Uncorrelated evolution between vocal and plumage coloration traits in the trogons: a comparative study


Juan Francisco Ornelas, Departamento de Biología Evolutiva, Instituto de Ecología, AC, Km 2.5 Carretera Antigua a Coatepec No. 351, Congregación El Haya, Apartado Postal 63, Xalapa, Veracruz 91070, México.
Tel.: +52 228 842 1800 ext. 3017; fax: +52 228 818 7809; e-mail: francisco.ornelas@inecol.edu.mx


The costs of bird song incurred in a diversity of ways may result in trade-offs in the production and maintenance of elaborate plumage ornaments. In this paper, we examine evolutionary trade-offs between acoustic and visual signalling in trogon birds (Trogonidae). Using multiple regressions with phylogenetically independent contrasts, we found that interspecific variation in male plumage coloration was not significantly predicted by song traits (reduced by PCA) or altitude. Although plumage coloration is expected to decrease with increases in song elaboration, both groups of variables were not related. Given that song and plumage coloration traits are likely targets of sexual selection, we also examined their relationships with sexual plumage dimorphism. We found that male carotenoid-derived coloration was positively related to sexual plumage dimorphism, suggesting that sexual selection on male carotenoid-derived coloration may be stronger than on melanin- or structurally based coloration, or than on acoustic traits. Comparative studies on other bird families accounting for the effects of phylogeny as well as environmental covariates are required to test the generality of our findings in trogons.


Evolutionary trade-offs, broadly applied to include negative correlations between groups of traits among species (reviewed in Roff & Fairbairn, 2007), comprise a central theme in evolutionary biology. However, comparative studies testing for evolutionary trade-offs between traits among species rarely use a phylogenetic approach (Stephens & Wiens, 2008). Traits associated with fecundity and survival are often subject to similar trade-offs because allocation of resources to one activity will limit the others. For instance, the cost of singing may result in trade-offs in the production and maintenance of elaborate plumage ornaments (Shutler & Weatherhead, 1990; Olson & Owens, 1998; de Repentigny et al., 2000; Badyaev et al., 2002). Under this scenario, trade-offs between song and plumage traits are expected if either trait is costly, because increasing elaboration at one trait requires decreasing elaboration at another, whereas no trade-offs are expected if both song and plumage traits are under strong sexual selection. Extensive reviews of empirical studies reveal that acoustic and visual signals are costly to develop or perform (reviewed in Gil & Gahr, 2002; Walther & Clayton, 2004), reliably indicate male quality to listeners (e.g. Schluter & Price, 1993; Nowicki et al., 1998, 2002) and that thus evolve under sexual selection pressures (e.g. Searcy, 1992; Hasselquist et al., 1996). Although numerous studies of sexual selection have shown that females prefer certain song and plumage traits and that such enhance mating success (reviewed by Andersson, 1994; Catchpole & Slater, 1995), the interactions and evolutionary trade-offs between these two kinds of traits are not well understood (Shutler & Weatherhead, 1990; Badyaev et al., 2002).

We used comparative methods to examine evolutionary trade-offs between song and plumage coloration traits in trogons (Aves: Trogonidae), a small family of 39 extant bird species (Sibley & Monroe, 1990) (hereafter members of the family are referred as ‘trogons’). Trogons exhibit a diversity of song patterns ranging from highly repetitive to more melodious songs (Collar, 2001). Songs are composed mainly by monosyllabic hooting calls generally arranged in low-pitched series that speed up or become louder or softer toward the end (Johnsgard, 2000). Plumage coloration and sexual dichromatism vary widely across species (Collar, 2001), from sexually monomorphic species, such as the Cuban trogon (Priotelus temnurus), to one of the most extravagant examples of plumage dimorphism, the resplendent quetzal (Pharomachrus mocinno). One of the most remarkable aspects of plumage in trogons is the high level of iridescence and, in most, the moderate degree of sexual dichromatism, which exists in spite of seemingly obligatory monogamy (Johnsgard, 2000). Male and female plumage coloration in trogons is derived from three mechanisms: carotenoid pigments, melanin pigments and/or feather microstructure (Johnsgard, 2000), each of which is probably under different selection pressures (Owens & Hartley, 1998). Colours that result from these mechanisms may have different costs associated with their production, i.e. condition dependence of colour content, and hence signal different aspects of individual condition (Badyaev & Hill, 2003; Siefferman & Hill, 2005). In this paper, we explore the relationship between song and male plumage coloration traits after removing phylogenetic effects. If driven by production and maintenance costs, female preference, and if evolutionary changes in song limit the expression of plumage coloration, we predicted a negative relationship between the two sets of variables because of expected trade-offs (e.g. Iwasa & Pomianowski, 1994).

