The plumage of the European starling (S. vulgaris) is gray-black with whitish spots, but it can prominently feature metallic blue, green or purple feather areas (Figure 1A). Epi-illumination light microscopy demonstrated that the colors originate from the feather barbules, which branch from either side of the barbs and partly overlap each other (Figures 1B,C). The barbules consist of a row of more or less flat cells, each exposing a surface with size ∼30 × 60 μm². The hue can differ between neighboring cells, but within a cell the hue varies only slightly, which indicates that the anatomical variation within a single barbule cell is small (Figures 1D,E).

As shown by electron microscopy, the barbule cell border is marked by a layer of melanosomes below a keratin cortex layer (Figures 2A,B). Numerous melanosomes exist in the barbule’s central area, but they are randomly distributed (Figures 2A,B). The TEM data from previous literature yielded virtually identical values for the melanosome diameters of the green and purple barbules: D_m = 250 nm and D_m = 254 nm, respectively, (Durrer and Villiger, 1970). Our SEM images of cross-sectioned barbules (Figure 2B) also yielded similar, but slightly larger mean diameters of 273 ± 20 nm and 271 ± 25 nm for the green and purple barbules, respectively. The cortex thickness of the green and purple barbules varies over the width of the barbule and was estimated to be 275 ± 30 nm and 336 ± 27 nm, respectively.

To visualize the melanosomes of the starling’s feather barbules in situ, we applied epi-illumination microscopy on barbules immersed in oil (Figure 2C). The micrographs show ∼1.5 μm long, rod-shaped melanosomes, arranged in clusters that are generally oriented parallel to the barbule axis (Figures 2C,D; Figure 2C shows an example of a melanosome cluster with a deviant orientation). The micrographs yielded an interdistance of the melanosome rodlets in the green barbules of 349 ± 25 nm and for the purple barbules 351 ± 22 nm, meaning that the melanosomes are on average spaced apart by D_i = ∼80 nm.

The anatomy suggests that the barbule’s cortex and melanosome layer form a simple layer-based interference reflector that causes the structural coloration. We tested this hypothesis by applying imaging scatterometry (Figure 2F). Local illumination of a purple barbule by a narrow aperture light beam indeed produced a scatterogram with a very restricted light spot. This signature of a very directional reflection validates the assumption. We note, however, that frequently the local scatterogram patterns deviated from an ideal, single spot. Clearly, the barbule is not flat throughout its width, as also evident from the anatomical data (Figures 2A,B).

We calculated the barbule’s reflectance spectra with an effective-medium-multilayer model (Figure 3). We treated the barbule structure as a multilayer, consisting of a keratin cortex (thickness c, facing air) and a single layer of rodlet-shaped melanosomes (diameter D_m, interdistance D_i) in a keratin matrix (see inset of Figure 3A). The colored feathers differ in the thickness of the barbule cortex, as for blue-green and red-violet feathers c = 290 nm and c = 330 nm, respectively (Figure 2 and Durrer and Villiger, 1970). For our initial calculations, we took for the cortex thickness c = 310 nm together with a melanosome diameter D_m = 250 nm as a starting point, as these values should produce reflectance values in the visible wavelength range. We used these values and the refractive indices of keratin and melanin in Eq. 1 to calculate the real and imaginary effective refractive index profiles, Re⁢(n~eff) and Im⁢(n~eff), along the coordinate perpendicular to the barbule surface. We then investigated the influence of varying parameters within a realistic parameter range chosen to be slightly larger than the derived values.

At first, we investigated three cases, where the melanosome interdistance is D_i = 0, 50, and 100 nm. By applying a transfer matrix-based calculation procedure, we then obtained the reflectance and transmittance spectra (Figures 3B,C). The reflectance spectra of Figure 3B show two distinct bands, peaking in the UV, at ∼340 nm, and in the green, at ∼510 nm. The reflectance amplitude slightly decreases with an increasing melanosome separation (Figure 3B), together with a slight increase of the transmittance (Figure 3C), but the spectral shapes hardly depend on D_i. To check the effective medium results, we performed FDTD calculations for the case D_m = 250 nm, D_i = 0 nm, and c = 310 nm, for both TE- and TM- polarized light, yielding reflectance spectra that are virtually identical to those of Figure 3B (Supplementary Figure 1).

