A tale of winglets: evolution of flight morphology in stick insects

The evolutionary transition between winglessness and a full-winged morphology requires selective advantage for intermediate forms. Conversely, repeated secondary wing reductions among the pterygotes indicates relaxation of such selection. However, evolutionary trajectories of such transitions are not well characterized. The stick insects (Phasmatodea) exhibit diverse wing sizes at both interspecific and intersexual levels, and thus provide a system for examining how selection on flight capability, along with other selective forces, drives the evolution of flight-related morphology. Here, we examine variation in relevant morphology for stick insects using data from 1100+ individuals representing 765 species. Although wing size varies along a continuous spectrum, taxa with either long or miniaturized wings are the most common, whereas those with intermediate-sized wings are relatively rare. In a morphological space defined by wing and body size, the aerodynamically relevant parameter termed wing loading (the average pressure exerted on the air by the wings) varies according to sex-specific scaling laws; volant but also flightless forms are the most common outcomes in both sexes. Using phylogenetically-informed analyses, we show that wing size and body size are correlated in long-wing insects regardless of sexual differences in morphology and ecology. These results demonstrate the diversity of flight-related morphology in stick insects, and also provided a general framework for addressing evolutionary coupling between wing and body size. We also find indirect evidence for a ‘fitness valley’ associated with intermediate-sized wings, suggesting relatively rapid evolutionary transitions between wingless and volant forms.

SWD and SSD. For example, if selection on male-biased mobility and on female-biased 126 fecundity were coupled, we might expect an inverse correlation between SWD and SSD. We 127 accordingly assess overall patterns of sexual dimorphism among phasmid species within 128 phylogenetic and allometric contexts. Scaling of wing loading 158 The loss of aerodynamic capability was assessed using wing loading, the ratio of body weight to 159 total wing area. We sampled total wing area (Aw), body mass (m) and L from 23 males and 21 160 females of field-collected and captive-bred insects from 36 species (Supplementary Dataset 2). 161 Digital images were obtained dorsally for insects placed on horizontal surfaces with all legs 162 laterally extended; projected areas of fully unfolded wings were manually extracted using 163 Photoshop (CS6, Adobe Systems Inc., San Jose, CA, USA). AW, Lw and L were measured using 164 ImageJ (Schneider et al., 2012). The scaling of wing loading (pw) with Q was analyzed for both 165 sexes. First, we examined the allometric scaling of body mass based on the formula: Bayesian phylogenetic reconstruction 181 We used three mitochondrial genes ( A total of five morphological traits was used in phylogenetic analyses (L and Q of both sexes 219 and sex-averaged L, ΔQ, and ΔL). First, we calculated the phylogenetic signals (λ) for all 220 characters using the maximum-likelihood approach implemented in Phytools (Pagel, 1999;221 Revell, 2012). This model was compared with alternative models where λ was forced to be 1 or 0 222 in order to find the best-fitting model. The best-fitting model was found using the likelihood ratio whereby the better fitting model has the highest log-likelihood score, Lh (Pagel, 1997(Pagel, , 1999227 Freckleton et al., 2002). When λ = 0, this suggests trait evolution is independent of phylogenetic 228 association, which is equivalent to generalized least square (GLS) model. We also assessed the 229 evolutionary contexts of morphological traits with maximum-likelihood ancestral state 230 reconstruction using 'fastAnc' function in Phytools (Revell, 2012).

231
For the species that lacked molecular data, we added them as polytomous tips to the node 232 representing the latest common ancestor on the tree. We then generated 100 random trees with 233 randomly resolved polytomous tips. Each new node was added using the function 'multi2di'

256
Sex-specific variation in flight-related morphology 257 Among all sampled insects, ~44% of females and ~51% of males were winged. Relative wing 258 size (Q) varied continuously from complete winglessness (Q = 0) to fully-sized wings (i.e., Q ≈ 259 0.85; Fig. 3a,b). For both sexes, the relative frequency of Q was bimodally distributed with a continuously varying SWD and female-biased SSD (Fig. 3g). For A. tanarata group, the 276 reduction in coefficients of wing and body size toward higher altitudes was sex-specific ( Fig.   277 3d,e). Males showed a relatively higher extent of wing reduction, leading to a reversal of SWD 278 from male-to female-biased (Fig. 3g).

