Myology of the forelimb of Majungasaurus crenatissimus (Theropoda, Abelisauridae) and the morphological consequences of extreme limb reduction

Scapulocoracoid

The scapulocoracoids of abelisaurids are characterized by large, elliptical coracoids and long, broad scapular blades with parallel edges. This provides large potential attachment areas for much of the pectoral musculature (Fig. 2). The insertion of Serratus superficialis is marked in Majungasaurus on the ventral edge of the medial aspect of the scapular blade by a shallow groove that extends distally from the approximate midpoint of the scapular blade (Fig. 2B). This groove is developed to various extents in other ceratosaurians; in Aucasaurus (MCF PVPH 236) it begins slightly more proximally on the blade whereas in Carnotaurus (MACN‐CH 894) and Ceratosaurus (UMNH VP 5278, BYUVP 13024) it is more extensive, covering two‐thirds of the scapular blade. The groove typically exhibits a well‐defined dorsal edge proximally that becomes indistinct at the distal end of the scapular blade. The medial surface of the distal end of the scapular blade is marked by shallow grooves in a Y‐shape that delineate the boundaries between the insertions of Serratus profundus, Rhomboideus, and the origin of Subscapularis (Fig. 2B).

Subscapularis originated broadly from the medial surface of the scapular blade, its dorsal and ventral edges bounded by the groove for Serratus superficialis ventrally and dorsally by a series of small grooves located at approximately the midpoint of the scapular blade, as in Carnotaurus (Fig. 2B). Because ceratosaurians do not possess a distinct medial mid‐scapular ridge as seen in early theropods (Burch, 2014) and their scapular blades do not narrow proximally before the acromial expansion, Subscapularis had a much more proximally extensive origin than seen in other theropods. The small, dorsal grooves of Majungasaurus and Carnotaurus may indicate the ventral extent of the Trapezius and Levator scapulae musculature, which insert along the dorsal edge of the scapular blade (Fig. 2B) but, because they are only present for a short distance in the middle of the blade, this relationship is not clear. However, in Ceratosaurus (UMNH VP 5278) this insertion is distinctly marked by a long, narrow, depressed, and highly rugose groove extending along the entire dorsal surface of the scapular blade. As the scapula expands into the acromion this groove becomes less rugose and distinct, and likely represents the area of this scar devoted to the insertion of Trapezius. Levator scapulae and Trapezius are thought to be lost in more derived members of Theropoda with a subhorizontal scapular blade (Jasinoski et al. 2006; Burch, 2014), but are here reconstructed due to the presence of osteological correlates in a wide variety of large‐bodied theropod dinosaurs.

The ventral surface of the scapula in Majungasaurus is marked by a distinct and slightly rugose lump near the glenoid that was previously identified as the scar for the insertion of Triceps brachii scapularis (Burch & Carrano, 2012). This lump is present as a distinct tubercle on the right scapula of Carnotaurus (MACN‐CH 894), but it is elongated distally into a prominent fin on the left scapula of Carnotaurus as well as in Aucasaurus (MCF PVPH 236). A similar rugose lump is also found on the scapula of Ceratosaurus, but it is located further from the glenoid fossa. The fin‐like morphology of this feature in some abelisaurids makes it an unlikely attachment for Triceps brachii scapularis, which has a limited tendinous insertion in archosaurs. It is much more probable that this lump and/or fin represents the area of origin of Scapulohumeralis posterior (Fig. 2). The lack of a mid‐scapular medial ridge would have limited the medial extent of Scapulohumeralis posterior in ceratosaurians, causing the fin‐like structure to develop to provide additional attachment area for this muscle. In addition, the lateral surface of the glenoid lip is rugose in Ceratosaurus, indicating an origin of Triceps brachii scapularis close to the edge of the glenoid. In Majungasaurus there is a small amount of rugosity between the lump and the edge of the glenoid lip, but in abelisaurids this origin is generally indistinct. The origin of Scapulohumeralis anterior is also difficult to place in ceratosaurians because the area dorsal and posterior to the glenoid fossa is smooth and lacks any kind of ridge demarcating the attachment of this muscle. This, along with a lack of any osteological correlate on the humerus in ceratosaurs, makes reconstruction of this muscle highly uncertain in ceratosaurs, and thus it is omitted at this point.

The broad, wide scapular blade of Majungasaurus also provides a large area of origin on its lateral surface for the Deltoideus scapularis (Fig. 2A). As with Subscapularis, the lack of narrowing of the scapular blade allows for a much wider origin proximally and a more even distribution of muscle fibers over the length of the blade. The acromial expansion is the location of origin of Deltoideus clavicularis, and in Majungasaurus it is long and slopes gently, creating a broad, triangular surface posterodorsal to the subacromial depression (Fig. 2A). This area narrows to a point at the tip of the acromion, allowing some muscle fibers to originate slightly anterior to the glenoid fossa. The subacromial depression of Majungasaurus and Carnotaurus is divided into two portions by a low ridge. The smaller portion, restricted to the scapula, forms a long oval along its proximal edge and represents the site of origin for Supracoracoideus accessorius (Fig. 2A). The rest of the Supracoracoideus complex arose from the larger part of the depression, which is formed primarily by the coracoid but extends onto a small part of the proximal scapula just beyond the line of fusion between the two bones.

