Reconstructing locomotor ecology of extinct avialans: a case study of Ichthyornis comparing sternum morphology and skeletal proportions

(a) Data acquisition and processing

The sternum of I. dispar was reconstructed as an idealized composite (figure 1a) based upon microcomputed tomography (μCT) scans of FHSM VP-18702, NHMUK PV A 905 and KUVP 119673 [38]. Retrodeformation methods are provided in the electronic supplementary material.

• Download figure
• Open in new tab
• Download PowerPoint

Following the detailed landmarking approach of [37] using 32 landmarks and nine sets of semi-landmarks, we incorporated the three-dimensional composite reconstruction of the Ichthyornis sternum into a dataset of three-dimensional CT and surface scanned sterna from 112 neornithine species, placing landmarks in Avizo (Thermo Fisher Scientific, Waltham, MA, USA). This dataset includes species with a wide range of locomotor styles across 62 neornithine family level clades and captures multiple independent origins of many major locomotor specializations across Neornithes, such as foot-propelled diving, soaring, burst flight and flightlessness. Locomotor capabilities were scored using the classification scheme of [37], using data from Cornell's Birds of the World [39]. Our phylogenetic comparative analyses used the composite tree from [40], which combines the age estimates and topology of [41] with the genus-level resolution from the supertree distribution of [42]. We added two extinct species from the original [37] dataset (Raphus cuculattus and Pinguinis impennis) using the ‘bind.tip' function in the R package phytools [43] in R 4.1.3 [44]. Ichthyornis dispar was appended to the tree using a custom function, with a minimum age estimate of 83.5 Ma from [18,22].

(b) Geometric morphometric analysis

Landmarks were aligned on the sternum of Ichthyornis using generalized Procrustes superimposition with the ‘gpagen’ function in the ‘geomorph' R package [45], to produce a general metric for sternum shape. This step translates, scales, and rotates a landmark constellation to a common centroid position, size and orientation, minimizing the bending energy distance of each landmark from the mean shape of the sample, and allowing semi-landmarks to slide along their tangent vectors [46,47]. Sternum centroid size, calculated as the square root of the summed squared distances of each landmark from the centroid position, was used as a proxy for sternum size [46]. We evaluated the relationship between sternal size and body mass for the dataset using phylogenetic generalized least squares (PGLS) regression [48]. We estimated the body mass for Ichthyornis based on a range of skeletal measurements from the specimen FHSM VP-18702, the most complete specimen used to construct the composite sternum, and incorporated these measurements into equations from [49] to yield a range of body mass estimates (electronic supplementary material, table S1). We included the minimum, mean and maximum body mass estimates for Ichthyornis into our analyses. The PGLS regression was performed using the ‘gls' function in the ‘nlme' R package [50] using a ‘corPagel' phylogenetic correlation structure in the ‘ape' package [51].

To assess the geometric similarity between the Ichthyornis sternum and those of crown birds, we visualized the major axes of sternum shape variation across our dataset with a principal component analysis (PCA) on the Procrustes coordinates using the ‘gm.prcomp' function in geomorph [45], and we mapped the phylogeny to the PCA to create a phylomorphospace [52]. We also calculated pairwise Euclidean distances between Ichthyornis and all other birds in the full multivariate shape space, to determine which neornithine taxa exhibit sternal shapes most similar to Ichthyornis.

