1. Ecology
  2. Evolutionary Biology

Complex plumages spur rapid color diversification in kingfishers (Aves: Alcedinidae)

  1. Chad M Eliason  Is a corresponding author
  2. Jenna M McCullough
  3. Shannon J Hackett
  4. Michael J Andersen
  1. Grainger Bioinformatics Center, Field Museum of Natural History, United States
  2. Negaunee Integrative Research Center, Field Museum of Natural History, United States
  3. Department of Biology and Museum of Southwestern Biology, University of New Mexico, United States
https://doi.org/10.7554/eLife.83426
Share this article
Cite this article
  1. Chad M Eliason
  2. Jenna M McCullough
  3. Shannon J Hackett
  4. Michael J Andersen
(2023)
Complex plumages spur rapid color diversification in kingfishers (Aves: Alcedinidae)
eLife 12:e83426.
https://doi.org/10.7554/eLife.83426
  2. Download BibTeX
  3. Download .RIS

Abstract

Colorful signals in nature provide some of the most stunning examples of rapid phenotypic evolution. Yet, studying color pattern evolution has been historically difficult owing to differences in perceptual ability of humans and analytical challenges with studying how complex color patterns evolve. Island systems provide a natural laboratory for testing hypotheses about the direction and magnitude of phenotypic change. A recent study found that plumage colors of island species are darker and less complex than continental species. Whether such shifts in plumage complexity are associated with increased rates of color evolution remains unknown. Here, we use geometric morphometric techniques to test the hypothesis that plumage complexity and insularity interact to influence color diversity in a species-rich clade of colorful birds—kingfishers (Aves: Alcedinidae). In particular, we test three predictions: (1) plumage complexity enhances interspecific rates of color evolution, (2) plumage complexity is lower on islands, and (3) rates of plumage color evolution are higher on islands. Our results show that more complex plumages result in more diverse colors among species and that island species have higher rates of color evolution. Importantly, we found that island species did not have more complex plumages than their continental relatives. Thus, complexity may be a key innovation that facilitates evolutionary response of individual color patches to distinct selection pressures on islands, rather than being a direct target of selection itself. This study demonstrates how a truly multivariate treatment of color data can reveal evolutionary patterns that might otherwise go unnoticed.

Editor's evaluation

This important work advances our understanding of the factors that affect the speed of colour evolution in birds and the resulting diversification patterns. It provides compelling evidence that more complex plumage coloration can lead to rapid colour evolution in kingfishers, and will pave the way for more comprehensive analyses that fully embrace the multidimensional nature of colour variation. Hence, the results will be of broad interest to ornithologists and evolutionary biologists in general.

https://doi.org/10.7554/eLife.83426.sa0

eLife digest

Birds are among the most colorful animals on Earth. The different patterns and colors displayed on their feathers help them to identify their own species, attract mates or hide from predators.

The bright plumages of birds are achieved through either pigments (such as reds and yellows) or structures (such as blues, greens or ultraviolet) inside feathers, or through a combination of both pigments and structures. Variation in the diversity of color patterns over time can give a helpful insight into the rate of evolution of a species. For example, structural colors evolve more quickly than pigment-based ones and can therefore be a key feature involved in species recognition or mate attraction.

Studying the evolution of plumage patterns has been challenging due to differences in the vision of humans and birds. However, recent advances in technology have enabled researchers to map the exact wavelengths of the colors that make up the patterns, allowing for rigorous comparison of plumage color patterns across different individuals and species.

To gain a greater understanding of how plumage color patterns evolve in birds, Eliason et al. studied kingfishers, a group of birds known for their complex and variable color patterns, and their worldwide distribution. The experiments analyzed the plumage color patterns of 72 kingfisher species (142 individual museum specimens) from both mainland and island populations by quantifying the amount of different wavelengths of light reflecting from a feather and accounting for relationships among species and among feather patches.

The analyzes showed that having more complex patterns leads to a greater accumulation of plumage colors over time, supporting the idea that complex plumages provide more traits for natural or sexual selection to act upon.

Moreover, in upper parts of the bodies, such as the back, the plumage varied more across the different species and evolved faster than in ventral parts, such as the belly or throat. This indicates that sexual selection may be the evolutionary force driving variation in more visible areas, such as the back, while patterns in the ventral part of the body are more important for kin recognition.

Eliason et al. further found no differences in plumage complexity between kingfishers located in island or mainland habitats, suggesting that the isolation of the island and the different selection pressures this may bring does not impact the complexity of color patterns. However, kingfisher species located on islands did display higher rates of color evolution. This indicates that, regardless of the complexity of the plumage, island-specific pressures are driving rapid color diversification.

Using a new multivariate approach, Eliason et al. have unearthed a pattern in plumage complexity that may otherwise have been missed and, for the first time, have linked differences in color pattern on individual birds with evolutionary differences across species. In doing so, they have provided a framework for future studies of color evolution. The next steps in this research would be to better understand why the island species are evolving more rapidly even though they do not have more complex plumage patterns and how the observed color differences relate to rapid rates of speciation.

Introduction

Understanding spatial and temporal trends in phenotypic diversity continues to be an important challenge in evolutionary biology. Colorful signals in birds are a good case study for a rapidly evolving phenotype that shows variation at broad spatial (Cooney et al., 2022) and phylogenetic scales (Cooney et al., 2019). Birds produce colorful plumage patterns with a combination of two mechanisms: light absorption by pigments and light scattering by feather nanostructures (Shawkey and D’Alba, 2017). Whereas melanin- and carotenoid-based coloration are produced by chemical pigments and absorb light waves, structural colors are produced by the physical interaction of light waves and nanometer-scale variations in the feather integument (Prum, 2006). Within birds, structural colors produce a wide array of color, including blue-green colors, glossy blacks, and iridescence. Because structural colors are more evolutionarily labile than pigment-based colors, they have faster evolutionary rates (Eliason et al., 2015) and are considered key innovations in some clades (e.g., African starlings; see Maia et al., 2013b). In addition to how color is produced, birds also vary in where they deploy colors in their plumage (Stoddard and Prum, 2011). Yet, studying color pattern evolution has been historically difficult due to our inability to perceive UV color (Eaton, 2005) and challenges with quantifying and analyzing complex color patterns (Mason and Bowie, 2020).

While the color of individual patches can be influenced in different directions by multiple selective factors (Cuthill et al., 2017), the deployment of color in distinct patterns appears to be constrained developmentally (Hidalgo et al., 2022). Since selection can only act on existing variability, such as distinct plumage patches across a bird’s body, ancestrally shared developmental bases of plumage patterns might act as a brake on color evolution (Price and Pavelka, 1996; Hidalgo et al., 2022; but see Felice et al., 2018). For example, in a hypothetical, uniformly colored species with strong developmental constraints that limit independent variation in color among patches, selection on the color of any single patch would cause the whole plumage to change in tandem. By contrast, if a species is variably colored (i.e., patchy, and therefore has a more complex plumage) with few constraints on the direction of color variation for different patches, selection can act on different aspects of coloration (Brooks and Couldridge, 1999). On a macroevolutionary scale, we would predict greater color divergence in a clade with an ancestrally complex plumage pattern because there is more standing color variation among patches upon which selection can act. On a more microevolutionary scale, however, intraspecific plumage complexity (i.e., the degree of variably colored patches across a bird’s body) could be a key innovation that drives rates of color evolution in birds and should be considered alongside ecological and geographic hypotheses.

Islands have been considered natural laboratories for studying evolution because they often lack natural predators and competitors due to their geographic isolation (Losos and Ricklefs, 2009). Compared to life history (Covas, 2012; Losos et al., 1998; Novosolov et al., 2013), behavior (Buglione et al., 2019; Roff, 1994), and morphological traits (Clegg and Owens, 2002; Wright et al., 2016), signals used in mating and social contexts have been less commonly explored in the context of island evolution. Yet, previous work has shown increased color polymorphism in island snails (Bellido et al., 2002; Ożgo, 2011) and lizards (Corl et al., 2010). Within birds, island species tend to be less sexually dimorphic and have simpler songs (Price, 2008). Island birds have also been shown to have darker colors and simpler plumage patterns (Bliard et al., 2020; Doutrelant et al., 2016). Under a species-recognition hypothesis, these shifts are thought to be driven by reduced competition on islands (Martin et al., 2015; Doutrelant et al., 2016), as fewer competitors would lower the risk of hybridization and cause a reduction in signal distinctiveness on islands (West-Eberhard, 1983). Despite these advances, we lack a detailed understanding of color evolution within, rather than between, island and mainland clades. For example, are changes in plumage color on islands also accompanied by bursts in phenotypic evolutionary rates, as has been shown for morphological traits in other groups (Millien, 2006; Thomas et al., 2009; Woods et al., 2020)?

Two hallmarks of kingfishers (Aves: Alcedinidae) are their complex plumage patterns (Eliason et al., 2019) and their island distributions (Andersen et al., 2018; McCullough et al., 2019). Kingfishers encompass a wide variety of colors—from the aquamarine-colored back of the common kingfisher (Alcedo atthis) to the brilliant silver back of the southern silvery-kingfisher (Ceyx argentatus), as well as the purple rump of the ultramarine kingfisher (Todiramphus leucopygius). They also run the gamut of plumage complexity, including intricate scalloped plumage of the spotted kingfisher (Actenoides lindsayi) and the contrasting hues of the black-backed dwarf-kingfisher (Ceyx erithaca). The family is widely distributed across the globe, but their center of diversity is the Indo-Pacific, including island clades in Wallacea and Melanesia that have recently been highlighted for their high diversification rates (Andersen et al., 2018). These same island clades, specifically within the woodland kingfisher genus Todiramphus and Ceyx pygmy-kingfishers, also have elevated color diversity (Eliason et al., 2019) and complex geographic histories. These genera include many allopatric, island-endemic taxa, as well as harboring a high degree of sympatry on islands (Andersen et al., 2015). For example, there are 10 species of kingfishers that occur on the Indonesian island Halmahera, 5 of which are in the genus Todiramphus. There are also multiple instances of sympatry in Ceyx, including on New Guinea, the Philippines, and the Solomon Islands (Andersen et al., 2013). Smaller population sizes, isolation, and genetic drift could potentially explain high rates of color evolution in island kingfishers, making them an ideal system to investigate the interplay between key innovations (complex plumages) and geographic isolation (i.e., spatial opportunity) in driving rapid color evolution.

In this study, we implement geometric morphometric techniques to investigate complex plumage pattern evolution across kingfishers. We hypothesized that potential constraints limiting where and how color is produced on a bird’s body should also limit evolutionary changes between species. Specifically, if complex plumages are a key innovation enabling rapid rates of color evolution (Prediction 1), and if plumage complexity is lower on islands (Prediction 2), then insularity and plumage complexity should both influence the direction and rate of change of plumage coloration (Prediction 3; see Figure 1B). We tested these predictions using UV–vis reflectance spectrophotometry of museum specimens and multivariate comparative methods. Our study of the interplay between the arrangement of color patches, interspecific competition, and geography sheds light more broadly on the role of spatial opportunity in phenotypic evolution.

Figure 1 with 2 supplements see all
Download asset Open asset
Illustrative guide to methods used to study kingfisher plumage coloration.

