Endoskeletal structure in Cheirolepis (Osteichthyes, Actinopterygii), An early ray‐finned fish

Abstract As the sister lineage of all other actinopterygians, the Middle to Late Devonian (Eifelian–Frasnian) Cheirolepis occupies a pivotal position in vertebrate phylogeny. Although the dermal skeleton of this taxon has been exhaustively described, very little of its endoskeleton is known, leaving questions of neurocranial and fin evolution in early ray‐finned fishes unresolved. The model for early actinopterygian anatomy has instead been based largely on the Late Devonian (Frasnian) Mimipiscis, preserved in stunning detail from the Gogo Formation of Australia. Here, we present re‐examinations of existing museum specimens through the use of high‐resolution laboratory‐ and synchrotron‐based computed tomography scanning, revealing new details of the neuro‐cranium, hyomandibula and pectoral fin endoskeleton for the Eifelian Cheirolepis trailli. These new data highlight traits considered uncharacteristic of early actinopterygians, including an uninvested dorsal aorta and imperforate propterygium, and corroborate the early divergence of Cheirolepis within actinopterygian phylogeny. These traits represent conspicuous differences between the endoskeletal structure of Cheirolepis and Mimipiscis. Additionally, we describe new aspects of the parasphenoid, vomer and scales, most notably that the scales display peg‐and‐socket articulation and a distinct neck. Collectively, these new data help clarify primitive conditions within ray‐finned fishes, which in turn have important implications for understanding features likely present in the last common ancestor of living osteichthyans.

Cheirolepis Agassiz, 1835 is the earliest occurring taxon that can be unequivocally placed within Actinopterygii. It is represented by articulated specimens from the late Eifelian of Scotland, the Givetian of Nevada (Reed 1992;Arratia and Cloutier 2004) and the Frasnian of Canada, as well as by scales from the Givetian of Germany (Gross 1973) and Eifelian-Givetian of Belarus, Latvia and Estonia (Blieck and Cloutier 2000; Mark-Kurik 2000; Luk sevi cs et al. 2010). As part of his original description, Agassiz (1835) erected three species on the basis of Scottish material: the type species C. trailli from Orkney, C. uragus from Gamrie and C. cummingae from Cromarty. A further three species were described by M'Coy (1848): C. velox, C. macrocephalus (both from Orkney) and C. curtus, from Lethen Bar. These species were subsequently revised by Egerton (1860) and Traquair (1888), with only C. trailli retained. Whiteaves (1881) first reported Canadian material from the Frasnian Miguasha Lagerst€ atte and assigned it to the new species C. canadensis. An additional species, C. schultzei, was erected by Arratia and Cloutier (2004) for specimens from the Denay Limestone of Nevada, first reported by Reed (1992) as Cheirolepis cf. C. canadensis. C. trailli and C. canadensis were comprehensively reviewed by Pearson and Westoll (1979) and Arratia and Cloutier (1996).
The affinities of Cheirolepis with bony fishes generally, and actinopterygians specifically, have not always been apparent. On the basis of its micromeric, non-overlapping scales, Agassiz (1835) grouped Cheirolepis with Acanthodes Agassiz, 1833 and Cheiracanthus Agassiz, 1835 in his Acanthodidae, a placement upheld by M'Coy (1855) and Egerton (1860). Dissenters from this view included Miller (1841) and M€ uller (1846), who considered the combination of characters in Cheirolepis sufficiently unique to merit placement in its own group. Similarities between Cheirolepis and 'palaeoniscoids' were first articulated by Pander (1860), although he too regarded Cheirolepis as a member of its own distinct group. Traquair (1875) noted conspicuous differences between the structure of Cheirolepis and acanthodians, including several points relating to scale morphology. The most compelling similarity between these groups (scale micromery) was also rejected based on Egerton's (1864) description of micrometric squamation in the Carboniferous 'palaeoniscoid' Myriolepis clarkei Egerton, 1864. The structure of the fins, shoulder girdle and skull bones led Traquair (1875) to align Cheirolepis with taxa that are now assigned to Actinopterygii, a position universally accepted since. More recently, cladistic analyses have consistently resolved Cheirolepis as the sister taxon of all other ray-finned fishes (Gardiner 1984;Coates 1999;Cloutier and Arratia 2004;Gardiner et al. 2005;Friedman and Blom 2006;Friedman 2007;Friedman et al. 2007;Long et al. 2008;Brazeau 2009;Swartz 2009;Zhu et al. 2009;Choo 2011;Davis et al. 2012;Giles et al. 2015a). As only a small part of the endoskeleton of Cheirolepis has been described (Fig. 1;Pearson and Westoll 1979), very limited comparisons can be drawn with other well-known early actinopterygians such as Mimipiscis and Moythomasia. This makes it impossible to understand the evolution of key actinopterygian characters during the early history of the group, particularly endoskeletal structures with a major impact on osteichthyan, and gnathostome, phylogeny more generally. The paucity of endoskeletal data outside of Mimipiscis and Moythomasia has led to a situation where early actinopterygian relationships are investigated almost exclusively on the basis of dermal characters (Gardiner and Schaeffer 1989;Cloutier and Arratia 2004;Friedman and Blom 2006;Long et al. 2008;Swartz 2009;Choo 2011), in stark contrast to the more comprehensive character sets used when examining early gnathostome and sarcopterygian interrelationships (Zhu and Yu 2002;Friedman 2007;Brazeau 2009;Zhu et al. 2009;Davis et al. 2012;Zhu et al. 2013;Giles et al. 2015a). Attempts to expand the actinopterygian character set (Coates 1998(Coates , 1999Hamel and Poplin 2008;Giles and Friedman 2014) have increased the number of informative endoskeletal characters, but these are yet to be incorporated into many analyses due to the large amounts of missing data associated with taxa known principally from dermal material (Choo 2011).
