Wing pointedness associated with migratory distance in common-garden and comparative studies of stonechats (Saxicola torquata)


Maude W. Baldwin, Department of Organismic and Evolutionary Biology and the Museum of Comparative Zoology, Harvard University, 26 Oxford Street, Cambridge, MA 02138, USA.
Tel.: 617 496 9389; fax: 617 495 5667; e-mail:


Migration promotes utilization of seasonal resources, and the distance flown is associated with specific morphologies, yet these relationships can be confounded by environmental factors and phylogeny. Understanding adaptations associated with migration is important: although migration patterns change rapidly, it is unclear whether migratory traits track behavioural shifts. We studied morphometrics of four stonechat populations representing a migratory gradient and raised under common-garden conditions. With multivariate analyses, we identified wing traits that differed clearly from general size trends, and used phylogenetic comparative methods to test the prediction that these traits correlated with migratory distance in captive and wild populations. Pointedness differed among populations, changed independently from overall body size, and was correlated with migration distance. Migration in stonechats may lead to deviations from allometric size changes, suggesting that birds may adapt morphologically to selection pressures created by their own behaviour in response to changing environmental conditions.


The extraordinary ability of birds to migrate long distances is facilitated by multiple traits, including morphological specializations that correlate with reduced energy expenditure in the wild (Bowlin & Wikelski, 2008). Comparative studies of migratory and sedentary birds have shown differences in both flight feather length and skeletal traits (Calmaestra & Moreno, 2000; Leisler & Winkler, 2003). Among the features of the wing and tail associated with the suite of aerodynamically important morphological traits called the migratory functional complex (Leisler & Winkler, 2003) wing pointedness in particular is implicated in long-distance flight performance (Mönkkönen, 1995), as pointed wings are aerodynamically favourable for prolonged, fast flight (Berthold, 1996; Hedenström, 2002; Bowlin & Wikelski, 2008). However, pointedness may come with trade-offs, such as a loss of manoeuvrability (Swaddle & Lockwood, 2003), and the rounded wings of many resident species and some migratory juveniles may help birds forage, perform courtship displays or avoid predators (Alatalo et al., 1984).

If wing pointedness conveys high benefits and costs, changes in migratory behaviour should be associated with morphological change. Migratory behaviour is often considered labile and in some instances can be gained and lost quite rapidly (Sutherland, 1998; Rappole et al., 2003; Newton, 2008). For example, in formerly sedentary house finches introduced to the eastern United States, migratory behaviour developed in only 60 generations (Able & Belthoft, 1998). Instances of reduced migratory behaviour are also common and are presumably associated with progressively milder winters (Newton, 2008). The selection pressures associated with increased or decreased migration may shape the wing in opposite ways: if rounded wings provide greater manoeuvrability than pointed wings, it is predicted that when species become more sedentary, their wings will evolve to become less pointed (Rolshausen et al., 2009). Species that increase migration distance, in contrast, will evolve more pointed wings, if, for instance, energy expenditure during migration is reduced and subsequent reproductive success is increased.

Migratory populations are particularly vulnerable to current global change (Sanderson et al., 2006) and their success at coping with environmental change will partly depend on their physical and physiological ability to sustain newly acquired migratory behaviour (Van Noordwijk et al., 2006). The time scale at which morphological change occurs depends in part on the degree to which relevant traits can be individually modified (Van Noordwijk et al., 2006). For example, wing pointedness requires the selective elongation or shortening of specific sets of flight feathers compared to others (primaries compared to secondaries), and can potentially evolve faster if the coupling of feather length to overall body size is weak. These possibilities can be explored using comparative phylogenetic analyses.

Interspecific (Mönkkönen, 1995; Leisler & Winkler, 2003) and intraspecific (Pérez-Tris & Tellería, 2001; Fiedler, 2005; Arizaga et al., 2006; Miláet al., 2008) studies have shown that, as predicted, migrants’ wings are generally more pointed than non-migrants’, although in some instances empirical data do not match these predictions (Mulvihill & Chandler, 1991; Mönkkönen, 1995; Fiedler, 2005). However, many other factors are known to affect wing morphology. For instance, wing length may depend on day length or nutritional conditions at the time when flight feathers were grown (Björklund, 1996; Dawson et al., 2000; Hall & Fransson, 2000) as well as on age (Fiedler, 2005). Such confounding factors can be controlled if birds representing different migration behaviours are raised under common-garden conditions (Van Noordwijk et al., 2006). Furthermore, phylogenetic relationships among taxa are known to often increase Type I error rates when phylogeny is ignored and standard statistical analyses are used (Rohlf, 2006). Recently, phylogenetic comparative methods have been employed in large-scale morphological studies of migratory and sedentary bird populations across genera (Mönkkönen, 1995; Calmaestra & Moreno, 2001) or across species within a genus (Marchetti et al., 1995; Voelker, 2001; Böhning-Gaese et al., 2003; Kaboli et al., 2007). Phylogenies for closely related species or subspecies can be difficult to resolve, making intraspecific comparisons challenging. Yet incorporating phylogenetic information is essential if the trait data are to be normalized for common descent (Freckleton et al., 2002; Garland et al., 2005).

