Data for Scaling and developmental integration of a massively exaggerated secondary sexual sensory trait in the crane fly Leptotarsus costalis

Methods Samples and morphometric data L. costalis specimens were captured using an entomological net in swampy areas with patches of tall grass near Smiths Lake Field Research Station in early November 2023, and frozen at -80°C. The adult sex ratio was highly male-biased, resulting in uneven sample size for males vs. females. We dissected the flies by removing all legs, both wings and both antennae. We then imaged several traits of each female (N = 23) and male (N = 114) specimen, including the tibia length of fore-, mid-, and hind-legs, wing length, antenna length, head width, head length, thorax width and thorax length (Table 1; Fig. 2A-F). All sample appendages except for the female antenna were mounted on a glass petri dish with a 1 cm grid as a scale. An image was then taken using a digital camera (Olympus TG-6). The female antenna was mounted on a micrometer slide using a glycerol mounting medium and imaged using a microscope camera (Leica Flexacam C1) mounted on a Leica MZ16A stereoscope (Wetzlar, Germany), with gooseneck lights used to illuminate the samples. Table 1: Trait measurements taken from the samples and specifications of how measurements were conducted. Trait Description Foreleg Tibia From the femur-tibia joint to the tibiotarsal joint on the foreleg Midleg Tibia From the femur-tibia joint to the tibiotarsal joint on the midleg Hindleg Tibia From the femur-tibia joint to the tibiotarsal joint on the hindleg Wing From the main spine of the wing at the intersection of the most mesial wing-vein to the most distal apical vein margin Antenna From the tip of the scape (base of the pedicel) to the tip of the final flagellomere Head Width The maximum distance between the eyes as viewed from above Head Length From the base of the head to the base of the proboscis Thorax Width Between the lateral mesoscutal setae Thorax Length From the tip of the first thoracic tergite to the tip of the third thoracic tergite We measured all images in ImageJ (Schneider et al., 2012). If structures were curved, they were measured using a segmented line. Some females (N = 5) had to be discarded due to incomplete antennae. Repeatability of the trait measurements was calculated by re-measuring a random subset of males (N = 30) and females (N = 18, excluding females with missing or damaged antennae) 3 times from the same images. We quantified sensilla density for a subset of males (N = 30) and females (N = 8) to determine any differences in the sensilla density between sexes. Antenna microstructures were quantified using a Zeiss compound microscope (Axioskop 40), with a camera (DinoEye). DinoCapture 2.0 software (AnMo Electronics, 2016) was used to draw a rectangular area around a section of the antenna covering the entire width of the antennomere and counted the sensilla present within that area. Calibration was done using a stage micrometer. Statistical analysis We used R version 4.4.1 (R Core Team, 2024) for all statistical analyses. The lmer function from the package lme4 (Bates et al., 2015) was used to estimate trait repeatability, using a linear mixed model with a random effect of sample ID and no fixed effects. Repeatability was then calculated as the ratio of the variance component for sample ID (i.e., individual) to the total random effects variance. Repeatability was above 95% for all traits (Suppl. Table 1). To investigate the developmental integration of the antenna, we carried out Principal Components Analysis (PCA) on the correlation matrix for the morphological traits. Separate PCAs were carried out for each sex using the function prcomp() in base R. Developmental decoupling of the antenna would be reflected in low covariation (i.e., loadings of dissimilar magnitude) between the antenna and other morphological traits on the shape PCs (PC2….PC9)—in other words, low covariation between relative antenna length and the relative sizes of other morphological traits. As a quantitative measure of the degree of developmental decoupling of the male antenna, we calculated the absolute values of all trait loadings on the shape PCs (PC2...PC9). We then calculated the difference between the loading of antenna length and the mean loading of all other traits on each PC. A large difference indicates either that the antenna exhibits a large loading while other traits exhibit small loadings or vice versa, indicating low covariance between the antenna and other traits on that PC. Finally, we calculated the mean of these differences across all shape PCs to obtain the developmental decoupling value, which ranges between 0 (no developmental decoupling) and 1 (complete developmental decoupling). For comparison, we carried out a similar calculation for each of the other male traits and all female traits. To examine the robustness of these patterns of covariation, we used a jackknife approach to recalculate the male and female PCAs while sequentially excluding each individual sample. For each jackknife iteration, we recalculated developmental decoupling value as described above, and then obtained the jackknife mean (Meanjackknife) and standard error (SDjackknife) for this value. We calculated 95% confidence limits for the developmental decoupling value of each trait as Meanjackknife +/- 1.96 × SDjackknife. We also used the PCA results to identify a single-trait index of body size for static allometry analysis. PC1 of a morphological trait matrix represents body size, and the trait that loads most strongly on this PC provides the best single-trait summary of body size variation. We found that wing length loaded most strongly on PC1 in both sexes, and therefore used wing length in analyses of static allometry. OLS and SMA are two commonly used models for linear regression, and differ in that SMA assumes error on both axes, while OLS assumes that measurement error only occurs on the Y-axis (Warton et al., 2006). This leads to underestimation of slopes when using OLS on datasets where both axes have measurement error (Warton et al., 2006; Peig and Green, 2009). However, OLS is more appropriate if error on the Y-axis is at least 3-fold greater than error on the X-axis (McArdle, 1988). In allometric analyses, OLS and SMA are both used frequently. Hence, for comparison, we used both regression models. These regressions were carried out using log-transformed antenna length and wing length values. OLS analysis was carried out using base R, while SMA analysis was carried out using the package smatr (Warton et al., 2012). Confidence intervals for SMA plots were generated using a bootstrap method (1000 iterations).