280
Sex-specific flight reduction 281 Scaling of wing area with wing length was nearly isometric, with an exponent (b) of 282 approximately 1.84 in both sexes (Fig. 4a). The allometric scaling of insect mass with respect to 283 body length was, however, sex-dependent, with females exhibiting a higher slope coefficient and where `= X / Y is the ratio of the slope coefficient between two sexes. Given that `= 2.89 297 and = 1.84 (Table 1), then rQ equals 1.78, suggesting that QF should be 78% greater than 298 9 QM to attain the same wing loading (Fig. 4e). This outcome may partially underlie the high 299 frequency of female-biased SWD found in long-winged taxa (see Discussion).

301
Variation in wing loading can also be presented as a three-dimensional landscape relative 302 to wing and body size. The allometric effect is stronger in females, whereas males exhibit a 303 smaller lower boundary for wing loading (Fig. 5a,b). Projecting the species richness distribution 304 onto these landscapes demonstrates clustering of taxa on the wing loading functional landscape 305 (Fig. 5c,d). Both sexes showed two major clusters associated with low and high wing loadings, 306 corresponding to long-winged and miniaturized-wing morphologies, respectively. The majority 307 of long-winged females were allometrically constrained to values of wing loading between 10 -308 0.5 Nm -2 < pw < 1 Nm -2 , whereas long-winged males clustered near a value of 10 -1 Nm -2 , with a  Fig. S4).

327
Based on PGLS results, there was a significant inverse correlation between Q and L in 328 long-wing insects (Q > 0.33) of both sexes (Fig. 7a,b; Table 3), which supported our initial 329 hypothesis on evolutionary coupling between wing and body size. In addition, sex-averaged 330 body size was coupled with the extent of both SWD and SSD in long-wing species (Q > 0.33 in 331 both sexes), suggesting opposite trends of variation in SWD and SSD along the gradient of sex-332 averaged body size (Fig. 7c). An exemplar of this correlation is demonstrated in Fig. 7d and e, 333 whereby increases in SSD and SWD both lead to greater sexual differences in wing loading.

334
Short-wing insects generally lacked significant correlation between wing and body size ( Fig.   335 7a,c). Across all winged species, variation in female traits contributed more substantially to 336 intersexual differences, as shown by the predominant roles of female Q and L values in 337 determining variation in SWD and SSD, respectively (Supplementary Fig. S6, Table S2).

342
Most winged phasmid species possess either small or large wings (Fig. 3b) mating. The inverse correlation between wing and body size in male stick insects (Fig. 7a,b) 379 suggests that selection for flight has limited the evolution of larger body size. Similar selection In winged phasmids, SSD and SWD are significantly correlated but not within either short-411 or long-wing species ( Fig. 7c; Supplementary Fig. S3), reflecting the interaction between 412 multiple selective forces within sex-specific ecological contexts (Fig. 8a). The evolutionary 413 intercorrelation between SSD and SWD is generally underexplored for most other insects.    Fig. 5a and Fig. 5b, wing loading relationships for females and males, 642 respectively between the two sexes; females typically have higher wing loading than males and 643 stronger allometric effects relative to body length. Fig. 5c and Fig. 5d, contours of the wing 644 loading landscape for females and males, respectively, as overlaid with hexagonal bins for 645 species counts (Fig. 3d); wing loading distribution differs substantially between the sexes.   Table 3.  Tables   677  678  679  680   Sex  C1  C2  a  b Male -2.08 (0.14) -0.24 (0.03) 1.84 (