Among limbed (and many limbless) amniotes, elements of the shoulder girdle and sternum partially encircle the ribcage to form a structural yoke that indirectly connects the forelimbs to the axial skeleton. Herein referred to as the pectoral apparatus, this skeletogenous nexus demonstrates enormous taxonomic variation and multiple (independent) instances of elements appearing to have been lost and/or insensibly combined. Furthermore, the pectoral apparatus includes both dermal and replacement bones. These features underline the difficulty of determining homology for various pectoral components.

Historical observations from both the fossil record and ontogenetic studies initially provided a well-structured scenario of the morphological transformation of mammals from the earliest stem synapsids. However, subtle yet important evolutionary details are only now being understood through the use of experimental methods and detailed developmental analyses at the level of the cell.

Among reptiles, sister group to mammals, the situation is less clear. Despite an increased knowledge of embryology and phylogenetics, and an abundance of new palaeontological data, the prevalent hypothesis regarding the homology of the reptilian coracoid has remained largely unchanged for more than 80 years. Unlike in mammals, there has been little consideration given to the development of the reptilian pectoral apparatus as a whole. This paper has three main objectives: (1) to review the morphology and development of the pectoral apparatus in both extinct and extant amniotes; (2) to re-address homology of the reptilian coracoid; and (3) to re-evaluate the scenario of coracoid evolution in amniotes.

Fig. 1 Evolution of the tetrapod pectoral apparatus. Among basalmost non-digit-bearing tetrapods (A), the pectoral apparatus is dominated by dermal elements; the scapulocoracoid is small and obscured laterally. With the advent of digits (B), taxa demonstrate (more ...)

With the evolution of digit-bearing limbs (Fig. 1B) elements linking the head skeleton with the pectoral apparatus were lost, resulting in a decoupling of the dermal girdle from the skull (Clack, 2002). Furthermore, the once modest scapulocoracoid adopted a more prominent and laterally facing profile. Further reduction of the dermal girdle (in particular the cleithrum) and enlargement of the primary girdle characterizes more deeply nested non-amniote tetrapods (Fig. 1C). Among these and more derived tetrapods, the singular scapulocoracoid is partitioned into two elements, a dorsal scapula and a ventral coracoid.

Among basal amniotes (Fig. 1D,E) the lone coracoid is replaced by two discrete components, a cranial procoracoid and a caudal metacoracoid (see Table 1 for a listing of coracoid synonyms). For modern taxa, the scapula is the most widely conserved element of the pectoral apparatus, absent only in some lepidosauromorphs (e.g. ophidians). Monotremes are unique in retaining both the procoracoid and the metacoracoid, whereas therian mammals (marsupials and eutherians) and reptiles (including birds) each maintain only one of the two elements; in therians the procoracoid no longer forms a discrete entity and the metacoracoid is a rudiment that fuses with the scapula, giving rise to the coracoid process. Conversely, in reptiles it has long been assumed that the metacoracoid has disappeared and only the procoracoid remains. The procoracoid is understood to be equivalent to the coracoid of non-amniote tetrapods. Consequently, modern authors typically refer to this element in reptiles as simply the coracoid (e.g. McGonnell, 2001), a designation adopted hereafter to avoid unnecessary confusion.

Table 1 Synonyms (and source listings) of the cranialmost and caudalmost coracoid elements present in basal amniotes and monotremes, and their inferred homologues in therians and reptiles

In comparison with the rest of the pectoral apparatus, the sternum is often poorly preserved or entirely absent from fossil taxa. This observation is usually explained in the context of histology; unlike other pectoral components that ossify, the sternum often remains cartilaginous. Notwithstanding its poor palaeontological representation, the sternum is considered characteristic of virtually all tetrapods [including lissamphibians and most amniotes, with the exception of turtles (although see below) and ophidians], with preserved material dating back to at least the Late Permian (256–248 million years ago; Vaughn, 1955).

Prior to the late 1800s, it was commonly held that of the two coracoid elements present in basal amniotes and monotremes, only the caudalmost element, the metacoracoid, was retained by modern reptiles and therians (e.g. Parker, 1868; Flower, 1876; Fig. 2A). The cranialmost element, the procoracoid, was assumed to be lost. Soon, however, new developmental and palaeontological evidence began to confuse the issue. Howes (1887, 1893) observed that the developing scapula of a generic rabbit consisted of three discrete centres – one giving rise to the blade and scapular spine, one forming the coracoid process and one contributing to the glenoid. He interpreted these centres as corresponding to the scapula, procoracoid and metacoracoid of monotremes, therefore suggesting that the procoracoid was not lost (Fig. 2B).

