Reassessment of the Triassic archosauriform Scleromochlus taylori: neither runner nor biped, but hopper

Skull

A detailed skull reconstruction was prepared based primarily on NHMUK R3146A and R3556 (Fig. 18). Note that Fig. 2 illustrates previous reconstructions of Scleromochlus’s skull, all scaled to uniform length, so as to facilitate comparisons. NHMUK R3146, R3556 and R3557 preserve relatively complete skulls that demonstrate the general shape and structure of the skull, but the upper jaw tip, narial region, palate, and occiput are incompletely known or unknown. The skulls demonstrate that in dorsal view there was a broad temporal region, a slight convex curve in the orbital region, and a straight taper from the orbital region to the jaw tip. NHMUK R3146A seems to demonstrate that the anterior end of the upper jaw curved with a small radius to a pointed tip, markedly different from that which Benton (1999) reconstructed. The right premaxilla of NHMUK R5589, upon which Benton seems to have based his interpretation of a broadly rounded premaxillary tip, does not preserve a finished anterolateral margin of the premaxilla and so does not provide evidence of a rounded tip, and likewise the upper jaw tips of NHMUK R3146B, R3556 and R3557 all appear truncated or broken and do not preserve finished tip margins. There is also no evidence of a concave jaw margin lateral to the antorbital fenestra, and the concave curve reconstructed by Benton may be resulted from combining a broadly rounded premaxillary tip with the more posterior parts of the skull. The straight taper of the upper jaw reconstructed here would have resulted in a premaxillary-maxillary tooth row that was closely aligned with, and just lateral to, the dentary tooth row.

No specimen preserves evidence of external nares. The suggestion of Benton & Walker (1985) that the naris had a lateral flange as a specific adaptation to living in sandy environments cannot be supported, whereas the suggestions of Woodward (1907) and Benton (1999) that the external nares were small and terminal are probably correct because evidence of the nares probably would be present in the preserved skulls if the nares were larger or more posteriorly placed.

The skulls of NHMUK R3146, R3556 and R3557 demonstrate that the antorbital fenestra was oval to subtriangular with a rounded anterior margin, and that of NHMUK R3146A demonstrates that its anterior margin was bounded by a prominent ridge. The shape of the posterior margin is unclear because the preorbital bar consisting of prefrontal and lacrimal is broken in all specimens. In some specimens there may be traces of bone preserved within the antorbital fenestra, which Sereno (1991) and Benton (1999) interpreted as representing an antorbital fossa; however, neither Sereno nor Benton provided any evidence or argumentation that the bone traces within the antorbital fenestra are internal portions of the maxilla or nasal forming an antorbital fossa. Note that although there is little evidence of the palate other than traces of the vomers, pterygoids, and possible palatine fragments, Benton (1999; Fig. 2G) reconstructed the palate as having very large internal nares that extended as far back as the posterior margin of the antorbital fenestrae. If the external nares were small and terminal, then there is no reason to think that the internal nares would be so extensive. In the end, I can find no evidence that bone traces within the antorbital fenestrae are not palatal elements, and thus no evidence that there were antorbital fossae.

The orbits, as preserved in NHMUK R3146 and R3556, appear as anteroposteriorly elongate ovals in dorsal view. The orbit also appears oval in lateral view in the skull reconstruction (Fig. 18), in which the dorsoventral diameter of the orbit between the narrow frontals forming the dorsomedial margins and the laterally placed jugals forming the ventrolateral margins is roughly equivalent to the anteroposterior diameter such that the orbits would have been subcircular and faced dorsolaterally. Dorsoventral compression of the skull would have distorted the dorsolaterally facing subcircular orbits into anteroposteriorly elongate ovals.

The orbit and its eyeball were large relative to the skull. That might suggest that Scleromochlus specimens were immature given the marked negative allometry of orbit and eyeball size relative to skull size common in diapsids. However, no specimen seems to exhibit size independent indicators of skeletal immaturity (e.g., unfused bones, incompletely ossified bones with simple shapes; Bennett, 1993, 1995), and the bones of the pelvic girdle of NHMUK R3557 appear fused and even the small distal tarsals of NHMUK R3556 seem to be well ossified. Therefore, all Scleromochlus specimens are interpreted as mature adults, in which case although the sclerotic ring is not known the large orbit and presumably large eyeball suggest that Scleromochlus could have been crepuscular (Schmitz & Motani, 2010).

Woodward (1907) and Sereno (1991) reconstructed sizeable upper temporal fenestrae within the area identified here as paired parietals, but I can find no evidence of such fenestrae. Benton & Walker (1985) and Benton (1999) reconstructed slit-like upper temporal fenestrae just below the plane of the parietals; however, the preserved shape of the left squamosal of NHMUK R3146A suggests that the upper temporal bar separating the upper and lower fenestrae was close to the inferolateral ends of the preserved squamosal. The unidentified displaced element on the left of NHMUK R3146A is interpreted here as part of the upper temporal bar.

Benton (1999) reconstructed the basioccipital of Scleromochlus as extending posterior to the posterior margin of the parietals as can be seen by combining his dorsal and palatal views of the reconstructed skull (Fig. 19). However, combining information from the dorsal and ventral slabs of NHMUK R3556 provides no evidence that the basioccipital was in such a position, and neither do other specimens, and NHMUK R3556 suggests that the basisphenoid was angled anteroventrally and the basioccipital extended only slightly behind the parietals.

NHMUK R3146A demonstrates that the quadrate region of Scleromochlus formed a roughly right angled posterior corner of the skull in dorsal view. The quadrate extended dorsomedially to contact the squamosal extending ventrolaterally from the broad parietals. Benton (1999) reconstructed the quadrate as slanting anterolaterally in dorsal view and anteroinferiorly in lateral view, presumably based on the anterolaterally slanting left posterior margin of the skull of NHMUK R3146B on the dorsal slab. However, that slanting margin seems to be an artifact, and the right quadrate and quadratojugal of NHMUK R3146A and the right quadrate and quadratojugal of NHMUK R3146B are at nearly right angles with the jaw margin in dorsal view such that the quadrate would have been roughly vertical in lateral view.

NHMUK R3146 and R3556 demonstrate that the mandibular rami were slender and straight anteriorly and met in a short symphysis that tapered to a sharp point in ventral view and seems to have tapered upward to the point in lateral view. Combining information from the dorsal and ventral slabs of NHMUK R3146A shows that the posterior parts of the rami curved medially to some extent and the jaw articulations were near their posterior ends with only short retroarticular processes (Fig. 6). Benton & Walker (1985) and Benton (1999) reconstructed the mandible with long retroarticular processes, which would suggest that Scleromochlus exhibited powerful depression of the mandible, but neither offered an explanation as to why Scleromochlus might have had a long retroarticular process or powerful depression. The more posterior position of the jaw articulation reconstructed here would mean that the upper and lower jaws were effectively longer and the gape greater for a given angle of depression than if the jaw had been shorter.

Benton (1999) reconstructed the mandible as shallow anteriorly but deeper posteriorly with a convex dorsal margin beneath the orbit, and characterized it as “slipper-shaped,” whereas Benton & Walker (1985) and Sereno (1991) reconstructed the mandible as shallow in lateral view. Benton’s (1999) interpretation seems to have resulted from misinterpretation of the upper and lower jaws of NHMUK R3556 on the ventral slab (Fig. 8), whereas that specimen is interpreted here as providing no evidence of a deep posterior mandible. Note that the right upper and lower jaws of NHMUK R3146B on the ventral slab present a similar appearance to that of NHMUK R3556 in that the upper jaw lies lateral to the ramus and the two might be misinterpreted as evidence of a broad ramus in lateral view (see Benton, 1999: fig. 2b).

Huene (1914) and Benton (1999) reconstructed an external mandibular fenestra and Benton & Walker (1985) illustrated one with a dashed outline. Neither Huene nor Benton & Walker presented evidence of a fenestra and Benton’s (1999) interpretation of a possible fenestra seems to have been based on the right ramus of NHMUK R3557 on the ventral slab. However, based on my examination of NHMUK R3556 and R3557, neither specimen provides evidence of a fenestra, for example, finished edges on the bones that would surround a fenestra, and there is no reason to think that fenestrae would have been present in the shallow rami.