(A) Flow chart depicting our process: (1) spectrophotometry of 22 plumage patches on closed-wing museum specimens, (2) conversion of data to tetrahedral colorspace coordinates, and (3) different ways we analyzed these data across the kingfisher phylogeny. We analyzed how individual patch colors evolved using multivariate comparative methods (3a). To estimate complexity at the intraspecific level (3b), we calculated three different metrics for each tip in the phylogeny: average pairwise distance among color patches (metric c1); the color volume (i.e., range) of all plumage patches in colorspace (metric c2); and the number of contiguous color patches that would be perceived as the same color by a bird (metric c3). We calculated interspecific rates of overall plumage color evolution using multivariate rate tests (3c). (B) We predicted faster rates of color evolution on islands (blue line) and in species with more complex plumages. Yet, there are examples of cases in which this relationship may be reversed (e.g., see insets showing species pairs with simple plumages and diverse colors, left, as well as complex plumages and similar colors, right). Illustrations created by Jenna McCullough.

Results

Holistic assessment of plumage color variation

Plumage coloration is highly multivariate, varying both within feathers, among feather regions on a bird, between sexes, and among species. To visualize trends in these data, we conducted a partitioning of variance analysis that revealed two distinct modes of color variation within kingfishers: (1) clades with complex color patterns that partition color variance more among patches than among species or individuals (e.g., Corythornis, Alcedininae) and (2) clades that vary more among species (e.g., Todiramphus, Cerylininae; Figure 2B). Chromatic variation among sexes was negligible for most clades (Figure 2B). Evolutionary rates of color were unevenly distributed across the body, with dorsal regions evolving faster than ventral ones (Figure 2C). This differs from several previous studies illustrating rapid rates of ventral plumage evolution in tanagers (Shultz and Burns, 2017), manakins (Doucet et al., 2007), fairy-wrens (Friedman and Remeš, 2015), and antbirds (Marcondes and Brumfield, 2019). This could indicate that dorsal plumage patches are under stronger sexual selection in kingfishers, as rapid rates of display trait evolution are thought to be associated with more intense sexual selection (Irwin et al., 2008; Seddon et al., 2013; Merwin et al., 2020). Rump, cheek, and throat patches showed the highest levels of phylogenetic signal (Figure 2D), suggesting that these patches are more taxonomically informative than crown or wing plumage coloration. To visualize major axes of variation in overall plumage color patterning, we used a phylogenetic principal components analysis (pPCA), with per-patch color coordinates as variables (N = 66). We plotted the first two pPC scores that together accounted for >50% of color variation in the clade, revealing extensive color pattern variation in the group (Figure 3; see Figure 3—figure supplement 1 for non-phylogenetic PCA results).

Figure 2
Download asset Open asset
Perceptually uniform colorspace and color variation in kingfishers.

(A) Color data, with points being the average of three plumage patch measurements for each individual (N = 3101). Colors are estimated from a human visual system using spec2rgb in pavo (Maia et al., 2013a). Distance between patches is proportional to the just noticeable differences (JNDs), assuming a UV-sensitive visual system (Parrish et al., 1984). (B) Proportional color variance among patches in an individual (violet), between sexes in a species (orange), and among species in a clade (green). Low variation between sexes was further confirmed with a multivariate phylogenetic integration test (r-PLS = 0.88, p < 0.01). Clades with more complex plumages (e.g., Alcedininae) tend to have a higher proportion of among-patch variation. (C) Distribution of multivariate evolutionary rates and (D) phylogenetic signal of color evolution across the body (darker colors indicate higher values).

Figure 3 with 1 supplement see all
Download asset Open asset
Color pattern morphospace of kingfishers.

Bird images show depictions of color in a human visual system based on spectral measurements over a grid of phylogenetic principal components analysis (pPCA) coordinates. Axes shown are pPC axes 1 and 2, together accounting for >50% of plumage color variation in the clade.

A novel approach for estimating plumage complexity

To test our hypothesis that intraspecific plumage complexity facilitates interspecific color divergence, we required species-specific estimates of plumage color complexity. For each species, we calculated plumage complexity for both chromatic (i.e., hue and saturation) and achromatic components (i.e., lightness) of plumage patches in three ways: (1) as the mean pairwise distance among all patches in colorspace; (2) as the color volume (or lightness range for achromatic plumage components; see Methods) enclosing all points for a species, and (3) as the number of uniquely colored contiguous patches on the body, assessed using just noticeable differences (JNDs >1 threshold) for a folded-wing plumage configuration (see Figure 1A, section 3b). The latter two metrics are similar to a recent method (Eliason et al., 2019) of calculating color complexity of plumages as the number of contiguous body regions sharing the same color mechanism (e.g., melanin-based or structural coloration), but they are based on continuous reflectance values instead of discrete color data (i.e., presence or absence of a given color mechanism). With this metric, higher differences between adjacent patches yielded higher plumage complexity scores (see Figure 1A, section 3b). Estimates of plumage complexity were strongly correlated among different complexity metrics for chromatic components of plumage coloration, but less so for achromatic variation (Figure 1—figure supplement 1).

Species with complex plumages have higher rates of color evolution

Plumage complexity of an individual bird and interspecific differences in coloration are typically thought of as distinct axes of color diversity. Yet, species that have evolved several patches have more degrees of freedom to vary, potentially leading to faster rates of color evolution among species. However, this is not necessarily the case, as there are examples within kingfishers that show simple plumages yet high color divergence, as well as complex plumages with little evolutionary divergence (Figure 1B). Here, we attempt to link plumage complexity with interspecific rates of color variation using multivariate approaches typically only applied in the field of geometric morphometrics. To determine rates of overall plumage evolution, we used a recent time-calibrated phylogeny (McCullough et al., 2019) that incorporated thousands of ultraconserved elements (Faircloth et al., 2012) and fully sampled the avian order Coraciiformes (kingfishers, bee-eaters, rollers, and allies). Next, using multivariate color data, we estimated species-specific multivariate rates of evolution using the R package RRphylo v. 2.6.3 (Castiglione et al., 2018). Because we predicted that insularity results in faster rates of plumage color evolution, we included insularity as a covariate in our phylogenetic analyses. Comparing species-specific rates of plumage color evolution with intraspecific complexity metrics, we found that rates of color evolution were higher in species with more complex plumages (Figure 4B, Table 1; see Supplementary file 1b for sex-specific results). For achromatic variation, body mass and lightness range (c2) significantly explained increases in rates, but folded-wing achromatic complexity (c3) did not (Table 1). Although complexity metrics were correlated (Figure 1—figure supplement 1), variance inflation factors (VIFs) were not extreme (all <5), and phylogenetic generalized least squares (PGLS) results were stable after dropping each complexity variable from the reduced models (Figure 4—figure supplement 1). These results were further confirmed using a well-established multivariate method for comparing lineage-specific rates (Denton and Adams, 2015) based on binary complexity scores (Supplementary file 1c; see Methods for details).

Figure 4 with 3 supplements see all
Download asset Open asset
Species with complex plumages have faster rates of color evolution.

(A) Phylogeny showing evolution of plumage color complexity, with edge colors corresponding to ancestral states of plumage complexity (mean interpatch color distance within a species, corresponding to metric c1 in Figure 1) and edge lengths proportional to color evolutionary rates (see legend). Tip colors correspond to different island systems (see legend in B), with continental species in gray. (B) Significant relationship between color evolutionary rates and plumage complexity (p < 0.01). Effect of island-dwelling (p = 0.02) is indicated by line type (dashed: continental, solid: island species). See Table 1 for statistical results and Figure 4—figure supplement 3 for results with analyzing achromatic variation in plumage.

Table 1
Plumage complexity predicts rates of color evolution among species.

Models were fit using PGLS in the phylolm R package, with species-specific evolutionary rates as the response variable and complexity metrics (c1, c2, and c3), island-dwelling, natural log body mass, and number of sympatric species as predictors. The best-fitting models were determined using a stepwise AIC-based procedure using the phylostep function in phylolm. Significant predictors in the most parsimonious models are indicated in bold. See Supplementary file 1b for sex-specific results and Figure 4—figure supplement 1 for results with the full model and alternate submodels.

ResponsePredictorEffectpλR2
Chromatic rateMean interpatch distance (c1)0.41 ± 0.13<0.010.000.45
# unique patches (c3)0.24 ± 0.130.07......
Insularity0.49 ± 0.200.02......
Achromatic rateLightness range (c2)0.31 ± 0.11<0.010.000.14
ln body mass0.23 ± 0.110.05......

Taken together, our findings are consistent with the idea of multifarious selection providing more axes for ecological or phenotypic divergence in complex color signals among species, and can eventually lead to speciation (Nosil et al., 2009). However, recent work in wolf spiders has revealed that signal complexity per se can be a direct target of sexual selection (Choi et al., 2022). Another possibility in kingfishers is that body size is driving the evolution of plumage complexity, as signal complexity has been shown to decrease with body size in iguanian lizards (Ord and Blumstein, 2002) and in passerine birds (Cooney et al., 2022). Interestingly, the kingfisher species with the most complex plumages are also among the smallest birds in the family, the pygmy-kingfishers, such as the indigo-banded kingfisher (Ceyx cyanopectus) and southern silvery-kingfisher (C. argentatus, Figure 4A). We found some support for this hypothesis, as most chromatic complexity metrics were significantly lower in large-bodied species, whereas achromatic complexity was not linked to body size (Table 2). An alternative hypothesis is that species on islands have more complex plumages, and therefore insularity is indirectly driving color divergence. However, plumage complexity metrics were not significantly divergent between islands and mainland taxa (Figure 4—figure supplement 2, Table 2), suggesting that insularity and plumage complexity are independent drivers of color variation in the group.

Table 2
Predictors of plumage complexity.

Models were fit for both chromatic (i.e., hue and saturation) and achromatic variables (i.e., plumage lightness) using PGLS in the phylolm R package. Different complexity metrics (see Figure 1 for details) were set as the response variable, and island-dwelling, ln body mass, and number of sympatric species were used as predictors. The best-fitting models were determined using a stepwise AIC-based procedure (i.e., using the phylostep function). Significant predictors are indicated in bold. See Supplementary file 1d for sex-specific results.

Data typeResponsePredictorEffectpλR2
ChromaticInterpatch dist. (c1)ln mass−0.34 ± 0.140.020.410.08
Color volume (c2)ln mass−0.33 ± 0.140.020.180.07
# unique patches (c3)ln mass−0.31 ± 0.150.050.510.05
AchromaticLightness range (c2)Insularity−0.60 ± 0.270.030.000.11
Lightness range (c2)# symp. species−0.32 ± 0.120.01......
# unique patches (c3)Insularity−0.41 ± 0.260.120.150.03

Island kingfishers have higher rates of color evolution

Colonization of islands is expected to result in shifts in both the direction of phenotypic change (i.e., convergent evolution when species colonize islands) and also the magnitude of change (i.e., elevated rates of phenotypic evolution on islands versus the mainland; Millien, 2006). To test these ideas, we first evaluated whether islands act as distinct selective regimes that drive convergent change toward particular colors using a distance-based PGLS (d-PGLS) approach developed for morphometric data (Adams, 2014a), but suitable for color data as well. Results of this analysis showed weak support for the prediction that island colonization has caused convergent evolution of color (p = 0.09; Figure 5, Table 3), while lightness showed no significant difference between mainland and island species (p = 0.59; Table 3). This is distinct from a previous study showing predictable trends toward darker plumages on islands (Doutrelant et al., 2016). However, we did find that achromatic complexity was significantly lower on islands (Table 2).

