Discrete colour polymorphism in the tawny dragon lizard (Ctenophorus decresii) and differences in signal conspicuousness among morphs


  • Data deposited at Dryad: doi:10.5061/dryad.1b2q0

Correspondence: Luisa Teasdale, Department of Zoology, University of Melbourne, Victoria 3010, Australia. Tel.: +613 8341 7431; fax: +613 8344 7909; e-mail: l.teasdale@student.unimelb.edu.au


Intraspecific colour variation is common in nature and can vary from the coexistence of discrete colour variants in polymorphic species to continuous variation. Whether coloration is continuous or discrete is often ambiguous and many species exhibit a combination of the two. The nature of the variation (discrete or continuous) has implications for both the genetic basis of the colour variation and the evolutionary processes generating and maintaining it. Consequently, it is important to qualify the existence of discrete morphs, particularly in relation to the animal's visual system. In this study, we quantified male throat colour variation in Ctenophorus decresii tawny dragon lizard and tested for morphological and ecological correlates of the colour variants. We confirmed that discrete throat colour morphs can be defined based on colour and pattern analyses independent of the human visual system. We also found that the colour variants differed in their conspicuousness from the background, to the lizard's visual system, which has implications for signalling. However, the morphs did not differ in morphology or microhabitat use, which suggests that these characteristics are not involved in the evolutionary maintenance of the polymorphism.


Colour variation within species is common in nature and is often associated with geographic distribution, environmental conditions or individual traits such as age, sex, health or social status (e.g. Rand & Andrews, 1975; Hoglund et al., 2002; Hoekstra et al., 2004). Colour polymorphism is a special case where discrete, genetically determined colour forms coexist within an interbreeding population at frequencies too common to be maintained by recurrent mutation (Ford, 1945; Huxley, 1955). It is important to determine whether colour variation is continuous or discrete for two reasons. First, the genetic basis of discrete and continuous colour variation may differ considerably. Discrete colour variation that is fixed for life at sexual maturity is often the result of few genes with simple Mendelian inheritance (Sinervo & Zamudio, 2001; Sinervo et al., 2001; King, 2003) whereas continuous colour variation is often condition-dependent and may have much more complex genetic architecture (Sinervo & Svensson, 2002). Second, the nature of the colour variation and underlying genetic basis has implications for how the colour variation is maintained. Continuous variation within populations is often associated with variation in environmental conditions, including maternal effects (Tschirren et al., 2012). Discrete variation, however, is more often maintained through processes such as frequency dependent selection (Sinervo & Calsbeek, 2006). Thus, determining the nature of colour variation is important for understanding the evolutionary processes generating and maintaining the variation, and its adaptive significance.

In many species where discrete colour morphs have been described, continuous variation may also be present within the morphs (Thompson & Moore, 1991; Vercken et al., 2008). Guppies, for example, are highly polymorphic but also show condition-dependent colour variation (Hughes et al., 1999; Eakley & Houde, 2004; Olendorf et al., 2006). Similarly, although male tree lizards (Urosaurus ornatus) can be categorized into discrete throat colour morphs based on the colours present, individuals within these groups can vary markedly in the amount of certain colours shown (Thompson & Moore, 1991). Furthermore, in many cases the distinction between continuous and discrete colour variation can be unclear. For example, whether discrete female colour morphs exist in the European common lizard (Lacerta vivipara) has been the subject of recent contention (Vercken et al., 2007, 2008; Cote et al., 2008).

One problem has been reliance on subjective classification of colour morphs, which ignores colour variation within morphs and does not determine whether variation is discrete or continuous objectively or in terms of the visual processing system of the appropriate receiver. This also assumes that nonhuman species categorise signals or stimuli in the same way as humans, which is unlikely to be consistently true. Over the last two decades, however, there has been a growing appreciation of the importance of objectively measuring colour and taking receiver vision into account (e.g. Endler, 1990; Bennett et al., 1994; Osorio & Vorobyev, 2008). More recent attempts to quantify colour variation objectively have used a range of approaches. Most often this involves analyses based on spectrophotometry, but reflectance spectra can only be taken from point samples of colour patches, and do not provide information on the entire colour pattern (i.e. spatial variation in colouration within the colour patch). An alternative is to use digital image analysis, and increasingly studies are using this approach to analyse animal markings. In many potentially polymorphic species morphs differ in colour or pattern or a combination of the two. For example, the African cuckoo finch, Anomalospiza imberbis, and its various host species show extreme intraspecific variability in egg colour and pattern. Recent studies have used a combination of visual modelling based on reflectance spectra and pattern analysis based on digital images to determine the cues used in the host rejection behaviour of this species and to quantify egg appearance and diversity (Spottiswoode & Stevens, 2010, 2011). Although the eggs broadly fall into different categories to human eyes, the variation is actually continuous when considering avian vision (Spottiswoode & Stevens, 2012). In addition, human vision misses a range of potentially important information, such as ultraviolet reflectance and that patterns can comprise multiple uncorrelated attributes (e.g. marking size, distribution and contrast; Stoddard & Stevens, 2010). It is therefore essential in studies of polymorphic colouration to objectively establish the existence of discrete morphs and to characterize them using a combination of objective measures of both colour and pattern.