Fig. 2 Competing hypotheses of coracoid homology. (A) Among many early workers (e.g. Parker, 1868) the procoracoid of basal taxa was assumed to be lost in more deeply nested forms. Consequently, the coracoid process of therians and the coracoid element of modern (more ...)

An alternative opinion was expressed by Lydekker (1893), who compared the development of the coracoid process of a sloth with the dual coracoid elements of what was then considered a basal reptile (namely †Dicynodon sp., now considered to be a basal synapsid). He concluded that the procoracoid was homologous with the therian coracoid process of the scapula, and that the caudalmost element, what he named the metacoracoid, was the only coracoid element of reptiles (Fig. 2C). This was later rebutted by Broom (1899), who returned to the notion that modern amniotes, with the exception of monotremes, had lost the procoracoid.

Although aspects of the debate continued, e.g. Gregory & Camp (1918; see also Hanson, 1920) argued that basal amniotes originally had three coracoid elements – the epicoracoid, coracoid and metacoracoid, the work of Williston (1911, 1925) is often cited as the basis for our modern interpretation of reptilian coracoid homology (Broom, 1912; Case & Williston, 1913; Hanson, 1920; Romer, 1922, 1956). Williston examined a variety of what were then considered to be fossil ‘reptilian’ taxa (most of which have since been recollected as Amniota outgroups) and noted that whereas the scapula and procoracoid were generally fused, the metacoracoid, if present at all, was often poorly sutured or completely unattached. Williston and others (e.g. Broom, 1912; Case & Williston, 1913) interpreted this as evidence for the anagenetic disappearance of the metacoracoid among modern reptiles.

Adopting the interpretation of Williston, Romer (1956; see also 1922) expanded the argument. Observing that non-amniote tetrapods (e.g. †Seymouria spp., lissamphibians), therians and modern reptiles have only one coracoid element, whereas some basal amniotes (e.g. †Dimetrodon spp.; see below) had two. Romer formulated an evolutionary scenario (Fig. 3) in which the procoracoid was the primitive element. This procoracoid was retained by reptiles and some synapsids (e.g. monotremes). Among basal synapsids a second more caudal element, the metacoracoid, developed. In therians the procoracoid disappears whereas the metacoracoid is retained, albeit in a reduced form as the coracoid process of the scapula. Among those fossil reptiles that demonstrated both coracoid elements, Romer suggested that for some taxa this reflected a proximate genealogy with synapsids. For others he stated ‘[w]e must assume either that the general reptile stem early acquired a second coracoid element which survived only in synapsids, other reptiles rapidly losing it again, or that parallelism occurred, with the development of a second coracoid in two or more lines. The second assumption is the more reasonable, but the situation is far from clear.’ (Romer, 1956, p. 309). Indeed, the situation remains uncertain, for as alluded to above, revised phylogenetic hypotheses of Amniota (Fig. 4) have weakened some of the fundamental underpinnings to the theory of reptilian coracoid homology as established by Williston (1911, 1925).

Fig. 3 Romer's interpretation of amniote primary girdle evolution. Building on the work of Williston (1911, 1925), Romer hypothesized that the primitive coracoid of non-amniote tetrapods such as †Seymouria, the procoracoid, was retained as the coracoid (more ...)

Fig. 4 Revised amniote phylogeny based on the work of Brochu (2001), Lee (1996), Luo et al. (2002), Rieppel & Reisz (1999) and Ruta et al. (2003). As the phylogenetic position of turtles remains confused, both competing hypotheses are presented (i.e. (more ...)

Unique among modern amniotes, adult monotremes retain each of the scapula, procoracoid, metacoracoid, clavicle, interclavicle and sternum (Fig. 5A,B). Despite the rarity of available material, morphogenesis of the monotreme pectoral apparatus has been well documented (Klima, 1973, 1985). During development, each ipsilateral scapula, procoracoid and metacoracoid arises from a common homogeneous endochondral condensation, the coracoid–scapular plate (Fig. 5C; Klima, 1973). The glenoid (for articulation with the humerus) develops at the intersection of the scapula and metacoracoid, to the exclusion of the procoracoid. Midventrally, these elements are bridged by a midline T-shaped interclavicle.

Fig. 5 Monotreme pectoral apparatus. An adult Ornithorhynchus anatinus pectoral apparatus in ventral (A) and left lateral (B) views. Among skeletally mature monotremes, the pectoral apparatus is composed of six elements: an unpaired sternum and interclavicle, (more ...)

