Chemical characterization of pterosaur melanin challenges color inferences in extinct animals

Melanosomes (melanin-bearing organelles) are common in the fossil record occurring as dense packs of globular microbodies. The organic component comprising the melanosome, melanin, is often preserved in fossils, allowing identification of the chemical nature of the constituent pigment. In present-day vertebrates, melanosome morphology correlates with their pigment content in selected melanin-containing structures, and this interdependency is employed in the color reconstruction of extinct animals. The lack of analyses integrating the morphology of fossil melanosomes with the chemical identification of pigments, however, makes these inferences tentative. Here, we chemically characterize the melanin content of the soft tissue headcrest of the pterosaur Tupandactylus imperator by alkaline hydrogen peroxide oxidation followed by high-performance liquid chromatography. Our results demonstrate the unequivocal presence of eumelanin in T. imperator headcrest. Scanning electron microscopy followed by statistical analyses, however, reveal that preserved melanosomes containing eumelanin are undistinguishable to pheomelanin-bearing organelles of extant vertebrates. Based on these new findings, straightforward color inferences based on melanosome morphology may not be valid for all fossil vertebrates, and color reconstructions based on ultrastructure alone should be regarded with caution.

The fitting of CPCA 3590 spectra (Fig. 2D) yielded several bands but the most diagnostic ones occur centered at about 1328 cm −1 and 1575 cm −1 (Table S2). Despite the former is slightly shifted to the left, both bands are similar to those of synthetic and Sepia officinalis melanins (Sigma-Aldrich M2649) from our experiments ( Fig. 2A) and reported in the literature. For synthetic melanin (Sigma-Aldrich M8631), the two most intense peaks occur at 1387 cm −1 and 1580 cm −1 , whereas for the Sepia-derived melanin, they are centered at about 1389 cm −1 and 1578 cm −1 . Other compounds present in the sample, such as carbonates and phosphates, may be influencing the spectra and affecting the melanin peak intensities (Fig. 2E,F). Furthermore, the broad bandwidth is reflective of the heterogeneity/disorder of eumelanin structure 24 , which may have incorporated metals, especially Ca and Mn, among its oligomer sheets 32 . Moreover, the slight shift in the Raman peaks may also indicate that T. imperator eumelanin went through a substantial change, possibly related to loss of functional groups, or may also be derived from the C-N stretching from the indole ring 33 . Regardless of these features, both bands can be confidently assigned to the stretching and plane vibrations of C-C, C-OH, C-N, C-O from pyrrole and indole rings 25,27-29,31,33  3 2− (at 1086 cm −1 ) and PO 4 3− (at 965 cm −1 ), which is consistent with a bioapatite variety. The second spectrum is that of the headcrest, and is similar to the two shown below, those of synthetic and Sepia melanins, respectively. (B) Raman spectra from dark bands of the headcrest exhibiting a signal variation between the two regions of measurement (red circles indicate the two points where four measurements were made) (C). (Supplementary Information). The less intense peaks in CPCA 3590 may result from the trace amounts of melanin. In contrast to carbon-rich compounds that exhibit overtone scattering bands (second-order peaks) above 2400 cm −1 34 , these peaks are absent, supporting the argument that they are derived from eumelanin 28,29 . The previous interpretation of phosphatization of the headcrest and microbodies 35 is also supported by RS.
Identification of eumelanin by chemical degradation and high-performance liquid chromatography (HPLC). The first modern attempt to chemically characterize melanins in the fossil record used synchrotron X-ray to identify trace metals alleged to be unique markers of the presence of eumelanin 36 . Since then, several attempts have been made using this methodology to evaluate different types of fossils 37 . However, this method did not survive scrutiny, as several taphonomic processes are able to concentrate metals and originate similar patterns under synchrotron light sources 13,37 . More recently, chemical fingerprints identified by time-of-flight secondary ion mass spectrometry (ToF-SIMS) was successfully employed to characterize both pheo-and eumelanins in fossils 4,10,15,16,[38][39][40] . Direct and conclusive chemical evidence, however, requires carrying out chemical degradation of the organic matter by alkaline hydrogen peroxide oxidation, which, if melanin is present, generates specific and unique chemical markers 3,41,42 .
In this study, samples of Tupandactylus imperator headcrest (CPCA 3590) were oxidized by alkaline hydrogen peroxide after demineralization, in accordance to standard protocols 3,41,42 (Materials and Methods; Supplementary Information). HPLC analysis of the oxidation products yielded melanin markers PTCA, PDCA, and PTeCA (Figs 3, S4, Table S3). Although their levels were trace, they were significantly higher than those in the adjacent sediment, and their identification was also confirmed by liquid chromatography-mass spectrometry (LC-MS, See Fig. S5). It is noteworthy that the level of PTeCA is much higher than that of PTCA, with a PTeCA/ PTCA ratio being 2.41 ± 0.16, which is characteristic of highly cross-linked eumelanin 3,5 . Ergo, the results of HPLC (Fig. 3A) indicate that most of the T. imperator eumelanin is derived from the crosslinking of DHICA and DHI fractions (Fig. 3B). Although the contribution of PDCA from DHI units cannot be excluded, because PTCA occur twice as much (Fig. 3C), it is suggested that this moiety is more involved in structural alterations 5,42 .

