Main

Feathers are the most complex integumentary appendages in vertebrates1. Most feathers in modern birds possess an axial shaft from which lateral barbs and barbules branch. Much is known about the anatomy, developmental biology and genomic regulation of these structures, but their deep evolutionary origin is controversial2,3,4. Feathers and feather-like integumentary structures have been reported in many theropod dinosaurs (including birds)3,5 and ornithischians, such as Psittacosaurus6, Tianyulong7 and Kulindadromeus8. Feather- or hair-like structures, termed pycnofibres9, have also been reported in several pterosaur specimens9,10,11,12,13, but their nature is not resolved.

Here, we report remarkably well-preserved pycnofibres in two anurognathid pterosaurs, and demonstrate—using evidence from morphology, chemistry and macroevolutionary analyses—that the preserved pycnofibres bear key features of feathers: monofilaments, two types of non-vaned grouped filaments, bilaterally branched filaments that were previously considered unique to maniraptoran dinosaurs, and preserved melanosomes with diverse geometries. Both specimens studied are from the Middle–Late Jurassic Yanliao Biota (around 165–160 million years ago14). NJU–57003 (Nanjing University) is a newly excavated specimen from the Mutoudeng locality. CAGS–Z070 (Institute of Geology, Chinese Academy of Geological Sciences), which has been noted briefly for its feather-like branched pycnofibres13, is from the Daohugou locality. Both specimens are near-complete and well-articulated, with extensive soft tissues (Figs. 1 and 2 and Supplementary Figs. 1–5). Both specimens are identified as anurognathids15 (see Supplementary Text for osteological descriptions).

Fig. 1: Integumentary filamentous structures in CAGS–Z070.
figure 1

a, Overview, showing extensive preservation of soft tissues. b–p, Details of the integumentary filaments in the regions indicated in a on the head and neck (b–d, i and j), forelimb (f and g), wing (l and m) and tail (o and p), and illustrated reconstructions of the filaments (type 1 filament (e), type 2 filament (h), type 3 filament (k) and type 4 filament (n)). Scale bars: 20 mm in a, 10 mm in b, 500 µm in c and i, 100 µm in d, 1 mm in f, l, m and p, 200 µm in g and j, and 5 mm in o.

Full size image
Fig. 2: Preservation, microstructure and chemistry of the integumentary filamentous structures in NJU–57003.
figure 2

a, Laser-stimulated fluorescence6,40,41 imaging highlights extensive preservation of the soft tissues (black areas). b–f, Details of the integumentary filaments in the regions indicated in a on the head and neck (b and c), wing (d and e) and tail (f). g,h, Scanning electron micrographs of the monofilaments on the neck (sample 10; g) and hindlimb (sample 39; h) (Supplementary Fig. 1a) show densely packed, elongate and oblate melanosomes. i, FTIR absorbance spectra of the monofilaments, monofilaments with sediment matrix, and sediment matrix in NJU–57003 (Sample 15; Supplementary Fig. 1a) compared with spectra from a feather of Anchiornis (from ref. 18), an extant Marabou stork feather (from ref. 19), and black and red human hair melanosomes (from ref. 22). Scale bars: 20 mm in a, 1 mm in b, c and e, 5 mm in d and f, and 1 μm in g and h.

Full size image

Preserved soft tissues include structural fibres (actinofibrils) and pycnofibres. Structural fibres—common in the pterosaur wing membrane9,12,16—are observed only in the posterior portion of the uropatagium in CAGS–Z070 (Fig. 1o,p). As reported elsewhere, they are parallel to subparallel and closely packed. Individual fibres are 0.08–0.11 mm wide (around 5 fibres per mm) and at least 1.9 mm long. Pycnofibres are preserved extensively in both pterosaur specimens (especially CAGS–Z070; Figs. 1 and 2 and Supplementary Figs. 1, 4 and 5) and are discriminated from structural fibres based on their curved morphology and overlapping arrangement. In the posterior portion of the uropatagium in CAGS–Z070, pycnofibres co-occur with structural fibres; oblique intersections reflect superposition of these features during decay (Fig. 1o,p).

