Cellular structure of dinosaur scales reveals retention of reptile-type skin during the evolutionary transition to feathers

Fossil feathers have transformed our understanding of integumentary evolution in vertebrates. The evolution of feathers is associated with novel skin ultrastructures, but the fossil record of these changes is poor and thus the critical transition from scaled to feathered skin is poorly understood. Here we shed light on this issue using preserved skin in the non-avian feathered dinosaur Psittacosaurus. Skin in the non-feathered, scaled torso is three-dimensionally replicated in silica and preserves epidermal layers, corneocytes and melanosomes. The morphology of the preserved stratum corneum is consistent with an original composition rich in corneous beta proteins, rather than (alpha-) keratins as in the feathered skin of birds. The stratum corneum is relatively thin in the ventral torso compared to extant quadrupedal reptiles, reflecting a reduced demand for mechanical protection in an elevated bipedal stance. The distribution of the melanosomes in the fossil skin is consistent with melanin-based colouration in extant crocodilians. Collectively, the fossil evidence supports partitioning of skin development in Psittacosaurus: a reptile-type condition in non-feathered regions and an avian-like condition in feathered regions. Retention of reptile-type skin in non-feathered regions would have ensured essential skin functions during the early, experimental stages of feather evolution.

scales form distinct transverse rows (Supplementary Fig. 3e-f).These scales correspond to the transverse abdominal scales typical of modern crocodilians and some squamates 7 .
Prior to this study, most of our understanding of Psittacosaurus skin derived from specimen SMF R4970, which shows preserved scaled skin covering almost its entire body 7 .Consistent with NJUES-10, SMF R4970 shows a small number of feature scales in the flank of the thorax and transverse rows of quadrangular scales in the abdomen 7 .The basement scales in SMF R4970 also form a hexagram pattern, but only on the limbs 7 and not in the torso (despite extensive preservation of the basement scales).The presence of the hexagram pattern on the torso of NJUES-10 therefore likely represents inter-or intraspecific variation.The absence of this pattern on the limbs of NJUES-10, on the other hand, may reflect poor skin preservation (Supplementary Fig. 2).
The preserved scales in SMF R4970 are larger than those in NJUES-10.The basement and feature scales are 1.4-2.3mm and 3-4 mm wide, respectively, in the thorax and abdomen region of SMF R4970, whereas these scales are 0.8-1.2mm and 1.5-2 mm wide, respectively, in NJUES-10.Given that the femur in SMF R4970 (150 mm; measured from ref. 8 ) is almost twice as long as in NJUES-10 (78 mm; Supplementary Table 1), the overall difference in scale sizes likely reflects different growth stages-in modern crocodiles, the scales increase in size with growth 9 .

Supplementary Note 4. Skin colour of Psittacosaurus
The geometry of the fossil melanosomes in the skin of NJUES-10 (oblate to spheroidal and ca.0.2-0.4μm wide) is consistent with previous reports from other Psittacosaurus specimens.The latter includes fossilised skin melanosomes in SMF R 4970 (ca.400 nm long and 250 nm wide) 8 , PKUP V1050 (482.5 nm long and 279.3 nm wide) 10 and PKUP V1051 (546.3 nm long and 389.7 nm wide) 10 .
The skin melanosomes of Psittacosaurus have been interpreted as phaeomelanin-rich based on their ovoid shape; accordingly, the skin of Psittacosaurus has been interpreted as a brown colour 8 .A correlation between melanosome geometry and visible tissue colour, however, has been established only for mammalian hair and maniraptoran feathers 10 .This correlation does not hold true for extant reptilian scales 10 and possibly not for the skin of most archosaurs.For instance, in the Nile crocodile 9 , the black iguana 11 and the pterosaur Tupandactylus 12 , melanosomes from the skin morphologically resemble the (low aspect ratio) phaeomelanosomes in extant feathers but are instead rich in eumelanin.Given the lack of chemical evidence for melanin in any specimen of Psittacosaurus, it is therefore not possible to determine skin colour.
It is, however, possible to comment on colour patterning.The distribution of fossil melanosomes within the skin varies across the torso (Supplementary Fig. 9).Melanosomes can be absent or can occur in only the uncornified epidermis layer or in both the uncornified and cornified layers (Figs.4-5 and Supplementary Figs.6-8).In extant crocodiles, such different distributions correspond to white, intermediate grey and black scale colours, respectively 9 .This lateral variation in melanosome distribution therefore suggests spatial variation in skin tone.The inferred individual skin tones can persist over 1 mm laterally (Supplementary Fig. 9), indicating at least millimetre-scale, macroscopic colour patterning.
Colour patterning in Psittacosaurus has been inferred for specimen SMF R 4970, where the skin is preserved organically 8 .The interpreted pattern is based on observed differences in tone, which were considered to reflect differences in original melanin density 8 .This approach is not applicable to our specimen, in which the skin is mineralised and shows no apparent variation in tone.
It is not possible to characterise in detail the colour pattern in NJUES-10 as this would require extensive sampling of the skin.High-resolution CT with the necessary resolution to visualise the spatial distribution of the mouldic melanosomes is currently not available.

