Pigments versus structure: examining the mechanism of age-dependent change in a carotenoid-based colour


  • Simon R. Evans,

    Corresponding authorCurrent affiliation:
    1. Department of Animal Ecology, Evolutionary Biology Centre, Uppsala University, Uppsala, Sweden
    • Edward Grey Institute, Department of Zoology, University of Oxford, Oxford, UK
    Search for more papers by this author
  • Ben C. Sheldon

    1. Edward Grey Institute, Department of Zoology, University of Oxford, Oxford, UK
    Search for more papers by this author

Correspondence author. E-mail: simon.evans@ebc.uu.se


  1. Within-population colour variation is widespread in animals, yet the determinants of variable coloration have been relatively neglected by ecologists.

  2. Age-dependent expression of conspicuous coloration is prevalent, particularly in birds. Such patterns can be generated by multiple combinations of demographic heterogeneity or within-individual change; longitudinal analyses are necessary to establish the importance of these processes.

  3. Further, although pigment-based colours are composite traits, produced by multiple component mechanisms (e.g. feather microstructure and carotenoid pigmentation), the contributions of these mechanisms to components of age dependence are rarely considered, even though doing so may yield information about the ecological causes for age-dependent coloration.

  4. We used a large-scale, longitudinal study of carotenoid-based plumage coloration in great tits (Parus major) to show age dependence of plumage coloration is driven almost exclusively by within-individual effects in the first 2 years of life.

  5. Using wavelength-specific analyses, we show that feather microstructure, while sensitive to annual variation, is independent of age, with increased carotenoid deposition driving changes in coloration. However, estimates of local carotenoid availability did not explain the change in coloration within individuals, suggesting that pigment availability may not be limiting.

  6. We thus show that it is individual-level changes in the pigment component of carotenoid-based coloration that determines age-dependent colour expression in great tits. More generally, our study highlights the utility of wavelength-specific analyses in determining the mechanisms underlying changes in expression of composite colour traits.


Biologists have long been aware of within-population variation in colour expression (Darwin 1859; Wallace 1881) and, while the consequences of variable coloration have received considerable attention (see Dale 2006), how this variation is generated has been relatively neglected. This is unfortunate, because understanding the processes underlying variable coloration may help to explain the evolutionary consequences of colour expression and the ecological constraints that exist (Evans & Sheldon 2012). Carotenoid-based colours provide a notable exception, with interest in the determinants of expression being driven by their potential condition dependence (carotenoids cannot be synthesized de novo and must therefore be sequestered from the environment: Goodwin 1984), such that they have become model traits for the study of colour expression and signalling (see McGraw 2006).

In recent years, colour expression has been shown to be dynamic at multiple timescales (e.g. Delhey & Kempenaers 2006; Evans, Summers & Sheldon 2012). Many conspicuous colour patches have potential signal functions (e.g. ornamental coloration), such that variable expression can have wider ecological implications (e.g. interactions between conspecifics; Bradbury & Vehrencamp 1998). Such selective processes, mediated by interactions between individuals, are understood to be particularly potent evolutionary drivers, potentially resulting in rapid evolution (Wolf, Brodie & Moore 1999). In such cases, the information content of the signal, and thus any selective benefit, will be determined by its sensitivity to spatial and temporal variation (Cornwallis & Uller 2010). For example, any signal of genetic quality (Fisher 1930) is likely to be undermined by phenotypic plasticity. As such, determining the causes for variable colour expression is an important aim for biologists seeking to understand the evolution of animal coloration.

Conspicuous colour traits are typically positively age dependent, particularly in birds, where periodic regrowth of integumentary tissue during moult provides an opportunity for colour expression to be altered (e.g. Delhey & Kempenaers 2006; Evans et al. 2010). However, various processes could generate such patterns (e.g. selective mortality, within-individual change) and cross-sectional analyses are unable to distinguish between these (Cam et al. 2002). For example, greater colour expression in adults could be driven by higher mortality of less colourful individuals (Hõrak et al. 2001) or by individuals increasing colour expression as they age (Delhey & Kempenaers 2006; del Val, Quesada & Senar 2010). Furthermore, given that first-years typically exhibit lower colour expression (e.g. Evans et al. 2010), it is possible that the first- to second-year moult alone drives the overall positive age dependency. Using longitudinal analyses, previous work on the blue tit (Cyanistes caeruleus: Delhey & Kempenaers 2006) and the great tit (Parus major: del Val, Quesada & Senar 2010) has shown that plumage colour expression increases from first- to second-years but comparable studies are missing and the trajectory of age-dependent expression at older ages remains largely untested (though see Evans, Gustafsson & Sheldon 2011).