Pycnofibres are categorized here into four types. Type 1 occurs around the head, neck, shoulder, torso, all four limbs and tail of both specimens (Figs. 1c–e,o,p and 2b,c,f). It comprises curved monofilaments that are 3.5–12.8 mm long and 70–430 μm wide. Some short, distally tapering examples discriminate between dark-toned lateral margins and light-toned axial regions, especially near the filament base where the light-toned axis is wider, suggesting a tube-like morphology (Fig. 1c–e). Type 2 is preserved in the neck, proximal forelimb, plantar metatarsus and proximal tail regions of CAGS–Z070. It consists of bundles of curved filaments of similar length that appear to form brush-like structures at the distal ends of thicker filaments (2.0–13.8 mm long and 80–180 μm wide) (Fig. 1f–h). These brush-like structures may represent individual thick filaments or fused proximal regions of thinner distal filaments. Type 3 occurs around the head of CAGS–Z070. It comprises straight to slightly curved, distally tapered, central filaments (4.5–7.0 mm long and 50–450 μm wide) with short lateral branches that diverge from the central filament near the midpoint (Fig. 1i–k). There are five type 3 filaments identified on the head, next to five similar filaments that are probably of the same nature but obscured by overlapping filaments (Supplementary Fig. 5b). Type 4 occurs on the wing membrane of both specimens. It comprises tufts of curved filaments (2.5–8.0 mm long and 70–130 μm wide) that diverge proximally (Figs. 1l–n and 2d,e), in contrast with the clear separation between type 1 filaments (Fig. 1o,p).

Filamentous integumentary structures in extant and fossil vertebrates commonly contain melanin-bearing organelles (melanosomes). Scanning electron microscopy (SEM) of the filamentous structures of NJU–57003 reveals densely packed microbodies 0.70 ± 0.11 μm long and 0.32 ± 0.05 μm wide (Fig. 2g,h, Supplementary Figs. 4a–f, 6 and 7 and Supplementary Table 2). As with most melanosome-rich fossil feathers17,18,19, energy-dispersive X-ray spectroscopy spectra of the filaments are dominated by a major peak for carbon (Supplementary Fig. 8). These carbonaceous microbodies resemble fossil melanosomes in terms of their geometry, dense packing, parallel alignment relative to the long axis of the integumentary structure (that is, barbules in Paraves) and preservation within the matrix of the filament (see Supplementary Text). Most of the microbodies are oblate and morphologically similar to those that are usually interpreted as phaeomelanosomes in fossils17 (Fig. 2h). Rod-shaped examples—usually interpreted as eumelanosomes in fossils17 (Fig. 2g)—are rare.

Fourier transform infrared spectroscopy (FTIR) of samples of pterosaur filaments shows four major peaks unique to the filaments (Fig. 2i). These peaks are consistent with the absorption regions of amide I at around 1,650 cm−1 (principally the C=O asymmetric stretching vibration with some C–N bending), amide II at around 1,540 cm−1 (a combination of N–H in-plane bending and C–N and C–C stretching, as in indole and pyrrole in melanin and amino acids) and aliphatic C–H stretching at 2,850 and 2,918 cm−1 (ref. 20). These peaks also occur in spectra obtained from extant feathers19,21, fossil feathers of the paravian Anchiornis18, and melanosomes isolated from human hair22. Furthermore, spectra of the pterosaur filaments more closely resemble those of pheomelanin-rich red human hair in the stronger absorption regions at around 2,850 and 2,918 cm−1 and higher-resolution spectra in the region around 1,500–1,700 cm−1 than those from eumelanin-rich black human hair and the ink sac of cuttlefish22. This, together with the SEM results, suggests that the densely packed microbodies in the pterosaur filaments are preserved melanosomes. The amide I peak at 1,650 cm−1 is more consistent with α-keratin (characteristic of extant mammal hair23) than β-keratin (the primary keratin in extant avian feathers20,24). This signal may be original or diagenetic; the molecular configuration of keratin24 and other proteins25 can alter under mechanical stress and changes in hydration levels.

