To remove carotenoids and thus reveal the underlying colour of barbs, we placed five feathers from each bird in glass tubes, added 2ml of acidified pyridine (Hudon & Brush 1992; three drops of HCl in 50ml pyridine) and capped the tubes. We then incubated the solution for 3h at 95°C. As a control, we performed the same procedures on white chicken feathers. After cooling to room temperature, feathers were removed from the solution, allowed to air dry, and taped in stacks of five to gloss-free black construction paper before spectral measurements were obtained (see below).

To disrupt structural colour, we attached separate groups of de-pigmented and intact yellow feathers from each bird in stacks of five to black construction paper and white index cards, and saturated them with Cresol (Sigma, St Louis, MI) using the blunt end of a pair of forceps. Cresol has a refractive index (RI) identical to that of keratin (RI1.54). The scattering of light as it moves between materials of different RIs causes structural colour (Prum 1999). Thus this treatment prevents structural colour reflectance in barbs by replacing air (RI=1.00) with cresol and eliminating the variation in the RI that causes colour production. Many structurally coloured tissues thus become transparent upon treatment with cresol (Mason 1923).

We obtained spectral reflectance curves while the feathers were saturated (see below). The construction paper and cards became saturated with cresol in the area where the feathers were attached, and this background was visible and hence recorded spectrophotometrically when the feathers became transparent owing to cresol treatment. To account for these changes in the background coloration, we took spectral measurements before and after they were treated with cresol.

We obtained reflectance curves using an Ocean Optics S2000 spectrometer (range 250–880nm; Dunedin, FL, USA) with a UV (deuterium bulb) and a visible (tungsten–halogen bulb) light source. Using a block sheath that excluded ambient light, we held a bifurcated micron fibre optic probe at a 90° angle 5mm from the yellow feather surface, creating a measurement area of 2mm in diameter. All data were generated relative to a white standard (WS-1, Ocean Optics). We used OOIbase software to record and average 20 spectra sequentially, and recorded and averaged measurements from five arbitrarily chosen points on each sample.

We next treated feathers with cresol, a liquid with the same RI as keratin that removes the structural component of coloration. Whether de-pigmented or naturally white, white feathers became transparent after cresol treatment (figure 2a). Yellow feathers treated with cresol became nearly transparent with a faint yellow cast (figure 2b) when placed against a black background. When these cresol-treated yellow feathers were placed against a white background, however, they once again appeared bright yellow (figure 2c). On a black background, coloration comes only from the direct reflection of light from the carotenoid pigments and this produces minimal coloration. On a white background, carotenoids absorb reflected light from the white background and again appear bright yellow. We obtained similar results when we applied the same methods to yellow feathers of prothonotary warblers (Prothonotaria citrea), cedar waxwings (Bombycilla cedrorum), and great crested flycatchers (Myiarchus crintus) (M. D. Shawkey, unpublished data). Thus, these results appear to apply to yellow carotenoid-pigmented feathers generally.

The yellow feathers of male American Goldfinches have the carotenoid pigments canary xanthophylls a and b (Stradi et al. 1995; McGraw et al. 2001), which are purported to produce the yellow feather colour. Here we show, however, that the yellow feathers of goldfinches, and probably of other species, also have white structural coloration that plays a key role in producing their yellow and UV reflectance peaks. Our observations show that carotenoids function primarily to absorb light at wavelengths from approximately 400 nm to 500nm from the structural white colour of the barbs.

Our microscopic examinations revealed that carotenoid pigments are scattered haphazardly throughout the keratin substrate of feather barbs (figure 3a). This spatial arrangement may allow carotenoids to absorb light from the white keratin substrate while weakly reflecting yellow light, resulting in the pure yellow coloration seen in figure 3b. Thus, carotenoids probably serve as semi-transparent filters in yellow feathers. The mechanisms of white colour production in feather barbs are largely unknown (Prum 1999), however, and elucidation of these mechanisms by precise physical modelling is needed before we can state definitively how white tissue and carotenoids interact to produce colour. Our results clearly show that both white structural colour and pigments are needed for the production of yellow colour, and these results have several important implications for the study of plumage coloration.

First, while the characterization of colour displays as pigment-based or structural has heuristic value and should continue to be used, researchers should understand that many pigmentary mechanisms have a structural component. Here, we have shown that carotenoid displays rely critically on underlying white structural coloration.

Second, the blurring of the distinction between mechanisms of production may add to the complexity of feather signalling properties. It has been shown that changes in red and yellow coloration, particularly in hue and saturation (i.e. colour purity), are proportional to changes in types and concentrations of carotenoids in the feather (Hill 2002; Saks et al. 2003). In turn, birds in poor condition and with greater parasite loads tend to be duller and have lower carotenoid concentration than those in good condition (Hill 2002). More carotenoids can cover the white structure more completely than fewer carotenoids, resulting in a more saturated and pure colour. Thus, that saturation and purity of carotenoid coloration would dependably reflect pigment concentration, which in turn could signal foraging ability or parasite load (Hill 2002). Other elements of yellow coloration, however, such as UV chroma may be affected by variation in structural, as well as pigmentary components. Indeed, differences in UV reflectance between male and female yellow-breasted chats (Icteria virens) cannot be explained by differences in carotenoid concentration alone, and thus are presumably influenced by differences in structural coloration as well (Mays et al. 2004).

The brightness, or total reflectance, of carotenoid-containing tissues is probably also strongly influenced by structural components of barbs, as the reflection of light (as opposed to the subtraction of light) is primarily a function of the underlying white structural colour. Recent studies of variation in achromatic brightness of white plumage patches suggest a signalling function for white plumage (Doucet et al. 2005). The production mechanisms and extent of variation in white coloration, however, remain largely unknown (Mennill et al. 2003). Clearly, more information on these topics is needed. Perhaps the structural and pigmentary components of feather coloration signal different, or redundant, information that is assessed simultaneously by the receiver. In this way, single feather patches could contain multiple ornaments (Candolin 2003).