Here, we capitalize on data from a long-term breeding program of four distinct populations of stonechats (Saxicola torquata) that migrate along a gradient of distances (Fig. 1) (Helm et al., 2005; Helm, 2009) and are the subject of recent molecular phylogenies (Illera et al., 2008; Woog et al., 2008; Zink et al., 2009). We used 16 morphological measures from birds that were raised and kept under identical (common-garden) conditions in traditional multivariate analyses to identify traits that deviated considerably from overall body size trends and could thus represent adaptations for migration. Then, we used phylogenetic comparative methods to look for evidence of correlated evolution of these candidate traits with migration behaviour. We tested whether the results from the captive populations can be generalized by adding wing measures from two field populations and one group of museum skins. One of the field populations, Saxicola dacotiae, has acquired residency on the Canary Islands (Fig. 1) (Illera et al., 2008) and thus offers an additional test for our hypotheses. Specifically, we aim to (i) assess the relative independence of wing traits from overall changes in body size; (ii) quantify the relationship between these size-independent wing traits and migration distance while accounting for phylogenetic relationships; and (iii) test the robustness of the association between morphological traits and distance flown by comparing birds from captive common-garden conditions to birds from the field. Furthermore, we used information on sex and age of birds to examine ideas about possible differences in selection pressures.

Figure 1.

 Phylogeny of the seven populations (Illera et al., 2008; Woog et al., 2008) and average migratory distance. The map represents the population sources and direction of migration. Triangles indicate resident populations. The tree is the majority rule consensus tree from a posterior distribution of trees generated in MrBayes.

Our study adds to the avian wing shape literature by combining data from common-garden and field populations with phylogenetic information. More broadly, it provides an example of covariation between an ecomorphological trait and an aspect of life history in vertebrates. In addition, it speaks to debates on allometry and selection (Phillips & Shine, 2006; Drake & Klingenberg, 2008) and contributes to our understanding of the role of behavioural changes in shaping the evolution of other aspects of the phenotype (Huey et al., 2003; Duckworth, 2006).

Materials and methods

Captive birds in a common-garden setting

We first examined an extended set of morphological measurements of stonechats (S. torquata) kept in aviaries in Andechs, Germany. The birds originated from four source populations (Helm, 2003, 2009; Helm et al., 2005) that had originally been considered subspecies: although recent taxonomic revisions suggest that some are distinct species (Wink et al., 2002), debate remains (del Hoyo et al., 2005; Illera et al., 2008). Migratory behaviour is markedly different among these groups (Table 1): African stonechats (S. t. axillaris) from equatorial Kenya (0°14′S, 36°0′E) are residents and defend territories year-round (Dittami & Gwinner, 1985), whereas Siberian stonechats (S. t. maura) from Kazakhstan (51.5°N, 63°E) are long-distance migrants that spend a large portion of the year migrating to and from wintering grounds in India, China and northeast Africa (Raess, 2008). Between these two extremes lie the European stonechats (S. t. rubicola) from Austria (48°14′N, 16°22′E) that migrate a comparatively shorter distance to northern Africa, and the British stonechats (S. t. hibernans) from Ireland (52°N, 10°W), a portion of whose population overwinters along the coast of Britain and Ireland (Helm et al., 2006). Both types of British stonechats were included here as they were not morphologically different. Birds were either collected as nestlings in the field or were offspring of these populations and hatched between 1998 and 2005 in outdoor aviaries in Germany. All young were taken from their nests after 5 days and raised on a standard diet (Gwinner et al., 1987). After completing their juvenile development, birds were housed in individual cages (Helm et al., 2005; Helm, 2009).

Table 1.   Average migratory distance flown by each population in the study.
Source population and species/subspecies nameDistance (km)References
Austria (Saxicola torquata rubicola)1700(Helm et al., 2006)
German (Saxicola t. rubicola)1700(Helm et al., 2006)
Irish (S. t. hibernans)1200(Helm et al., 2006)
Siberian (S. t. maura)2600(Raess, 2008)
Kenyan (S. t. axillaris)0(Dittami & Gwinner, 1985)
Canary Islands (S. dacotiae)0(Illera & Diaz, 2006)
Madagascar (S. t. sibilla)? (suspected resident or local migrant)(Keith et al., 1992)

In these birds, we measured 16 traits (Table 2) that are related to different functional complexes (Leisler & Winkler, 1985, 2003). The dataset consisted of 226 birds, including 28 African, 48 Austrian, 34 Irish and 17 Siberian stonechats for which we obtained a full complement of measures. One investigator (MWB) measured all birds, mostly over a period of 6 weeks in winter before moult, using methods employed in other extensive morphological studies (Fiedler, 1998; Leisler & Winkler, 2003, 2009). We recorded body mass and fat score, and measured tarsus and bill with digital callipers. We omitted bill and claw length because these parts are cut in captivity. Measures of flight feathers included those of the primaries (connected to the manus), those of the secondaries (connected to the ulna), and those of the rectrices (tail feathers). The length of the third primary (counted from the outside) was measured with a ruler with an attached pin; the wing chord and the distance from the shoulder to the first secondary were measured with a wing chord ruler. All other tail and wing measurements were taken with laminated millimetre paper. Primary distance measurements (the distance from the tip of the longest primary to the tip of the other primaries) were measured, in succession, on the folded wing (Lockwood et al., 1998). Distal distance is the distance between the tip of the longest primary and the outermost or first primary, which is a very short feather in stonechats. Kipp’s distance is the length from the tip of the first secondary to the tip of the longest primary, measured on the folded wing.