NHMUK R3146A and R3557 preserve traces of teeth. NHMUK R3146A exhibits three premaxillary alveoli, but probably had two additional alveoli anteromedial to the three, which agrees with the five premaxillary teeth reconstructed by Benton & Walker (1985) and Sereno (1991). NHMUK R3146A also exhibits seven maxillary alveoli, whereas NHMUK R3557 seems to have had eight maxillary teeth. Thus, Scleromochlus had at least 13 teeth in the upper jaw, but it might have had more. Huene (1914) reconstructed 14 and 13 teeth in the upper and lower jaws, respectively, Benton & Walker (1985) 18 and 16, Sereno (1991) 20 and 16 and Benton (1999) 15 each. The best preserved tooth of NHMUK R3146A seems not to have been fully erupted, but if reconstructed with the widest part of the crown slightly above the margin of the alveolus, then the length of the tooth probably would not exceed its diameter, which would agree with the teeth of NHMUK R3557 (see Benton, 1999: fig. 10d). NHMUK R3146A suggests that there may have been variation in the size of premaxillary alveoli and so variation in the diameter and perhaps the length of the premaxillary teeth, but the maxillary alveoli vary little suggesting little to no variation in maxillary tooth size. Sereno (1991) and Benton (1999) reconstructed the teeth as long and posteriorly curved like theropod teeth, but there is no evidence of curvature or marked lateral compression or evidence that the teeth were as long as reconstructed by those authors. Given the small size of the teeth, they probably would have been suited to holding small invertebrate prey before swallowing whole, but would not have been suited to biting off pieces of large food items or processing food items in the mouth before swallowing.

Reconstructions of the skull of Scleromochlus show it as strongly dorsoventrally compressed with height:width ratios ranging from 0.5 to 0.62 (Fig. 2; Supplemental Information). Dorsoventrally compressed skulls are found in terrestrial vertebrates that exhibit sprawling postures and often bring the skull into contact with the substrate, for example, frogs and lizards, whereas erect sauropsids that hold their skull above the substrate do not have dorsoventrally compressed skulls, for example, theropods with height:width ratios >1.15. The strongly dorsoventrally compressed skull of Scleromochlus is inconsistent with an erect posture, but is consistent with a sprawling posture.

Vertebral column and ribs

Huene (1914) and Benton (1999) reconstructed the vertebral series as consisting of eight cervical vertebrae and 12–13 and 16–17 dorsal vertebrae, respectively (Table 3). NHMUK R3146A & B provide information as to the length of the presacral vertebral column, but only NHMUK R3556 preserves more than short isolated series of vertebral centra and the number of dorsal vertebrae is unclear. Here the cervical series is interpreted as consisting of seven vertebrae based on NHMUK R3556, which preserves seven cervical vertebrae in articulation with the basioccipital anteriorly and dorsal vertebrae posteriorly, and in which the positions of the right coracoid on the ventral slab and anterior end of the left scapula on the dorsal slab mark the position of the cervico-dorsal transition. The cervicals have anteroposteriorly broad neural spines, with that of the axis being larger than the subsequent cervicals. NHMUK R3146A and B also seem to have seven cervicals based on the relative positions of the posterior skull margin and the cervico-dorsal transition.

The cervical series of NHMUK R3556 suggests a length of ~2.2 mm/segment, and one cervical centrum of NHMUK R3146A on the ventral slab has a length of ~2.3 mm. NHMUK R3146A does not exhibit articulated anterior dorsal centra, but the length from the 1st to 8th dorsal ribs on the articulated left side of the ribcage on the dorsal slab suggests a length of 2.8 mm/segment. A series of three dorsal vertebral centra of NHMUK R3914 on the ventral slab, which are interpreted as dorsals 11–13, has a length of 3.1 mm/segment, and a series of three mid-dorsal vertebral centra of NHMUK R3557 on the ventral slab has a length of ~3.5 mm/segment. However, the three dorsal vertebrae of NHMUK R3557 just anterior to the sacrum on the dorsal slab have a length of ~2.6 mm/segment, and the three sacral vertebrae of NHMUK R3146A on the dorsal slab have a length of ~2.3 mm/segment. The various measurements suggest that the anterior dorsal vertebrae were longer than the cervicals, posteriormost dorsals, and sacrals. If one reconstructs the dorsal vertebral series as 49 mm long from the cervico-dorsal transition to the first sacral as in NHMUK R3146A, and assumes that the anterior 13 dorsal vertebrae had an average length of 2.8 mm/segment based on the spacing of the anterior eight ribs, then there would be 12.6 mm remaining for a series of posterior dorsals gradually decreasing in length to meet the ~2.3 mm long sacrals. The average segment length of the series would be ~2.5 mm, and so there would be five such dorsals, bringing the number of dorsals to 18, whereas Benton (1999) reconstructed the dorsal vertebral series as ~42 mm long and consisting of 17 vertebrae.

Woodward (1907), Huene (1914) and Benton (1999) reconstructed the sacrum as consisting of four vertebrae. Only NHMUK R3146B preserves articulated sacral vertebrae and ilia in dorsal view such that the articulation can be observed, and it clearly shows there are at least three sacral vertebrae. However, the ilia do not seem to be long enough to fully accommodate articulation with a fourth sacral. Similarly, the positions of the ilia and adjacent vertebral centra on NHMUK R3556 suggest that there were only three sacral vertebrae. Therefore, I interpret Scleromochlus as having only three sacrals.

No specimen preserves a complete caudal series. I accept Benton’s (1999) interpretation of the preserved caudal vertebrae as indicating a minimum of ~35 vertebrae, but can find neither evidence that tail was not longer than 35 vertebrae nor evidence of terminal caudal vertebrae. Benton (1999) reconstructed the caudal vertebral series as ~80 mm long, whereas my measurements of NHMUK R3557 suggest that the tail was at least 90 mm long and may have been longer.

Woodward (1907: fig. 3b) and Huene (1914: fig. 9) described and illustrated disarticulated chevrons no longer than the mid-caudal vertebral centra on NHMUK R3557. Both described them as delicate, though Huene stated that Woodward illustrated them as too massive. I saw no evidence of chevrons on any specimen. However, I accept that Woodward and Huene saw small chevrons on NHMUK R3557, and suspect that the impressions were degraded by the time Walker made the PVC casts, indeed Huene’s description that they were more delicate than Woodward illustrated might have been based on observations of partially degraded impressions. In his skeletal reconstruction, Huene (1914), who illustrated the tail with 62 caudal vertebrae, illustrated the 24th caudal vertebra as bearing the last chevron, and illustrated caudal neural spines as small and virtually absent by the 40th vertebra. He commented that transverse processes were absent in mid-caudal vertebra, and both he and Woodward (1907) described the tail as slender. I concur that the tail probably was rather slender, and although it probably could have been used as a dynamic stabilizer of body orientation it probably did not have enough mass to be an effective counterbalance for the head, neck, and trunk when in a static bipedal pose.

The dorsal ribs of NHMUK R3146A extend laterally up to 9.4 mm from the midline, curve slightly posteriorly, diverge from one another laterally and exhibit a gentle posterior curvature reminiscent of the ribcage of Phrynosoma cornutum (Hodges, 2001) and seem to have curved ventrally only near their lateral ends. The dorsal ribs of NHMUK R3914 extend up to 8.8 mm from the midline and that of the 13th vertebra was at least 7 mm long. NHMUK R5589 preserves an isolated dorsal vertebra and rib in articulation in posterior view, which shows that the rib was straight and extended laterally 9.7 mm from the midline without significant ventral curvature. There is no evidence that the straightness of that rib resulted from compression, and combining that information with the gentle posterior curvature (~17 mm radius) of the ribs of NHMUK R3146A suggests that the dorsal ribs were straight in anterior view and curving posteriorly in dorsal view for much of their length. In addition, the isolated ribs of NHMUK R3146A and R5589 (Fig. 7) exhibit similar curvatures (~17 mm radius) and do not suggest that the ribs were curved in posterior view in addition to the curvature in dorsal view. Ribs are preserved on most specimens, but I found no examples with radii of less than ~10 cm and no ribs curved ventrally to a significant degree so as to result in a rounded rather than dorsoventrally compressed trunk. The smaller radius of a few ribs may reflect the transition between the neck and trunk. Ribs straight in anterior view would have resulted in a broad, flat dorsal surface, and so the trunk presumably was dorsoventrally compressed. One might argue that the lack of curvature resulted from crushing of the specimen, but crushing could not explain the fact that the isolated rib of NHMUK R5589 preserved in posterior view was straight in posterior view or the fact that the radii of the isolated ribs of NHMUK R3146B and R5589 ranged from 10 to 17.5 mm. The dorsoventrally compressed ribcage, like the dorsoventrally compressed skull, is inconsistent with erect posture, but is consistent with sprawling.

