Size differences among canaries, goldfinches and allies may explain correlated evolution of song and colour

1. Introduction

There is a long history of questioning whether different types of sexual signals evolve in a correlated manner, dating back to Darwin, who noted that drab bird species would sometimes have extremely elaborate songs, while colourful species could have simple songs [1]. Evolutionary trade-offs between the elaboration of different types of sexual signals have been hypothesized if, for example, resource limitation prevents investing in more than one type of signal [2], if the evolution of an elaborate signal decreases the usefulness of another type of signal [3,4] or if constraints affecting a signal modality increase selection on another signal modality [1]. The trade-off between the elaboration of different types of sexual signals is often referred to as the transference hypothesis. Alternatively, species’ differences in the strength of sexual selection could cause positively correlated evolution among different types of sexual signals. Positively correlated evolution would be due to species under weak sexual selection having simple signals and species under strong selection evolving more elaborate signals across sensory modalities (e.g. visual, acoustic).

Empirical research has found diverse patterns of correlated evolution between sexual signals, and it is not clear why these patterns differ from one clade to another. Most comparative tests used colour ornamentation and acoustic signals in birds. While in some clades there is evidence for positively correlated evolution of colour ornamentation and the elaboration of acoustic signals [5–11], in others negatively correlated evolution may have occurred [12–14], and in others, their evolution was largely independent [15–18]. Gomes et al. [18,19] also found that, in estrildid finches, visual and acoustic signals have evolved in association with different ecological or social predictors (e.g. song performance with investment in reproduction, colour ornamentation with gregariousness). Therefore, they suggested that whether or not different signals evolve in a correlated manner may depend on how, in a given clade, the selective pressures and constraints affecting each type of signal happen to be co-distributed across species [18].

To evaluate why different signals may evolve in a correlated manner, we tested how body size and habitat type affect the evolution of both song and colour in canaries, goldfinches and allies. Body size and habitat type may plausibly influence the evolution of acoustic and visual signals. Body or bill size can, for example, hinder the performance of rapidly modulated vocalizations [20–23]. Furthermore, across birds in general, differences in size among species are also associated with life-history and behavioural traits that may affect sexual selection (e.g. longevity [24–26], relative brain size [27,28]). Also, densely vegetated habitats may, for example, conceal visual signals, change the light environment in which those signals are perceived [29] or cause sound refraction that degrades the temporal structure of acoustic signals [30]. Depending on the effects of body size or habitat on signal evolution, and on how body size or habitat are distributed across species, they could influence patterns of correlated evolution among signals.

Species’ differences in song and colour ornamentation were previously characterized for canaries, goldfinches and allies, and evidence has also accumulated for sexual selection on their song and colour (reviewed in [31]). These species differ in song complexity, from continuous singers with a large variety of syllable types in each song, to species with just one or a few syllable types per song [32,33]. They also differ in song motor performance, with larger species singing low-performance trills (i.e. with slow trill rates and/or narrow frequency bandwidth) and smaller species singing either low- or high-performance trills [23]. This triangle-shaped co-distribution of body size and song motor performance indicates an upper limit of song performance that decreases with increasing size [23]. Carotenoid-based colour ornamentation is very common in the breast of canary relatives, showing male-biased sexual dichromatism typical of sexual signals, and species differ strongly in breast colour saturation (from whitish to vividly coloured) but less so in hue (yellow in most species and only rarely red [34]). Carotenoid-based colour is also common dorsally in the rump, but the rump colour is more similar between the sexes and does not show male-biased sexual dichromatism, suggesting non-sexual functions [34]. Breast colour saturation thus allows an objective comparison of homologous ornamentation across canaries, goldfinches and allies, without the need to use an indirect index such as sexual dichromatism, which is not as easily interpretable [35]. We, therefore, use within-song syllable diversity and motor performance of trills to quantify song elaboration, and male breast colour saturation to quantify colour ornamentation. We test for the possibility of correlated evolution between these signals, and assess how body size or habitat influence their pattern of correlated evolution.