In many species discrete colour morphs are also associated with different reproductive or ecological strategies, which allow the morphs to coexist, either because, on average, they have equal fitness or due to frequency dependent selection (Sinervo & Svensson, 2002; Roulin, 2004; Gray & McKinnon, 2007; McKinnon & Pierotti, 2010). Specifically, each colour morph may evolve different suites of morphological, physiological, behavioural or ecological traits to maximize their fitness (Sinervo & Svensson, 2002). A notable example is the rock paper scissors dynamics of the side-blotched lizard, Uta stansburina, in which morphs are maintained by frequency dependent selection (Sinervo & Lively, 1996). In this species there are three male throat colour morphs, each associated with differences in reproductive strategy, morphology and steroid levels. Morphs adopting different reproductive or ecological strategies often show differences in microhabitat use (Skúlason & Smith, 1995). Consequently, morphs may be polymorphic for sexual signals such as male throat colouration in lizards, but also in cryptic dorsal colouration which may lead to differential predation risk (Lancaster et al., 2007). Once we have determined whether discrete colour morphs exist within a population the next step in understanding the evolution of the colour morphs and the evolutionary processes maintaining the polymorphism is to identify morphological and ecological correlates of the colour variation.

Here, we quantify variation in male throat colouration of Ctenophorus decresii (the tawny dragon lizard; Duméril & Bibron, 1837) and test whether discrete morphs can be defined using colour and pattern analysis of digital photographs in addition to visual modelling of spectral data to assess the colour variation. This species is a small, sexually dimorphic agamid lizard, which exhibits considerable variation in male throat colour both within and between populations. Males have combinations of UV-blue, grey, cream, black, orange and yellow throat colours. In contrast, female throat coloration is much more subtle. Females have cream throats with varying amounts of grey reticulation and a yellow flush in some individuals during the breeding season. Tawny dragons are also highly territorial and the males employ complex behavioural displays, which include push ups, head-bobbing and expanding the throat in aggressive encounters, all of which emphasize the throat signal (Gibbons, 1979). This suggests that throat colour may potentially function as an important sexual signal in this species. The environment they inhabit is highly variable and offers the potential for individuals to occupy a number of different microhabitats and visual environments even within populations; therefore, we might expect different throat colour morphs to be correlated with a number of other traits. C. decresii has been shown to vary in the conspicuousness of individual dorsal colouration, which has been linked to increased predation risk; however, it is not known whether this dorsal variation is associated with variation in the throat colour signal (Stuart-Fox et al., 2003). Therefore, we tested whether the male throat signal varies in conspicuousness to conspecifics, and whether throat colouration is associated with differences in dorsal conspicuousness to avian predators. We also tested whether the colour variants of C. decresii differ in morphology and microhabitat use.

Materials and methods

Study species and sites

We sampled Ctenophorus decresii from two populations in the Flinders Ranges, South Australia: Telowie Gorge Conservation Park (33°2′9.09″ S 138°7′47.96″ E) and Flinders Ranges National Park, Wilpena Pound (31°22′39.14″ S 138°35′55.67″ E), in October and November 2010. Both populations are representative of the core part of the species distribution in the central Flinders Ranges (Figure S1 in the Supporting Information). The throat varies substantially in both colour and pattern (i.e. the degree of reticulation) within populations (Houston, 1974) and is fixed at maturity (Osborne, 2004; Stuart-Fox, unpublished data). Males have blue-grey dorsal coloration, a thick black dorso-lateral stripe and a thin cream to orange stripe along the upper lip and along the top anterior border of the dorso-lateral stripe (Fig. 1b, c). Populations in the central Flinders Ranges all show similar throat colour variation as the two populations in this study, although populations in the far southern and eastern parts of the species range show different throat colours and appear to be monomorphic (Claire Mclean, unpublished data).

Figure 1.

Examples of Ctenophorus decresii males allocated to four throat colour classes (a) (1) grey, (2) grey and yellow, (3) grey, yellow and orange and (4) grey and orange and the (b) dorsal and (c) lateral male colouration. (Photos: Luisa Teasdale)

Male dragons were captured by noosing (Telowie n = 15; Wilpena: n = 39). We recorded GPS coordinates for where the lizard was first seen, measured cloacal temperature to 0.1 °C and recorded microhabitat data (methods described below). We then returned the animals to the field station where we took photographs, spectral measurements and morphometrics (methods described below). The lizards were kept overnight, marked with a unique florescent elastomer code and released the next day into the rock crevice nearest to the point of capture.

Morphology and colour pattern

For each lizard, we measured mass (to the nearest 0.1 g), snout to vent length (SVL), vent to tail tip (both to the nearest mm), hind limb length (body wall to the tip of the longest toe) and head depth, width and length (to the nearest 0.01 mm). Head length was measured from the angle of the lower jaw to the tip of the snout, head width was measured at the widest point of the head and depth was measured perpendicular to this point (Figure S2). We took the residuals of each measurement against SVL and used the residuals in subsequent analyses.