Osteoderms

The posterior trunk of NHMUK R3146A preserves transverse striations visible under low angle illumination, which Woodward (1907) and Huene (1914) interpreted as gastralia. Benton (1999: p. 1433; see also Benton & Walker, 1985: p. 213) argued that they were dorsal structures that lay over the vertebral column and interpreted them as “probably very thin scutes or horny scales.” I concur that they were dorsal, but it is not clear what Benton meant by scute or scale because although he did not label the structures in his figure 2, the similar structures of NHMUK R3557 and R4823 in his figures 4 and 6 were labeled as osteoderms and yet his cladistic analysis coded Scleromochlus as lacking osteoderms. Furthermore, Romer (1956) used scute and scale to refer to both cornified epidermis and dermal bone, and both cornified epidermal shields of the chelonian shell and dermal bones of crocodilian dorsal body armor are commonly referred to as scutes. It is unlikely that the structures in Scleromochlus were cornified epidermis because they occur on four of the six specimens of Scleromochlus, whereas keratinized structures (e.g., claw sheathes of unguals, epidermal body scales) seem not to have been preserved in any specimens of Scleromochlus or other taxa from the Lossiemouth Sandstone. Therefore, the structures are interpreted as osteoderms, i.e., bony structures formed within the dermis.

The best preserved articulated series of osteoderms is that of NHMUK R4823 in which there are bilateral pairs of osteoderms with each element having an anteroposterior length of ~1.15 mm and a transverse width of ~7.3 mm, thus roughly two segments of osteoderm per vertebral segment. I found no evidence of additional osteoderms lateral to the bilateral pair. NHMUK R3146A and R4823 demonstrate that the osteoderms were closely spaced with no gap evident between one segment and the next. NHMUK R3557 demonstrates that the osteoderm elements were firmly connected to one another because it preserves a series of osteoderms still in tight articulation even though significantly displaced to the side in a disturbed area.

The osteoderms seem to have been restricted to the posterior half of the trunk. The anterior margin of the osteoderms of NHMUK R3146A lies behind the 7th dorsal rib, roughly in line with the knee when the hindlimb is folded at rest. In NHMUK R4823, it is not possible to count dorsal vertebrae or ribs, but based on the position of the anterior margin of the preserved osteoderms relative to the positions of the humerus and knee on the ventral slab, it appears that the anterior margin is in a similar position. In NHMUK R3557 an articulated series of osteoderms lies more anteriorly on the ventral slab, but it is displaced to the left within the area of disturbance and so does not provide clear evidence of a more anterior extent.

The osteoderms covering the posterior trunk formed a structure here termed a culet (in armory, a culet is a piece of plate armor consisting of small, horizontal plates or lames articulated so as to protect the small of the back or buttocks). Although the culet could have had a protective function, there is no evidence that one segment could normally slide over another and so the firmly connected osteoderms would have severely limited lateral flexion of the posterior trunk such that the culet probably was responsible for the reduced lateral flexion of the lumbar vertebral column evident in some specimens.

Pectoral girdle and forelimb

NHMUK R3914 preserves the best example of the pectoral girdle of Scleromochlus, which consists of a long scapula and a robust coracoid (Figs. 14 and 15). The scapula is strap-like, ~12.1 mm long, and lies more or less parallel to the vertebral column with its anterior end curving laterally toward the humerus. NHMUK R3146A exhibits a comparable scapula lying parallel to the vertebral column, but its anterior end curves medially and the proximal ends of the humeri are quite close together, probably because dorsoventral compression drove the glenohumeral joints medially. The right coracoid on the dorsal slab of NHMUK R3914 disappears into the medial wall of a depression in the matrix without any evidence of a termination medial to the depression, and the small crack passing medially from the lateral end of the coracoid on the ventral slab might be an indirect trace of an elongate medial part of the coracoid. Combining information from the dorsal and ventral slabs shows that the coracoid had a robust lateral end and was at least ~5.8 mm long (Fig. 15), thus ~52% of scapula length even if the crack on the ventral slab and the transversely oriented subcylindrical mass had nothing to do with the coracoid. The shoulder joint had a large preserved range of motion, ~150° in the horizontal plane, and there is no reason to think that motion would have been restricted in the vertical plane.

The forelimb, though much shorter than the hindlimb, was long relative to the body. The humerus had a prominent proximally positioned deltopectoral crest. Although impressions of the humerus on several casts suggest that the humeral shaft was rather slender, they probably do not reflect the full diameter of the shaft, and the impression of NHMUK R4824 on the dorsal slab shows the shaft diameter was ~1.2 mm (compare to tibia and fibula midshaft diameters of NHMUK R3556 of 1.75 and 0.80 mm, respectively). Like the humeral impressions, impressions of the radius and ulna may not reflect the full diameters of their shafts, but the impression of NHMUK R4824 on the dorsal slab shows the ulnar shaft was ~0.95 mm in diameter with rather thick walls. The elbow’s preserved range of motion is from fully flexed to ~87°, but it seems likely that it could have been extended considerably further. The manus was small, and it seems unlikely that it could have reached the mouth to pass food items to the mouth or assist in manipulating food, so the forelimb presumably was used primarily for locomotion.

Pelvic girdle and hindlimb

The pelvic girdle was shallow with rather short processes and an imperforate acetabulum. The longest ilium preserved is that of NHMUK R3556, 8.4 mm long, and the ilia are shorter than the series of four adjacent vertebrae in all specimens. Benton (1999: fig. 4C; see also Fig. 11A) illustrated the pelvic girdle of NHMUK R3557 as preserving a pubis and ischium that extended ~4.3 and ~7.9 mm from the center of the acetabulum, respectively. Such a pelvic girdle with ischium longer than pubis is similar to that of Euparkeria, whereas Benton’s (1999: fig. 14; see also Sereno, 1991: fig. 18c) skeletal reconstruction shows a theropod-like pelvic girdle with pubis and ischium extending ~8.5 and ~6.8 mm from the center of the acetabulum, respectively. I found no evidence of a long pubis in any specimen, and a long pubis would be inconsistent with the dorsoventrally compressed trunk.

The femur of NHMUK R3557 has a weakly inturned hemispherical femoral head with the articular surface covering much of the proximal end, which combined with the imperforate acetabulum probably would not have permitted the femur to swing in a parasagittal plane, and even if the hip joint could be adducted to lie close to a parasagittal plane, further adduction for control movements would be impossible and the joint would be unsuitable for erect locomotion. Even if one assumed that the hip joint morphology was not a hindrance to such movement, the prominent tubercle on the pubis would have been. Instead, the hip morphology suggests the femur could be flexed anteriorly, extended posteriorly, and abducted laterally to a considerable degree (~68° in NHMUK R3146B), but could not be adducted to swing fore and aft in a parasagittal plane. Thus, the hip joint and femur were not suited to erect bipedal locomotion with parasagittal gait, but sprawling to semi-erect postures were possible. The knee joint consists of a distal femur with two prominent condyles separated by an intercondylar groove and a rather flat proximal tibia, and has been interpreted as a hinge joint that allowed flexion and extension without significant rotation (Padian, 1984; Benton, 1999). I agree with that interpretation. The preserved range of motion is from fully flexed to ~68°, but there is nothing in the morphology of the distal femur and proximal tibia to suggest that the knee could not be extended to ~180°, such that the knee would be suited to be both leaping and erect running.

The interpretation of the tarsus has been critical to the interpretation of the locomotion and relationships of Scleromochlus, and the two right tarsals of NHMUK R3557 preserved lying in contact with one another on the ventral slab have been central to the interpretation of the tarsus. Both Woodward (1907) and Huene (1914) illustrated the upper, smaller of the two and interpreted it as a calcaneum with prominent tuber, and Martin (1983) concurred. Padian (1984) illustrated that tarsal and the larger tarsal of NHMUK R3556 and interpreted them as lateral and medial distal tarsals, respectively, like those of pterosaurs, but seems to have overlooked the small distal tarsals of NHMUK R3556 that can be seen on the polyurethane cast that he made. Sereno (1991) illustrated both tarsals of NHMUK R3557 and interpreted the larger as distal tarsal 3 and the smaller as distal tarsal 4, but also seems to have overlooked the small distal tarsals of NHMUK R3556 and did not comment on the larger tarsal of NHMUK R3556 figured by Padian, which should have been visible on Sereno’s latex peel because it was made after Padian’s polyurethane cast. Benton (1999) described distal tarsals 1–4 of NHMUK R3556 and so interpreted the two tarsals of NHMUK R3557 as proximal tarsals, but did not comment on Padian’s illustration of the larger proximal tarsal of NHMUK R3556 and reversed Woodward’s and Huene’s identifications based on comparison to the proximal tarsals of Lagosuchus (Sereno, 1991: fig. 9). Benton (1999: fig. 13d) illustrated the two proximal tarsals of NHMUK R3557 as of nearly the same size, and Sereno (1991: fig. 17) illustrated the larger medial tarsal as only slightly larger than the smaller. However, comparison of the tarsals of NHMUK R3557 with the larger tarsal of NHMUK R3556 reveals that both Sereno and Benton overlooked the pointed tip of the larger tarsal of NHMUK R3557, i.e., the triangular feature in Fig. 12B extending toward the bottom of the figure from beneath the body of the smaller tarsal. Therefore, the larger proximal tarsal was roughly twice the size of the smaller.