The ultrastructural and chemical features of the pterosaur filaments confirm that they are hair- or feather-like integumentary structures. The four types of filaments described here show distinct distributions and morphologies. They are clearly separated from the sedimentary matrix by sharp boundaries (Supplementary Fig. 4g–i). There is no evidence that one or more filament type(s) were generated taphonomically (for example, through selective degradation or fossilization, or superimposition of filaments). For instance, although type 1 and 4 filaments occur widely in both specimens, type 4 occurs only in the wings, while type 1 occupies the remaining body regions. Type 1 filaments are thus not degraded products of type 4, and type 4 filaments do not represent superimposed clusters of type 1 filaments. Filament types 2 and 3 occur only in CAGS–Z070. Type 3 occurs only in the facial area and is associated with type 1, where types 2 and 4 are not evident. Type 3 filaments are thus not degraded type 2 or 4 filaments. Central filaments of type 3 are morphologically identical to the short, distally tapering filaments of type 1, but the branching filaments are much thinner (<40 μm for type 3 versus >70 μm for type 1) and shorter (<0.6 mm versus >3.5 mm, respectively) than the type 1 filaments. The branching filaments are thus unlikely to reflect superimposition of clusters of type 1 filaments. In contrast, the distal ends of type 2 filaments are similar, and have a similar distribution pattern, to type 1 filaments. An alternative interpretation—that type 2 filaments might represent superimposition of type 1 filaments at their proximal ends—is unlikely (see detailed discussion in the Supplementary Text). Feathers and feather-like integumentary structures have been reported in non-avian dinosaurs, although debate continues about their true nature2. These structures have been ascribed to several morphotypes—some absent in living birds3,5—and provide a basis to analyse the evolutionary significance of pterosaur pycnofibres. The pterosaur type 1 filaments resemble monofilaments in the ornithischian dinosaurs Tianyulong and Psittacosaurus and the coelurosaur Beipiaosaurus: unbranched, cylindrical structures with a midline groove that widens towards the base (presumed in Beipiaosaurus)3,5. The pterosaur type 2 filaments resemble the brush-like bundles of filaments in the coelurosaurs Epidexipteryx and Yi3,5,26: both comprise parallel filaments that unite proximally. The morphology and circumcranial distribution of pterosaur type 3 filaments resemble bristles in modern birds1, but surprisingly do not correspond to any reported morphotype in non-avian dinosaurs. The type 3 filaments recall bilaterally branched filaments in Sinornithosaurus, Anchiornis and Dilong, but type 3 filaments in Dilong branch throughout their length, rather than halfway along the central filament(s), as in the pterosaur structure3,5. The pterosaur type 4 filaments are identical to the radially branched, downy feather-like morphotype found widely in coelurosaurs such as Sinornithosaurus, Beipiaosaurus, Protarchaeopteryx, Caudipteryx and Dilong3,5.