To quantify colour pattern variation, we took photographs of the ventral side of each lizard using a Canon PowerShot SX1-IS digital camera (saved in raw format; Stevens et al., 2007). Each photo included a ruler for scale and a standard grey card (Micnova).

We also measured the spectral reflectance of three different body regions for each male, the throat, the dorsum and the dorsal stripe (Figure S2) using an Ocean Optics USB2000 + spectrometer and PX-2 Pulsed Xenon light source, both connected to a probe via a bifurcated fibre optic cable. Measurements were an oval point sample 3 mm × 4 mm, expressed relative to a 99% diffuse white reflectance standard and taken at a 45° angle to the surface of the lizard. As body temperature can affect reflectance in reptiles (Gibbons & Lillywhite, 1981; Cooper & Greenberg, 1992), we heated the lizards under a heat lamp prior to taking spectral measurements. The average body temperature of lizards immediately after measurements were taken was 32.5 ± 0.27 °C, which was closely matched to the average body temperatures at capture in the field (33.1 ± 0.33 °C). We took 2–4 measurements for each body region and these were averaged where appropriate. For the throat, we measured both the primary (central) and secondary (surrounding) colours. In some cases (31.5% of throat measurements), we were unable to obtain accurate spectral measurements for a component of the throat colour due to fine patterning (e.g. a measurement for the yellow when the throat colour was fine yellow and grey reticulations). In these instances, we compared the RGB image values of the colour in question to similar colours on other males sampled. To do this we averaged the spectral measurements for males with RGB values within ± 10 RGB units based on the calibrated images (see Segmentation analysis; n = 2–5 males in each case) and used these averages to replace the missing data.

We also measured background reflectance of representative rock samples at each of the study sites. Adult C. decresii are almost exclusively found on rocks (Houston, 1974). Samples included a variety of different rock types found at the two locations including pound quartzite at Wilpena and rhynie sandstone at Telowie. We also measured the reflectance of lichen found on the rock samples. For each sample multiple measurements were taken (3 to 4) and the mean reflectance of the backgrounds (rock and lichen) for the analyses were used; as the rocks were generally finely speckled with lichen a mean measurement would be a reasonable representation of the overall rock colour viewed from a short distance.

Microhabitat sampling

We measured a number of microhabitat features within a circular plot 3 m in radius around the point where each lizard was first sighted. We measured the distance to the nearest vegetation, the nearest crevice and the diameter of the rock the dragon was found on. By eye we estimated the percent cover of each of vegetation less than 1 m tall, dead vegetation, rocks less than and greater than 50 cm in diameter (in any dimension) and canopy cover as one of six categories (< 10%, 10–24%, 25–49%, 50–74%, 75–89%, > 90%). We also counted the number of shrubs between 1 and 3 m tall within the plot. These features were chosen as potential indicators of territory quality as in rock-dwelling lizards, including the closely related ornate rock dragon (Ctenophorus ornatus), higher quality territories have more large rocks and less vegetation, which makes them potentially easier to defend (LeBas, 2001). The number of shrubs in each plot was log-transformed and the mid-point of the classes for the estimates of cover variables were used to meet model assumptions (see 'What are the correlates of morph type?').

To determine whether the lizards were selecting their microhabitat nonrandomly with respect to the habitat available, we estimated the same microhabitat measurements for 95 randomly selected 3 m radius plots at Telowie and 86 at Wilpena. These plots were selected by walking a random number of metres (max. 50 m) along a transect (i.e. path/creek) and then a random number of metres perpendicular to the transect (max. 20 m) which incorporates the area in which the lizards were captured.

Analysis of colour pattern

Each male captured was visually allocated into four classes or ‘morphs’ based on the presence and absence of yellow and orange on the throat: 1. grey, 2. yellow and grey, 3. orange and yellow; 4. orange and grey (Fig. 1a). However, the composition of the throat signal was highly complex and variable in this species. To verify the visual classification, and to determine whether there are four discrete morphs rather than continuous variation, we quantified colour pattern by estimating the proportion of each colour (an image segmentation analysis), and the pattern marking size, contrast and diversity (an image granularity analysis) as well as spectral properties of the throat colours, as they would be perceived by the lizards (receptor quantum catches). All of these measurements are independent of the human visual system.

Segmentation analysis

We first linearized the images with respect to radiance and calibrated them to reflectance information (which removes any variation due to differences in illumination between photos) using a custom program in MATLAB (The MathWorks, Inc., Natick, MA, USA) (see Stevens et al., 2007). Each throat image was rescaled to 148.2 pixels per cm, so that they were all on the same scale, and then cut out from its background and placed on a white background (500 by 500 pixels).

We then developed a ‘segmentation’ analysis to objectively quantify the proportion of yellow, orange and grey on the throat of each individual (e.g. Figure S3). This analysis extracts portions of the image based on the RGB values of each pixel according to set threshold values. First, the RGB values were standardized to remove absolute variation in pixel values (which corresponds to brightness) by calculating each pixel value as a proportion of the total (i.e. proportion red (pR) = R/(R+G+B)), where R, G, B are the unstandardised values for a given pixel.