2. Methods

We used data on song syllable diversity and motor performance of canaries, goldfinches and allies from [23], and data on breast colour saturation from [34]. These previous articles compiled data for song plus colour, and molecular phylogenetic information, for 35 species in the genera Acanthis (A. flammea), Carduelis (C. carduelis, C. citrinella), Chloris (C. chloris), Crithagra (C. albogularis, C. atrogularis, C. burtoni, C. capistrata, C. citrinelloides, C. dorsostriata, C. flaviventris, C. gularis, C. leucopygia, C. mennelli, C. mozambica, C. rufobrunnea, C. striolata, C. sulphurata, C. totta), Linaria (L. cannabina), Serinus (S. alario, S. canaria, S. canicollis, S. pusillus, S. serinus) and Spinus (S. atratus, S. barbatus, S. magellanicus, S. pinus, S. psaltria, S. spinescens, S. spinus, S. tristis, S. xanthogastrus, S. yarrellii). These species were formerly classified as Serinus (canaries and allies) and Carduelis (goldfinches and allies), before the discovery that Serinus and Carduelis were polyphyletic genera in a single clade [36,37]. They represent 51% of the species formerly classified as Serinus or Carduelis [38]. The remaining did not have either song recordings at the National Sound Archive of the British Library [23,32,33] or skins at the Natural History Museum of London [34].

Methods on extraction of syllable diversity and motor performance are detailed elsewhere [23]. Briefly, we analysed up to five song recordings from the National Sound Archive of the British Library for each species, selected based on sound quality; representative spectrograms for each species are in [32,33]. We used spectrograms in Avisoft-SASLab Pro (Avisoft Bioacoustics, Berlin, Germany) to identify which syllables (i.e. sound elements or tight groups of elements temporally separated from others within the song) were or were not repeated to form trills, and evaluated within-song syllable diversity as the mean number of syllables per song times the proportion of non-repeated syllables. In trills (i.e. repetitions of the same syllable), we used automatic measurement tools of Avisoft-SASLab Pro to separate syllables, measure their duration and measured maximum and minimum frequency (frequencies at which the amplitude fell below −21 dB from the peak amplitude in the syllable). From these data, we computed the trill rate of each trill, and computed the frequency bandwidth of the trilled syllables on a ratio scale to avoid overestimating the bandwidth of trills using higher sound frequencies relative to those at lower frequencies [39]. We used double quantile regression [40] to compute double quantile vocal deviation (hereafter, DQ vocal deviation) for each trill, to evaluate motor performance in relation to the trade-off between singing with fast trill rate or wide frequency bandwidth (see figure 1a of [23]). We averaged DQ vocal deviation for every trill in each recording and then for every recording of each species, to obtain a species-level value of song motor performance. For Serinus totta, we did not have data on DQ vocal deviation because there were no trills in its recordings, and for Carduelis spinescens, the value of DQ vocal deviation was a clear outlier [23]. The sample size for analyses involving DQ vocal deviation was therefore 33 rather than 35 species.

• Download figure
• Open in new tab
• Download PowerPoint

Methods on colour measurements are detailed elsewhere [34]. Briefly, we used reflectance spectrophotometry on up to three male skins for each species at the Natural History Museum of London to measure their most vividly coloured patch of plumage in the breast. We computed colour saturation as the distance to the achromatic centre of a tetrahedral colour space defined by a model of the avian UV visual system, where coordinates are the relative stimulation of the four avian cone types [41,42]. This measure of saturation (r) quantifies perceived chromaticity of colour irrespective of brightness or hue. The dataset per species of song traits and colour saturation is in the electronic supplementary material.

We used molecular phylogenies made by Cardoso et al. [23], based on DNA sequences for up to 14 loci and inferred with Bayesian models of molecular evolution. A set of 1000 post burn-in trees were extracted and rooted to obtain chronograms. Detailed methods are in Cardoso et al. [23], and the 1000 rooted chronograms are in their electronic supplementary material.

We tested if song and colour evolved in a correlated manner with phylogenetic generalized least square (PGLS) regressions [43] of syllable diversity or DQ vocal deviation on male breast colour saturation. We used song as a dependent variable and colour as a predictor because male colour saturation was shown to have a high phylogenetic signal (λ > 0.99 [34]), while these song traits show more labile evolution [23,33]. Here, as in all subsequent analyses, we used the pgls function in the R package caper (v. 1.0.1 [44]), estimating the parameter λ of the PGLS model to adjust the phylogenetic correction to the degree of phylogenetic signal in trait associations [45]. All PGLS models were run for each of the 1000 chronogram trees, to incorporate phylogenetic uncertainty and, following [46], we averaged the 1000 model results weighted by their Akaike weights. We checked that results remained identical if instead using non-weighed model averaging. We report weight-averaged values of standardized regression coefficients (β _st), p and also the model λ. We checked that, in these and all following PGLS models, the weighted 95% confidence intervals of β _st did not encompass zero for effects with weight-averaged p-values < 0.05, and did encompass zero for effects with p > 0.05. Because earlier work found a negative correlation between sexual dichromatism and song complexity across cardueline finches [12], we repeated these two PGLS models using sexual dichromatism in the breast (male minus female saturation) rather than male saturation.