The biceps tubercle of Majungasaurus is low but distinct and clearly marks the point of origin for the coracoid head of Biceps brachii (Fig. 2A). It is located close to the coracoid lip of the glenoid fossa and helps to define the margin of the subglenoid fossa. The subglenoid ridge intersects with the margin of the coracoid a short distance from the point of inflection of the posteroventral process, creating a wide, subrectangular subglenoid fossa for the origin of Coracobrachialis (Fig. 2A). This fossa is set off from the lateral surface of the coracoid by the prominent subglenoid ridge and faces slightly posteroventrally. A low rugosity on the anterior edge of the coracoid in one specimen of Majungasaurus (FMNH PR 2836, left side) may represent an area of origin of Pectoralis (Fig. 2A); this broad muscle arose from presumably cartilaginous sternal elements but in some theropod taxa it may have also had an origin from the coracoid, as in Struthio (Jasinoski et al. 2006). A scar for Pectoralis may also be located on the right coracoid of Aucasaurus (MCF PVPH 236), which also has a low rugosity along the ventral portion of its margin. On the medial surface of the Majungasaurus coracoid there are no indications of the boundaries of the origin of Subcoracoideus (as in most other theropods), but in Ceratosaurus (UMNH VP 5278) there is a large, subcircular, lightly striated depression just anterior to the coracoid foramen that indicates that this muscle arose from this area of the expansive coracoid in ceratosaurians (Fig. 2B).

Humerus

Abelisaurids have stout, robust humeri that exhibit a low to moderate degree of scarring from the musculature. Nevertheless, there are several prominent features that are identifiable as muscle attachment sites. Majungasaurus, Carnotaurus, and Aucasaurus all possess well‐developed internal tuberosities that project substantially beyond the medial border of the humeral head (Fig. 1). This is the insertion site for Subscapularis and Subcoracoideus, two of the main adductors of the forelimb (Fig. 3). Although these two muscles typically insert with a common tendon, separate insertions in abelisaurids may be indicated by a midline ridge present in some specimens (right humerus of FMNH PR 2836 of Majungasaurus and right humerus of MACN‐CH 894 of Carnotaurus) that divides the proximal surface of the internal tuberosity into slightly anterior‐ and posterior‐facing surfaces. In this case it would be expected that the anterior face would be occupied by the insertion of Subcoracoideus, being the more anteriorly located muscle, and the posterior face would be occupied by the insertion of Subscapularis.

Distal to the proximal surface of the internal tuberosity, a narrow, distinct oblong depression is present on the medial margin of its anterior surface in FMNH PR 2423, representing the site of origin of the humeral head of Biceps brachii (Fig. 3A,B); a small rugosity near this position is also present in the right humerus of Aucasaurus. The posterior surface of the internal tuberosity is marked in Majungasaurus (represented in both FMNH PR 2423 and the right humerus of FMNH PR 2836) by a small rugose area that corresponds to the insertion point of Scapulohumeralis posterior (Fig. 3C).

The humerus FMNH PR 2423 of Majungasaurus possesses a deep fossa on the anterior surface of the humeral shaft just distal to the humeral head, which represents the wide, fleshy insertion site of Coracobrachialis (Fig. 3A). The distal extent of this fossa is somewhat restricted, potentially limiting the distal range of this insertion to slightly less than half the length of the deltopectoral crest. The area available for the insertion of Pectoralis on the medial surface of the deltopectoral crest is also limited given its low, broad morphology, so the insertion would have been restricted to a narrow area at the apex of the crest (Fig. 3A,B). However, at the tip of the deltopectoral crest its rugose edge is extremely wide and provides a large attachment site for the insertion of the Supracoracoideus complex of muscles. The greater tubercle is distally displaced, leaving a short stretch of the edge of the crest for the insertion of Supracoracoideus accessorius (Fig. 3A,D).

The posterior surface of the deltopectoral crest in abelisaurids exhibits numerous large, lumpy rugosities that do not obviously correspond to muscle attachment sites in this area in early theropods. Although their morphology is unusual, it is very similar across all abelisaurid taxa, and smaller tubercles in an equivalent area also appear in the humerus of Ceratosaurus (UMNH VP 5278). In Majungasaurus there are two distinct tubercles preserved in multiple specimens: a proximal one close to the greater tubercle, and a distal tubercle separated from the edge of the deltopectoral crest by a flat, unmarked surface. Although these tubercles are located near the area of insertion for the Deltoideus musculature, other osteological correlates in the area support alternate insertions for these muscles. The posterior surface of the greater tubercle is covered by light rugosity in Majungasaurus and slightly larger rugosities in Aucasaurus, suggesting a limited insertion for Deltoideus scapularis in this area not associated with the tubercles (Fig. 3C,D). Deltoideus clavicularis typically has a wide insertion over the lateral surface of the deltopectoral crest, and in Ceratosaurus this area is heavily striated, extending up to the edge of the posterior tubercles, which form a ridge extending proximodistally. This ridge is aligned proximodistally with a smaller, more distally located ridge and together they may represent the triceps ridge, from which Triceps brachii lateralis originates. Ceratosaurus also lacks a distinct furrow for the insertion of Latissimus dorsi that is found in more basal taxa; instead, it is likely that the ‘tubercles’ in Ceratosaurus are a product of a more proximally located insertion of Latissimus dorsi that is interacting with the triceps ridge in this taxon to create a highly rugose area. Although it is somewhat difficult to clearly assign the tubercles in Aucasaurus and Carnotaurus to this arrangement, the tubercles of Majungasaurus more closely align with the morphology of Ceratosaurus, and thus a similar arrangement is tentatively reconstructed here (Fig. 3).

