Elaborate plumage patterning in a Cretaceous bird

Integumentary patterns and colors can differentiate species, sexes, and life changes and can inform on habitat and ecology. However, they are rarely preserved in the fossil record. Here, we report on an extremely well-preserved specimen of the Cretaceous bird Confuciusornis with unprecedented complexity, including small spots on the wings, crest, and throat. Morphological and chemical evidence suggest that these patterns are produced by melanin, but unusual preservation prevents assignment of specific colors. Based on comparisons with extant birds, these patterns were likely used for camouflage, although other functions including sexual signaling cannot be ruled out. Our data show that even more elaborate plumage patterns than the spangles in Anchiornis and stripes in Sinosauropteryx were present at a relatively early stage of avian evolution, showing the significance of coloration and patterning to feather evolution.


Taxonomic status of CUGB 1401
CUGB P1401 (China University of Geosciences, Beijing) is differentiated from Eoconfuciusornis zhengi by possessing a much larger deltopectoral crest with a foramen at its centre and a posterior sternal margin with diminutive trabeculae and twin concavities (incisurae; Fig. S1) rather than with a straight to convex margin (Zhang et al., 2008). A large deltopectoral crest (Fig. S1) is seen in confuciusornithid specimens referred to Confuciusornis, Changchengornis and Jinzhouornis Zhang et al., 2008).
The new specimen is differentiated from Changchengornis hengdaoziensis by the presence of a large round foramen perforating this large deltopectoral crest ( Fig. S1; . It shares this perforated crest with all named species of Confuciusornis and Jinzhouornis. The two named species of Jinzhouornis (Hou, 2002) have alternatively been proposed to be junior synonyms of Confuciusornis sanctus (Chiappe et al., 2007(Chiappe et al., , 2008Marugán-Lobón et al., 2011) or diagnosably distinct (Li et al., 2010a). One concern with the taxonomic status of Jinzhouornis is that all characters proposed to be diagnostic for this genus are also present in Confuciusornis with the exception of an elongate rostrum. The two elements of Jinzhouornis yixianensis that could be reliably measured were nearly identical in proportions to other Confuciusornis specimens. Furthermore, there are abundant specimens of Confuciusornis sanctus with rostri of the proportions of Jinzhouornis zhangjiyinensis and Jinzhouornis yixianensis (e.g., Chiappe et al., 1999, figs. 8, 15;GMV 2130), which may support previous recommendations of junior synonym status. Furthermore, the skull of the holotype of Jinzhousornis zhangjiyingensis does not appear confidently associated with the rest of the skeleton as the cervical series is conspicuously truncated by a large crack separating it from the skull.
The new specimen is differentiated from Confuciusornis feduccai and Confuciusornis dui but similar to Confuciusornis sanctus in the presence of relatively deep rather than shallow incisures on the posterior sternal margin; it also differs from these taxa in size (Fig. S1; Table S1; Zhang et al., 2009). The holotype of C. chuongzhous consists of only a partial hind limb. It and that of C. suniae are identical in relative proportions and morphologies to C. sanctus (Table S1; Chiappe et al., 1999Chiappe et al., , 2008Marugán-Lobón et al. 2011). Confuciusornis jianchangensis is slightly smaller than CUGB P1401 based on published measurements (Li et al., 2010a). While its taxonomic status has not been revisited, it suffers the same issues identified for previously proposed species of the Confuciusornithidae (Chiappe et al., 2008;Marugán-Lobón et al., 2011). Table S1. Measurements of the new specimen (CUGB 1401) compared to holotype and referred confuciusornithid specimens. Direct measurements for 8 specimens were added to the supplementary data table from Chiappe et al. (2008). Data on the holotype specimens of Confuciusornis feducciai and Confuciusornis dui are from Zhang et al. (2009), and those for Changchengornis hengdaoziensis, from Ji et al. (1999). Values are sorted by femur length, followed by humerus and ulna length.   Preserved feathering in specimens referred to Confuciusornis sanctus illustrating plumage differences from CUGB 1401 including the absence of an elongate patterned crest: A) Chanchengornis holotype specimen as illustrated in 2) IVPP 11374, 11375 referred to Confuciusornis sanctus, C) Eoconfuciusornis holotype specimen 1, D) the one other specimen referred to Confuciusornis showing secondary coverts with spangles as illustrated in Zheng (2009).