Here, we use computed tomography (CT) to examine endoskeletal anatomy in Cheirolepis. This study uses material previously noted as preserving endoskeletal structures, but which could only be described on the basis of surface morphology (Pearson and Westoll 1979; Figs 1-3). Lab-and synchrotron-lCT permit the description of internal features of the specimens without recourse to destructive techniques, which could not be applied to such rare material. The specimens described herein preserve a largely complete braincase, a hyomandibula and pectoral fin endoskeleton. In the light of these data, we also revisit the neurocranium and associated bones in two other Devonian actinopterygians sometimes hypothesized to diverge outside the clade comprising Mimipiscis, Moythomasia and more derived actinopterygians (Friedman and Blom 2006;Long et al. 2008;Swartz 2009;Choo 2011): the Givetian Howqualepis Long, 1988 and the Famennian Tegeolepis Miller, 1892. Collectively, these new data allow us to test the suitability of Mimipiscis as a model of primitive ray-fin anatomy while also clarifying patterns of character evolution early in actinopterygian history.
Howqualepis. The specimen of Howqualepis rostridens examined here, AMF65495, is that of a near-complete fish, missing only the snout and the anal fin. The material is from the lacustrine Mt. Howitt locality, south-east Australia, and has been dated by Young

Methods
Several specimens were selected for CT scanning. NHMUK P.62908, which preserves the braincase of Cheirolepis, was scanned using synchrotron radiation X-ray microtomography at the I12 beamline of the Diamond Light Source, Didcot, UK, using an 80 keV monochromatic beam, CdW04 scintillator of 0.9 mm thickness, 4008 9 2672 pco.4000 camera and 3000 projections of 0.04-second exposure collected through 180°rotation. From the projections, slice images were reconstructed with an in-house filtered back projection reconstruction algorithm (Titarenko et al. 2010). The resulting voxel size was 12.35 lm. The part (NMS.1877.30.5), in which the parasphenoid and parts of the ethmoid region are preserved ( Fig. 2A), was not scanned due to its large size and high aspect ratio. NMS.1956.19, which preserves the endoskeletal fin girdles and hyomandibula of Cheirolepis, was scanned at the Imaging and Analysis Centre, NHMUK, using a Metris X-Tek HMX ST 225 CT System with a 2000 9 2000 pixel detector, tungsten reflection target and 3142 projections. Volumes were created with CTPro V2.1. AMF65495, which preserves a mould of the braincase and hyomandibula, as well as dermal material of Howqualepis, was scanned using the same machine at NHMUK, using a copper filter.
Following scanning, the data were reconstructed and segmented manually in Mimics version 15.01 (http://biomedical.materialise.com/mimics; Materialise, Leuven, Belgium). As the preservation in AMF65495 is mouldic, a mask of the air was generated, producing a 'virtual' cast. The use of this method, rather than producing a latex peel, minimizes any risk of damage to the specimen and results in a permanent record that will not deteriorate (original latex peels of the material, made in the 1980s, have degraded badly and been lost; J. Long pers. comm. 2013). Meshes were exported as .PLY surface files and were exported to and imaged in Blender (Garwood and Dunlop 2014). PLY files of the braincase of Cheirolepis, the fin and hyomandibula of Cheirolepis and the cast of the Howqualepis specimen are made available online (Giles et al. 2015b). These files can be easily opened and manipulated in free programs such as Meshlab (http:// meshlab.sourceforge.net; Cignoni et al. 2008).
Meshlab was used to downsample and prepare the two portions of the Cheirolepis braincase for 3D printing. The model was upscaled by a factor of five and printed using a Zprinter 350 at the Hull York Medical School, York, UK. These models aided in the interpretation of the braincase.
The parasphenoid of Tegeolepis is preserved as a negative impression. We studied a positive cast of this mould made using two-part flexible dental casting compound. Unlike standard latex peels, this material releases easily from matrix. For photography, this cast was coated with a sublimate of ammonium chloride.
Several specimens were studied under immersion, and photographed with a Nikon SLR camera with a polarizing filter. Line drawings were produced by hand. Emended diagnosis. See Pearson and Westoll (1979, p. 390) with the following amendments. Cheirolepidid with lozenge-shaped parasphenoid lacking ascending processes. Spiracle housed in groove. Open groove for dorsal aorta on basicranium. Differs from other species of Cheirolepis in the following features: extrascapulae do not contact each other at midline, head of dermohyal projects above operculum, elongate spiracular slit.

Neurocranium
General features. The specimen of Cheirolepis containing the neurocranium studied here is preserved in part and counterpart (NMS.1877.30.5/NHMUK P.62908;Figs 2A, 3A). The bulk of the braincase is preserved in the part. The neurocranium has been dorsoventrally compressed, with the loss of internal anatomy, and is preserved as two parts: the base of the ethmosphenoid region, and the dorsal part of the ethmosphenoid region plus the otic region and occipital arch. The break between these two components occurred along the interorbital septum. Detachment of the parasphenoid and ethmosphenoid is common in specimens of Cheirolepis, with isolated examples found in several specimens (e.g. NMS.1877.30.5/NHMUK P.62908 (Sp.2a); NHMUK P.4051a/b; MCZ 6039; NHMUK P.66863 (BMP.41410); Fig. 3). The occipital region has rotated backwards during compaction such that the posterior face of the occiput is now oriented dorsally (Fig. 4). A similar style of preservation appears to characterize Pearson and Westoll's (1979, fig. 11a) 'Sp. 13 ' and UMZC.425 (Pearson and Westoll 1979, fig. 4a). The basioccipital plate (lying between the vestibular fontanelles, and extending from the occiput to the ventral otic fissure) has shifted slightly to the anatomical right. Abbreviations: bhc, buccohypophyseal canal; bpt, basipterygoid process; dent, denticle field of parasphenoid; epsa, efferent pseudobranchial artery; nas.c, nasal capsule. Anterior to top. Scale bar represents 2 mm.
