A total of 398 adult butterflies emerged from the rearing treatments (N: low = 206, high = 192). As expected, all body size measurements were highly correlated with each other (wingspan to body length: r = 0.66, wingspan to forewing length: r = 0.71, body length to forewing length: r = 0.75). Thus, we chose a single measurement, wingspan, to assess the morphological response to rearing temperature. In contrast to our expectation based on the temperature-size rule, high rearing temperature individuals had larger wingspans than did low rearing temperature individuals (mean ± 1 SE (mm): low = 124.12 ± 0.69, high = 159.17 ± 1.13; Fig. 1, t = 169.13, p < 2 x 10− 16).
We found statistically significant interaction effects in all models with any aspect of flight behavior as a response. The addition of a random effect of flight trial did not substantially improve the performance of models testing for a rearing x testing temperature interaction (speed: DAICc = 2.1, model weight = 0.74; time: DAICc = 2, model weight = 0.74; distance: DAICc = 1.2, model weight = 0.64; in all cases, the model with the lower AIC did not include the random effect), thus we proceeded with simpler GLMs. For the response of flight speed, butterflies in the low/low and high/high treatments flew significantly slower than did the individuals who experienced different rearing and testing treatment temperatures (GLM: rearing temperature, F = 8.57, P = 0.004, testing temperature, F = 30.15, P = 7.17 x 10− 08, interaction, F = 113.57, P < 2.2 x 10− 16; mean ± 1 SE (cm/s): low/low: 3.61 ± 0.06, high/high: 4.08 ± 0.19, low/high: 7.01 ± 0.31, high/low: 5.27 ± 0.21; Fig. 2). When we decomposed speed into its two components (time flying and distance traveled), we also observed statistically significant interaction effects, and all treatment combinations were different from each other (Figs. 3, 4).
However, butterflies from the low/low and high/high treatments flew longer (GLM: rearing temperature, F = 0.06, P = 0.81, testing temperature, F = 25.80, P = 5.84 x 10− 7, interaction, F = 378.84, P < 2.2 x 10− 16; mean ± 1 SE (seconds): low/low: 1,166.85 ± 19.54, high/high: 1,058.25 ± 38.41, low/high: 542.04 ± 20.35, high/low: 663.50 ± 23.50; Fig. 3) and for greater distances (GLM: rearing temperature, F = 155.33, P < 2.2 x 10− 16, testing temperature, F = 57.33, P = 2.61 x 10− 13, interaction, F = 1265.70, P < 2.2 x 10− 16; mean ± 1 SE (cm); low/low: 4,093.00 ± 14.53, high/high: 3,689.48 ± 32.87, low/high: 3,192.82 ± 17.24, high/low: 3,060.72 ± 17.79; Fig. 4) than individuals in the other treatments. Thus, butterflies that experienced matching rearing and testing temperatures, whether low or high, flew further and for longer than animals from the other treatments (Figs. 3, 4), but at lower speeds (Fig. 2).
Because animals reared at the higher temperature were larger than those reared at the lower temperature, we then ran a related set of models in which the categorical variable “rearing temperature” was replaced with the continuous variable “wingspan,” which allowed us to examine the role of body size itself (wingspan) on movement behavior. For this set of models, inclusion of the random effect of flight trial did generally improve model performance (speed: DAICc = 0.1, model weight = 0.53; time: DAICc = 12.5, model weight = 0.99; distance: DAICc = 125.2, model weight > 0.99; in all cases, the model with the lower AIC included the random effect). Similar to the previous set of analyses, we found statistically significant interaction effects for all movement responses. We found a statistically significant effect of the interaction between wingspan (which results from rearing temperature, Fig. 1) and flight testing temperature on speed (GLMM: wingspan, F = 2.80, P = 0.10, testing temperature, F = 43.22, P = 8.18 x 10− 10, interaction, F = 35.12, P = 2.07 x 10− 8, Fig. 5).
However, within each of the four rearing/testing temperature combinations, wingspan did not obviously influence speed (Fig. 5). The time spent flying was also affected by the interaction between wingspan and testing temperature (GLMM: wingspan, F = 1.87, P = 0.17, testing temperature, F = 22.83, P = 3.43 x 10− 6, interaction, F = 18.74, P = 2.29 x 10− 5, Fig. 6). Again, while there was variation due to temperature treatments, within those treatments, time flying was relatively consistent across wingspans (Fig. 6), with the exception of “high/high” animals, who showed a negative relationship between wingspan and time in flight. Finally, we observed that distance traveled was also affected by the interaction between wingspan and testing temperature (GLMM: wingspan, F = 0.19, P = 0.66, testing temperature, F = 8.83, P = 0.003, interaction, F = 5.95, P = 0.02, Fig. 7), but again, wingspan per se did not affect distance traveled within the treatment groups.