The origin of Triceps brachii medialis does not usually leave a scar on the bone surface, but it is large and fleshy, covering much of the posteromedial surface of the humerus (Burch, 2014). The humerus of Majungasaurus is smooth in this area and lacks any ridges or tuberosities that would potentially limit the area of origin, so it is here reconstructed with a similar morphology to the basal condition (Fig. 3). The origin of Brachialis is also difficult to position with accuracy because it, too, lacks any defining scars in early theropods, in which it arises linearly from the distal part of the deltopectoral crest and the anterolateral humeral shaft. The low apex of the deltopectoral crest wraps around the humeral shaft to a more anteriorly placed position in Majungasaurus, and so the origin is reconstructed in a position that is located closer to the midline of the anterior surface (Fig. 3A). The antebrachial muscles arising from the humeral ect‐ and entepicondyles would have likely done so in close association with one another, and there is little indication that their arrangement in Majungasaurus would have differed from that in a more basal taxon. However, the entepicondyle of both Majungasaurus and Aucasaurus projects far anteriorly, placing the origin of the muscles attaching to it more medially than in earlier theropods (Fig. 3A). It is difficult to locate a true ectepicondyle in Majungasaurus, although a slight depression on the posterolateral surface of the shaft just proximal to the articular surface may represent the area from which the ectepicondylar muscles originated (Fig. 3C,D). The relative posterior placement of the ectepicondyle and the proximal placement of the insertion of Supinator (see below) indicate that the origin of Supinator had migrated more proximally on the humeral shaft. In Aucasaurus, the radius possesses a proximal projection of this muscle attachment and concurrently exhibits a rugose ridge just proximal to the ectepicondyle, likely representing a proximally shifted origin for this muscle.

Antebrachium

The abelisaurid antebrachium has a highly unusual reduced morphology distinct from that of other theropods. Although the antebrachial elements are very short, they are also very robust, with several large, rugose muscle scars. When articulated, the two bones lock tightly together and no long‐axis rotation is possible for either element. Most of the supinators and pronators of the forearm also have a role in flexion or extension and evidence from the scars in Majungasaurus indicates that they have shifted toward this function rather than disappearing. The exception to this is Pronator quadratus, which only has a function in pronation of the forearm and manus and thus would likely have been lost in abelisaurids. Additionally, the status of Pronator accessorius is unknown; there is a single, undifferentiated Pronator muscle in paleognaths (Beddard, 1898), and the fusion of two similarly acting muscles is also a common stage seen in limb reduction of lepidosaurs (Fürbringer, 1870). In the limited muscular space available in the abelisaurid forelimb, Pronator accessorius may have become lost or fused with Pronator teres. The insertion of Pronator teres on the radius of Majungasaurus is represented by a small triangular rugosity near the proximal end of the anterior surface that extends into a low ridge along the shaft of the bone (Fig. 4A,C). Just lateral to this is a very large triangular rugosity previously referred to as the lateral triangular rugosity and identified as the insertion of Supinator (Burch & Carrano, 2012). Inserting on the anterolateral surface of the radius in other theropods, Supinator is a major flexor of the forearm and this action seems to be retained in Majungasaurus. The rugosity runs the length of the radius and is widest proximally where the inclination and slight anterolateral projection of the radial articular surface gives it the most proximal insertion of the antebrachial muscles (Fig. 4).

Another previously named rugosity, the posterolateral triangular rugosity (Burch & Carrano, 2012), is smaller and located just above the distal articular surface at the posterolateral corner of the radius (Fig. 4B,D). This rugosity represents the combined insertion of Abductor radialis and/or Extensor carpi radialis (ECR). These two muscles are closely associated in tetrapods and may have become fused in the reduced forearm of Majungasaurus. In early theropods, Abductor radialis inserts on the lateral surface of the radius and ECR inserts on the radiale. No ceratosaur is known to possess a radiale, so the insertion of this muscle must have shifted or the muscle itself was lost in these taxa. In lepidosaurs with reduced forelimbs, muscles of the antebrachium such as ECR are typically present even when the bones to which they are attached have been lost, instead shifting their attachment sites to other nearby bones (Abdala et al. 2015). In birds, ECR inserts on the carpometacarpus in an area equivalent to the base of metacarpal I (George & Berger, 1966) and, although this may have served as an insertion point in ceratosaurs with more robust manual morphology, supported by a tubercle on the proximolateral surface of metacarpal I in Ceratosaurus (USNM 4735), the first metacarpal of abelisaurids is small and already serves as the point of attachment for another muscle (see below). If this muscle has not been lost, the only other location for its insertion is on the distal radius, as it is in some turtles (Abdala et al. 2008) and teiid lizards with reduced forelimbs (Abdala et al. 2015). The substantial rugosity on the distal radius does not obviously correspond to any other antebrachial muscle and may represent the fusion and common insertion of these two muscles in abelisaurids, where it would serve primarily as an extensor of the forearm.