The otico-occipital region of the braincase of Cheirolepis is formed as one ossification, as in Mimipiscis and Moythomasia (Gardiner 1984). Although the crushed nature of the specimen makes it difficult to determine the location or extent of different ossification centres, the boundaries between the occipital and otic/orbital regions of the braincase can be deduced from the presence of conspicuous fissures.
Occipital region. The occiput is preserved lying face up on the surface of the rock, with the openings for the foramen magnum and notochord clearly visible (Fig. 4). The foramen magnum (fm; Fig. 4B) is ovoid and approximately twice the size of the notochord (not; Fig. 4B). As the floor of the foramen magnum is incompletely mineralized along the midline, the two openings appear confluent. Incomplete mineralization of the floor of the foramen magnum is also typical in Mimipiscis and Moythomasia (Gardiner 1984, p. 189). Pearson and Westoll (1979, p. 345) incorrectly identified these openings as accommodating the notochord and dorsal aorta. The floor of the notochordal canal is largely complete, with the exception of a slot-shaped cavity on the midline, which likely represents incomplete fusion of paired parachordal plates, as in the Gogo actinopterygians (Gardiner 1984; Fig. 5B). Because the braincase is flattened, the notochordal canal cannot be traced anteriorly, and the relationship of this canal with the ventral otic fissure is unknown. The posteriormost parts of the occipital arch, presumably including the articular areas for the first pharyngobranchials, are preserved in the counterpart.
Lateral to the floor of the foramen magnum, the occiput is pierced by a small canal (focn; Figs 4, 5). This travels anterolater-ally to open on the lateral face of the occiput (focn; Fig. 6), and comparison with other ray fins suggests that this would have transmitted the occipital nerve. This opening was cautiously identified by Pearson and Westoll (1979, fig. 1c) as accommodating a spinal nerve. A distinct ridge on the lateral face of the occiput runs dorsally from the level of the occipital nerve (oims 2 ; Fig. 6). Such a ridge is also present in Mimipiscis and Kansasiella Poplin, 1975(Poplin 1974, and is tentatively identified in Cheirolepis as the ridge for the insertion of the second intermuscular septum. A modest craniospinal process is visible on the left side of the specimen, lateral to the posterior dorsal fontanelle (crsp; Figs 4-6). In life, this would have formed the dorsolateral corner of the occipital plate behind the otico-occipital fissure.
The dorsal surface of the braincase is smooth, with no evidence of a fossa bridgei (Fig. 4). Although present in later actinopterygians, a fossa bridgei is also absent in Mimipiscis and Moythomasia (Gardiner 1984), and poorly developed in Kentuckia deani (Eastman, 1908) (Rayner 1951). The dorsal roof of the braincase is poorly preserved anterior to the ventral otic fissure, particularly along the midline, and the presence or absence of an anterior dorsal fontanelle cannot be determined. Mineralization of the upper surface of the braincase is complete behind the ventral otic fissure, with the exception of the posterior dorsal fontanelle (pdf; Fig. 4B); the unfinished areas anterior and lateral to the fontanelle represent areas where the bone is either too thin or too poorly mineralized to be fully resolved by the scan.
The otico-occipital fissure (oof; Figs 5, 6) is completely open, as in most other early osteichthyans, and can be traced anteriorly along the ventrolateral face of the braincase before intersecting the vestibular fontanelle. A slight expansion in the line of the fissure presumably marks the exit of the vagus nerve (X; Fig. 5B). It is not possible to determine whether this is divided into a dorsal and ventral portion, as in Mimipiscis and Moythomasia (Gardiner 1984). This foramen, and indeed much of the fissure, is difficult to trace on the anatomical right of the specimen, presumably due to the slight lateral displacement of the occipital plate. The ovoid vestibular fontanelles (v.fon; Fig. 5B) are at least twice the relative length of those in Mimipiscis and Moythomasia, and are more similar in size to the vestibular fontanelles of Carboniferous actinopterygians (e.g. Kentuckia Rayner, 1951;Coccocephalichthys Whitley, 1940(Watson 1925Poplin and V eran 1996);Pteronisculus White, 1933(Nielsen 1942Patterson 1975;Coates 1998)) and sarcopterygians (e.g. Youngolepis Chang andYu, 1981 (Chang 1982); Eusthenopteron Whiteaves, 1881 (Bjerring 1971;Jarvik 1980); Gogonasus Long, 1985 (Holland 2014)). The large fontanelles presumably formed a point of weakness about which the otic region collapsed, and the fontanelles may in fact appear slightly larger than their original size. As in Mimipiscis, Moythomasia, Kansasiella, Coccocephalichthys, Lawrenciella Poplin, 1984(Hamel and Poplin 2008), Boreosomus Stensi€ o, 1921(Nielsen 1942 and Luederia Schaeffer and Dalquest, 1978, the vestibular fontanelles are clearly separated from the ventral otic fissure by a substantial bridge of bone. The occipital portion of the ventral otic fissure (vof; Figs 4, 5, 7), which delimits the region anteriorly, is very straight.