Ever since Darwin, plumage sexual dimorphism has often been considered to be the result of strong sexual selection on males, and thus to be mainly the result of evolutionary changes in male coloration. Population studies have confirmed that male plumage coloration patterns in sexually dimorphic species are subject to sexual selection (e.g. Price, 1984; Hill, 1990; Møller & Birkhead, 1994), and sexual dimorphism in plumage colours has been used as an index of the intensity of sexual selection in comparative studies (e.g. Read & Harvey, 1989; Møller & Birkhead, 1994; Owens & Bennett, 1994; Owens & Hartley, 1998; Dunn et al., 2001). Furthermore, there is extensive empirical evidence suggesting that the content of colour signals is condition dependent in the males and females of many taxa (reviewed in Andersson, 1994; Amundsen, 2000). Carotenoids are generally regarded as the most clearly condition-dependent form of plumage coloration in males and females (Gray, 1996; Hill, 2000). Although melanin-based coloration seems less affected by environmental conditions than carotenoid-derived coloration (e.g. McGraw & Hill, 2000), its expression might be limited by access to rare amino acids (reviewed in Jawor & Breitwisch, 2003) or minerals (McGraw, 2003). Similarly, the production costs associated with iridescent structural coloration seem also to depend on nutritional condition (McGraw et al., 2002; Hill et al., 2005; Siefferman & Hill, 2005). If sexual selection, either by male–male competition or by female choice, operates on the basis of condition-dependent and reliable indicators of quality (Andersson, 1982; Kodric-Brown & Brown, 1984; Grafen, 1990), carotenoid-derived colours are expected to increase in lineages subjected to intense sexual selection, whereas melanin and structurally based coloration may do so if coupled with an additional reliability-enforcing mechanism (Andersson, 1986; Hill, 1990, 1993; Gray, 1996; Badyaev & Hill, 2000; Jawor & Breitwisch, 2003; Siefferman & Hill, 2005).

Here, we tested the relationship between male plumage coloration (carotenoid, melanin and structural) and sexual plumage dimorphism as an indirect measure of the intensity of sexual selection. If sexual selection on plumage coloration is stronger on males than on females, a positive relationship between plumage coloration and sexual plumage dimorphism is expected; a link between some of the male plumage coloration traits and plumage sexual dimorphism would be a strong indication of sexual selection on males. Moreover, elaboration in carotenoid-derived versus melanin-based or structural coloration is expected if the former type of coloration, due to its greater conspicuousness, plays a more significant role in sexual displays relative to the latter (see Badyaev et al., 2002). Alternatively, sexual selection may favour signals of quality in both sexes, but carotenoids might not be the best signals. Most trogons consume large amounts of fruit and so have access to an abundance of carotenoids that would result in weaker sexual selection for carotenoid-derived coloration (Gray, 1996). A final prediction with sexual plumage dimorphism is that if both song and plumage coloration traits are under strong sexual selection, we expect that elaboration in song and male plumage coloration traits will be associated with higher levels of sexual plumage dimorphism.

The specific goals of this study were to: (1) produce a well-supported molecular phylogeny for Trogonidae using mitochondrial and nuclear DNA sequence data, and (2) use the phylogeny in combination with comparative methods to detect trade-offs between song and plumage coloration traits and (3) evaluate how these sets of traits correlate with plumage sexual dimorphism as an index of the intensity of sexual selection.


Data collection

Measuring song features

We obtained most recordings for 27 of the 39 species recognized by Sibley & Monroe (1990) from the Macaulay Library of Natural Sounds, the Borror Laboratory of Bioacustics, and from several commercially available tapes (Hardy, 1983; Hardy et al., 1987; Coffey & Coffey, 1990, 1993; Ross et al., 1997; Álvarez-Rebolledo, 2000; Keller, 2001; Álvarez-Rebolledo et al., 2003). The source, location, recorders and date for all of the recordings analysed are reported in Table S1.

Spectrograms of recordings for all individuals (mean ± SE 8.9 ± 1.1 individuals/taxon, = 240 individuals) were generated using Canary sound analysis software (version 1.2.4, Charif et al., 1995). Sampling rate was set at 22.25 kHz, and spectrograms were made with a 174.8-Hz filter bandwidth, 2.9-ms temporal resolution and a frame length of 512 points. Although time and frequency are traded off against one another when generating a spectrogram, the specifications of filter bandwidth and frame length are intermediate, and therefore maximize the resolution between the frequency and temporal properties of spectrograms (Charif et al., 1995). We used one to three songs per individual in recordings with clearest long series that had been collected at different locations and/or on different dates, and assumed that each represented a different individual bird. Alternatively, when field notes by recordists confirmed that more than one bird was present, we selected one song from each individual. Altogether, we included 569 trogon songs in our analysis (mean ± SE 21.1 ± 2.8 songs/taxon).

Spectrograms were first examined by naked eye to identify syllable types as units, on the basis of distinct tracings on the spectrograms. Syllables were letter coded according to González & Ornelas (2005) and categorized by C.G. for further analysis. We also measured and estimated 18 acoustic variables on each spectrogram (Fig. 1) and averaged values for each individual. Acoustic measurements were made directly from on-screen spectrograms by using on-screen measurement cursors in Canary (cursor resolution = two digits). We included most temporal, frequency and structural aspects of song. Some of the song measures were included because of energetic and physiological constraints such as neuromuscular fatigue, muscle recovery and oxygen consumption (reviewed in Gil & Gahr, 2002; Podos et al., 2004). Because song output could potentially vary among species owing to differences in singing rate or in the duration of each sound produced (Wasserman & Cigliano, 1991; Price & Lanyon, 2004) or both, we also measured total song duration and intersound pause duration to explore these possibilities. Lastly, syllable versatility as a measure of song elaboration was estimated as the number of syllable types multiplied by the number of transitions from one syllable type to another divided by the total number of syllables (see also Ornelas et al., 2002). A full description of all acoustic measurements is given in Table S2.