Figure 5
Download asset Open asset
No support for convergence of color patterns on islands.

Phylogenetic principal components analysis (pPCA) plot with points colored by continental (gray) and island species (see legend). Distance-based PGLS analyses suggest island and mainland species are not significantly different in plumage coloration (F = 1.84, p = 0.09). See inset for interpretation of pPC values and Table 3 for full statistical results.

Table 3
Multivariate plumage color is not significantly different on islands.

Results of multivariate distance-based PGLS (d-PGLS) tests testing for convergence in overall plumage coloration on islands. Both chromatic (i.e., hue and saturation) and achromatic plumage variables (i.e., lightness) were considered. p values were calculated with a permutation approach using 999 iterations. See Figure 5 for details and Supplementary file 1e for sex-specific results.

ResponsePredictorFpNtraitsNspecies
Multivariate colorInsularity1.840.096672
# sympatric species0.350.94......
In body mass1.140.32......
Multivariate lightnessInsularity0.830.592272
# sympatric species1.140.32......
In body mass1.670.11......

To test for an ‘island effect’ on rates of color evolution, we treated individual patches as geometric morphometric ‘landmarks’ and compared multivariate evolutionary rates between insular and continental species using rate ratio tests (Denton and Adams, 2015). When considering the island effect alone on rates of color evolution, we found that species distributed on islands have faster rates of color evolution (σcont2 = 0.13, σisland2 = 0.23, p = 0.02; Figure 4A) but similar rates of light-to-dark evolution compared to continental species (σcont2 = 0.84, σisland2 = 0.91, p = 0.72). To further test the possibility that the observed rapid color evolution on islands is the result of reproductive character displacement occurring within islands (e.g., see Losos and Ricklefs, 2009), we included the number of sympatric kingfisher species as a predictor in our PGLS models. The number of sympatric lineages ranged from 1 to 9 on islands, and 6–38 for mainland taxa (see Dryad). Neither overall plumage color patterns (Table 3) nor rates of plumage evolution (Table 1) were significantly associated with the number of sympatric species. Thus, rather than interspecific competition driving color diversity, intraspecific competition or genetic drift may instead be driving rapid rates of color evolution in island kingfishers.

Discussion

We lack a cohesive understanding of how plumage color patterns evolve in birds. This study is the first attempt to link intraspecific color variation among patches to interspecific color variation among species. We find support for higher rates of plumage evolution in clades with more complex plumages (Figure 4). This supports the idea that plumage complexity, rather than uniformity, provides more phenotypic traits for natural selection to act upon. In addition, we find that island lineages have faster rates of plumage evolution (Table 1), but not more complex plumages (Table 2, Figure 4—figure supplement 2), than continental lineages.

Colonization of novel geographic areas can promote either shifts in mean phenotype or changes in rates of phenotypic evolution (Collar et al., 2009). Changes in rates associated with island colonization have been described in lizards (Pinto et al., 2008) and mammals (Millien, 2006). Pinto et al., 2008 found that rates of morphological evolution were not elevated in Caribbean anoles compared to mainland species, but they did show differences in morphospace (i.e., convergence). In birds, Doutrelant et al., 2016 measured coloration in 4448 patches of 232 species (including eight kingfisher species) and found that island-dwelling species have darker colors and fewer color patches (i.e., less complex plumages) than mainland species. This differs from the results of our study, as we found no difference in achromatic (Table 3) or chromatic plumage complexity between mainland and island species (Table 2, Figure 4—figure supplement 2). Rather, it is the rates of evolution that increase once kingfishers colonize islands (Figure 4B). This suggests decoupling between the effects of complexity and insularity on color evolution rates, and is consistent with previous work showing elevated rates of morphological evolution being independent of the acquisition of a key innovation, such as gecko toepads (Garcia-Porta and Ord, 2013). However, this does not answer the question of why plumage color among island lineages would differ more than among mainland lineages.

Contrary to the prevailing view that island species should have reduced diversity of mate recognition signals (West-Eberhard, 1983), island kingfishers have more variable plumages than their continental relatives (Figure 4B, Table 1). Other examples of this pattern of elevated signal diversity on islands include Anolis lizards (Gorman, 1968) and Tropidurus lizards (Carpenter, 1966) that both show high diversity in dewlap displays. Historical explanations for why colors might evolve include increased conspicuousness for mating displays and more efficient species recognition (Andersson, 1994; Doucet et al., 2007). The species-recognition hypothesis predicts reduced signal distinctiveness (i.e., low amounts of plumage variation) on islands (West-Eberhard, 1983). This is because of the lack of potential competitors and conspecifics on islands that would otherwise put selective pressures on color patterns, for example through reproductive character displacement (Drury et al., 2018). However, we found no support for this idea, as evolutionary rates of coloration were not significantly associated with the number of sympatric species (Table 1). Another mechanism that could explain the observed rapid color evolution on islands is divergence in abiotic factors among islands. Structural coloration, responsible for vivid blues, greens, and purples, is salient in kingfishers (Stavenga et al., 2011; Eliason et al., 2019). Compared to pigment-based colors, structural colors exhibit some of the fastest rates of color evolution known in birds (e.g., hummingbirds; Eliason et al., 2019; Venable et al., 2022). Structural and melanin-based forms of coloration may have thermal benefits to birds (Rogalla et al., 2022), and both molt speed (Griggio et al., 2009) and dietary protein availability (Meadows et al., 2012) have been shown to influence structurally colored signals. Thus, divergence in food availability or climate among island populations could be driving rapid shifts in coloration. Future work will be needed to tease apart the relative roles of genetic drift, competition, and abiotic factors in driving color evolution on islands. The kingfisher genus Todiramphus harbors several ‘superspecies’—monophyletic groups of allopatric and morphologically distinct taxa (Mayr, 1963)—and therefore could be an ideal study system for clarifying the roles of ecology and constraint in driving color diversity on within and between islands.

Studying color pattern evolution has been historically difficult, due, in part, to an inability of humans to perceive UV color (Eaton, 2005) and difficulties with measuring and analyzing complex color patterns (Mason and Bowie, 2020). Recent work has showed that changes in plumage complexity are associated with shifts in light environment (Maia et al., 2016; Shultz and Burns, 2013). Here, plumage complexity was treated as a response variable rather than as a predictor of overall color divergence between lineages. Our results therefore provide a contrast to previous work in looking at a potential developmental constraint—how plumage patches are arranged on the body (Price and Pavelka, 1996)—and its causal influence on evolutionary trends of color divergence. A caveat with our approach is that it does not consider color patterning within feathers. For example, the species with the least complex plumage according to the mean interpatch color distance (metric c1) is the pied kingfisher (Ceryle rudis), despite its conspicuous black and white barring/spotting across its body and even within individual feathers. We hope that researchers will consider the morphometrics approach we take here, as well as assess its potential strengths and weaknesses, in future studies on the evolution of complex color patterns in nature.

In this study, we collected a large amount of spectral data (9362 measurements of 142 individuals in 72 species) in a diverse family of birds notable for their complex plumages and rapid speciation on islands. The major finding of our study is that complex plumage patterns enable faster rates of color evolution compared to simpler, uniform plumages. Colonization of islands, independent of plumage complexity, resulted in further divergence of coloration among species. More broadly, these results highlight the interplay between a potential key innovation (i.e., plumage complexity) and geographical opportunity for allopatric speciation in birds. It also highlights the need for incorporating multidimensional aspects of plumage patterns in such analyses. Further research is needed to test whether complex plumages are more common in clades that are speciating rapidly and if complexity is itself a direct target of sexual selection (e.g., Choi et al., 2022).

Materials and methods

Measuring feather color

Request a detailed protocol

Using a UV–Vis spectrophotometer (Ocean Optics) operating in bird-visible wavelengths (300–700 nm), we measured reflectance spectra at normal incidence in triplicate for 22 patches (see Figure 1—figure supplement 2) in 72 species, including both males and females, from museum specimens. In total, we obtained 9303 spectra for 142 individuals (available on Dryad). We averaged three spectra per patch per specimen and converted values into avian tetrahedral colorspace (u, s, m, and l channels) using the vismodel function in pavo (Maia et al., 2013a), based on a UV-sensitive visual system (Parrish et al., 1984). We converted quantum catches into perceptually uniform XYZ coordinates (Pike, 2012) for use in downstream comparative analyses (Figure 2A). The distance between pairs of coordinates in this colorspace is proportional to the just noticeable difference (JND). As these data only capture variation in chromatic aspects of coloration, we also assessed achromatic variation by calculating luminance as the quantum catch for the double cone. We used photosensitivity data for the blue tit (Hart et al., 2000) due to the limited availability of sensitivity data for other avian species, and we further accounted for receptor noise following Olsson et al., 2018. We converted luminance values into a scale where distances between pairs of measurements are equivalent to JND values by subtracting ln(0.01) from the ln luminance values and dividing this by the Weber fraction (ω = 0.1), following Pike, 2012. Although it is possible, in theory, to combine chromatic and achromatic channels of plumage variation in a single analysis (Pike, 2012), we opted to analyze them separately because they are likely under different selection pressures (Osorio and Vorobyev, 2005) and we wanted to be able to compare our results with previous work on island bird coloration (Doutrelant et al., 2016).

Assessing color variation

Request a detailed protocol

To gain an understanding of how color varies across different levels of organization (e.g., plumage patches), we performed a taxonomic analysis of variance (Derrickson and Ricklefs, 1988). Briefly, we fit a linear mixed model in MCMCglmm (Hadfield and Nakagawa, 2010) using colorspace XYZ coordinates as amultivariate response, with random effects for plumage patch, sex, and species. We set rather uninformative priors (ν = 2 for random effects, ν = 0.002 for residual covariance) and ran the Markov chain Monte Carlo (MCMC) chains for 106 generations, discarding 25% as burn-in. We ran two chains and assessed convergence by plotting the Gelman–Rubin diagnostic (Gelman and Rubin, 1992) using gelman.plot in the R package coda 0.19.4. From the fitted models, we calculated the sum of mean posterior variances for each colorspace coordinate and estimated the proportional amount of variation explained at each level of organization by dividing each variance by the total variance (see Dryad for R code). To visualize color pattern diversity, we used pPCA to reduce the dimensionality of our color data set. We performed pPCA using the phyl.pca_pl function in R (Clavel et al., 2019) based on the covariance matrix. Phylogenetic PCA has been criticized because of the influence of component selection bias when used in downstream comparative analyses (Uyeda et al., 2015), therefore we also performed an ordinary PCA with the prcomp function in R, with similar parameters as above.