Supplementary Note 5. The lack of feather preservation in NJUES-10
Bristle-like integumentary structures interpreted as feathers have been reported on the tail of specimen SMF R 4970 8,13 .NJUES-10 does not preserve any tail feathers, despite extensive preservation of skin in the torso.In an illustration by Li et al. (Extended Data Figure 3; 2014) 10 , two specimens with fossilised skin (PKUPV1050 and PKUP V1051) also appear to lack tail feathers; confirmation of this would require further inspection of the specimens and thus is not considered further here.
The lack of tail feathers in NJUES-10 may be a biological feature, representing sexual, ontogenetic or interspecific variation.Indeed, based on the relationship between femoral length and age 4 , NJUES-10 may have been a three-year-old juvenile whereas SMF R 4970 (femur length is ca.150 mm; measured from ref. 8 ) is consistent with a seven-year-old subadult/adult.
Alternatively the apparent lack of tail feathers may be a taphonomic artefact.It has been demonstrated experimentally that, during soft tissue silicification, the volume of silica precipitated correlates with the amount of available silica-binding functional groups, specifically the hydroxyl, amino and carboxyl groups 14 .These functional groups are common in decaying protein-rich tissues that include both scales and feathers.Without decay, however, these functional groups in scales and feathers are likely inaccessible for binding with silica, due to the polymerized and highly crosslinked nature of their protein structure 15 .In NJUES-10, the silicified scales are located almost exclusively in the torso and rarely the limbs and the tail, even though the latter were almost certainly covered by scales in vivo 8 .This heterogeneous preservation in the torso suggests heterogeneous rates of decay of the scales between the torso and the limbs/tail.In fact, this pattern is consistent with decay controlled by endogenous microbes 16 .After invading the body cavity from the gut 16 , endogenous microbes may have been responsible for degrading the protein structure of the scales in the torso, thereby creating new molecular sites for binding silica 14,15 .Feathers (and scales) from the limbs and tail, on the other hand, are relatively distal to the primary source of decay microbes and may have had experienced limited decay relative to the scalation of the torso.As a result, scales in the limbs and tail would have possessed fewer functional groups available for binding silica and thus a lower potential for silicification.  in the Supplementary information.Abbreviations: 'FL', forelimb (humerus plus radius); 'HL', hindlimb (femur plus tibia).All measurements are lengths in mm.

Figure 1 .
Sedimentary structures associated with NJUES-10.a Overview of the specimen.Arrow indicates location of e. b Vertical section showing soft-sediment deformation of the laminae underlying the vertebrae.Arrows indicate lateral thinning and truncation of the laminae caused by the impacting carcass at deposition; arrowhead indicates normal grading in a sedimentary lamina.under daylight (a) and UV (b) light; dashed lines indicate the approximate position of the polished section.c SEM image of the polished section; the white and grey arrows indicate absence of melanosomes and melanosomes present in only the lower skin layer, respectively.d-e A fossil skin sample (sampling location shown in Fig. 2b) under daylight (d) and UV (e) light; dashed lines indicate the approximate position of the polished section.f SEM image of the polished section; the black and grey arrows indicate melanosomes present in both the upper and lower skin layers and melanosomes present in only the lower skin layer, respectively.

Table 1 | Measurements of NJUES-10 long bones and comparison with other Psittacosaurus specimens.
Data for the other specimens were taken from ref.