Developmentally the interclavicle is exceptional, receiving contributions from two skeletogenetic sources: an unpaired endochondral condensation (the pars chondralis interclaviculae) that combines with paired intramembranous condensations (the pars desmalis interclaviculae) to form a composite anlagen (Fig. 5C; Klima, 1973, 1987). A pair of slender clavicles form along the cranial margin of the interclavicle, each derived from a single intramembranous condensation. The skeletally mature sternum includes a cranialmost manubrium sterni, followed by a series of variably co-ossified block-like sternebrae (and in tachyglossids, but not Ornithorhynchus anatinus, a terminal xiphisternum), all of which develop from a longitudinal pair of condensations, the sternal bands (Klima, 1973).

In contrast to adult monotremes, the pectoral apparatus of therians is composed of fewer osteological components, with only the scapula and sternum present in all forms. Adult therians lack a discrete interclavicle (although see below), the clavicles may be reduced or completely absent (e.g. ungulates, peramelid marsupials; Flower, 1876; see also Hall, 2001), and the coracoid elements are either rudimentary or vestigial (Klima, 1973, 1985, 1987).

The scapula is the only therian pectoral apparatus element to articulate with the forelimbs. Although morphologically variable in profile (Fig. 6A–C), virtually all therian scapulae have a coracoid process adjacent to the glenoid and at least one conspicuous scapular spine that divides the lateral surface into two relatively large, shallow fossae (Sánchez-Villagra & Maier, 2002). The distal terminus of the scapular spine gives rise to a second process, the acromion, that comes into close proximity or articulates with the clavicle, when present.

Fig. 6 Therian scapula morphology and pectoral apparatus development. Adult (right) scapulae of (A) Canis, (B) Priodontes and (C) Mus in lateral view, demonstrating the morphological variation of this element. (D) Early during skeletogenesis the cellular condensations (more ...)

Like monotremes, therians develop a coracoid–scapular plate early during morphogenesis that becomes dominated by the presumptive scapula. Although the procoracoid precursor is initially present, it does not develop beyond a rudiment, merging with the sternal bands and a median (unpaired) endochondral condensation (Fig. 6D; discussed below), or a vestige nested between the clavicle and manubrium sterni. Among various marsupial lineages (e.g. dasyurids) these vestiges, the praeclavia (Klima, 1987), are relatively common, and may remain cartilaginous or ossify with age. For eutherians such vestiges (termed suprasternals; Hanson, 1919; Klima, 1968, 1972, 1987) are rare and considered to be atavisms.

Similar to the procoracoid, the metacoracoid primordium develops only as a rudiment. Initially it joins the procoracoid in contacting the manubrium sterni. This connection is transient, however, and eventually disappears, with the metacoracoid remaining firmly affixed to the scapular condensation (Fig. 6A). Ultimately, the metacoracoid gives rise to the coracoid process. In some taxa (e.g. the marsupial Monodelphis domestica, the insectivores Crocidura russula and Suncus etruscus) the metacoracoid condensation also contributes to the glenoid (Großman et al. 2002; Sánchez-Villagra & Maier, 2002).

The acromion develops prior to the scapular spine, arising from a cartilaginous primordium that splits off the scapula (Großmann et al. 2002; Sánchez-Villagra & Maier, 2002, 2003). In most taxa, the scapular spine subsequently forms by appositional ossification of the scapular perichondrium (e.g. Monodelphis, Crocidura; Sánchez-Villagra & Maier, 2002; Großmann et al. 2002). However, in the rodent Mesocricetus auratus, the spine reportedly develops from a cartilage precursor stemming from the acromion (Großman et al. 2002).

Experimental work on the laboratory rodent, Mus musculus, illustrates that normal development of the mature scapula is dependent on expression of Pax1, Hoxc6 and Emx2 genes. The acromion is greatly affected in mutants deficient for Pax1 (undulated mutants), and may be completely absent, present and grossly deformed, or present and fused with the clavicle (compare Fig. 6E and F; Timmons et al. 1994; Dietrich & Gruss, 1995). By contrast, defects of the rest of the scapula including the scapular spine involve a relatively minor reduction in size. Similarly, the scapular blade remains virtually unchanged in mice with a mutation in the platelet-derived growth factor α gene receptor (PDGFαR; required for normal patterning of the somites) or Hoxc6, whereas the acromion (PDGFαR; Soriano, 1997), or acromion and glenoid (Hoxc6; Pellegrini et al. 2001) are altered. In another dramatic deviation, the scapular blade fails to form in Emx2 homozygous knockout mutants, with only the coracoid process, acromion and glenoid of the scapula developing (Pellegrini et al. 2001). Data from the mouse mutants are consistent with the individuation of pectoral elements, which is consistent with an independent evolutionary origin.

The bulk of the sternum is derived exclusively from condensations of lateral plate mesoderm, the sternal bands. As noted earlier, the cranial end of the sternum, the manubrium sterni, receives contributions from three distinct sources: the procoracoid rudiments, the sternal bands and an unpaired cartilaginous condensation (Fig. 6D). Based on location, morphology and histology of the unpaired condensation, Klima (1987) identified this unmatched rudiment as the homologue of the pars chondralis interclaviculae of monotremes. As for the scapula, mutations in Pax1 lead to defects of the sternum (Timmons et al. 1994).