The medial elements of the crus and tarsus of basal archosauriforms were significantly larger than the lateral elements, see for example Proterosuchus, Euparkeria, Rutiodon, Riojasuchus and Lagosuchus (Sereno, 1991; figs. 3, 4, 6, 7 and 9), with the tibia significantly larger than the fibula and the astragalus significantly larger than the calcaneal body from which the tuber, if present, projected. In Scleromochlus, too, the tibia was significantly larger than the fibula, i.e., in NHMUK R3556 tibial and fibular shaft diameters are 1.75 mm and 0.8 mm, respectively, and the one proximal tarsal was roughly twice the size of the other (i.e., 4.9 × 3.5 mm vs. 3.5 × 2.3 mm). On that basis, the larger proximal tarsal is interpreted as medial and the astragalus and the smaller tarsal as lateral and the calcaneum with prominent tuber. So identified, they compare well with the astragalus and calcaneum of Proterosuchus, Euparkeria, Rutiodon and Riojasuchus (Sereno, 1991), in which the astragalus was larger and subrectangular and the calcaneum was smaller and bore a prominent tuber, but they do not compare well with those of Lagosuchus (Sereno, 1991). Based on the morphology of the astragalus and calcaneum and comparisons to the tarsi of Proterosuchus, Euparkeria, Rutiodon, Riojasuchus and Lagosuchus, the tarsus of Scleromochlus had a crurotarsal joint. Sereno (1991: p. 37) acknowledged that the calcaneum “indeed does resemble a left crurotarsal calcaneum in lateral view, but this resemblance must be misleading because the bone clearly belongs to the right tarsus.” Such reasoning was illogical because the outline of a left crurotarsal calcaneum in lateral view would not be significantly different from that of a right in medial view, and Sereno gave no justification for interpreting the exposed face of the bone as lateral rather than medial. The preserved range of motion of the ankle is ~100° from folded close to the crus to moderately extended, but there is no reason to think that it could not have been extended to nearly 180°.

The morphology of the metatarsus with its four closely appressed elongate metatarsals suggests that flexion and extension at the metatarsophalangeal joints was somehow important in locomotion with the metatarsus capable of acting as a discrete segment of the hindlimb. As such, the pes should be considered semiplantigrade (Gebo, 1992). However, the arguments of Clark et al. (1998) against digitigrady in Dimorphodon weintraubi would apply as well to Scleromochlus because the metatarsophalangeal joints do not exhibit any modifications to suggest that they were suited to a habitual digitigrade posture with the metatarsus held at ≥45° to the substrate. Digits I–IV increased in length and diverged only slightly, digit IV was ~70% of the metatarsus length, and the proximal phalanges markedly longer than the more distal (Woodward, 1907; see also Benton, 1999: fig. 13b). Such toes combined with the elongate closely appressed metatarsals would result in a long and narrow pes, unsuited to erect digitigrady.

Skeletal reconstruction

A skeletal reconstruction of Scleromochlus incorporating the new evidence described and discussed above is shown in Fig. 20 (measurements in Supplemental Information). The proportions of the precaudal axial skeleton of the reconstruction were based primarily on NHMUK R3146A and caudal length was based on NHMUK R3557. The resulting proportions differ significantly from those of Sereno’s (1991: fig. 18b) and Benton’s (1999: fig. 14) reconstructions. Limb proportions were based on comparisons of NHMUK R3146A, R3914 and R3556, and are close to those of Sereno’s and Benton’s reconstructions. The skull and trunk are dorsoventrally compressed and the pelvic girdle shallow resulting in a body ~150% as wide as high, much as is found in extant lizards that spend much of their time with their trunk in contact with the substrate. The skeleton is reconstructed in what is interpreted as a typical sprawling resting posture with the brachia directed laterally, the antebrachia directed anteriorly, and the hindlimbs folded compactly and directed anterolaterally with the hip, knee, and ankle joints fully flexed. Such a posture is consistent with the reconstructed highly mobile shoulder joint, the hip joint that can be abducted in the horizontal plane but cannot swing in a parasagittal plane, the crurotarsal ankle, and the plantigrade to semiplantigrade pes. Such a posture is also consistent with the preservation of Scleromochlus skeletons in dorsal view with sprawling fore and hindlimb postures, articulated and undisturbed except by settling of the sand. Those who favor erect bipedal body postures such as Sereno (1991: fig. 18b) and Benton (1999: fig. 14) reconstructed should present arguments as to how the narrow ribcage and erect limbs of such erect bipeds could have been crushed into the preserved positions of the specimens.

Erect bipedal runner?

Woodward (1907) interpreted Scleromochlus as a bipedal runner but reconstructed it in a quadrupedal pose, which suggests that he viewed Scleromochlus as a sprawling to semi-erect facultative biped rather than as an obligate biped. However, Padian (1984) interpreted Scleromochlus as an obligate erect digitigrade biped, and subsequent authors (Gauthier, 1986; Sereno, 1991; Benton, 1999, 2006; Fraser, 2006) accepted that interpretation though Benton (1999: fig. 15) suggested that erect quadrupedal poses were also possible. Padian stated that Scleromochlus exhibited a bowed femur, fibula reduced to splint-like form, mesotarsal ankle with astragalus and calcaneum joined to the tibia, and elongate closely appressed metatarsus that implied digitigrade stance, features that formed a subset of a suite of hindlimb features that Padian (1980: 96–99; 1983), citing Coombs (1978), identified as correlates of erect digitigrade bipedal locomotion in pterosaurs (the other features were: femur shorter than tibia, double condyle knee with well developed intercondylar sulcus, median symmetry of the pes, and reduction of outer digits). However, no evidence or argumentation beyond lists of features has been offered to support that interpretation as a whole or specific aspects of it (e.g., bipedality, digitigrady, or erect limb posture), and Padian (1984) was incorrect in that the femur was not significantly bowed (and frogs also have somewhat bowed femora), the fibula was not splint-like, the ankle was not mesotarsal and the astragalus and calcaneum were not fixed to the tibia, and the ankle does not include pterosaur-like medial and lateral distal tarsals. Furthermore, no attempt has been made to explain the “high degree of specialization” that Woodward (1907: p. 144) found “truly astonishing” in what he took to be a Triassic dinosaur. Indeed, if Scleromochlus was an erect digitigrade bipedal runner, then its extreme cursorial adaptation remained unmatched for many million years until ornithomimid dinosaurs evolved.

Woodward (1907) implied that Scleromochlus’s limb proportions alone were sufficient to interpret it as a bipedal runner. Multivariate analysis of skeletal measurement data often have been used to reconstruct the locomotor or predatory behavior of fossil taxa by comparisons to extant and fossil taxa (Van Valkenburgh, 1987; Carrano, 1997, 1999; Elissamburu & Vizcaíno, 2004; Weisbecker & Warton, 2006; Shockey, Croft & Anaya, 2007; Samuels & Van Valkenburgh, 2008; Samuels, Meachen & Sakai, 2013; Janis, Buttrill & Figueirido, 2014; Janis & Figueirido, 2014; Chen & Wilson, 2015), and so a principal component analysis of trunk and limb segment lengths of Scleromochlus reconstructions, selected theropod dinosaurs, and representative extant vertebrates of known locomotor type was done to test Woodward’s idea.

The principal component analysis resulted in three components that explain 81.4%, 8.5% and 6.8% of variance, respectively. Bivariate plots of the components with polygons drawn around the ranges of different types of vertebrates are shown in Fig. 21. In the plot of Components 1 & 2 representing 89.9% of variance, most taxa are distributed in a broad arc with jerboas at the top, Zapus and Dipodomys in the middle, and lizards and toads at the bottom, whereas the frogs stretch leftward from Zapus and Dipodomys. The theropods Sinornithoides, Sinornithomimus, Struthiomimus and Compsognathus are in a similar arc slightly to the left with Sinornithoides above near jerboas, Sinornithomimus in the middle within the range of frogs and Struthiomimus nearby within the range of Zapus and Compsognathus below near Basiliscus and Bufo. Archaeopteryx is at lower left, well away from other taxa presumably because of its long forelimbs. The Scleromochlus reconstructions fall within the range of frogs (near the White-lipped Tree Frog and Crawfish Frog), near the range of Zapus, and one is right next to Sinornithomimus. In the plot of Components 2 & 3 representing 15.3% of variance, frogs are upper middle, toads and lizards overlap at left, jerboas are at right, and Scleromochlus is within the range of Dipodomys with Sinornithomimus nearby and Struthiomimus is within the range of Zapus somewhat below.