The orange component of the image was then extracted by subtracting the green image layer values from the red layer (i.e. pR-pG for each pixel). This produced a new image that was thresholded into binary format (orange pixels were encoded by a 1 and other regions by 0), and used to calculate the proportion of the total throat area covered by orange. For yellow the process was similar, but we subtracted both the red and the blue layer, which left the yellow regions. The proportion grey was the area of throat minus the orange and yellow regions. The thresholds used to determine the colour layers in this analysis were 0.25 for red and 0.15 for yellow. These values were chosen based on an initial analysis of a subset of the images.

Granularity analysis

To analyse the pattern component of the throat signal, we used a ‘granularity’ analysis similar to that recently used to analyse cephalopod camouflage (e.g. Barbosa et al., 2008; Chiao et al., 2009) and avian egg-shell markings (e.g. Spottiswoode & Stevens, 2010; Stoddard & Stevens, 2010). This analysis calculates the contribution of different marking sizes to a given pattern, and bandpass filters a square image of the center of the throat into a subset of seven images of different spatial frequencies (small to large markings sizes). The throat images used were greyscale, corresponding to the green layer of the calibrated reflectance image (as this should broadly approximate the region of the spectrum used by birds and lizards for achromatic vision, which is important in pattern assessment). Following this, the total amount of ‘energy’ is calculated for each filtered image, as the sum of the squared pixel values in each image divided by the number of pixels in the image. The amount of energy in each image vs. spatial frequency produces a granularity spectrum (see Chiao et al., 2009; Stoddard & Stevens, 2010 for further details).

From this spectrum, we determined (1) the image containing the highest energy, which corresponds to the predominant marking size of the pattern; (2) the proportion of the total energy across the spectrum corresponding to this filter size, which reveals how much the main marking size contributes to the overall throat pattern (high values indicates that the pattern is dominated by this marking size) and (3) the total energy across all filter sizes (the amplitude of the spectrum), which provides a measure of overall pattern contrast, with higher values indicating more contrasting markings (see Chiao et al., 2009; Stoddard & Stevens, 2010). This analysis was conducted as part of the custom written program in MATLAB.

Ctenophorus decresii also has a black chest patch the size of which has been shown to be associated with aggressiveness and contest outcome in a closely related species (Osborne, 2005). Chest patch size (i.e. area) was measured from the ventral photographs using ImageJ (Abramoff et al., 2004).

Colour analysis: receptor quantum catches

Prior to analysis, all spectral data (lizards and backgrounds) were smoothed and averaged over each 5 nm interval within the range of 300–700 nm, the approximate visible spectrum of birds and most diurnal lizards (Vorobyev et al., 1998; Loew et al., 2002). To quantify male throat colour variation as perceived by conspecific lizards, we estimated the relative stimulation of lizard photoreceptors (receptor quantum catches) for both the central and surrounding throat colours. This analysis resulted in five variables (quantum catches) for each colour patch, one for each of the four single cones used for colour perception and the double cone which is thought to be used for luminance perception (Osorio & Vorobyev, 2005).

Receptor quantum catches (Qi) were calculated using the following equation (Vorobyev et al., 1998):

display math

where λ is wavelength, Ri(λ) is the spectral sensitivity of cone i, S(λ) is the spectral reflectance of the colour patch and I(λ) is the spectrum of light entering the eye (irradiance).

For these calculations, we used data on the spectral sensitivities of a closely related lizard, Ctenophorus ornatus (Barbour et al., 2002). Barbour et al. (2002) only identified cones with three visual pigment types, a short wavelength-sensitive (SWS), medium wavelength-sensitive (MWS) and long wavelength-sensitive (LWS) type. However, a fourth ultraviolet-sensitive (UVS) photoreceptor has been found in all other related diurnal lizards (e.g. Loew et al., 2002; Bowmaker et al., 2005). The UVS photoreceptor may have been missed as they are generally low in number in the retina and the sample sizes for the microspectrophotometric data in Barbour et al. (2002) were very small. We therefore conducted the calculations with a fourth cone, a UVS photoreceptor (λmax = 365 nm; Figure S4b) (Loew et al., 2002), as it is likely that the lizards can see in UV wavelengths. The spectral sensitivities were corrected for the transmission of associated oil droplets (as described in Chan et al., 2009). We used an irradiance spectrum representing full sunlight because the habitat is open, and the lizards are diurnal (Figure S4c) (Stuart-Fox et al., 2003).

Conspicuousness against the background

We assessed the conspicuousness of the throat, as perceived by conspecific lizards against the background, and of the dorsal surface (mid-dorsal and dorsal stripe), as perceived by predators (i.e. birds). We also assessed the contrast between the primary and secondary throat colour of each individual. To estimate contrasts, we applied the model of Vorobyev & Osorio (1998), which estimates the discriminability of two colours (e.g. the lizard and the background; Fig. 1c) in units of just noticeable differences (JNDs) and assumes that visual discrimination is limited by photoreceptor noise. The model produces a measure of the chromatic (colour) contrast (ΔS) based on the four single cones used for colour perception and a measure of achromatic (luminance) contrast (FD) based on the double cone used for luminance perception (Vorobyev & Osorio, 1998). Contrast measures provide an indication of conspicuousness or colour matching to the background with values less than one ‘just noticeable difference’ (JND) indicating that the lizard colour is indistinguishable from the background.