To test the predictive power of sternum morphology for the presence or absence of various locomotor capabilities of birds, we performed a phylogenetic flexible discriminant analysis (pFDA) on the Procrustes scores of the Neornithes + Ichthyornis dataset. using the ‘phylo.FDA' function developed by [53]. Similar to linear discriminant analysis, pFDA uses categorical data—in this case, presence or absence of a locomotor mode—and multivariate shape data to estimate the posterior probability that a fossil taxon exhibited particular ecological traits (PP_trait) while accounting for phylogenetic non-independence in the multivariate dataset. Following [54], we conducted our pFDA using the regression scores produced from the best-supported Procrustes distance-based PGLS (D-PGLS) model, which identifies the variables that are significantly correlated with sternum morphology and illustrates the independent effects of each locomotor variable on sternum shape in Neornithes, while controlling for potential allometric and phylogenetic influences. Since the D-PGLS analyses correlations between shape and known locomotor variables, Ichthyornis was not included in this model, and D-PGLS regression scores for Ichthyornis were predicted by projecting their Procrustes shape coordinates onto the coefficient vectors of the D-PGLS using custom R functions [54]. We ran our discriminant function using the extant, or ‘training' dataset to predict the posterior probability (PP_trait) of each locomotor variable in Ichthyornis. To account for unequal variance in locomotor trait occurrence across our extant dataset, we subsampled the dataset at random for an equal number of taxa exhibiting ‘presence' and ‘absence' for each trait and ran 100 replicates of the pFDA on each variable. From this range of 100 estimates, we extracted the mean PP_trait value, as well as the first and third quartile of PP_trait estimates. We considered a mean PP_trait of <0.33 as ‘probably absent', PP_trait of>0.67 as ‘probably present', and PP_trait values in-between as uncertain, or equivocal. We also evaluated the accuracy of the pFDAs through their correct classification rates (CCR) for neornithine taxa with known ecologies. We considered traits to be reliably estimated when they showed median correct classification rates of 0.67 or higher. Both the CCR and PP_trait were considered when evaluating the pFDA predictions.

(c) Analysis of skeletal proportions

We collected the following 21 measurements from skeletal elements across the avian appendicular skeleton (electronic supplementary material, figure S1) from a landmarked dataset [55]: humerus length, humerus proximal width, humerus distal width, ulna length, ulna proximal width, radius length, carpometacarpus length, carpometacarpus proximal width, coracoid length, humeral articulation facet, sternal articulation facet, scapula length, sternum length, keel height, synsacrum length, ilium length, ischium length, pubis length, femur length, tibiotarsus length, and tarsometatarsus length (electronic supplementary material, table S2). This dataset, mainly taken from [55], comprises 145 neornithine taxa, which differ slightly from those examined in our sternum analysis owing to data availability but reflect similar taxonomic and locomotor breadth. The phylogeny was obtained in the same way as in the sternum analysis described above, with Ichthyornis appended at the base. Species in this dataset were coded for locomotor mode in the same way as in the sternum analysis, but with an additional category, ‘underwater locomotion', which includes all birds assigned as either wing-propelled or foot-propelled divers (which were also used as a separate categories). Measurements for Ichthyornis were included as the mean value of measurements from a range of Ichthyornis specimens. The following Ichthyornis specimens were used: ALMNH:Paleo:3316, BHI 6420, BHI 6421, FHSM VP 18702, KUVP 2281, KUVP 2300, KUVP 25469, KUVP 157821, KUVP 25471, KUVP 25472, KUVP 2284, KUVP 119673, KUVP 123459, NHMUK A 905, MSC 7841, MSC 5937, MSC 6201, MSC 6202, YPM 1461, YPM 1733, YPM 1450, YPM 1740, YPM 1741, YPM 1742 [22]. It should be noted that these specimens show a broad size distribution owing to well-documented intraspecific variation within Ichthyornis [18,22]. However, given that no individual specimen (including the most complete FHSM VP 18702) preserved all elements without breakage, computing mean values provides a sufficient estimate of the true proportions of Ichthyornis given the data available. The full dataset of measurements was log-transformed and then subjected to Mosimann's transformation, which computes size-corrected geometric shape vectors for each species by subtracting the log-geometric size (calculated as the mean across all measurements) from the log-transformed measurements [56]. The dataset was then run through a PCA to visualize the position of Ichthyornis relative to other birds in appendicular phylomorphospace [52]. Euclidean distances between Ichthyornis and all other species were calculated as an index of the body shape differences between Ichthyornis and species in our neornithine sample, within the full multivariate measurement space. We then tested the relationship between skeletal proportions and locomotor modes using distance-based PGLS models [57]. We performed model selection through stepwise regression, choosing the best fitting model based on the statistical significance of each variable after sequentially removing non-significant variables [58], as there is currently no adequate Akaike information criterion-based model selection method for D-PGLS. We then conducted a pFDA on the regression scores from the best supported D-PGLS model to predict the locomotor ecology of Ichthyornis, with the same criteria described above, to assess the predictive power of skeletal proportions for locomotor ecology compared with sternum shape alone.

