Non-integumentary melanosomes can bias reconstructions of the colours of fossil vertebrates

The soft tissues of many fossil vertebrates preserve evidence of melanosomes—micron-scale organelles that inform on integumentary coloration and communication strategies. In extant vertebrates, however, melanosomes also occur in internal tissues. Hence, fossil melanosomes may not derive solely from the integument and its appendages. Here, by analyzing extant and fossil frogs, we show that non-integumentary melanosomes have high fossilization potential, vastly outnumber those from the skin, and potentially dominate the melanosome films preserved in some fossil vertebrates. Our decay experiments show that non-integumentary melanosomes usually remain in situ provided that carcasses are undisturbed. Micron-scale study of fossils, however, demonstrates that non-integumentary melanosomes can redistribute through parts of the body if carcasses are disturbed by currents. Collectively, these data indicate that fossil melanosomes do not always relate to integumentary coloration. Integumentary and non-integumentary melanosomes can be discriminated using melanosome geometry and distribution. This is essential to accurate reconstructions of the integumentary colours of fossil vertebrates.

M elanin is a key component of visual signals in animals through its incorporation into integumentary patterning 1 . In vertebrates, melanin occurs in skin and its derivatives as discrete micron-sized membrane-bound organelles termed melanosomes 2 . Physical and chemical evidence of melanin has been used to infer the plumage colours of fossil birds and feathered dinosaurs (reviewed in ref. 3 ) and integumentary coloration in fossil marine reptiles 4 (but see refs. [5][6][7]. Reconstructions of original skin colour rely on evidence that fossil microstructures are melanosomes (and not, for example, bacteria 5 ) and that fossil melanosomes derive solely from the integument. Melanosomes occur, however, in various internal organs in extant vertebrates [8][9][10] and thus melanosomes preserved in the body outline of fossils will not always indicate integumentary colour. Given the high recalcitrance of melanin and melanosomes 11 , non-integumentary melanosomes (defined here as those from internal organs and tissues and excluding the eyes) may persist during fossilization, but this remains to be demonstrated; the decay microenvironment within carcasses and the external environment often differ markedly in chemistry 12,13 . If nonintegumentary melanosomes are decay-resistant and abundant in vivo, reconstruction of integumentary coloration based on the presence/absence, shape and distribution of fossil melanosomes requires that melanosomes from different tissue sources can be discriminated. Recent studies of colour reconstruction 14,15 have opted to avoid sampling the torso, in attempts to avoid sampling non-integumentary melanosomes. This does not, however, guarantee that melanosomes from the limbs and tail are integumentary because non-integumentary melanosomes may be redistributed throughout parts of the body during decay. In key biotas such as Jehol, fossils are often partly disarticulated by the action of bottom currents close to the lake floor and can preserve evidence for rupturing of organs and redistribution of their contents throughout the body 16 . Intuitively, such disturbance of exposed carcasses by current activity could also potentially distribute melanosomes from degraded organs inside the body. Whether such redistribution is likely in the absence of physical disturbance (e.g., when organs decompose or the carcass ruptures from gas pressure) is unknown. Resolving these critical issues requires a systematic and comprehensive test of (1) whether nonintegumentary melanosomes are abundant and can survive decay and (2) under what taphonomic scenarios are they likely to redistribute.
Here, we address these issues using histology, fluorescence microscopy, scanning electron microscopy (SEM) and alkaline hydroxide peroxide oxidation (AHPO) 11 of tissues from freshly killed and experimentally decayed extant frogs coupled with data from fossils. Frogs are an ideal test case as melanosomes have been reported in internal organs of extant taxa [8][9][10] and fossil examples are known that preserve melanosomes 13 . Our data show that non-integumentary melanosomes are abundant in some extant vertebrates, have a high fossilization potential, and can redistribute throughout the body if disturbed by bottom currents, and must thus be considered in studies of fossil colour based on preserved melanosomes. Non-integumentary and integumentary melanosomes can, however, be discriminated on the basis of their geometry and spatial distributions.

Results
Abundance of non-integumentary melanosomes. Fontana-Masson histological staining is an established technique for the identification of melanin in histological sections 17 : melanosomes are stained black, and other tissue components, hues of pink to yellow (Fig. 1a-v). Fontana-Masson, however, can also stain argentaffin granules, which can occur in liver tissue and in the epithelial lining of the digestive tract. To discriminate between melanosomes and argentaffin granules, we prepared histological sections of tissues from an albino specimen of Xenopus laevis (albino animals lack melanosomes but can possess argentaffin granules 18 ). The results (Fig. 1g-j) reveal no evidence for argentaffin granules. The regions stained black in the histological sections of the wild-type (melanin-bearing) specimens thus correspond to melanin.
