4.1 How do femoral shape changes correlate with locomotor habits and body size across the Triassic–Jurassic boundary? 4.1.1 Covariation between locomotor modes and body size

An increase in femoral robusticity – increase of width relative to length – correlates with an increase in body size in terrestrial tetrapods (Biewener, 1989; Campione & Evans, 2012; Carrano, 1998; Etienne et al., 2020; Mallet et al., 2020), meaning that our results enabled us to study the shape variation linked to locomotor mode and size in early archosauriforms through early crown archosaurs (Figures 3a-i, 4a,b). Moreover, our results demonstrate that the increase of femoral robusticity in early archosauriforms was coupled with the fourth trochanter being located closer to the mid-shaft (i.e., more distally located) among the most robust femora. This is typically recognized as a signal of graviportality rather than cursoriality; i.e., a morphology favoring production of greater hip joint torques rather than larger ranges of femoral motion during retraction (Carrano, 1999; Coombs, 1978; Parrish, 1986). Therefore, our results highlight a covariation between the traits linked to locomotor mode and body size among early archosauriforms (Figures 3a-i, 4a,b). Furthermore, our findings demonstrate that there was no restriction of locomotor mode depending on body size for Triassic–Jurassic archosauriforms, meaning that a bipedal habit (e.g., Table 2) was not restricted to small, more cursorial animals and that a quadrupedal habit was not exclusive to more graviportal ones (Figures 3a, 4a,b). However, the distinction between locomotor modes was clearer among robust femora than between gracile femora (Figure 3a). One of the main morphological differences between locomotor modes was femoral curvature (Figures 3g, 4b). The bipedal sauropodomorphs Plateosaurus and Mussaurus had an anteriorly bowed femur, whereas the quadrupedal aetosaurs Typothorax and Paratypothorax had a nearly straight one (Figure 3a; “Pla, Mus, Typ, Par,” 3G). Femoral curvature is negatively correlated with increased body size in dinosaurs (Carrano, 2001), and we demonstrated that this applies more broadly to archosauriforms. However, we found the opposite trend in our limited sample of crocodylian ontogeny (Figure 3a, 3g; see also Hedrick et al., 2021). Femoral curvature was enhanced plesiomorphically in avemetatarsalians with the origin of an erect limb posture and perhaps bipedalism (Hutchinson, 2001). This femoral curvature was retained by early bipedal sauropodomorphs and was subsequently lost by gigantic quadrupedal sauropods later during the Jurassic and Cretaceous (Carrano, 2001; Hutchinson, 2001). Thus, femoral curvature in our sample was mainly impacted by locomotor mode but also by body size, explaining why the distinction between bipedal and quadrupedal early archosauriforms was stronger among heavier animals than among lighter ones (Figure 3a). Nevertheless, this observation highlights an important morphological convergence in the specialization to graviportality—or at least to a greater body size—at the femoral level between bipedal avemetatarsalians and quadrupedal pseudosuchians (Figures 3a, 6). Our finding contributes to previous inferences from other skeletal elements (except for Kubo & Kubo, 2016) that avemetatarsalians were already morphologically disparate at the end of the Triassic and did not have a smaller body size in general than pseudosuchians (Brusatte et al., 2008a, 2008b; Foth et al., 2016, 2021; Kubo & Kubo, 2016; Stubbs et al., 2013; Toljagić & Butler, 2013).