Unlike all other early actinopterygians in which the condition is known (but see below for a reinterpretation of Howqualepis), the dorsal aorta in Cheirolepis is not enclosed in a midline canal. Instead, the aorta was accommodated by a groove on the basioccipital (da; Fig. 5B), as in sarcopterygians (e.g. Youngolepis (Chang 1982) and Acanthodes (Miles 1973;Davis et al. 2012)). The aortic groove is deeper than observed in members of these outgroups. This bifurcation occurs some way posterior of the vestibular fontanelles and is positioned more posteriorly than in other early ray fins. The lateral dorsal aortae almost immediately split again; grooves for the first epibranchial artery travel anterolaterally towards the vestibular fontanelles (epi I; Figs 5, 7), and the carotids continue anteriorly towards the ventral otic fissure (lda; Figs 5, 7). Although a similar arterial branching pattern is observed in Kentuckia (Rayner 1951), Coccocephalichthys (Poplin and V eran 1996), Lawrenciella (Hamel and Poplin 2008) and Luederia (Schaeffer and Dalquest 1978), the epibranchial arteries occupy a longer portion of the basioccipital in Cheirolepis. Similarly, the proportion of the braincase carrying the lateral dorsal aortae is longer in Cheirolepis than in Mimipiscis and other early actinopterygians. As the dorsal aorta is unfloored, the position of exit of the second epibranchial arteries is unknown. In Kentuckia, Coccocephalichthys, Lawrenciella, Luederia, Kansasiella (Poplin 1974), Boreosomus and Pteronisculus (Nielsen 1942), these leave through a single or paired opening from the floor of the aortic canal. The branches of the lateral dorsal aortae and first epibranchial arteries are widely separated in Cheirolepis, and the basioccipital plate as a whole is broader than the corresponding area in Lawrenciella, Kansasiella and the Gogo actinopterygians. Immediately posterior to the ventral otic fissure, a shallow groove branches from the lateral dorsal aorta before turning posterolaterally and entering the braincase (hy; Figs 5, 7). A similarly placed groove and foramen in other early actinopterygians (e.g. Mimipiscis: Gardiner 1984, fig. 50; Lawrenciella: Hamel and Poplin 2008, fig. 9) has been interpreted as housing the orbital artery. Positionally, this is implausible: it is unlikely that the orbital artery would turn posteriorly, away from the orbit. The posterolateral orientation of the groove and comparison with other extant and extinct gnathostomes (e.g. Chlamydoselache Allis, 1923; Janusiscus Giles et al., 2015a) suggest that this represents the path of the efferent hyoid artery.
A small peg is located on the midline immediately after the divergence of the dorsal aorta. Comparison with Mimipiscis and Moythomasia (Gardiner 1984) indicates that this likely marks the attachment point for the aortic ligament (alig; Fig. 5B). The roof of the aortic groove is pierced by an anterodorsally directed canal that opens into the notochordal canal. The canal lies on the midline, but as it is developed as a distinct groove, it is unli-kely to represent a gap between the basioccipitals, and the opening is too posterior to be the aortic ligament. It may represent a branch of the occipital artery (?boca; Fig. 5B).
Otic and orbital regions. The deformation of the specimen has caused the lateral face of the neurocranium to be flattened out onto a level with the rest of the ventral surface. Therefore, the otic region may appear somewhat wider than it would in life. The lateral commissure (the transverse otic process; see revised terminology in Giles et al. 2015a) is anteroposteriorly broader and slightly longer than in other early actinopterygians (lcom; Fig. 5B The articular area for the hyomandibula is positioned behind the spiracular groove on the posterior face of the broad postorbital process (hmf; Figs 4B, 5B). The hyomandibular facet has been distorted and now faces posteriorly. The facet on the right of the specimen is partially obscured by a displaced, possibly spiracular, toothplate (tp; Fig. 5B).
A deep gutter marking the course of the jugular vein is visible on the lateral side of the otic region, below the shelf formed by the hyomandibular facet (jug; Fig. 5B). The groove bends dorsolaterally around the prominent parampullary process (p.amp; Fig. 5B), as in Mimipiscis and Moythomasia (Gardiner 1984), and the first suprapharyngobranchial likely articulated with this region. The posterior entrance of the jugular canal into the lateral commissure presumably also transmitted the hyomandibular trunk of the facial nerve. The exit of the glossopharyngeal nerve cannot be identified. The otic region anterior to the postorbital process has collapsed, obscuring the anterior opening of the jugular canal and the trigeminofacialis chamber.
The floor and hind walls of the orbit are not preserved. Due to the manner in which the specimen is broken, the interorbital septum (ios; Figs 4B, 5B) is largely incomplete. The precise width of the interorbital septum is unclear, but appears wider than in Mimipiscis and Moythomasia (Gardiner 1984). Details of the orbital roof are not preserved, but deep grooves are visible on the dorsal surface for the overlying supraorbital canals of the dermal skull roof (soc; Fig. 5B).
Although comparison with other early actinopterygians suggests the interorbital septum would originally have been completely mineralized, separation of the sphenoid and ethmoid from the occipital and orbitotemporal regions is fairly common in specimens of Cheirolepis. The neurocranium has broken through the ventral otic fissure and basiphenoid pillar and along the interorbital septum. Consequently, the parasphenoid and ventralmost parts of the ethmosphenoid are preserved in the counterpart.