Discussion
Previous microstructural analyses of CPCA 3590 identified the subspherical microbodies as autolithified bacteria 35 , an assumption mainly based on (i) their comparable size with modern microorganisms; (ii) the lack of the typical organization patterns often seen in melanosomes; (iii) the presence of supposed extracellular polymeric substances and (iv) the putative ongoing cellular division 35 . This interpretation was debated in subsequent publications -questioned by some 43 and favored by others [13][14][15][16] . As pointed out by the authors that favored the bacterial alternative, the physical aspects of these microbodies, such as morphology, distribution, and size, were insufficient to completely eliminate the hypothesis of both endogenous or exogenous bacteria 17,18,[44][45][46] . However, it is widely accepted that chemical analysis can significantly aid in their identification 10,[13][14][15][16]45,47,48 .
Although microbes are rarely preserved, the fossilization of animal soft-tissues usually involves the presence of microorganisms that alter geochemical processes at the microscopic level, inducing the precipitation of several minerals 44,49,50 . Experiments simulating diagenesis in microbes indicate that although their molecular signatures can be slightly altered 51 , their morphology often exhibits significant changes, mostly in the form of body deflation or partial degradation 52 . These features are absent in our sample, in which microbodies are predominantly solid particles. Furthermore, T. imperator microbodies also show several characteristics that are consistent with being melanosomes, such as absence of morphotype diversity, no evidence of binary fission and lack of bacterial by-products (such as honeycomb-like structures) 53 , distinct chemistry differences in chemical composition between former soft tissue and matrix 54 , and limited microbody distribution (Supplementary Information).
Energy dispersive spectroscopy (EDS) data show that T. imperator microbodies are Ca-and P-rich, suggesting that they are composed of calcium phosphate 35 . In addition, SR-μXRF indicates the presence of Ca, Cu, Fe, Mn and Zn (Fig. S3). Since phosphatization is the common type of bacterial preservation 50 , these microbodies could indeed represent phosphatized microorganisms. However, the chemical signatures revealed in our study show that the CPCA 3590 headcrest contains eumelanin. Therefore, the combination of morphological and chemical analyses confirms an unequivocal identification of the microbodies as melanosomes.
Since the seminal work of Vinther et al. 55 , inferences about the color patterns of fossil animals rely mainly on melanosome morphology 13 , in spite of other studies that indicate the lack of correlation between melanosome shape and their melanin content 56,57 . While the connection between shape and color is unresolved [13][14][15]58 , it is commonly invoked that high-aspect-ratio ("sausage-like") melanosomes contain eumelanins (black to dark brown in color), whereas globular, low-aspect-ratio melanosomes normally reflect the presence of pheomelanin (rufous red to pale yellow). Moreover, statistical analyses testing the correlation between melanosome morphology and color of extant birds demonstrated a high (up to 82% accuracy) predictive potential for animals in which hues are mainly determined by melanins 13,59,60 .
A recent contribution 59 , however, demonstrated that a similar predictive model cannot be extrapolated to lepidosaurs, turtles, and crocodiles, whereas it is reasonably accurate for bird feathers and mammalian hair. As such, these latter animals would present a high diversity of melanosome morphologies and usually a clear correlation between different morphotypes, the type of melanin they contain and, as a consequence, expressed color 59 . The transition between the primitive low melanosome diversity displayed by lizards, turtles and crocodiles and the pattern displayed by present-day mammals and birds would have been driven by a distinct physiological shift 59 . Alternatively, this change in pattern would be a consequence of the loss of the chromatophore complex, responsible for the color diversity of amniotes showing the primitive condition 13 . The chromatophore system might be superfluous for animals in which color patterns are expressed in well-developed integumentary structures, such as feathers and fur 13 .
It would, however, be expected that pterosaurs did not depend on chromatophores to express color patterns, as these archosaurs were also covered by a dense layer of supposedly keratinous filamentous structures that were potentially homologous to feathers 61 . The analyses of Li et al. 59 included two pterosaur specimens, in which the microstructure of the hair-like coverage showed a low diversity of low-aspect-ratio melanosomes, more consistent with what is observed in lepidosaur, turtle and crocodile skin than to feathers or mammal fur. A similar pattern is also displayed in Tupandactylus imperator headcrest (based on CPCA 3590). Morphologically, the vast majority of the microbodies revealed by SEM images would be identified as pheomelanin-like melanosomes. In spite of that, the chemical degradation performed yielded the specific markers of eumelanin (i.e. PTCA, PDCA, and PTeCA) in concentrations compatible with highly cross-linked eumelanin 3,5,42 , and the absence of the specific markers for pheomelanin. Thus, these results imply that a clear distinction between high-aspect-ratio eumelanosomes and spherical phaeomelanosomes is not valid for pterosaurs, and by extrapolation, for amniotes that share the primitive condition of low melanosome diversity. Indeed, CPCA 3590 organelles are remarkably similar to internal eumelanosomes 58 from basal-most vertebrates, such as the amphibians, cyclostomes 38,39 and cuttlefish 3 . Consequently, any color inference in animals presenting the plesiomorphic condition based on melanosome morphology would be equivocal, as ellipsoidal, low-aspect-ratio bodies can contain both pheo-and eumelanins. We should also stress that the circumstances surrounding the physiological shift proposed by Li et al. 59 are still obscure, and it would be precipitate to imply that animals such as non-avian dinosaurs shared with birds a straightforward correlation between melanosome morphology and their pigment content.
Most chemical surveys of fossil pigments have thus far identified eumelanosomes and eumelanin fingerprints. The reason for the low occurrence of pheomelanin or pheomelanosomes is still unknown; it remains possible that pheomelanin preservations may not be as robust as that of eumelanin 62 . Although other classes of biochromes (e.g. carotenoids and porphyrins) are relatively common in sedimentary deposits, these compounds are extremely frail and prone to chemical alterations 1,63,64 . For instance, following deposition, porphyrins, and carotenoids readily experience several chemical reactions such as oxidation and polymerization, transforming them into long chains of hydrocarbons 1,65,66 . Because melanin is a highly conserved polymer 5,10 , and eumelanins are the most common class of melanins in nature, it is expected that this pigment is present in the majority of exceptionally preserved fossils 12,13 . (2019) 9:15947 | https://doi.org/10.1038/s41598-019-52318-y www.nature.com/scientificreports www.nature.com/scientificreports/ Studies that identified pheomelanin have rarely found corresponding microbodies preserved in three dimensions. As a consequence, phaeomelanic colorations were often based on the recognition of melanosome external molds 67 or chemical fingerprints 40 . Despite the latter approach being a more reliable way to identify pheomelanin, the former possesses serious issues to color inferences. Our results support this claim, as spherical and subspherical microbodies can potentially bear one or both types of melanin pigments or be composed mainly by one type of moiety. This may be true to some dinosaurs, such as Sinosauropteryx 67 , Anchiornis 68 , Yi qi 69 , and Psittacosaurus 70 , whose color inferences were based solely on the morphology of molds, with no further chemical and/or statistical support.
The extremely selective nature of fossilization has the effect of building a virtually insurmountable barrier between post-diagenetic remains of organisms and living beings. The recent recognition of the persistence of melanins and melanin-containing organelles in the fossil record 3,42 allowed reconstructions of color patterns of extinct animals. However, many paleocolor studies relied basically on the microbody morphology, raising questions about the validity of their outcomes. Correspondingly, our results strongly support these disputes. Since melanins are directly involved in complex social and ecological behaviors, such as camouflage, intraspecific recognition, and sexual display, their correct characterization can sum to the understanding of the biology of extinct animals 13 , and color reconstruction cannot rely solely on microstructural analysis 13,71 .