On the anterior surface of the antebrachium, closely associated tubercles on the proximal edges of the radius and ulna correspond to the common insertion of Biceps brachii and Brachialis, as in early theropods (Fig.��4A). In Majungasaurus the ulnar tubercle (previously called the anterolateral ridge; Burch & Carrano, 2012) is slightly larger than the radial tubercle and this disparity is exaggerated in Aucasaurus and Carnotaurus, indicating that the ulnar attachment may be the primary insertion for this common tendon in abelisaurids. Anterior to the interosseus process of the ulna, Flexor digitorum longus profundus (FDLP) had a long, narrow origin extending along the shaft of the ulna distal to the tubercle for Biceps brachii (Fig. 4A,D). There is a distinct notch in the margin of the otherwise dramatically flaring distal articular surface, just anterior to the radial articular facet, to allow for passage of the tendon of this muscle into the manus. On the posterior surface of the antebrachium, a small Abductor pollicis longus muscle likely took origin from the facing surfaces of the radius and ulna, near the ridges for the interosseus membrane (Fig. 4A,D). Although the manus of Majungasaurus is very reduced, this muscle is often retained in extant taxa with reduced manus (McGowan, 1982; Berger‐Dell'mour, 1983; Abdala et al. 2015), so it is here reconstructed as present. The tendon of this muscle would have reached the manus via a shallow groove in the flaring distal margin of the radius, between the ulnar articular facet and the posterior projection of the radial distal articular surface.

A major muscle attaching to the posterior surface of the ulna but lacking a scar is Anconeus. The ulna of Majungasaurus has a wide, smooth posterior surface that provides a large potential area of insertion for this muscle (Fig. 4B), which has a large, fleshy insertion in early theropods. Inserting on the posterior surface of the ulna, Anconeus would have been one of the primary extensors of the forearm in abelisaurids. The other major antebrachial extensor was Triceps brachii, which inserted on the olecranon process of the ulna (Fig. 4B,C). The olecranon is extremely reduced in Majungasaurus, barely projecting beyond the level of the proximal articular surface, and lacks any striations on its posterior surface. Nevertheless, the insertion of Triceps brachii is extremely consistent among tetrapods and thus it is reconstructed here in accordance with this conserved topology. On the anterolateral surface of the ulna, a low ridge probably demarcates the posterior border of the insertion of Epitrochleoanconeus (Fig. 4C). This muscle's fleshy insertion was restricted to the anterior half of the ulna in early theropods, but some forelimb muscles have been shown to extend proportionately farther distally in a taxon with reduced limbs (Livezey, 1992a), so this muscle may have extended along the relatively short distance of the ulnar shaft in Majungasaurus.

The ceratosaurian wrist contained, at most, one ossified carpal; although no carpals are known from the articulated forelimbs of Majungasaurus, Ceratosaurus (Gilmore, 1920; Carrano & Choiniere, 2016) or Limusaurus (Xu et al. 2009), a single rounded bony element that appears to be a carpal can be seen in the articulated arm of Aucasaurus (MCF‐PVPH 236) and in the recently described Eoabelisaurus (Pol & Rauhut, 2012). A small round bone is also present on the distal ulna of Carnotaurus (MACN‐CH 894) in nearly the same location as preserved in the Aucasaurus arm, but the manus of this taxon is jumbled as preserved so it is unknown whether this represents an homologous element. The transformation of the carpus into cartilage and retention of bony metacarpals has been described in extant lepidosaurs with reduced forelimbs (Fürbringer, 1870), although it is not found in taxa that still possess a well‐developed manus, as in Ceratosaurus. Muscles that typically attach to the carpals are either lost or their attachments shift to a nearby bone when their attachment sites become cartilaginous, as has been described in lepidosaurs (Berger‐Dell'mour, 1983; Abdala et al. 2015). It is most common for distal attachments of muscles inserting on the carpals to shift to the proximal metacarpals (these attachments will be discussed below), but Flexor carpi ulnaris may have taken a more proximal insertion in Majungasaurus (Fig. 4A). Both adult Majungasaurus ulnae (FMNH PR 2836 and UA 9860) possess large round pits on the anterior expansion of the distal articular surface, and in the smaller ulna (MAD 10061) this is represented by a shallow divot in the same area. This pit appears to have been the site of a large tendinous attachment, but most of the muscles inserting on the ulna do so via long, fleshy attachments. Of the muscles attaching to the carpus in this region, only Flexor carpi ulnaris typically has a large enough tendon to make such a pit. An attachment to the distal ulna for Flexor carpi ulnaris has also been described in teiid lizards with reduced forelimb (Abdala et al. 2015). This insertion of Flexor carpi ulnaris would have resulted in the loss of its action on the carpus and the anterior projection of the distal ulna would have made this muscle a strong flexor of the antebrachium.