Quantifying plumage complexity

Request a detailed protocol

To estimate plumage complexity at the individual level, we obtained pairwise perceptual distances (proportional to just noticeable differences, JNDs) using the coldist function in pavo (Maia et al., 2013b). Since we measured color of 22 patches, this resulted in a 22 × 22 matrix for each individual. Next, using these chromatic and achromatic distances, we calculated plumage complexity in three ways (see Figure 1A, section 3b). First, we averaged these distance matrices by species and calculated mean interpatch distances (metric c1) using the R dist function. Second, we calculated the total colorspace volume occupied by an individual’s plumage (metric c3) as a 3D volume for XYZ colorspace coordinates. We used the convhulln function in the R package geometry v. 0.4.6.1 to calculate this metric, which can also be described as the range of luminance values for achromatic variables. Third, we calculated complexity as the number of distinct contiguous patches in folded-wing body configuration (metric c3). To do so, we converted species color distance matrices to binary scores indicating whether pairs of patches are perceptibly distinct (JND >1) or not (JND <1), resulting in a pairwise color distance matrix (MJND), with 0 indicating patches that are distinct and 1 indicating patches that would be perceived as the same by a bird. For each individual, we multiplied MJND by the adjacency matrix Madj. In Madj, 1 indicates patches that are adjacent on a bird’s body and 0 indicates non-adjacent patches. Multiplying MJND by Madj results in a matrix with 1 if patches are both adjacent and indistinguishable in colorspace (Figure 1A). Finally, we converted these final matrices into igraph objects using graph_from_adjacency_matrix and determined the number of distinct plumage regions using the components function in igraph (Csardi and Nepusz, 2006). We only considered a folded-wing plumage configuration because this is how a bird would be typically seen by a conspecific and because folded-wing complexity scores were highly correlated with spread-wing complexity (r = 0.98).

Understanding the tempo and mode of color evolution

Request a detailed protocol

To understand evolutionary trends on a per-patch basis, we compared phylogenetic signal and rates of color evolution within each individual plumage patch (N = 22, Figure 2C, D) using distance-based comparative methods (Adams, 2014b; Denton and Adams, 2015). Next, to account for phylogenetic signal at the overall plumage level, we fit multivariate Brownian motion and Pagel’s λ models for all color traits (i.e., 22 lightness variables, 66 color variables) using the fit_t_pl function (Clavel et al., 2019) in RPANDA v. 2.1 (Morlon et al., 2016). We compared models using generalized information criteria. The best-fitting model was a Pagel’s λ model for both achromatic and chromatic plumage components (Supplementary file 1a), thus we used these λ estimates to transform branch lengths of the phylogeny prior to running multivariate comparative analyses. Although variance tests revealed some color variation attributable to sex differences (Figure 2B), multivariate phylogenetic integration tests (Adams et al., 2014c) revealed significant correlations between male and female plumage coloration for both plumage color (r-PLS = 0.88, p < 0.01, N = 57) and lightness (r-PLS = 0.93, p < 0.01, N = 57). Therefore, for primary analyses, we used a combined data set with male and female color data averaged together for each species. However, we also included results for the males-only (N = 63) and females-only data sets (N = 68; see Supplementary file 1a–e).

Testing ecological predictors of plumage color variation

Request a detailed protocol

To the prediction that species on islands have less complex plumages, as predicted by the species-recognition hypothesis (West-Eberhard, 1983), we used phylogenetic linear models (Ho and Ané, 2014). For the models, complexity metric was set as the response variable and the predictors were insularity and the number of sympatric species (additional parameters: method = "lambda", lower.bound = 1e−10; see Dryad for R code). Body mass has recently been shown to explain variation plumage complexity of passerine birds (Cooney et al., 2022), therefore we also included ln body mass (in grams) as a covariate in our regression models (species averages obtained from Dunning, 2007). We removed non-important variables from the models using a bidirectional AIC-based stepwise procedure in phylostep. To further test whether species on islands are convergent in their overall plumage color patterns, we used a multivariate d-PGLS approach (Adams, 2014b) implemented in the prodD.pgls function of geomorph v. 4.0.4 (Adams et al., 2013). For d-PGLS models, either multivariate color or multivariate lightness was set as the response variable, and the predictor variables were insularity, ln body mass, and the number of sympatric species. We assessed significance using a permutation approach with 999 iterations.

Determining drivers of evolutionary rate variation in plumage color

Request a detailed protocol

To test our predictions that rapid rates of color evolution are associated with complex plumages and insularity, we again used PGLS models implemented in the phylolm R function (Ho and Ané, 2014). We ran PGLS models with species-specific rates as the response and complexity metrics (c1, c2, and c3), ln body mass, insularity, and the number of sympatric species as predictors. We included several complexity metrics in the same analyses since these metrics are likely capturing somewhat different aspects of plumage complexity. As an example, a hypothetical bird with a mostly black plumage and a single red patch would result in a high color volume despite a low mean interpatch distance. We fit models for both plumage variable types (chromatic and achromatic) and sexes (males and females) using the options method = "lambda" and lower.bound = 1e−10 and determined the best-fitting model using a stepwise AIC-based procedure with phylostep (see Supplementary file 1f and g for results of all models tested). Because some complexity metrics were correlated (Figure 1—figure supplement 1), we assessed multicollinearity with VIFs. Briefly, following Mundry, 2014, we (1) re-formulated the best-fitting PGLS model with each complexity metric set as the response variable rather than a predictor, (2) estimated R2 values for each model with the R2.lik function in rr2 v. 1.0.2 (Ives, 2019), and (3) calculated VIFs as 1/(1R2). To further test the sensitivity of PGLS estimates to potential multicollinearity, we re-fit the best-fitting models after dropping each complexity metric in turn (see Figure 4—figure supplement 1 for results). All variables were scaled prior to PGLS analysis to make coefficients comparable across models (i.e., as effect sizes).

In addition to this PGLS approach, we also used a well-established method (Denton and Adams, 2015) for comparing rates among groups with high and low plumage complexity scores. This analysis required binary estimates of complexity, therefore we used kmeans clustering (centers = 2) to derive binary complexity scores. We then input these values as a predictor of rate variation into the compare.evol.rates function (Denton and Adams, 2015), with multivariate color (N = 66 traits) or lightness (N = 22) as the response. We used the permutation option with 999 iterations to assess significance of the relationship between predictors (complexity or insularity) and rates of color evolution. We used a similar approach with insularity as a predictor of rate variation.

Data availability

Data sets and R code used in analyses within the manuscript are available at: https://doi.org/10.5061/dryad.5mkkwh78v.

The following data sets were generated
    1. Eliason C
    2. McCullough J
    3. Hackett S
    4. Andersen M
    (2023) Dryad Digital Repository
    Complex plumages spur rapid color diversification in kingfishers (Aves: Alcedinidae).
    https://doi.org/10.5061/dryad.5mkkwh78v
The following previously published data sets were used
    1. Eliason C
    2. Andersen M
    3. Hackett S
    (2020) Dryad Digital Repository
    Data from: Using historical biogeography models to study color pattern evolution.
    https://doi.org/10.5061/dryad.3680n0c
    1. Eliason C
    2. McCullough J
    3. Andersen M
    4. Hackett S
    (2021) Dryad Digital Repository
    Accelerated brain shape evolution is associated with rapid diversification in an avian radiation.
    https://doi.org/10.5061/dryad.ffbg79cs6

References

  8. Book
    1. Andersson M
    (1994) Sexual Selection
    Princeton, NJ: Princeton University Press.
    https://doi.org/10.1515/9780691207278
  13. Conference
    1. Carpenter CC
    (1966) Comparative behavior of the Galápagos lava lizards (Tropidurus)
    The Galápagos: Proceedings of the Galápagos International Scientific Project. pp. 269–273.
    https://doi.org/10.1525/9780520328389
    1. Csardi G
    2. Nepusz T
    (2006)
    The igraph software package for complex network research
    InterJournal, Complex Systems 1695:1–9.
    1. Gorman GC
    (1968)
    The relationships of Anolis of the Roquet species group (Sauria: Iguanidae)-III. Comparative study of display behavior
    Breviora 284:1–31.
  61. Book
    1. Mundry R
    (2014) Statistical issues and assumptions of phylogenetic generalized least squares
    In: Garamszegi LZ, editors. Modern Phylogenetic Comparative Methods and Their Application in Evolutionary Biology: Concepts and Practice. Berlin, Heidelberg: Springer. pp. 131–153.
    https://doi.org/10.1007/978-3-662-43550-2
  72. Book
    1. Price TD
    (2008)
    Speciation in Birds
    Greenwood Village, CO: Roberts & Co.
  73. Book
    1. Prum RO
    (2006) Anatomy, physics, and evolution of structural colors
    In: Hill GE, McGraw KJ, editors. Bird Coloration. Volume I: Mechanisms and Measurements. Cambridge, Massachusetts: Harvard University Press. pp. 295–353.
    https://doi.org/10.2307/j.ctv22jnscm

Decision letter

  1. Kaspar Delhey
    Reviewing Editor; Max Planck Institute for Ornithology, Germany
  2. Christian Rutz
    Senior Editor; University of St Andrews, United Kingdom
  3. Kaspar Delhey
    Reviewer; Max Planck Institute for Ornithology, Germany

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Complex plumages spur rapid color diversification in island kingfishers (Aves: Alcedinidae)" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Kaspar Delhey as Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Christian Rutz as the Senior Editor.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) The two main topics linked to evolutionary rates of plumage colour, island-dwelling, and plumage complexity need to be better introduced and linked in the introduction. This is a point that was made in one form or another by the three reviewers.

2) The concept of plumage complexity, which is central to the paper, requires a more comprehensive explanation. In particular, how it differs from evolutionary divergence to avoid any potential circularity.

3) The methods used to analyse differences in colour evolutionary rates constitute some of the most novel aspects of the paper. However, these are difficult to follow. The authors need to cater for two different types of readers here:

(a) Those that only want to broadly understand the methods used and

(b) those that will want to adapt the methods to their own study system.

To deal with (a) Rev 4 suggestion of a flowchart figure illustrating the methods provides an elegant way to summarise the approach used. For (b) much more detail is needed regarding the multiple ways in which analyses can vary, this ranges from information regarding arguments used within specific functions to better explanations as to why certain methodological decisions were taken.

4) There are many statistical analyses in the paper. Some may be better suited than others (e.g. the pitfalls of using PCA suggest that maybe they are less well suited) and some seem somewhat redundant. The authors should try to streamline this as much as possible, potentially discarding ancillary analyses that do not deal with the main questions.

Reviewer #1 (Recommendations for the authors):

Specific comments (ln refer to line numbers)

Ln 112-128 Is the PCA needed to describe colour variation? It does not seem to add much and PCA has been recently criticised in the context of comparative analyses and other multivariate approaches (some of which are implemented in the paper) are preferred (Uyeda et al., 2015; Adams and Collyer, 2018).

Ln 139 high phylogenetic signal == species diagnostic traits. I am not sure that this should be the case, the high phylogenetic signal would indicate that closely related species have similar colours, and hence such species groups would share rather than differ in these traits. Thus, species identity among closely related species would not be easily signalled by such colours.

Ln 179-180 and in passerine birds see (Cooney et al., 2022).

Ln 140-187 Why did you not try to use geomorph::compare.evol.rates to compare multivariate evolutionary rates between species with low and high plumage complexity? I know this approach only deals with tips and not nodes, but it is a bit unsatisfactory to use one approach for one question and another one for the other given that the procedures are quite different.

Ln 225 why intra-specific competition?

Ln 236 p=0.08 should not be interpreted as no support (see (Muff et al., 2021)), and it would be good to state the direction of this effect in the text as well (does it lean towards convergence or away from it?) as most readers will not be familiar with this particular statistic.

Ln 238 with 'drabber' do you mean darker (less light)? I would be careful with using terms such as drab or bright (dark and light would be better) without a clear definition first.