The wide distribution of theropods and the overlap with the ranges of frogs and Zapus shows that limb proportions alone are not sufficient to interpret an animal as an erect bipedal runner. Struthiomimus (~4 m long, ~150 kg) and Sinornithomimus (~2.5 m, ~25 kg) are much larger than the frogs (~14 cm snout-vent length) and Zapus (~24 cm, ~30 gm) that they plot near, and although they share very long hindlimbs with tibia longer than femur with frogs and jumping mice, the effects of scaling are such that similar limb and trunk proportions alone in Zapus and the 160 times longer and 5,000 times more massive Struthiomimus cannot be informative (Gasc, 2001; Iriarte-Díaz, 2002; Biewener, 2005). Thus, although Scleromochlus plots close to Struthiomimus and Sinornithomimus, it cannot be assumed that Scleromochlus was an erect bipedal runner. Because Scleromochlus (~17 cm) was similar in size to the frogs, lizards and rodents, the fact that it does not plot close to jerboas or lizards strongly suggests that it was not an obligate bipedal bounder or a quadrupedal runner, whereas the fact that it plots close to frogs and Zapus suggests that its locomotion was similar to quadrupedal hopping or quadrupedal bounding.

Osteological information can be used to put the principal component analysis results in context and constrain the interpretation of locomotor type. Scleromochlus’s imperforate acetabulum and weakly inturned hemispherical femoral head would not have permitted the hindlimb to adopt erect postures and swing in a parasagittal plane, and even if one assumed that the hip joint morphology was not a hindrance to such movement, then the prominent tubercle on the pubis would have been. Instead, the hip morphology and preserved range of motion indicate that the hip was highly mobile and could be flexed anteriorly, extended posteriorly, and abducted laterally to a considerable degree and so was suited to sprawling to semi-erect postures. The double condyle knee suggests that Scleromochlus was capable of powerful knee extension, but such knees and capabilities are not limited to bipedal runners and are not necessarily inconsistent with sprawling to semi-erect postures. The crurotarsal ankle with prominent calcaneal tuber, though not necessarily mechanically unsuited to digitigrade bipedality, is typical of sprawling to semi-erect archosauriforms and is not found in erect digitigrade bipedal archosaurs. The closely appressed equilength metatarsals suggest locomotory specialization, but the metatarsophalangeal joints do not exhibit any modifications to suggest that they were suited to a habitual digitigrade posture with the metatarsus held at ≥45° to the substrate, and the asymmetrical pes with digits diverging slightly seems to have been semiplantigrade. Together these features contradict the interpretation of Scleromochlus as an erect digitigrade bipedal runner and suggest it was suited to sprawling to semi-erect locomotion.

There is also an evolutionary reason to think that Scleromochlus was not an erect digitigrade bipedal runner. Cursorial adaptations have a metabolic cost and will only be evolved if they provide selective advantage. In extant tetrapods with marked cursorial adaptations that advantage comes from improved prey acquisition and/or predator avoidance, and that presumably was the case in extinct forms as well. Cursorial adaptations usually have been coevolved by similarly sized predators and prey in escalating competition over the course of millions of years (Bakker, 1983); extant examples of such coevolution include cheetahs preying on small to medium antelopes and lions preying on large antelopes, zebras, and buffalos, and fossil examples include dromaeosaurines preying on ornithomimosaurs. Such coevolving cursorial predators and prey are similarly sized because maximum running speeds of tetrapods are roughly proportional to body size (Alexander, 1977; Weitz & Levin, 2006), so large animals do not need marked cursorial adaptations to run down or outrun significantly smaller animals and small animals even with marked cursorial adaptations cannot run down or outrun significantly larger animals. As for Scleromochlus, it is unlikely that extreme cursoriality would have aided in prey acquisition, for example, no extant vertebrates seem to be specialized for running down small invertebrate prey. It is also unlikely that extreme cursoriality would have aided in predator avoidance because Erpetosuchus, Saltopus and juvenile Ornithosuchus were significantly larger (≥3.5 × body length) than Scleromochlus. In the end, the interpretation of Scleromochlus as an erect digitigrade bipedal runner must be rejected because of the problems discussed above and the absence of support from the principal component analysis.

Erect jerboa-like bounder?

Woodward (1907), presumably prompted by C.W. Andrews’ comment in the appended discussion, proposed jerboa-like bipedal bounding as a possible alternative to bipedal running. Benton (1999, 2006) also considered it a possible alternative, whereas Benton & Walker (1985; see also Witton, 2014) argued for bounding and made no mention of running. Jerboas and kangaroo rats typically inhabit arid, sandy environments, and Benton & Walker (1985) supported their interpretation of Scleromochlus as a jerboa-like bounder by stating that it shared several features with desert living lizards (i.e., nares nearly closed by lateral flanges, posterior flange of squamosal and quadratojugal protecting tympanic region, countersunk mandible, and flattened metatarsals) and was adapted to and lived in a sandy dune environment. Those four features do not seem to be present in Scleromochlus and were not discussed by Benton (1999), but despite that he interpreted Scleromochlus as a possible “sand hopper.” However, as discussed above, the depositional environment provides no support for an interpretation of Scleromochlus as living in a sand, desert, or dune environments, and so the depositional environment cannot be used to support an interpretation of Scleromochlus as a jerboa-like bounder. The only remaining argument for interpreting Scleromochlus as a jerboa-like bounder is the marked limb length disparity and very long hindlimbs, and as discussed above the hip, ankle, and metatarsophalangeal joint morphology are unsuited to erect digitigrade bipedality with the hindlimb swinging parasagittally, whereas the hindlimb was suited to sprawling to semi-erect postures. Like erect bipedal running, the interpretation of Scleromochlus as a jerboa-like bounder must be rejected because of the problems discussed above and the absence of support from the principal component analysis.

Quadrupedal hopper

Having rejected arboreality, erect bipedal running, and jerboa-like bounding, two further alternatives remain. The sprawling quadrupedal Scleromochlus might be interpreted as a hopper like frogs or as a facultative bipedal runner like some lizards.

Hopping in frogs starts from a sprawling resting pose with the hindlimb fully flexed at the hip, knee, crurotarsal, and tarsometatarsal joints such that the extensor muscles are fully stretched maximizing passive tension. The extension phase begins as the body is tilted upward by extension of the forelimb because leaping at an angle of 45° above horizontal maximizes distance (Peters, Kamel & Bashor, 1996). Flexion of the glenohumeral joint may briefly assist the powerful extension of the hindlimb in accelerating the body forward and upward, but the forelimb soon clears the substrate. Full extension of the hip, knee, crurotarsal, and tarsometatarsal joints is attained just before the pedal digits clear the substrate, and the frog flies through the air with hindlimbs extended backward. Experimental removal of the tail of kangaroo rats and jumping mice results in loss of pitch control such that the animal tumbles in the air, demonstrating that their tails provide such control (Svihla & Svihla, 1933; Bartholomew & Caswell, 1951; Libby et al., 2012). Although frogs lack tails, it is likely that the long posteriorly extended hindlimbs provide pitch and yaw control in lieu of a tail. Unfortunately, experiments to demonstrate that frog hindlimbs provide such control by removal of the hindlimbs are not feasible. The recovery phase begins as the forelimbs are extended shortly before landing, and upon landing the forelimbs forcefully decelerate and stabilize the body as it pivots downward and the hindlimbs are fully flexed to return to a stable resting pose in case another hop is necessary. Note that toads, which typically have shorter hindlimbs than other frogs and so cannot hop as far, are specialized for bounding, continuous quadrupedal hopping without return to a stable resting pose (Reilly et al., 2015).

The skeleton of Scleromochlus was well suited to frog-like hopping. The hindlimb resembles that of frogs in having a highly mobile hip joint, long femur and somewhat longer crus, and strong knee with prominent condyles. The four closely appressed equilength metatarsals were presumably functionally equivalent to the elongate tibiale and fibulare of frogs and the elongate navicular and calcaneum in leaping Galago and Tarsius, and markedly different from the metatarsus of lizards in which metatarsals diverge and are of varying lengths. In frogs, the pes is semiplantigrade in that the elongate tibiale and fibulare are usually held slightly above the substrate when the hindlimb is fully flexed, and that may have been the case with the metatarsus of Scleromochlus as well. The resting pose seen in Scleromochlus specimens, while markedly different from that of Leptopleuron, resembles that of frogs in having the hindlimbs folded compactly and directed anterolaterally with the hip, knee and ankle joints fully flexed such that the extensor muscles would be maximally stretched with passive tension maximized, ready to hop if startled (Fig. 18). The strong pectoral girdle with long scapula and robust coracoid, highly mobile glenohumeral joint, and long and strong forelimb with prominent deltopectoral crest and olecranon process could tilt the body upward and change direction in anticipation of hopping, contribute to initial acceleration, and decelerate and control the body as it returned to a resting pose upon landing.