Volumes of non-integumentary melanosomes were calculated as follows. First, the volume of each organ was measured in the laboratory. The volume of melanosomes in each organ was then estimated using histological sections, whereby the area of tissue occupied by melanin in each section was converted to a volume using the known thickness of the histological sections; see Methods). Volumes of non-integumentary melanosomes are 1.7 cm 3 for Xenopus, 0.36 cm 3 for Kaloula and 0.01 cm 3 for Osteopilus (Table 2); these values correspond to 3.9%, 0.6% and 0.1% of body volume, respectively. For all taxa, melanosomes are most abundant in the liver (84-98% of total melanosomes) and least abundant in the spleen (0.0005-1.45% of total melanosomes; Table 2).
Models of integumentary colour in fossils should consider the potential of contributions from non-integumentary melanosomes if the latter are abundant in extant vertebrates. The abundance of non-integumentary melanosomes was assessed as follows. The layer of non-integumentary melanosomes in the Libros frogs has a mean thickness of ca. 25 µm (measured in SEM from polished vertical sections) 12 . The volume of melanosomes required to produce a layer of this thickness over the body outline is 0.126 cm 3 for Xenopus, 0.038 cm 3 for Kaloula and 0.028 cm 3 for Osteopilus (Table 2). Remarkably, Xenopus and Kaloula therefore possess sufficient non-integumentary melanosomes to define a layer over the entire body that is 338 μm thick and 234 μm thick, respectively, i.e., far in excess of the thickness of the layer in the fossils. Osteopilus has sufficient non-integumentary melanosomes to define a 25-μm-thick layer over the torso, but not the torso plus the limbs. In extant frogs, integumentary melanosomes typically form patchy layers <10 μm thick (melanophores (the dominant source of integumentary melanosomes) often occur in clusters) 9 . Thus, in the fossil frogs, and in extant Xenopus and Kaloula, non-integumentary melanosomes vastly outnumber those from the skin and should dominate the preserved melanosomes.
The fate of non-integumentary melanosomes during decay. Scanning electron micrographs of the tissues of the experimentally degraded frogs reveal abundant rod-shaped and spherical to oblate microstructures (Figs. 1aa, af, ag, al and 2a-c). The rodshaped microstructures are interpreted as decay bacteria as no rod-shaped structures were identified in the tissues in untreated specimens. The spherical to oblate microstructures, however, are not definitively coccoid bacteria, as they and melanosomes cannot   be discriminated in scanning electron micrographs. In fluorescence images of the degraded tissues, rod-shaped microstructures stain green ( Fig. 2d-f) and can be identified as bacteria (in stationary growth phase) on the basis of their fluorescence response 22 , rod-shaped morphology, size (1.5-2.5 µm long), spatial association with identical rod-shaped microstructures that stain orange-red (representing bacteria in active growth phase), and location within regions of filamentous material, presumably bacterial extracellular polymeric saccharides. The spherical, micron-sized structures in the fluorescence images comparable to those visible under SEM quench fluorescence and appear dark ( Fig. 2d-f), as with melanosomes in extant animals [23][24][25] . This plus the melanin-rich nature of our tissue samples and the histological evidence for melanosomes in the degraded tissues from the torso (Fig. 1e, o, u) and thigh ( Fig. 1f, p, v), the dark microstructures in the fluorescence images can be most parsimoniously interpreted as melanosomes. Melanin can fluoresce when strongly oxidized using hydrogen peroxide 25,26 , but that procedure was not used in this study.
Our laboratory experiments thus confirm that nonintegumentary melanosomes survive the decay process; as with melanosomes from the skin, they show no visible morphological alteration during decay. Our results also show that melanosomes usually remain dispersed through, but inside, in the torso during early decay. After 12 weeks, non-integumentary melanosomes are abundant in degraded tissue from the torso (Fig. 1e, o, u, aa, af, al). They also occur, albeit rarely, in the thigh region of all three taxa ( Fig. 1f, p, v, aa, ag). Given that our experiments did not involve agitation or other disturbance of carcasses, our data suggest that substantial redistribution of melanosomes in fossils is unlikely where carcasses are deposited prior to extensive decay and not subjected to disturbance while resting on the sediment surface, e.g., by bottom currents. More prolonged pre-depositional decay and/or post-depositional disturbance of carcasses, however, is likely to result in more extensive redistribution of melanosomes inside the body. This hypothesis can be tested using fossil material (see below).