Ln 239 I am not sure that these results add much, the comparison between structural and pigmentary colours is a bit redundant (mechanisms were largely identified due to properties of the reflectance spectrum). I think that including these results detracts from the main message, similarly, I am not sure about the relevance of sex differences. Differences in mean colour between island and mainland species -if included- should come earlier in the more descriptive part of the results. Why did you not use geomorph::procD.pgls for this analysis? As an aside, how did you deal with sex differences in the other analyses? I might have missed it but, do they represent both males and females or only one sex? Please clarify.

Ln 280 "… competition between individuals of the same species is driving color diversity…" this needs further explanation, what would the mechanism be?

Ln 307-310 Why is your node+tips approach inherently better than a tips-only approach, other than by being more conservative? It could also be that including nodes increases type II error rates simply due to the inherent uncertainty associated with the estimation of ancestral traits (which often end up being an average of the tips). I think that a bit more justification is needed here.

Ln 312 nor achromatic variation.

Ln 332 you averaged three spectra per patch per specimen, right?

Ln 344-46 PCA using a covariance or correlation matrix? Covariance would be the appropriate choice here as all variables are measured in the same units. This is not clear from the R code provided (at least to me). For the non-phylogenetic PCA probably the covariance matrix was used as this seems the default option for function prcomp, but I am not sure what the case is for the phylogenetic PCA function used. Nevertheless, this should be clear in the text.

Ln 346 pPCA has been criticised before see (Uyeda et al., 2015; Adams and Collyer, 2018), and this is particularly relevant when choosing to discard some PCs from subsequent analyses.

More information on the MCMCglmm priors is needed, from the code I get the impression that default priors were used (i.e. I see no prior argument specified in the code, or am I misunderstanding something)? Such default priors are usually not good for random effects. Also, it would be good to state that models converged and how this was assessed. Also, the multivariate Wald tests used in these MCMCglmm models should be better explained in the text.

The function geomorph::compare.evol.rates has two methods to establish whether evolutionary rate ratios between two regimes differ significantly from 1, the default ("simulation") and an alternative ("permutation"). In my experience using this function to analyse evolutionary rates of colour, "simulation" yields much lower p-values than "permutation" (unpubl. data). Based on reading (Adams and Collyer, 2018) and on direct communication with the main author of the geomorph package (Dean Adams) it seems that "permutation" is a better alternative to achieve an appropriate level of type I errors. Based on the R code provided it seems that the authors have used the default option "simulation". I think that the analyses using this function should be re-run with the alternative "permutation". I suspect that the evolutionary rate ratio will not differ significantly from 1 in that case. This should not matter, the result in itself is interesting regardless of the outcome, although it could mean that some key conclusions of the paper will have to be changed.

Figure 1. This figure could be deleted or moved to the suppl material, it does not add much to the general story.

Figure 3C It would be good to label data points that correspond to nodes vs tips in the figure to illustrate differences in results. Would it be possible to also include island/mainland dwellings in this analysis? It should be doable to reconstruct ancestral states for this trait as well right?

Figure 7 I am not sure this figure should be included in the main text, it is hard to visualize the patterns.

References

Adams, D.C. and Collyer, M.L. 2018. Multivariate Phylogenetic Comparative Methods: Evaluations, Comparisons, and Recommendations. Syst. Biol. 67: 14-31.

Cooney, C.R., He, Y., Varley, Z.K., Nouri, L.O., Moody, C.J.A., Jardine, M.D., et al. 2022. Latitudinal gradients in avian colourfulness. Nat. Ecol. Evol. 6: 622-629.

Muff, S., Nilsen, E.B., O'Hara, R.B. and Nater, C.R. 2021. Rewriting Results sections in the language of evidence. Trends Ecol. Evol. 37: 203-210.

Olsson, P., Lind, O. and Kelber, A. 2018. Chromatic and achromatic vision: parameter choice and limitations for reliable model predictions. Behav. Ecol. 29: 273-282.

Siddiqi, A., Cronin, T.W., Loew, E.R., Vorobyev, M. and Summers, K. 2004. Interspecific and intraspecific views of color signals in the strawberry poison frog Dendrobates pumilio. J. Exp. Biol. 207: 2471-2485.

Uyeda, J.C., Caetano, D.S. and Pennell, M.W. 2015. Comparative Analysis of Principal Components Can be Misleading. Syst. Biol. 64: 677-689.

Reviewer #2 (Recommendations for the authors):

Line 16: I think I know what you mean but this is not actually clear. By phenotypic, you mean increased rates of change per speciation event, correct? Whereas by species diversification you just mean increased speciation rates, correct? Make this clearer.

Line 25: I'm immediately wondering how plumage complexity is calculated and whether it de facto will tend to correlate with color diversity.

Lines 38-40: I'm not sure this is how the island rule is phrased. Isn't it more that organisms tend to shift notably in body size? Otherwise, how do we define a large-bodied colonist vs. a small-bodied colonist? Suggest making this sentence more general.

Line 58-59: I'm not sure what you mean here. If a color variant for a species with uniform coloration had higher fitness then yes, the whole plumage would change in tandem, but if the mutation involved a patch changing color, then the whole plumage wouldn't change in tandem. The way this is worded makes it sound like a uniformly colored species will be unable to evolve plumage complexity.

Line 63: are the authors going to define plumage complexity as the number of plumage patches that are different from one another? Because, if so, this will QED correlate with color variation. Revisiting this point after a few reads of the manuscript, I realize the authors are working across intraspecific and interspecific scales in this sentence. This needs revision here.

Lines 96-98: a lot of this feels really circular. Some attention is needed here. Cutting some words, this sentence says, "promote color divergence…contributing to high rates of color evolution and disparity". How color divergence differs from rates of color evolution differs from color disparity is not clear.

Line 122: change to "sex" in Figure 2, not "individual".

Line 127: I'm confused. Isn't Tanysiptera monomorphic?

Lines 124-128: this immediately makes me wonder whether type (i) tends to be allopatric more often than does type (iii). I think the answer is yes.

Line 137: This is in line with a recent Merwin et al. study on parrots (https://bmcecolevol.biomedcentral.com/articles/10.1186/s12862-020-1577-y).

Lines 138-139: High phylogenetic signal in X vs Y does not mean that X is "more diagnostic of species" than Y. Re-word to clarify.

Lines 152-157: Ok, it does seem like the authors define plumage complexity as the number of uniquely colored patches.

Lines 180-187: I think there are a few key results here that deserve more space.

Line 206: the result is not that they are de novo evolving new colors, correct? Rather, they are shifting around in color space more quickly. Clarify.

Lines 209-225: There are four substantial issues in this paragraph. First, anytime I hear about creating two separate trees from one to run a test I get leery. I'm not sure what the justification is here. Second, how many island species occur with a sympatric relative? I believe that the answer is not many, so we wouldn't expect interspecific competition to be relevant, correct? Third, I'm fairly sure these RPANDA functions can take a geography object to account for this varying sympatry/allopatry. No mention is made of whether this was done or, if so, how. Fourth, RPANDA isn't cited. I may be missing something, but I also believe the last sentence here is not warranted. It could also mean that drift on islands best explains the observed pattern.

Line 237: in what taxon?

Line 247: I'm confused. Earlier you said this author DID find predictable patterns towards duller plumage. You also say it again in the discussion below.

Line 289: What island is this? Does New Guinea count as an island?

Lines 273-293: I am not following the logic in this paragraph. I see no reason why not finding evidence for interspecific competition influencing plumage patterns makes any statement about the relevance of intraspecific competition. This is further confused by the first discussion in the manuscript of sympatry and allopatry, which is then not actually tied into the mechanisms of interest (competition and character displacement). This is a weak point of the paper as it stands.

Lines 300-309: I'm still a little confused about the differences between plumage complexity and evolutionary rates (it would be helpful to keep reminding the reader that one is within a species, and one is between species), but I think some of what the authors are saying here can be attributed to differences between a phylogenetic independent contrasts-esque approach (what the authors use), and a phylogenetic generalized least squares approach (what others tend to have used). The former tends to be more conservative in my experience. Per a comment above, I'm withholding too many thoughts about alternative tests until I get a clearer picture of what the authors want to test, but I suspect this could be done as a phylogenetic t-test, where the test is whether plumage complexity differs between island taxa or not.

Figure 7: very challenging to see. It might be clearer with a white background.

Figure S6: "Effect of data analysis on rate-complexity correlations". Data analysis is too vague. Can you come up with a more specific figure legend?

Line 43: errant parentheses.

Lines 92-94: run-on sentence, passive voice. Revise.

Line 103: missing a word after color.

Line 291: are->is

Line 411: the website for this paper incorrectly lists Ian PF Owens as just PF Owens. If you check the original manuscript, it correctly lists it as IPF Owens.

Reviewer #3 (Recommendations for the authors):

My line-by-line suggestions follow, I hope are helpful:

Line 15: This is entirely pedantic, but children typically leave the cradle, whereas oceanic island lineages often remain confined there until extinction or replacement (as in Wilson's taxon cycle). Maybe there's a better analogy out there.

Lines 57-58: Constraints on morphospace need not constrain evolutionary rates, see Goswami et al. 2014.

Line 60: Defining complexity can be a difficult topic to agree on in evolutionary biology and I think that considering the centrality of pattern complexity to the study, this would benefit from being discussed in more detail. Complexity at one level (phenotype) may not match with complexity at others (biochemistry, development, genome). I could see a working definition in the methods, as well as a supporting citation or two as a way to resolve this issue.

Lines 66-84: Can this paragraph be condensed and combined with the one below? I'm not seeing how it's relevant enough to the paper's key ideas to merit this much-lit review text. I don't want to hold this against the authors though, because I'm getting a subtle impression that a previous reviewer might have requested it, and nothing is more infuriating than having reviewers from different journals disagree with each other.

Line 82-84: I don't disagree with this outright – it does seem to be the case in starlings – but I would consider this to be a hypothesis that is actively being researched rather than something that can be asserted. In many avian taxa, structural color seems to be static as white, but maybe that's also pedantic.

Color Ordination and Comparison: I'm fond of the authors' approach of summarizing color as a multivariate trait. However, I'm not altogether certain I understand it precisely because different parts of it are described in different places throughout the manuscript. Perhaps a figure panel with a simple flow chart connecting the color and comparative analyses performed would be helpful to readers.

Lines 125-128, Figure 2B: I had some difficulty placing this analysis in the context of the study's design or tying it to any of the main hypotheses or questions. If the interpretation is that this highlights which clades tend to be the most complex, maybe a sentence about that could go in the figure legend?

Lines 213-222: I found this part of the paperless more convincing than the rest due to issues described in Uyeda et al. (2016, Syst Biol). Even with a phylogenetic PCA, we should expect biases in model fit when individual principal components are used, particularly around OU models (also see Adams and Collyer 2018, Syst Biol). So while I would interpret Figure 5B in the same way as the authors and have no reason to disagree with them, I am skeptical it allows us to really reject competition as an explanatory mechanism here.

Lines 253-254: This is really interesting and I'd love to see more written about this idea!

Lines 273-293: This is tangentially related to the previous comment, reading the discussion it felt like we dive into adaptive explanations of color without much space to discuss drift or liability/availability of different color mechanisms (these are however discussed in great detail in the introduction).