We used the model calculations detailed in Siddiqi et al. (2004). We derived receptor quantum catches (as above), using either the lizard or bird spectral sensitivity data (Figure S4a and b) and the same irradiance spectrum (Figure S4c). Broadly, there are two conserved visual systems within birds in terms of visual pigment spectral sensitivity. These are ultraviolet sensitive (UVS) with a UVS cone peak sensitivity of around 360 nm and violet sensitive (VS) with a VS cone peak sensitivity of around 410 nm (Hart & Hunt, 2007). The main predators of C. decresii are raptors, such as the Nankeen Kestrel (Falco cenchroides) and kookaburras (Dacleo novaeguineae), which have a VS visual system and corvids, such as the grey butcherbird (Cracticus torquatus), which potentially have a UVS visual system according to a study by Hart & Hunt (2007) (Gibbons & Lillywhite, 1981; Stuart-Fox et al., 2004). We present results based on the UVS visual system to capture any variation in how UV sensitive predators would perceive the lizard colours. However, results are qualitatively the same using a VS visual system as there is little UV reflectance in the lizard colours in these populations. The bird spectral sensitivities were also corrected for transmission of associated oil droplets (Figure S4a) (Endler & Mielke, 2005).

We applied the von Kries transformation to the cone catches to account for receptor adaptation to the light environment, which contributes to colour constancy (Vorobyev & Osorio, 1998). We assumed that photoreceptor noise (ωi) for the LWS photoreceptor = 0.05 and then derived ωi for remaining photoreceptor classes (Stuart-Fox et al., 2003; Siddiqi et al., 2004). We used a ratio of 1 : 1.4 : 2 : 2.6 for the four avian photoreceptor classes of a UVS visual system (Hart, 2001). We used a ratio of 1 : 1 : 3.5 : 6 for the four single receptor classes for the model of the dragon UVS visual system, based on the relative photoreceptor frequencies in Barbour et al. (2002).

Statistical analysis

We first assessed whether the objective measures of colour and pattern corroborated the visual classification of the dragons into four throat colour morphs, using a discriminant function analysis (DFA; PROC DISCRIM; SAS 9.2). This analysis generates a linear combination of the variables (Canonical variables) that maximizes the probability of correctly assigning observations to their predetermined groups (in this case four ‘morphs’). We used the three throat colour proportions, the three throat pattern variables and the five quantum catches (four single cones and a double cone) for both the primary and secondary throat colours as response variables (16 variables in total). All variables met the assumptions of this and subsequent analyses.

We then tested whether the morphs differed in morphology, microhabitat and contrast conspicuousness using five separate manovas (PROC GLM; SAS 9.2). Two of the manovas for the morphological (including chest patch size) and microhabitat variables, respectively, and three for conspicuousness: (1) throat colour compared with the background (primary and secondary); (2) the primary throat colour compared with the secondary throat colour and (3) the dorsal colour (dorsal and dorsal stripe) compared with the background. The manovas included both chromatic (colour) contrast (ΔS) and the achromatic (brightness) contrast (FD) variables in each case. Throat colour morph, site (Telowie or Wilpena) and the interaction between morph and site were the factors in each model. These analyses were repeated in a mancova using the two continuous canonical variables from the discriminant function analysis (i.e. Canonical variable 1 and 2) rather than the categorical ‘morph’ variable as the continuous variables capture variation within as well as between morphs and may better represent the existing colour variation.

To test whether the dragons were using the available habitat at random we conducted a manova comparing the multiple microhabitat variables for the dragons to those within the random plots. Plot type (lizard or random), site and the interaction between them were the dependent variables.


Are there discrete colour morphs?

As detailed above, all dragons captured were categorized into one of four throat colour classes based on the presence and absence of yellow and orange (Fig. 1). The discriminant function analysis of the 16 colour and pattern variables of the throat (Table 1) was able to discriminate between the four morph classes; however, there was some overlap of the 95% confidence ellipses (Fig. 2). Multivariate discrimination based on the 16 traits was highly significant (Wilks’ λ = 0.026, F39,107 = 6.71, < 0.0001) and cumulatively the first two canonical variables explained 98.3% of the variation. Canonical Variable 1 (Can 1) discriminated the morphs primarily based on the amount of orange: as Can 1 increases the amount of orange decreases and the amount of grey increases (Table 1). Canonical Variable 2 (Can 2) discriminated the morphs based on the amount of yellow: as Can 2 increases the amount of yellow decreases (Table 1). In univariate tests 12 of the 16 variables differed significantly between the morphs (Table 1). The proportion of grey, orange and yellow was each significantly different between the four morphs (F3,48 = 25.41, 28.68 and 20.59, respectively, and < 0.0001 for each). Total energy, which is a measure of overall pattern contrast, was also significantly different with the grey morph having greater pattern contrast than the other three morphs (F3,48 = 4.47, P = 0.0076).