Non-integumentary melanosomes in fossil amphibians. Previous study of the soft tissues of frogs from the Late Miocene Libros biota revealed the presence of carbonaceous microbodies ( Fig. 3g-i) 12 . The microbodies were originally interpreted as fossil bacteria 12 , but reinterpreted as fossil melanosomes based on geochemical data 13 . This interpretation is further supported herein by our AHPO analysis, which confirms the presence of PTCA and PTeCA (a marker for diagenetically altered eumelanin 11 ) in the fossil soft tissues ( Table 1).
The melanosomes occur as layers in association with phosphatized soft tissues interpreted as the mid-dermal Eberth-Katschenko (E-K) layer. This interpretation is discussed in detail in ref. 12 ; in brief, the fossil layer is identical to the E-K layer of extant frogs in terms of its anatomical distribution, thickness, collagen-rich composition, the geometry of its upper and lower surfaces, and the presence of perforations identical in diameter and geometry to the centrifugal pillars of the dermis. Selective phosphatization of this tissue in the fossils reflects its phosphate-rich composition in vivo. This phosphatized tissue layer is both under-and overlain by layers of melanosomes ('internal' and 'external', respectively). These soft tissue layers have different characteristics in hand specimen (Fig. 4): the outer melanosome layer is light brown in colour, 2-5 µm thick, can be patchy or continuous and can preserve patterning (Fig. 4c); the phosphatized mid-dermis is white to cream in colour, 15-30 µm thick, and continuous; and the inner melanosome layer is dark brown in colour, 15-200 µm thick, and continuous. Both the outer melanosome layer and the phosphatized mid-dermal layer are present on dorsal and ventral sides of the specimen.
Our results demonstrate that the melanosomes in different layers are from different tissues. Melanosomes in the outer layer are 0.35-0.5 µm spheres; the inner layer includes sub-layers comprising 0.9-1.1 µm ovoids and ca. 0.5 µm spheres, respectively ( Fig. 3i) 12,13 . The outer melanosome layer derives from the upper dermis and epidermis, and the thicker inner layer, from internal tissues.
Other amphibians from Libros and other localities 27,28 also exhibit discrete layers of melanosomes. Thin patchy layers over the body outline represent integumentary melanosomes; nonintegumentary melanosomes form a thicker layer that is usually confined to the torso (Fig. 3a-f), where they can define the positions of internal organs 12,[27][28][29] . Critically, 93% of the 73 Libros frogs studied possess non-integumentary melanosomes (see Supplementary Data 1) and in 25%, the latter extend from the torso into the limbs 29 (Fig. 4). Thus, it cannot be assumed a priori that melanosomes from the body outlines of fossil vertebrates represent only integumentary melanosomes. A mechanism to discriminate integumentary and non-integumentary melanosomes is required. Tissue-specific melanosome geometries. Our results show that melanosomes vary in geometry among body tissues within individual taxa (Table 3); except for the liver, lung and spleen of Osteopilus (which have similar geometries), these differences are statistically significant (Table 4). Critically, differences in the sizes of melanosomes from the skin and the liver (the primary source of melanosomes) are statistically significant for all three taxa (Table 4). We predict that collapse of soft tissues during decay would generate diagnostic spatial distributions of melanosomes, typically a thick layer of melanosomes sandwiched between thin layers of melanosomes of different geometry; where melanosomes have not redistributed, limbs should show a thin layer of melanosomes of uniform size. Fossil frogs from Libros exhibit the predicted vertical partitioning of melanosomes (Fig. 3g-i): nonintegumentary and integumentary melanosomes differ significantly in geometry (Student's t-test p = 3.42e −13 (length); p = 7.96e −8 (aspect ratio)); non-integumentary melanosomes from adjacent layers differ in geometry (p = 1.67e −17 (length); p = 1.28e −8 (aspect ratio)).