How changes in summary colour metrics correspond to shifts in the shape of reflectance spectra is largely unknown (Evans, Summers & Sheldon 2012), so we have little knowledge of the mechanisms underlying variation in plumage coloration. This is unfortunate because it has been increasingly recognized that pigment-based colours consist of two components that combine to determine reflectance: (i) a highly reflective and relatively achromatic (i.e. white) background, for which reflectance is determined by the physical microstructure of the feathers and limited melanin content and (ii) carotenoid pigments that are deposited into the feather structure and selectively absorb violet-blue (400–500 nm) light (Shawkey & Hill 2005; Shawkey et al. 2006; Isaksson et al. 2008; Jacot et al. 2010). It is intuitively appealing to assume that chromatic changes are driven by altered carotenoid deposition, given the relatively achromatic response that is expected from altered feather microstructure. However, chromatic indices that correlate with the carotenoid content of feathers (e.g. ‘carotenoid chroma’: Montgomerie 2006; Butler, Toomey & McGraw 2011) implicitly control for the broad-spectrum reflectance intensity of the plumage, implying that more reflective (i.e. brighter) backgrounds will yield greater chromaticity for a given concentration of carotenoids. Thus, observed changes in plumage reflectance, as described by summary variables (e.g. tristimulus colour scores: Montgomerie 2006), may be driven by a combination of altered carotenoid concentration and background reflectance. By improving our understanding of the mechanistic basis to colour variation, we can determine whether interpretations of the information content of carotenoid-based colour signals based on trait ontogeny (e.g. Endler 1983; Hill 1990) are empirically supported in the wild, and assess whether their composite nature is exploited to convey multiple messages within a single colour patch.

The ventral plumage of the great tit is a characteristic yellow and expression of this trait has received considerable attention (e.g. Evans et al. 2010 and references therein). A recent meta-analysis of the age dependence of this trait found that first-years exhibit lower expression than older birds (Evans et al. 2010), a pattern that has alternatively been attributed to both within-individual change (del Val, Quesada & Senar 2010) and selective mortality (Hõrak et al. 2001). However, how it is generated in terms of proximate mechanisms (e.g. carotenoid deposition versus feather microstructure) is not known, and the possibility of changes in coloration at older ages has not been explored. We make use of a large dataset of great tit plumage reflectance measures, collected across 4 years, to examine age-dependent expression of this trait in detail. We quantify plumage reflectance using independent metrics of chromatic (‘colour’) and brightness reflectance variation that correspond with current knowledge of avian visual sensitivities (Osorio & Voroboyev 2005; Kelber & Osorio 2010). Furthermore, we examine the wavelength-specificity of these changes by partitioning the bird-visible spectrum into narrow wavebands, which can then be analysed in parallel to provide an alternative perspective on changes in reflectance and from which changes in the component mechanisms governing plumage reflectance can be assessed. Having demonstrated that carotenoid deposition underlies the shift in chromaticity, we use individual-level analyses of spatial heterogeneity in habitat quality to assess whether estimates of local carotenoid availability can explain variation in chromaticity across moults.

Materials and methods

Study site and species

Data were collected between May 2008 and June 2011 (i.e. over four consecutive years) on a free-ranging population of great tits in Bagley Wood (51°42′N, 1°15′W), near Oxford, UK. Five hundred and ten nestboxes have been in place since January 2007, arranged in 12 spatially separated plots (Evans & Sheldon 2012). Birds were caught between September and March using baited mist nets, whilst breeding birds were caught at the nestbox during chick provisioning (May–June). Unringed birds were individually marked with a numbered aluminium leg-ring and sex was assigned based on standard plumage characteristics (Svensson 1994). First-years were identified by possession of distinctive, non-adult greater wing coverts (Svensson 1994). For individuals first caught as adults (i.e. ≥2 years old), we assigned minimum possible age based on the assumption that they were 2 years old when ringed (see, e.g. Brommer, Wilson & Gustafsson 2007; Bouwhuis et al. 2009). Of the 1500 individuals in our dataset, 193 (12·9%) were assigned estimated rather than exact ages in this way. For winter-caught birds, age was scored based on the age they would be in the coming breeding season, rather than absolute age since birth, because individuals moult into a new set of body feathers in late summer and maintain this plumage through the winter and the following breeding season. Thus, individuals caught in September that hatched during the preceding spring were scored as 1-year-olds (first-years).

Plumage reflectance

Plumage reflectance measurements were made using a field-portable spectrophotometer (USB4000; Ocean Optics, Dunedin, Florida) in combination with a xenon lamp (PX-2; Ocean Optics). These were connected via a bifurcated fibre-optic probe, at the objective end of which was fitted a cylindrical probe-holder (Andersson & Prager 2006) that ensured measurements were taken at a uniform distance from the feather surface (3·0 mm) and that ambient light was excluded. Measurements were taken at the mid-point of a line between the sternum and the right shoulder, using a coincident normal geometry (Andersson & Prager 2006). Reflectance was assessed relative to a dark standard and a white Spectralon tile. Three measurements were taken [themselves averages of 12 readings, calculated using the spectrasuite (v.12.2; Ocean Optics) software] and the mean of these was used as the plumage reflectance measure for that sample.