4.1.2 Similar femoral disparity between avemetatarsalians and pseudosuchians

Our dataset did not enable us to compare shifts of disparity across the Triassic–Jurassic transition because it mostly included archosauriforms from the Late Triassic. However, our results demonstrated that the femoral disparity of early avemetatarsalians was as high as that of pseudosuchians among the Late Triassic archosauriforms we sampled (Figures 3a, 6). Femoral robusticity ranged from the gracile morphology of the small lagerpetid Kongonaphon—similar to the most gracile pseudosuchians of our sample (the non-crocodylian crocodylomorphs Terrestrisuchus and TTU-P11443)—to the robust morphology of the heavy bipedal sauropodomorphs Mussaurus and Plateosaurus, similar to the most heavily-built pseudosuchians Crocodylus, aetosaurs, Revueltosaurus, Riojasuchus, Nundasuchus, loricatan NMMNH P-36144, suchian NMT RB187, and phytosaurs (Figure 3a; “Kon, Ter, Crm, Mus, Pla, Cro, Typ, Par, Rev, Rio, Nun, Lor, Suc, Phy,”). Additionally, both locomotor modes were represented in the two clades (Figures 3a, 6). Even if the locomotor mode was restricted to bipedal for the most robust avemetatarsalian femora and quadrupedal for the most robust pseudosuchian femora, our results showed that some avemetatarsalians—at least Asilisaurus and Teleocrater—also were assigned to a quadrupedal locomotor mode and some pseudosuchians—at least Shuvosaurus—were assigned to a bipedal locomotor mode (Figure 3a; “Asi, Tel, Shu,” 6; Table 2). Thus, the femoral disparity of early avemetatarsalians in comparison with pseudosuchians seems higher than previously suspected in other morphological studies that included femoral characters among other bones (Brusatte et al., 2008a, 2008b; Kubo & Kubo, 2012). Indeed, our finding could indicate that femoral disparity was underappreciated by studies which showed significant differences in disparity between the two clades in the Late Triassic by relying either on the whole skeleton (Brusatte et al., 2008a, 2008b) or ratios between limb element lengths (Kubo & Kubo, 2012). Furthermore, a substantial number of studies of the disparity of pseudosuchians and non-archosaurian Archosauriformes around the Late Triassic and Early Jurassic were based on cranial characters. Hence these studies did not account for limb disparity in relative to locomotor habit and body size, which are often cited as central aspects in the faunal turnover across the Triassic–Jurassic boundary (Foth et al., 2016, 2021; Singh et al., 2021; Stubbs et al., 2013; Toljagić & Butler, 2013).

Brusatte et al. (2008a, 2008b) found that dinosaurs and ornithodirans as a whole had a lower disparity than pseudosuchians in the Late Triassic using a cladistic character dataset and a principal coordinate analysis including characters from the whole skeleton, whereas we found a similar level of disparity between these two clades when studying femoral shape variation using 3D GMM and PCA (Figures 3a, 6). Despite the inherent differences between our two approaches, here we have shown that avemetatarsalian (femoral) disparity could be enhanced by the inclusion of the early avemetatarsalians Asilisaurus and Teleocrater, as also shown by Toljagić and Butler (2013), who investigated pseudosuchian disparity using cranial characters. Asilisaurus and Teleocrater are early, possibly quadrupedal (Table 2) avemetatarsalians that were not known in 2008 (Nesbitt et al., 2011; 2017). Using the ratio between relative forelimb and hindlimb length and between metatarsal III and femur length, Kubo and Kubo (2012) measured greater morphological cursoriality among ornithodirans than pseudosuchians, mostly because the bipedal pseudosuchian Poposaurus had a lower “cursoriality index” than ornithodirans did. However, we found that Poposaurus showed a similar femoral robusticity (i.e., morphological cursoriality) to other cursorial avemetatarsalians (Figure 3a; “Pop,” 6). Moreover, we found that Shuvosaurus, another bipedal pseudosuchian, and Terrestrisuchus, a quadrupedal pseudosuchian, had a higher morphological cursoriality of the femur than Poposaurus (Figure 3a; “Shu, Ter, Pop”). The addition of Shuvosaurus and Terrestrisuchus could impact the findings of Kubo and Kubo (2012) on the relative length between the metatarsal III and femur since they were not included in their study. Nevertheless, our results indicate that the femoral morphology of Poposaurus and some specimens of Shuvosaurus was not as unambiguously bipedal as in avemetatarsalians. We inferred this result to most likely be caused by phylogenetic inertia, due to a combination of specialization to bipedalism and anatomical features specific to pseudosuchians, the vast majority of which in the Late Triassic were (plesiomorphically) quadrupedal animals. In addition, using metatarsal III and femur length, Kubo and Kubo (2012) indicated that sauropodomorphs were still more cursorial than large pseudosuchians, whereas our findings showed that larger avemetatarsalians and pseudosuchians, such as sauropodomorphs, Riojasuchus, aetosaurs, early diverging suchian NMT RB187 and loricatan NMMNH P-36144 and Nundasuchus, had similar levels of femoral specialization to body size (i.e., femoral robusticity and fourth trochanter's position; Figure 3a; “Pla, Mus, Rio, Typ, Par, Suc, Lor, Nun,” 6). These findings are somewhat incongruent but may indicate that specialization to a heavy weight similarly impacted the femoral morphology—independently of femoral length—in each clade and that the 3D morphology of metatarsal III should be investigated further.