We tested if male breast colour saturation, syllable diversity or DQ vocal deviation were predicted by species differences in body size or in vegetation density of habitats. As in earlier work, we used body length (midpoint from ranges in [38]) as the measure of size because this measure is available for all species and correlates strongly with both body mass and bill size [33], and we took a score of vegetation density for each species from [47], which categorized the typical vegetation of habitats as 1-open, 2-semiclosed or 3-closed. These data are also provided in the electronic supplementary material. We ran PGLS multiple regression models with breast colour or a song trait as the dependent variable and, as predictors, body size and vegetation density. Finally, we tested again for an association between syllable diversity or DQ vocal deviation and, as a predictor, male breast colour, but now controlling for effects of body size and vegetation density, including them as additional predictors in the PGLS multiple regression model.

3. Results

Male breast colour saturation did not predict species’ differences in within-song syllable diversity (β _st = 0.12, p = 0.52; figure 1a ), but species with more saturated breast colour on average sang trills with higher motor performance (β _st = 0.43, p = 0.01; figure 1b ). The estimated λ for both PGLS models was zero, indicating a weak phylogenetic signal for the associations between colour and song. When using sexual dichromatism in the breast, rather than male saturation, again coloration did not predict syllable diversity (β _st = 0.16, p = 0.37, model λ = 0.001), while more dichromatic species tended to sing higher performance trills (β _st = 0.32, p = 0.07, model λ = 0.001).

There was a non-significant trend for larger species to have less syllable diversity in songs (partial β _st = −0.33, p = 0.06; figure 2a ). Larger species sang trills with lower motor performance (partial β _st = −0.56, p = 0.001; figure 2b ), and also had less saturated breast colour (partial β _st = −0.42, p = 0.01; figure 2c ). These PGLS models did not find associations between the vegetation density of habitats and either colour or song traits (table 1).

• Download figure
• Open in new tab
• Download PowerPoint

Controlling for body size and vegetation density, male breast colour did not predict species differences in syllable diversity or in the motor performance of trills (both partial |β _st| < 0.21, both p > 0.21; table 2).

4. Discussion

Song performance and colour ornamentation evolved positively correlated with each other across canaries, goldfinches and allies, and both evolved in a correlated manner with body size, but likely for different reasons. In the case of song performance, larger and heavier bills may limit the speed of bill movements [48] that are required for modulating sound frequency [49–52], contributing to slower trill rates or narrower frequency bandwidths. Since body and bill sizes are strongly correlated in these canary relatives [33], the triangle-shaped co-distribution of body size and song performance that we found, whereby smaller species can sing either low- or high-performance trills but larger species sing low-performance trills, strongly suggests that size limits maximum motor performance (see also [23]). In the case of colour, it is less clear whether body size per se should limit ornamentation ([53], but see also [54]) but, across birds in general, body size is associated with some life-history and behavioural traits that can weaken sexual selection. For example, body size is often indicative of longevity [24–26] or of larger relative brain size [27,28], both of which are associated with more investment in long-term pair bonds and less in extra-pair paternity [55,56].

