Rapid evolution of elaborate male coloration is driven by visual system in Australian fairy-wrens (Maluridae)

Colour measurement and analysis

To measure plumage reflectance, we used an Avantes AvaSpec-2048 spectrometer equipped with an AvaLight-XE pulsed xenon light source and standardized with a WS-2 white reference. We examined vouchered museum specimens of 24 species in the family Maluridae at the Australian National Wildlife Collection (ANWC), with a sampling goal of five individuals from each (see Appendix S1, mean = 2.7 females, 3.6 males). We were unable to measure female specimens from Chenoramphus grayi, and this changed our total number of taxa to 23 species for analyses of female colour and dichromatism. For each specimen, we measured 11 plumage patches (Fig. 1) at a perpendicular probe angle, averaging across three repeated measurements. For species with coloration that differed between the mantle and scapulars (e.g. M. elegans), we measured the scapulars as well for the purposes of calculating a complete colour span. To ensure that the males we measured were adults in breeding condition, we sampled only individuals that showed enlarged testes and complete skull ossification when such data were available (as they are for most specimens at ANWC).

There are many metrics that may be used for quantifying coloration (Montgomerie, 2006; Butler et al., 2011); however, no single metric was appropriate for this study. Some metrics were avoided because they are effective for only a single mechanism of colour (e.g. λR₅₀; Friedman et al., 2011), or cannot be mapped onto a phylogeny due to their circular character space (e.g. θ; Stoddard & Prum, 2008; Maia et al., 2013a). As one of our aims was to test for an effect of visual system on colour evolution, it was necessary to intentionally utilize metrics that lacked explicit visual modelling so as to avoid combining the visual model and reflectance as a composite character on the phylogeny (McLennan & Brooks, 1993).

We used Endler's segment classification to describe hue (LM hue and MS hue; Endler, 1990; see Appendix S2). LM hue is the relative difference in reflectance and chroma between long- and medium-wavelength segments, and MS hue is the difference between medium- and short-wavelength segments. A limitation of this approach is that it cannot describe variation in UV reflectance. Consequently, we included analyses of UV chroma (Montgomerie, 2006) to describe this axis of variation. We also included measurements from Montgomerie (2006) that describe saturation and mean brightness. As estimates of plumage brightness are sensitive to measurement error, we performed all measurements in a darkened room at the ANWC. We performed all colour analyses in the r package pavo (version 0.5-1, Maia et al., 2013b).

To describe the degree of chromatic contrast among patches for each species, we developed an analogue of the colour span measurement described by Endler & Mielke (2005). Our measure (hereafter PCA span) is similar to colour span in that it takes a specimen's mean pairwise Euclidean distance among the reflectance spectra for its patches in colour space. However, to avoid incorporating visual system into this measure (see above), we used a colour space derived from a principle component analysis (PCA) of our reflectance spectra (centred to remove brightness; Maia et al., 2013a,b). Thus, PCA span represents the mean distance among patches in a PCA space defined by the first two principle components, which together described 92% of colour variation (see loadings in Fig. S1). We repeated the comparative analyses below using both colour span and colour volume under an average UVS visual system, which performed similarly to one another and to PCA span (Stoddard & Prum, 2008).

To describe sexual dichromatism for use in our comparative analyses, we calculated mean Euclidean distances between male and female colours as plotted in PCA space for each patch. We then took an aggregate ‘net dichromatism’ score that is the sum of mean Euclidean colour distances across all patches, compared between sexes.

Colour analysis using visual models

Birds have four retinal cone classes, and thus, their extent of potentially visible colours can be represented as a tetrahedron or as a spherical projection, with each axis representing the relative stimulation of a cone class (Endler & Mielke, 2005; Stoddard & Prum, 2008). We used visual models to estimate relative receptor stimulation values for each patch's reflectance spectrum in its native visual system (Vorobyev et al., 1998; Maia et al., 2013b), allowing us to place each measured patch in colour space as estimated for an avian UVS or VS observer.

To describe sexual dichromatism in an avian visual context, we calculated mean Euclidean distances between male and female colours as plotted in Stoddard & Prum's (2008) tetrahedral colour space based on either a UVS or VS observer model. This measure is intended to represent distance in character space, and not discriminability under ideal conditions. Following Avery et al. (2014), we also examined 1000 bootstrap replicates of male–female comparisons for each species to produce a confidence interval around the mean degree of dichromatism. We refer to species as ‘dichromatic’ below when the confidence interval of colour differences among sexes did not overlap with the extent of differences within sexes.