(a) Body mass estimates

Body mass of Ichthyornis was estimated based on skeletal measurements of FHSM VP-18702, the most complete specimen, including the coracoid maximum glenoid diameter, coracoid maximum lateral length, coracoid least width and femur maximum length, following [49] (electronic supplementary material, table S1). After removing maximum and minimum estimates, the mean body mass estimate for FHSM VP-18702 was computed as 355.82 g. This falls within the range of body mass estimates for Ichthyornis generated by [22], with the smallest individuals estimated at 117.15–181.79 g and the largest at 426.64–486.18 g.

(b) Relative sternal size of Ichthyornis

As also observed in the Neornithes dataset of [37], our analysis of sternum size allometry showed that sternal size increases isometrically with body size (R² = 0.89, p < 0.0001). The body mass and sternal size of Ichthyornis are close to the median values for Neornithes in our sample, and sternal size is close to our mean value for Neornithes. The relative sternal size to body size of Ichthyornis is marginally smaller than predicted for a bird of its size (mean residual = −0.03 (−0.015 to −0.048)), and falls within the ranges of continuous flappers, and underwater swimmers and divers (figure 2). Within this group, Ichthyornis shows similar sternum-to-body size ratios to flying foot-propelled diving birds like the anhinga (Anhinga anhinga) and the aerial seabirds like the black skimmer (Rynchops niger) and the common tern (Sterna hirundo).

• Download figure
• Open in new tab
• Download PowerPoint

(c) Geometric morphometric analysis of sternal shape

The birds with the most similar sternal shape to Ichthyornis, based on Euclidean distances in shape space, are the short-tailed shearwater Ardenna tenuirostris (0.189), the white-tailed tropicbird Phaethon lepturus (0.190), the shoebill Balaeniceps rex (0.191), the light-mantled albatross Phoebetria palpebrata (0.194), the Marabou stork Leptoptilos crumeniferus (0.206), the African grey hornbill Lophoceros nasutus (0.213), the Abyssinian ground hornbill Bucorvus abyssinicus (0.226) and the anhinga A. anhinga (0.228) (table 1). The furthest bird from Ichthyornis in sternum shape space was the elegant crested tinamou Eudromia elegans, with a distance of 0.63.

The first four principal components (PCs) explained approximately 74% of sternal shape variance in our sample. Subsequent axes each accounted for less than 5% of total shape variance and thus are not discussed further.

Shape variation across the first four PC axes is almost identical to that described by [37]. PC1 (30.84%) presents negative values corresponding to narrow, elongated sterna with lateral trabeculae, and positive values corresponding to shorter, square-shaped sterna with cranially projected keels and concave caudal borders (figure 3a). Ichthyornis falls on the positive side of PC1, and is very close to A. anhinga, as well as the albatross Pho. palpebrata, the Titicaca grebe Rollandia microptera, and the shoebill B. rex. PC2 (23.86%) shows negative values corresponding to narrow, elongated sterna, and positive values corresponding to shorter, wider sterna with four sternal notches in continuous flappers. Ichthyornis plots on the low positive end of PC2, near the anhinga A. anhinga, the white-tailed tropicbird Pha. lepturus, the short-tailed shearwater A. teniurostris, and the Leach's storm petrel Oceanodroma leucorhoa (figure 3a; electronic supplementary material, S2). PC3 (11.18%) differentiates quite clearly between sterna with keels and without, with the most positive values associated with shallow keels and square-shaped caudal sternal borders, and the most negative values associated with taller, rounder keels and rounded caudal borders. Ichthyornis plots closest with the Leach's storm petrel O. leucorhoa, the large-tailed nightjar Caprimulgus macrurus, the ground hornbill B. abyssinicus and the masked booby Sula dactylatra on the negative end of PC3 (figure 3b; electronic supplementary material, S2). Finally, PC4 (7.6%) loosely differentiates between sternal keel angle and sternal notch depth, with positive values associated with shorter, rounder keels and shallower sternal notches, and negative values associated with deeper, cranially projected keels and deeper sternal notches. Ichthyornis plots on the low positive end of PC4 near birds with deeper, mildly anteriorly projecting keels like B. abysinnicus (figure 3b).