The filamentous integumentary structures in our anurognathid pterosaurs are thus remarkably similar to feathers and feather-like structures in non-avian dinosaurs. Intriguingly, cylindrical (type 1), radially symmetrical branched (types 2 and 4) and bilaterally symmetrical branched (type 3) filaments clearly coexisted in individual animals; these structures may represent transitional forms in the evolution of feathers, as revealed by developmental studies3,5. These new findings warrant revision of the origin of complex feather-like branching integumentary structures from Dinosauria to Avemetatarsalia (the wider clade that includes dinosaurs, pterosaurs and close relatives)4,27. The early evolutionary history of bird feathers and homologous structures in dinosaurs, and the multiple complex pycnofibres of pterosaurs, is enigmatic. A previous study concluded that the common ancestor of these clades bore scales and not filamentous integumentary appendages2, but this result emerged only when the filaments of pterosaurs were coded as non-homologous with those of dinosaurs and there are no morphological criteria for such a determination. The presence of multiple pycnofibre types and their morphological, ultrastructural and chemical similarity to feathers and feather-like structures in various dinosaurian clades confirms their probable homology with filamentous structures in non-avian dinosaurs and birds. Comparative phylogenetic analysis produces equivocal results: maximum-likelihood modelling of plausible ancestral states against various combinations of branch length and character transition models (Supplementary Text, Supplementary Fig. 9 and Supplementary Table 3) reveals various potential solutions. The statistically most likely result (Fig. 3 and Supplementary Table 3; highest log-likelihood value) shows that the avemetatarsalian ancestors of dinosaurs and pterosaurs possessed integumentary filaments, with the highest likelihood of possessing monofilaments; tufts of filaments (especially brush-type filaments) are less likely ancestral states. This confirms that feather-like structures arose in the Early or Middle Triassic. The alternative tree for Dinosauria, with Ornithischia and Theropoda paired as Ornithoscelida28, produces an identical result.

Fig. 3: Phylogenetic comparative analysis of integumentary filament and feather evolution in pterosaurs and archosaurs.
figure 3

The phylogeny is scaled to geological time, with recorded terminal character states for each species and estimated ancestral character states at the lower nodes. This model is the most likely of the maximum-likelihood models, based on minimum branch lengths and transitions occurring as all-rates-different, but other results with lower likelihoods show scales as ancestral. The ancestral state reconstruction shows a combination of monofilaments, tuft-like filaments and brush-type filaments as the ancestral state for Avemetatarsalia and Dinosauria. The estimated ancestral state for Theropoda comprises all five feather states. Numbered small vertical arrows indicate the earliest occurrences of feather types 2–6. Two hypotheses for the timing of avian feather origins are indicated: early origin, at the base of Avemetatarsalia in the Early Triassic (A) or late origin, at the base of Maniraptora in the Early–Middle Jurassic (B).

Full size image

Present these modelling data with caution for two reasons: (1) the tree rooting method can influence the result (Supplementary Table 3), favouring results in which either scales are the basal condition or non-theropod feather-like structures and feathers evolved independently (Supplementary Fig. 9 and Supplementary Table 3); and (2) there is no adequate way to model the probabilities of evolution of all six feather types, or to model the probabilities of transitions between the six different feather types.

The discovery of multiple types of feather-like structures in pterosaurs has broad implications for our understanding of pterosaur biology and the functional origin of feather-like structures in Avemetatarsalia29,30. Potential functions of these structures include insulation, tactile sensing, streamlining and colouration (primarily for camouflage and signalling), as for bristles, down feathers and mammalian hairs29,30,31,32. Type 1, 2 and 4 filaments could shape a filamentous covering around the body and wings (Fig. 4) that might have functioned in streamlining the body surface to reduce drag during flight, as for modern bat fur or avian covert feathers31,33. Type 1 and 2 filaments occur in considerably high densities, particularly around the neck, shoulder, hindlimb and tail regions where the high degree of superposition prevents easy discrimination of adjacent fibres. This, along with the wide distribution and frayed appearance, resembles mammalian underfur adapted for thermal insulation34,35. Despite the less dense packing of type 4 filaments on the wings, the morphology of the structures is consistent with a thermoregulatory function: down feathers can achieve similar insulation to mammalian hair with only about half the mass, due to their air-trapping properties and high mechanical resilience, and are effective in retaining an insulating layer of still air36. This may optimize the encumbrance of the large wing area to wing locomotion16. Type 3 filaments around the jaw (Fig. 4) may have had tactile functions in, for example, prey handling, information gathering during flight, and navigation in nest cavities and on the ground at night, similar to bristles in birds37.

Fig. 4: Reconstruction of one of the studied anurognathid pterosaurs.
figure 4

Yuan Zhang

This pterosaur exhibits diverse types of pycnofibres distributed in different body parts.