Figure 1: This is a great figure, I really enjoyed it.

Figure 3: When I see an ancestral state reconstruction like this where nearly all change is independently derived, it makes me skeptical of the model. Highly complex patterns seem to be prevalent in the Alcedininae, but I couldn't find any ancestral nodes with complexity scores half as high as extant species. Since these ancestral states are used as data in downstream analyses, it might be worth testing the robustness of those analyses to uncertainty in those states, or at least describing that uncertainty.

Figure 4: The tip colors seem to refer to a legend that is no longer in Figure 3, but in Figure 6.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your article entitled "Complex plumages spur rapid color diversification in kingfishers (Aves: Alcedinidae)" for further consideration by eLife.

Your revised article has been evaluated by Christian Rutz (Senior Editor) and Kaspar Delhey (Guest Reviewing Editor). The article has been improved, but there are some remaining issues that need to be addressed, as outlined below:

Ln 23 "island insularity" revise.

Ln 29-31 "Importantly, we found that island species did not have more complex plumages than their continental relatives. Thus, complexity may be a key innovation that facilitates response to relaxed (or divergent) selection pressures on islands." We cannot see how the second sentence follows the first, if island bids do not have higher plumage complexity, how can complexity be a key innovation that facilitates island living?

Ln 145 chromatic variation includes hue and saturation.

Ln 145 "achromatic color" does not make much sense; suggest referring to chromatic and achromatic variation throughout the paper rather than chromatic and achromatic color.

Ln 146 similarly "brightness" tends to be used in a multitude of ways, maybe replace with lightness or light-to-dark variation.

Ln 166-169 "To determine rates of overall plumage evolution, we used a phylogeny by McCullough et al. (2019), which incorporated thousands of ultraconserved elements (Faircloth et al., 2012) for a fully sampled, time-calibrated phylogeny of the avian order Coraciiformes (kingfishers, bee-eaters, rollers, and allies)." Revise sentence, "…we used a phylogeny…for a phylogeny…" something is off here, verb lacking.

Ln 230 maybe "theory" is too strong here, replace it with "idea"?

Ln 253 Is this the case? You mention Table 3 which refers to rates of colour evolution, not plumage complexity. This needs some clarification, as in Table 2 you do show that island species do not have more complex plumages. As a matter of fact, some sentences in this paragraph contradict the previous one.

Ln 390 some justification for the inclusion of body mass here may be good.

Ln 400 typo "drivers".

Table 3 We do have some concerns with multicollinearity here. As shown in Figure S1, different estimates of plumage complexity seem strongly intercorrelated. Thus, when your best model for chromatic variation identifies c1 and c3 as important predictors, the effects that your model quantifies constitute the effects of one predictor controlling for variation in the other. Now, if both are strongly correlated there is not that much independent variation that is relevant to explain. We think that it would be important to fit models with each predictor separately in order to check that results follow the same pattern. Moreover, this would mean reporting more complete results independent of whether the model is the "best" model identified by AIC. Potential future meta-analysts will need this information regardless of statistical significance or AIC. Please present the results from all models tested.

https://doi.org/10.7554/eLife.83426.sa1

Author response

Essential revisions:

1) The two main topics linked to evolutionary rates of plumage colour, island-dwelling, and plumage complexity need to be better introduced and linked in the introduction. This is a point that was made in one form or another by the three reviewers.

We have revised the introduction to make these connections more clearly. Now, we first introduce bird coloration (paragraph 1), followed by the challenges and importance of studying complexity (para 2) and a discussion of how islands are ideal systems to test evolutionary hypotheses (para 3). Finally, we move into our study system (para 4) and the hypothesis/predictions we are testing (para 5).

2) The concept of plumage complexity, which is central to the paper, requires a more comprehensive explanation. In particular, how it differs from evolutionary divergence to avoid any potential circularity.

We changed the text to clarify that complexity is an intraspecific measure, while rates are an interspecific measure (e.g., see lines 140, 170, 222, 668).

3) The methods used to analyse differences in colour evolutionary rates constitute some of the most novel aspects of the paper. However, these are difficult to follow. The authors need to cater for two different types of readers here: (a) Those that only want to broadly understand the methods used and (b) those that will want to adapt the methods to their own study system. To deal with (a) Rev 4 suggestion of a flowchart figure illustrating the methods provides an elegant way to summarise the approach used. For (b) much more detail is needed regarding the multiple ways in which analyses can vary, this ranges from information regarding arguments used within specific functions to better explanations as to why certain methodological decisions were taken.

Following these suggests, we have now added a new workflow figure (Figure 1) that we hope will clarify any questions the reader may have of our approach. We have also added several details in the Methods/Results sections to justify our methods and provide the reader with a blueprint for replicating the approach (e.g., see lines 336-359).

4) There are many statistical analyses in the paper. Some may be better suited than others (e.g. the pitfalls of using PCA suggest that maybe they are less well suited) and some seem somewhat redundant. The authors should try to streamline this as much as possible, potentially discarding ancillary analyses that do not deal with the main questions.

We removed the Bayesian models (MCMCglmm), as 2 reviewers had concerns about whether the analyses were necessary for supporting the main points of the manuscript. We further streamlined the methods as much as possible, for example, we now analyze color evolutionary rates and color convergence on islands in a unified PGLS framework (discussed in Methods, e.g., lines 377-390, 391-402).

Reviewer #1 (Recommendations for the authors):

Specific comments (ln refer to line numbers)

Ln 112-128 Is the PCA needed to describe colour variation? It does not seem to add much and PCA has been recently criticised in the context of comparative analyses and other multivariate approaches (some of which are implemented in the paper) are preferred (Uyeda et al., 2015; Adams and Collyer, 2018).

We felt this provided a neat way to visualize variation in complex plumage patterns across the group (as also mentioned by Reviewer 3). As such, we would prefer to leave this component in the manuscript.

Ln 139 high phylogenetic signal == species diagnostic traits. I am not sure that this should be the case, the high phylogenetic signal would indicate that closely related species have similar colours, and hence such species groups would share rather than differ in these traits. Thus, species identity among closely related species would not be easily signalled by such colours.

We changed this line to read “are more taxonomically informative than crown or wing plumage” (lines 133-134) to clarify our point.

Ln 179-180 and in passerine birds see (Cooney et al., 2022).

We appreciate the reviewer bringing this reference to our attention–we have added it at line 184.

Ln 140-187 Why did you not try to use geomorph::compare.evol.rates to compare multivariate evolutionary rates between species with low and high plumage complexity? I know this approach only deals with tips and not nodes, but it is a bit unsatisfactory to use one approach for one question and another one for the other given that the procedures are quite different.

We appreciate this comment and agree it is best to not mix-and-match methods. We were trying to account for continuous variation in plumage complexity rather than bin species into high or low values of complexity. As such, in the revised version, we treat complexity and insularity together in a unified PGLS framework, and then validate these results (as supplemental material) using compare.evol.rates with high/low complexity bins, as suggested. The results are generally in agreement with the PGLS analyses of species-specific rates, showing faster rates for more complex plumages for chromatic, but not achromatic, color.

Ln 225 why intra-specific competition?

This is just one hypothesis that could be tested in future work. We further point out that genetic drift could also explain rapid color evolution on islands (e.g., lines 103, 219, 266).

Ln 236 p=0.08 should not be interpreted as no support (see (Muff et al., 2021)), and it would be good to state the direction of this effect in the text as well (does it lean towards convergence or away from it?) as most readers will not be familiar with this particular statistic.

Point noted. We changed the text to “weak support.”

Ln 238 with 'drabber' do you mean darker (less light)? I would be careful with using terms such as drab or bright (dark and light would be better) without a clear definition first.

We changed the text to “darker” instead of drabber.

Ln 239 I am not sure that these results add much, the comparison between structural and pigmentary colours is a bit redundant (mechanisms were largely identified due to properties of the reflectance spectrum). I think that including these results detracts from the main message, similarly, I am not sure about the relevance of sex differences. Differences in mean colour between island and mainland species -if included- should come earlier in the more descriptive part of the results. Why did you not use geomorph::procD.pgls for this analysis?

We have removed this aspect of the manuscript, as 2 of 3 reviewers and the editors had concerns about it. As suggested by this reviewer, we feel it did not add very much value, and the paper is now more streamlined as a result. We did not use procD.pgls for this because, to the best of our understanding, that analysis requires an averaged value for a species, and we wanted to look at within-species variation (e.g., between sexes, among plumage patches), hence the MCMCglmm approach.

As an aside, how did you deal with sex differences in the other analyses? I might have missed it but, do they represent both males and females or only one sex? Please clarify.

In our original analysis of rates and complexity, we had averaged males and females. However, we appreciate the reviewer pointing out that we had not made this clear, or justified the approach. We now discuss this detail in the Methods (lines 370-376), provide results of a test for sexual dichromatism showing a strong correlation between male and female coloration justifying our approach of taking species averages (line 685), as well as include statistical results for males and females analyzed separately (e.g., see Supplementary files 1a-1e).

Ln 280 "… competition between individuals of the same species is driving color diversity…" this needs further explanation, what would the mechanism be?

We removed the line and reframed this section (see lines 245-270).

Ln 307-310 Why is your node+tips approach inherently better than a tips-only approach, other than by being more conservative? It could also be that including nodes increases type II error rates simply due to the inherent uncertainty associated with the estimation of ancestral traits (which often end up being an average of the tips). I think that a bit more justification is needed here.

This is a good point about error associated with ancestral states. We’ve opted to use a tip-based approach that treats species-specific rates in a PGLS framework, alongside the geomorph::compare.evol.rates approach.

Ln 312 nor achromatic variation.

We have now integrated achromatic color variation into the manuscript.

Ln 332 you averaged three spectra per patch per specimen, right?

Correct, we have added this detail at line 303.

Ln 344-46 PCA using a covariance or correlation matrix? Covariance would be the appropriate choice here as all variables are measured in the same units. This is not clear from the R code provided (at least to me). For the non-phylogenetic PCA probably the covariance matrix was used as this seems the default option for function prcomp, but I am not sure what the case is for the phylogenetic PCA function used. Nevertheless, this should be clear in the text.

We used covariance matrices, and we now indicate this in the Methods (line 332).

Ln 346 pPCA has been criticised before see (Uyeda et al., 2015; Adams and Collyer, 2018), and this is particularly relevant when choosing to discard some PCs from subsequent analyses.

We are only using PCA/pPCA to visualize trends in plumage complexity now. Both PC and pPC traits show similar trends, thus we’d prefer to leave these analyses in place. That said, we added an additional line about the caveats of pPCA (see lines 333-335).

More information on the MCMCglmm priors is needed, from the code I get the impression that default priors were used (i.e. I see no prior argument specified in the code, or am I misunderstanding something)? Such default priors are usually not good for random effects. Also, it would be good to state that models converged and how this was assessed. Also, the multivariate Wald tests used in these MCMCglmm models should be better explained in the text.

In response to comments made by the editors and Reviewer 2, we have removed this analysis from the manuscript.

