Functional constraints on the number and shape of flight feathers

Nine to eleven primary feathers is recovered as the ancestral trait of crown birds and all volant species in our dataset fall within this range (Fig. 1; a symmetrical transition rates model, ΔAICc = 1.66; SI Appendix, Fig. S1). In some flightless birds, there is a functional transformation in the primary feathers resulting in a change in their number. In two flightless Palaeognathae families, ostriches (Struthionidae) and rheas (Rheidae), and in all penguins (Neognathae; Sphenisciformes), the number of primaries increases (16 to 62 feathers), while in other flightless Palaeognathae families, cassowaries and emus (Casuariidae) and kiwis (Apterygidae), there is a dramatic decrease in the number of primaries including their complete lost. All other secondary flightless modern birds (Neoaves), in which the loss of flight is a relatively recent evolutionary occurrence (<35 Mya), the plesiomorphic condition of nine to eleven primary feathers is retained, for example, in the Flightless Cormorant (Nannopterum harrisi), Kākāpō (Strigops habroptila), and many flightless Rallidae species (SI Appendix, Table S1). In contrast, in all flightless crown birds with more than eleven or less than nine primaries, the loss of flight occurred in the distant past (>35 Mya); for example, the oldest flightless fossil penguins are 62 Mya (24) and flightless stem Struthiornithiformes have been found in deposits that are approximately 40 Mya (25). This also suggests that in most occurrences in Neornithes, the loss of flight makes species vulnerable to extinction as most lineages are short-lived (i.e., there are few living flightless lineages with long evolutionary histories).

Among non-avian pennaraptorans, nine to eleven primary feathers is also recovered as the ancestral trait and is present in the two basal-most taxa examined, Caudipteryx (Oviraptorosauria) and Zhenyuanlong (Dromaeosauridae). The paravian clade Anchiornithinae (Anchiornis, Serikornis, and Eosinopteryx) exhibits a substantial increase in the number of primary feathers (15–25) compared to the plesiomorphic condition (Fig. 1). Among Aves, some specimens of Archaeopteryx preserve 12 primary feathers, but all other sampled non-neornithine birds have nine to eleven primaries (n = 16 tested taxa; SI Appendix, Table S1).

Among crown birds, the degree of primary feather vane asymmetry is a strong predictor of flight ability (r² = 0.53, P < 0.001; phylogenetic generalized least squares, PGLS; Fig. 2 and SI Appendix, Table S2). The degree of primary asymmetry in volant crown birds ranges from 2.50 to 6.63 (1.00 = no asymmetry; mean ± SD: 3.84 ± 0.89; n = 50 taxa), while in flightless birds, the range is 1.00 to 2.79 (1.61 ± 0.60; n = 24 taxa; Fig. 2). Only three flightless taxa in our sample overlap with volant species: Auckland Islands Teal (Anas aucklandica), Falkland Steamer-Duck (Tachyeres brachypterus), and Aldabra White-throated Rail (Dryolimnas cuvieri aldabranus). The high variability in this value among crown birds reflects variation in flight style and the high phenotypic plasticity of this trait.

Among non-avian pennaraptorans, Caudipteryx shows a relatively low value (2.03) and Anchiornis shows an extremely low value (1.02) indicating almost complete symmetry, which almost certainly indicates the absence of flight in these taxa (Fig. 2). Microraptor shows a high value (4.15), corresponding with the widespread inference that this taxon was volant (6, 26, 27). Archaeopteryx and other stem birds in our sample show relatively high values (2.72 to 4.50; 3.76 ± 0.48; n = 13 taxa; SI Appendix, Table S3), indicating primary vane asymmetry in the range of extant volant crown birds. Only one fossil taxon, the enantiornithine Orienantius ritteri, falls within the area of overlap between the volant and flightless crown birds (asymmetry value = 2.72), which likely means it was either secondarily flightless as a relatively recent evolutionary occurrence or little dependence on flight, as in some waterfowl (e.g., Slavonian Grebe Podiceps auritus, asymmetry value = 2.50) and Bustards (Otididae; e.g., Little Bustard Tetrax tetrax, asymmetry value = 2.68).