• Download figure
• Open in new tab
• Download PowerPoint

The best multivariate D-PGLS model, in which shape is the dependent variable, and size and locomotor traits are independent variables, was formulated as: sternum shape ∼ body mass + sternum size + forelimb propulsion + aerial flight + terrestriality + cursoriality + burst flight + foot-propelled swimming and diving + soaring (R² = 0.45, $R_{\mathrm{adj}}^2=0.402$; electronic supplementary material, table S3), indicating that these locomotor variables are the best predictors of sternum morphology in Neornithes.

Our pFDA for sternum morphology generally showed strong predictive power for specific locomotor traits: soaring, foot-propelled swimming/diving, burst flight and cursoriality (figure 4a). However, sternal shape shows less predictive power for the capability to fly in the first place (figure 4a, see aerial flight and forelimb propulsion). CCR varied across variables, with the highest value for burst flight (median = 0.9) and the lowest for aerial flight (0.54). The pFDA predicted that Ichthyornis exhibited foot-propelled swimming or diving and soaring and that burst flight and cursoriality were absent (table 2; figure 4a). Partial R² from the D-PGLS models for each variable does not directly correlate with CCR in all cases. Soaring has a relatively high partial R² and CCR; however, aerial flight also has a relatively high partial R² and the lowest CCR in our sample (table 2). Therefore, these two metrics represent different aspects of significance, and while partial R² can indicate a statistical correlation between shape and an ecological variable, it does not necessarily provide confidence in predicting ecological states.

• Download figure
• Open in new tab
• Download PowerPoint

Forelimb propulsion and aerial flight are predicted as present in Ichthyornis based on sternal shape, however with fairly low CCR (table 2; figure 4a). Foot-propelled swimming or diving was inferred as present in Ichthyornis, with a median CCR of 0.76 (0.6–0.86) and high probability (PP_present = 0.92; table 2; figure 4a). Soaring flight was estimated with a high degree of certainty (CCR = 0.88) and was inferred as present in Ichthyornis with a high PP_present (= 0.99). Burst flight was also estimated with a high degree of certainty (CCR = 0.9) and was inferred as absent with a very low PP_present (= 3.61 × 10⁻⁵; table 2; figure 4a). High correct classification rates and probabilities indicate that these locomotor variables can be confidently inferred from sternum morphology and that their presence and absence, respectively, are strongly supported. Cursoriality was also estimated with confidence (CCR = 0.75) and inferred as absent in Ichthyornis with low PP_present (= 0.05; table 2; figure 4a). Terrestriality was estimated with poorer confidence (CCR = 0.59) and inferred as marginally present in Ichthyornis, with a PP_present of 0.69 (table 2; figure 4a).

(d) Linear skeletal proportions analysis

The birds with most similar overall skeletal proportions to Ichthyornis, based on Euclidean distances, are the Auckland teal Anas aucklandica (1.095), the black skimmer R. niger (1.097), A. anhinga (1.098), the swallow-winged puffbird Chelidoptera tenebrosa (1.105), the booby S. dactylatra (1.128), the collared puffbird Bucco capensis (1.156), the green woodhoopoe Phoeniculus purpureus (1.163) and the common pochard Aythya ferina (1.187) (with A. anhinga also appearing in the list of birds with highly similar sternum morphology to Ichthyornis, table 1). The furthest from Ichthyornis was the southern brown kiwi Apteryx australis, with a distance of 4.76 (table 1; electronic supplementary material, figure S2b). A comparative scatterplot of Euclidean distances based on both sternum morphology and skeletal proportions is provided in figure 5, showing that the birds with the greatest overall similarity to Ichthyornis, considering both sternal morphology and skeletal proportions, are mainly an assemblage of water-associated birds, especially the anhinga (A. anhinga), the black skimmer (R. niger) and the short-tailed shearwater (A. tenuirostris).