Plumage coloration was scored using SWS ratio and double-cone; respectively, chromatic and brightness indices of variation in plumage reflectance (Evans et al. 2010), which are calculated using a model of avian colour vision that corresponds with current understanding of avian visual processing (Osorio & Voroboyev 2005; Kelber & Osorio 2010). Previous analyses have shown that these two indices of reflectance variation are largely independent (see Evans et al. 2010). We used Hadfield's (2005) SPEC package to calculate estimated cone catches for each of the five cone types present in the avian retina (four single-cones and the double-cone). Single-cone photosensitivities were averages of species with UVS-type vision, provided in the TetraColorSpace package (Stoddard & Prum 2008). For the double-cone, we used photosensitivity data for the blue tit (Hart et al. 2000), due to the limited availability of double-cone sensitivity data for other species. Because the lighting environment is expected to be variable in natural conditions, we used standard daylight (D65) as the incident lighting spectrum, which represents light of an intermediate composition between blue sky and forest shade. SWS ratio compares the quantum catch for the SWS cone (which has peak sensitivity in the violet-blue region of the spectrum, where the reflectance spectrum of great tit ventral plumage exhibits a trough; see Fig. 3a) to the mean quantum cone catch of the other three single-cones (UVS, MWS and LWS):

display math

Spectra with a large contrast between the sectors of high [UV (320–400 nm) peak and yellow-red (500–700 nm) plateau] and low (violet-blue trough) reflectance (see Fig. 3a), which will have greater chroma (saturation), will have higher scores for SWS ratio. In this way, SWS ratio provides a bird vision-focussed measure of chromatic variation in plumage reflectance (see Evans et al. 2010). Alternatives are available (e.g. Stoddard & Prum 2008), but our approach yields a single metric of chromatic variation and thus avoids the interpretational difficulties that can arise when multiple metrics are analysed concurrently (see Evans et al. 2010). How well SWS ratio corresponds to opponency mechanisms of avian subjects is uncertain: data on this are currently lacking (though see Ventura et al. 2001; Rocha et al. 2008). However, research on other tetrachromatic taxa has shown this opponency channel to exist in vivo (Rocha et al. 2008). To achieve approximate normality, we used the natural logarithm of SWS ratio.

Age-dependent plumage reflectance

We calculated annual mean values for each individual sampled on multiple occasions within a single moult. Given that SWS ratio exhibits a linear seasonal (i.e. within-moult) decline (Evans, Summers & Sheldon 2012), we also calculated mean Septemberday values for each year-individual combination. Septemberday is defined as the number of days elapsed since the 31 August (i.e. for measurements on 1st September, Septemberday = 1: Evans et al. 2010). Significant annual variation exists in this population with respect to both plumage chromaticity and achromaticity (Evans & Sheldon 2012). Given that sample sizes are unevenly distributed with respect to year, population-level annual changes could drive apparent age dependency if not explicitly controlled for (see Discussion). We calculated year-standardized scores by subtracting the annual mean from each measure; the overall (across all years) mean was then added so that values were biologically meaningful.

Following van de Pol's (van de Pol & Verhulst 2006; van de Pol & Wright 2009) within-subject centring approach, we calculated within-individual age terms by subtracting an individual's mean age value from each observation age (i.e. xij − xj', where xij is the age value of measurement i from subject j and xj′ is the mean age value of subject j). We constructed linear mixed effects models (LMMs) of year-standardized chromatic (SWS ratio) and brightness (double-cone) variation in plumage reflectance, in which individual identity was fitted as a random effect to control for repeated measurements of individuals. Initial models included Septemberday, sex and mean individual age, along with linear and quadratic terms for within-individual age and their interactions with sex. The quadratic within-individual age term allowed us to test for non-linear responses to age; whilst a non-linear age response could conceivably be modelled by various functions, the quadratic provides a flexible descriptor of non-linearity. To aid comparison across traits, dependent variables were transformed into z-scores.

To illustrate the contribution of within-individual change – as well as the influence of demographic heterogeneity – to the pattern of age dependency in plumage chromaticity observed at the population level (Fig. 1a), we follow the methodology of Rebke et al. (2010). Population-level phenotypic expression (P) is decomposed into within-individual change (s), selective disappearance (d) and selective appearance (a):

display math


display math

is the difference between mean expression at one age (Vx) and the previous age (Vx−1);

display math

is the difference in mean expression for individuals represented at both the previous age (vx−1) and the focal age (vx);

display math

gives the difference between the mean of surviving individuals at the earlier age (vx−1) and the mean of all individuals at that earlier age (i.e. including non-survivors); and

display math

gives the difference between the two means at the older age (i.e. Vx is the population-level mean at age x, while vx is the mean of individuals at age x that were also sampled at the previous age).

Figure 1.

Population-level age dependence (mean ± SE) of (a) chromatic (SWS ratio) and (b) brightness (double-cone) variation in reflectance of ventral plumage in the great tit.