Discussion
This study highlights a potential risk in the use of fossil melanosomes to infer original coloration in ancient animals. Melanosomes preserved within the body outline of a fossil animal may not have imparted integumentary coloration. Contrary to suggestions that melanosomes decay rapidly 5 , our experiments confirm that vertebrate melanosomes routinely survive decay. Analysis of modern frogs shows that non-integumentary melanosomes are highly abundant and most remain in situ during early decay provided carcasses are not disturbed. Many fossils, especially those deposited rapidly after death in a quiescent environment, and especially those buried soon after death, may thus be expected to retain original distributions of non-integumentary melanosomes. Our fossil analysis demonstrates, however, that under certain circumstances, non-integumentary melanosomes can redistribute from the torso into the limbs; many frogs from Libros show evidence of disturbance by weak currents during residence on the lake floor, viz. disarticulation of small distal limb bones and association of fossils with plant fragments 29 . Redistribution should thus be anticipated in contexts where the biostratinomy of the fossils or the sedimentological context indicates current activity.
Even if they remain in situ, melanosomes preserved in the torso will derive from both the internal organs and the skin and thus cannot be assumed to reflect integumentary pigmentation. Compounding this, the plane of splitting typically passes through preserved soft tissues (as evidenced by similar distributions of soft tissues on both part and counterpart of a specimen; Fig. 5a-f), thus exposing non-integumentary melanosomes on the surface of a fossil. Importantly, preservation of non-integumentary melanosomes does not necessarily preclude preservation of original integumentary colour patterning (Fig. 4c), especially where melanosomes have not redistributed and in body regions where redistribution is unlikely. The latter includes feathers that do not overlap the torso, in particular where melanosomes are localized to the barbules and are embedded in an organic matrix, i.e., the degraded remains of the feather keratin. Studies of fossil melanin Our results show that melanosomes from the internal organs and skin in extant frogs can, in some instances, be discriminated based on geometry. Sympathetic to this, the fossils studied show vertical partitioning of melanosomes into size-specific layers. offer the best opportunity to prevent erroneous sampling of nonintegumentary melanosomes. Layering of melanosomes has also been identified in the eyespots of fossil animals and interpreted as a taxonomic signal 30 . A single visually uniform layer of melanosomes (potentially with mixing of melanosomes of different size and geometry) in fossil specimens, however, may be original or result from melanosome dispersal.
Diverse fossil vertebrate taxa preserve visual evidence for nonintegumentary melanosomes and thus our results have broader applications within vertebrates, including taxa that possess feathers or hair. Examples include the conspicuous dark patch in the anterior torso of the ichythyosaur Stenopterygius quadriscissus 31 and dark patches in the torso of the bat Palaeochiropteryx tupadiodon 32 . Numerous examples in fossils from the Jehol biota include the dark patches in the torso of Sinosauropteryx prima 33 , Sinornithosaurus 34 and Epidexipteryx 35 . Preliminary analyses of extant reptiles confirm that melanosomes are abundant in various internal tissues (Fig. 6). Future studies of diverse extant vertebrates will characterize the preservation potential, abundance and geometry of non-integumentary melanosomes, and will test the feasibility of other mechanisms to identify the source of melanosomes, e.g., the chemistry of the melanosomes and their embedding matrix. Discriminating between the melanosomes from different sources is essential to accurate reconstruction of melanin-based pigmentation in fossil vertebrates.

Methods
Decay experiments. The morphological decay of frogs has been studied in detail previously 36 . Our laboratory experiments had two objectives: (1) to assess whether non-integumentary melanosomes survive early decay, and (2) to assess whether these melanosomes redistribute within a carcass during early decay. We conducted a series of pilot experiments using extant frogs to identify the most appropriate conditions for our experiments; the goal was not to attempt to replicate the full complexity of natural settings 37 but to identify conditions in which carcasses would remain intact during decay, allowing the decay of melanosomes within the body cavity to be tracked. Adult specimens of the African-clawed frog, Xenopus laevis (Anura: Pipidae), the Asian-painted frog, Kaloula pulchra (Anura: Microhylidae), and the Cuban tree frog, Osteopilus septentrionalis (Anura: Hylidae) were euthanased via immersion in water with added 3 g/l tricaine methane sulphonate. The first pilot experiment used natural lake water plus an inoculum of lake sediment and was carried out at 25°C. This experiment was unsuccessful as all carcasses bloated rapidly due to high rates of gas production by decay bacteria and ruptured within 96 h, releasing internal tissues (and their melanosomes) into the decay medium. The second pilot experiment used distilled water with no sediment and was carried out at 15°C, but all carcasses ruptured within 1 week. The third pilot experiment involved freezing freshly killed carcasses at −80°C for 24 h in order to inhibit the growth of (and rate of gas production by) decay bacteria and thus prevent bloating; carcasses were then decayed as per the second phase of pilot experiments. Carcasses in this experiment did not rupture. Dissection of specimens after 3 days revealed that all major organs showed a similar fidelity to those in specimens that had not been frozen; there is no evidence that freezing affected the rate of tissue degradation and/or the redistribution of non-integumentary melanosomes.