Manus

Majungasaurus retains four digits in its manus, each bearing one or two phalanges. The phalanx of digit IV is fused to the metacarpal in FMNH PR 2836, and it is unknown whether this is a typical morphology or this phalanx was sometimes mobile. Both the left and right hands preserved in MCF‐PVPH 236 of Aucasaurus also exhibit fusion within digit IV, so here it is assumed that the phalanx of digit IV was also non‐mobile in Majungasaurus. The hands of early dinosaurs possessed several layers of short digital flexors and extensors for fine control of the fingers (Burch, 2014), but Majungasaurus certainly lacked a need for such fine control. The intrinsic manual muscles are the first to be lost and individual muscles with similar actions often fuse in lepidosaurs with reduced forelimbs (Fürbringer, 1870; Berger‐Dell'mour, 1983; Abdala et al. 2015). This, combined with the loss of the bony carpals in ceratosaurs, points to the loss of the superficial layers of each muscle group, which arise from the carpals. However, there are several features that suggest that the deep layers of the short digital muscles were present in the first three digits. On the ventral surface, metacarpals I–III each have a wide, projecting, proximal lip that likely served as the site of origin for Flexor digitorum brevis (FDB) of each digit (Fig. 5B). The slips for digits II and III would have divided to insert on the slightly striated projections at either corner of the proximal edge of the proximal (or only) phalanx. Although these projections are small, a split insertion is also supported by the proximal phalanges of Aucasaurus, which possess very large tubercles in these locations for the attachment of these tendons. Digit I has only one phalanx, and the ventral surface of phalanx I‐1 has a major central ridge and proximal tubercle. The FDB slip to digit I typically does not divide and inserts only on the medial side of the phalanx, so it is reconstructed with this morphology here (Fig. 5B). The central tubercle of phalanx I‐1 would have been the insertion point of the tendon of Flexor digitorum longus (FDL) to digit I (Fig. 5B). The FDL tendons to the other digits would have inserted on the proximal surfaces of the distal phalanges. It is uncertain whether digit III possessed one or two phalanges, so the FDL tendon is tentatively reconstructed as inserting on the fused potential second phalanx of this digit (Fig. 5B).

Dorsally, Extensores digitores breves (EDB) are the only muscles that insert on the distal phalanges and extend each digit, so they would have potentially been present in all digits possessing mobile phalanges. However, the EDB are also one of the first muscle groups lost among lepidosaurs with reduced forelimbs (Abdala et al. 2015), so it is possible they were also lost in Majungasaurus. If present, these muscle slips likely originated from the wide, flat dorsal surfaces of metacarpals I–III and traveled a short distance to insert on the proximal edge of the distal phalanx of each digit (Fig. 5A). The tendons of Extensor digitorum longus (EDL) inserted on the base of each metacarpal, just proximal to the origin of EDB (Fig. 5A). Although an insertion by EDL on metacarpal IV is phylogenetically equivocal and not reconstructed in the early theropod Tawa (Burch, 2014), an attachment is reconstructed here based on the relative metacarpal sizes. In Tawa, metacarpal IV is very small relative to the other metacarpals but in abelisaurids, metacarpal IV is as large or larger than the other metacarpals. Additionally, abelisaurids apparently exhibited an extreme range of manual extension (as preserved in Aucasaurus), which would be better served by having an extensor slip attaching to each digit.

The relatively large metacarpal IV would have also served as an attachment site for the insertion of Extensor carpi ulnaris (Fig. 5A). This muscle is reconstructed as inserting on both the pisiform and the lateral‐most metacarpal in early theropods, and solely on the lateral‐most metacarpal once the pisiform is lost in more derived taxa (Burch, 2014). The proximal surface of metacarpal IV in Majungasaurus is highly angled such that the lateral side projects proximally, and this morphology is even more exaggerated in the manus of Aucasaurus. This projection corresponds well to the insertion of Extensor carpi ulnaris, where it would primarily provide extension and ulnar deviation of the wrist. On the opposite side of the manus, the tendon of Abductor pollicis longus would have inserted on the medial side of the first metacarpal (Fig. 5A). Although Majungasaurus lacks the prominent medial projection of the first metacarpal seen in early theropods, the proximal edge of the medial surface is characterized by a flat, slightly projecting surface for the insertion of this muscle. Two small manual muscles responsible for detailed movements of the digits, Abductor pollicis brevis and Abductor digiti minimi, were likely lost from the reduced abelisaurid manus. The digits of Majungasaurus probably did not have much autonomy, so it is unlikely that these muscles would have been maintained in the manus.

Comparative myology

The scapulocoracoid morphology of Majungasaurus suggests few major shifts in the musculature of the shoulder girdle from the basal condition (Fig. 6). The most prominent of these are the changes caused by the parallel scapular margins in abelisaurids, which have the greatest effect on the origins of Deltoideus scapularis and Subscapularis. The relative narrowing of the distal scapular blade (i.e. loss of the distal flare) and the relative widening of the proximal scapular blade results in a shift in the distribution of the muscle fibers such that a much larger proportion of the muscle is located proximally. The origin of Scapulohumeralis posterior was also shifted proximally in abelisaurids, whereas Ceratosaurus (UMNH VP 5278), Limusaurus (IVPP V15923), and Masiakasaurus (UA 9160) maintained a more plesiomorphic origin further from the edge of the glenoid. These muscles are the primary retractors of the humerus and by shifting their origins proximally, the length of their moment arms, and thus the torque they would be able to generate, was reduced. This is somewhat counteracted in both Deltoideus scapularis and Scapulohumeralis posterior by the slight distal displacement of their insertion sites on the humeral shaft, which served to lengthen their lever arms. The insertion for Deltoideus scapularis in abelisaurids also is very poorly developed relative to that in more basal taxa including Ceratosaurus, which possesses a robust, rugose greater tubercle.