We tested the effect of ulna length and flight ability on the number of secondary feathers. Flight ability was categorized into three groups: 1) volant crown birds (n = 275 taxa), 2) flightless crown birds (n = 22 taxa), and 3) fossil taxa inferred to be volant based on primary count and vane asymmetry (n = 8 taxa; SI Appendix, Table S4). The three anchiornithines (Anchiornis, Serikornis, and Eosinopteryx) were analyzed separately since primary number and vane symmetry suggests these taxa were most likely flightless. The model that best explained the variation in the number of secondary feathers included ulna length, flight ability, and their interaction (ΔAICc = 26.20; r² = 0.41, P < 0.001; PGLS; SI Appendix, Table S5). Ulna length is a strong predictor of secondary number among all tested taxa. Among volant taxa, including crown birds and fossils, there is no difference in the relationship between the number of secondary feathers and the length of the ulna (P = 0.26). In contrast, flightless crown birds show a significantly different relationship (P < 0.001) in which most flightless taxa have on average more feathers per unit length of the ulna compared to volant taxa (Fig. 3).

Most crown birds have five to 11 pairs of rectrices. The model that best explains variation in the number of rectrices in neornithines includes flight ability (volant versus flightless), hydrodynamic ability (non-hydrodynamic versus hydrodynamic), and their interaction (ΔAICc = 3.55; r² = 0.15, P < 0.01; PGLS; SI Appendix, Table S6). Among volant crown birds, non-hydrodynamic species have fewer rectrices (3 to 13; 6.09 ± 1.39; n = 66 taxa; SI Appendix, Table S7) than hydrodynamic ones (6 to 11; 7.94 ± 1.61; n = 16 taxa), but among flightless crown birds non-hydrodynamic species have more rectrices (5 to 50; 27.71 ± 21.27; n = 7 taxa) than hydrodynamic ones (6 to 20; 10.50 ± 5.15; n = 8 taxa). Palaeognaths that lack both flight and hydrodynamic ability are characterized by the greatest number of rectrix pairs.

In contrast to crown birds, the number of rectrices among non-neornithine paravians (n = 24 taxa) shows greater variation. The ancestral condition in Paraves is estimated to be around 25 pairs of rectrices, while that of the Neoaves is approximately seven pairs (Fig. 4). Some long-tailed taxa have an even greater number of feathers, for example, the Anchiornithinae (estimated at about 50 pairs) and Archaeopteryx (20 to 27 pairs), but others have less, for example, Microraptor and Jeholornis prima (nine pairs) and Jeholornis palmapenis (six pairs) (28). Among some stem pygostylians, there is a significant decrease in the number of rectrices (e.g., Confuciusornithiformes and Enantiornithes). Several taxa in these clades have only one pair of rectrices, for example Confuciusornis sanctus (29, 30) and Protopteryx fengningensis (31). The Songlingornithidae and Hongshanornithidae (Ornithuromorpha), which are closer to modern birds, have five to six pairs, falling within the range of volant crown bird.

Our data indicate that all volant neornithines have between nine and eleven primary feathers (Fig. 5). Further research is required to understand this narrow range and limited variation, although we hypothesize this constraint may be related to feather shape and aerodynamics with possible developmental influences. Furthermore, it appears this trait does not change rapidly in response to the loss of flight resulting in a discrepancy between the observed phenotype and locomotor ability among birds that have recently lost flight (<35 Mya). In contrast, among all birds in which flight was lost deep in evolutionary time (>35 Mya; Fig. 6), the number of primary feathers has become modified, this occurring in different ways depending on the new function of the forelimbs and their plumage, or the lack of function (atrophy; e.g., kiwis; Fig. 1). For example, among penguins, the primary feathers no longer function as a lift producing surface and have thus become small, short, stiff, and dense in response to their new function in thermoregulation and smoothing the forelimb into a paddle for aquatic locomotion (Fig. 5D).