A B
F I G . 6 . Occipital portion of the braincase of Cheirolepis trailli NHMUK P.62908 in lateral view; this part of the braincase has rotated such that the dorsalmost part of the occiput, identified by the foramen magnum, is now at the posterior edge of the specimen. A, three-dimensional rendering of braincase. B, interpretive drawing of braincase. Anterior to left. Abbreviations: crsp, craniospinal process; fm, foramen magnum; focn, foramen of the occipital nerve; oims 2 , origin of second intermuscular septum; oof, otico-occipital fissure. Scale bar represents 2 mm.
The basiphenoid portion of the ventral otic fissure is incompletely resolved, but faint notches for the orbitonasal arteries are apparent posterior to the basipterygoid processes (nona; Fig. 8B, D), as in Mimipiscis (Gardiner 1984) and Coccocephalichthys (Poplin and V eran 1996). The basisphenoid pillar in this specimen is split open through the hypophyseal fossa, and the buccohypophyseal canal (bhc; Fig. 8B, D), which travels through the basisphenoid to open on the ventral surface of the parasphenoid, is visible. Lateral to the buccohypophyseal opening are two anterolaterally directed canals, which can also be traced on to the parasphenoid. Comparison with Mimipiscis (Gardiner 1984), Moythomasia and Lawrenciella (Hamel and Poplin 2008) suggests these transmitted the efferent pseudobranchial arteries (epsa; Fig. 8B, D). The buccohypophyseal canal and openings of the efferent pseudobranchials were both identified by Pearson and Westoll (1979, fig. 1). The entrance of the palatine artery into the basisphenoid is marked by an anteroventrally directed canal immediately in front of the ventral otic fissure (pal.a; Fig. 8B).
The sides of the basisphenoid extend laterally as small basipterygoid processes (bpt; Fig. 8B, D), as noted by Pearson and Westoll (1979). These processes are entirely endoskeletal and are not in contact with the parasphenoid; a dermal component to the basipterygoid processes is seen in Kentuckia (Rayner 1951), Kansasiella (Poplin 1974), Pteronisculus and Boreosomus (Nielsen 1942). The spiracular groove continues onto the basisphenoid (spig; Fig. 8B) behind the basipterygoid processes.
Due to the high level of dorsoventral compression, no details of the endocast can be described. As in most fish fossils preserved in carbonate matrix, the otoliths cannot be discerned, having presumably dissolved during fossilization.
Ethmoid region and parasphenoid. Very little of the ethmoid region is preserved, particularly above the level of the suborbital shelf. Additionally, much of the ventral part of the ethmoid is preserved in the counterpart. Anteriorly, the subnasal shelf flares dorsally. Two distinct notches mark the articular areas for the autopalatine (apal; Fig. 8B, D).
Parasphenoids are preserved in a number of specimens (e.g. NHMUK P.62908/NMS.1877.30.5; NHMUK P.60499; NMS.1892. 8.60; NHMUK P.66863 (BMP.41410); NHMUK P.4051a/b; MCZ 6039; Fig. 3) and are separated from the bulk of the braincase in most instances. It is possible that these isolated parasphenoids represent individuals with poorly ossified or unossified endocrania. The parasphenoid is complete in NHMUK P.62908/ NMS.1877.30.5 and is wholly preserved in the counterpart (Fig. 3A). As described by Pearson and Westoll (1979), this bone is simple, lacking a complex anterior margin and ascending processes, and quite unlike the ossifications seen in Mimipiscis and Moythomasia (Gardiner 1984; Choo 2011). Although the posterior part of the bone flares slightly behind the efferent pseudobranchial openings, there is no evidence of an ascending process. Ascending processes are also lacking in Mimipiscis. A shagreen of denticles covers much of the ventral surface of the parasphenoid in Cheirolepis (Fig. 3A, D).
A vomerine toothplate is preserved on the left side of the ethmoid (vom; Fig. 8D). The toothplate is lozenge shaped and is covered in fine denticles. The vomer is small, resembling those of Mimipiscis (Gardiner 1984) and Moythomasia (Gardiner and Bartram 1977), rather than the enlarged vomers of sarcopterygians (e.g. Eusthenopteron (Jarvik 1980); Powichthys (Jessen 1980; Cl ement and Janvier 2004)). It appears to be in life position, although it has been crushed somewhat into the ethmoid.

Hyomandibula
The left hyomandibula is preserved within NMS.1956.19 and is not visible on the surface. Although unbroken, it has been displaced to the region of the right scapulocoracoid. The hyomandibula is fairly slender and gently curved, and is firmly fused to the dermohyal (i.e. no sutures are apparent between these two ossifications in tomographs), as in Mimipiscis and Moythomasia (dhy; Fig. 9D). The dermohyal appears proportionately longer A B F I G . 7 . Close-up of the ventral surface of the braincase of Cheirolepis trailli NHMUK P.62908 showing the lateral dorsal aortae and hyoid arteries. A, three-dimensional rendering of braincase. B, interpretive drawing of braincase. Anterior to top. Abbreviations: da, dorsal aorta; epi.I, first epibranchial artery; hy, hyoid artery; oof, otico-occipital fissure; lda, lateral dorsal aorta; vof, ventral otic fissure; v.fon, vestibular fontanelle. Scale bar represents 5 mm. than in other early actinopterygians, although as the distal part of the hyomandibula appears to be unmineralized, its precise proportion cannot be determined. A distinct process on the medial face of the hyomandibular projects above the dermohyal (proc, Fig. 9C, D). Comparison with other actinopterygians suggests that this is too proximal to be the opercular process, and reference to Pearson and Westoll's (1979, fig. 20a

Pectoral fin endoskeleton
Both pectoral endoskeletal girdles are preserved in NMS.1956.19, although both are broken and thus cannot be exhaustively described. The left girdle is visible on the surface and was partially described by Pearson and Westoll (1979, fig. 14a), although their reconstruction of a complete scapulocoracoid (Pearson and Westoll 1979, fig. 12) is almost entirely hypothetical. The preserved scapulocoracoid in Cheirolepis is mineralized as a single ossification (scpc; Fig. 10C). The area of attachment to the cleithrum is fairly broad and elongate (art.cleith; Fig. 10D). There is no evidence of a ventral process nor a mesocoracoid arch. Thus, the scapulocoracoid appears to have had only a simple attachment to the cleithrum (art.cleith; Fig. 10D The anterior process (apr; Fig. 10D) in Cheirolepis appears extensive, although it is fragmented in both girdles. As a result, it is difficult to trace the course of any blood vessels. The processus lateralis identified by Pearson and Westoll (1979, fig. 14, Pr.l) appears to be an artefact on the surface of the specimen, and the processus ventralis ( fig. 14, Pr.v) is in fact the misplaced metapterygium.