Materials and Methods
Specimen CPCA 3590 is preserved in a grayish-color laminated limestone typical of Crato Formation beds 35 (Supplementary Information). This fossil is comprised of a fairly complete skull with headcrest's soft tissues, which allows it to be assigned to the tapejarid species Tupandactylus imperator (for taxonomic details, see Pinheiro et al. 70 ). This specimen is permanently housed in the Paleontological collection of the Centro de Pesquisas Paleontólógicas da Chapada do Araripe (CPCA, Crato, Ceará, Brazil). The headcrest's soft tissue was examined using a scanning electron microscope (SEM). Elemental mapping was carried out using synchrotron radiation-micro X-ray fluorescence (SR-µXRF). The molecular content was examined using Raman spectroscopy (RS) and high-performance liquid chromatography (HPLC). The latter technique was performed to quantitate melanin degradation products, PTCA, PDCA, and PTeCA after treatment by alkaline hydrogen peroxide oxidation of demineralized samples of CPCA 3590 5,41 . To confirm the identification of PTCA and PTeCA, LC-MS of extracts of oxidation products was performed according to previously described methods (see Glass et al. 3). Following the image acquisition using SEM, melanosomes and minerals were measured using ImageJ 72 , and statistical analysis was performed using Past 3.06 73 . SR-µXRF mapping was processed using PyMCA 5.1.1 software and Raman spectra were processed using Renishaw Wire 4.1 and Wire 4.4, and Origin 9.6.0.172. Analyses were performed at the Duke University Chemistry Department Mass Spectrometry Facility, Brazilian Synchrotron Light Laboratory (LNLS) and Institute of Chemistry of the University of São Paulo (IQ-USP). See SOM2 for details on material and methods.

Data availability
No datasets were generated or analyzed during the current study.