Despite the considerable changes in the morphology (three long, clawed manual digits) and proportions of the hand (much longer than the humerus to typically shorter) that occurred during the evolution of non-avian to neornithine pennaraptorans, nine to eleven primaries is also recovered as the ancestral pennaraptoran condition. Assuming that the strong relationship recovered among crown birds between the number of primary feathers and flight was also valid in non-neornithine pennaraptorans, as suggested by the fact that, despite considerable morphological variation in the hand, all non-neornithine birds crownward of Archaeopteryx also fall within this range, this finding suggests that flight evolved very early in this clade, prior to the divergence between the basal-most taxa we examined (Caudipteryx and Zhenyuanlong; Fig. 5I). This suggests that some form of volant ability evolved at the base of Pennaraptora, as previously claimed by some authors (1, 35, 36) and supported by the inference that scansoriopterygids, a group most commonly recovered as basal oviraptorosaurs (37), were likely volant (38, 39). Although some authors have inferred that flight may have evolved multiple times, the only previous study to provide support for this hypothesis in fact only claims the potential for powered flight in pennaraptorans whose phylogenetic relationships suggest multiple origins. This study relied on wing loading estimates (including estimates for taxa in which no primary feathers are actually preserved) and specific lift but did not incorporate feather morphology (27). This highlights the challenges in inferring ecology and locomotor function in extinct animals from a non-holistic approach. While not criticizing the methods of Pei et al. (27), and in fact, their conclusions remained tentative, as do ours, we argue it is impossible to assess flight potential in non-avian pennaraptorans without examining the structure of the feathers forming the wing itself.

This finding highlights the existence of critical gaps in the fossil record pertaining to the earliest evolution of the feathered wing, relevant for understanding the origins of volant abilities in Pennaraptora. The early appearance of volant abilities in pennaraptorans inferred by these data suggests that in most lineages, volant capabilities were lost secondarily early on. The presence of nine primary feathers in the basal oviraptorosaur Caudipteryx suggests a volant ancestor while the proportionately short forelimb and small wing surface and robust and elongate hindlimbs preclude flight in this taxon, together supporting previous hypotheses that this taxon was secondarily flightless (40). The numerous primaries present in anchiornithines together with their phylogenetic position (Fig. 5J) may also indicate a secondary loss of flight and a change in the function of these feathers as occurred in Palaeognathae and Sphenisciformes. These results contrast with at least one previous study that infers the potential for powered flight in Anchiornis (in fact, the results of this study place this taxon on the cut off between powered flight potential and flightlessness, and thus does not strongly support for the potential powered flight) (27), although there is no consensus regarding flight in this clade and at least one previous study considered Anchiornis non-volant based on feather morphology (34, 41).

Among crown birds, the number of primaries is uniform within each species and is therefore considered an important taxonomic characteristic (42–45), although there are extremely rare exceptions (46). Differences in the number of primary feathers in specimens of Archaeopteryx, 11 [e.g., the 10th specimen—WDC-CSG-100 (47)] versus 12 [e.g., the 11th specimen (48)], may suggest these specimens do not represent a single taxon (49–51). However, intraspecific variation may have been more common among stem birds. Primary number does not support a close relationship between Archaeopteryx and anchiornithines, which some studies suggest form a single clade (52, 53), resolved as both avian and non-avian, depending on the analysis.

The degree of primary vane asymmetry measured here (differences in vane width) reflects a combination of barb length and barb angle. Our data from secondarily flightless modern birds indicate that vane width asymmetry in the primaries of most flightless birds differs from that in volant birds, congruent with previous findings (54). Although in contrast with the conclusions drawn by Feo et al. (8) and studies utilizing their dataset (55, 56), in which the primaries of volant and flightless birds were found to be similar, this discrepancy most likely reflects both differences in the method of measurement and the taxa sampled. For example, flightless palaeognaths and penguins were not sampled by Feo et al. (8), and most flightless species in their sample lost flight in a relatively recent evolutionary time (<35 Mya) and thus, likely still retain functionally asymmetric primaries.

Due to the narrow overlap between volant and flightless species (Fig. 2A), the degree of primary vane asymmetry appears to be a trait that responds to selective pressures at a faster rate over evolutionary time compared to primary number, both in response to the refinement of flight abilities and its secondary loss. Therefore, vane asymmetry is a more reliable predictor of volant ability in fossil taxa. Of the three extant flightless taxa in our dataset that fall within the range occupied by volant taxa, the two ducks belong to genera that include both volant and flightless species. Although overlapping with the lower range of values occupied by volant taxa, the ratio in these flightless species is still lower than in the volant members of their genus; for example, the degree of asymmetry measured in the flightless Falkland Steamer-Duck is 2.64 compared to 3.15 in the Flying Steamer-Duck (Tachyeres patachonicus). The third taxon, the Aldabra White-throated Rail, probably represents the most rapid and recent loss of flight among crown birds; this change occurred between subspecies at most 80,000 to 130,000 years ago (57). The degree of asymmetry measured in the flightless Aldabra White-throated Rail is 2.63 compared to 4.27 in the volant nominate subspecies Dryolimnas c. cuvieri, attesting to the rapid rate of phenotypic change in this trait in response to changes in locomotor behavior and feather function.