Unique to mysticete (toothless) cetaceans, the sternal bands fail to form during embryogenesis. Among adults, however, the sternum is present, derived exclusively from the procoracoid rudiments and the pars chondralis interclaviculae (Klima, 1978, 1985, 1987, 1990).

Fig. 7 Evolution of the synapsid primary girdle. Among basal synapsids (A), the primary girdle consists of the scapula and the large, plate-like procoracoid and metacoracoid. All three elements contribute to the glenoid. Among more deeply nested non-mammalian (more ...)

Among reptiles the morphological diversity of the pectoral apparatus is immense. In the interest of clarity, this review recognizes three structurally similar groups: ornithodirans (birds and their ancestors), non-avian diapsids (a paraphyletic group consisting of lepidosauromorphs, crocodylians and their fossil relatives) and Testudines (turtles).

Fig. 8 The avian primary girdle and sternum. Although the scapula and coracoid initially begin as a common condensation, the two elements separate from one another and ossify independently. Furcula not depicted. Cartilage shaded. Abbreviations: cor (coracoid), (more ...)

Using the domestic fowl Gallus gallus, it has been established experimentally that the scapula is derived, at least in part, from an embryological source distinct from the rest of the pectoral apparatus. Whereas the coracoid and sternum are derivatives of lateral plate mesoderm, chick-quail transplantation experiments by Chevallier (1975, 1977; see also Beresford, 1983) have shown that most of the scapula arises from somites. More recently, Huang et al. (2000) demonstrated that the proximal-most portion of the scapula (the area around the glenoid), like the coracoid, is derived from lateral plate mesoderm. Other experimental observations further reinforce this idea. While Talpid³ chick mutants demonstrate several primary girdle defects, the deviant scapula morphology (resembling an open chevron with tooth-like projections) is considerably more severe than the modestly foreshortened coracoid (Fig. 9; Ede & Kelly, 1964). Work examining the role of Pax1 found that repression of this gene by application of BMP-2- and -4 soaked beads resulted in defects of the proximal part of the scapula and (in some instances) the coracoid, including coracoid aplasia (Hofmann et al. 1998). Localized application of retinoic acid (via bead implants) or tissue grafts of the polarizing region of an exogenous limb bud to the developing wing buds of Gallus enlarges the Hoxc6 (formerly X1Hbox 1) gradient and may result in fusion of the scapula with the coracoid, as well as duplication or foreshortening of the coracoid and absence of the procoracoid process (Oliver et al. 1990; see also Williams, 2003). Thus, patterning of the procoracoid process appears to be at least partially independent from the rest of the coracoid.

Fig. 9 Normal and abnormal development of the scapula and coracoid in 11-day-old Gallus embryos stained for cartilage, with all other elements omitted and limbs removed. Normal embryos, in (A) dorsal and (B) lateral views. Talpid3 mutant embryos, in (C) dorsal (more ...)

Whereas modern researchers have largely overlooked the development of the procoracoid process, early anatomists recorded that it developed independent of the rest of the avian coracoid (Fig. 10A; Parker, 1868; Newton & Gadow, 1893–1896). Indeed, Parker remarked that ‘the præ-coracoid (procoracoid) is always segmented from the head of the coracoid’ (p. 143), and noted that it may become mineralized independent of the latter. Lindsay (1885) and Broom (1906b) later examined the development of the procoracoid process in Struthio camelus (the ostrich; Fig. 10B) but arrived at conflicting interpretations. Whereas Lindsay argued that the procoracoid process developed independent of the coracoid, according to Broom the procoracoid process represents a ‘descending process of the scapula’ (p. 357). Broom further argued that the procoracoid process could not be homologous with the procoracoid of basal amniotes as such a bone was not present in any immediate ancesstors of modern avians (although see below).

Fig. 10 Development of the avian procoracoid process. Many avian taxa have been observed to develop a discrete skeletal centre for the procoracoid process (A), a medial projection that contacts the furcula. (B) In Struthio the procoracoid process is hypertrophied (more ...)

Fig. 11 Evolution of the ornithodiran primary girdle. (A) Among non-avian ornithodirans (e.g. †Deinonychus) the primary girdle consists of a strap-like scapula and a large, plate-like coracoid. (B) The coracoid of the basal avian †Archaeopteryx (more ...)

Fig. 12 Non-avian reptile scapulae and coracoids. (A) The pseudosuchian (crocodylian) Alligator and (B,C) the lepidosauromorphan (squamate) Shinisaurus. (C) Evidence from an osteological study of Shinisaurus indicates that the procoracoid process of the scapula (more ...)