4.1.3 Locomotor mode prediction based on femoral morphology and its evolutionary importance

Variations of femoral head rotation, shaft curvature, and fourth trochanter symmetry (as represented by PC2) in our sample were more driven by locomotor mode attribution than clade membership, even though the phylogenetic signal was significantly strong (Figures 3a,f-i, 6; Table 2). This was highlighted by the quadrupedal avemetatarsalian Teleocrater lying close to pseudosuchians in the morphospace, much as a subset of bipedal pseudoschians Shuvosaurus and Poposaurus lay close to avemetatarsalians (Figure 3a “Tel, Shu, Pop,” 6; Table 2). Therefore, 3D femoral morphology appears useful for locomotor mode estimation, especially given that (1) 93.1% of specimens accompanied by a priori knowledge of locomotor modes were correctly estimated; (2) angles associated with femoral head rotation and distal condyles (i.e., crista tibiofibularis and lateral condyle) were both significantly associated with “known” and estimated locomotor modes (Figure 5a,b; Table 2). It is generally uncommon that both a fossilized hind- and forelimb are found preserved together in Late Triassic archosauriforms and in the vertebrate fossil record in general, sometimes with little evidence that they belonged to the same individual, which is problematic for estimations based on relative length between different segments from the appendicular skeleton. Therefore, our study adds to the understanding of locomotor mode predictions based on a single limb element and provides an alternative to estimations using femoral and/or humeral minimal circumference (McPhee et al., 2018).

Interestingly, both specimens of the silesaurid Asilisaurus were estimated as quadrupedal (Figure 3a; “Asi,” 6; Table 2). However, all other Silesauridae were estimated as bipedal (Figure 3a; “Sid, Sil,” 6; Table 2). This estimation is not congruent with the previously suggested locomotor mode of Silesaurus, which was described as a quadruped based on its limb proportions and trunk length (Fechner, 2009; Kubo & Kubo, 2012; Grinham et al., 2019; Table 2), although Silesaurus was originally described as a biped (Dzik, 2003). Piechowski and Dzik (2010) and Piechowski and Tałanda (2020) speculated that occasional bipedalism was possible for Silesaurus because its center of mass was presumed to be situated near its sacrum/hips, but this has never been quantified or compared with other bipeds/quadrupeds. Hence, the locomotor mode of Silesauridae remains uncertain even though our data bring new evidence for considering the controversial question of silesaurid locomotion, as well as suggesting that further analysis of locomotion of the clade using quantitative evidence based on other osteological elements than the femur alone may be warranted.

We made a similar observation with the lagerpetid Dromomeron gregorii, to which we initially assigned an ambiguous locomotor mode; same as its smaller relative D. romerii (Nesbitt et al., 2009). We inferred D. gregorii to be a quadruped except for one (out of three; the most mature juvenile specimen TMM 31100 1308; Figure 3a; “Dro”; Table 2). In contrast, Grinham et al. (2019) assumed that this D. gregorii was a facultative biped. Postosuchus and Riojasuchus also were estimated as either bipedal or quadrupedal in our analysis depending on the specimen (Table 2). Riojasuchus was described as possibly being a facultative biped given the prominent lesser trochanter on the femur, the shortened forelimb morphology and the relative lengths of digits between hind and forelimb in ornithosuchids (Walker, 1964; Baczko et al., 2020), which may explain why our estimation was different for each specimen. Bishop et al. (2020) obtained similar results for Riojasuchus using estimated mass properties and relative hindlimb and forelimb lengths, but noted the controversial nature of this taxon's locomotor habit, whereas Grinham et al. (2019) assumed Riojasuchus to be an obligate quadruped. However, Postosuchus was described, assumed, or estimated as an obligate biped in several recent studies (Bishop et al., 2020; Grinham et al., 2019; Weinbaum, 2013).