Table 2.   Means of size-adjusted characters (divided by the cube root of the mean body mass and log-transformed) for the four captive populations of stonechats.
Siberian femalesSiberian malesAustrian femalesAustrian malesIrish femalesIrish malesAfrican femalesAfrican males
Alula to tip of longest primary2.973.033.002.992.942.952.952.96
Beak height0.380.380.380.340.370.360.390.35
Beak width0.630.600.620.580.560.540.670.63
Digit 31.651.651.691.671.671.641.631.63
Distal distance2.592.692.592.612.542.572.492.52
Inner notch, 2nd primary2.
Kipp’s distance1.891.941.871.851.841.841.621.64
Length of 3rd primary (counted from outside)
Outer notch, 3rd primary2.362.412.332.322.272.292.312.33
Carpal joint (shoulder) to tip of first secondary3.
Tail gradation0.110.220.390.
Tail length3.003.042.972.962.922.933.003.00
Wing length3.303.353.

We analysed the data from the common-garden birds with principal component analysis (PCA) and discriminant analysis to extract major axes of covariation and to compare by sex and population (Leisler & Winkler, 1985; Lockwood et al., 1998) before adjusting for size. We then analysed size-adjusted data with discriminant analysis to remove the general isometric increase in characters with body mass by log-transforming the trait value divided by the cube root of the mean body mass for each population and sex (Leisler & Winkler, 1985). Using body mass is problematic, as it can fluctuate with condition and between seasons, but this strategy is nevertheless preferable to using the tarsus as a structural measure of size (Fiedler, 2005).

Three traits were identified by multivariate procedures as deviating from general size trends and subsequently examined in phylogenetic comparative analyses. In a preliminary comparative analysis on the common-garden populations, we took phylogeny into account while estimating the correlation between morphometric traits and migration by using phylogenetic generalized least squares (PGLS) in the program BayesTraits (Pagel, 1997, 1999). The program estimates the extent to which phylogeny accounts for the variance and covariance among trait values, given as the parameter λ (phylogenetic signal). λ may assume values between 0 and 1. Under a random walk model of evolution, when λ = 1, the amount of expected covariance among trait values, is proportional to the length of the branches shared, from the most recent common ancestor to the root. When λ = 0, the data covary independently of the phylogeny, which is tantamount to assuming a polytomy, although variance is still expected to be proportional to the total path length from root to tip for a given taxon. Before using maximum likelihood methods to test for trait correlation in the common-garden dataset, we estimated the maximum likelihood (ML) value of λ, a key parameter of the model. Disparate combinations of evolutionary rate and process can produce similar patterns in phylogenetic signal (Revell et al., 2008). Moreover, different traits can evolve under different evolutionary rates and processes (Freckleton et al., 2002) making a priori assignments of λ difficult to justify. Using a tailored value of λ is important: traits correlated merely because of common ancestry would appear spuriously associated if phylogenetic information were ignored, but for other traits that covary independently of the tree, accounting for phylogeny will increase error rates on average (Freckleton et al., 2002).

To decide which value of λ to use, we determined the strength of ML evidence for the estimate of λ with nested likelihood ratio tests (LRT = 2*(log-likelihood1– log-likelihood2), which has a χ2 distribution with degrees of freedom equal to the difference in the number of parameters). The ML estimate of λ was used in the subsequent trait-correlation model when the P-value was < 0.05; otherwise, 1, the default, was used. As there is no clear biological explanation for values greater than 1, we used 1 in these cases as well. Then, to test for trait correlation, the log-likelihoods from two runs of the random walk model were compared, the first with trait correlation allowed and the second with trait correlation forced to zero.

The phylogenetic framework used in the comparative analyses of the four common-garden populations above was based on two recent Saxicola phylogenies (Illera et al., 2008; Woog et al., 2008). To re-create a tree with branch lengths, we generated an unrooted ML tree using Phylip 3.67 (F84 model of sequence evolution) (Felsenstein 1989) and cytochrome b sequence data obtained from a previous study (Illera et al., 2008).

Evolutionary trends in an extended set of populations

To create our extended dataset, we added data from three populations of wild birds. Stonechats on the Canary Islands (S. dacotiae) on Fuerteventura are site-faithful (Illera & Diaz, 2008) and live in arid, seasonal habitats at mid-latitudes (28°46′N, 14°31′W; (Illera & Diaz, 2008). In the fall of 2008, Juan Carlos Illera measured Kipp’s distance, wing length, distal distance and body mass from nine males and five females. Stonechats in Germany (51°N, 6°3′E; S. t. rubicola) have been under extensive field study since 1976 (Flinks et al., 2008) and are fully migratory (Helm et al., 2006). Heiner Flinks collected data on Kipp’s distance, wing length and distal distance from 21 males and 17 females in the summer of 2008, and as individual mass measures were not available, we used the mean body mass from the same population during summer (Flinks et al., 2008). From the literature, we recorded mean Kipp’s distance, wing length, and body mass from stonechats from Madagascar (19°S, 46°E; S. t.sibilla) (Woog et al., 2008). Unfortunately, no detailed information exists about migration in this population but the birds are thought to be resident or local migrants (Keith et al., 1992). Data were collected from skins (with sample sizes of between 12 and 40 birds, depending on the measurement), but field data suggest similar dimensions in the wild population (F. Woog, personal communication).