Figure 1.

 Fragment of a trogon song showing components and descriptive terminology as well as some of the measurements taken for each song.

Measuring plumage coloration

Variation in sexual plumage dimorphism was estimated as an indirect measure of the intensity of sexual selection on song and plumage coloration traits (see also Price & Lanyon, 2004). We did not use mating system itself as an indicator of the relative strength of sexual selection because this aspect of life history is not well known for most trogons (Johnsgard, 2000; Collar, 2001). Sexual plumage dimorphism was estimated in two ways. First, to provide a measure that is independent of our data and that is unbiased to the coloration origin (i.e. carotenoid-, melanin- or structurally based), we used the sexual dichromatism index (hereafter SD) according to Gray (1996). We recorded dichromatism per se (i.e. dichromatic and not dichromatic), rather than scoring each sex separately and then comparing the scores. This was performed because measures of dimorphism simply required the observer to record the degree of difference between the sexes rather than make a subjective judgement about which sex is, for example, ‘brighter’ than the other (Owens & Hartley, 1998; Dunn et al., 2001). Dichromatism was scored for each of 22 body parts (Fig. 2) using the trogon plates from The Birds of the World (Collar, 2001). To compute SD by species the number of dichromatic body parts were summed and converted to proportions (number of dichromatic body parts divided by 22). Note that SD is often referred to as overall plumage–coloration dimorphism, although several nonplumage parts are included (bare parts).

Figure 2.

 Body parts in males and females scored for (a) sexual dichromatism (SD), (b) plumage–colour dimorphism (PSD) and (c) in males for components of plumage coloration (i.e. melanin, carotenoid and structural) using colour plates from The Birds of the World (Collar, 2001).

Second, the extent of plumage–colour dimorphism (hereafter PSD) was measured according to Owens & Bennett (1994) on a scale from zero (monomorphic) to 28 (maximum dimorphism). Six independent observers were provided with the trogon plates and the outline of a trogon body divided into 14-feathered parts (nonfeathered body regions, undertail feathers and wings were excluded; Fig. 2). Each body part was scored: 0 = no difference in colour of a given body part; 1 = no difference in the basis of the colour but a difference in the intensity of the colour (e.g. the same melanin-derived pigment is present in both sexes but in different hues); and 2 = difference in the overall basis of the colour (e.g. structural colour present in one sex but not the other, or carotenoid-derived colour in one sex but a mixture of carotenoids and melanins in the other). We followed Gray (1996) and Owens & Hartley (1998) for the diagnosis of plumage colours. Namely, we assumed that bright yellows, oranges, reds and greens were due to carotenoid-derived pigments; that blacks, browns, greys, rufous and dull reds were due to melanin-based pigments; and that iridescent blue, black, purple and green were due to structural colours. Scoring the origin of green coloration is more ambiguous than scoring red or black because green coloration can be produced in several ways (Fox & Vevers, 1960; Brush, 1978). Then, to be conservative when assigning a carotenoid origin to coloration, noniridescent bright green was scored as carotenoid/structural (i.e. yellow plus blue), whereas iridescent green, dull green and olive were scored as melanin/structural. The overall PSD score was then the mean sum of scores for the 14 body parts between the six independent observers. Given that this measure is based on scoring by six observers, we calculated the repeatability (r that ranges from 0 to 1) across observers and individual plates using one-way anova (see, e.g. Lessels & Boag, 1987 for how to calculate repeatability). The analysis of interobserver reliability on PSD indicates that scorers were consistent (F6,2016 = 6.94, < 0.0001, = 0.49).

Lastly, male plumage coloration due to carotenoid-derived pigmentation, melanin-based pigmentation and coloration due to structurally based coloration was measured using the same trogon plates and the outline of a male trogon body divided into 17 parts (Fig. 2). Each body part was scored based on the origin of colour (carotenoid, melanin or structural). If a body part was covered by two colours that differed in origin, the body part was scored as half for both origins. The body parts with colour of a given origin were then summed and converted to proportions (e.g. number of body parts with carotenoid-derived coloration divided by 17).