Phylogenetic and comparative methods

As our sampling of the Maluridae included not only Malurus species, but also a large proportion of Stipiturus and Amytornis, we used the more completely sampled species tree recently inferred by Lee et al. (2012; their fig. 5), which includes these groups. This tree was constructed using a BEST analysis (Liu, 2008) on one mitochondrial and 16 nuclear loci, and its branch lengths are proportional to time in generations as estimated by BEST's coalescent model (Liu & Pearl, 2007). Previous phylogenetic studies of Malurus have relied on Driskell et al. (2011; Ödeen et al., 2012; Delhey et al., 2013), whose results differ primarily in the relationships among the chestnut-shouldered group.

To test whether the evolution of male PCA span, female PCA span or net dichromatism was driven by visual system, we examined the fit of Ornstein–Uhlenbeck (OU) models as implemented in the r package OUwie (version 1.42; Butler & King, 2004; O'Meara & Beaulieu, 2014). This framework is capable of comparing models in which traits are (i) evolving randomly following Brownian motion (BM), (ii) evolving towards a single optimum (OU1), (iii) evolving towards several optima depending on the selective regime (OUM) or (iv) evolving towards several optima at differing rates depending on the selective regime (OUMV). In this context, support for the BM and OU1 models indicates that the trait is evolving independently of selective regime, whereas support for the OUM and OUMV models indicates evidence that the trait is evolving towards different optima defined by the selective regime (Butler & King, 2004; e.g. Maia et al., 2013a).

We constructed selective regimes for the species’ visual system based on UV/V sensitivity reported by Ödeen et al. (2012). We defined a discrete character for species’ geography following Johnson et al. (2013), scoring species as breeding in Australia or PNG based on range maps available from BirdLife (BirdLife International & NatureServe 2011). In each case, we estimated ancestral states for regimes using unordered parsimony and likelihood in Mesquite (version 2.75; Maddison & Maddison, 2011), and we repeated our analyses across each of the most parsimonious reconstructions (Fig. S2). Using these selective regimes, we fit each of the models described above to data on male colour span, female colour span and net dichromatism. We compared the fit of these models to our data using the Akaike information criterion corrected for sample size (AICc) and converted these AICc values to proportional Akaike weights indicating relative model support.

To compare the rates of evolution across the 11 plumage patches that we measured, we used the r package geiger (version 2.0.3) (Harmon et al., 2008). We fit BM and OU1 models to our colour data and the tree using 100 000 MCMC generations under flat priors, and compared the posterior distributions of the model's σ² parameter, which is a common measure of evolutionary rate (see Adams, 2013 and refs. therein). Rate parameters estimated from BM and OU1 models were highly congruent (compare Fig. 1b, Fig. S3, Fig. S4), and thus, parameters from the BM model are reported throughout, as these require fewer assumptions (Beaulieu et al., 2012).

Ancestral state reconstruction for PCA span was performed using phytools (Revell, 2012) and is included primarily as a visual aid.

Colour space and sexual dichromatism

The major differences between the gamut of female and male plumage coloration in Maluridae are in regions of intense short-wavelength reflectance or a combination of UV and long-wavelength reflectance; we observed few female plumage spectra with these characteristics (Fig. 2). Interestingly, the whole of UVS species in Malurus appears to cover a similar range of potential colours as VS species. However, UVS species showed greater distances among patches as measured by PCA span (Fig. 3) and colour span. Grasswren species (Amytornis) exhibited a considerably smaller range of colours than other species in this study (Fig. 2) and were largely restricted to melanin-derived colours including blacks, browns and rusty reds.

Sexual dichromatism was not prevalent in the grasswrens, but was very prevalent in the fairy-wrens (Malurus). For example, 44% of sampled grasswren species showed high degrees of dichromatism for the breast patch, compared to 100% of fairy-wren species (Fig. S5A). Interestingly, net dichromatism was highest for two VS species, M. cyanocephalus and M. leucopterus. All Malurus species exhibited sexual dichromatism on the breast, cheek, crown, flank, mantle and rump patches (e.g. Fig. S5A). The degree of sexual dichromatism for each species was nearly identical when comparing between UVS and VS observers (e.g. Figs S5A vs. S6).

Evolutionary rate

Evolutionary rate, as indicated here by the σ² parameter of BM models, varied considerably among patches (Fig. 1a). The primaries, tail and wing coverts were consistently the most conserved plumage patches across all colour metrics and in both males and females. Among males, rates of evolutionary change along the LM axis of hue were greatest on dorsal plumage patches, whereas rates along the MS axis were greatest on ventral plumage patches and the mantle (Fig. 1a). The rates of evolutionary change in male UV chroma were greatest on the mantle, breast and flank. Rates for saturation were greatest on the mantle and rump, whereas rates for brightness were greatest on the cheek, mantle and ventral patches. In contrast, the rates of plumage evolution were highest in females for the cheek and rump patches across measurements of hue and chroma, but highest on ventral patches when measuring saturation and brightness. The rates of evolution for the mantle patch, among the highest for males in all measurements, were uniformly low in females, despite the gain of short-wavelength dorsal plumage in females of M. amabilis and M. lamberti (Joseph et al., 2013).