Before subjecting the combined dataset of field and wild birds to phylogenetic comparative analyses, we examined variation within and among populations on the original data. We used linear mixed models to test for correlations with measures of size (such as body mass and wing length) and for possibly confounding effects of age, sex and their interactions with population. Comparisons were made over all birds, both wild and captive, (n = 278), excluding those from Madagascar for which age information was not available. We grouped birds into young (wearing their first set of flight feathers, i.e., up to the onset of the first post-nuptial moult in yearlings) and adult (any age afterwards). Results are given as Wald statistics, which are asymptotically distributed like a chi-square distribution. We found widespread effects of age and sex. Therefore, we used data from adults only for more detailed analyses and treated the sexes separately.

For the comparative analyses on the dataset with the seven populations, we combined the sequence data used in recent phylogenies (Illera et al., 2008; Woog et al., 2008) and downloaded data for two mitochondrial genes, cytochrome B (CYTB) and NADH dehydrogenase subunit 2 (ND2) for the seven species of stonechats from GENBANK. We built a tree that included five additional species (two stonechats, from Réunion and Tanzania; Saxicola caprata, pied bushchat; Saxicola rubetra, whinchat and, as an outgroup, Myrmecocichla cinnamomeiventris, mocking cliff chat). Both sequences were present for all species, except for birds in the German population; for them, only CYTB was available. We aligned the sequences in Clustal X (Higgins & Sharp, 1988; Larkin et al., 2007), concatenated them using BioEdit (Hall, 1999), and generated a posterior distribution of trees in MrBayes (Huelsenbeck & Ronquist, 2001). We used the GTR model and ran the program for 10 million generations, sampling every 1000 trees. We removed samples from the first 2.5 million generations as burn-in, pruned the five outgroup species and created a majority rule consensus tree (Fig. 1).

Based on this phylogeny, we subjected the data of the six populations with known migration distances (excluding Madagascar) to Bayesian phylogenetic regression analyses. We regressed average distance flown against the average of each of the three identified wing traits for each population, using the ‘Continuous’ program in BayesTraits. Here, we ran the Markov chain Monte Carlo (MCMC) simulations for 100 million generations and sampled every 1000 generations. To determine which value of λ should be used, we first compared the harmonic mean of the log-likelihoods of the regression when λ was set at the default value of 1 to the harmonic mean when λ was allowed to be estimated every iteration. We calculated the Bayes factor by multiplying by two the difference between the last harmonic mean of the log-likelihoods from each run. A Bayes factor > 2 is considered positive evidence, and > 5 is strong evidence (Raftery, 1996). We only used the estimated λ when the Bayes factor favoured it with a value larger than 5. An approximation of the significance of the regression was determined by plotting the posterior distribution of the β parameter values (the slope of the regression of the log of migratory distance on the mean size-adjusted trait) and considering the fraction of β values that cross zero (the null hypothesis). The posterior distribution was visualized in JMP (SAS, 1989–2007).

To investigate how controlling for size affected the results, we used non-size-adjusted data. We first ran a multiple regression on the six populations used above with migration distance as the dependent variable and both Kipp’s distance and body mass as independent variables. We also added data from the Malagasy population, a distinct but close relative to the Kenyan population, to guard against the possibility that our Kenyan population was an outlier. As solid data on migration in the Malagasy population are lacking, instead of including S. t. sibilla in the regression of trait and migratory distance, we explored how the relationships between Kipp’s distance and wing length or body mass were changed by the addition of the Malagasy data.


Captive birds in a common-garden setting

The stonechat populations differed in overall size, as illustrated by body mass (Fig. 2). Kenyan resident stonechats had the highest, and Siberian long-distance migrants had the lowest body mass. Before controlling for size, the first two axes of the PCA (Table S1A) on the means of the populations of the 127 captive birds with complete sets of measurements described 89% of the total variation. Interestingly, the first axis, interpreted as representing overall size, had a negative relation to Kipp’s distance and a very low (although still positive) relation to distal distance. Thus, although populations loaded on PC1 according to overall size, Kipp’s distance was distinguished from the other measurements, even in the non-size-adjusted data.

Figure 2.

 Body mass (g), separated by sex, of the four groups of captive stonechats.

The non-size-adjusted discriminant analysis on the 127 captive birds (Table S1B) produced two axes that together explained 88% of the total variation. The first axis, which explained 76% of the total variance, separated the larger Kenyan birds from the northern populations (Fig. 3), mainly because of the contrast between multivariate size and Kipp’s distance. The second axis (12%) separated both the sexes and the Siberian birds with migrants and males scoring higher. It correlated positively to the greatest extent with distal distance, and to some degree with Kipp’s distance and wing length. These two axes deviated clearly from an orthogonal relationship corresponding to a correlation (cosine) of −0.56. In the size-adjusted discriminant analysis (Table S1C, Fig. 4), the first axis accounted for 53% of the total variance and correlated positively primarily with Kipp’s distance and distal distance, and most negatively with wing width and tail gradation. The second axis (33%), almost orthogonal to the first, was explained by wider, longer and notched wings (Table S1C).

Figure 3.

 Graph of non-size-adjusted discriminant analysis. The first two axes explain 88% of the total variation. The first axis (multivariate size) separates the larger African birds (G, H) from the other populations and explains 76% of the total variance. Migrants and males scored higher on the second axis (12%), which correlated with distal distance, Kipp’s distance, and wing length.