The two primary feather traits studied here (number and symmetry) change at very different rates in response to the secondary loss of flight (Fig. 6). This difference likely reflects the degree of phenotypic flexibility inherent to each trait. A flexible trait that changes relatively rapidly over evolutionary time, such as feather asymmetry, may provide reliable information regarding the current locomotor ability of the studied taxon. In contrast, a trait characterized by low evolutionary flexibility which changes at a slow rate, such as number of primary feathers, informs on the ancestral trait in a studied clade. For example, Caudipteryx, which is inferred to be secondarily flightless by our analyses, retains the flight-related trait of having between nine and 11 primary feathers, yet the degree of vane asymmetry falls clearly within the range of flightless taxa. We suggest that the origin of this contradiction lies in the differences in phenotypic flexibility between the two studied traits: The first, the number of primary feathers, represents the ancestral trait (low flexibility), while the latter, degree of vane asymmetry, represents the current locomotor ability (high flexibility). The phenotypic variation observed in vane asymmetry is likely due to the fact that changes can be easily produced through variations in molecular gradients (58). In contrast, modifications to the number of primaries requires changes to both feather number and either feather size or proportions of the hand.

Although Caudipteryx exhibits low vane asymmetry consistent with flightlessness, its primaries are still asymmetric to a small degree, greater than that observed in anchiornithines. The vane asymmetry ratio in Caudipteryx corresponds to values present in flightless crown bird taxa, in which flight was lost relatively recently in evolutionary time, for example, the Titicaca Grebe (Rollandia microptera; 12.4 Mya) and Snoring Rail (Aramidopsis plateni; 13.1 Mya; Fig. 6). This is consistent with the lack of change observed in the low flexibility trait, primary number, in all these taxa. In contrast to high flexibility traits (in this case vane asymmetry), the absence of change in a trait with low flexibility (in this case primary number) reveals the factors underlying the tested trait (in this case flight) by preserving the relationship between the trait and the original pressures that shaped it, even if these pressures are no longer present (in this case, the relationship between volant ability and the number of primaries). Although the Anchiornithinae (Anchiornis, Serikornis, and Eosinopteryx) are resolved as having a volant ancestor (as indicated by the fact nine to eleven primary feathers is recovered as the ancestral condition in Pennaraptora; Fig. 1), members of this clade have both low vane symmetry (59) and a high number of primary feathers; therefore, this clade is inferred to represent a secondary loss of flight ability that occurred relatively less recently in evolutionary time compared to Caudipteryx (similar to the flightless Palaeognathae and Sphenisciformes).

These cases of incompatibility between morphology and function (e.g., the flightless Caudipteryx having between nine and 11 primary feathers) may be at the forefront of future studies which will also consider developmental biology and the molecular mechanisms underlying the observed relationships. New studies on extant birds in this direction may provide insights regarding the mechanisms that determine rates of change in these features in response to changes in function and reveal underlying developmental controls.

Another morphological trait of the primary feathers, for which information regarding its degree of flexibility among the crown birds is available, is primary moult strategy, whether sequential or non-sequential moult. This trait shows even higher flexibility than vane asymmetry (Fig. 6), such that there is no flightless crown bird that moults its primary feathers in a sequential strategy (60). Although there are species in which non-sequential moult developed prior to the loss of flight (e.g., Anatidae and Rallidae) (61), others clearly show a rapid change in this trait in response to the loss of flight, for example, the Flightless Cormorant, a species that lost the flight ability about 5.3 Mya (SI Appendix, Table S8). Unfortunately, information regarding moult strategy in non-neornithine pennaraptorans is definitively known in only one fossil taxon, Microraptor (60, 62, 63). The discovery of additional fossils preserving active primary moult will be critical for better understanding the evolution of flight among dinosaurs (62).

Unlike the number of primary feathers, which is unaffected by body size among pennaraptorans (including crown birds), the number of secondaries is affected by ulna length, the bone on which these feathers attach. This difference reflects the direction of growth of these two feather tracts: Extension of the primaries can be achieved by lengthening the feathers themselves because they grow in a direction almost horizontal to the wing, for example, hummingbirds (Trochilidae; Fig. 5E), swifts (Apodidae), and swallows (Hirundinidae). In contrast, to extend the area of the wing formed by the secondary feathers, the perpendicular growth of these feathers relative to the wing bones on which they attach requires additional secondary feathers, for example, albatrosses (Diomedeidae), gulls (Laridae), and pelicans (Pelecanidae; Fig. 5B). Note that in many species, both mechanisms act together (e.g., Alcidae and Meropidae).