Fig. 13 Basal reptilian pectoral apparatus as exemplified by †Captorhinus. Adult in (A) lateral and (B) ventral views (C) subadult in ventral view. The dashed line separating the procoracoid from the scapula denotes that these elements are often fused (more ...)

In adults, the primary girdle forms a triradiate structure with slender dorsal, caudal and ventromedial processes (Fig. 14A,B). The glenoid is positioned at the crux. Whereas the column-like dorsal ramus is widely accepted as the scapula, the identity of the remaining processes has been the subject of debate. The ventromedial or acromial process has been identified as the homologue of the procoracoid (e.g. Parker, 1868; Korringa, 1938; Gaffney, 1990; deBraga & Rieppel, 1997) or alternatively as an outgrowth of the scapula (e.g. Romer, 1956; Walker, 1947, 1973; Lee, 1997). The caudal ramus (the coracoid) has been presumed to represent either the metacoracoid (e.g. Parker, 1868; Gaffney, 1990; deBraga & Rieppel, 1997; Lee, 1997) or the procoracoid (Walker, 1947, 1973).

Fig. 14 Testudines pectoral apparatus. Macroclemys (=Macrochelys), adult in (A) lateral and (B) ventral views; (C) subadult in ventral view. †Proganochelys, adult in (D) lateral and (E) ventral views. Abbreviations: act (turtle acromial process), cor (more ...)

During the earliest stages of chondrogenesis, the presumptive scapula, acromial process and coracoid of each side develop from either a unified trifurcate condensation or a trifurcate condensation in which the caudal ramus is initially isolated (Fig. 14C; Walker, 1947; Rieppel, 1993; Sheil, 2003, 2005). In at least one taxon (Chrysemys picta marginata), a supracoracoid foramen transmitting a neurovascular bundle is present early during development, nested between the presumptive scapula and coracoid (Walker, 1947). As the scapula and coracoid continue to differentiate, the neurovascular bundle is liberated from the now cartilaginous complex and the foramen disappears. In all taxa studied thus far (e.g. Apalone spinifera, Chelydra serpentina) each ramus begins to ossify from a separate centre within the condensation. The notion of the scapula developing independent of the other replacement bones is supported by the extirpation experiments of Burke (1991b). The majority of Chelydra embryos that had cervical somites 8–12 removed demonstrated scapular aplasia or incomplete scapula development, whereas the acromial process, coracoid and elements of the limb appear to have been minimally affected.

In view of their unusual morphology, the genealogical relationship of turtles to other amniotes is difficult to resolve unequivocally. Historically, however, they have long been considered members of Proganosauria (demonstrating the anapsid condition, a skull without temporal fenestrations; Fig. 15A). Among the most deeply nested non-turtle fossil proganosaurs (e.g. †pareiasaurs and †procolophonids), the pectoral apparatus closely resembles the morphology previously noted for basal synapsids (namely †′pelycosaurian grade synapsids′), with a robust, plate-like procoracoid and metacoracoid (Lee, 1996, 1997; deBraga, 2003). As in basal synapsids, the glenoid may receive contributions from the metacoracoid and scapula alone, or include minor involvement of the procoracoid. Among basal proganosaurs (e.g. †mesosaurids, †millerettids), the scapula and coracoid element(s) are typically fused together into a single element (Gow, 1972; Modesto, 1999). Unfortunately, few data are available on the ontogeny of fossil proganosaurs and it remains unclear if both coracoids are present and fuse together during early skeletogenesis or if one element fails to form.

Fig. 15 Evolution of the turtle primary girdle. (A) The proganosaur hypothesis as proposed by Lee (1996), wherein Testudines share a close common ancestry with †pareisaurs. According to this hypothesis, the evolution of the acromial process of turtles (more ...)

Corresponding to the burgeoning interest in turtle evolution, several efforts have been made to resolve the identification of the acromial process. At present, two competing views are held: the acromial process as an outgrowth of the scapula or as the homologue of the procoracoid (see below). Assuming Testudines as proganosaurs, it has recently been proposed that the evolution of the acromial process may be traced phylogenetically from †procolophonids – which lack any conspicuous scapular processes – to †pareiasaurs – which demonstrate an incipient ridge-like projection – to modern turtles and allies (Fig. 15A; Lee, 1996, 1997; although see Rieppel, 1996, for a contradictory interpretation of the same data). According to this hypothesis, the advent of the acromial process (in †pareiasaurs) precedes loss of the procoracoid.