Reconstructing trogon phylogeny

Deducing phylogenetic relationships within Trogonidae has been the focus of several molecular studies (Espinosa de los Monteros, 1998, 2000; Johansson, 1998; Johansson & Ericson, 2003, 2005; Sorenson et al., 2003; Moyle, 2005). Our phylogenetic analysis focused mainly on the relationships within the New World trogons, which includes Pharomachrus, Euptilotis, Trogon and Priotelus. We obtained DNA sequences from 22 individuals representing nearly all extant species of New World trogons (Sibley & Monroe, 1990) and from eight representatives of the Old World trogon taxa (Harpactes and Apaloderma). We used sequences from three mitochondrial and four nuclear genes for constructing a molecular phylogeny. Most cytochrome b (cyt b; 1143 bp) and 12S ribosomal RNA (12S; 1016 bp) sequences were obtained from the study by Espinosa de los Monteros (1998). Sequences for the NADH dehydrogenase subunit 2 (ND2; 1042 bp) and the recombination activating protein 1 (RAG-1; 2873 bp) were taken from the study by Moyle (2005). Sequences of the glyceraldehyde-3-phosphate deshydrogenase intron 11 NADPH (358 bp) and the myoglobin intron 2 (734 bp) were taken from Johansson & Ericson (2005) and β-fibrinogen intron 7 (939 bp) was obtained from Fain & Houde (2004). Finally, cyt b and 12S sequence data for P. mocinno, Trogon citreolus, Trogon melanocephalus and Trogon massenawere not available in published analyses and were obtained from Quintero-Rivero & Espinosa de los Monteros (unpublished data).

Our phylogenetic analyses included a total of 30 trogon taxa and 143 gene sequences, ranging from 2159 to 8105 bp. DNA sequence data sets were only partially overlapping in their sampling of genes and taxa. We included only a single individual or terminal taxon per species, given that our study does not focus on the details of species-level relationships within genera. In several cases, we combined data from different genes that had been sequenced for different individuals of the same species. The dissimilar origin of some sequences, however, is a potential source of error for phylogenetic reconstruction (e.g. mixture of different phylogroups and fusion of nonsister lineages). As a result of combining partially overlapping data sets from different studies, many taxa lacked data for one or more genes. On average, each trogonid species had data for 4.8 of 7 genes (range 2–7 genes) and lacked data for 31.9% of their characters (range 0–71.4%). Although some taxa had data from only two genes, recent empirical studies and simulations (e.g. Wiens, 1998, 2003; Wiens et al., 2005) suggest that including characters with incomplete taxon sampling can improve phylogenetic accuracy, relative to excluding these sets of characters entirely. Thus, we are confident that the cladogram resulting from the combined data set would produce a robust hypothesis supported by taxonomic information and available data. GenBank accession numbers for used sequences are reported in Table S3.

Combined data from all seven genes were analysed using maximum likelihood (ML) and Bayesian methods. ML was used as a selecting criterion for phylogenetic reconstruction and implemented in paup* version 4.0b 10 (Swofford, 2002). The computer program Modeltest 3.04 (Posada & Crandall, 1998) was used to statistically compare successively nested models and to determine the appropriate model of sequence evolution for our data set. The best-fit model for the complete data matrix corresponded to the GTR + I′ + invariant sites. The likelihood settings used were Gamma distribution shape = 0.4492, proportion of invariable sites = 0.4575, base frequencies (A = 0.3188, G = 0.2869, C = 0.1816, T = 0.2127) and rate matrix (1.6295, 8.0770, 1.7310, 0.5030, 20.6357 and 1.0000). A molecular clock was not enforced, and the starting branch lengths were obtained using the Rogers–Swofford approximation method. The tree space searches were performed with the Ratchet algorithm (Nixon, 1999) in paup* Rat (Sikes & Lewis, 2001). We performed 20 repetitions of the Ratchet with 200 iterations per run, as suggested by Sikes & Lewis (2001). For the resulting tree we estimated clade robustness with 500 full heuristic bootstrap replications.

Bayesian analyses were also implemented because Bayesian inference allows specific evolutionary models to be assigned for each of the data partitions (Huelsenbeck et al., 2002). Bayesian trees and posterior probabilities for clade reliability were generated with MrBayes, version 2.0 (Huelsenbeck & Ronquist, 2001) using the Markov chain Monte Carlo with the Metropolis–Hasting algorithm. Each gene fragment was declared as an individual partition. The best-fit model for each of the DNA sequence data sets, as suggested by Modeltest, corresponded to: GTR + I′ + invariant sites for cytochrome b, HKY + I′ for the fibrinogen, K80 + I′ for myoglobin, TVM + I′ + invariant sites for NADH2, HKY for NAPDH and TrN + I′ + invariant sites for RAG1 and 12S. The Bayesian search was run for one million generations, with four parallel chains and sampling every 100 generations. The burn-in value was set when −ln L scores reached an asymptote. A majority-rule consensus tree after burn-in was used to calculate the posterior probabilities for each node. We have two reasons for using the DNA partitions in a total-evidence analysis. First, this strategy maximizes the informativeness and explanatory power of the data. The use of different partitions simultaneously may interact positively to resolve different nodes in a phylogenetic tree. Slowly evolving partitions might provide characters to resolve deeper splits, whereas rapidly evolving ones might resolve the tips of the tree. Second, independent analyses for the different partitions had been performed by the authors from whom we compiled the sequences (Espinosa de los Monteros, 1998; Johansson, 1998; Johansson & Ericson, 2005; Moyle, 2005). Three of the original studies produced consistent topologies, whereas Moyle’s (2005) differed only in the in-group rooting. The unrooted network, however, showed no significant differences from the other topologies.