Frog hindlimbs typically are more heavily muscled than those of lizards because all the force needed for leaps of 10 body lengths or more (Zug, 1972; Gomes et al., 2009) is applied in the brief acceleration before becoming airborne (~140 ms in Rana pipiens, Peters, Kamel & Bashor, 1996), whereas runners apply force with every stride. Therefore, a hopping Scleromochlus would be expected to have significantly larger hindlimb muscles than a running one and the forces on the posterior dorsal vertebrae would be much greater than in a runner. That fact probably explains the presence of the culet of closely spaced dorsal osteoderms restricted to the posterior half of the trunk, which seems to have severely limited lateral flexion of the posterior dorsal vertebrae. The culet would have stiffened the posterior trunk, enabling it to withstand the stresses associated with the powerful extension of the hindlimbs, and it probably was functionally equivalent in that regard to the urostyle and long ilia of frogs. It has been suggested that frogs lack tails because they would limit upward body tilt and thereby distance hopped (Handrigan & Wassersug, 2007), and if frog hindlimbs provide pitch and yaw control, then a tail would add mass yet contribute nothing to hopping. However, the slender tail of Scleromochlus might not have significantly limited upward body tilt, and a hopping Scleromochlus might have retained a tail for pitch and yaw control, because m. caudofemoralis was an important hip extensor, or perhaps because loss would produce pleiotropic effects.

The energy costs of running locomotion in mammals are equivalent for quadrupeds and bipeds (Roberts et al., 1998), and costs of bipedal bounding in jerboas and kangaroo rats are also roughly equivalent to running though slow is less efficient and fast more efficient (Webster & Dawson, 2004). Therefore, rodents presumably did not evolve bipedal bounding to reduce energy costs. Instead, it is likely that bounding evolved for predator avoidance (Bartholomew & Caswell, 1951). Jerboas and kangaroo rats are preyed upon by larger animals (e.g., foxes) that presumably could run them down when straightaway bounding, but given their small mass and inertia jerboas and kangaroo rats can accelerate and change direction much faster than larger animals, and their erratic bounding consisting of series of powerful leaps with repeated unpredictable changes of direction aids in predator avoidance. Selection for predator avoidance could also explain the evolution of quadrupedal hopping in Scleromochlus. As noted above, Scleromochlus was presumably preyed upon by Erpetosuchus, Saltopus and juvenile Ornithosuchus, which were significantly larger (≥3.5 × body length) and would have been faster straightaway runners. However, the combination of small mass and inertia and heavily muscled hindlimbs would enable a hopping Scleromochlus to perform series of powerful leaps with repeated unpredictable changes of direction so as to evade the predators.

The interpretation of Scleromochlus as a quadrupedal hopper proposed here would require long hindlimbs and a large hindlimb muscle mass but would not require unusual muscular or physiological sophistication. It would require control of the amount and balance of force produced by the two hindlimbs during hindlimb extension to vary distance and direction, but because frogs, grasshoppers, crickets, leafhoppers, and jumping spiders can perform those tasks there is no reason to think they would be beyond the capabilities of Scleromochlus. In addition, the interpretation would not represent a “truly astonishing high degree of specialization,” because longer hindlimb bones and increased muscle mass presumably would be rather simple to evolve.

The other alternative interpretation is that of Scleromochlus as a sprawling facultative bipedal runner like extant Basiliscus, Liolaemus and Tropidurus, which are habitually quadrupedal but run bipedally when startled with semi-erect gait and widely spaced pedes (Rand & Marx, 1967; Glasheen & McMahon, 1996; Rocha-Barbosa et al., 2008). The widely spaced pedes provide lateral (roll) stability and their long tail provides longitudinal (pitch and yaw) stability. Although the principal component analysis did not place close to Basiliscus, it is possible that Scleromochlus was capable of sprawling facultative bipedal running; however, such an interpretation would not explain the presence of the culet of osteoderms, the closely appressed equilength metatarsals, or the frog-like resting pose in Scleromochlus.

Bipedally running lizards with widely spaced pedes exhibit the same lateral undulations of the vertebral column as quadrupedal lizards and salamanders, and such undulations are desirable because they increase the stride length and speed and enable trunk musculature to contribute to locomotion. As discussed above, the culet of dorsal osteoderms would have stiffened the trunk of Scleromochlus, and such stiffening would be undesirable in a runner because it would prevent the lateral undulations of the vertebral column, reducing stride length and speed and preventing involvement of trunk musculature in locomotion. Bipedally running lizards have plantigrade pedes with spreading metatarsals of varying lengths, and the closely appressed equilength metatarsals of Scleromochlus would be unnecessary and unexpected if it was a sprawling facultative bipedal runner. The frog-like resting pose with the fully flexed hindlimb and fully stretched extensor muscles also would be unnecessary and unexpected in such a runner.

Clemente et al. (2008) found that bipedal running in lizards does not confer greater running speed or endurance compared to quadrupedally running relatives, but does result in greater initial acceleration. Perhaps longer hindlimbs have been selected for in some lizards to improve initial acceleration and so predator avoidance, and once the hindlimbs are significantly longer than the fore it is easier to run bipedally than quadrupedally. However, it seems likely that Scleromochlus evolved even longer and more heavily muscled hindlimbs for even greater initial acceleration and ended up flying through the air like a frog. Although the frog-like hopper and lizard-like facultative bipedal runner interpretations of Scleromochlus are both largely consistent with its morphology, the culet of osteoderms, closely appressed equilength metatarsals, and resting pose with fully flexed hindlimb are important components of frog-like hopping that would be disadvantageous or unnecessary in a bipedal runner, and on that basis and supported by the principal component analyses, the frog-like hopper interpretation is accepted here.

Replication of Ezcurra’s (2016) analyses

Attempts at replicating Ezcurra’s (2016) Analyses 1–3 revealed inconsistencies. The NEXUS file in his Supplemental Information was modified by the deletion of the MacClade and Mesquite commands and the addition of delete commands in the PAUP block to replicate Ezcurra’s Analyses 1–3 with 98, 97 and 81 taxa respectively, and when analyzed resulted in most parsimonious trees of 2,832, 2,948 and 2,828 steps, respectively, all markedly longer than the 2,651, 2,664 and 2,646 steps reported by Ezcurra. The PAUP block in the NEXUS file included the command “pset = mstaxa variable,” which results in the heuristic search setting “multistate taxa interpretation depends on “{}” vs. “()” designation” such that characters enclosed by braces, “{}”, are treated as uncertain and characters enclosed by parentheses, “()”, are treated as polymorphic. When the pset command was deleted from the PAUP block, then the heuristic search setting “multistate taxa interpreted as uncertainty” was applied, which resulted in most parsimonious trees of 2,651, 2,664 and 2,646 steps, respectively, the same as reported by Ezcurra. The strict consensus trees from Analyses 1 and 2 when replicated both with and without the “pset = mstaxa variable” command were the same as those in Ezcurra’s figure 52. The two treatments of multistate characters thus resulted in consensus trees with identical topologies but different lengths.

Ezcurra (2016) conducted his analyses with TNT, a program with which I am unfamiliar; however, it seems likely that if Ezcurra obtained shortest trees of 2,651, 2,664 and 2,646 from his Analyses 1–3 when using TNT, then that program was also treating all multistate characters as uncertain rather than treating those bounded by braces as uncertain and those bounded by parentheses as polymorphic.

Ezcurra (2016: pp. 271, 274) described the 81 taxon Analysis 3 as resulting in 40 most parsimonious trees of 2,646 steps and Consistency Index (CI) = 0.2978, but illustrated a strict consensus tree “after the a posteriori pruning of Kalisuchus rewanensis and Asperoris mnyama” in his Figures 48 and 50; thus the statistics do not apply to the illustrated tree. Analysis 3 when replicated with the “pset = mstaxa variable” command resulted in 1,028 most parsimonious trees of 2,828 steps and the strict consensus tree was broadly similar to that in Ezcurra’s Figures 48 and 50 but with Asperoris mnyama, Dongusuchus efremovi, Dorosuchus neoetus, Euparkeria capensis, Yarasuchus deccanensis, and the clade Archosauria + Proterochampsia in a polytomy, and with Cuyosuchus huenei, Fugusuchus hejiapanensis, Kalisuchus rewanensis, Sarmatosuchus otschevi, and the clade Erythrosuchidae + Eucrocopoda in a polytomy. Analysis 3 when replicated without the “pset = mstaxa variable” command resulted in 1008 most parsimonious trees of 2,646 steps and the same consensus tree topology.

