed (~7 days after pupation), and when black melanin pigments were starting to pepper the center of the wings (~8 days after pupation). In the forewings, Days 1, 3, and 5 were at 12, 36, and 60 hours post-pupation. In the hindwings these stages were sampled at 24, 48, and 72 hours. Samples hybridized to microarrays included three replicates each of each race, stage, and wing section for forewings (3 replicates x 2 morphs x 3 wing sections x 5 stages, with one replicate wing missing for Day 1 H. e. petiverana = 87 samples) and four replicates of each stage and race for hindwings (4 replicates x 3 races x 5 stages = 60 samples). Total RNA was extracted and converted to cDNA. Cy3-labeling of samples, hybridization, and array scanning was performed according to NimbleGen protocols (2008): for the forewings this was performed at the City of Hope Functional Genomics Core, while the hindwings were run separately at NCSU.Samples were hybridized to NimbleGen HD2 12-plex arrays. These arrays include 12 identical subarrays with 135,000 60 bp probes each, each hybridizing a separate sample. Samples were distributed across arrays to prevent repeat conditions as much as possible and to space similar conditions in different regions of the slide. The array design involved two classes of probes. First there was a tiling component involving 89,310 probes tiled across three genomic intervals. Results from the tiling data were used for the initial discovery of the optix gene and are not the focus of the present study. The second component involved a representation of a set of 12,450 transcript contigs at 1-6X coverage for a total of 40,046 probes, with a mean coverage of 3-4 probes per contig. The number of probes for each contig depended on the ability to create suitable probes according to NimbleGen probe selection criteria and was limited by the small size of some transcripts and the minimum spacing criterion of 15 bp apart. Sequences of low complexity and high repeats with the rest of the genome (>5X representation), determined by comparison against 1.6 MB of genomic sequence available at the time, were avoided for designing probes. An additional 3,248 random probes were placed on the array for quality control.