Based on the results of the analyses across all age classes, we created two smaller datasets, one representing first- and second-years, the other representing individuals at least 2 years old (hence second-year measurements were present in both datasets). Analysing these two datasets separately allowed us to determine the extent to which the lower chromaticity of first-years (Fig. 1a) was the exclusive driver of the positive age dependency we observed across all ages (see Results). Given the inevitable decline in sample sizes with increasing age and the potential power limitation this may impose on statistical assessment of late-life change, we supplemented our within-subject centred analysis of the adult dataset with a repeated measures-based assessment, examining whether there was statistical support for a decline in plumage chromaticity after the within-individual peak (4 years old: Fig. 2). Despite pooling post-peak ages (i.e. ≥5 years), we only had 16 individuals that were sampled at 4 years of age and again in a subsequent year. One of these individuals was measured at two post-peak ages – when aged five and again the following year – and for this individual we used the later measurement. We tested for a late-life decline in plumage chromaticity using a one-tailed matched-pairs t-test and, as previously, used year-standardized values of SWS ratio.

Figure 2.

Contribution of within-individual and demographic processes to population-level age dependency of plumage chromaticity (SWS ratio). Each of the three trajectories representing a component process shows the age-dependent pattern that would be observed if that process alone was acting. The population-level change observed across any two age classes is the sum of the changes in the three component processes. Chromaticity scores were year-standardized.

Spectral compartmentalization

To examine the wavelength-specificity of the change in reflectance across the first- to second-year moult, we collated the first- and second-year reflectance spectra of the 236 individuals that were repeat-measured across this moult. We divided the bird-visible spectrum (320–700 nm: Andersson & Prager 2006; Montgomerie 2006) into 10-nm-width wavebands, for each of which we calculated the median reflectance value. For individuals measured multiple times within a single year (i.e. moult), we followed our procedure for the summary colour variables in calculating the within-moult mean, such that each individual is represented only once at each age. As for the summary colour variables, each waveband was year-standardized to control for population-level annual change, by subtracting the annual mean from each value. The overall (across years) mean was then added, to generate biologically relevant reflectance values for which the effects of population-level annual change had been eliminated. We conducted two-tailed matched-pair t-tests for each 10-nm waveband.

Given that the shape of reflectance spectra is determined by a combination of broad-spectrum reflectance, controlled by feather microstructure and melanin content, and wavelength-specific light filtration by carotenoids deposited into the feathers (Shawkey & Hill 2005; Shawkey et al. 2006; Isaksson et al. 2008; Jacot et al. 2010), we can expect wavebands to be non-independent. Thus, there will be a considerable degree of repeatability to our analyses. However, covariance across wavebands is representative of the underlying spectrum: if reflectance changes dramatically over a narrow spectral range, our spectrophotometer-based spectral measurement approach can reproduce this (as we see at the ‘transition wavelengths’, where the reflectance shifts from relatively high to low reflectance values across a narrow bandwidth; Fig. 3a). Thus, we can be confident that our wavelength-dependent responses are representative of the reflectance spectra of natural objects and the processes we aim to describe.

Figure 3.

(a) Mean ventral plumage reflectance spectra of first- and second-year great tits measured at both ages (nindividuals = 236). (b) The proportional change in mean reflectance across the first- to second-year moult transition. Points are shaded according to the result of a two-tailed matched-pairs t-test for each 10-nm waveband that examined the change in reflectance across the first- to second-year moult. The inset figure is a plot based on the same procedure but where annual change at the population level was not eliminated.

Local carotenoid availability

Having shown that the change in chromaticity across the first- to second-year moult is driven by increased carotenoid deposition, we examined whether aspects of individuals' immediate environment can explain the change in chromaticity they exhibit. Oak-feeding caterpillars are known to be a rich source of carotenoids for great tits during the spring and early summer months which precede the annual moult (Partali et al. 1987). Thus, if carotenoid deposition is environmentally limited, the local availability of caterpillars may influence the increase in chromaticity that individual great tits can achieve. Whilst caterpillar density can be measured directly (Zandt 1994; van Noordwijk, McCleery & Perrins 1995), the abundance of oak (Quercus spp.) – the preferred food source for those lepidopteran species heavily preyed upon by great tits (Summerville et al. 2003) – is a surrogate measure for which data can be more easily collected across a large area. Indeed, the amount of oak vegetation is an established correlate of caterpillar density (Wint 1983; Foss & Rieske 2003).

Following natal dispersal and recruitment, great tits are highly philopatric (Harvey, Greenwood & Perrins 1979) and only a small minority of individuals (16·7%) have been recorded in more than one nestbox plot during the 4 years of our study. To gain precise locations of each individual, we used those birds recorded as breeders when they were 1 or 2 years old (225 of 236 repeat-measured individuals). The location of each nestbox was recorded using a handheld GPS unit. A forestry atlas of the site was available (Bagley Wood is actively managed for timber production), detailing the species composition of the canopy within each of 297 compartments across the site (mean compartment size: 0·67 Ha; range: 0·01–7·24 Ha). Using Geographic Information Systems (GIS) software, we digitally mapped these compartments and, for each nestbox, calculated the average oak composition (as a proportion of total canopy cover) within a 75 m radius, based on the results of a previous spatial analytical study on a nearby great tit population that established the 75 m buffer to be a relevant scale for estimating oak availability (Wilkin, Perrins & Sheldon 2007; see also Hinks 2010). Our measure of oak density can thus vary between 0 (areas where oak was completely absent) and 1 (where the canopy consisted solely of oak species). Across all nestboxes, oak density values ranged from 0 to 0·87 (mean = 0·32). For individuals recorded as breeders in both years, we used the mean oak composition of the two nestboxes (distances between breeding sites are typically small: Harvey, Greenwood & Perrins 1979).