In order to replicate the strict consensus tree in Ezcurra’s (2016) Figures 48 and 50, I added Asperoris mnyama and Kalisuchus rewanensis to the delete command in the PAUP block and repeated the analysis. This 79 taxon analysis when replicated with the command “pset = mstaxa variable” resulted in 24 most parsimonious trees of 2,821 steps and the strict consensus tree lacked the polytomies discussed above and matched the topology of the strict consensus tree in his Figures 48 and 50, whereas when replicated without the command “pset = mstaxa variable” resulted in 24 most parsimonious trees of 2,639 steps and CI of 0.2982 and the same consensus tree topology. Thus the reason for the a posteriori deletion of the two taxa was presumably cosmetic, replacing the polytomies with dichotomous branching, but why did Ezcurra not report the statistics from that analysis that produced the strict consensus tree he illustrated? In the end, it is remarkable that Ezcurra’s 79 taxon data matrix produced only 24 most parsimonious trees and a fully resolved strict consensus tree. However, it should be remembered that appearances can be deceiving and its dichotomous branching is but two taxa away from major polytomies.

Homogeneity testing and data matrix quality

Bull et al. (1993) showed that the probability of recovering the correct phylogeny decreases as the percent of inconsistent characters in a data set increases, and stated that data set partitioning and homogeneity testing are appropriate if one lacks confidence in a data set’s quality. Where significant homogeneity is found it is inappropriate to combine the data set partitions without modification or further justification. I lacked confidence in the quality of Ezcurra’s (2016) 79 taxon 600 character data set, so I partitioned it into Cranial, Postcranial, Forelimb and Hindlimb partitions and subjected them to homogeneity testing (see explanation in Materials and Methods). The results revealed significant incongruence in 90% of homogeneity tests at the α = 0.01 probability level (Table 4) whereas, for example, Bennett’s (2013) Revised Data Set had 0% incongruence. Although homogeneity testing may reveal incongruence related to homoplasy between one set of characters and others, just as Bennett (2013) used it to detect homoplasy in the coding of a suite of hindlimb characters that Padian (1980, 1983) asserted were associated with cursorial digitigrade bipedal locomotion in pterosaurs, here the significant incongruence in 90% of homogeneity tests indicates that there is much character inconsistency and significantly reduced probability of recovering the correct phylogeny. As a result, the data set and its phylogenetic signal can be considered to be of low quality.

The low quality of Ezcurra’s (2016) data set seems to result from following the general trends in recent phylogenetic analyses of increased numbers of taxa and characters (Table 5). Increasing numbers of taxa may result from increasing taxonomic breadth and/or using multiple species as terminal taxa instead of a single composite taxon to represent higher level taxa (e.g., family or order). Multiple individually coded species will usually have more missing data than a composite taxon, in which case a consequence of their use is increased missing data in the data matrix [Note that although Ezcurra (2016) and other authors used both MISSING (?) and GAP (–) symbols in their data matrices, GAP symbols are ignored in PAUP* unless DATATYPE is DNA, RNA, or PROTEIN, thus gaps are treated the same as missing]. In the case of Ezcurra’s (2016) data set, the percentage of missing data is very high, 58% in the 98 taxon matrix and 49% in the 79 taxon matrix (Table 5), and the high percentage presumably contributed to the low quality of the data set. Wiens (2003) showed that missing data could cause problems either by having insufficient characters coded for a taxon so that its position could not be accurately determined or by having insufficient taxa coded for a character so that its distribution could not be accurately determined, and it seems likely that Ezcurra’s (2016) data set suffered from both problems.

Ezcurra (2016) also followed the trend of increasing the number of characters (Table 5): Brusatte et al. (2010) had 187 characters including 47 new ones; Nesbitt (2011) had 412 characters including 114 new ones; and Ezcurra (2016) had 600 characters including 96 new ones. Comparison of the homogeneity testing of Ezcurra’s (2016) 79 taxon data matrix, in which 90% of homogeneity tests were significantly incongruent (Table 4), and Nesbitt’s (2011) 77 taxon data matrix, in which 60% of tests were significantly incongruent, suggests that the decreased quality of Ezcurra’s (2016) matrix is directly related to the 188 additional characters. However, it is probably not the number of characters, per se, but rather the quality of the added characters that affects the overall quality of the data matrix.

Because paleontologists have been compiling phylogenetic analyses to reconstruct the relationships of archosauromorphs for over 30 years, most of the good characters have already been identified. New characters whether created de novo or by atomizing old characters (i.e., splitting one character into multiple characters that code for parts of the original character) have often described simple shapes or proportions such that the derived character state may occur in multiple unrelated species. In addition, characters describing simple shapes or proportions taken from previous analyses of restricted taxonomic breadth in which they were consistent may turn out to be highly inconsistent when applied across a broader range of taxa. Such characters are phenetic, and parsimony analyses using such characters will tend to group taxa on the basis of similarities of shape and proportion rather than possession of shared unique structures inherited from common ancestors (Sneath & Sokal, 1973). Many of the characters in Ezcurra’s (2016) and Nesbitt’s (2011) analyses are phenetic with character state distributions exhibiting multiple convergent acquisitions of the derived condition, thus their analyses are hybrid phenetic-phylogenetic analyses, and both acknowledged as much: Nesbitt (2011: p. 85) stated regarding his Char. 93 “the distribution of this character is not straightforward, but it may support small clades” and Ezcurra (2016: p. 386) stated regarding his Char. 130 “this condition does not seem to have a priori a clear phylogenetic signal, but it may be informative at low taxonomic levels.” Ezcurra (2016) discussed the bimodal distribution of character CIs, but seems to have been unaware that homoplastic characters should be excluded or reformulated to reduce homoplasy as much as possible.

Ezcurra’s results regarding the Pterosauria

Ezcurra (2016: pp. 338–339) stated that his phylogenetic data set constituted “the best data matrix compiled so far to test the position of pterosaurs within Archosauromorpha because of the broad sample of Permo-Triassic species, including the undoubted pterosaur Dimorphodon macronyx.” I will grant that the taxonomic scope of Ezcurra’s data set, extending from Petrolacosaurus to basal dinosaurs, was the broadest at the time that included pterosaurs—Benton (1999) used Euparkeria as the outgroup, Brusatte et al. (2010) used Erythrosuchus, Euparkeria and the Proterochampsidae, Nesbitt (2011) used Mesosuchus and Prolacerta and Bennett (1996, 2013) used the Lepidosauromorpha as outgroup. However, Ezcurra’s (2016) data set seems inadequate to test the phylogenetic position of pterosaurs in that among other things: Ezcurra (2016) included only one pterosaur (Brusatte et al. used Dimorphodon and Pteranodon; Nesbitt (2011) used Dimorphodon and Eudimorphodon and Bennett’s (2013) Updated Data Set used Anurognathus, Peteinosaurus and Eudimorphodon); Dimorphodon was neither basal nor a typical Early Jurassic pterosaur; Ezcurra (2016) made several errors in coding Dimorphodon; and Ezcurra (2016) did not take the opportunity to test the relationships of Scleromochlus taylori, which had often been advanced as a sister taxon of the Pterosauria or of Pterosauria + Dinosauromorpha (Padian, 1984; Gauthier, 1984, 1986; Sereno, 1991; Benton, 1999, 2004; Brusatte et al., 2010).

Not only was Ezcurra’s (2016) data set not the best to test the position of pterosaurs, Ezcurra (2016) also ignored Bennett’s (2013) demonstration through data set partitioning and homogeneity testing of incongruence between a partition consisting of a suite of hindlimb characters that Padian (1980, 1983) asserted were associated with cursorial digitigrade bipedal locomotion and partitions consisting of cranial, postcranial axial, forelimb, and other hindlimb characters, respectively, and with all other characters collectively at the α = 0.01 probability level. Ezcurra (2016) ignored my further demonstration that the source of the incongruence was homoplasy in the coding of pterosaurs for the cursorial characters by removing each taxon from the data matrix one at a time and subjecting the remaining taxa to homogeneity testing. Ezcurra (2016) ignored my assertion that those homoplastic hindlimb characters had misled the earlier analyses (Gauthier, 1984, 1986; Benton, 1990; Sereno & Arcucci, 1990; Sereno, 1991) from which Bennett (1996) had drawn characters and codings, and presumably misled the subsequent analyses (Benton, 1999; Langer & Benton, 2006; Hone & Benton, 2007, 2008; Irmis et al., 2007; Nesbitt et al., 2009; Brusatte et al., 2010; Nesbitt, 2011) that also used those homoplastic characters. Ezcurra’s (2016) analysis used those homoplastic characters as well, so there is no reason not to think that it was also misled. Bennett’s (2013) interpretation of those hindlimb characters as homoplastic may or may not be correct, but until it is refuted or accepted any serious discussion of the relationships of pterosaurs to other archosauromorphs should address it.