The insertion site for Subscapularis is very well developed in abelisaurids, although the orientation of the internal tuberosity is such that its projection creates a line of action that is more advantageous for medial rotation than humeral retraction. The projecting internal tuberosity may also be driven by the insertion of Subcoracoideus, which has a large potential area of origin on the medial surface of the coracoid, providing a substantial capacity for humeral adduction. The shift of more of the muscle fibers of Subscapularis proximally also improved this muscle's role as an adductor of the humerus, though it would have still remained a secondary action. Abelisaurids maintain the wide triangular area of origin for Deltoideus clavicularis as seen in Ceratosaurus as well as in earlier theropods (e.g. Dilophosaurus, UCMP 37302), but the reduction of the deltopectoral crest greatly limited the potential area of insertion as well as nearly eliminated any protractional capabilities of this muscle.

The role of the humeral protractor in abelisaurids is filled by the Supracoracoideus complex of muscles, whose large potential area of origin from the subacromial depression and lateral surface of the coracoid indicates substantial development of this muscle group in ceratosaurs. Because the projection of the deltopectoral crest is extremely low in abelisaurids, this muscle group would have inserted nearly flush with the humeral shaft and thus had a shorter moment arm for protraction. Its moment arm for abduction was also reduced due to the medial deviation of the deltopectoral crest onto the anterior shaft of the humerus. Protraction of the humerus was also assisted by Coracobrachialis. Abelisaurids retained a relatively large area of origin for this muscle as well as a subglenoid fossa that faces primarily posteroventrally, which resulted in a more direct line of action for this muscle. Although the insertion is somewhat restricted in Majungasaurus, the humeri of Aucasaurus possess large, low rugose areas distal and medial to the anterior fossa that may indicate a larger area of insertion in this taxon.

The Pectoralis muscle of Majungasaurus may have had a broad origin that included the anterior edge of the scapular blade, but the reduced deltopectoral crest again pulled the point of insertion close to the humeral shaft and shortened the moment arm for this muscle. Similarly, the insertion of Latissimus dorsi moved proximally relative to its position in early theropods, thus shortening the lever arm of this muscle for retraction of the humerus, but this arrangement was also present in Ceratosaurus (UMNH VP 5278) and Limusaurus (IVPP V15924). Triceps brachii retained all three of its heads in abelisaurids, although the scapular head seems to have had a smaller, less well‐developed, and less rugose origin than in other large theropods, including Ceratosaurus. There is no indication of reduction in either of the other two heads, and scarring around the origin of the lateral head seems to indicate that this part was well developed. However, the olecranon processes on the ulnae of Majungasaurus and Aucasaurus are very reduced and hardly project beyond the proximal articular surface. This is not the case in Carnotaurus, which possesses an extremely well‐developed olecranon process bearing slight striations on its posterior surface.

The lack of any rotation of the antebrachial elements relative to each other would have resulted in the loss or modification of all the antebrachial muscles with roles in pronation or supination. In Majungasaurus and Aucasaurus the entepicondyle projects anteriorly, reorienting the lines of action of these muscles to flexion. The ectepicondyle is less well‐developed in both taxa, although in Aucasaurus it is not quite as reduced as in Majungasaurus, and it is oriented more posteriorly, enhancing the extensor action of most of the muscles originating from this epicondyle. The exception to this is Supinator, a flexor, whose origin had to move proximally in order to maintain this role. In Carnotaurus, both epicondyles are large and retain their medial and lateral orientations, similar to the arrangement in Ceratosaurus and other early theropods.

The insertions of the antebrachial muscles of abelisaurids were highly modified and difficult to compare to the basal condition. Furthermore, little is known about the antebrachium of abelisauroids or about any transitional stages in the evolution of this bizarre morphology within the clade. Even the basal abelisaurid Eoabelisaurus possesses a radius and ulna that are morphologically more similar to those of early theropods than to those of more derived abelisaurids (Pol & Rauhut, 2012). Although slightly more elongate than those of Majungasaurus and Carnotaurus, the antebrachial elements of Aucasaurus bear the same general morphology and arrangement of scars and tubercles, though with a few notable differences. The radius of Aucasaurus lacks a distinct posterolateral triangular rugosity for the insertion of Abductor radialis and Extensor carpi radialis and, although the fleshy insertion of Abductor radialis may not have left a scar on the posterior surface of the radius, the tendinous insertion of ECR likely would have. It is possible that in Aucasaurus the insertion of ECR was located on the medial surface of metacarpal I; this insertion corresponds with the area of insertion in birds (George & Berger, 1966), and it is also the likely area of insertion in Ceratosaurus and other ceratosaurs that had well‐developed hands but lacked an ossified carpus. In this case, the muscle would have retained its ability to extend and possibly abduct the wrist.

Another distinct scar in Aucasaurus is a large rugosity located distally on the medial surface of the ulna. This rugosity does not obviously correspond to any muscle, but it may represent a distal insertion of Epitrochleoanconeus in this taxon. Flexor carpi ulnaris may have had a different insertion point in Aucasaurus and Carnotaurus, both of which lack the anterior pit on the distal articular surface present in Majungasaurus. This is possibly related to the round bone found at nearly the exact same location at the anteromedial corner of the distal articular surface of the ulna in the right arm of Aucasaurus and the left arm of Carnotaurus. Located in the vicinity of the insertion of Flexor carpi ulnaris (FCU), this bone may either represent a carpal or potentially a sesamoid found in the tendon of FCU, assisting it to wrap around the projecting distal articular surface of the ulna. In either case, it is hypothesized here that FCU inserted on this bone, which was then connected distally to the lateral side of metacarpal IV via a ligament or the continuation of the FCU tendon. The insertion of Extensor carpi ulnaris (ECU) also had an attachment to the pisiform in early theropods, but it also attached to the lateral‐most metacarpal, which typically possesses a lateral fin that relates to this insertion (Burch, 2014). This morphology is present in the metacarpal IV of Ceratosaurus (USNM 4735), but this metacarpal does not display increased development of this fin due to the loss of the pisiform. Both the left and right metacarpal IVs preserved in the holotype of Aucasaurus, however, bear extremely well‐developed lateral processes that also project proximally and, although the metacarpal IV of Majungasaurus is relatively reduced, it still has a distinct proximal inclination laterally.