Figure 4.

 Graph of size-adjusted canonical discriminant analysis. The first axis (53%) correlates primarily with Kipp’s distance and distal distance, and the second axis (33%) is explained by wider, longer and notched wings.

These combined analyses identified Kipp’s distance, distal distance and wing length as traits that differed most clearly from general size trends in the captive birds, and therefore as candidates for the selective effects of migration distance on character size. We examined these traits for correlations with migratory behaviour (Table 3), scoring residents as 0, shorter distance and partial migrants, 1, and long-distance migrants, 2. The degree of phylogenetic signal (λ) varied among these traits and between the sexes (Table 3). The mean Kipp’s distance for each category correlated significantly with migratory score in males (P < 0.001; n = 4) and in females (P = 0.029; n = 4). Distal distance showed the same pattern (P < 0.001 for males, P = 0.011 for females). Wing length was significantly correlated with migration scores for males (P = 0.043), but not for females (P = 0.139).

Table 3.   Likelihood ratio tests for the estimate of the λ parameter for the four captive populations, and for the test for correlation between the trait and migratory score. The likelihood ratio statistic and the associated P-value (obtained from a chi-square distribution) are shown. The maximum likelihood estimate of lambda was used when it was significantly different and smaller than the default value (1). Most traits showed significant correlation with migratory score.
SexTraitML estimate of λLikelihood ratio statistic (λ different than 1)P-value for estimate of λλ (implemented in model)Test for correlation (likelihood ratio statistic)Test for correlation (P-value)

Evolutionary trends in an extended set of populations

We tested the robustness of the correlations between migration and wing measures by adding data from free-living Canary Island and German stonechats (Figs 1 and 5). In the extended dataset before accounting for phylogeny, body mass was significantly correlated with wing length (r = 0.64; P < 0.001; n = 278) and to lesser degree, distal distance (r = 0.33; P < 0.001; n = 263), but not with Kipp’s distance (r = 0.02; P = 0.779; n = 276). Similarly, when analysed within the populations, only Kipp’s distance was independent of measures of size such as mass or wing length. Kipp’s distance was correlated with distal distance only (r = 0.50; P < 0.001; n = 261). Consequently, distal distance was associated with wing pointedness as well as with overall size.

Figure 5.

 Overview of trait sizes of seven populations of stonechats. Box-plots show body mass, Kipp’s distance, and Kipp’s distance adjusted using body mass, for adult males (left) and females (right). Labels on x-axis identify populations. For the Malagasy birds, mean trait values are taken from the literature (Woog et al., 2008).

We also tested for confounding effects of sex and age class on the traits under study. Body mass was higher in adult than in young stonechats (Wald1 = 15.6; P < 0.001) and higher in males than in females (Wald1 = 5.3; P = 0.021; interactions not significant, P > 0.1). Before controlling for size, wings were much longer in old compared to in young birds (Wald1 = 64.0; P < 0.001), and in males compared to in females (Wald1 = 149.6; P < 0.001). The populations differed slightly in the way age differences, but not sex differences, affected wing length (interaction age and population: Wald5 = 13.0; P = 0.024). After controlling for size, only the difference between the sexes (Wald1 = 15.2; P < 0.001), not between age groups, persisted (Wald1 = 0.0; P = 0.864). Kipp’s distance was longer in males than in females (raw data: Wald1 = 18.9; P < 0.001; size-adjusted: Wald1 = 7.5; P = 0.006), but age groups did not differ (P > 0.6). Sex influenced Kipp’s distance differently in the populations (interaction sex and population: Wald5 = 18.7; P = 0.002). Distal distance was greater in males than females (raw data: Wald1 = 119.3; P < 0.001; size-adjusted: Wald1 = 49.4; P < 0.001), and in adults than in juveniles (raw data: Wald1 = 59.5; P < 0.001; size-adjusted: Wald1 = 9.9; P = 0.002; interactions not significant, P > 0.25).

For the application of phylogenetic comparative methods, we constructed a consensus tree (Fig. 1). This tree agreed with the previous phylogenetic analyses of Saxicola. The only node with < 95% posterior support separated the German population from the Austrian and Irish population. As this node had only 51% posterior support, we collapsed it to a polytomy (Fig. 1).

Migratory distance evolved in close association with wing shape characteristics. Results from the Bayesian estimation of PGLS of size-adjusted data from the six populations supported the correlations seen in the subset of the captive birds. In the regression, distance flown was strongly correlated with Kipp’s distance for both sexes (r2 = 0.81 for males, r2 = 0.52 for females; Table 4A) and with distal distance, although to a lesser degree (r2 = 0.59 for males, r2 = 0.31 for females). For wing length, however, the regression explained a very low proportion of variance (r2 = 0.05 for males, r2 = 0.09 for females). Figure S1 shows the Bayesian posterior distribution of the regression’s slope parameter β (mean ± SD = 11.6 ± 4.0) for Kipp’s distance in males. The regression is significant: of 100 001 MCMC generations sampled, only 774 of the runs produced a value of β less than zero (P = 0.008). The regression was also significant, but less so, for Kipp’s distance in females (P = 0.05), and distal distance in males (P = 0.036) but not females (P = 0.139), nor for wing length for either sex (males, P = 0.352; females, P = 0.280). In the aforementioned regression with the extended population set, we used λ values of 1, as estimating λ for each run was not supported (Table 4A). However, as the earlier test for trait correlation in the common-garden birds had indicated that females had a lower value of λ (Table 3) and thus a lower amount of the covariance was accounted for by the phylogeny, we ran additional regression analyses for both sexes with an estimated λ to see how the values compared. Allowing λ to be estimated during the MCMC (Table 4B) increased the r2 and the significance, most drastically for the Kipp’s distance in females (r2 = 0.84) and distal distance in females (r2 = 0.61).