1. Palaeontology and genealogy: Regardless of genealogical scenario, a three-element primary girdle is plesiomorphic for all amniotes. Within each of Synapsida and Reptilia one coracoid element appears to be independently lost. However, fossil material representing various ontogenetic stages of both basal synapsids (e.g. specimens of †Dimetrodon; Romer & Price, 1940) and reptiles (†Captorhinus; Holmes, 1977) demonstrates that this apparent loss is an example of two skeletal centres coalescing to form a single mature morphology. Thus, while the procoracoid is present and discrete among subadults, it generally fuses with the scapula (and the sutures disappear) in larger ‘adult-sized’ individuals (see Fig. 13). Consequently the scapula element of skeletally mature specimens in effect represents the scapula + procoracoid. By contrast, the metacoracoid frequently remains independent and discrete (Williston, 1925; Romer & Price, 1940; see also Gaffney, 1990).

2. Topography and connectivity: Several arguments can be made on the basis of the position and association of the various elements. Among basal amniotes the glenoid is dominated by the metacoracoid, with a decreasing contribution from the scapula and (when apparent) the procoracoid. According to the theory of Williston (1925) and Romer (1956), the glenoid of modern reptiles is taken over by the formerly minimally contributing procoracoid. By contrast, the revised hypothesis presented herein proposes that the glenoid of extant reptiles, similar to basal forms, remains dominated by the metacoracoid.

Of the two coracoid elements of basal amniotes, only the procoracoid contacts both the clavicle and the interclavicle. In Testudines (including †Proganochelys), the acromial process shares a common topology with the procoracoid of basal amniotes and retains a comparable contact with the clavicle–interclavicle homologues (namely the epiplastron and entoplastron; Gaffney, 1990; Fig. 14A). Among avians, the procoracoid process also demonstrates a similar connection with the intramembranously derived furcula. In both these representative amniotes, the process contacting the dermal element(s) is held to be the procoracoid, secondarily joined with either the scapula (Testudines) or coracoid (Aves). Gaffney (1990) also observed that the interpreted position of supracoracoideus muscle attachment on the procoracoid of basal amniotes (sensuRomer, 1922; Holmes, 1977) corresponds to the same muscle origin on the acromial process of the turtle coracoid.

3. Development and experimental embryology: Among therians, the sternum and primary girdle elements receive contributions from multiple embryonic condensations. For instance, the skeletally mature scapula element minimally represents a coalescence of the metacoracoid rudiment with the scapular condensation, and the manubrium sterni integrates components of the procoracoid, pars chondralis interclaviculae rudiment and sternal band. These developmental data are supplemented by evidence gleaned from the study of mutant mouse strains. In Emx2 mutants, the coracoid process, acromion and glenoid develop, whereas the blade of the scapula fails to form (Pellegrini et al. 2001). This indicates that Emx2 plays a role in patterning the scapular condensation but not the metacoracoid rudiment. By contrast, normal development for derivatives of the metacoracoid (e.g. coracoid process, glenoid and acromion) is dependent on Pax1, PDGFαR and Hoxc6 (Timmons et al. 1994; Soriano, 1997; Pellegrini et al. 2001; Fig. 6E,F). At the present time, however, the embryological source of the mammalian scapula (somites and/or lateral plate mesoderm) remains to be determined.

Among lepidosauromorphs the presence of a discrete procoracoid has been reported (Fig. 12C; Conrad, 2006), although in most instances it is unclear if the element in question is different from the coracoid element proper or the epicoracoidal cartilage. In the reptilian lineage Pseudosuchia there is (presently) no evidence for the development of more than one coracoid element, even transiently. However, neither group has been explored experimentally so future efforts will undoubtedly yield important data.

Among avians, the procoracoid process of the coracoid, as noted above, occupies a comparable position and set of articulations with the procoracoid of basal amniotes. Furthermore, it develops independent of the coracoid, and fuses with the latter at some later point of skeletogenesis (Fig. 10A; Parker, 1868). Experiments on the avian Gallus that enlarge the Hoxc6 gradient result in a reduction or complete absence of the procoracoid process (a derivative of the procoracoid), along with defects of the glenoid region of the scapula (possible procoracoid or metacoracoid origin) and (meta)coracoid malformations (Oliver et al. 1990). Other experimental work (including somite extirpation, chick-quail grafting and gene expression labelling) also demonstrates that the avian scapula receives contributions from more than one embryological source, namely somites and lateral plate mesoderm (see Chevallier, 1977; Timmons et al. 1994; Hofmann et al. 1998; Huang et al. 2000; Pellegrini et al. 2001). By contrast, the (meta)coracoid is exclusively lateral plate. Huang et al. (2000) suggested that the non-somitic source of the scapula was homologous with the metacoracoid. It should be noted, however, that the procoracoid is a comparable candidate. Similar to scapula development in mice, the gene Pax1 has been observed to play a role in patterning of the normal avian scapula element (Hofmann et al. 1998; Huang et al. 2000), whereas Hoxc6 is involved in regulation of the procoracoid and metacoracoid derivatives. Indeed, mutations of Pax1 and Hoxc6 in Gallus embryos result in primary girdle defects, minimally including the proximal part of the scapula, and in more extreme cases, the coracoid (Oliver et al. 1990; Hofmann et al. 1998). This indicates that similar to mice, Pax1 and Hoxc6 play a role in specifically patterning derivatives of the metacoracoid and (at least in avians) the procoracoid.