The intrinsic manual musculature was likely reduced to a single layer in all ceratosaurs given the lack of more than one ossified carpal. However, in lepidosaurs, Flexores digitorum breves superficiales originate from an annular ligament (Russell & Bauer, 2008), so it is possible that this layer was retained at least in ceratosaurs with well‐developed manus and possessed a similar non‐bony origin. The manus of Aucasaurus possesses a digit I that is more reduced than that of Majungasaurus, consisting of a pyramidal metacarpal lacking a phalanx or even a distal articular surface for a phalanx. In this case, the short flexors would have been reduced to only the second and third digits. However, unlike that of Majungasaurus, the hands of Carnotaurus and Aucasaurus both seem to have possessed unguals. In Aucasaurus, the distal phalanx of digit II possesses a distal articular surface and a pit for a collateral ligament on its medial condyle, and the manual elements known from Carnotaurus seem to include unguals, although further preparation of the specimen is necessary to confirm this. An ungual present on digit II in these taxa would serve as the insertion point for the tendons of Flexor digitorum longus and Extensor digitorum brevis for this digit, as in earlier theropods. Aucasaurus and Ceratosaurus both possess very large tubercles on the proximal ventral surfaces of phalanges II‐1 and III‐1; in both taxa these tubercles are greatly enlarged relative to the overall length of the phalanx, and indicate substantial flexor musculature despite the small size of the phalanges.

Consequences of extreme forelimb reduction in abelisaurids

As is typical in tetrapod limb reduction (Fürbringer, 1870; Lande, 1978), forelimb reduction in abelisaurids is greatest in the distal elements. Abelisaurids take this to the extreme among taxa that retain all of the distal limb elements; although their antebrachial and manual elements are among the shortest relative to body size in Theropoda, their scapulocoracoids do not exhibit reduction in comparison with basal ceratosaurs or indeed other large theropods. The retention of large scapulocoraoids in nonavian theropod taxa with reduced limbs may be due to a close developmental association between the scapular blade and the axial skeleton (Kuijper et al. 2005; Valasek et al. 2011; Dececchi & Larsson, 2013), but some of the muscles attaching to the scapulocoracoid possibly had important roles in other activities. Those muscles attaching to the neck (e.g. Levator scapulae) could have played a part in feeding, as may be the case in extant crocodylians (Meers, 2003), and those muscles attaching to the ribs (e.g. Serratus muscles) could have had a role in respiration, as they appear to have in extant birds and crocodylians (Codd et al. 2005; Munns et al. 2012). Although several studies have investigated the functional morphology of respiration and craniocervical muscle dynamics in nonavian dinosaurs (e.g. O'Connor & Claessens, 2005; Snively & Russell, 2007; Tsuihiji, 2010; Tickle et al. 2012), these studies focused on the axial musculature, and the role of the muscles attaching to the pectoral girdle has not been explored. Furthermore, these muscles predominantly attach to the perimeter of the scapulocoracoid, so their potential areas of origin would be increased through lengthening of the scapular blade, not by widening it, as is demonstrated in the scapula of Majungasaurus. Muscles attaching to the flat surfaces, such as Deltoideus scapularis and Subscapularis, maintain relatively wide, fleshy sites of origin. At the same time, movement of the origins of these muscles toward the glenoid fossa, as well as reduction of the deltopectoral crest in the humerus, result in an overall shortening in moment arms of most of the muscles crossing the glenohumeral joint. A shorter moment arm reduces the torque a muscle can produce for a given action but creates a large angular displacement for the distal end of the bone. That the shoulder of Majungasaurus was well‐suited for a wide range of motion is also supported by the presence of a bulbous, hemispherical humeral head, which would have allowed the humerus to move in nearly any direction.

The morphology of the antebrachial elements of Majungasaurus is distinct from that of any extant tetrapod, including those with reduced limbs. Besides their unusual shape, the presence of substantial muscle scars on their surfaces indicates the retention of a well‐developed musculature and a lack of vestigialization of the distal elements. The radius and ulna of the vestigial forelimbs of the kiwi and emu have been described as simple (McGowan, 1982) and ‘essentially featureless’ (Maxwell & Larsson, 2007: p. 428) but, in ostriches, which make use of their forelimbs for display and other purposes (Davies, 2002), the radius and ulna retain scars and distinct intermuscular lines (S.H. Burch, pers. obs.). These differences in osteology are reflected in the development of the musculature: the antebrachial musculature of the former two species, particularly that of the emu, is highly reduced and many muscles have been lost altogether (McGowan, 1982; Maxwell & Larsson, 2007), whereas the ostrich retains a nearly full complement of antebrachial muscles (S. H. Burch, in prep.). The forelimbs of emus and kiwis are also highly osteologically and myologically variable, which has been suggested to be a result of increased evolutionary drift after a loss of function (McGowan, 1982; Maxwell & Larsson, 2007), but in the sample of Majungasaurus that has been collected thus far, the morphology of the forelimb elements is relatively consistant. In lepidosaurs with forelimbs that show extreme reduction to a single digit, the distal muscles have undergone a reduction and fusion similar to that seen in the emu, and the osteology is similarly indistinct (Berger‐Dell'mour, 1983; Abdala et al. 2015). The intrinsic manual musculature is some of the most variable among lizards with reduced forelimbs, and the most likely to be lost even when the digits to which they plesiomorphically attach are retained (Abdala et al. 2015). On the other hand, muscles of the brachium and antebrachium are typically plesiomorphically retained even when the elements they attach to are reduced, suggesting that these muscles can be reconstructed with some confidence (Abdala et al. 2015). Although distinct muscle scars present on the antebrachium in Majungasaurus provide evidence against extensive fusion of these muscles, the actions of the individual antebrachial muscles have all converged on either flexion or extension, limiting the potential movements of the elbow.