Full size image

Methods

Sampling

The specimen NJU–57003 is represented by two fragmented slabs, both containing original bone, fossilized soft tissues and natural moulds of bones. Each slab was glued together along the fissures by fossil dealers with the fossil on the surfaces untouched. The specimen CAGS–Z070 is represented by a single unbroken slab. Small flakes (1–3 mm wide) of samples with preserved integument and/or enclosing sediments were carefully removed from the inferred integumentary filaments from different parts of NJU–57003 (Supplementary Figs. 1a and 4a–c) using a dissecting scalpel. This method was used to avoid sampling from degraded products of other tissues, such as dermis, epidermis or even internal organs. Most samples were not treated further; the remainder were sputter-coated with gold to enhance SEM resolution (Fig. 2g,h and Supplementary Figs. 4a–f and 6). All of the experiments described below were repeated to validate the results.

SEM

Samples were examined using a JEOL 8530F Hyperprobe at the School of Earth Sciences, University of Bristol, and an LEO 1530VP scanning electron microscope at the Technical Services Centre, Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences. Both instruments were equipped with a secondary electron detector, back-scattered electron detector and energy-dispersive X-ray spectrometer.

Measurements of melanosomes

The geometry of melanosomes was measured from SEM images using the image-processing programme ImageJ (available for download at http://rsbweb.nih.gov/ij/). We measured the maximum short and long axis lengths of melanosomes that were oriented perpendicular to line of sight. From these data, we calculated the mean and coefficient of variation of the long and short axis, and mean aspect ratio (long:short axis). Based on the proposed taphonomic alteration of fossil melanosome size (shrinkage up to ~20% in both length and diameter)38,39, we modelled potential diagenetic alteration by enlarging the original measurements by 20%.

FTIR microspectroscopy

Samples of the filamentous tissues and associated sediments were removed separately from NJU–57003 and placed on a BaF2 plate without further treatment. The infrared absorbance spectra were collected using a Nicolet iN10 MX infrared microscope (Thermo Fisher Scientific) with a cooled HgCdTe detector, at the School of Earth Sciences, University of Bristol. The microscope was operated in transmission mode with a 15 µm × 15µm aperture. Ten spectra were obtained from the filamentous tissues. The spectra show consistent results. The example presented in Fig. 2 shows the highest signal-to-noise ratio and was obtained with 2 cm−1 resolution and 2,000 scans.

Fluorescence microscopy

Selected areas with extensive soft tissue preservation in NJU–57003 were investigated and photographed using a Zeiss Axio Imager Z2 microscope with a digital camera (AxioCam HRc) and fluorescence illuminator (514 nm light-emitting diode) attached, at the Technical Services Centre, Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences.

Laser-stimulated fluorescence imaging and data reduction protocol

Laser-stimulated fluorescence images were collected using the protocol of Kaye and co-workers40,41. NJU–57003 was imaged with a 405 nm 500 mW laser that was projected into a vertical line by a Laserline Optics Canada lens. The laser line was swept repeatedly over the specimen during the exposure time for each image in a dark room. Images were captured with a Nikon D610 digital single-lens reflex camera fitted with an appropriate long-pass blocking filter in front of the lens to prevent image saturation by the laser. Standard laser safety protocols were followed during laser usage. The images were post-processed in Photoshop CS6 for sharpness, colour balance and saturation.