Rates of hue evolution were roughly three- to six-fold lower in females than in males. Overall, dorsal patches were consistently more evolutionarily labile in males than females (Figs 1B, S3 and S4). The evolution of ventral patches largely followed this pattern for measures of hue, but evolved at equally low rates for saturation and equally high rates for brightness (Fig. 1a). Rates of colour evolution for the primaries, tail and wing coverts were similarly low in both sexes across measurements.

Modelling selective regimes: visual system and geography

To define ancestral selective regimes for use in OU analyses, we used several methods to reconstruct ancestral states for visual system and geography (see above). The unordered parsimony method showed equivocal support for two most parsimonious histories for both visual system and geography (Fig. S2), and as such, we repeated each OU analysis across both of the two most parsimonious histories for its selective regime (Fig. 4). Likelihood models in which gains and losses were equally probable (mk1) supported two independent gains of UVS in Malurus; however, likelihood models with unequal rates for gains and losses (mk2) showed more equivocal results (Fig. S2). However, both likelihood models supported two independent malurid radiations in PNG as in Schodde (1982).

Multi-optimum OU models based on a visual system selective regime strongly outperformed single-optimum and BM models in explaining the evolution of male coloration, but not female coloration (Fig. 4). In the OU framework, this indicates evidence for a correlation between visual system and male PCA span and was observed to a weaker extent for net dichromatism. In contrast, single-optimum OU and BM models were sufficient to explain the phylogenetic distribution of female coloration span. These results were consistent across the two most parsimonious reconstructions of UV sensitivity (Fig. S2). Furthermore, these results were supported in additional analyses conducted using male colour span and colour volume.

We conducted similar analyses with selective regimes based on geographic range following Johnson et al. (2013). In these analyses, multi-optimum models showed slightly better fit than single-optimum models, showing at best weak support for a correlation between geography and female coloration, and virtually none for a correlation with male coloration or sexual dichromatism (Fig. 4). These results were consistent across the two most parsimonious reconstructions of geographic range (Fig. S2). However, analyses conducted with colour span and colour volume showed stronger support for a relationship between female coloration and geography.

Evolutionary rate and dichromatism

The rates of evolution were greater for male plumage coloration than for female plumage coloration across nearly every measurement and patch, often by a wide margin (Figs 1B and S3). The differential rate of male and female plumage evolution we observe suggests a greater role for sexual selection in driving phenotypic diversification of colour in males than in females. There is currently a wide body of evidence supporting rapid rates of evolution for male nuptial characters across many clades (Emlen et al., 2007; Arnegard et al., 2010; Martin & Mendelson, 2014; Klaczko et al., 2015). Indeed, studies have implicated sexual selection in driving this divergence, showing that male and not female coloration evolves more rapidly in sexually dichromatic species (Seddon et al., 2013; but see Price & Whalen, 2009). However, Price & Eaton (2014) showed that for a clade of New World blackbirds, female coloration at times evolved twice as rapidly as male coloration, often with female divergence associated with changes in mating system. Our finding of the reverse pattern is in stark contrast to that finding, and also to recent estimates that rates are similar for both sexes in Maluridae (Johnson et al., 2013). When considered solely in terms of the presence or absence of ornamentation, male coloration appears constant relative to female coloration, which differs considerably among species (Karubian, 2013). However, our analyses show that when considered in terms of the colour of that ornamentation, male ornaments are far more variable and evolving far more rapidly.

In Maluridae, most of the striking differences between the male and female range of plumage colours are in the UV (Fig. 2). Indeed, many of the intense short-wavelength colours observed in males were wholly absent in females. Thus, it is likely these were missed by the visual plumage scoring performed in preceding studies (Johnson et al., 2013). This may explain why we observed a greater evolutionary rate for male colour measures rather than equal rates for both sexes, and it highlights the importance of considering the evolution of species’ colour patterns in the context of their visual system (see Eaton, 2005). However, it is worth noting that for several metrics that do not include variation in the UV range (LM hue and MS hue), evolutionary rates were also substantially greater for males than for females (Fig. 1b).