• Download figure
• Open in new tab
• Download PowerPoint

The best multivariate D-PGLS model (here, appendicular proportions are referred to as ‘shape') is: shape ∼ body size + aerial flight + forelimb propulsion + underwater locomotion + soaring + flap-gliding + terrestriality + maneuverability, indicating that these locomotor variables are the best predictors of skeletal proportions in Neornithes (electronic supplementary material, table S4). This model had an R² of 0.518 ($R_{\mathrm{adj}}^2=0.49$), which is slightly higher than that for our sternum model (R² = 0.45, $R_{\mathrm{adj}}^2=0.402$).

The pFDA shows that skeletal proportions have generally weaker predictive power (low posterior correct classification rates) for inferring specific locomotor traits, compared with estimates based on sternum morphology (figure 4b). However, skeletal proportions show stronger predictive power for the capability to fly in the first place (CCR = 0.75 for forelimb propulsion, = 0.71 for aerial flight; table 3; figure 4b). The pFDA estimated forelimb propulsion and aerial flight to be present in Ichthyornis, with high PP_present (=0.89 for forelimb propulsion, 0.71 for aerial flight; table 3; figure 4b). More specific locomotor traits in the model, including soaring, terrestriality, flap-gliding and maneuverability were estimated as present, however, these showed overall unreliable correct classification rates, ranging from 0.5 for flap-gliding to 0.62 for terrestrial locomotion (table 3; figure 4b). Underwater locomotion was also estimated with a low correct classification rate (=0.58), and was predicted as equivocal in Ichthyornis, with PP_present = 0.599 (table 3; figure 4b).

(a) Ecological inferences from sternal morphology and skeletal proportions

Assessing the locomotor capabilities of Mesozoic stem birds is critical to reconstructing the origin and refinement of avian flight. Ornithurines, like Ichthyornis, represent the closest-known stem-group relatives of Neornithes and provide important evolutionary context for inferring locomotor capabilities of the earliest crown birds [18,22,59–61]. Overall, our three-dimensional analyses of sternum morphology appear to be considerably more useful for predicting the specific locomotor traits of extant birds, while our analyses of skeletal proportions show more confident predictions of flight itself (figure 4a,b). This discrepancy may be explained in part by the necessary reliance on overall skeletal structure for flight to occur, probably more so than on the intricacies of sternal shape. The more detailed shape information of the sternum captured through three-dimensional landmark data appears to denote greater ecological signal, indicative of specific locomotor ecologies (also see [37]).

Our results show that sternum shape can be used to predict specific locomotor abilities in extinct birds—more so for specialized traits like soaring, burst flight, and foot-propelled swimming than for more phylogenetically widespread traits, like terrestrial locomotion. Flight itself is less well-estimated by sternum shape, possibly because ‘flightless' birds are enormously disparate. They range from flightless ratites that have lost numerous skeletal specializations for flight [62], to penguins, which use their highly specialized forelimb apparatus and sternal keel for wing-propelled diving, to numerous additional taxa from across Neornithes that lost flying ability relatively recently, including flightless auks such as the great auk P. impennis, the Titicaca grebe R. microptera, and the Auckland teal Anas aucklandica [63–65]. Flightlessness may not be clearly reflected in the sternal features of taxa in the latter category, emphasizing the challenge of estimating flight capacity from shape-based morphometric analyses alone. We included the ‘forelimb propulsion' category in addition to ‘aerial flight' as an attempt to address this issue, creating a category that includes all birds that use their forelimbs for propulsion through any fluid medium, thereby including both flightless wing-propelled diving penguins and flying birds (see description of categories in [37]). However, neither forelimb propulsion nor aerial flight variables were confidently predicted by sternum morphology, and body size, as well as sternum size relative to body size [66,67], both appear to be better predictors for the presence or absence of flight than sternum shape alone [37]. The predictive power of appendicular proportions appears significantly greater for predicting the presence or absence of flight (as characterized by both aerial flight and forelimb propulsion), but a substantially lower degree of reliability for predicting locomotor mode in fossil birds. Several studies have shown that various avian skeletal measurements are more strongly correlated with phylogeny than with ecology or flight style, especially in wing proportions; thus, inferring ecology for fossil taxa from these measurements should be done cautiously [11,68–70].