The function geomorph::compare.evol.rates has two methods to establish whether evolutionary rate ratios between two regimes differ significantly from 1, the default ("simulation") and an alternative ("permutation"). In my experience using this function to analyse evolutionary rates of colour, "simulation" yields much lower p-values than "permutation" (unpubl. data). Based on reading (Adams and Collyer, 2018) and on direct communication with the main author of the geomorph package (Dean Adams) it seems that "permutation" is a better alternative to achieve an appropriate level of type I errors. Based on the R code provided it seems that the authors have used the default option "simulation". I think that the analyses using this function should be re-run with the alternative "permutation". I suspect that the evolutionary rate ratio will not differ significantly from 1 in that case. This should not matter, the result in itself is interesting regardless of the outcome, although it could mean that some key conclusions of the paper will have to be changed.

We were using geomorph 4.0.4 (which we now indicate in line 387) that has the default option set as a permutation. We clarify this in the Methods (see lines 408-410).

Figure 1. This figure could be deleted or moved to the suppl material, it does not add much to the general story.

We would prefer to leave it in place, as Reviewer 3 pointed out they liked the figure and we feel it is useful for visualizing variation in plumage complexity across kingfishers.

Figure 3C It would be good to label data points that correspond to nodes vs tips in the figure to illustrate differences in results. Would it be possible to also include island/mainland dwellings in this analysis? It should be doable to reconstruct ancestral states for this trait as well right?

We appreciate this suggestion. We have redone this figure to show both complexity (color of branches) and rates (length of branches). Our new tip-based rate analyses further incorporate both insularity and complexity (e.g., see lines 391-402).

Figure 7 I am not sure this figure should be included in the main text, it is hard to visualize the patterns.

Considering this comment and the suggestion of Reviewer 2 and the editors, we have removed this figure from the manuscript, along with the corresponding RPANDA::fit_t_comp analyses.

Reviewer #2 (Recommendations for the authors):

Line 16: I think I know what you mean but this is not actually clear. By phenotypic, you mean increased rates of change per speciation event, correct? Whereas by species diversification you just mean increased speciation rates, correct? Make this clearer.

We cut this line from the manuscript.

Line 25: I'm immediately wondering how plumage complexity is calculated and whether it de facto will tend to correlate with color diversity.

We appreciate this point. To point out to the reader that these two metrics could in theory be either negatively or positively correlated, we added examples of sister kingfisher species showing simple plumages and divergent colors, as well as complex plumages with similar colors (see Figure 1B).

Lines 38-40: I'm not sure this is how the island rule is phrased. Isn't it more that organisms tend to shift notably in body size? Otherwise, how do we define a large-bodied colonist vs. a small-bodied colonist? Suggest making this sentence more general.

We removed this sentence, as it was not needed in our revised Introduction.

Line 58-59: I'm not sure what you mean here. If a color variant for a species with uniform coloration had higher fitness then yes, the whole plumage would change in tandem, but if the mutation involved a patch changing color, then the whole plumage wouldn't change in tandem. The way this is worded makes it sound like a uniformly colored species will be unable to evolve plumage complexity.

We agree this was confusing as originally written. We added text to better justify our argument (e.g., lines 57–62: “For example, in a hypothetical, uniformly colored species with strong developmental constraints that limit independent variation in color among patches, selection on the color of any single patch would cause the whole plumage to change in tandem. By contrast, if a species is variably colored (i.e., patchy, and therefore has a more complex plumage) with few constraints on the direction of color variation for different patches, selection can act on different aspects of coloration (Brooks and Couldridge, 1999).”)

Line 63: are the authors going to define plumage complexity as the number of plumage patches that are different from one another? Because, if so, this will QED correlate with color variation. Revisiting this point after a few reads of the manuscript, I realize the authors are working across intraspecific and interspecific scales in this sentence. This needs revision here.

We changed this paragraph to make our points clearer, as well as removed some redundant lines (see lines 52-68).

Lines 96-98: a lot of this feels really circular. Some attention is needed here. Cutting some words, this sentence says, "promote color divergence…contributing to high rates of color evolution and disparity". How color divergence differs from rates of color evolution differs from color disparity is not clear.

We simplified this to read “Smaller population sizes, isolation, and genetic drift could promote high rates of color evolution within island kingfishers” (lines 102-103).

Line 122: change to "sex" in Figure 2, not "individual".

Fixed.

Line 127: I'm confused. Isn't Tanysiptera monomorphic?

Generally, yes. We added a multivariate phylogenetic integration test that reveals a significant correlations between male and female plumage coloration (see lines 370-373). As such, we report the results of males and females averaged together in the main text, but we also include results for the males-only and females-only data sets in the supplement (see Supplementary files 1a-1e).

Lines 124-128: this immediately makes me wonder whether type (i) tends to be allopatric more often than does type (iii). I think the answer is yes.

This is a good question. We hope to tackle this idea in future research within the Todiramphus genus.

Line 137: This is in line with a recent Merwin et al. study on parrots (https://bmcecolevol.biomedcentral.com/articles/10.1186/s12862-020-1577-y).

Thank you, we have added this citation.

Lines 138-139: High phylogenetic signal in X vs Y does not mean that X is "more diagnostic of species" than Y. Re-word to clarify.

We corrected this to refer to the higher taxonomic value of characters with high versus low phylogenetic signal.

Lines 152-157: Ok, it does seem like the authors define plumage complexity as the number of uniquely colored patches.

We appreciate this comment, and we modified the text to clarify that we are counting up the “number of uniquely colored contiguous patches on the body” (see line 145).

Lines 180-187: I think there are a few key results here that deserve more space.

We have now added an additional result showing a significant effect of body size on plumage complexity and discuss this finding in line 187-189.

Line 206: the result is not that they are de novo evolving new colors, correct? Rather, they are shifting around in color space more quickly. Clarify.

We changed the text to “species distributed on islands have faster rates of chromatic color evolution.”

Lines 209-225: There are four substantial issues in this paragraph. First, anytime I hear about creating two separate trees from one to run a test I get leery. I'm not sure what the justification is here. Second, how many island species occur with a sympatric relative? I believe that the answer is not many, so we wouldn't expect interspecific competition to be relevant, correct? Third, I'm fairly sure these RPANDA functions can take a geography object to account for this varying sympatry/allopatry. No mention is made of whether this was done or, if so, how. Fourth, RPANDA isn't cited.

: Based on this and other comments made by Reviewer 1, we have opted to remove this analysis from the manuscript. However, we have gone through and made sure we cite functions at first use to credit the work of other researchers.

I may be missing something, but I also believe the last sentence here is not warranted. It could also mean that drift on islands best explains the observed pattern.

We changed the last sentence to incorporate this idea that drift on islands may explain the observed pattern.

Line 237: in what taxon?

We are not sure what the reviewer is referring to. Original line 237 stated “mainland and island species (Figure S5). This is distinct from Doutrelant et al. (2016), who showed predictable trends…” We hope that the revised paragraph has addressed this concern (e.g., see lines 195-205).

Line 247: I'm confused. Earlier you said this author DID find predictable patterns towards duller plumage. You also say it again in the discussion below.

We removed the lines about the earlier study.

Line 289: What island is this? Does New Guinea count as an island?

We do not count New Guinea as an island in our analyses because of recent and repeated connections to Australia across the Sahul Shelf during Pleistocene eustacy. The island is Halmahera Island, as we now note in the manuscript (line 100).

Lines 273-293: I am not following the logic in this paragraph. I see no reason why not finding evidence for interspecific competition influencing plumage patterns makes any statement about the relevance of intraspecific competition. This is further confused by the first discussion in the manuscript of sympatry and allopatry, which is then not actually tied into the mechanisms of interest (competition and character displacement). This is a weak point of the paper as it stands.

We changed the line about intraspecific competition and reframed the paragraph to focus on competition alongside other non-biotic factors that could explain these patterns (see lines 245-270).

Lines 300-309: I'm still a little confused about the differences between plumage complexity and evolutionary rates (it would be helpful to keep reminding the reader that one is within a species, and one is between species), but I think some of what the authors are saying here can be attributed to differences between a phylogenetic independent contrasts-esque approach (what the authors use), and a phylogenetic generalized least squares approach (what others tend to have used). The former tends to be more conservative in my experience. Per a comment above, I'm withholding too many thoughts about alternative tests until I get a clearer picture of what the authors want to test, but I suspect this could be done as a phylogenetic t-test, where the test is whether plumage complexity differs between island taxa or not.

We have reworked our analytical pipeline. This involved removing the node-based approach and replacing it with PGLS-like approaches (phylolm for univariate traits like our complexity metrics, d-PGLS for multivariate coloration).

Figure 7: very challenging to see. It might be clearer with a white background.

We removed this figure in response to comments made by R1 and the editors.

Figure S6: "Effect of data analysis on rate-complexity correlations". Data analysis is too vague. Can you come up with a more specific figure legend?

This figure has been removed.

Line 43: errant parentheses.

Fixed.

Lines 92-94: run-on sentence, passive voice. Revise.

We changed it to: “Although the family is widely distributed across the globe, their center of diversity is the Indo-Pacific, including island clades in Wallacea and Melanesia that have recently been highlighted for their high diversification rates (Andersen et al., 2018).”

Line 103: missing a word after color.

We have added “divergence” after color.

Line 291: are->is

Fixed.

Line 411: the website for this paper incorrectly lists Ian PF Owens as just PF Owens. If you check the original manuscript, it correctly lists it as IPF Owens.

Fixed.

Reviewer #3 (Recommendations for the authors):

My line-by-line suggestions follow, I hope are helpful:

Line 15: This is entirely pedantic, but children typically leave the cradle, whereas oceanic island lineages often remain confined there until extinction or replacement (as in Wilson's taxon cycle). Maybe there's a better analogy out there.

We appreciate the point, and this line has now been removed, following suggestions by the other reviewers on how to revise the Introduction.

Lines 57-58: Constraints on morphospace need not constrain evolutionary rates, see Goswami et al. 2014.

We appreciate this important point. We have added a more recent reference (Felice et al., 2018) near this point (line 57) that discusses the idea that rates need not be constrained by development or integration.

Line 60: Defining complexity can be a difficult topic to agree on in evolutionary biology and I think that considering the centrality of pattern complexity to the study, this would benefit from being discussed in more detail. Complexity at one level (phenotype) may not match with complexity at others (biochemistry, development, genome). I could see a working definition in the methods, as well as a supporting citation or two as a way to resolve this issue.

We added several details to the Methods (lines 337-359) and a new workflow figure illustrating the ways we are calculating plumage complexity (see Figure 1). We hope these changes have addressed this concern.

Lines 66-84: Can this paragraph be condensed and combined with the one below? I'm not seeing how it's relevant enough to the paper's key ideas to merit this much-lit review text. I don't want to hold this against the authors though, because I'm getting a subtle impression that a previous reviewer might have requested it, and nothing is more infuriating than having reviewers from different journals disagree with each other.

We condensed and combined the two paragraphs mentioned. We think it reads much cleaner now (see lines 36-51).

Line 82-84: I don't disagree with this outright – it does seem to be the case in starlings – but I would consider this to be a hypothesis that is actively being researched rather than something that can be asserted. In many avian taxa, structural color seems to be static as white, but maybe that's also pedantic.

We added a caveat (italicized) – “are considered key innovations in some clades (e.g., African starlings; see Maia et al., 2013b)”

Color Ordination and Comparison: I'm fond of the authors' approach of summarizing color as a multivariate trait. However, I'm not altogether certain I understand it precisely because different parts of it are described in different places throughout the manuscript. Perhaps a figure panel with a simple flow chart connecting the color and comparative analyses performed would be helpful to readers.