To explore the relationship between local carotenoid availability and plumage chromaticity, we examined how well local oak density at the time of moult predicted plumage coloration of first- and second-years. As an estimate of local carotenoid availability prior to the moult in which first-year plumage is acquired (the post-fledging moult), we used the oak density score of the natal nestbox. Of the 1639 individuals sampled as first-years, 485 were ringed as nestlings, so we knew their site of origin. We constructed a LMM of first-year plumage chromaticity, with Septemberday, sex and natal oak density included as fixed effects, and brood-of-origin included as a random effect to control for the non-independence of siblings. To estimate contemporary carotenoid availability at the time second-year plumage is grown we again used the nesting location of individuals recorded as breeders. We constructed a general linear model of second-year plumage chromaticity, including Septemberday, sex and contemporary oak density in the initial model.

Statistical analyses were conducted in jmp v.9.0.0 (SAS Institute, Cary, NC, USA) and based on restricted maximum likelihood. For all models we report results for the minimum adequate model generated by a backwards selection procedure in which a saturated model is first fitted. Terms were removed in order of increasing significance and following rules of marginality to produce a final model composed of terms where P < 0·1.


The model of plumage chromaticity found a highly significant increase as individuals age, with a mean shift of 0·45 standard deviations per year, which exceeds the sexual dichromatism (Table 1a). This linear change was independent of sex, and there was no support for a nonlinear age dependence. For the model of plumage brightness (Table 1b) both interaction terms including sex were statistically significant, indicating that the age trajectory of plumage brightness is sex-dependent. However, given the marginal statistical support for these effects and their relatively small effect sizes, we focussed our attention on further quantifying the positive age dependence of plumage chromaticity.

Table 1. Models examining the within-individual effects of age on chromatic (SWS ratio) and brightness (double-cone) aspects of plumage coloration in the great tit
 d.f.EstimateSE F P Exclusion sequence
  1. Terms retained in the final models are shown in bold. Parameter estimates for terms including sex represent the value for females. Individual identity was included as a random effect, and response variables were standardized (z-scores) before analysis to aid comparison.

(a) Year-standardized SWS ratio (n = 1941)
Septemberday 1, 1933 0·0008 0·0003 8·26 0·004  
Sex [f] 1, 1309 0·1944 0·0220 78·3 0·0001 
Individual mean age 1, 985 0·3466 0·0286 147 < 0·0001 
Within-individual age 1, 477 0·4469 0·0452 97·6 0·0001 
(Within-individual age)21, 1865−0·07020·07570·8590·3543
Within-individual age × sex [f]1, 4660·01340·04530·0880·7672
(Within-individual age)2 × sex [f]1, 18340·10040·06982·070·1501
Final modelR2adj. = 43·9%
b) Year-standardized double-cone (n = 1941)
Septemberday1, 19300·00010·00030·03930·84281
Sex [f] 1, 1847 0·3523 0·0246 205 < 0·0001  
Individual mean age1, 13860·01170·03060·1470·7022
Within-individual age 1, 508 −0·0989 0·0474 4·35 0·038  
(Within-individual age)2 1, 1720 0·0104 0·0714 0·021 0·885  
Within-individual age × sex [f] 1, 509 −0·0962 0·0474 4·12 0·043  
(Within-individual age)2 × sex [f] 1, 1720 0·1544 0·0714 4·68 0·031  
Final modelR2adj. = 36·9%

Partitioning the population-level age dependency into the three contributing processes – selective disappearance, selective appearance and within-individual change – demonstrated that, from a longitudinal perspective, the first-to second-year moult transition is by far the main contributor to the positive within-individual age dependency (Fig. 2). In comparison, the effects of demographic heterogeneity (selective disappearance and appearance) on population-level patterns are minor. We therefore repeated our analysis using two new datasets: the first incorporating only first- and second-year birds, the other excluding first-year birds. We fitted only a linear within-individual age term in the initial model for younger great tits, as the dataset incorporates only a single moult transition (first- to second-year). The results (Table 2) show that it is the first- to second-year transition that generates the positive within-individual age effect on plumage chromaticity found across all ages: across this moult transition, plumage chromaticity exhibited a highly significant within-individual increase in SWS ratio, equivalent to nearly one standard deviation (Table 2a). There was no evidence for across-moult change in plumage chromaticity at older ages (Table 2b). Further, when pooling individuals at or beyond 5 years of age, there was no evidence for a late-life decline in plumage chromaticity after 4 years of age (t15 = −0·845; P = 0·206).