Colour ornamentation and song motor performance appear to have evolved in a correlated manner due to their associations with body size, since no correlated evolution between colour and song was apparent when controlling for species differences in size. It is plausible that, to some extent, the positive correlation between song performance and colour ornamentation results from both traits responding independently, but in somewhat similar ways, to variation in the strength of sexual selection because there is evidence for sexual selection on these two traits in canary relatives [31]. Their colour saturation reflects the concentration of carotenoid pigments in feathers [57] and, at least in some species, colour saturation functions as a sexual signal [58–60]. Song performance too, involving fast frequency modulation and/or syllable rate, is preferred by females or, most often, mediates male–male aggressive communication in several species (e.g. [61–67]), including canary relatives [68–70]. Therefore, the correlation between song performance and colour ornamentation may partially result from them responding to species differences in sexual selection, which is compatible with the traditional hypothesis of positively correlated evolution between sexual signals. However, the associations of song performance and colour ornamentation with body size are, at least in part, likely for different reasons (physical constraint for song performance [23], and possible influences of life history for colour). This means that the correlated evolution between song and colour was, at least in part, contingent on how different constraints and selective pressures affected each type of signal and how they happened to be co-distributed across species. The role of constraints limiting song motor performance in larger species is particularly evident, since small species can evolve either high or low motor performance whereas large species appear constrained to only have low motor performance (figure 2b ; see also figure 4a in [23]). Our results, thus, support the hypothesis that whether or not sexual signals evolve in a correlated manner is influenced by how different constraints and selection pressures affecting each signal covary across species in a particular clade [18].

In addition to body size, we also considered habitat vegetation density as a potential influence on the evolution of signals, but it did not predict species’ differences in the song or colour traits studied here. We showed before that species of canaries, goldfinches and allies from more densely vegetated habitats sing fewer trills, likely because vegetation-caused refraction degrades the temporal structure of acoustic signals [30], but that vegetation density does not predict species differences in song syllable diversity or in finer aspects of trill structure (trill rate, frequency bandwidth and overall trill motor performance [23]). For colour ornamentation, vegetation density influences the amount and type of ambient light [29] and can, thus, influence colour evolution (reviewed in [71]). For example, in Phylloscopus warblers and in female fairy-wrens, species’ differences in colour ornamentation are predicted by the amount of light in their habitat, which in turn is determined by vegetation density [72,73]. However, unlike those species, the vast majority of canaries, goldfinches and allies are not territorial and they make frequent vagrant movements to forage for seeds [74], meaning that individuals travel and interact in habitats with variable vegetation, which perhaps explains the lack of association with ornamental colour saturation.

Unlike previous work on cardueline finches [12], we found no evidence for negatively correlated evolution between syllable diversity in songs and either ornamental colour saturation or sexual dichromatism. Negative correlations between the evolution of avian ornamentation and the elaboration of acoustic signals appear to be rare, having been reported in only three studies. First, Badyaev et al. [12] found a negative correlation between sexual dichromatism and a metric of song complexity that includes syllable diversity, across a sample of 33 cardueline finches. Here, using 35 canary relatives, a subclade of cardueline finches rather than the entire family, we could not confirm this finding, indicating that negatively correlated evolution of song and colour may not be a general phenomenon within the subfamily Carduelinae. Second, Beco et al. [13] found negative correlations between aspects of song complexity and, especially, the luminance of dorsal plumage in antwrens. Antwren colour ornamentation is, however, based on darkly melanized plumage, with males typically more uniformly dark than females, such that high luminance may often indicate less rather than more ornamentation. Third, Marcolin et al. [14] found negatively correlated evolution between sexual dichromatism and different aspects of call complexity in parrots.

This scarcity of examples for negatively correlated evolution between avian colour ornamentation and the elaboration of acoustic signals does not mean that the opposite, positively correlated evolution, is a general pattern. On the contrary, past studies and our results indicate that uncorrelated [15–18] and positively correlated [5–11] evolution between colour ornamentation and aspects of acoustic signal elaboration are both common in birds. Instead of a broadly applicable pattern of correlated evolution between avian visual and acoustic signals, what appear to exist are different clade-specific patterns of correlated evolution. A challenge for the future is to explain these differences among clades. The hypothesis that evolutionary correlations between sexual signals are largely contingent on how the constraints and socioecological factors influencing each type of signal are co-distributed across species [18] is supported by our finding that body size explains the correlated evolution of song performance and colour in canary relatives. This more contingent view of how sexual signals become correlated may help understand clade-specific evolutionary patterns.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The dataset and code for the analyses in this article are in the electronic supplementary material [75].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

G.C.C.: conceptualization, data curation, formal analysis, writing—original draft, writing—review and editing; J.M.A.: data curation, writing—review and editing; P.G.M.: conceptualization, data curation, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by Fundação para a Ciência e a Tecnologia (grants POCI/BIA-BDE/58301/2004 and PTDC/BIA-BEC/105325/2008, and contract doi:10.54499/2022.07867.CEECIND/CP1730/CT0005).