(b) Locomotor ecology of Ichthyornis

Our study provides, to our knowledge, the first statistical support for a foot-propelled swimming and soaring ecology in Ichthyornis, which is compatible with the marine environment in which most of its fossils have been found. Its sternum morphology is most similar to that of several extant taxa that regularly soar and are associated with aquatic environments, including pelagic shearwaters and tropicbirds, the wading shoebill and the Marabou stork, soaring seabirds such as albatrosses, and the anhinga, a soaring suliform waterbird. Appendicular proportions of Ichthyornis, by contrast, show the strongest similarities with aquatic birds with long wings and short legs, such as ducks and terns, as suggested by [17,18,22], as well as gannets and boobies (Suliformes). Certain morphological features previously used to suggest swimming adaptations in Ichthyornis, such as foot proportions [22] are not included in our linear measurement dataset owing to the unreliable nature of pedal phalanx identification. Perhaps foot proportions alone may provide stronger signal for foot-propelled swimming than the remainder of the skeleton, however skeletal proportions excluding the feet have been used extensively to reconstruct ecology [11,12,68–73], making it a valid dataset for comparison with sternum morphology.

A few surprising taxa are found to be morphologically similar to Ichthyornis in either sternal morphology or skeletal proportions individually—hornbills (Bucerotidae) like the African grey hornbill L. nasutus and the Abyssinian ground hornbill B. abyssinicus, and the shoebill B. rex for sternum shape, as well as puffbirds (Bucconidae) like the swallow-winged puffbird C. tenebrosa and the collared puffbird Bucco capensis for skeletal proportions—are neither aquatic nor soaring. Hornbill and shoebill sterna have deep, anteriorly projected keels similar to Ichthyornis, a feature loosely associated with swimming [37], although these species do not swim. Therefore, the presence of these features in hornbills and the shoebill may be related to the structural accommodation of some other combination of traits, like the articulation or fusion of the furcula with the sternal keel, suggesting the interaction of ecological and functional traits to shape morphology, and that further ecological traits should be considered when interpreting sternum ecomorphology. Similarly, puffbirds are small, arboreal birds with short legs and long wings much like soaring seabirds; so, while this small hindlimb-to-forelimb ratio is indeed shared between puffbirds and Ichthyornis, these taxa differ dramatically in size and ecology. Further, continuous flapping, which is the primary locomotor mode used by hornbills and puffbirds, may be considered a ‘generalist' mode, with different morphotypes conferring the same ecology [74], and was not found to be significantly correlated with either sternum morphology or skeletal proportions in our analyses (tables 2 and 3; electronic supplementary material, S3 and S4). These observations strongly emphasize the value of considering multiple aspects of skeletal anatomy together, as well as body size, when interpreting ecology in fossil taxa. When this is done, Ichthyornis is generally supported as being most similar to an assemblage of water-related birds (figure 5).

Our results suggest that Ichthyornis was morphologically best suited for soaring and foot-propelled swimming, however, it was probably not specialized for foot-propelled diving like loons (Gaviidae) and grebes (Podicipedidae). The most striking signal of foot-propelled diving in neornithines is a greatly enlarged cnemial crest on the tibiotarsus, but the cnemial crest of Ichthyornis better resembles that of soaring and foot-propelled swimming darters and gannets (Suliformes) [22], which are only mildly enlarged. [22] also highlighted several hindlimb traits associated with foot-propelled swimming, namely a reduced femoral trochanter, a very short tarsometatarsus, and elongated pedal phalanges [22,70,75–77]. In particular, the elongated pedal phalanges of Ichthyornis contrast with its remarkably shortened tarsometatarsus, such that the longest phalanx represents more than 60% of the total tarsometatarsal length, as in living foot-propelled swimmers like the mallard Anas platyrhynchos, the Eurasian coot Fulica atra or the northern gannet Morus bassanus [22]. Although we did not include pedal phalanx measurements in our study owing to data availability and identifiability issues, the interpretations of [22] are consistent with our statistical findings.