Table 1. Univariate results from the discriminant function analysis testing for the presence of discrete male throat colour morphs in Ctenophorus decresii (DF = 3,48)
  F P Canonical Variable 1Canonical Variable 2
  1. Significant P-values are in italics.

Proportion (Segmentation analysis)
Grey25.41 <0.0001 0.6110.635
Orange28.68 <0.0001 −0.8380.074
Yellow20.59 <0.0001 0.245−0.868
Pattern variables (Granularity analysis)  
Total energy (etot)4.47 0.0076 0.3290.386
Proportion energy (eprop)2.120.11040.0560.343
Filter size (predominant marking size)0.560.6437−0.1920.026
Quantum cone captures  
Primary throat colourU0.330.80470.137−0.067
S3.43 0.0242 0.4260.134
M21.52 <0.0001 0.7910.103
L3.85 0.0151 −0.461−0.045
Secondary throat colourU2.99 0.0399 −0.2500.308
S5.07 0.0039 0.0110.584
M5.53 0.0024 0.4350.334
L4.20 0.0102 0.062−0.521
D3.48 0.0228 0.165−0.436
Figure 2.

Discriminant function analysis of 16 throat colour and pattern variables testing for the presence of discrete male throat colour morphs in Ctenophorus decresii (DF = 3,48). Canonical Variable 1 relates to the extent of orange and Canonical Variable 2 relates to the extent of yellow. Morph categories are surrounded by 95% confidence ellipses.

The frequency of each morph at each site also differed significantly (F3 = 11.81, = 0.001). At Telowie the yellow morph was the most abundant, whereas at Wilpena orange-yellow was the most common.

What are the correlates of morph type?

Conspicuousness of the throat to a lizard

The morphs differed significantly in the conspicuousness of the primary and secondary throat colours compared with the background (Wilks’ λ = 0.587, F12,114 = 2.12, = 0.021) and in contrast between the primary and secondary throat colour (Wilks’ λ = 0.713, F6,90 = 2.77, = 0.016) as perceived by a conspecific. The grey morph was chromatically indistinguishable from the background (i.e. JND < 1) and the two colours (grey and cream) comprising the grey morph's throat are chromatically indistinguishable (i.e. JND < 1), although they differ moderately in brightness (Fig. 3). Morphs with orange as the primary colour (i.e. orange and orange-yellow), appear to be more chromatically conspicuous against the background (i.e. JND > 1; Fig. 3a) and orange, orange-yellow and yellow morphs all have colour combinations with higher chromatic contrast between primary and secondary colours than the grey morph (all JND > 1; Fig. 3a).

Figure 3.

Contrast (mean JND's (Just Noticeable Difference) ± SE) of primary and secondary male throat coloration against a rock and lichen background and between the primary and secondary throat colours (‘both throat colours’) (a) chromatic (colour) contrast (ΔS) (b) achromatic (‘brightness’) contrast (FD).

All four morphs were brighter than the background (Fig. 3b) and achromatic contrast of either the primary throat colour or secondary throat colour against the background did not differ significantly between morphs (primary: F7 = 1.97, = 0.132; secondary: F7 = 4.74, = 0.959). However, the yellow and orange of the orange-yellow morph have a lower achromatic contrast (FD) than the primary and secondary colour combinations of the other morphs (Fig. 3b).

In terms of continuous colour variation, conspicuousness of the primary and secondary throat colours against the background were positively associated with the amount of orange (Can 1: F1 = 7.22, = 0.01) and amount of yellow (Can 2: F1 = 16.34, < 0.001) respectively. Consistent with differences between morphs, individuals with more orange had higher achromatic contrast between primary and secondary throat colours than individuals with only grey (Can 1: F1 = 6.29, = 0.016).

Overall, the results indicate that the grey morph was least conspicuous and orange and orange-yellow morphs the most conspicuous with conspicuousness of individuals within categories most strongly associated with the amount of orange.

In all analyses (for morphs or continuous colour variation), there were significant differences between the sites in conspicuousness against the background but not in the contrast between primary and secondary throat colour. There were no significant interactions between colour (morph, Can 1 or Can 2) and site in any of the analyses.

Conspicuousness of dorsal coloration to birds

We found significant differences between the morphs in the conspicuousness of the dorsal colouration against the background, as perceived by birds (Wilks’ λ = 0.416, F12,69 = 2.26, = 0.018); however, the values were all less than 1 JND. In terms of continuous variation, there were no differences among morphs in dorsal conspicuousness based on the extent of orange (Can 1: F4,29 = 0.762, = 0.087) or the extent of yellow (Can 2: F4,29 = 0.893, = 0.494) and no interaction between colour (morph, Can 1 or Can 2) and site although there were differences in conspicuousness between sites.