As with the avian procoracoid process, the acromial process of Testudines is topographically, connectively and developmentally comparable with the procoracoid. During skeletogenesis each of the turtle scapula, acromial process and coracoid arises from separate centres of ossification (Parker, 1868; Rieppel, 1993; Sheil, 2003). Removal (extirpation) of cervical–thoracic somites results in aplasia of the scapular blade but has little effect on the development of the acromial process, glenoid region or coracoid (Burke, 1991b). The data clearly imply a separate origin for each of the acromial process and scapula blade, one that may readily be explained by integration of the procoracoid rudiment into the scapula. This parallels the fusion/integration of the procoracoid and scapula in basal diapsids and synapsids.

The second topic relates to the apparent absence of the sternum in turtles. The view that turtles lack this pectoral element is widely held (e.g. Parker, 1868; Romer, 1956) inasmuch as the ventrally positioned plastron develops entirely via intramembranous ossification (Parker, 1868; Gilbert et al. 2001). Nevertheless, Walker (1947) suggested that a connective tissue band (the acromialcoracoid ligament; Wyneken, 2001) linking each ipsilateral acromial process and coracoid might be homologous with the sternum. Walker offered four observations to support his hypothesis: (1) ligaments may represent vestiges of cartilage (e.g. in undulated mutant mice a ligament may replace the acromion [Timmons et al. 1994; see also Hall (1970) for a discussion of the relationship between fibroblasts (producing ligaments) and chondroblasts (producing cartilage)]; (2) the acromialcoracoid ligament and sternum are both fundamentally paired structures; (3) the acromialcoracoid ligament and sternum arise in association with the coracoid; and (4) the acromialcoracoid ligament and sternum share a general topography with the rest of the pectoral apparatus. However unlike the sternum, acromialcoracoid ligaments do not contact the ribs and do not extend cranially to the clavicular homologues (the epiplastra) or caudally past the coracoids. Walker countered these differences by remarking that acromialcoracoid ligaments represent vestiges (and therefore are not developed to the same extent as a sternum) and that turtle ribs are fully integrated into the carapace (having become ensnared by the carapacial ridge; see Ruckes, 1929; Burke, 1991a; Gilbert et al. 2001). Moreover, as has been demonstrated for other parts of the pectoral apparatus (see above), the mere absence of a skeletal counterpart does not necessarily speak to definitive non-existence. This notion arguably receives indirect support from a study conducted on the marine reptile clade †Plesiosauria.

Similar to Testudines, among †plesiosaurs a discrete sternum is both unreported and architecturally problematic. Whereas turtles demonstrate a well-ossified plastron, in †plesiosaurs the presence of a sternum seems to be precluded by the enlarged and ventrally displaced coracoids. However, unlike turtles, in which the ribs are integrated into the carapace, in †plesiosaurs the ribs make no obvious connections with any preserved skeletal elements. Nicholls & Russell (1991) reviewed the issue of the sternum in †plesiosaurs and, based on data gleaned from sternal development, function and phylogeny, found that the suggestion of such an element was anatomically justified and functionally required. Without a sternum, the †pleisosaur girdle would probably be unable to withstand tensile forces associated with cranially directed movements of the forelimbs (Nicholls & Russell, 1991). Nicholls and Russell postulated that the position of the unmineralized (and thus unpreserved) sternum was dorsal (visceral) to the coracoid. Whereas these data do not speak directly to the presence or existence of a sternum in Testudines (indeed the architecture of the turtle pectoral apparatus fundamentally differs from that of †plesiosaurs in the internalization of these elements with respect to the ribcage), they suggest that connective tissues in this region (e.g. the acromialcoracoid ligament) might represent a latent homologue (sensuStone & Hall, 2004). Finally, it is worth observing that regardless of genealogical affinity (proganosaur or diapsid), the ancestors of modern turtles did demonstrate ossified rib cages requiring a sternal underpinning.