Considering that our study did not test for facultative bipedalism, this may highlight why some taxa were misclassified. Explanations other than facultative bipedalism include phylogenetic history, with only some clades retaining a plesiomorphically quadrupedal morphology for their femora whereas other skeletal elements indicate bipedalism. Hence, when estimations of locomotor mode based on femoral morphology only are ambiguous, estimated mass properties (Bishop et al., 2020) and other bones from both girdles (Grinham et al., 2019; Kubo & Kubo, 2012; McPhee et al., 2018) and the vertebral column (Bishop et al., 2020; Christian & Preuschoft, 1996; Jones et al., 2021; Padian, 2008) should be analyzed to better characterized locomotor habit of extinct archosauriforms.

Ontogenetic differences in locomotor mode did not seem to affect femoral morphology, as suggested by both adult and juvenile specimens of Mussaurus being estimated as bipedal in our study (Table 2), contrary to the hypotheses of Otero et al. (2019), Bishop et al. (2020), and Chapelle et al. (2020), who estimated juvenile Mussaurus as being quadrupedal and adults as bipedal using mass properties and limb bone relative lengths and circumferences. Similarly, it is not possible to demonstrate a shift in locomotor mode linked with growth (assessed via a limited cross-sectional metapopulational sample) between the shortest and longest femora of Postosuchus and Riojasuchus (Tables 1, 3). However, we found that the juvenile specimens of Mussaurus, Coelophysis, and Tawa were located closer to the quadrupedal morphospace than the adult specimens (Figure S6; “Mus, Coe, Taw”). Similarly, the most mature individual of Dromomeron was closer to the bipedal morphospace than the quadrupedal one (Figure S6), leading to the most mature specimen of Dromomeron being estimated as bipedal, as discussed above in regard to facultative bipedalism (Table 2). Furthermore, we found the same ontogenetic spread along femoral specialization to locomotor mode in the extant crocodylian Crocodylus, with juvenile individuals laying closer to the bipedal morphospace than the adult one, while still being consistently estimated as quadrupedal (Figure S6; “Cro”). Hence, we infer this ontogenetic femoral disparity to be linked to a shift in how locomotor functional constraints were distributed across the appendicular skeleton toward adult stages, but not to a strict shift of locomotor mode across ontogeny. Nevertheless, our results indicate that those specimens should be investigated further using other approaches that can estimate shifts of locomotor mode and center of mass across ontogeny (e.g., Bishop et al., 2020; Otero et al., 2019), and ideally explain such shifts and locomotor function itself using fundamental biomechanical processes and mechanisms.

Regardless, our results raise an additional question prompted by available data and inferences: when did (obligate) bipedalism evolve in archosaur lineages? First, the estimation of locomotor mode regarding pterosaurs and lagerpetids, which were recently suggested to be sister taxa (Ezcurra et al., 2020), as well as silesaurids, is controversial (Padian, 1983, 2008; Grinham et al., 2019; Mazin et al., 2003; Mazin & Pouech, 2020; McCabe & Nesbitt, 2021; Piechowski & Dzik, 2010; Piechowski & Tałanda, 2020; Witton, 2015). Secondly, Lagosuchus clearly was bipedal like all early dinosaurs seem to have been (Bishop et al., 2020; Grinham et al., 2019), and Archosauria was ancestrally quadrupedal (Figure 1), with this plesiomorphic condition retained by Teleocrater among avemetatarsalians. Certainly all origin(s) of obligate bipedalism in Pseudosuchia were independent acquisitions (e.g., Bates & Schachner, 2012; Gauthier et al., 2011). Hence the above question can be reframed as, when did the dinosaur lineage first become bipedal? The ancestral locomotor mode on the avemetatarsalian lineage remains ambiguous (under maximum parsimony assumptions; see Figures 6, 8) until the Dinosauriformes (Dinosauria + Silesauridae + Lagosuchus) node. This ambiguity would be removed or reduced if some taxa with indeterminate locomotor modes were reassigned as facultative bipeds or if, as has been suggested, some Silesauridae independently reverted to quadrupedalism (see Grinham et al., 2019), which is, however, curiously contradicted by our findings for Asilisaurus being estimated as quadrupedal vs. Silesaurus being estimated as bipedal (Table 2). Our results suggest that a fresh look at the origin(s) of bipedalism within Avemetatarsalia is sorely needed through a combined approach including biomechanics, functional morphology, and phylogenetics (see also McCabe & Nesbitt, 2021).