Phylogenetic position of Scleromochlus

The results of the three parsimony analyses of the 80 taxon 600 character data matrix consisting of Ezcurra’s (2016) 79 taxon data matrix plus Scleromochlus (Fig. 22; Supplemental Information) show Scleromochlus in positions other than the generally accepted sister taxon relationship with dinosauromorphs. Analysis #1 with characters ordered or unordered as described in Ezcurra’s (2016) Nexus file found 96 most parsimonious trees of 2,873 steps and CI = 0.3376. Note that all strict and 50% majority-rule consensus trees resulting from analyses #1–3 are shown in the Supplemental Information, and that in analyses #2 and 3, resulting trees were initially rooted near Pamelaria dolichotrachela, which made comparisons to trees from analysis #1 difficult, so Petrolacosaurus kansensis was designated as ancestor to make comparisons easier. The 50% majority-rule consensus tree differs from the strict consensus tree of Ezcurra’s 79 taxon analysis only in that Scleromochlus is a sister taxon of Tarjadia ruthae + Archeopelta arborensis and Jaxtasuchus salomoni + Doswellia kaltenbachi within the Doswelliidae, whereas the strict consensus tree has Scleromochlus in a polytomy with Vancleavea campi, Tarjadia, Archeopelta and Jaxtasuchus + Doswellia. Those topologies require the reversal of characters 558 and 559, distal tarsals 1 and 2 present or absent, respectively, and if such reversals of losses are viewed as unremarkable then we can conclude that Scleromochlus was a doswelliid.

Gauthier (1984) used the character “distal tarsal 1 lost and 2 slow to ossify” as a synapomorphy of the clade consisting of the most recent common ancestor of the Erythrosuchidae and crown group Archosauria and all its descendants, and Benton (1985, 1990) used the character “tarsus consists of four elements” (i.e., astragalus, calcaneum, and distal tarsals 3 and 4) as a synapomorphy of the same clade. Sereno & Arcucci (1990) stated the condition of distal tarsals 1 and 2 was unknown in erythrosuchids and used the character “distal tarsals 1 and 2 lost” as a synapomorphy of the clade Euparkeria capensis + Archosauria; however, Cruickshank (1978; see also Gower, 1996) showed that Erythrosuchus africanus had lost distal tarsals 1 and 2. Although Benton (1999) described the distal tarsals 1–4 of Scleromochlus taylori, his analysis used Euparkeria as the outgroup, his only distal tarsal characters pertained to the shape of distal tarsal 4 and its articulation with metatarsal V, and he inexplicably ignored the implication that Scleromochlus was either outside the clade Erythrosuchidae + Archosauria or exhibited a remarkable atavism in the reversal of the loss of distal tarsals 1 and 2. Benton (2004) commented on a 24 taxon 95 character analysis that used Hyperodapedon as the outgroup, but seems not to have published the data matrix. He used the characters distal tarsals 1 and 2 absent as synapomorphies of the clade Erythrosuchus + Euparkeria + Proterochampsidae + Avesuchia, which included Scleromochlus as the sister taxon to the clade consisting of the common ancestor of the Pterosauria and Dinosauromorpha and all its descendants, but did not comment on the reversal of the loss of distal tarsals in Scleromochlus that such a position presumably would have required.

The loss of distal tarsals 1 and 2 in the clade consisting of the most recent common ancestor of the Erythrosuchidae and Archosauria and all its descendants seems to be an evolutionary event equivalent to the loss of the cleithrum in primitive amniotes, manual digit V in pterosaurs, and all teeth in birds, that is, complete loss of a skeletal structure or structures and modification of the adjacent skeletal elements to adapt to and compensate for the loss. There are no known instances of reversals of the losses of the cleithrum, manual digit V, or teeth, and no known instances of reversals of the loss of distal tarsals 1 and 2 in archosauriforms, so it is remarkable that anyone would think it no problem evolutionarily to reverse the loss of distal tarsals 1 and 2 in Scleromochlus simply because it is most parsimonious in a cladistic analysis. Until an explanation for reversal of the loss of distal tarsals 1 and 2 in Scleromochlus can be found, it is reasonable to assume that it is evolutionarily most likely that the presence of distal tarsals 1 and 2 in Scleromochlus was not the result of the reversal of losses.

In order to explore alternatives to the result of Analysis #1, a second analysis was done with Chars. 558 and 559 set to irreversible. Analysis #2 found >1,000 most parsimonious trees of 2879 steps and CI = 0.3369. The strict and 50% majority-rule consensus trees from analysis #2 are identical to those of analysis #1 with Scleromochlus within the Doswelliidae, and the greater length reflects multiple convergent losses of distal tarsals 1 and 2 in other taxa instead of the above noted reversals of the losses in Scleromochlus. However, that also seems unlikely because such a result would require the retention of distal tarsals 1 and 2 in Scleromochlus while multiple closely related lineages underwent convergent losses of the tarsals.

A third analysis was done with Chars. 558 and 559 set to irreversible and weighted as 2 in order to reduce the parsimoniousness of convergent losses of the distal tarsals. Analysis #3 found >1000 most parsimonious trees of 2,886 steps and CI = 0.3369. The 50% majority-rule consensus tree has Scleromochlus as a sister taxon of a clade consisting of the common ancestor of Euparkeria, Erythrosuchidae, Proterochampsia and Archosauria and all its descendants, in a position that is as far up the tree as possible without multiple convergent losses of the distal tarsals in related taxa (see dashed line in Fig. 22). Note that although the composition of that clade is equivalent to the clade Erythrosuchidae + Eucrocopoda, the ingroup relationships are different from those in Ezcurra’s 79 taxon analysis. The strict consensus tree has Scleromochlus in the same position and most taxa within the Eucrocopoda in a large polytomy.

Although analyses #1–3 do not seem to suggest exactly where in the archosauriform tree Scleromochlus belongs, the most important message from the analyses is that there is no support in any of them for a placement of Scleromochlus as sister taxon of either the Pterosauria or Dinosauromorpha (Sereno, 1991) or as sister taxon to the Pterosauria + Dinosauromorpha (Benton, 1999, 2004) or as a sister taxon in a tritomy with Lagerpeton and the Dinosauriformes (Ezcurra et al., 2017). In the end, it must be concluded that Scleromochlus was not close to dinosauromorphs.

In analyses #1–3, the pterosaur Dimorphodon macronyx was placed as sister taxon of the Dinosauromorpha. Although Ezcurra (2016) opined that his analysis was the best data matrix compiled so far to test the position of pterosaurs within the Archosauromorpha, there are errors in the coding of Dimorphodon and other problems discussed above. As such, neither Ezcurra’s (2016) analysis nor the three 80 taxon analyses should be viewed as providing significant support for the Pterosauria as a sister taxon of the Dinosauromorpha (Padian, 1984; Gauthier, 1984, 1986; Sereno, 1991; Benton, 1999, 2004; Brusatte et al., 2010; Nesbitt, 2011).

Bennett (1996) argued that a suite of hindlimb characters that Padian (1980, 1983) asserted were associated with cursorial digitigrade bipedal locomotion in pterosaurs were convergent with those of dinosaurs; however, I used a posteriori recoding of a set of functionally related characters, which was deemed an inappropriate methodology and gave others an excuse to ignore the conclusion that pterosaurs were not ornithodirans. It is notable that Jiang et al. (2016) did the same thing, a posteriori recoding of a set of functionally related characters in order to examine their effect on the evolution of ichthyosauriforms, but their methods and results have not been challenged. However, Bennett (2013) demonstrated through homogeneity testing of Bennett’s (1996) data set after division into five partitions, that the partition consisting of the suite of characters that Padian (1980, 1983) asserted were associated with cursorial digitigrade bipedal locomotion was incongruent with other partitions and all other characters at the α = 0.01 probability level. To date, no one has suggested that Bennett (2013) used inappropriate methodology, but the conclusions have been ignored.

The character codings for Scleromochlus used in the 1996 and 2013 cladistic analyses were based on my uncritical acceptance of Padian’s (1984) interpretation of Scleromochlus’s hindlimb morphology, Gauthier’s (1984, 1986) and Sereno’s (1991) codings of Scleromochlus from their cladistic analyses, and Huene’s (1914) interpretation of Scleromochlus as an arboreal leaper rather than on examinations of the specimens or casts, so the codings needed to be corrected. Analysis of Bennett’s (2013) 19 taxa 134 character Updated Data Matrix with the coding of Scleromochlus corrected on the basis of the new information and interpretation in the present article found a single most parsimonious tree of 274 steps and CI = 0.5876 (Fig. 23), which differs from that of Bennett (2013: fig. 3) in that Scleromochlus is the sister taxon of the clade Pterosauria + (Erythrosuchidae + Eucrocopoda). That result compares well with the results of Analyses #1–3 of the 80 taxon 600 character data matrix discussed above in that Scleromochlus is not an ornithodiran, and compares well with Analysis #3 in that Scleromochlus is outside the clade Erythrosuchidae + Eucrocopoda. Although Scleromochlus is close to the Pterosauria in the cladogram, its crurotarsal ankle and terrestrial lifestyle suggest that it was not particularly close to the last non-volant ancestor origin of pterosaurs.