The glenoid fossa is narrow and elongate (gf; Fig. 10D). It bears several moderately well-developed articular facets for the radials. The radials are preserved as hollow perichondral sheaths, and those of the left fin have been shifted out of the scapulocoracoid plane. The radials of the right fin are completely disarticulated. The propterygium is stout and is clasped by four fin rays (propt, f.r; Fig. 10C). Unlike all other early actinopterygians, the propterygium is not pierced by the propterygial canal. The metapterygium is large and robust, although not noticeably elongate (mpt; Fig. 10C). It bears articular facets to support two distal radials (art.fac; Fig. 10C), although these are not preserved; these were probably cartilaginous, as in Polypterus and possibly Mimipiscis (Gardiner 1984). The shape of the metapterygium is more similar to that of actinopterygian outgroups such as Acanthodes (Coates 2003, fig. 3b) than other actinopterygians such as Mimipiscis (Gardiner 1984, fig. 137). However, the presence of an elongate metapterygium in sarcopterygians makes it difficult to assess whether Cheirolepis has retained the primitive condition or if its metapterygium is secondarily reduced. Five radials sit between the propterygium and metapterygium (rad; Fig. 10C).

Body scales
Several patches of body scales are preserved in NHMUK P.62908, including a number found in articulation with each other but inside the braincase (Fig. 11). Although the external surface of the scales is poorly resolved, the internal surface is well preserved. The scales are rhomboid, with a pronounced anterodorsal process, and bear a distinct midline keel (ant.d, k; Fig. 11B). Most importantly, the scales display peg-and-socket articulation, a feature previously considered absent in Cheirolepis trailli (Pearson and Westoll 1979;Gardiner 1984;Arratia and Cloutier 1996). As in Psarolepis and acanthodians, but unlike the scales of most osteichthyans (Qu et al. 2013), the rhomboid scales of Cheirolepis bear a distinct neck between the crown and base (n; Fig. 11B).

Comparative morphology
Howqualepis. The braincase of Howqualepis is known in most detail from AMF65495, where it is preserved in ventral view but is dorsoventrally compressed (Long 1988, fig.  16). As described by Long, the ascending processes are elongate, and end in a fairly blunt edge rather than tapering (asc.pr; Fig. 12B). Although depicted as having a pointed apex in Moythomasia (Gardiner 1984, fig. 7), unbroken ascending processes in this genus also tend to terminate rather bluntly (SG pers. obs.). The processes in Howqualepis do not appear to be denticulated. The basipterygoid processes are well developed and are entirely endoskeletal (bpt; Fig. 12B). The shape of the parasphenoid denticle field is difficult to interpret, but the denticles are largely restricted to the anterior portion of the parasphenoid. The postorbital process in Howqualepis is incompletely preserved. The median dorsal aorta is housed in an open groove, rather than an enclosed canal (da; Fig. 12B).
Tegeolepis. The parasphenoid of Tegeolepis is preserved in dorsal view in CMNH 5518. It is slender and is more similar in shape to the parasphenoid of Cheirolepis than that of any other early actinopterygian (Fig. 13). The parasphenoid is narrow anteriorly and tapers to its narrowest point just in front of the buccohypophyseal canal. It bears two small basipterygoid processes (?bpt; Fig. 13B), but their skeletal origin is unclear. Posteriorly, the parasphenoid flares into two short ascending processes (asc.pr; Fig. 13B), reminiscent of those of Howqualepis. No details of the parasphenoid denticle field can be described.
The parasphenoid described here bears little relation to that reconstructed by Dunkle and Schaeffer (1973, fig. 1) for another specimen (CMNH 8244), in which the lateral edges of the bone are strongly fluted, the ascending processes are very narrow, and the posterior of the bone is developed as a long stalk that would have extended below the otic and occipital regions. Based on observations of this specimen, we question the interpretation of CMNH 8244 as a parasphenoid, and note that the structure in question is not intact.

A B
F I G . 1 1 . Scales of Cheirolepis trailli NHMUK P.62908 in ventral view. A, three-dimensional rendering of scales. B, interpretive drawing of scales. Abbreviations: ant.d, anterodorsal process; k, keel; n, neck; p, peg; s, socket. Anterior to right. Scale bar represents 0.5 mm.

