In Bicyclus anynana, two eyespots are present on the dorsal forewing (anterior, A, and posterior, P) and seven on the ventral hindwing (eyespots E1–E7). In separate experiments, we selected for the relative size (the ratio of total eyespot diameter divided by the distance between the centers of A and P; ) or the color composition (the size of the black color ring relative to total eyespot diameter) of two eyespots simultaneously. Selection for eyespot size targeted the two dorsal forewing eyespots . The results of size selection showed that selection to uncouple A and P also uncoupled the relative sizes of anterior and posterior groups of ventral hindwing eyespots, and that the division between the two groups of ventral hindwing eyespots lay between E4 and E6 . Thus, we chose to select on the color composition of E4 and E6 because: 1) the size experiment showed that these two eyespots could be decoupled, even when they were targeted only indirectly by selection (see Results and Discussion); and 2) color composition is a function of total eyespot diameter, so it was necessary to choose two eyespots of initially similar size and color composition before selection.
The starting population for each experiment was derived from an outbred stock maintained in the laboratory for > 100 generations at high Ne . For each trait we established three types of lines from the stock population: 1) antagonistic selection lines where two eyespots were selected in opposite direction (large A and small P, 'Large-Small,' or the opposite combination 'Small-Large'; E4 gold and E6 black, 'Gold-Black,' or the opposite combination 'Black-Gold'); 2) concerted selection lines where two eyespots were selected in the same direction ('Gold-Gold,' or 'Black-Black'; 'Large-Large,' or 'Small-Small'); and 3) unselected control (UC) lines.
To estimate eyespot color composition, total diameter and the diameter of the inner black disc of E4 and E6 were measured in the proximal-distal axis, along the midline of each eyespot (parallel to the wing veins). Color composition of an eyespot was estimated from the orthogonal (Type II) regression of black disc diameter on total eyespot diameter. In this analysis, positive and negative residuals indicate relatively black and relatively gold eyespots, respectively. To identify individuals for selection, we performed an additional orthogonal regression of E4 color composition on E6 color composition (from the previous regression). The regression line is similar to the axis of the first principal component. For concerted selection, we ranked the residuals parallel to the regression line (similar to individual loadings on the first principal component); extreme positive or negative residuals indicated that both eyespots were relatively black or relatively gold, respectively. For antagonistic selection, we ranked the residuals perpendicular to the regression line (similar to individual loadings on the second, orthogonal, principal component axis); extreme positive residuals indicated individuals where E4 was relatively black compared to E6, and extreme negative residuals indicated individuals where E6 was relatively black compared to E4. Selection for eyespot size was made on the basis of the additive combination of the ranks for each individual eyespot as explained in . There were two replicates of each selection and each UC line, with one additional UC replicate in the size experiment. To estimate correlations between eyespot characters in the base population, we calculated Pearson product-moment correlations between the target eyespots in the starting population for each experiment.
Female butterflies were selected for 10 generations. For the color composition experiment, 1056 females were measured and randomly split into two groups and 40 females were randomly drawn from each group to establish the UC lines. Within each group, the remaining females were ranked and the 40 females with the most extreme phenotypes in each direction were selected, so that each selection line was replicated twice in total. For the size experiment, 2254 females were measured and randomly split into two groups after 40 females were drawn to establish the first UC line . All selected females were placed with ~50 randomly chosen males and allowed to lay eggs. Rearing conditions for the stock population and selected lines have been described previously . We maintained similar selection intensities in both experiments for generations 1–10: each line, per generation, we measured 140–240 females for color composition (mean ± SE: 209 ± 5) and 15–200 females for size (mean ± SE: 173 ± 3). We selected 40 females per line every generation in the color experiment; in the size experiment the number decreased to 35 females between generations 5–10 .
Each generation, trait values for each replicate line were calculated relative to their respective UC lines. A least-squares regression of selection response on cumulative selection differential was used to estimate the rate of response to selection for each replicate. We used nested analysis of covariance (ANCOVA) to test for differences in the rate of response between selection directions, with 'cumulative selection differential' as the covariate and 'replicate within direction' as a random effect. Tukey's multiple comparisons were used to compare responses of individual replicates when there was a significant interaction between the two main factors in the nested ANCOVA. Analyses were conducted using PROC GLM in SAS version 6.12 (SAS Institute, 1996).
Induction of ectopic eyespots
Damage to pupal wings in the first 18 hours after pupation induces formation of ectopic eyespots on the dorsal forewing  and can be used to assay the threshold response of wing tissue to eyespot-inducing signals [46, 53]. To determine the effect of selection for eyespot color composition on this threshold response, we damaged the pupal wings of females in each of the Black-Black, Gold-Gold, and UC lines and measured the color composition of the resulting ectopic eyespots. The experiment was conducted during generation 14 (lines were maintained under random mating conditions after generation 10, with 40 females and 55 males selected at random and ~250 offspring reared per line per generation).
To reduce the size of native eyespots (A and P) on the dorsal forewing and increase visibility of the induced ectopic, we first pierced the foci of the anterior and posterior eyespots on the left dorsal forewing of each individual 4.5 hours after pupation with a finely sharpened tungsten needle . We then induced ectopic eyespots by piercing each of the left forewings at a site in the fourth wing cell (immediately distal to the normal location of the eyespots) 17 hours after pupation. We returned operated pupae to 27°C, froze newly emerged adults (after their wings had fully hardened), and measured the total diameter and inner black diameter of ectopic eyespots and the interpupil distance between the reduced A and P eyespots on each manipulated (left) wing. We also measured the total diameter and black disc diameter of P on the right dorsal forewing, and the interpupil distance on this unmanipulated (right) wing.
We used ANCOVA to test for differences in color composition of ectopic eyespots among Black-Black, Gold-Gold, and UC lines. With 'total ectopic diameter' as the covariate, we examined the fixed effect of 'selection direction (control, black, or gold)' and the random effect of 'replicate line nested within direction' on the size of the ectopic black disc. None of the interaction terms (selection direction by covariate and random effect by covariate) were significant and we used Tukey's HSD to test for differences in the adjusted means between Black-Black, Gold-Gold, and UC lines.
Estimates of genetic covariation in the stock population
We used a paternal half-sib breeding design  to estimate quantitative genetic parameters in our outbred stock population. We randomly selected 100 virgin males from the stock population at adult eclosion; each male was mated sequentially to 2 virgin females. At hatching, ~30 eggs per female were transferred to mesh rearing cages and fed on young maize plants ad libitum until pupation. Full-sib offspring were reared together but densities were kept low to minimize interaction and competition between individuals. Rearing cages were moved every 4 days to randomize environmental effects within the growth chamber. Emerging adult offspring were allowed several hours for their wings to expand and fully harden before being frozen for later analysis.
Five female offspring were randomly selected from each of 174 full-sib families (representing 87 sires who successfully produced offspring by two dams each) and dorsal forewing eyespots A and P, and ventral hindwing eyespots E4 and E6 were measured as described above. We used our nested breeding design (full-sib dam families nested within sire families) to estimate sire, dam, and progeny variance and covariance components in SAS 6.12. Genetic correlations between eyespot pairs (and their standard errors) were calculated according to ; to eliminate potential inflation of the estimates by maternal effects, genetic correlations were calculated from the among-sire variance components only .