Phylogenetic macroevolutionary analysis

To analyse the evolution of feather characters, data were compiled on known integumentary characters across dinosaurs and pterosaurs. The basic data were taken from the supplementary data of Barrett et al.2, comprising 84 dinosaurs (33 ornithischians, 7 sauropods and 44 theropods (including 4 Mesozoic birds)). To this dataset, we added four pterosaurs. Barrett et al.2 scored taxa for three integumentary states (scales, filaments and feathers) in their macroevolutionary analyses. We checked and followed these basic categories and added three more. We then cross-referenced these six categories against the feather morphotypes defined by Xu et al.42 The categories used herein are: scales (1; not included in Xu et al.42), monofilaments (2; morphotypes 1 and 2 in Xu et al.42), brush-like filaments associated with a planar basal feature (3; morphotypes 4 and 6 in Xu et al.42), tufts of filaments joined basally (4; morphotype 3 in Xu et al.42), open pennaceous vane, lacking secondary branching (5; morphotype 5 in Xu et al.42) and closed pennaceous feathers comprising a rachis-like structure associated with lateral branches (barbs and barbules) (6). There was some uncertainty over feathers coded herein as type 3, which could correspond to morphotype 6, or morphotypes 4 and 6 in Xu et al.42. However, the only taxa coded with these as the most derived feather type are Sordes pilosus and Beipiaosaurus inexpectus. These taxa belong to separate clades; thus, the calculation of ancestral states is not affected by how our feather type 3 is coded (that is, whether treating morphotypes 4 and 6 of Xu et al.42 in combination or separately).

As in previous studies2, we used maximum-likelihood approaches to explore trait evolution. There are many methods to estimate ancestral states for continuous characters, but choices are more limited for discrete characters, such as here, where only maximum-likelihood estimation of ancestral states is appropriate43. We calculated maximum-likelihood reconstructions of ancestral character states using the ‘ace’ function of the ape R package44, with tree branch lengths estimated in terms of time, derived using the ‘timePaleoPhy’ function in the paleotree package45 and the ‘DatePhylo’ function in the strap R package46. These enabled us to assess the results according to three methods of estimating branch lengths: the ‘basic’ method, which makes each internal node in a tree the age of its oldest descendant; ‘equal branch length’ (equal) method, which adds a pre-determined branch length (often 1 Myr) to the tree root and then evenly distributes zero-length branches at the base of the tree; and minimum branch length method, which minimizes inferred branching times and closely resembles the raw, time-calibrated tree. A problem with the ‘basic’ branch length estimation is that it results in many branch lengths of length zero, in cases where many related taxa are of the same age. In these cases, we added a line of code to make such zero branch lengths equal to 1/1,000,000 of the total tree length. A criticism of the minimum branch length method is that it tends to extend terminal branching events back in time, especially when internal ghost lineages are extensive2, but this is not the case here, and the base of the tree barely extends to the Triassic–Jurassic boundary.

We ran our analyses using three evolutionary models with different rates of transition between the specified number of character states (six here)—namely, an equal-rates model, an all-rates-different model and a symmetrical model. These were calculated using the ace function in ape2 and the add.simmap.legend function of the R package ‘phytools’47.

In a further series of analyses, we attempted a different approach to the ancestral state modelling, by recording all feather type traits found in each taxon (see Supplementary Results), so coding multiple trait values for taxa that preserve multiple feather types. This did not shed much light on patterns of evolution of feather types because the multiple trait codings (for example, 1,2 or 2,5,6) were each made into a new state, making 14 in all, and these were not linked. Therefore, the six multiply coded taxa that each had feather type 6 were represented as six independent states and their evolution was tracked in those terms. Furthermore, we attempted to separate the six characters so they would track through the tree whether recorded as singles or multiples in different taxa; however, we did not have the information to enable us to do this with confidence because of gaps in the coding. In terms of reality, these multiply coded taxa still represent an incomplete sample of the true presence and absence of character states. By chance, many coelurosaurs are not coded for scales (1) or monofilaments (2); yet, it is likely they all had these epidermal appendages. Therefore, attempting to run such multiple codings, with characters either as groups or coded independently, encounters so many gaps that the result is hard to interpret. Our approach is to code the most derived feather in each taxon. This is also incomplete because of fossilization gaps, but at least it represents a minimal, or conservative, approach to trait coding and hence to the discoveries of macroevolutionary patterns of feather evolution. Complete fossil data might show wider distributions of each feather type and hence deeper hypothesized points of origin. Complete coding of feather types would of course allow each trait to be tracked in a multiple-traits analysis.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.