Table 4.   Bayesian analysis of the regression of log(migratory distance) on mean size-adjusted morphological trait for the four common-garden birds and the two field populations. To decide whether to estimate λ, the harmonic means were compared, and a Bayes factor was calculated: λ was estimated if the Bayes factor was > 5. As the program estimates λ every iteration, a distribution of λ values is produced, there is not a single estimate of λ per run, as in the maximum likelihood analyses. The posterior distribution of beta (the slope of the regression) was calculated from 100 001 MCMC simulations. The mean and standard deviation were calculated from the distribution, and the P-value is approximated by the percentage of the distribution that crosses zero (the null hypothesis). (A) Test for λ value and regression values when λ = 1. (B) Values when λ is estimated during every iteration.
TraitHarmonic mean (Default, λ = 1)Harmonic mean, λ estimatedBayes factorλMean of betaStandard deviation of betar2# Posterior beta < 0P-value (# beta < 0)/100 001
Female Kipp−13.472−12.370−2.20517.0564.6780.52250000.050
Female Wing−13.181−14.1461.93116.60712.2700.091280370.280
Female Dist−13.011−14.1182.213110.04310.4550.306139330.139
Male Kipp−10.763−11.1430.761111.5693.9980.8057740.008
Male Wing−14.528−15.8902.72414.31912.8610.046352270.352
Male Dist−11.313−12.3702.115118.9649.5420.58935850.036
Traitr2P-value (# beta < 0)/100 001
Female Kipp0.8400.007
Female Wing0.1460.280
Female Dist0.6130.037
Male Kipp0.8910.002
Male Wing0.0590.338
Male Dist0.6410.029

Multiple regression of migratory distance on non-size-adjusted Kipp’s distance and body mass were carried out on data from males only, revealing a tight association (R2 = 0.70); in addition, the ML test for trait correlation was highly significant (P < 0.001). The mean slope parameter (β) for Kipp’s distance was 0.76 ± 0.37 (SD) and for body mass was −0.26 ± 0.18. Using the approximations of significance described earlier, the slope of Kipp’s distance crossed zero 2122 times (P = 0.021), and the slope for body mass crossed 4704 times (P = 0.047) (Fig. S2A,B). Adding the data for the second African population from Madagascar did not substantially change the trends we saw in the other six populations (Table 5). For instance, in the non-size-adjusted regression of Kipp’s distance on body mass, the results were similar both with and without the Malagasy birds (excluding: r2 = 0.30; including: Malagasy r2 = 0.33).

Table 5.   Bayesian analysis of the regression of the non-size-adjusted means of two traits, for all populations with and without the Malagasy birds included. Trait one is the dependent variable; trait two is the independent variable. The posterior distribution of beta (the slope of the regression) was calculated from 100 001 iterations of the MCMC simulation. The mean and standard deviation were calculated from the distribution, and the P-value is approximated by the percentage of the distribution that crosses zero.
Malagasy included?Trait 1Trait 2r2Mean of betaStandard deviation of beta# simulations where posterior distribution of λ < 0P-value
NoKippwing length0.1820.3640.379145900.146
YesKippwing length0.1990.3530.333120800.121
NoKippbody mass0.2970.2620.293135770.136
YesKippbody mass0.3260.3000.24788890.089


Our analysis of migratory traits in stonechats combines data from a common-garden study, from the field, and from molecular phylogenetics. The findings support the hypothesis that traits may respond to selection driven by a change in behaviour such as avian migration. Wing pointedness, an aspect of morphology relevant to migration (Mönkkönen, 1995; Bowlin & Wikelski, 2008), correlates closely with migratory distance and has evolved independently of overall body size, whereas no such trends were apparent for wing length. Our study provides clear answers to questions regarding the independence of traits from changes in body size, the relationship between these size-independent traits and migration distance, and the robustness of this relationship in regard to differences between captive common-garden and field conditions. In addition, our data on effects of sex and age supports current ideas of different selection pressures. We discuss our results in the following three main points.

1. Relative independence from size trends: the Kipp’s and distal distances deviated considerably from allometric change, as determined by classical multivariate analyses of 16 morphometric traits in four stonechat populations that were kept under common-garden conditions (Figs 3 and 4). Kipp’s distance (i.e., the distance from the longest primary to the first secondary) is of known aerodynamic importance (Mönkkönen, 1995; Bowlin & Wikelski, 2008; Miláet al., 2008), whereas the role of distal distance (i.e., distance from the longest to the smallest, outermost primary) in aerodynamics is less clear, but it may affect flight performance via aspect ratio or via changes in the leading edge of the wing. The increase in pointedness in stonechats as migratory distance increased provides strong evidence for non-isometric change in migration traits, apparent even in non-size-adjusted data: for instance, in the large African residents, Kipp’s distance was considerably shorter than in the small Siberian migrants (Fig. 5). Recent studies of other passerine taxa (Fiedler, 2005; Arizaga et al., 2006; Miláet al., 2008) have also documented more pointed wings in migratory subspecies; however, in these studies, migrants were also heavier, and absolute Kipp’s distance increased with body size.