Melanosome sampling and assessment of morphology
We collected data on melanosome morphology from the fossil as previously described (Li et al., 2010b), with the exception that we took two (rather than one) samples from each location. Each of the samples was randomly assigned to either morphological (SEM) or chemical (Raman) analysis, such that we obtained both types of data from the same location on the fossil (see Fig. S3; Table S3 for details). The fossil was not exposed to glues or other chemicals prior to analysis.
Morphological samples were sputter-coated with silver and viewed on a ZEISS SUPRA-55 VP field emission SEM at China University of Geosciences, Beijing. Melanosomes were preserved in both three dimensions and as moldic impressions. For consistency, and to avoid potential differences between the two preservational styles (Clarke et al., 2010), we only measured impressions. We measured length and diameter of fossil melanosomes from these images using ImageJ and used these measurements to calculate morphological variables (mean, coefficient of variation, skewness of length, diameter and aspect ratio) as before. We visually examined images ( Fig. S4) and plotted mean melanosome measurements against those of extant samples ( Fig. S6; Li et al., 2012).

Melanosome morphology
Six samples (o, r, w, a1, e1, k1) were outside the range of length, diameter or both for extant melanosome samples (Fig. S6). All of these samples were characterized by large, round morphologies and (in some cases) unusual honeycomb-like arrangements (e.g., sample k1, Fig. S4). In some cases (e.g., sample o, Fig. S4), these morphologies were colocalized with more rod-like morphologies that were of the typical size for melanosomes. Although both morphology and distribution in the remainder of the samples were in the range of modern melanosomes, these unusual features suggest that this fossil may have been preserved in a way that distorted melanosome features or introduced artifacts. We therefore did not use morphology to reconstruct colour of this specimen. These results support the need for caution in colour reconstruction (McNamara et al., 2013), but do not invalidate prior or future work using more conventionally preserved specimens.

Raman analyses
We compared Raman spectra from the fossil and matrix samples to extant melanin extracted from modern bird feathers, the common keratinolytic bacterium Bacillus licheniformis, carbon controls, including carbon black and fossil plants (Peteya et al., 2017), as well as bituminous coal and graphite as additional carbon controls and squid (Sepia) melanin as an additional melanin control. Raman spectra of all but one extant melanin sample were highly similar to one another and to previous spectra of eumelanin (Figs. 3, S3-4;Galván et al., 2013). They were characterized by a large peak at ~1578 cm/-1 and a smaller one at ~1360 cm/-1. A third, shorter peak was present in avian melanin samples at ~1178 cm/-1 but at ~988 cm/-1 in Sepia. Spectra were characterized by numerous sharp peaks for keratin and by weak background peaks for the bacteria and matrix, and in all cases were dissimilar to those of extant melanin (Figs. 3, S3-4; Table  S2-3). Peaks in the Raman spectra from the fossil samples closely matched those of extant eumelanin, but not those of keratin, bacteria or matrix (Figs. 3, S3-4; Table S2-3). Peaks for carbon black, bituminous coal, and graphite also differed, both in the location of the second main peak (~1590 cm/-1) and peak morphology (Figs. 3, S3-4; Table S3). However, fossil plants that likely lacked eumelanin have been shown to have similar Raman spectra to eumelanin, so Raman spectroscopy remains an inconclusive technique without the support of other chemical or morphological techniques (Peteya et al., 2017).
Raman curves from pheomelanic feathers of Rhode Island Red rooster Gallus gallus had numerous weak peaks, and none matched those of the other samples (Fig. S5). Previous studies showed distinctive Raman patterns for synthetic pheomelanin (Galván et al., 2013), but whether they are also characteristic of natural pheomelanin is only weakly supported. Indeed, the chemical distance of synthetic from natural eumelanin (Liu et al., 2014, Xiao et al. 2018 suggests that results from synthetic melanin should be interpreted with caution. Additionally, while Raman spectra from natural pheomelanin sources have been reported (Galván and Jorge, 2015;Galván et al., 2017), these signals are weak and difficult to characterize. We have tested pheomelanized samples extracted from Rhode Island Red rooster feathers and orange zebrafinch cheek feathers (in which pheomelanin composes over 99% of the total melanin (McGraw and Wakamatsu, 2004)) using both a green 532 nm laser ( Figure S5) and an IR 785 nm laser, none of which gave distinctive pheomelanin peaks. In any case, our Raman spectral patterns were consistent across black, brown, and iridescent samples ( Figure S5), including some that are known to contain a percentage of pheomelanin (Liu et al., 2014). The broad peaks of eumelanin may mask pheomelanin peaks in these samples, although the lack of distinctive pheomelanin peaks in feathers with greater concentrations of pheomelanin makes this unlikely. Moreover, the inability to detect eumelanin/pheomelanin mixtures using Raman spectroscopy suggests that other techniques (e.g. ToF-SIMS (Lindgren et al., 2014) or VUV-LDMS (Liu et al., 2014)) should be used in conjunction with Raman for distinguishing colours based on melanin chemistry. Indeed, the latter technique discriminates black from brown colours reasonably well. While chemistry from iridescent feathers is highly variable (Liu et al., 2014) their morphology is consistent (Li et al., 2012), suggesting that a combination of both morphological and chemical data may be a promising avenue in the further reconstruction of fossil colour.