Evolutionary history of archosauriform locomotor modes under maximum parsimony assumption: colored in blue, bipedal; gradient, indeterminate; orange, quadrupedal. Squares represent character optimizations and circles are ancestral state reconstructions. Silesauridae shown as bipedal but see text for controversy over locomotor mode(s)

4.2 Impact of locomotor mode and body size on features commonly used in cladistic analyses

We have shown that the 3D morphological variation of the femur linked to locomotor modes follows the inferred phylogeny, and that the variation linked to body size was strongly convergent between the avemetatarsalians and pseudosuchians (Figure 6). These observations enabled us to isolate which femoral characters and character states that are commonly used in archosaur phylogenetics might have homoplastic distributions corresponding to changes in body size, and also identify features that may vary more strongly with differences in locomotor mode (Figure 6).

The widening of the proximal end of the femur along the mediolateral axis was related to the variation of femoral robusticity, and influenced the medial and lateral sides of the epiphysis, but not the anteromedial and anterolateral tubers (Figures 3a-e, 4a, 6). Consequently, the posteromedial tuber appeared larger than the anteromedial tuber on the more robust theoretical shape, which is a phylogenetic character usually attributed to most crocodylians, aetosaurs, Revueltosaurus, and ornithosuchids (Nesbitt, 2011; Novas, 1996; Figure 6). These two tubera are usually coded as equal in size in sauropodomorph dinosaurs, which shared a similar level of femoral robusticity as taxa mentioned above. Hence, we consider the variation of this phylogenetic character as homoplastic because it appeared convergently in pseudosuchians and avemetatarsalians (Figures 3a-e, 4a, 6), which future studies should further analyze and consider. A similar observation is made on the medial edge of the fourth trochanter, which was rounded among robust femora and sharper on gracile ones (Figures 3a-c, 4a); this anatomical variation, used in some cladistic analyses (e.g., Bennett, 1996; Nesbitt, 2011), also appears homoplastic at least within archosauromorphs.

However, the distal ridge of the fourth trochanter had a steep slope on bipedal femora and was more symmetric on quadrupedal femora (Figures 3g, 4b). A steeper slope of this distal ridge is characteristic of almost all Triassic dinosaur clades including Herrerasaurus but not most other theropods (Langer & Benton, 2006; Nesbitt, 2011). Accordingly, such theropods did not have a distal ridge as steep as that in bipedal sauropodomorphs, suggesting a continuous trait among saurischians (Figure 3a). This asymmetry in the fourth trochanter among dinosaurs was named “semi-pendant” by Langer and Benton (2006) and interpreted as reflecting an increase in muscular stress on the distal part of the femur in early dinosaurs. A pendant fourth trochanter has long been assumed to correlate with the connection to a secondary tendon of the M. caudofemoralis longus (Dollo, 1888; Hutchinson, 2001). Its covariation with locomotor habits hints at a link with the origin of bipedalism and a more adducted, upright limb posture (Figures 3a,f-g, 4b). Ornithischians displayed a more extreme state of this morphology, with a pendant fourth trochanter that has a reversed distal slope (Dollo, 1888; Hutchinson, 2001; Persons & Currie, 2020). However, because of the gap in the Triassic ornithischian fossil record (Irmis et al., 2007; Müller & Garcia, 2020), it is difficult to investigate the evolution of this feature and how it relates to the semi-pendant state of saurischians alongside the origin(s) of bipedalism. Nevertheless, our observation that Lesothosaurus, a small ornithischian from the Early Jurassic, and the saurischians Plateosaurus, Mussaurus, and Herrerasaurus all had a similar slope of the distal ridge of the fourth trochanter, despite having pendant to semi-pendant morphologies, respectively might illuminate the evolutionary history of the fourth trochanter among dinosaurs and how the reversed-distal slope of this muscular attachment appeared. However, this character state could also be plesiomorphic because of the ancestral diapsid presence of the secondary “tendon of Sutton” (Dollo, 1888; Hutchinson, 2001). Nevertheless, our finding that the fourth trochanter might have at least two distinct components of morphological variation that are often coded in phylogenetic analyses, with the medial ridge being homologous and the distal slope homoplastic, could inspire follow-up research, including phylogenetic analyses (Figures 4b, 6).