Cheirolepis and other early actinopterygians
Although broadly resembling other that of actinopterygians, the braincase of Cheirolepis bears clear similarities with members of actinopterygian outgroups, in particular the sarcopterygians. These features, outlined below, are presumably osteichthyan symplesiomorphies (Fig. 14). The most conspicuous of these is the uninvested dorsal aorta; an enclosed midline canal for the aorta has previously been considered an actinopterygian synapomorphy (Rayner 1951;Miles 1973;Patterson 1975;Gardiner 1984;Long 1988;Friedman and Brazeau 2010). Although present in all other early actinopterygians (with the exception of Howqualepis), the dorsal aorta is uninvested in sarcopterygians (e.g. Eusthenopteron, Jarvik 1980; Youngolepis, Chang 1982) and acanthodians (e.g. Acanthodes, Miles 1973). The condition of the dorsal aorta in the early osteichthyans Ligulalepis (Basden and Young 2001) and Psarolepis (Yu 1998) is unknown, as the only described braincases for these taxa lack the ventral occipital plate. Other features that appear plesiomorphic for actinopterygians apparent in Cheirolepis include the relative position of bifurcation of the dorsal aorta on the basioccipital, which bears a greater resemblance to sarcopterygians such as Youngolepis; and the imperforate propterygium (as noted by Pearson and Westoll 1979; Friedman and Brazeau
2010; contra Sallan 2014), a character shared with chondrichthyans, with the presence of a propterygial canal previously considered to be an actinopterygian synapomorphy.  1984), suggesting that this feature might also be primitive for actinopterygians. However, the lack of preservation of the occipital plate in stem sarcopterygians, coupled with the absence of vestibular fontanelles in many sarcopterygian taxa (e.g. Qingmenodus Lu and Zhu, 2010;Styloichthys Zhu and Yu, 2002;coelacanths, Forey 1998;megalichthyid tetrapodomorphs, Fox et al. 1995), leaves open the question of whether the large fontanelles of Cheirolepis are plesiomorphic or secondarily derived. In addition to differences in the endoskeleton, the dermal skeleton of Cheirolepis has also long been considered atypical of other actinopterygians, given the large number of plesiomorphic osteichthyan features it retains. These features include unconsolidated rostral bones, a tectal, elongate jaws, lobed base to pectoral fins, epichordal lobe in caudal fin, and lack of acrodin tooth caps (Gardiner 1963;Pearson and Westoll 1979;Pearson 1982;Friedman and Brazeau 2010). Despite differing from Mimipiscis and other actinopterygians in certain key respects, Cheirolepis possesses several features that unite it with other actinopterygians: the presence of a basipterygoid fenestra in the palatoquadrate (Pearson and Westoll 1979, fig. 7c, d), a single dorsal fin (Pearson and Westoll 1979, fig. 16; Arratia and Cloutier 1996, fig. 4a) and the absence of a jugal canal (Pearson and Westoll 1979, fig. 20;Arratia and Cloutier 1996, fig. 6;Arratia and Cloutier 2004, fig. 4;Friedman and Brazeau 2010). The generalized conditions outlined above confirm the position of Cheirolepis as an early diverging member of Actinopterygii.

Parasphenoid evolution in early actinopterygians
New data on the parasphenoid in Tegeolepis, as well as new descriptions of the parasphenoids of Mimipiscis toomsbi and M. bartrami Choo, 2011, allow a detailed understanding of the evolution of the parasphenoid in early actinopterygians. Ascending processes of the parasphenoid, previously considered to be a derived actinopterygian character confined to actinopterygians crownwards of Mimipiscis (Patterson 1982;Gardiner 1984;Coates 1999;Gardiner et al. 2005), are present in all actinopterygians crownwards of Cheirolepis (Node D of Fig. 14; Choo 2011). The slight lateral extensions posterior to the buccohypophyseal canal in Cheirolepis may represent homologues of the ascending process of later actinopterygians. The absence of this process in Mimipiscis thus appears to be a secondary loss. Several groups of sarcopterygians bear ascending processes of the parasphenoid, or expanded denticle plates that occupy a similar region as actinopterygian ascending processes (Jarvik 1980;Jessen 1980;Chang 1982). However, patterns of character distribution given current hypotheses of osteichthyans interrelationships suggest that these are independently derived relative to ascending processes in actinopterygians.

Implications for the actinopterygian character suite
The recognition of features previously thought to be restricted to sarcopterygians in the braincase of an early actinopterygian, most particularly the uninvested dorsal aorta, indicates that these characters represent osteichthyan plesiomorphies. Cheirolepis lacks several key features previously identified as actinopterygian synapomorphies, allowing a new understanding of the sequence of character evolution early in the history of the group (Fig. 14). Because these data do not change the phylogenetic position of Cheirolepis, only reinforce existing interpretations, this character distribution can be summarized by mapping key neurocranial and endocast characters onto a cladogram based on recent analyses of early actinopterygian and osteichthyan relationships (Choo 2011;Zhu et al. 2013;Xu et al. 2014;Fig. 14).