The transmittance increases with increasing wavelength, which is immediately recognized as the hallmark of melanin absorption. To assess the possible effect of absorption on the reflectance, we assumed that the melanin refractive index is real instead of complex. The reflectance spectrum then obtained, for the case D_i = 0 (Figure 3B, green curve), has a slightly higher amplitude than the spectrum resulting with the complex refractive index, and its peak wavelength is a few nm shifted, but overall the effect of melanin absorption on the reflectance spectrum can be considered to be minor. When fully neglecting melanin absorption, given a reflectance of at most a few percent (Figure 3B), the transmittance becomes almost 100%, as expected (Figure 3C, green curve).

In studies of the structural coloration by melanosomes, the layers are often assumed to create discrete, stepwise changes in refractive index (see, e.g., Doucet et al., 2006). In our effective medium approach, we implemented more realistic, gradual changes in the refractive index by interpolation and thresholding with a very fine mesh. We investigated the difference between gradual and stepwise changes of the refractive index by comparing an array of contiguous solid rodlets, diameter D_m = 250 nm and D_i = 0 nm, with a few discrete layers, homogeneously filled with the same total amount of melanin (Figures 3D–F). Figure 3D shows the real and imaginary effective refractive index profiles for four cases: (i) the rodlets with gradually changing melanin fraction f_m with averaged melanin content (πD_m²/4)/D_m = πD_m/4; (ii) a discrete layer with pure melanin, f_m = 1 and thickness πD_m/4; (iii) a discrete layer with reduced melanin, f_m = π/2 and thickness D_mπ/2; (iv) a discrete layer with even less melanin, f_m = π/4, and thickness D_m. In all cases, the location of the center of the melanosome layers was identical (Figure 3D). The reflectance spectra calculated for the four cases have similar shapes and peak wavelengths (∼510 nm) in the visible wavelength range, but the reflectance magnitudes of the extrema differ slightly (Figure 3E). As expected, the transmittance spectra are virtually identical, because the total melanin content is identical (Figure 3F).

The total melanin content is not constant when varying the diameter of the melanin rodlets (Figures 3G–I). We considered three cases of the melanosome diameter: D_m = 250, 300, and 350 nm, with interdistance D_i = 0 nm, keeping a constant cortex thickness c = 310 nm. Again, the detailed shape of the resulting reflectance spectra varies, but the spectral location of the UV stays about the same and the band in the green wavelength range (up to about 630 nm) varies only in its detailed shape (Figure 3H). In the long wavelength range, the reflectance peak moves bathochromic (to longer wavelengths) due to the increased optical pathlength resulting from the thicker melanin layer. Also, with an increasing rodlet diameter, the melanin content increases, causing a decrease in the transmittance (Figure 3I).

We finally investigated the effect of the cortex thickness. In addition to the values c = 290 and 330 nm informed by EM measurements of Figure 2B and literature (Durrer and Villiger, 1970), we also applied slightly smaller c = 250 nm and larger thicknesses c = 370 nm to investigate the effect on the color response. For all cases, D_m = 250 nm and D_i = 0 nm are kept constant (Figures 3J–L). The resulting reflectance spectra have all obtained approximately the same shape, but the peak wavelengths of the four cases, 455, 496, 536, and 577 nm, respectively, are shifted ∼40 nm with respect to each other along the wavelength abscissa (Figure 3K). The transmittance spectra are virtually identical again (Figure 3L). We can conclude from the various parameter values that the cortex thickness is the most sensitive parameter that determines the barbule’s reflectance spectrum and thereby its color.