The final experiment used five specimens of Xenopus (four wild type, one albino) and two specimens each of Kaloula and Osteopilus. Frog carcasses were frozen at −80°C for 48 h, placed in individual glass vessels containing 650 ml of distilled water and degraded at 15± 1°C for 12 weeks in a Memmert IPS260 constant temperature incubator.
Reptiles for scanning electron microscopy (see below) were supplied as carcasses.
Ethics. The authors have complied with all relevant ethical regulations. Euthanasia of frogs was approved by the University of Bristol Animal Ethics Committee via authorization UIN/13/017 and carried out by the University of Bristol Animal Services Unit.
Histology. The abundance of melanin in internal organs was assessed using histological sections. Small (ca. 5 mm 3 ) samples of liver, lung and spleen were dissected from freshly killed frog specimens and the volume of each organ (and that of the entire frog) calculated by immersion in water. Samples of degraded tissue were also removed from within the abdominal body cavity of degraded specimens. Tissue Table 4 Student's t-test analysis of differences in melanosome geometry between each pair of tissue types listed in Table 3 Xenopus melanosome length Xenopus melanosome aspect ratio  Image analysis. In histological sections stained using Fontana-Masson, melanosomes appear black. In order to assess the abundance of melanosomes in each organ, the area of tissue occupied by melanosomes was calculated by analyzing digital images of histological sections using the Threshold Colour function in ImageJ (www.imagej. nih.org). Ten images were analyzed for each tissue type in each taxon. These data were used to estimate the volume of melanosomes in each organ based on the measured organ volume (see above) and thickness of histological sections (9 µm). The mean thickness of the internal melanosome layer in fossil frogs from Libros is 25 μm 12 . The volume of a 25-µm-thick layer corresponding to the torso, and entire body, respectively, was calculated for each extant taxon.
Fluorescence microscopy. Acridine orange staining is an established technique for discrimination of bacterial cells and eukaryotic tissue in histological sections 17,38 .
Scanning electron microscopy (SEM). Samples of fresh and decayed tissue were prepared using standard protocols 40 and examined using a Hitachi S-3500N variable pressure SEM at an accelerating voltage of 15 kV. Long and short axes were measured for melanosomes from the liver, lung, spleen and skin in the extant frogs, and from melanosomes in the two layers in the fossils.
Alkaline hydroxide peroxide oxidation (AHPO). Alkaline hydrogen peroxide oxidation is a unique chemical assay for melanin that produces diagnostic chemical markers (e.g., pyrrole-2,3,5-tricarboxylic acid (PTCA), pyrrole-2,3-dicarboxylic acid (PDCA) and pyrrole-2,3,4,5-tetracarboxylic acid (PTeCA), which are derived from the dihydroxyindole parent subunit of the melanin molecule) that allow identification and quantification of melanin in modern and fossil materials 11,19 . Freeze-dried samples of liver, lung and spleen of X. laevis (9-17 mg) were homogenized in water with a Ten-Broeck homogenizer at a concentration of 10 mg/ml and 200 µl aliquots were dried in a dessicator and subjected to acid hydrolysis with 6 M HCl (0.5 ml) at 110°C for 16 h 40 . This acid hydrolysis removes proteins and small molecules that might interfere with the assay and thus increases specificity of the biomarkers 40 . The resulting insoluble materials including melanin were collected by centrifugation and washed once with water as described 40 . Then the residues were subjected to alkaline hydrogen peroxide oxidation (AHPO) 38 . Aliquots (100 µl) of X. laevis suspensions were also subjected to hydroiodic acid hydrolysis to analyze 4-amino-3-hydroxyphenylalanine (4-AHP), a specific biomarker for phaeomelanin 39 . A sample of soft tissues from a fossil frog from the Late Miocene Libros biota (MNCN 63776) was finely ground with a mortar and pestle and weighed. The powder was subjected to acid hydrolysis and then AHPO as above. In a-f the plane of splitting passes medially through the soft tissues, thus exposing non-integumentary melanosomes at the surface and producing near-identical distributions of soft tissues in part and counterpart. In g, the dissimilar distribution of soft tissues in part and counterpart demonstrates that the plane of splitting is not precisely medial within the soft tissues; non-integumentary melanosomes may still be exposed at the surface. a