Interestingly, our results show that the relationship between the number of secondary feathers and the length of the ulna is also related to flight ability (Fig. 3). In some flightless species, secondary feathers have atrophied and their number is decreased significantly (e.g., cassowaries and kiwis; Fig. 5C). In other flightless species, the function of flight feathers has changed, and as a result, the number of feathers in relation to the size of the wing has also become modified, such that there is a greater number of secondary feathers for a given ulna length (e.g., penguins; Fig. 5D). Two out of the three tested anchiornithine genera (Serikornis and Eosinopteryx) are interpreted as revealing a similar pattern. These taxa show an increase in the number of secondary feathers relative to ulnar length that is disproportionate with size (Fig. 3). Comparison with extant taxa indicates that this is a trait that apparently does not support specialized volant abilities and is consistent with our other findings that indicate anchiornithines were non-volant.

Among crown birds, the tail can play an important aerodynamic function, helping to produce lift and drag forces, thus aiding to maintain stability and balance, and facilitating turning and slow flight such as landing and take-off (64, 65). However, even in volant birds, rectrices have a relatively wider range of functions compared to remiges (13, 65–67). They are also utilized in forms of locomotion beyond flight, such as in woodpeckers (Picidae) and other tree creeper taxa in which they are used for mechanical support during vertical climbing (13, 67). These feathers may alternatively serve as ornaments (e.g., Long-tailed Widowbird Euplectes progne and Superb Lyrebird Menura novaehollandiae), produce sound by vibration (e.g., Gallinago snipes and Anna’s Hummingbird Calypte anna) (13), and therefore be involved in displays. Although these functions may also occur in the remiges in rare instances, it is far more common in the rectrices. Wing flight feather morphology is tightly constrained by flight in volant taxa, whereas unspecialized flight does not rely upon the tail permitting the effects of sexual selection to be more prevalent in rectrices. As a result, there is enormous variation in the size and morphology of individual tail feathers. This wider range of functions suggests evolutionary signals in the tail feathers should be inherently weaker or more complex.

Among volant neornithines, the species with the highest number of rectrices investigated in this study is the Pin-tailed Snipe (Gallinago stenura; 13 pairs; Fig. 4). The large number of rectrices in this genus, especially the extra rectrices on the edge of the tail, are used for making sound during flight displays and not specifically for locomotion (neither aerodynamic nor hydrodynamic) (13). The volant crown bird with the least number of rectrices is Emuwrens (Stipiturus; three pairs; Fig. 5G); in this genus, the feathers are elongated and modified, suggesting the tail is mainly used in visual communication. Recently flightless non-hydrodynamic species have rectrix pair counts similar to volant birds (e.g., 5 to 6 pairs in the Kākāpō and flightless rails). Interestingly, hydrodynamically capable flightless crown birds also have a similar number of rectrices as their volant relatives. For example, flightless steamer-ducks have nine rectrix pairs compared to eight pairs in flying steamer-ducks; the extinct flightless Great Auk (Pinguinus impennis) had six pairs, similar to other alcids; and flightless cormorants have seven pairs, as in the volant Great Cormorant (Phalacrocorax carbo). Among these aquatic taxa, we found that rectrices are characterized by a rigid rachis and are apparently used in aquatic locomotion, maintaining the hydrodynamic contour of the body and helping to generate thrust underwater. Several penguin species also have well-developed rectrices, which contrasts with their otherwise small feathers (Fig. 5D). However, in the flightless and non-hydrodynamic Palaeognathae rectrices are numerous (most taxa have between 40 and 50 pairs), resembling and probably functioning like body feathers (Fig. 5C).