We measured reflectance spectra in single cells of various colored feather barbules of S. vulgaris using a microspectrophotometer (MSP). Figures 4A,B show MSP spectra from barbule cells of a green and purple feather, respectively. Although the shape of the different spectra is similar, they are shifted with respect to each other by several tens of nm, and also the amplitude varies quite noticeably. The latter may be a consequence of the measurement procedure, because the objective of the microspectrophotometer projects a light beam with a considerable aperture onto the barbule. Since the barbule and its multilayer reflector is not perfectly flat, part of the reflected light will not be captured by the objective. In other words, even when the local cortex thickness and melanosome layer remain constant and the spectral shape thus stays the same, the amplitude of the measured reflectance will vary with the shape of the barbule surface.

In order to estimate the local cortex thicknesses, we compared two examples of measured spectra with spectra calculated with the above model. To obtain a close fit, we adjusted the parameters. Taking again D_m = 250 nm, satisfactory fits for the case of a blue-green and a purple barbule were obtained with c = 300 nm and c = 390 nm, respectively, (Figure 4C).

Whereas the reflectance spectra measured locally from within a single barbule cell area vary (Figures 4A,B), the coloration observed visually will be the sum of local reflection spectra. This can be illustrated by measuring the reflectance with a bifurcated probe, which integrates the reflections from an area on the order of 1 mm². The different colored feather areas then yield clearly distinct reflectance spectra (Figure 4D).

Compared to the European starling, the Cape starling (L. nitens) has a much more impressive coloration, with striking metallic colors throughout its entire iridescent plumage (Figure 5A). The head, body and wing feathers are colored variously blue-green (Figure 5B), while the tail is uniformly blue (Figure 5C). Micrographs show that individual barbule cells are again rather uniformly colored, but the color of adjacent cells can differ substantially (Figures 5D,E).

The Cape starling has hollow melanosomes (Craig and Hartley, 1985; Maia et al., 2013). A transmission electron micrograph of a barbule cross-section of a related glossy starling, the Superb starling Lamprotornis superbus, shows a single layer of hollow melanosomes, which line the entire barbule perimeter like the melanosomes of S. vulgaris barbules; in the barbule interior some randomly arranged hollow melanosomes are seen [Figure 6A; from Durrer and Villiger (1970)]. Scanning electron micrographs of cross-sections of the investigated L. nitens barbules show essentially the same organization (Figure 6B) and gives a cortex thickness of about 118 ± 25 and 185 ± 31 nm for the blue and the green barbule, respectively.

Durrer and Villiger described the melanosomes of four Lamprotornis species as platelet-shaped, with a compact melanin sheet that envelopes a honeycomb-like air-filled space (Durrer and Villiger, 1970). Epi-illumination microscopy on feather barbules of L. nitens immersed in oil visualized the melanosomes (Figure 6C). The resulting images differed remarkably from those of the European starling in that they show a distribution of different melanosome shapes with a majority of doughnut-like shapes intermingled with small bars. These different shapes were also visible in scanning electron micrographs of the surface of an L. nitens feather barbule (Figure 6D).

The immersion images suggest that the melanosomes are less uniform in size and shape than the melanosomes of S. vulgaris. The images show the appearance of bars and doughnut-like shapes in Figures 6C,D, which might either suggest a mixture of melanosome shapes ranging between rodlets and platelets (Figure 6E), or a disordered arrangement of platelets alone. An extensive comparative study on the melanosomes of African starlings demonstrated that the melanin granules of Lamprotornis may be short and thick as well as slender and elongated (Craig and Hartley, 1985). Evolutionary transitions between rodlets and platelets have been recently suggested for starlings (Maia et al., 2013; Rubenstein et al., 2021). Furthermore, the interior of the melanosomes is presumably not fully hollow as TEM studies showed the existence of minor cross-linking elements (Figure 6E).

Whatever the fine structure, the single layer of melanosomes together with the keratin cortex will create a multilayer reflector. We investigated this by mounting a blue feather piece in the imaging scatterometer. Narrow beam illumination of a small barbule area yielded a quite discrete spot, demonstrating that locally the barbule indeed approximates an ideal reflector (Figure 6F). Similar as with the European starling, we often obtained scatterograms with more distributed patterns, indicating a varying flatness of the barbules’ multilayers.