Table 2. Models describing the within-individual effects of age on plumage chromaticity (SWS ratio) in great tits in (a) first- and second-years, and (b) second-years and older birds
 d.f.EstimateSE F P Exclusion sequence
(a) Year-standardized SWS ratio (1–2-year-olds; = 1639)
  1. Terms retained in the final models are shown in bold. Parameter estimates for terms including sex represent the value for females. Individual identity was included as a random effect, and response variables were standardized (z-scores) before analysis to aid comparison.

Septemberday 1, 1602 0·0016 0·0003 26·6 0·0001 
Sex [f] 1, 1300 0·1978 0·0233 72·1 0·0001 
Individual mean age 1, 1565 0·7927 0·0657 146 0·0001 
Within-individual age 1, 231 0·9528 0·0705 183 0·0001 
Within-individual age × sex [f]1, 230−0·07320·07131·060·3051
Final modelR2adj. = 57·4%
(b) Year-standardized SWS ratio (2+ year olds; n = 796)
Septemberday 1, 776 −0·0017 0·0006 9·41 0·002  
Sex [f] 1, 553 0·2562 0·0364 49·4 0·0001 
Individual mean age1, 441−0·00020·05400·0000·9964
Within-individual age1, 2250·01580·06540·0580·8105
(Within-individual age)21, 653−0·01730·10900·0250·8742
Within-individual age × sex [f]1, 216−0·02020·06590·0940·7593
(within-individual age)2 × sex [f]1, 7840·03640·10180·1280·7211
Final modelR2adj. = 51·8%

Examining the wavelength-specific changes in plumage reflectance that accompany the first- to second-year moult transition (Fig. 3) suggests that the change in chromaticity was driven exclusively by increased carotenoid deposition into the feathers. Repeat-measured birds exhibited a highly significant reduction in reflectivity in the violet-blue (400–500 nm) region of the spectrum, with the multi-peaked shape consistent with this being caused by increased levels of carotenoids in the feathers (Fig. 3b; Zsceile, White & Beadle 1942; Ruban, Horton & Young 1993; see also Evans et al. 2010; Evans, Summers & Sheldon 2012). Beyond the violet-blue region of the spectrum, we found little support for a change in plumage reflectance, as would have been expected if broad-spectrum reflectance was altered via modifications to feather microstructure or melanin content. These analyses also serve to demonstrate the importance of controlling for population-level temporal effects – the inset in Fig. 4b, which is based on raw reflectance values (without accounting for annual change at the population level), shows that a very different conclusion on underlying mechanisms would have been reached if year effects were neglected.

Figure 4.

The relationship between carotenoid availability (as estimated by local oak density) and the change in plumage chromaticity shown by great tits across the first- to second-year moult. Chromaticity scores were year-standardized prior to calculating the difference across the moult, to control for annual variation in expression.

Having established that increased carotenoid deposition into the feathers was responsible for the increase in plumage chromaticity across the first- to second-year moult, we assessed whether the increase in coloration was predicted by local carotenoid availability, using the 225 repeat-measured individuals of known breeding location. Linear regression analysis showed a small, non-significant effect of carotenoid availability – as estimated by local oak density – on the across-moult change in chromaticity (β = 0·0006 ± 0·0006; F1,223 = 0·986; P = 0·322; Fig. 4). However, across-moult change is dependent on chromaticity at the two ages, so effects of carotenoid availability on chromaticity at either age may be obscured. Our analyses of plumage chromaticity at the two ages showed divergent results with respect to local carotenoid availability at the time of moult. The positive relationship between natal carotenoid availability and plumage chromaticity of first-years (i.e. after the post-fledging moult) was weakly significant (Table 3). However, plumage chromaticity of second-years was not related to local carotenoid availability density (Table 4), suggesting that the availability of pigments in the environment was not a limiting factor for carotenoid-based plumage coloration of adult (2+) birds.

Table 3. Model examining whether the availability of carotenoids in the immediate natal environment predicts plumage chromaticity (SWS ratio) of first-year great tits of known natal origin (n = 485)
 d.f.EstimateSE F P
  1. The brood-of-origin was included as a random effect to control for the non-independence of siblings. The response variable was year-standardized and z-transformed to allow comparison of effect sizes with previous models.

Septemberday1, 481−0·00210·000515·10·0001
Sex [f]1, 421−0·23350·039734·7< 0·0001
Natal oak density1, 2230·00480·00234·480·035
Final modelR2adj. = 62·4%
Table 4. Model examining whether the availability of carotenoids in the immediate contemporary environment predicts plumage chromaticity (SWS ratio) of second-year great tits (n = 450)
 d.f.EstimateSE F P Exclusion sequence
  1. The response variable was year-standardized and z-transformed to allow comparison of effect sizes with previous models.