Great idea, we have implemented it in a new Figure 1.

Lines 125-128, Figure 2B: I had some difficulty placing this analysis in the context of the study's design or tying it to any of the main hypotheses or questions. If the interpretation is that this highlights which clades tend to be the most complex, maybe a sentence about that could go in the figure legend?

Great idea, we added the line “Clades with more complex plumages tend to have a higher proportion of among-patch variation” to the legend (lines 685-686).

Lines 213-222: I found this part of the paperless more convincing than the rest due to issues described in Uyeda et al. (2016, Syst Biol). Even with a phylogenetic PCA, we should expect biases in model fit when individual principal components are used, particularly around OU models (also see Adams and Collyer 2018, Syst Biol). So while I would interpret Figure 5B in the same way as the authors and have no reason to disagree with them, I am skeptical it allows us to really reject competition as an explanatory mechanism here.

We have opted to address competition in an alternative way (e.g., including the number of sympatric species as a predictor of color rate variation in our PGLS models).

Lines 253-254: This is really interesting and I'd love to see more written about this idea!

Thank you for the encouraging remark.

Lines 273-293: This is tangentially related to the previous comment, reading the discussion it felt like we dive into adaptive explanations of color without much space to discuss drift or liability/availability of different color mechanisms (these are however discussed in great detail in the introduction).

Good point. We have now condensed much of the introduction (see lines 36–51) and added further discussion about color mechanisms and rates of evolution (e.g., see lines 256-270: “Another mechanism that could explain the observed rapid color evolution on islands is divergence in abiotic factors among islands…”).

Figure 1: This is a great figure, I really enjoyed it.

We appreciate this positive comment about this figure.

Figure 3: When I see an ancestral state reconstruction like this where nearly all change is independently derived, it makes me skeptical of the model. Highly complex patterns seem to be prevalent in the Alcedininae, but I couldn't find any ancestral nodes with complexity scores half as high as extant species. Since these ancestral states are used as data in downstream analyses, it might be worth testing the robustness of those analyses to uncertainty in those states, or at least describing that uncertainty.

Thank you very much for pointing this out. We realized there was an error in that we were showing ancestral estimates derived from a scaled tree, but the tree shown was unscaled, thus making the ancestral states appear “off”. We would like to point out that our analyses were correct, it was just an issue with visualization. The issue has been corrected in the new figure (see Figure 4).

Figure 4: The tip colors seem to refer to a legend that is no longer in Figure 3, but in Figure 6.

We fixed this in the revised version (old Figure 3 was merged with Figure 4, and the island legend added; see new Figure 4).

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Ln 23 "island insularity" revise.

We removed "island"

Ln 29-31 "Importantly, we found that island species did not have more complex plumages than their continental relatives. Thus, complexity may be a key innovation that facilitates response to relaxed (or divergent) selection pressures on islands." We cannot see how the second sentence follows the first, if island bids do not have higher plumage complexity, how can complexity be a key innovation that facilitates island living?

We rephrased to:

"Thus, complexity may be a key innovation that facilitates evolutionary response of individual color patches to relaxed (or divergent) selection pressures on islands rather than being a direct target of selection itself" (see L29-32).

We hope this makes our point more clear–that complexity is not itself changing on islands but enables change of the color of individual patches under drift or selection on different islands.

Ln 145 chromatic variation includes hue and saturation.

Noted and added this distinction where applicable (e.g., L144, 660, 671).

Ln 145 "achromatic color" does not make much sense; suggest referring to chromatic and achromatic variation throughout the paper rather than chromatic and achromatic color.

We appreciate this comment and revised all instances throughout (e.g., to "achromatic variables" in lines 350, 405, 671, or "achromatic variation" in lines 157, 726).

Ln 146 similarly "brightness" tends to be used in a multitude of ways, maybe replace with lightness or light-to-dark variation.

We implemented the suggested wording throughout the manuscript.

Ln 166-169 "To determine rates of overall plumage evolution, we used a phylogeny by McCullough et al. (2019), which incorporated thousands of ultraconserved elements (Faircloth et al., 2012) for a fully sampled, time-calibrated phylogeny of the avian order Coraciiformes (kingfishers, bee-eaters, rollers, and allies)." Revise sentence, "…we used a phylogeny…for a phylogeny…" something is off here, verb lacking.

We changed the wording to "To determine rates of overall plumage evolution, we used a recent time-calibrated phylogeny (McCullough et al. 2019) that incorporated thousands of ultraconserved elements (Faircloth et al., 2012) and fully sampled the avian order Coraciiformes (kingfishers, bee-eaters, rollers, and allies)." We hope the editors agree this is more clear.

Ln 230 maybe "theory" is too strong here, replace it with "idea"?

We changed "theory" to "idea" as suggested (see L229).

Ln 253 Is this the case? You mention Table 3 which refers to rates of colour evolution, not plumage complexity. This needs some clarification, as in Table 2 you do show that island species do not have more complex plumages. As a matter of fact, some sentences in this paragraph contradict the previous one.

We carefully re-read this paragraph and reworked it to make our points more clear. For example, we moved the line about lizard dewlap diversity up to where we discuss our result of increased plumage diversity on islands (L251), added clarification (e.g., "low amounts of plumage variation" in L256), and removed redundant lines (e.g., a line in the middle of the paragraph starting with "However, we found the opposite…" was redundant with current L250).

Ln 390 some justification for the inclusion of body mass here may be good.

We added justification to L386:

"Body mass has recently been shown to explain variation plumage complexity of passerine birds (Cooney et al., 2022), therefore we also included ln body mass (in grams) as a covariate in our regression models (species averages obtained from Dunning, 2007)."

Ln 400 typo "drivers".

We fixed the typo.

Table 3 We do have some concerns with multicollinearity here. As shown in Figure S1, different estimates of plumage complexity seem strongly intercorrelated. Thus, when your best model for chromatic variation identifies c1 and c3 as important predictors, the effects that your model quantifies constitute the effects of one predictor controlling for variation in the other. Now, if both are strongly correlated there is not that much independent variation that is relevant to explain. We think that it would be important to fit models with each predictor separately in order to check that results follow the same pattern. Moreover, this would mean reporting more complete results independent of whether the model is the "best" model identified by AIC. Potential future meta-analysts will need this information regardless of statistical significance or AIC. Please present the results from all models tested.

We appreciate this comment. We now calculate variance inflation factors (which we not are mostly rather low, which a max of ~4) and present results after removing each complexity metric in turn. As suggested, we also include the full regression results. We describe our approach in Methods L409-415 and included new Figure 3—figure supplement 1 (showing the "full" model results and sub-models with complexity metrics dropped in turn) and Supplementary files 1f and 1g (results for all models tested). Effect sizes were ~stable under these alternative model formulations (Figure 3—figure supplement 1), suggesting our results are robust to the observed levels of multicollinearity.

We would like to point out that the new Supplementary files 1f and 1g are rather unwieldy (63 models tested in each table). As such, we would ask the editor(s) to consider whether the new Figure 3—figure supplement 1 is sufficient for addressing this important point about potential multicollinearity influencing our results.

https://doi.org/10.7554/eLife.83426.sa2

Article and author information

Author details

  1. Chad M Eliason

    1. Grainger Bioinformatics Center, Field Museum of Natural History, Chicago, United States
    2. Negaunee Integrative Research Center, Field Museum of Natural History, Chicago, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Funding acquisition, Writing – original draft
    For correspondence
    celiason@fieldmuseum.org
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8426-0373

  2. Jenna M McCullough

    Department of Biology and Museum of Southwestern Biology, University of New Mexico, Albuquerque, United States
    Contribution
    Writing – original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7664-3868

  3. Shannon J Hackett

    Negaunee Integrative Research Center, Field Museum of Natural History, Chicago, United States
    Contribution
    Funding acquisition, Writing – original draft
    Competing interests
    No competing interests declared

  4. Michael J Andersen

    Department of Biology and Museum of Southwestern Biology, University of New Mexico, Albuquerque, United States
    Contribution
    Funding acquisition, Writing – original draft
    Competing interests
    No competing interests declared

Funding

National Science Foundation (EP-2112468)

  • Chad M Eliason

National Science Foundation (EP-2112467)

  • Michael J Andersen

National Science Foundation (DEB-1557051)

  • Michael J Andersen

The funders had no role in study design, data collection, and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Ben Marks for assistance with bird specimens at the FMNH. We also thank Kristopher Menghi who helped with the collection of spectral data. This work was partially supported by grants from the National Science Foundation (NSF EP 2112468 to CME and SJH, NSF EP 2112467 and DEB 1557051 to MJA).

Senior Editor

  1. Christian Rutz, University of St Andrews, United Kingdom

Reviewing Editor

  1. Kaspar Delhey, Max Planck Institute for Ornithology, Germany

Reviewer

  1. Kaspar Delhey, Max Planck Institute for Ornithology, Germany

Version history

  1. Received: September 13, 2022
  2. Preprint posted: September 27, 2022 (view preprint)
  3. Accepted: March 22, 2023
  4. Version of Record published: April 21, 2023 (version 1)

Copyright

© 2023, Eliason et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 796
    Page views
  • 118
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Chad M Eliason
  2. Jenna M McCullough
  3. Shannon J Hackett
  4. Michael J Andersen
(2023)
Complex plumages spur rapid color diversification in kingfishers (Aves: Alcedinidae)
eLife 12:e83426.
https://doi.org/10.7554/eLife.83426

Categories and tags

Research organism

Further reading

  1. Listen to Chad Eliason talk about the evolution of colour in birds.

    Podcast

    More complex plumages result in more diverse feather colours in kingfisher birds

    1. Computational and Systems Biology
    2. Ecology

    Gut Health: The value of connections

    Vanessa Rossetto Marcelino
    Insight

    High proportions of gut bacteria that produce their own food can be an indicator for poor gut health.

    1. Ecology

    Engineering multifunctional rhizosphere probiotics using consortia of Bacillus amyloliquefaciens transposon insertion mutants

    Jingxuan Li, Chunlan Yang ... Zhong Wei
    Research Article

    While bacterial diversity is beneficial for the functioning of rhizosphere microbiomes, multi-species bioinoculants often fail to promote plant growth. One potential reason for this is that competition between different species of inoculated consortia members creates conflicts for their survival and functioning. To circumvent this, we used transposon insertion mutagenesis to increase the functional diversity within Bacillus amyloliquefaciens bacterial species and tested if we could improve plant growth-promotion by assembling consortia of highly clonal but phenotypically dissimilar mutants. While most insertion mutations were harmful, some significantly improved B. amyloliquefaciens plant growth-promotion traits relative to the wild-type strain. Eight phenotypically distinct mutants were selected to test if their functioning could be improved by applying them as multifunctional consortia. We found that B. amyloliquefaciens consortium richness correlated positively with plant root colonization and protection from Ralstonia solanacearum phytopathogenic bacterium. Crucially, 8-mutant consortium consisting of phenotypically dissimilar mutants performed better than randomly assembled 8-mutant consortia, suggesting that improvements were likely driven by consortia multifunctionality instead of consortia richness. Together, our results suggest that increasing intra-species phenotypic diversity could be an effective way to improve probiotic consortium functioning and plant growth-promotion in agricultural systems.