There were no significant differences in morphology between the morph classes (Wilks’ λ = 0.473, F24,114 = 1.39, = 0.126); however, there was a significant difference between the two study sites (Wilks’ λ = 0.548, F8,39 = 4.02, = 0.0015) with males at Wilpena having relatively larger heads. Consistent with the results based on morph types, there was no difference in morphology of individuals based on the extent of orange throat colouration (Can 1: Wilks’ λ = 0.851, F8, 40 = 0.87, = 0.547) or the extent of yellow (Can 2: Wilks’ λ = 0.891, F8,40 = 0.61, = 0.761).


There were no significant differences in microhabitat use between the four different morphs (Wilks’ λ = 0.518, F27,94 = 0.88, = 0.636) but there was a difference between the sites (Wilks’ λ = 0.432, F9,32 = 4.67, = 0.0005). Dragons at Telowie were generally found in areas with a denser canopy cover than at Wilpena. Similarly, there was also no difference in microhabitat use amongst individuals based on the extent of orange (Can 1 - Wilks’ λ = 0.846, F9,33 = 0.70, = 0.6039) or the extent of yellow (Can 2 - Wilks’ λ = 0.851, F9,33 = 0.64, = 0.4756).

When we compared the microhabitat used by the dragons (irrespective of morph) and the microhabitat available at the study sites we found a significant interaction between study site and plot type (dragon vs. random plot; Wilks’ λ = 0.877, = 6.11, < 0.0001). This shows that the microhabitat used by the lizards differed from that available at the study sites although there were minor differences in microhabitat preferences between the two sites. The lizards consistently chose microhabitats with lower than average canopy cover, number of shrubs and percent cover of dead vegetation and higher than average percent cover of shrubs less than a metre and of rocks greater than 50 cm in diameter. The dragons, however, were found in areas with more small rocks than the average available at Telowie but areas with less than the average available at Wilpena (Table S1).


Are there discrete morphs?

In many colour polymorphic systems morphs are based on human classifications and variation within morphs is rarely quantified. Although this does make research within these systems easier, often no analysis is done to ensure that these groupings adequately reflect the variation within the morphs (e.g. Arnegard et al., 1999). By using digital image analyses independent of the human visual system, as well as spectral information, we were able to verify our visual classification of the male throat colour morphs of C. decresii while accounting for any variation within morphs. Visual classification and colour and pattern analyses showed that there are four discrete male throat colour morphs present within both populations of C. decresii studied. These morphs vary in the presence and absence of yellow and orange and there appears to be two pure morphs (yellow or orange), one mixed morph (both yellow and orange) and one primarily grey morph with no yellow or orange present. The grey morph has grey and white or cream patterning whereas the yellow and/or orange appears to overlay grey patterning in the other three morphs. There was little evidence of an ultra-violet component to any of the throat colours studied in these populations.

Differences in the lizards’ perception of the throat colours (i.e. the quantum cone captures) supported the observed differences between morphs in colour and pattern from digital images. Although pattern is a vital component of many signals involved in visual communication, variation in a pattern component in colour polymorphic species has rarely been investigated in any detail and is often ignored. A recent study showed that in a rock dwelling cichlid, Labeotropheus fuelleborni, pattern was more important than colour in determining the outcome of aggressive encounters (Pauers et al., 2011) and in a number of species including the pygmy swordtail, (Xiphophorus nigrensis) bars on the flanks are also associated with dominance in aggressive encounters (Zimmerer & Kallman, 1837). Pattern can affect how the signal is perceived and depending on the environmental context can either increase or decrease signal efficacy (Endler, 1990). Further research is needed to determine whether pattern plays a role in communication in C. decresii.

As is the case in many polymorphic species (e.g. Urosaurus ornatus; Thompson & Moore, 1991) there was some variation in the proportion of the colours within each morph category (Fig. 3). It is currently unclear what mechanism is responsible for this variation in C. decresii; however, it is unlikely that it is a reflection of adult condition. Yellow-red colours in lizards are produced partially or largely by self-synthesized pteridine pigments (Steffen & McGraw, 2007; Weiss et al., 2012), which are unlikely to be condition dependent. Even when yellow-red colours are produced by externally acquired carotenoid pigments, there is little evidence that dietary limitation affects their expression in lizards (Olsson et al., 2008; Steffen et al., 2010). Moreover, in captive populations of this species, there was no evidence of changes in throat colour after it developed at sexual maturity (Stuart-Fox, 2002). It is possible that colour expression is environmentally influenced via maternal effects, incubation conditions or post-hatching conditions (e.g. diet) prior to sexual maturity. Alternatively, the relative proportion of each colour may be largely genetically determined rather than condition dependent, as in other polymorphic lizard species (Thompson & Moore, 1991; Sinervo & Svensson, 2002).

When visual signals are highly variable, one possibility is that they are used for individual recognition. Although discrete morphs can be objectively defined in C. decresii, there was substantial variation in both the relative proportion of colours and pattern within morphs. In the cordylid lizard, Platysaurus broadleyi, polymorphism in leg colouration was found to play no role in individual recognition (Whiting, 1999). There are, however, examples where the relative proportion of colour, rather than the colour itself, is evolutionarily important. In the tree lizard, Urosaurus ornatus, one of the five throat colour morphs consists of a blue patch surrounded by orange. Males of this morph were more dominant and more likely to win contests when their blue patch was larger relative to others (Thompson & Moore, 1991). We found no evidence that the relative proportion of colours (as indicated by the canonical variables) was associated with differences in morphology or microhabitat in this species although it may be associated with factors that we did not measure such as behavioural, physiological or reproductive traits.