The long axis of the femoral head was plesiomorphically more anteriorly oriented in pseudosuchians and more medially oriented in avemetatarsalians (Figures 4b, 6). This feature is known to distinguish the two clades without indicating a bipedal/quadrupedal locomotor mode, because quadrupedal dinosaurs that evolved after the Triassic–Jurassic transition did not return to the ancestral condition of an anteriorly oriented femoral head (Carrano, 2000; Hutchinson, 2001). However, we suggest that the functional significance of femoral head orientation may be underappreciated (Figures 3a,h-i, 4b, 5a). A commonly suggested functional explanation of this feature is that the anteriorly oriented femoral head correlates with a (plesiomorphically) more sprawled hindlimb posture and rotary gait whereas a medially oriented femoral head evolved in lineages having a more erect (adducted) limb posture and parasagittal gait (Bonaparte, 1984; Carrano, 2000; Charig, 1972; Demuth et al., 2020; Hutchinson, 2001, 2006). Our study did not address the difference of postures between sprawling to erect, and some archosauriforms were not “fully erect.” Thus, we considered a continuum in postures between sprawling to erect (e.g., see Gatesy, 1991 and Hutchinson, 2006) and a more erect limb posture as a prerequisite to bipedalism in both archosaur clades. In addition, the variation of femoral head orientation demonstrated that the bipedal pseudosuchian Shuvosaurus and the potentially bipedal pseudosuchian Postosuchus have a more medially oriented femoral head than other pseudosuchians, which were quadrupedal (or controversially so) (Figure 3a; “Shu, Pop,” Figures 3h-i, 4b, 5a, 6; Tables 2, S3). This was not the case for the bipedal pseudosuchian Poposaurus (Figure 3a; “Pop,” Figure 6). However, both Poposaurus specimens were close to the “least” bipedal femur of Shuvosaurus. We made the same observation with the quadrupedal (or potentially so) avemetatarsalians Teleocrater, Asilisaurus, and Dromomeron gregorii having a more anteriorly oriented femoral head than clearly bipedal avemetatarsalians (Figure 3a; “Tel, Asi, Dro,” Figures 4b, 6; Tables 2, S3). Thus, femoral head orientation could be even more closely related to locomotor mode and kinematics than previously thought. Analyses of joint mobility (e.g., Demuth et al., 2020) could test this possibility further.

We infer that the lesser trochanter is less important than femoral head orientation or bone curvature in the estimation of early archosaur locomotor modes. We showed that the lesser trochanter was more expanded proximo-anteriorly among bipedal archosaurs and most avemetatarsalians (Figures 3g-h, 4b, 6). A well-developed lesser trochanter evolved independently in different clades of dinosaurs and has been suggested to correlate with bipedalism, as it could allow a greater protraction and retraction of the hindlimb (Carrano, 2000; Gauthier, 1986; Novas, 1996). However, a lesser trochanter is absent in the bipedal pseudosuchians Shuvosaurus and Poposaurus (Nesbitt, 2007; Schachner et al., 2019), supporting the inference that this feature appeared only in bipedal avemetatarsalians, with parasagittal gait as a prerequisite (Carrano, 2000). Moreover, one specimen of Riojasuchus, perhaps a facultatively bipedal ornithosuchid (Baczko et al., 2020), resembled more quadrupedal pseudosuchians whereas the other specimen was closer to bipedal archosaurs, such as members of Poposauridae, in the morphospace, despite having a proximo-anteriorly developed lesser trochanter (Figure 3a; “Rio, Pop”).