We observed rapid rates of male plumage evolution on both dorsal and ventral plumage patches, whereas in females only ventral patches showed rapid rates of evolution, and only in our measure of brightness (Fig. 1). Such ventral patches are likely to be less conspicuous to predators in ground-foraging birds and should experience less intense selection for crypsis (Gomez & Théry, 2007). This suggests that dorsal plumage coloration in malurid females may be constrained by greater selection for crypsis, which is relaxed on ventral patches. However, many of the colour transitions among females for ventral patches reflect changes in the extent of rusty plumage in Amytornis and Stipiturus; in these genera, many species exhibit rusty flanks, which are likely cryptic in typical arid habitats. Such differential selection on females and males is a cornerstone of theory on sexual selection (Andersson, 1994) and has been observed to cause sexual dichromatism in many groups (Kunte, 2008; Price & Eaton, 2014).

Interestingly, the Maluridae do not appear to be a case in which dichromatism is driven by divergence in male plumage or by divergence in female plumage (Figuerola & Green, 2000; Kunte, 2008; Friedman et al., 2009). Rather, our results suggest that both sexes are evolving independently: at different rates and in response to different selective regimes (Johnson et al., 2013; Karubian, 2013). This is inconsistent with, but does not exclude, a genetic correlation model for the evolution of sexual dichromatism. In such a model, selection on males leads to both sexes gaining elaborate coloration that is subsequently suppressed in females (Kimball & Ligon, 1999) – this would require a monochromatic elaborate ancestor. It appears as though elaborate female coloration is suppressed in most fairy-wren species, with this suppression lost on several occasions (the alternative being that convergence with male plumage is selected or coincidental). Although an androgenic source of plumage control in females seems promising at first, previous observations have suggested that the development of sex-specific plumage coloration is under nongonadal (luteinizing hormone and/or genetic) control in many songbirds, with both plumage types expressed seasonally in gonadectomized songbirds, and expressed simultaneously in gynandromorphs (Kimball & Ligon, 1999; Agate et al., 2003). These observations have been supported by experimental evidence in Maluridae, which shows that testosterone treatment is not sufficient to induce male coloration in females of M. cyaneus (Peters, 2007), despite being sufficient to induce this phenotype in juvenile males of M. melanocephalus and M. cyaneus (Peters et al., 2000; Lindsay et al., 2011). The repeated transitions between dull and elaborate female plumage in Maluridae underline the need for further genomic and experimental study of the mechanisms of sexual dichromatism in birds (Karubian, 2013).

We found that colour evolution was most conserved for patches located on the wing and tail, both in males and in females (Fig. 1). These feathers are used more in flight than the contour feathers measured in other patches (Burtt, 1986), and each was a dark brown colour that varied little between species. Previous studies have shown in laboratory and natural experiments that melanin-pigmented feathers are more resistant to abrasion and bacterial degradation (Burtt, 1979; Barraclough & Sibley, 1980; Goldstein et al., 2004). Our results suggest that feather coloration was highly conserved on those patches most critical for flight and vulnerable to abrasive damage.

Visual system and coloration

We found strong support for OU models in which male PCA span was evolving dependent on visual system (Fig. 3), representing evidence of a correlation between these traits. Thus, our results further support a relationship between the evolution of a UVS visual system and the evolution of elaborate feather ornamentation in Maluridae. However, the direction of this relationship remains contentious. Ödeen et al. (2012) compared the use of short-wavelength plumage ornamentation to UV/V sensitivity and reported that gains of UV coloration most likely predated changes in visual system. Here, we show that the range and contrast of male colours expands in concert with the evolution of UVS (Fig. 4). Given the prevalence of extra-pair mating and sexual selection in fairy-wrens (Mulder et al., 1994), it is not difficult to imagine that in this case a change in sensory system might have lead female preferences to drive male ornaments towards a new optimum. However, it is also possible that other potential differences between UVS and VS species, such as the strength or optima of female preferences, could explain the expansion of male contrast in Maluridae.

Unlike Delhey et al. (2013), our study is not capable of differentiating between selection for increased signal variability vs. detectability to conspecific receivers. Delhey et al. (2013) showed that the fairy-wrens’ UVS visual system enhances the conspicuousness of elaborate male patterns, but does not enhance discriminability of intraspecific variation. Our study expands on this by showing that whereas males in species with a UVS visual system have more contrasting plumage, females tend to retain cryptic plumage and evolve at a slow rate relative to males (Higgins et al., 2001). The lack of a relationship between female contrast and UVS may suggest that selection for increased signal detectability is not typically mirrored in females – this would contrast with other lineages in which elaborate female plumage coloration has been maintained due to its function in social competition (Trail, 1990; Stutchbury & Morton, 2001; Murphy et al., 2009). Although this function of female colour may be largely absent in Maluridae, it could be present in several species in which females exhibit elaborate coloration. As suggested by Karubian (2013), behavioural studies are needed to address the roles of predation avoidance and crypsis for female coloration in this clade and to assess whether these differ between species in Australia and PNG.