Some researchers argue that in birds, phenotypes are generally conserved and allometric change is prevalent (Merilä & Björklund, 1999; Wink et al., 2002). In contrast, our stonechat data demonstrate non-allometric changes in body shape. Wing pointedness clearly correlates with migratory distance, and like known adaptive characters (e.g. beak dimensions (Merilä & Björklund, 1999)), appears to be decoupled from allometric change. In addition to the trends across populations, we also found evidence that wing pointedness is independent from size within populations. As Kipp’s distance did not correlate with body mass within most populations, it may be an evolutionarily labile and unconstrained trait (Merilä & Björklund, 1999; Van Noordwijk et al., 2006). Our findings also demonstrate that the first principal component may differ substantially from an ideal isometric size vector. Kipp’s distance loaded relatively high on this axis with a negative value, indicating that, in addition to size, the first component may contain significant shape information.

2. Persistent relationship between migratory traits and migration distance after accounting for phylogeny: the relationship between some wing measures and migratory distance became even clearer when we examined data from seven populations of stonechats using comparative phylogenetic analysis of size-adjusted traits (Figs 1 and 5). In regression analyses, variation in migratory distance was associated with a high proportion of variance in Kipp’s distance (r2 between around 0.5 and 0.8), with less variation in distal distance (r2 between 0.3 and 0.6), but with almost no variation in wing length (r2 between 0.05 and 0.1). Stonechats on the Canary Islands are highly resident but may have evolved sedentary lifestyles independently from African stonechats. Our prediction that Kipp’s and distal distances would be smaller (causing the wing to be more rounded) in the resident Canary Island stonechats was upheld: both the size-adjusted and non-size-adjusted measures were very close to those in the sedentary Kenyan birds, even though the Canary Island birds are nested within the migratory Paleartic clade (Fig. 1) (Illera et al., 2008). These trends in Kipp’s distance agree for the most part with studies across multiple genera of birds (Mönkkönen, 1995; Calmaestra & Moreno, 2001) as well as across closely related species groups that have incorporated phylogeny (Marchetti et al., 1995; Voelker, 2001). Kaboli et al. (2007) recently investigated museum specimens of 22 species of male wheatears (Oenanthe spp.) for which phylogenetic data were available for 12 Palearctic species (Aliabadian et al., 2007). Similarly, they found that a roundedness index (the distance from the first primary to the wingtip, divided by wing length) negatively correlated with migratory score. The wheatear study showed clear phylogenetic effects, as does the present study on stonechats, both in contrast to a recent comparative study of blackcaps (genus Sylvia) (Böhning-Gaese et al., 2003).

However, the sex differences in our estimate of λ (phylogenetic signal) (Tables 3 and 4) raise interesting questions about the interpretation of λ for these traits. Although Bayesian methods did not reject using the default value of 1 (i.e., random walk and strong influence of phylogeny) for either sex (Table 4A), the correlation in females between measures of pointedness and distance flown is far stronger when we use λ < 1. Female stonechats have shorter distal and Kipp’s distances: even after controlling for their overall smaller body size, they have more rounded wings than males. If pointedness is selectively advantageous for migration and under no other constraint, no sex difference would be expected in size-adjusted trait values or in λ. Sex differences in wing morphology have been noted in other species and were associated with differences in either distance (Fernández & Lank, 2007) or speed of migration (Dierschke et al., 2005), but in stonechats, the sexes winter in pairs and their migration patterns are not known to differ (Helm et al., 2006). A similar, unexplained sex difference was found in another small passerine species, the bluethroat (Arizaga et al., 2006).

One possible explanation is that the male aerial flight displays (Urquhart, 2002) could contribute to differences in wing shape. Pointedness could aid males in flight displays, a behaviour which across pipits, another songbird (genus Anthus), is correlated with migratory tendency (Voelker, 2001). Like stonechat females, pipit females also exhibited a lower degree of phylogenetic signal than males did in some wing measures (Voelker, 2001), but like in stonechats, Kipp’s distance remained significantly associated with migration in both sexes and across all tests. This emphasizes the difficulty of interpreting the biological meaning of phylogenetic signal as an indicator of certain types of evolutionary rate or even process (Revell et al., 2008). The differences in λ in a trait that we expected to be primarily influenced by migration (and therefore sex independent) show that multiple selection pressures may be implicated in the evolution of wing pointedness. Many studies confine analyses to a single sex, yet examining sex differences may be a useful way to tease apart contributions of different evolutionary processes shaping the wing.

3. Persistent relationships between migratory traits and migration distance in the field and captivity: the present study demonstrates a close relationship between wing pointedness and migration distance in four stonechat populations raised under common-garden conditions. As all birds were raised and most even bred in captivity, these findings are likely not a consequence of raising environment and imply genetic differences between populations. The strong correlation between migration distance and Kipp’s distance was further confirmed when additional data from free-living birds were included (Figs 1 and 5). Figure 5 shows empirical sizes of body mass and Kipp’s distance for adults of all seven study populations, as well as size-adjusted Kipp measures. The three field populations were substantially less heavy than captive birds, but, as observed in the captive populations, Kipp’s distance deviated clearly from body mass patterns. Even prior to size-adjustment, Kipp’s distance was much longer in the migratory German population than in the resident Canary Island birds and in the (presumably-resident) Malagasy stonechats. After controlling for size, measures of Kipp’s distance from captive and field birds were in good agreement.