Three of the trogon taxa (Pharomachrus fulgidus, Trogon bairdii and Trogon clathratus) were not included in our study of molecular relationships, or in any other molecular phylogenetic analysis (Espinosa de los Monteros, 1998, 2000; Johansson, 1998; Johansson & Ericson, 2005; Moyle, 2005). Given the lack of molecular data, we used plumage and song characters to predict their phylogenetic affinities. Evolutionary inter-relationships were also inferred to verify whether the species for which we had molecular data would be correctly placed using plumage and song characters only. We carried out independent analyses (i.e. molecular vs. nonmolecular data) because the available song recordings did not necessarily belonged to the same populations from which the DNA was gathered. The inclusion of nonmolecular characters when estimating a tree that is then used to analyse phenotypic data is somewhat contentious (e.g. de Queiroz, 1996; Zrzavy, 1997). By including these intrinsic characters we are assuming that they are heritable to some degree and, therefore, follow the basic logical principle in phylogenetics of descent with modification (for further discussion see Grandcolas et al., 2001). A detailed description of the methods used to select potentially useful song and plumage characters is given in Appendix S1. Here, we only briefly outline the methods used to estimate their phylogenetic relationships. Phylogenetic affinities of the three taxa were first estimated by performing maximum parsimony analyses of the entire data set (Table S4), the plumage data set and the song data set in paup*, with the molecular phylogeny obtained in our study as a topological constraint according to Price & Lanyon (2002). An additional analysis for the nonmolecular matrix consisted in 1 000 000 heuristic searches, with random addition of taxa and branch swapping made by the tree bisection-reconstruction (TBR) algorithm. Implicit weights (Goloboff, 1993) were applied to this matrix in order to minimize the homoplasy observed in the coloration characters. To evaluate the weighting effect we run five independent searches using different concavity factors (i.e. Goloboff’s K = 2, 4, 6, 8 and 10).

Comparative analyses

Our comparative analyses were divided into two groups: (1) bivariate analyses performed to establish the relationship between SD and PSD and between these indices and male plumage coloration due to carotenoid-derived pigmentation (C), melanin-based pigmentation (M) or coloration due to structurally based coloration (S) and (2) multivariate analyses to establish the relationship between measures of plumage dichromatism or male plumage coloration and song traits. Bivariate and multivariate analyses were performed on the 27 species for which we had data on all variables. Because closely related species tend to share similar trait values (i.e. exhibit a ‘phylogenetic signal’sensuBlomberg et al., 2003), comparative data sets may violate the assumptions of conventional statistical methods (Felsenstein, 1985; Grafen, 1989; Martins & Garland, 1991; Purvis et al., 1994; Garland et al., 2005). Thus, bivariate and multivariate analyses were based on phylogenetically independent contrasts (hereafter PICs) (Felsenstein, 1985; Garland et al., 1992) for all species in the data set. This method incorporates phylogeny into components of continuous variables among taxa to correct for statistical nonindependence caused by shared history (Harvey & Pagel, 1991). The most direct way of addressing evolutionary trade-offs among species (i.e. an evolutionary increase in trait A relative to the ancestral state is associated with an evolutionary decrease in trait B on the same branch of the phylogeny) is by inferring evolutionary increases and decreases in performance or fitness on a phylogeny using ancestral trait reconstruction (minimum evolution method) (Stephens & Wiens, 2008). We used PICs of raw trait values to correct for the phylogenetic relatedness of species. Our approach does not directly address evolutionary increases and decreases on performance or fitness, but it gives similar results to the minimum evolution approach under many circumstances (Martins & Garland, 1991).

Phylogenetically independent contrasts were calculated using the pdap: pdtree package of Mesquite (Midford et al., 2002; Maddison & Maddison, 2004). Because of correlations among acoustic characteristics (results not shown), we used a principal components analysis (PCA) to reduce the acoustic species data set into a small number of mutually independent variables and used the generated PCs in multiple regressions with PICs with species mean values of plumage variables and PC scores. Finally, we included altitude (mid-elevational range, metres above sea level) for each species in our analyses as a covariate because the environment affects visual signalling systems and song structure independently of phylogeny (e.g. Endler et al., 2005; Seddon, 2005).

Multiple regressions were performed with SD or PSD as the dependent variable and PCs and altitude as independent variables for independent contrasts derived from species data. We also performed multiple regressions with each component of male plumage coloration as dependent variable and PCs and altitude as the independent variables for independent contrasts derived from species data. Regression analyses of contrasts used models with no intercept (i.e. fitting regression lines through the origin; Garland et al., 1992). Based on the diagnostic of Garland et al. (1992), which checks for patterns in the plots of absolute values of standardized contrasts vs. their standard deviation, constant branch lengths worked the best and were used in all analyses (i.e. an explicitly punctuational model of character evolution). All statistical analyses were performed using spss Version 11.0.2 (SPSS, 2002) as well as pdtree.