The key morphological changes in the endoskeleton at successive nodes can be summarized as follows: 1. Node A, Osteichthyes: long olfactory tracts (Coates 1999;Friedman and Brazeau 2010;Brazeau and Friedman mandibula (Patterson 1982;Gardiner 1984;Coates 1999;Gardiner et al. 2005; reversals in Tegeolepis and some post-Devonian actinopterygians). 4. Node D: ascending process of the parasphenoid (Patterson 1982;Gardiner 1984;Coates 1999;Gardiner et al. 2005; reversals in Mimipiscis and Gogosardina). 5. Node E: enclosed dorsal aorta (Miles 1973;Coates 1999;Friedman and Brazeau 2010); propterygium pierced by propterygial canal (condition unknown in Howqualepis and Tegeolepis;Patterson 1982;Gardiner 1984;Gardiner and Schaeffer 1989;Coates 1999;Gardiner et al. 2005;Friedman and Brazeau 2010); pectoral fin endoskeleton does not project outside of body (Coates 1999; reversal in some post-Carboniferous actinopterygians). 6. Node F: enclosed spiracular canal (Patterson 1982;Gardiner 1984;Gardiner and Schaeffer 1989;Coates 1999;Gardiner et al. 2005). 7. Node G: anteriorly (as opposed to anterolaterally) directed olfactory tracts (Coates 1999; Giles and Friedman 2014); olfactory nerves carried in single midline tube (Coates 1999; Giles and Friedman 2014; reversal in some post-Carboniferous actinopterygians); optic lobes same width as cerebellum (Giles and Friedman 2014); corpus cerebellum of anterior and posterior semicircular canals ventral to endocranial roof (Giles and Friedman 2014); fossa bridgei (Miles 1973;Gardiner 1984;Gardiner and Schaeffer 1989;Coates 1999); expanded anterior dorsal fontanelle. (NB some or all of the latter four characters may occur at Node F, but as an endocast for Moythomasia has not been described this cannot be determined.) The endoskeleton of Cheirolepis differs from that of Mimipiscis in several important respects, including an uninvested dorsal aorta; large vestibular fontanelles; elongate basioccipital portion of the braincase; unelaborated parasphenoid lacking a multifid anterior margin and ascending processes; pectoral endoskeleton protruding from the body; and imperforate propterygium. In all of these respects, Cheirolepis more closely resembles early members of actinopterygian outgroups than Mimipiscis and other Devonian actinopterygians. The recognition of apomorphic features in the braincase of Mimipiscis, for example the loss of ascending processes of the parasphenoid, the presence of an elongate, fully enclosed dorsal aorta and the very small vestibular fontanelles (Gardiner 1984), further bring into question the suitability of Mimipiscis as representative of the plesiomorphic actinopterygian condition.
The description of the braincase and other endoskeletal anatomy of Cheirolepis adds more detail to our understanding of endoskeletal evolution in early actinopterygians. However, further data, for example from Howqualepis (Long 1988) and Gogosardina Choo et al., 2009 are needed to fully understand the early evolution of the group. Despite distinct apomorphies in its dermal skeleton (such as micromeric squamation), the endoskeleton of Cheirolepis appears to largely reflect the primitive osteichthyan condition. Cladistic analyses of early actinopterygian relationships have tended to rely almost exclusively on dermal characters (Gardiner and Schaeffer 1989;Cloutier and Arratia 2004;Friedman and Blom 2006), reflecting Gardiner and Schaeffer's (1989) view that neurocrania were too anatomically conserved to be of use in determining early actinopterygian relationships. Consequently, placement of taxa known largely or exclusively from endoskeletal remains impossible (e.g. Lawrenciella, Hamel and Poplin 2008;Kansasiella, Poplin 1974). Further problems arise from the variable nature of the dermal skeleton. Many dermal characters reflect the overall nature of the dermal skeleton as either an armour of regularly patterned plates (as in placoderms and osteichthyans) or fields of only regionally differentiated denticles or tesserae (as in acanthodians and chondrichthyans). Neurocrania are, by contrast, present in all vertebrates, and their rich complement of characters is thus comparable across almost the entirety of vertebrate diversity. The expansion of the actinopterygian character set to include more endoskeletal characters will help place problematic taxa and facilitate their routine inclusion into phylogenetic analyses of early actinopterygian relationships, thus increasing the sample set of Devonian and Carboniferous actinopterygians. Since Gardiner and Schaeffer's (1989) denouncement of neurocranial characters, the discovery of additional taxa preserving braincases (e.g. Lawrenciella, Hamel and Poplin 2008;Gogosardina, Choo et al. 2009), as well as work revealing new diagnostic characters within these structures (Coates 1998(Coates , 1999Giles and Friedman 2014), has made clear their importance for unravelling early actinopterygian relationships. The utility of neurocranial character sets has also been established as a tool in resolving systematic issues in other taxonomic groups (e.g. chondrichthyans, Coates and Sequeira 1998, Maisey and Anderson 2001lungfishes, Friedman 2007; tetrapodomorph fishes, Coates and Friedman 2010).

CONCLUSION
The endoskeleton of the Eifelian Cheirolepis trailli, from the Middle Old Red Sandstone of Scotland, is described here on the basis of lab-and synchrotron-based lCT. The unique combination of primitive osteichthyan and derived actinopterygian characters revealed in the endoskeleton confirms the position of Cheirolepis as the sister taxon to all other actinopterygians, and mirrors the atypical actinopterygian morphology seen in the dermal skeleton (Pearson and Westoll 1979). The presence of ascending processes of the parasphenoid in Howqualepis and Tegeolepis confirms that these appeared fairly early in actinopterygian evolution, and were secondarily lost or reduced in Mimipiscis, as concluded by Choo (2011).
The emerging picture of the endoskeleton in early rayfinned fishes is that it looked broadly like that of early sarcopterygians, with an uninvested dorsal aorta, large vestibular fontanelles, and an elongate basioccipital region. This is far removed from the endoskeletal anatomy of Mimipiscis, which shares more features with stratigraphically younger actinopterygians, as well as having secondarily lost or reduced key features. The endoskeleton of early sarcopterygians perhaps more accurately reflects aspects of the primitive osteichthyan, including those features outlined above. The discovery of further specimens of Cheirolepis preserving the braincase, particularly material that is three-dimensionally preserved, may provide a way of further corroborating these observations.