In addition to the main research questions, we also examined differences between age groups. Juveniles are predicted to have more rounded wings than adults to improve their chances of escaping from predators (Alatalo et al., 1984; Pérez-Tris & Tellería, 2001). We examined our data to test this prediction and those of a subsequent model (Miláet al., 2008) that predicts that short-distance migrants would show a larger age difference in roundedness than do residents and long-distance migrants, as in long-distance migrants the selective pressure for sustained flight at a lower cost would override benefits gained from increased manoeuvrability. However, in stonechats, Kipp’s distance did not differ as predicted between age groups across all populations. In contrast, distal distance was generally shorter in juveniles, and differences between the age groups corresponded closely to the trend that would be anticipated (Miláet al., 2008) (Fig. 6), if distal distance has aerodynamic benefits for long-distance flight.

Figure 6.

 Age differences in size-adjusted Kipp’s distance and distal distance. (a) Kipp’s distance, sexes pooled; (b) distal distance, sexes pooled. Populations are ordered along the x-axis according to migration distance.

Overall, the correlation in stonechats of wing shape measures with distance flown but not with measures of size provides strong evidence that some of the characters important for migration are independent, and thereby support the idea that migratory behaviour can be labile and may evolve in multiple ways (Winkler & Leisler, 2004). Winkler and Leisler suggest that the transition between sedentary and migratory populations can be accomplished with low selective costs because of a lack of strong correlations among key traits. Flexibility and an easy interchange between sedentary and migratory strategies are suggested in stonechats by recent behavioural data on Zugunruhe (migratory restlessness) (Helm, 2006; Helm & Gwinner, 2006) and by studies of other species in which low genetic distances between migrant and resident populations point to the rapid acquisition of new migratory behaviour (e.g. Pérez-Tris et al., 2004). However, the rate at which changes in wing shape track changes in migratory behaviour is less clear. Morphological changes in other avian traits such as beak shape have been shown to respond rapidly to environmental changes (Grant & Grant, 1995). Data from house finches suggest that wing pointedness might evolve relatively rapidly: proximal primary lengths in an eastern population that had become partially migrant within 60 generations were slightly shorter (causing a more pointed wing) than those in western populations, whereas no difference in overall wing length was seen (Egbert & Belthoff, 2003). Recently, a study of blackcaps (Rolshausen et al., 2009) demonstrates a decrease in migration distance over the last 30 generations associated with a change in multiple morphological traits, including pointedness. Consistent with these data, but over a longer period of time, wing pointedness in stonechats appears to track the loss of migratory behaviour: the Canary Island birds with very rounded wings are estimated to have diverged from the common Palearctic ancestor around 1.6 mya (for divergence time estimates of the other stonechat groups, see (Illera et al., 2008)). However, not all passerine birds studied always show this trend, although it is difficult to disentangle the relative roles of time or gene flow. Whereas within Europe migratory and sedentary blackcaps differed in wing pointedness, sedentary Atlantic Island blackcaps showed neither genetic differentiation (Pérez-Tris et al., 2004) nor changes in pointedness (Fiedler, 2005) when compared to migratory European populations. Similarly, the Irish and the Austrian stonechats used in our study showed no genetic divergence at the two loci used, due either to gene flow or recency of divergence. Although Irish stonechats differed from the Austrians in many traits related to overall body size, there was no difference in size-adjusted Kipp’s distance.

How rapidly organisms can morphologically adapt to changes in their environment is an area of active investigation (Hendry & Kinnison, 1999; Hendry et al., 2008). Shifts in dispersal patterns or migratory behaviour are predicted to be potentially important instigators of evolutionary change (Duckworth, 2009), and there is evidence that at least some species of birds are likely to complement rapid changes in migration by changes in key migratory traits. Our results corroborate recent work (Rolshausen et al., 2009) indicating that wing pointedness, important for efficient flight (Bowlin & Wikelski, 2008) is a labile trait that responds readily to selection. In view of the many threats to migratory birds, including shifts in timing or range of prey items at breeding grounds (Both & Visser, 2005) or predicted increases in Afrotropical desert barriers (Sanderson et al., 2006; Barbet-Massin et al., 2009), rapid adaptive changes in morphology because of selection imposed by changes in behaviour could be crucial for future survival under changing environmental conditions.


M. W. Baldwin thanks the group of Biological Rhythms and Behaviour at the Max Planck Institute for Ornithology and the Department of Organismic and Evolutionary Biology at Harvard University for support during the project. J.C. Illera and H. Flinks kindly captured and measured Canary Island and German stonechats. S.V. Edwards and Y. E. Stuart gave very helpful feedback on drafts of this work. We thank B. Leisler, W. Fiedler and K-H. Siebenrock for help and advice during the initial discussion of measurements and Emily Wheeler for editorial assistance. Three anonymous referees and Armando Caballero provided thoughtful comments which greatly improved the manuscript.