Septemberday1, 447−0·00220·00087·830·005 
Sex [f]1, 447−0·23730·038438·1< 0·0001 
Local oak density1, 4460·00250·00191·860·1741
Final modelR2adj. = 9·0%


Over 4 years, we collected plumage reflectance measurements from a nestbox-breeding great tit population to assess the age-dependent expression of the yellow ventral plumage, a carotenoid-based colour patch. By analysing both chromatic (‘colour’) and brightness aspects of reflectance variation separately (in accordance with current understanding of avian visual perception: Osorio & Voroboyev 2005; Kelber & Osorio 2010) we were able to demonstrate a marked contrast in age dependency: whereas evidence of change in plumage brightness was weak, plumage chromaticity exhibited a highly significant, positive age dependency. Further analysis of this relationship showed that it was driven exclusively by a marked increase in chromaticity during the first- to second-year moult. Comparison of wavelength-specific changes in plumage reflectance across the first- to second-year transition demonstrated that this change in plumage reflectance resulted from increased carotenoid deposition into plumage. However, local carotenoid availability, as estimated by oak density, did not explain individual variation in this increase.

The observed increase in plumage chromaticity across the first- to second-year moult was large, representing one standard deviation in a dataset of first- and second-years, and decomposition of the population-level change confirmed that within-individual change was the sole process driving this relationship. Nonetheless, it should be recognized that there remains uncertainty regarding the perceptibility of this age dichromatism from a conspecific perspective (Evans et al. 2010), which will impact on the potential for age status to be communicated by expression of this trait. Analysing the wavelength-specific changes in reflectance showed that this change in chromaticity is driven by increased levels of carotenoids being deposited into the feather, with feather microstructure being age-independent. Why first-years deposit lower levels of carotenoids into their plumage remains unclear. It may be that individuals have access to a smaller pool of circulating carotenoids during their first moult, which occurs within a few months of fledging (Jenni & Winkler 1994). This could arise due to fledglings moulting later in the summer than older birds (Cramp & Perrins 1993). If carotenoid availability is seasonally variable–for example, due to a decline in the availability of the carotenoid-rich caterpillars that great tits rely on as a major food source during late-spring and early summer (Slagsvold & Lifjeld 1985; Partali et al. 1987) – then later-moulting birds may experience greater pigment limitation. Nutritional limitation at the time of moulting has been applied in this species to explain a similar pattern with respect to wing length, another plumage-based trait, with first-year great tits having shorter wings than older birds (van Balen 1967). Such an explanation is consistent with the positive relationship we report between first-year plumage chromaticity and natal oak density. Nevertheless, the effect size was small and both non-random settlement of parents and differential timing of breeding could also contribute to this relationship (Evans & Sheldon 2012), suggesting that carotenoid availability may not impose so severe a limitation on colour expression in the wild as previously understood (see also Evans & Sheldon 2012).

Having shown that the marked increase in plumage chromaticity across the first- to second-year moult was driven by increased carotenoid deposition, it seems reasonable to suggest that carotenoid availability in the local environment may explain individual-level variation in the chromatic change. However, using a surrogate measure of the abundance of carotenoid-rich caterpillars in the immediate environment, we found no evidence that carotenoid availability influenced the change in chromaticity, suggesting that if environmental limitation of carotenoids does operate in natural conditions, it may be weaker than generally expected. It is possible that local oak density is a poor estimate of individual-level access to dietary carotenoids. However, whilst more direct measures of caterpillar abundance (Zandt 1994; van Noordwijk, McCleery & Perrins 1995) might be preferable, oak abundance is known to predict caterpillar density (Wint 1983; Foss & Rieske 2003; Summerville et al. 2003) and has been successfully used as a measure of habitat quality in studies of tit ecology (e.g. Wilkin, Perrins & Sheldon 2007; Wilkin, King & Sheldon 2009; Parker et al. 2011). Further, caterpillars are the richest source of carotenoids in the summer diet of the great tit (Eeva et al. 2010) and are thus likely to make major contributions to the internal carotenoid supplies of both fledglings and mature birds prior to moulting (Isaksson, von Post & Andersson 2007).

If individual ranges are larger than we have modelled then this may mitigate against dietary limitation in low quality habitats. However, a large majority of individuals (83·3%) are recorded within a single nestbox plot after natal dispersal, and the plot design is such that the dendrofloristic composition is relatively uniform within plots (Evans & Sheldon 2012). Thus, wider ranging individuals will often experience a similar environment as far as canopy composition is concerned. Furthermore, the moult into second-year plumage occurs immediately after the breeding season (Jenni & Winkler 1994), during which time the ranging abilities of individuals are tightly constrained by the species' territoriality (Krebs 1971) and the enormous time demands of raising offspring (Thomas et al. 2001). Lastly, the chosen spatial scale (75 m radius) has previously been shown to be the relevant scale for relating oak abundance to reproductive success in a nearby great tit population (Wilkin, Perrins & Sheldon 2007; Hinks 2010). It thus seems likely that our measure will relate to dietary carotenoid intake and, given that oak density predicted neither the change in chromaticity nor the chromaticity of the newly acquired second-year plumage, we conclude that dietary carotenoid availability was not a limiting factor. In combination with the results of a recent quantitative genetic analysis of plumage chromaticity in this population (Evans & Sheldon 2012), our results suggest that the importance of contemporary spatial environmental factors in determining carotenoid-based colour expression in natural conditions may have been over-estimated.