A revised scenario for the evolution of the amniote primary girdle is offered (Fig. 16), drawing on a global review of available data (documented above) and taking into consideration the role of cell condensations in skeletogenesis. It is argued that disruptions in preskeletogenic cell aggregation, proliferation and differentiation, and the acceleration (or retardation) of osteogenesis play vital roles in the evolution of the amniote primary girdle. Plesiomorphically, basal tetrapods have a unified scapulocoracoid (e.g. †Acanthostega) or a scapula and a single coracoid (e.g. †Seymouria). Among basal amniotes, the primary girdle consists of a scapula, procoracoid and metacoracoid. This transformation may be achieved as the normal osteogenic pattern of the condensation forming the membrane skeleton of the primary girdle elements (the coracoid–scapular plate, sensuKlima, 1973, 1987) is delayed and/or the failure (and delay) of preskeletogenic cells to differentiate, resulting in the formation of sutural contacts. The metacoracoid is more strongly influenced by osteogenic retardation, and unlike the procoracoid will often appear discrete even late in ontogeny.

Fig. 16 Evolution of the amniote scapula, procoracoid and metacoracoid; a new scenario. Amniote phylogeny based on the work of Brochu (2001), Lee (1996), Luo et al. (2002), Rieppel & Reisz (1999) and Ruta et al. (2003). Beginning with the unified scapulocoracoid (more ...)

Within Synapsida, the various outgroups leading to Theria (e.g. †Dimetrodon, monotremes) retain the three-part primary girdle. Among therians, phenotypic expression of both the procoracoid and the metacoracoid becomes dramatically diminished. This reduction reflects a decrease in preskeletogenic cell proliferation. The procoracoid rudiment detaches from the coracoid–scapular plate and either forms a vestigial element (the praeclavium of marsupials) or becomes subsumed into the manubrium sterni (eutherians). By contrast, the therian metacoracoid returns to the primitive condition of remaining joined with the scapula, giving rise to the coracoid process, acromion and probably contributes to the glenoid. At least among some eutherians, different genes appear to play roles in patterning the metacoracoid rudiment (e.g. Pax1, Hoxc6) compared with the scapula condensation (Emx2), consistent with the aforementioned homology argument.

Among reptiles, most skeletally mature individuals demonstrate only a single coracoid element, herein considered to be the metacoracoid. The procoracoid is not lost per se, but is retained as a rudiment that does not detach from the scapular portion of the coracoid–scapular plate. Retention of the combined scapula + procoracoid is in part achieved by a reduction in the duration of preskeletogenic cell mitosis (resulting in a decrease in size of the presumptive procoracoid) and an acceleration of osteogenesis (resulting in fusion of the procoracoid and scapula). Observations gleaned from experimental work on turtles and avians suggest that the reptilian scapula is a derivative of both the somites (forming the blade) and lateral plate mesoderm (forming the region around the glenoid).

Unique to avians the procoracoid rudiment may subdivide to join with both the metacoracoid (as the coracoid process) and possibly the scapula (at the glenoid). This partitioning of the procoracoid reflects either a further delay in osteogenesis or a delay in aggregation of the condensation or both. As in eutherians there is evidence to imply a patterning role of the procoracoid and metacoracoid components by Pax1 and Hoxc6. In the absence of data it remains unclear if crocodylians and most lepidosauromorphs retain the procoracoid rudiment (integrated into the scapula) or if this component is lost (i.e. a failure of the presumptive skeletogenic cells to proliferate and/or differentiate).

Whereas the newly conceived scenario of amniote pectoral evolution is well supported, important data are lacking in several key areas. As demonstrated, a comprehensive understanding of development, in particular the role of cell condensations, forms the critical underpinning to understanding the evolution of morphologically complex structures. Regrettably, our knowledge of the membranous skeleton in many reptilian taxa, especially lepidosauromorphs and crocodylians, remains incomplete. Another important area for consideration is the use of experimental embryology. With the exception of turtles, most non-avian reptiles are infrequent experimental subjects. Even among avians, the diversity of taxa commonly employed is overwhelmingly dominated by a single species, Gallus gallus. This is despite previous (and in the case of turtles, ongoing) successes with the use of tissue grafts and culturing, gene expression and extirpation experiments, and chimeric surgeries in various non-model taxa (Gans, 1985 and references therein; see also Burke, 1991b; Vincent et al. 2003; Nagashima et al. 2005). Increased application of such experimental methods holds great promise as a virtually untapped resource of developmental and evolutionary data.

We would like to express their sincerest thanks to Tim Fedak (http://www.figs.ca and Dalhousie University) for his excellent work on all 16 figures. T. Fedak and Dr A. P. Russell (University of Calgary) reviewed early drafts of this work and their input is greatly appreciated. Special thanks to Dr S. S. Sumida (California State University, San Bernardino) and an anonymous reviewer for their useful comments that significantly enhanced the text. This work was supported by an NSERC operating grant to B.K.H.