An anteriorly bowed (curved) femur is also a character used to distinguish the in-group relationships within archosaurs (Figures 3a, 4b, 6; Sereno, 1991), varying according to locomotor habit (Figure 3g) and we infer this feature to be a reliable predictor of locomotor mode. Femoral bowing is known to vary across all archosauriforms (Gauthier et al., 1988, more specifically with the origin of a more erect posture (Hutchinson, 2001) and body size variations (Biewener, 1983; Carrano, 2000). This feature is suggested to better predict mechanical bending stress related to a bipedal locomotor habit (Hutchinson, 2001), whereas a straightening of the shaft correlates with increased body mass in quadrupedal animals, except across crocodylian ontogeny (Biewener, 1983; Carrano, 2000, Hedrick et al., 2021; our results for Crocodylus). However, large bipedal archosaurs, such as sauropodomorphs and theropods, retained an anteroposteriorly bowed femur (Hutchinson, 2001). Thus, the variation of this feature has a strong functional implication and might be well suited to predict archosaur posture and locomotor mode, especially because bipedal pseudosuchians are thought to have had a more erect hindlimb posture similar to that of bipedal ornithodirans (Figures 3a, 4b, 6; Bates & Schachner, 2012; Nesbitt & Norell, 2006).

The angle between the lateral condyle and the crista tibiofibularis (Figure 2a) is known to distinguish ornithosuchids, aetosaurs, Revueltosaurus, phytosaurs and most avemetatarsalians with a rather obtuse angle from other archosauriforms such as Postosuchus, poposauroids, and crocodylomorphs, with a rather right angle (Nesbitt, 2011; Parker & Irmis, 2005; Parrish, 1986). However, our study demonstrates that, despite varying continuously rather than in a discrete manner, this angle is greater (more obtuse) in quadrupedal archosauriforms (most pseudosuchians) than in bipedal ones (most avemetatarsalians, Figures 3a, 4b, 5b; Table S3). Moreover, the phylogenetic signal of the variation between locomotor modes shows that the mean angle of this feature is significantly greater (more obtuse) in pseudosuchians than in avemetatarsalians (more acute), with exceptions for some taxa (Figures 3a, 5b, 6); contradicting prior ideas that avemetatarsalians, some pseudosuchians and phytosaurs shared the same angle (Nesbitt, 2011; Parker & Irmis, 2005; Parrish, 1986). Furthermore, the acuteness of this angle may be increased by the presence of a groove between the crista tibiofibularis and the lateral condyle among bipedal (or controversially so) poposauroids, dinosauromorphs, and Postosuchus, whereas this groove is absent in quadrupedal (or controversially so) phytosaurs, Euparkeria, aetosaurs, Revueltosaurus and Riojasuchus (Nesbitt, 2011). One possible explanation would be that the preservation of this groove may vary between specimens, subsequently affecting the acuteness of the angle, especially because of its proximity to the cartilaginous epiphyseal cap, which is not always well preserved in extinct archosauriforms (Bonnan et al., 2010; Holliday et al., 2010; Nesbitt, 2011). We did not observe a variation of the proximodistal width of bone epiphyses, at least not along the two first PC axes (Figures 3a-i, 4), meaning that the variation in the amount of preserved distal articular cartilage does not directly explain the morphological variation shown in our study, even if it may be visible among other PC axes. Furthermore, the potential link between the acuteness of the angle and the presence of a groove between the crista tibiofibularis and the lateral condyle raise the issue of correlation between these two characters in phylogenetic analyses. Thus, despite the potential impact of taphonomic factors, the angle between the lateral condyle and the crista tibiofibularis should be investigated further in order to better understand its evolutionary history and functional implications such as knee joint mobility and orientation and should be better integrated into phylogenetic studies.