Maximum likelihood and Bayesian analyses recovered the same phylogenetic inter-relationships among trogon species. Even though there were several short internodes, the monophyly of African trogons, Asian trogons, Quetzals, Caribbean trogons and ‘Brown’ and ‘Grey’ trogons was well supported (bootstrap and Bayesian posterior probabilities; see Fig. 3) and are similar to those postulated by Espinosa de los Monteros (1998) and Johansson & Ericson (2005). Predicted phylogenetic affinities based on song and plumage characters for the three taxa for which we lack molecular data are shown in Fig. 4. The tree topology estimated in this study was similar regardless of whether molecular and nonmolecular data were used during phylogeny estimation. The nonmolecular matrix yielded to the same most single parsimonious tree independently of the K value used. This tree was consistent with the phylogeny recovered from the DNA sequences (71% of the monophyletic groups were identified by both phylogenies). The phylogenetic position of P. fulgidus, T. bairdii and T. clathratus was identical in the molecular and in the nonmolecular tree. An independent source of corroboration for the evolutionary inter-relationships of these taxa is the recent molecular (only ND2) phylogeny published by DaCosta & Klicka (2008) that included all Trogon species.

Figure 3.

 Phylogenetic tree inferred independently from maximum likelihood and Bayesian analyses of the trogons data set. The phylogeny is rooted with representatives of Coliiformes (Collius) and Cuculiformes (Centropus, Geococcyx and Cuculux). Branch length is proportional to nucleotide substitution rate per site. Numbers are nodal support values obtained from bootstrap and Bayesian posterior probabilities respectively.

Figure 4.

 Phylogenetic tree used in testing for trade-offs between song and plumage coloration traits. Spectrograms show typical song patterns for taxa. Data for song recordings used in this study are given in Table S1. Predicted phylogenetic affinities for three trogon taxa (dotted branches) were based solely on song and plumage characters. Relationships among other taxa (solid branches) were constrained to the molecular phylogeny (Fig. 3). Four Harpactes species for which we had no song data were pruned from the final topology.

As shown in Fig. 4, trogon vocalizations were not very versatile and were composed of series, generally of repeated low-frequency sounds. For a more detailed description of the acoustic characteristics, see Table S5. For acoustic measurements, PCA generated four principal components explaining 79.6% of the variation. The contribution of each PC in explaining song variation among species is presented in Table 1.

Table 1.   Factor loadings for the first four principal components (PCA) derived from measurements of the acoustic properties of Trogonidae songs.
 Principal component analysis
  1. Predictor variables were first subjected to a PCA with the resulting orthogonal components used as predictor in phylogenetic comparative analyses. Bold denotes variables that make an important contribution to the component (factor loading > 0.4).

Variation (%)29.724.917.67.4
Total song duration (s)0.62−0.0010.530.42
Song output (s)0.430.160.680.13
Syllable duration (s)0.600.63−0.380.19
Pause duration among syllables (s)0.700.450.37−0.18
Duration of longest syllable (s)0.610.60−0.140.16
Duration of shortest syllable (s)−0.380.450.680.18
Maximum frequency (Hz)0.680.660.260.007
Minimum frequency (Hz)0.0050.58−0.290.41
Number of syllables0.0050.560.440.53
Number of syllable types−−0.27
Number of transitions0.580.300.550.10
Maximum number of notes per syllable−
Initial maximum frequency (Hz)0.650.670.110.002
Intermediate maximum frequency (Hz)0.670.680.210.007
Final maximum frequency (Hz)0.590.720.130.006
Frequency range (Hz)0.680.370.440.31
Syllable versatility0.590.500.53−0.15
Syllables per second0.690.420.150.10
n (individuals)240   

Sexual dichromatism (SD) in trogons ranged from 0 to 1 (monochromatic to fully dichromatic) and PSD ranged from 0 to 24.3 in a scale of 28 of maximum dimorphism. Components of male plumage coloration also varied among species, with carotenoid-derived plumage ranging from 12% to 29%, melanin-based plumage from 24% to 59%, and structurally based plumage from 12% to 59%. The complete plumage coloration data by species are available in Table S6.

Bivariate analyses on contrasts showed that indices of plumage dichromatism positively correlated (SD vs. PSD: t25 = 8.65, = 0.87, < 0.0001; = 26 contrasts). Among components of male plumage coloration, only melanin and structural coloration negatively correlated (t25 = −14.73, = −0.95, < 0.0001; = 26 contrasts). PICs also showed that only carotenoid-derived coloration increased with increased PSD (t25 = 2.68, = 0.47, = 0.013; = 26 contrasts). None of the three components of male plumage coloration correlated significantly with SD (> 0.05).

Phylogenetic multiple regression analyses showed that plumage dichromatism indices did not correlate with the acoustic properties of songs (defined by PC1, PC2, PC3 and PC4 respectively) and altitude (Table 2). However, the multiple regression with PICs on the relationship between components of male plumage coloration and the acoustic traits (PCs) and altitude showed that PC4 scores were negatively related to melanin-based coloration and positively related to structurally based coloration (Table 3). None of the PCs or altitude correlated significantly with the proportion of carotenoid-derived coloration in males.