Time-of-flight secondary ion mass spectrometry1
ToF-SIMS spectra showed peaks at the theoretical masses for eumelanin for all samples tested, including the extant samples, the CUGB P1404 fossil samples, and the matrix sample. Our PCA of the standardized relative intensities of all peaks did not separate the matrix sample from the fossil samples, although the wild turkey melanin sample also plotted with the CUGB P1404 samples ( Figure S8). We were therefore unable to corroborate the presence of preserved eumelanin using PCA. Additionally, Colleary et al. (2015) showed that standardized intensities of fossil melanin ToF-SIMS spectra do not plot with modern eumelanin intensities and that experimentally matured samples bridge the gap between modern and fossil melanin samples. We could not include matured samples in our analysis, so a gap between modern and fossil samples would be unsurprising if eumelanin is preserved. None of the ToF-SIMS spectra resemble previously reported pyomelanin -a microbial melanin -spectra (Lindgren et al., 2015a).

Matrix-assisted laser desorption/ionization mass spectrometry
Modern melanin samples, including the red-winged blackbird and Sepia melanin, presented a single peak at a mass of approximately 659.3 m/z (Fig. S9), which represents the combined masses of a melanin dimer and a sodium ion from the extraction buffer. This peak was also present in the fossil test sample, but not the carbon controls, the fossil plant samples, nor any of the CUGB P1401 samples (Figs. S10, S11). Some of the CUGB P1401 samples (b1, d1, e1, y) had a peak at a mass of approximately 357.1 m/z that could be a potassinated melanin monomer, but this peak was not present in either of the modern melanin samples. The 357 m/z peak is more likely an impurity. Additionally, MALDI spectra from the CUGB P1401 samples did not match any of the carbon controls. The chemical results for the Confuciusornis samples are therefore inconclusive. All other peaks present in the fossil samples and modern melanin samples can be attributed to salt cluster ions or other buffer-related contaminants.
Although it did not find melanin in the Confuciusornis samples, MALDI may be useful for future studies in fossil melanin, given that our test fossil sample did yield the 659.3 m/z eumelanin peak. However, this technique requires grinding the fossil samples, which are usually composed of a tiny bit of surface fossil material with underlying minerals, with a matrix, so larger fossil samples may be necessary to separate a melanin signal from the rock and matrix signals.1  . Raman spectra of eumelanin extracted from black chicken, black red-winged blackbird, brown Cooper's hawk, brown house wren, iridescent mallard, and iridescent wild turkey feathers, a Sepia eumelanin standard, melanin extracted from a Rhode Island red rooster feather composed of a high concentration of pheomelanin, the bacterium Bacillus licheniformis grown on a white (pigmentless) feather substrate, and keratin from a white sulphur-crested cockatoo feather.

Meleagris gallopavo
Bacillus licheniformis Keratin Fig. S6. Diameter vs length for extant avian melanosomes and as measured from CUGB 1401. Colour of triangles (extant samples) corresponds to feather colour. Black, grey, red-brown, blue, and purple-colored triangles represent black, grey, red-brown, penguintype, and iridescent feathers, respectively. Sample codes correspond to Fig. S1.      Non--original matrix N absent Rock2 Original matrix N absent Rock3 Original matrix N absent   Table S4. Peak fitting table for Raman spectra of feather keratin, the bacterium Bacillus licheniformis, and carbon controls. All peak locations were previously reported (Peteya et al., 2017), except those for bituminous coal and graphite.