4.3 Size effect and crocodylian ontogeny

We found that an evolutionary allometric relationship of increasing femoral robusticity and centroid size was significant, but weak. This effect was due to bipeds and quadrupeds having different allometric trajectories involving an increase of femoral robusticity (Figures 3a-e, 7a). This effect was already described before and is known to intensify with the phylogenetic breadth of a sample (Adams et al., 2013; Klingenberg & Froese, 1991; Mitteroecker et al., 2004). We show that, when accounting for groups (bipedal/quadrupedal), the correlation between the increase of femoral robusticity and centroid size was significant and strong among bipedal archosauriforms, but not among quadrupedal archosauriforms. Centroid sizes among the most robust quadrupedal femora were lower than among the most robust bipedal ones, with more scattered values along the morphological variation, whereas values were similar for the most gracile bones among the two groups (Figure 7a). We showed that this pattern does not result from the presence of juvenile crocodylians in the sample, because Revueltosaurus and Parringtonia also had low centroid sizes with highly robust femora (Figure 3a “Cro, Rev, Par,” Figure 7a, Figure S6). In addition, Dodson (1975) and Hedrick et al. (2021) demonstrated that the femoral robusticity in Alligator mississippiensis varied significantly along ontogeny, with the fourth trochanter migrating down the shaft toward the adult age, along with an increase of femoral disparity. This morphological variation is identical to the one we highlighted along the specialization to body size. Yet, we did not observe a separation between juveniles and the adult specimen of Crocodylus along the increase of femoral robusticity (as in PC1; Figure 3a; Figure S6; “Cro”), subsequently indicating the rather conservatively high robusticity in crocodile femora across ontogeny when compared with a larger taxonomic sample. However, we did observe a separation between juveniles and adult Crocodylus along the axes pertaining to femoral specialization to locomotor mode (as in PC2; Figure 3a; “Cro”). Thus, juvenile Crocodylus had straighter femora (i.e., lower anterior curvature) than the adult specimen, which is congruent with findings described under “femoral robusticity” in Alligator by Hedrick et al. (2021). Morphological variation of the femur that seemed to indicate a shift of estimated locomotor mode from bipedal to quadrupedal across ontogeny was also observed in extant crocodylians by McPhee et al. (2018; Caiman) and Bishop et al. (2020; Alligator). There is no known bipedalism in Crocodylia, even early in posthatching ontogeny, so these results are all anomalous in terms of identifying locomotor mode. The main difference we have highlighted with previous studies is that femora of juvenile Crocodylus showed a higher degree of specialization to a quadrupedal locomotor mode than adults (Figure 3a; Figure S6; “Cro”). This opposite trend in results could be explained by findings from Ijima and Kubo (2019) who recently discussed that growth parameters and variation of limb morphology along ontogeny were highly variable across the various extant clades of crocodylians, indicating that a trend observed in Crocodylus ontogeny may not be necessarily attributable to Alligator nor Caiman.

4.4 Convergence between semi-aquatic lifestyle and specialization to heavy weight

Our results showed that phytosaurs and the extant pseudosuchian Crocodylus shared a similar femoral morphology both in term of robusticity and specialization to locomotor habit (Figure 3a; “Cro,” Figure 6; Table 2). This morphological convergence was already described at the level of whole-animal morphology and suggested as an adaptation to a similar semi-aquatic lifestyle (Lautenschlager & Butler, 2016; Stocker & Butler, 2013). However, we showed that other taxa with robust femoral morphology and a probable terrestrial lifestyle, such as aetosaurs, Revueltosaurus, ornithosuchids, and non-sauropod sauropodomorphs, were recovered close to Crocodylus and phytosaurs in the morphospace, highlighting a morphological convergence between adaptations to weight support and a secondary semi-aquatic lifestyle, with similarly enlarged epiphyseal width and a rounded fourth trochanter near the mid-shaft; independently of locomotor habit (Figure 3a-e; “Typ, Par, Rev, Rio, Pla, Mus,” Figure 4a). Such morphological convergence between heavy terrestrial and semi-aquatic quadrupeds is intriguing because it evolved to serve opposite functions in relation to environmental factors (e.g., buoyancy in a low gravity environment vs. improved resistance to gravitational forces on land) and has already been studied among some massive animals through a microanatomical approach (Houssaye et al., 2016, 2021). We suggest that this convergence should be investigated further in archosauriforms using a similar approach coupled with biomechanical analyses in order to decipher specializations between these two functional constraints which seemed to drive the appearance of convergent femoral morphologies.