Table 2.   Phylogenetic multiple regressions with phylogenetically independent contrasts (PICs) of sexual dichromatism (SD) and plumage–colour dimorphism (PSD) indices with acoustic properties of trogon songs (defined by PC1, PC2, PC3 and PC4 respectively) and altitude.
DependentIndependentUnstandardized coefficientsStandardized coefficients95% Confidence intervals for b
bSE bβtPLower boundUpper bound
  1. Full model data for multiple regressions: SD, multiple r2 = 0.08, F5,21 = 0.39, P = 0.852; PSD, multiple r2 = 0.09, F5,21 = 0.39, P = 0.847. All branch lengths were equal to 1 and linear regressions were forced through the origin for PICs. Confidence limits that do not include zero indicate that the coefficient differed significantly from zero at α = 0.05, n = 26 in all cases. All values of P are for two-tailed tests.

PC4 0.0460.0470.260.980.336−0.0510.144
Altitude 5.003E−050.0000.160.720.4770.0000.000
PC3 0.2371.3840.070.170.865−2.6403.115
PC4 1.0821.1880.240.910.373−1.3873.552
Altitude 0.000410.0020.050.230.818−0.0030.004
Table 3.   Phylogenetic multiple regressions with phylogenetically independent contrasts (PICs) of male plumage coloration due to carotenoid-derived pigmentation (C), melanin-based pigmentation (M) and coloration due to structurally based coloration (S) with acoustic properties of trogon songs (defined by PC1, PC2, PC3 and PC4 respectively) and altitude.
DependentIndependentUnstandardized coefficientsStandardized coefficients95% Confidence intervals for b
bSE bβtPLower boundUpper bound
  1. Full model data for multiple regressions: C, multiple r2 = 0.14, F5,21 = 0.67, P = 0.653; M, multiple r2 = 0.41, F5,21 = 2.92, P = 0.037; S, multiple r2 = 0.44, F5,21 = 3.35, P = 0.022. Components of male plumage coloration (C = carotenoids, M = melanins and S = structural) are proportions. All branch lengths were equal to 1 and linear regressions were forced through the origin for PICs. Confidence limits that do not include zero indicate that the coefficient differed significantly from zero at α = 0.05, n = 26 in all cases. All values of P are for two-tailed tests.



Despite the widespread view that sexually selected condition-dependent traits in birds should be traded-off against one another, few comparative studies relating visual and vocally sexually selected traits have been conducted for passerine groups (Shutler & Weatherhead, 1990; de Repentigny et al., 2000; Badyaev et al., 2002). Our comparative study is only the third to explore relationships between song and plumage traits, and to our knowledge is the very first to do so in a nonpasserine bird assemblage. Our results, however, were not consistent with the prediction that visual and acoustic traits are traded-off against one another.

We used a large number of highly co-correlated acoustic measures reduced by PCA to analyse the relationships between these measures and aspects of plumage dichromatism and components of male plumage coloration with phylogenetic multiple regressions using PICs. Based on the results of multiple regressions with species contrasts, we found that acoustic PC scores were not significantly associated with plumage dichromatism (SD or PSD) scores, suggesting weak sexual selection in trogons for tightly correlated acoustic traits. Despite negative correlations between PC4 and melanin-based plumage traits (see Table 3), PC4 scores accounted only for 7.4% of acoustic variation. Complex vocal patterns (e.g. a repertoire of distinct syllable types and syllable versatility) are likely to be sexually selected (Podos et al., 2004), in which case PC1 and PC3 scores (see Table 1) should be negatively related to measures of plumage dichromatism or components of male plumage coloration. Such relationships were found only with PC4 that reflects song duration and number of syllables (effectively the same thing); acoustic measures that reflect song quality and investment in song production but not expected to reflect the intensity of sexual selection (Podos et al., 2004). Shutler & Weatherhead (1990) found that highly dimorphic wood warblers sang shorter songs; however, plumage dimorphism did not correlate with song complexity as measured by the number of notes and repertoire size. When removing monochromatic species that generally nest on the ground and thus have developed less conspicuous plumage because of predation pressures, they noted an increase in song complexity with an increase in the degree of sexual dichromatism. This is contrary to the trade-off hypothesis originally posed by Darwin (1871). In a comparative study for 123 North American oscines, de Repentigny et al. (2000) also found that plumage elaboration increases with song complexity, although the link between these traits was rather loose. Across cardueline finch species, Badyaev et al. (2002) reported that variation in song complexity (PC scores mainly associated with an increase in song length and frequency range) was strongly negatively related to elaboration of plumage ornamentation (carotenoid-derived plumage dimorphism). However, we know of no empirical evidence suggesting that song length and frequency range are correlates of mating success and thus subject to sexual selection in these cardueline finches.