Compared to crown birds, evolutionary trends in the number of rectrices in non-avian paravians (the number of tail feathers could not be accurately counted or estimated in any published oviraptorosaur) is additionally complicated by changes in the skeletal morphology of the tail that occurred during early avian evolution: namely, the evolution of the abbreviated tail and pygostyle (68, 69). Additional lineage-specific trends in tail length are also observed in oviraptorosaurs (decrease in the number of free caudal vertebrae and independent evolution of a pygostyle-like structure) and jeholornithiforms (increased number of free caudal vertebrae) (68). In some taxa, it appears that the rectrices may have either functioned more similar to body plumage (Anchiornithinae) or in display (e.g., the elongate paired rectrices in Confuciusornis; Fig. 5M). In some stem birds, sexual dimorphism of ornamental rectrices is hypothesized (70). However, vane asymmetry is observed in the rectrices of some taxa [e.g., Archaeopteryx (48)], indicating that these feathers had some aerodynamic function, as in many crown birds (65, 67).

The plesiomorphic pennaraptoran rectricial condition is a high number of feather pairs due to the elongate morphology of the plesiomorphic pennaraptoran tail. Therefore, even taxa in which the tail had some aerodynamic function but retained an elongate reptilian tail (e.g., Archaeopteryx), a greater number of rectrices was present than in volant birds with aerodynamic tails. The refinement of flight in long bony-tailed paravians is associated with a reduction in the number of rectrices, which may have served to reduce mass and drag (71); rectrices are distally restricted in Microraptor and Jeholornis (the dorsal tail feathers in Jeholornis are here considered modified coverts; Fig. 5 K and L). Notably, tail feathers are also distally restricted in basal oviraptorosaurs (e.g., Caudipteryx) (72).

The evolution of the pygostyle also co-evolved with a decrease in the number of rectrix pairs, with non-neornithine pygostylians having between one and six rectrix pairs. The taxa with the lowest number of rectrices (one pair; Confuciusornithiformes and many enantiornithines) are thought to lack rectricial bulbs (73), which function to spread the tail feathers during slow flight in birds. Rectricial fans consisting of several rectrix pairs may lack aerodynamic benefit in the absence of this structure, limiting these clades to ornamental tails without aerodynamic function (68, 69). In contrast, in species where the morphology of the pygostyle suggests the presence of rectricial bulbs, the number of rectrix pairs is consistent with that in volant neornithines (e.g., Ornithuromorpha, Sapeornis) (68).

This study explores the evolution of the remex and rectrix feathered tracts in dinosaurs, from Mesozoic pennaraptorans to extant modern birds, over a period of approximately 160 myr (1). Investigating the number of remex and rectrix feathers, and the degree of primary feather vane symmetry in a large dataset of living birds, we found that these morphological traits, especially those pertaining to the remiges, are driven by functional constraints related to locomotion, a fundamental trait of life (74). Applying these results to fossils supports previous hypotheses regarding the early evolution of flight in pennaraptorans and the single origin of flight in this clade. In addition, this suggests that the secondary loss of flight may be prevalent among extinct non-avian pennaraptoran dinosaurs and that early stages in the evolution of the pennaraptoran “wing” are not sampled in the currently available fossil record. However, we refrain from making conclusive statements regarding this controversy but rather present important new results that complicate and challenge the current interpretations. This study reveals the critical importance of feathers in understanding the ecology and evolution of pennaraptoran dinosaurs and their modern avian descendants. These findings highlight the importance of linking an animal’s morphology and function through analyses of large datasets built upon extant taxa in order to facilitate evolutionary inferences in extinct clades (60) and further demonstrates the importance of considering ecology, morphology, and biomechanical traits in order to improve our understanding of evolutionary trends (75).

In order to examine the relationship between the number of wing (remiges) and tail (rectrices) flight feathers and flight ability among extant birds, we used visual inspection to count the number of primary, secondary, and rectrix feathers in specimens (skins) available in Ornithology collections (SI Appendix, Fig. S2). Studied specimens belong to the collections of five natural history museums: 1) the Field Museum of Natural History, Chicago, Illinois, 2) the Tring Natural History Museum, Tring, United Kingdom, 3) the Natural History Museum of Czech Republic, Prague, Czech Republic, 4) the Steinhardt Museum of Natural History, Tel-Aviv, Israel, and 5) the National History Museum of Denmark, Copenhagen, Denmark. The tertials, the three to seven innermost secondary feathers, were counted in one sequence with the secondaries as by previous authors (76).