For the modeling of the reflectance spectra, we used anatomical data of the feathers of the Lesser blue-eared glossy-starling, L. chloropterus, as it is closely related to L. nitens and similarly colored (Durrer and Villiger, 1970; Craig and Hartley, 1985; Maia et al., 2013) in addition to the EM data of the investigated cross-sections. Figure 7A presents a diagram of a barbule with cortex thickness c and cylindrical, hollow melanosomes, membrane thickness d_m, air cavity thickness d_a, and width W_m, separated from each other by a distance W_i. For the modeling we used d_m = 70 nm and d_a = 85 nm (Figure 6; Durrer and Villiger, 1970). We first investigated the effect on the reflectance of an incomplete filling of the melanosome layer. For our initial calculations we took for the cortex thickness c = 160 nm, and we assumed, based on the electron micrographs, that the honeycomb of thin melanin membranes within the air cavities take up 10% of the volume, so that the inside air fraction is 0.9. The keratin filling factor of the melanosome layer g = W_i/W_m has the value g = 0 when the melanosomes touch each other. With an increase of the filling factor to g = 0.1 and g = 0.2, the refractive index contrast of the melanosome layer decreases, as shown in the refractive index profiles (Figure 7B). The calculated reflectance spectra show that this causes a decrease of the reflectance amplitude, but the peak wavelength remains virtually the same (Figure 7C).

We subsequently considered a change in the air cavity thickness from 75 to 95 nm (well sampling the range reported for various Lamprotornis species of 73 to 98 nm; Durrer and Villiger, 1970). This causes slight changes in both the amplitude and peak wavelength of the reflectance spectra (λ_max), but their shape remains about the same (Figure 7D). For d_a = 75, 85, 95 nm, λ_max = 472, 479, and 486 nm, respectively, i.e., a 10 nm increase of the air cavity thickness causes a 7 nm peak wavelength shift (Figure 7D).

To underscore the effective medium results, we performed FDTD calculations for the case d_m = 70 nm, d_a = 85 nm, and c = 160 nm, for both TE- and TM-polarized light (Supplementary Figure 2). The calculated TE- and TM-reflectance spectra slightly differ in amplitude, demonstrating the effect of anisotropy of the melanosomes. Nevertheless, the reflectance spectra obtained for different filling factors with effective-medium-multilayer modeling and FDTD were virtually identical (Figure 7C and Supplementary Figure 2).

Changes in the cortex thickness again have a substantial effect on the reflectance spectra (Figure 7E). For c = 140, 160 and 180 nm, the peak wavelength is λ_max = 449, 480, and 511 nm, respectively, i.e., an increase of the cortex thickness by 20 nm causes a 31 nm peak wavelength shift. Figure 7F shows the related transmittance spectra, where the melanin absorption spectrum is still well recognizable, but the transmittance in the blue-green wavelength range is reduced due to the higher reflectance in this band.

We measured reflectance spectra of the variously colored feather barbules of L. nitens with a microspectrophotometer (MSP) on local areas of single barbule cells. Figures 8A,B show MSP spectra from single barbule cells of a blue and green feather, respectively. As in the case of the European starling, the shape of the various spectra is similar, but their spectral location somewhat differs, and even more their amplitudes. This will again be due to the dependence on the local barbule shape as well as the multilayer dimensions.

We compared two of the measured spectra with spectra following from optical modeling. Satisfactory fits were obtained for the tail feather barbules using a cortex thickness c = 135 nm and for the covert feather barbules c = 188 nm (Figure 8C). Nearly identical fits to the different color spectra of L. nitens could also be obtained by changing the melanosome air cavity thickness and/or the melanin membrane thickness, but the necessary parameter variation was larger for the air cavity thickness and the melanin membrane thickness than the 53 nm change that was necessary in cortex thickness variation. This would be beyond the range reported in the literature. We furthermore measured reflectance spectra of the tail and covert feathers of L. nitens with a bifurcated reflection probe (Figure 8D). The spectra are representative for the observed colors and closely resemble the averaged reflectance spectra of Figures 8A,B.