We have shown that the relationship between song and plumage coloration traits was not significant, providing no empirical support for the trade-offs between condition-dependent sexually selected traits. However, lack of a significant association between these two sets of traits is insufficient to demonstrate the absence of trade-offs. Several explanations may account for our results. First, it is possible that the sexual traits we selected were poor predictors of male quality, and thus unable to reflect the intensity of sexual selection. This is unlikely because comparative analyses have shown that elaboration in song and plumage coloration traits can evolve under the pressure of sexual selection and are good predictors of male quality (e.g. Read & Weary, 1992, Birkhead & Møller, 1992). Second, trogons were expected to exhibit traits that are more conspicuous, less costly or that improve signal condition, and to reduce the display of other traits. However, we do not know whether temporal aspects of song- and/or frequency-related acoustic traits (linked to PC4) are sexually selected condition-dependent traits in trogons. In a comparative study of 34 emberizid species, Podos et al. (2004) hypothesized that songs produced with low trill rates and limited frequency bandwidth require minimal vocal tract movements and, therefore, should be easily produced. As songs increase in either frequency bandwidth or trill rate, the demands required of vocal performance are elevated (i.e. wider gapes, greater rates of beak opening and closing). According to the ‘vocal tract constraint’ hypothesis (reviewed in Podos et al., 2004), the acoustic features of most trogon species (see Fig. 4) suggest that the least costly acoustic signals have been selected for, except in the ‘grey’ trogons (see also Table S5) where the physical challenge of producing songs with faster syllable rates might be elevated. Third, an alternative explanation for the lack of trade-offs in the trogon data set is that the trade-off is there, but there is no power in the data set. With ‘grey’ trogons as a monophyletic group, we have a single evolutionary change of plumage and song and, therefore, no power to detect the trade-off. Under the trade-off hypothesis, we expect trogons to have either bright plumage and noncostly songs or drab plumage and costly songs. However, this alternative is unlikely because members of the ‘grey’ trogon clade (with the most costly songs) are, on average, the most dimorphic ones (see Table S6). Lastly, physical features of habitats may favour certain plumage pigmentation and thereby constrain the distribution of other types of pigments or structural colours. The negative relationship between altitude and melanin-based coloration in males is perhaps associated with variation in light conditions of the environment, i.e. melanin-based coloration is favoured in habitats with higher light availability (e.g. Endler & Théry, 1996; McNaught & Owens, 2002).

In the case of associations between indices of sexual plumage dichromatisn and components of male plumage coloration, bivariate analyses with species contrasts suggested a positive relationship between male carotenoid-derived coloration and PSD. Changes in melanin- and structurally based coloration were not significantly associated with PSD. These results support the prediction that sexual dichromatism would result from brighter carotenoid-derived coloration in males, whereas changes in melanin- and structurally based coloration would contribute less in variation in sexual dichromatism. That PSD scores positively correlated with carotenoid-derived coloration is a strong indication of sexual selection for carotenoid-derived coloration on birds that consume large amounts of fruits and so have easy access to carotenoids. This does not prove that sexual selection has operated on the basis of a condition-dependent honest indicator of quality (see Gray, 1996). Our results confirm in trogons what other workers have found for passerine birds, particularly granivores and insectivores (Gray, 1996; Badyaev & Hill, 2000). We add to these results that carotenoid ornamentation is positively associated with degree of dichromatism in frugivorous trogon taxa for which carotenoids are overly abundant. The association between carotenoid-derived male plumage coloration and PSD demonstrated here is easily incorporated into good genes or condition-dependent models of sexual selection. However, an alternative hypothesis is the sensory bias model of sexual selection, in which male traits are elaborated solely because of an innate female perception bias or perhaps because of an association between red, yellow or orange stimuli and important naturally selected ecological factors, for example, red fruits as food sources (Gray, 1996). These alternatives deserve further testing.

In summary, our study provides no support for the trade-off hypothesis. One explanation for these results is that because of the interaction between factors favouring the elaboration of song and plumage traits and others constraining it, these traits are uncorrelated across trogon species. More comparative studies need to be carried out with other species before the trade-off hypothesis can be applied to all birds or universally. Further documentation of broad evolutionary patterns will stimulate detailed between- and within-species studies of the proximate mechanisms that contribute to them, particularly for other Neotropical birds with rich diversity of colour and acoustic signalling.


We are grateful to J. A. Soha (Borror Laboratory of Bioacoustics, Ohio State University) and M. Guthrie (Macaulay Library of Natural Sounds, Cornell University) for allowing us to use a large number of their recordings; A. Cruz, M. L. Jiménez, L. Mendoza, O. Rojas, R. Villegas and M. E. Quintero for help in data gathering, K. Omland, J. J. Price, J. Rull, M. Sorenson and three anonymous reviewers for valuable comments on the manuscript and B. Delfosse for polishing our English version. This study was funded by grants from the following institutions: Comisión para el Uso y Conocimiento de la Biodiversidad (CONABIO) (research grant no. FB612/R174/98) (A.E.M.) and Instituto de Ecología, A.C. (Ref. 902-12-563) (J.F.O.). We are grateful to the artist, Richard Allen; the editors, J. del Hoyo, A. Elliott and J. Sargatal; and the publisher of Collar (2001)Handbook of the Birds of the World, Lynx Editions; for permission to use Allen’s trogon paintings in Fig. 2.