Due to the relative rarity of flightless extant birds, we sampled all flightless taxa available at each of these collections together with closely related volant taxa. In addition, we sampled at least one species from all 41 extant bird orders. Two orders were not tested for the number of rectrices, Apterygiformes (Kiwis), and Podicipediformes (Grebes), due to the inability to count or estimate the number of their extremely small size tail feathers. The sample size we tested reflects the degree of variation in the tested trait. For the analysis testing the number of primaries, which show a relatively low variation among birds, for orders that do not include flightless species, we examined only one or two species. For the analysis testing the number of rectrices, which show more variation, we examined a larger sample size in order to encapsulate variation and examine evolutionary trends. In contrast to the primaries and rectrices, the number of secondaries shows high variation, in some species even between individuals. Therefore, for the analysis testing the number of secondaries, we examined many more species than we examined for two other feather tracts (which typically only included one individual for each species).

Fossil specimens of pennaraptoran dinosaurs (non-avian and stem birds) preserving feathers stored in 17 paleontological collections were examined visually to determine the number of primary, secondary, and rectrix feathers. The studied specimens belong to the following collections: 1) Beijing Museum of Natural History, Beijing, China, 2) Natural History Museum, London, United Kingdom, 3) Dalian Natural History Museum, Dalian, China, 4) Geological Museum of China, Beijing, China, 5) Paleontological Center, Bohai University, Jinzhou, China, 6) Henan Geological Museum, Zhengzhou, China, 7) Institute of Vertebrate Paleontology and Paleoanthropology, Beijing, China, 8) Jinzhou Paleontological Museum, Jinzhou, China, 9) Las Hoyas Collection, Cuenca Museum, Cuenca, Spain, 10) The Museum of Beipiao, Sihetun, Chaoyang, Liaoning Province, China, 11) Natural History Museum, Berlin, Germany, 12) Nanjing Institute of Geology and Paleontology, Nanjing, China, 13) Palaeontological Museum of Liaoning, Liaoning, China, 14) Shandong Museum, Jinan, China, 15) Shandong Tianyu Museum of Nature, Pingyi, China, 16) Wyoming Dinosaur Center, Thermopolis, Wyoming (US), and 17) Yizhou Fossil and Geology Park, Liaoning, China. For most specimens, we relied on high-resolution photographs. For some flightless crown birds and fossil specimens with multiple feathers that we could not count precisely but could provide a good estimate, estimates of the number of feathers were calculated (species-specific information is detailed in SI Appendix, Tables S1, S3, S4, and S7 and Fig. S3).

The degree of primary vane asymmetry was calculated as the ratio between the inner (posterior, trailing edge) and outer (anterior, leading edge) vanes (SI Appendix, Fig. S2). This ratio is measured in the central part of the primary feathers but not in the innermost or outermost feathers, whose shape is different. For each individual, the ratio is measured at several points (≥3) and the final value determined as the mean of all the measured points. For each species, we tested three individuals, and the species-specific value is the mean of the three individuals. However, for several taxa, mainly flightless or other rare species, less than three individuals were available. For extant bird species, the degree of primary feather vane asymmetry was calculated for all taxa for which the number of primaries was tested with one exception, the Emu (Dromaius novaehollandiae) which lacks primary feathers. Fossil specimens were tested using the same method, although fewer individuals and feathers were available for each taxon (SI Appendix, Table S9 and Fig. S3).

The length of the ulna (mm; SI Appendix, Fig. S2) was measured for all specimens in which the number of secondary feathers was examined. The length of the ulna was used as an independent factor affecting the number of secondary feathers.

Because species traits are known to be phylogenetically conserved, and thus data from closely related species are not statistically independent, we used an independent contrasts analysis, which identifies evolutionarily independent comparisons (77). To account for phylogenetic non-independence, we conducted all analyses using a phylogenetic generalized least square (PGLS) regression (78). We examined the strength of phylogenetic non-independence using the maximum likelihood value of the scaling parameter Pagel’s λ (79) implemented in the R package “caper” (80). Pagel’s λ is a multiplier of the off-diagonal elements of the variance–covariance matrix, which provides the best fit of the Brownian motion model to the tip data, and ranges between zero (no phylogenetic signal) and one (phylogenetic signal that depends on branch length as in analysis of phylogenetic-independent contrasts). We then corrected for the effects of shared ancestry using the maximum likelihood value of λ. The phylogenetic tree (SI Appendix, Fig. S4 and Table S10) was obtained from an analysis of global bird diversity (81, 82) using a random sample of 1,000 trees that were downloaded from the BirdTree project (83), the Ericson backbone (Ericson all species: a set of 10,000 trees with 9,993 OTUs each). This tree is the most detailed time-calibrated phylogeny that is currently available at the species level. The consensus tree was built using BEAST version 1.8.4. Two taxa not included in this tree were manually added using information published in the literature, the extinct Great Auk (Pinguinus impennis) (84) and Aldabra White-throated Rail (Dryolimnas cuvieri aldabranus) (57). Analyses (two-tailed, critical α = 0.05) were performed using R (version 4.2.2; R Development Core Team 2016).