In addition to an enlarged sternal keel, the deltopectoral crest on the humerus in Ichthyornis is particularly large and probably cranially projected, a trait shared with volant neornithines, but is also proximodistally long, resembling extinct paravians and modern palaeognaths [18,20,22,78]. This combination of traits is rarely seen in both stem and crown birds, as an enlarged keel and long deltopectoral crest are often thought to be mutually compensatory, except in soaring birds [3,79]. Soaring birds typically have long deltopectoral crests, suited for slow but strong wingbeats. By contrast, non-soarers with faster wingbeats tend to exhibit shorter deltopectoral crests [78]. The long deltopectoral crest of Ichthyornis further bolsters our inference of soaring based on sternum geometry, in addition to the brachial index (= humerus length/ulna length) of Ichthyornis calculated by [22] showing similarities with soaring marine taxa, such as terns, gulls, shearwaters and tropicbirds. Interestingly, the recently described Maastrichtian ichthyornithine Janavis finalidens exhibits similar humeral proportions despite its much larger size, suggesting that similar locomotor habits may have been retained among ichthyornithines for at least 20 Myr [80]. Birds that are both strong fliers and foot-propelled swimmers, like the darter, tend to show intermediate morphology between flight-specialized and diving-specialized taxa [81]. Further, the anhinga and the short-tailed shearwater showed high similarity to Ichthyornis in both appendicular proportions and sternum shape, suggesting that they are overall the most similar morphologically to Ichthyornis (figure 5). These, as well as gulls and terns are the groups with closest overall similarity to Ichthyornis in both morphological categories and can be considered reasonable modern analogues for the taxon.

This is, to our knowledge, the first study to present strong statistical support regarding the locomotor ecology of Ichthyornis. [13] presented equivocal support for both ‘non-swimming' and ‘foot-propelled surface-swimming' ecology in Ichthyornis based on skeletal measurements, and documented notable morphometric similarities between plunge-divers, surface swimmers and non-swimmers (fig. 4 in [13]), demonstrating the difficulty of unravelling these disparate ecologies using only linear measurements. With the recent suggestion of foot-propelled swimming in Ichthyornis by [22] based on skeletal elements not previously preserved or described for Ichthyornis, such as the pedal phalanges, the possibility of a plunge-diving foraging ecology cannot be ruled out.

Overall, the similarity of Ichthyornis to soaring and swimming seabirds contrasts with the hypothesized ecology of the extinct hesperornithines (Ornithurae) that occur in the same deposits as Ichthyornis [17,61,70]. The bone walls of Ichthyornis' forelimb elements are relatively thin compared with Hesperornis, which may relate to sustained flight capability in Ichthyornis, or increased specialization for diving in Hesperornis [23,82,83]. Hesperornithes are regarded as the oldest-known lineage of diving avialans [61]; however, it is possible that the last common ancestor of Ichthyornis and Hesperornithes was aquatic—indeed, the crownwardmost non-ornithurine euornithean, Gansus yumenensis, has been hypothesized as a volant foot-propelled swimmer or diver [12], suggesting that the origin of foot-propelled swimming or diving may predate the origin of the major avialan clade Ornithurae, highlighting the possibility of a Cretaceous aquatic phase in the direct ancestry of crown birds [84,85].

Properly characterizing complex functional traits like locomotion is challenging. These traits are often simplified into discrete categories, which can exclude important intermediate or overlapping states. Further, form-function relationships are often estimated as direct correlations, rather than complex, nested relationships involving interdependencies. Most functional traits probably exist in multidimensional continuous space, resulting in several intermediate alternatives, with overlapping traits or trait states. This appears to be the case for Ichthyornis, which probably exhibited foot-propelled swimming morphology somewhere in between that of loons and darters, and a soaring morphology characteristic of soaring seabirds, resulting in a combination of traits that represent unique locomotor abilities, but that are difficult to discretize. While discrete functional categories are straightforward to implement into statistical analyses, contemporary multivariate methods are well-equipped to handle multidimensional functional data. Expanding our characterization of functional traits to include continua, and combined, nested or overlapping states would allow for more detailed and comprehensive hypothesis testing, and improve our ability to interpret ecological habits in the fossil record.