The highly reduced manus of Majungasaurus is unusual in that it retains the plesiomorphic ceratosaurian state of four digits. The first stages of limb reduction in tetrapods are typically a reduction in the number of phalanges and loss of pre‐ and post‐axial digits. In lepidosaurs, digit reduction and loss typically follows the sequence of I > V > II > (III, IV), with digits III and IV being retained in taxa that also possess a manus (Fig. 7A), whereas in theropod dinosaurs and birds, this sequence is V > IV > III >I > II (Fig. 7B; Shapiro et al. 2007). Although these two sequences differ in order, they both share the retention of one or two larger central digits subsequent to complete loss of the pre‐ and post‐axial digits. This type of forelimb reduction is found in other theropods, such as tyrannosaurs (Fig. 7C), but it is notably not present among ceratosaurs. Basal ceratosaurs and abelisauroids do show a substantial reduction in the medial and lateral digits relative to digits II and III (Gilmore, 1920; Xu et al. 2009; Pol & Rauhut, 2012) but, unlike the typical progression, these digits are not lost in derived abelisaurids and the metacarpals become nearly subequal in length (Burch & Carrano, 2012). Uniform reduction of the manus, in which the metacarpals are subequal and one or two short phalanges are present on each digit, is not as common and is found in large graviportal animals including tortoises, sauropods, and elephants (Fig. 7D,E; Shapiro et al. 2007). As it is virtually impossible for the abelisaurid forelimb to have been used in such a manner, the reasons for its convergence on this type of reduction are unclear. Intriguingly, this morphology can also be seen in the amphisbaenian Bipes and the skink Bachia, and may indicate the re‐evolution of lost digits (Brandley et al. 2008; Abdala et al. 2015). The retention of the medial and lateral metacarpals in abelisaurids may be a result of their roles as attachment sites for Abductor pollicis longus and Extensor carpi ulnaris, respectively, and the evidently large range of motion possible in the wrist of abelisaurids. The bulbous distal articular surfaces of the radius and ulna would have allowed not only substantial flexion and extension (the latter being preserved in the right forelimb of Aucasaurus), but also potentially a relatively wide range of abduction and adduction of the wrist.

With the loss of mobile phalanges in digit IV (and digit I in Aucasaurus), the small intrinsic muscles of the hand also disappeared in these digits. The extremely small size of the remaining phalanges would have hardly extended beyond the palm of the hand, yet the large ventral tubercles on the phalanges of Aucasaurus provide evidence that the short flexors of digits II and III were retained. As in the antebrachium, the condition of the manual muscles in extant flightless birds seems to reflect the utilization of the wing. Although the nonfunctional wings of kiwis and emus actually possess more phalanges in the major digit than is seen in other birds, they have lost all intrinsic manual muscles (McGowan, 1982; Maxwell & Larsson, 2007), whereas the ostrich retains all muscles present in volant birds and even exhibits primitive muscles not found in other avian taxa (Burch, 2014; S.H. Burch, in prep.). The flightless Galápagos Cormorant, which displays the greatest degree of wing reduction among extant neognaths, also retains a complete manual musculature and possesses an additional intrinsic muscle to the alula absent from other birds; although they do not engage in wing‐propelled diving, Galápagos Cormorants likely do make use of their wings and particularly the feathers attaching to the alula during dives for hydrodynamic stabilization (Livezey, 1992a). The potential retention of the intrinsic manual musculature in the central digits in the hand of abelisaurids suggests that these digits possess some function, even at their extremely reduced size.

Among extant flightless birds, functions of non‐vestigial wings include display, balance during turns in running, and shading the nest (Davies, 2002). The forelimbs of abelisaurids such as Majungasaurus did not have enough mass or span to be of use in the latter two functions, but even at their small size the forelimbs of abelisaurids may have been useful for species‐recognition or mating displays. Another possibility is that of stimulation of the partner during mating such as that performed by boid snakes, using their highly reduced hind limbs in a wide arc to brush the female (Murphy et al. 1978; Gillingham & Chambers, 1982). The osteology and myology of abelisaurids such as Majungasaurus indicate that the forelimb was able to be moved in a wide range of motion, which is potentially congruent with both of these functions. It is currently unknown whether the forelimbs of abelisaurids were sexually dimorphic structures, as might be expected if they played an important role in display or copulation, but future fossil discoveries may help to further evaluate these functional hypotheses.