Using PGLS, we examined several factors that may affect the number of remiges and rectrices and primary vane asymmetry among the extant bird dataset and compared these results with the findings we documented for pennaraptoran dinosaurs and stem birds. We examined 1) the effect of flight ability (independent categorical variable: volant versus flightless) on the degree of primary feather vane asymmetry (dependent continuous variable; univariate model); 2) the effect of ulna length (independent continuous variable), flight ability (independent categorical variable: volant versus flightless), and their interaction, on the number of secondary feathers (dependent continuous variable; multivariate model); and, 3) the effect of aerodynamic and hydrodynamic abilities and their interaction (independent categorical variables: volant versus flightless and hydrodynamic versus non-hydrodynamic species) on the number of rectrices (dependent continuous variable; multivariate model).

In order to study the evolution of remex and rectrix number and primary vane asymmetry, as well as to estimate and visualize the ancestral state of these traits, we used an ancestral trait reconstruction analysis under a continuous-time Markov chain model in the R package “phytools” (version 0.6-99; Phylogenetic Tools for Comparative Biology) (85). For this purpose, we built a cladogram (SI Appendix, Fig. S5) representing the evolutionary relationships of pennaraptoran dinosaurs, spanning from the early pennaraptoran oviraptorosaur Caudipteryx to modern birds (Neornithes), encompassing about 160 million years (1). Because no published phylogenetic analysis includes all taxon sampled here, the cladogram was built from a combination of the information published in several studies (37, 53, 86–95), while the Neornithes part was built based on an analysis of global bird diversity (81, 82) and BirdTree project (83), as described above.

Due to the nature of primary number variation, which is divided into three categories, taxa with more than 11 primaries, taxa with nine to 11 primaries, and taxa with less than nine primaries (SI Appendix, Table S1), and because of the centrality of this trait in our study, we repeated the ancestral trait reconstruction analysis using a categorical variable (three categories). In this analysis, we explored three models of discrete character evolution: 1) an equal transition rates model, 2) a symmetrical transition rates model, and 3) an all-rates different model. Then, we selected the best model based on the Akaike information criterion, modified for small sample sizes (AICc) (96). We selected a specific model only if it attained ΔAICc > 2.00 compared to the other models. Based on an analysis of global bird diversity (81, 82) and the BirdTree project (83), Tinamidae are placed at the base of the Palaeognathae. Due to the fact that recent avian molecular phylogenetic studies place this taxon deeper within the Palaeognathae (97), and due to the potential impact of this difference, we repeated this analysis. In the repeated analysis (SI Appendix, Fig. S6), we placed the Tinamidae as a sister of Apterygidae and Casuariidae, the Rheidae is a sister to the remaining Notopalaeognaths, and Struthionidae is a sister to all Palaeognathae (18, 98–100).

In addition, we used the same analysis for 400 species including all flightless crown birds in our sample. The purpose of this analysis was to estimate how long ago (Mya) each flightless taxon lost its flight ability (two categories: volant versus flightless; all are secondary flightless taxa). This time is calculated as the sum of all the lengths of the branches whose ancestral trait was “flightless” together with half the length of the first branch whose ancestral trait was “volant.” Since this value is an estimate, and in order to simplify the results of this analysis, we divided all flightless species into two categories: (I) those where the loss of flight probably occurred in the distant past (>35 Mya) and (II) those in which the loss of flight is a relatively recent evolutionary occurrence [<35 Mya; notably, this value represents roughly half the known fossil record of modern birds (101); SI Appendix, Table S8]. Of note, 35 Mya was chosen as a round number slightly older than the estimated lost flight timing by the oldest flightless avian lineage with modifications to feather asymmetry but not the number of flight feathers (New Guinea Flightless Rail Megacrex inepta; 34.0 Mya). This analysis allowed us to estimate the evolutionary time during which each tested morphological feature changes in relation to the loss of flight ability.

Y.K. and J.K.O. designed research; performed research; analyzed data; and wrote the paper.

The authors declare no competing interest.