Do morphs vary in conspicuousness?

The four morphs in C. decresii differed in the conspicuousness of their throat colouration against the background, as perceived by lizards. The grey morph was generally less conspicuous against the background than the other three morphs. The grey and cream of the grey morph were also chromatically indistinguishable whereas the other three morphs have more chromatically contrasting colour combinations. Orange and orange-yellow morphs were more chromatically conspicuous against the background, consistent with the amount of orange being positively related to background contrast.

These differences may affect how the signals are interpreted by conspecifics and may have implications for the potential function of the signals. Colour patterns that contrast highly against each other or against the background environment are more likely to be seen by conspecifics or may convey information faster during behavioural displays (Endler, 1992; Stuart-Fox & Ord, 2004). One possibility is that differences in chromatic contrast may be maintained by spatial heterogeneity in the backgrounds present, as well as by sexual selection (e.g. Gray et al., 2008). Differences in conspicuousness may therefore be associated with differences in signalling strategies, which potentially reflect differences in mating behaviour as in other throat colour polymorphic lizards (e.g. Sinervo & Lively, 1996). For instance, we might predict that the grey morph with the less conspicuous signal relies less on long-distance territorial signalling and may therefore maintain smaller territories, potentially associated with a mate-guarding strategy. Conversely, the conspicuous orange or orange-yellow morphs may maintain larger territories and exhibit higher levels of aggression, similar to the orange morph of the side-blotched lizard, Uta stansburiana (Sinervo & Lively, 1996). It is also possible that the different morphs select particular substrates on a microhabitat level to signal against and this should be further investigated.

Our data suggest that although the morphs differ in conspicuousness to conspecifics, there are unlikely to differ in predation risk based on dorsal conspicuousness alone. Chromatic contrast values were all less than one ‘just noticeable difference’ and the morphs did not differ in achromatic contrast as perceived by birds.

Morphological and environmental correlates of morph type

We found no evidence that the morphs differ in morphology or microhabitat use. The lack of size differences between morphs confirms that colour variation in the species is not ontogenic. While morph colouration does not vary with age it is likely that traits such as SVL and head size do. As these lizards have overlapping generations, substantial age-dependent variation within morphs may obscure differences between morphs. Ideally, individuals of the same age from captive populations, or mark recapture studies, should be compared to determine whether morphs differ in morphology or in other traits such as growth rate.

The morphs also did not differ in the surrounding microhabitat features we measured, several of which are thought to be associated with territory quality in rock dragons (LeBas, 2001). Comparison of the dragons’ microhabitat use and the available microhabitat at the study sites showed that the lizards are more often found in areas with lower canopy cover, abundance of shrubs and cover of dead vegetation and higher cover of small shrubs and large rocks. These results imply that although males make specific habitat choices, the colour morphs do not differ in their habitat use, as is the case in other polymorphic lizards such as the Dalmatian wall lizard, Podarcis melisellensis (Huyghe et al., 2007). It is possible that morphs differ in territory size rather than territory quality, as for U. stansburiana in which territory size is directly related to differences in each morph's reproductive strategy (Sinervo & Lively, 1996).


Our results show that discrete male throat colour morphs do exist within populations of C. decresii and that these groupings are supported by digital analysis of both colour and pattern as well as spectral modelling. The presence of discrete morphs suggests that the next step needed to understand the evolution, maintenance and adaptive significance of the colour variation is to examine throat colour heritability, test for correlations with behavioural and physiological traits (e.g. social behaviour, steroids, immunocompetence), and to examine changes in morph frequency and fitness over time. We also found substantial variation within morphs and future study is needed to determine whether this variation correlates with environmental conditions or individual traits such as physiology or behaviour. Morphs differed in conspicuousness against the background, which has implications for behavioural strategies; however, there was no evidence for correlated differences in dorsal colouration. We found no relationship between colour variation (discrete or continuous) and morphology or microhabitat use, which suggests that these traits do not play a role in maintaining the polymorphism. Further research is needed to determine which processes are involved and whether morphs differ in other behavioural, physiological and life history traits.


We thank Sally South, Claire Mclean and Adnan Moussalli for help in the field and Sandy Clarke for statistical advice. Funding was from the Australian Research Council (DP1092908). Martin Stevens was supported by a Biotechnology and Biological Sciences Research Council David Phillips Research Fellowship (BB/G022887/1). The research was conducted under the following permits: South Australia Department of Environment and Natural Resources Permit to Undertake Scientific Research (E25861-1), South Australian Wildlife Ethics Committee approval (18/2010), the University of Melbourne Animal Ethics Committee approval (1011760) and South Australian Licence for Teaching, Research or Experimentation Involving Animals (15/0231).