The importance of accounting for annual change in plumage reflectance was demonstrated by the inset figure examining wavelength-specific change using unmodified reflectance values (Fig. 3b). Failure to account for annual shifts in reflectance at the population-level would have led us to conclude that broad-spectrum reflectance also increases across the first- to second-year moult, a change attributable to feather microstructure (Jacot et al. 2010) and potentially also melanin content (Isaksson et al. 2008). Plumage brightness has recently been shown to exhibit considerable annual variation in this population (Evans & Sheldon 2012). Thus, as discussed by Evans & Sheldon (2012), background reflectance appears to be a highly plastic trait that is sensitive to large-scale environmental factors (e.g. climatic variation). Regardless of what drives this variation, studies examining variation in plumage reflectance will need to consider the potential for results to be confounded by such annual changes. Indeed, failure to account for annual variation in plumage chromaticity would have led us to conclude that chromaticity is positively age-dependent beyond the first- to second-year moult transition (unpublished results), a relationship that is driven by the confound between individual age and time.

The lack of change in plumage chromaticity beyond the first- to second-year moult contrasts with a recent study that also used within-subject centring (van de Pol & Verhulst 2006; van de Pol & Wright 2009) to examine the age-dependent expression of a conspicuous colour patch: the ornamental white plumage patches of collared flycatchers (Ficedula albicollis; Evans, Gustafsson & Sheldon 2011). In this case, expression was positively age-dependent amongst adults, although it should be noted that variable expression was in terms of the size of the colour patch, rather than its spectral composition. However, the mean annual increase reported by Evans, Gustafsson & Sheldon (2011) was very small and, given our much smaller sample size, it may be that statistical power would have been insufficient to detect an effect of this size; such sampling limitations are commonplace in studies of age-dependent phenotypic expression (Nussey et al. 2008). Furthermore, whilst the ornamental function of colour patches is well established for the collared flycatcher, the function of the yellow plumage of great tits remains contentious (e.g. Isaksson et al. 2008), such that both mechanistic and functional factors could explain the differences between these two studies. A long-term study of blue-footed boobies (Sula nebouxii) found that male foot colour, a carotenoid-based trait, is negatively age-dependent at the population level (Torres & Velando 2007; Velando, Drummond & Torres 2010), although the absence of longitudinal analyses make the contribution of within-individual change uncertain (Cam et al. 2002; van de Pol & Verhulst 2006), and other studies of carotenoid-based colour expression have suggested that investment increases with age (Candolin 2000; Cote et al. 2010). Phenotypic decomposition showed that a non-significant late-life decline occured in our population but the decline was small and, given the inherent difficulty of sampling old-age birds (Nussey et al. 2008), a much longer period of data collection would be required to assess this formerly (e.g. Bouwhuis et al. 2009; Evans, Gustafsson & Sheldon 2011).

Our results show that the age dichromatism of this trait (Evans et al. 2010) is driven by a longitudinal process in which individuals increase plumage chromaticity when they acquire second-year plumage, with chromaticity remaining static thereafter. This increase is driven by increased carotenoid deposition, with no apparent change in background reflectance, indicating that feather microstructure and melanin content are independent of age, despite apparent sensitivity to external environmental factors. However, we found no evidence that local carotenoid availability influences the change in chromaticity, leading us to conclude that carotenoid limitation, at least on the scale we analyse here, may be weak once individuals have attained full independence from their parents (great tits continue to care for their offspring for several weeks after they have fledged: Naef-Daenzer & Grüebler 2008). Our study highlights how wavelength-specific responses can be modelled by partitioning the spectrum into narrow wavebands that are then analysed in parallel, allowing the component mechanisms that determine phenotypic expression of composite colours to be assessed (see also Evans, Summers & Sheldon 2012). We believe this approach can be a useful tool for studies of coloration, particularly for pigment-based traits, for which a highly reflective background will be present alongside pigment molecules, with the two combining to determine the expressed colour.


We are grateful to St. John's College for allowing us to conduct research at the study site and to their staff for accommodating us; M. Munirul Islam for calculating the oak density values of nestboxes; Hannah Edwards, Camilla Hinde, Julian Howe, Caroline Isaksson, Adele Mennerat, Nicole Milligan, Sarah Roberts, Teddy Wilkin, Helen Wilkinson and Stuart Will for their assistance during breeding seasons; the National Environment Research Council (NE/D011744/1) and the European Research Council (AdG 250164) for funding (to BCS); Anne Charmantier & Charlie Cornwallis for advice with data analysis; and A. Lord and several anonymous reviewers for their comments that greatly improved previous versions of the manuscript.