A giant dapediid from the Late Triassic of Switzerland and insights into neopterygian phylogeny

A new Triassic neopterygian is described on the basis of a large three-dimensional neurocranium from the Rhaetian (Late Triassic) of the Kössen Formation (Schesaplana, Grisons, Switzerland). CT scanning reveals neurocranial features similar to Dapedium, suggesting that this new genus, Scopulipiscis saxciput gen. et sp. nov., was deep-bodied and potentially durophagous, although no associated dental material is known. An expanded phylogenetic analysis of actinopterygians resolves Dapediidae as a clade (inclusive of Tetragonolepis), although fails to recover any characters supporting the monophyly of the genus Dapedium. Dapediids are resolved as stem holosteans, filling a conspicuous gap in early neopterygian relationships. Pycnodonts, previously suggested as either stem teleosts or the sister group to dapediids, are resolved as a clade on the neopterygian stem. Similarities between the new taxon described here and Dapedium provide insights into morphological disparity within early members of the group—suggesting that the ecological expansion of dapediids originated prior to the End-Triassic extinction—as well as contributing to a growing understanding of endocranial anatomy in Palaeozoic and Early Mesozoic actinopterygians.


Introduction
Ray-finned fishes (Actinopterygii) constitute the majority of living vertebrate diversity, displaying a huge disparity of body forms filling a wide variety of ecological niches [1,2]. Almost all extant actinopterygians (more than 30 000 species) belong to a single group, the Teleostei, with non-teleostean actinopterygians comprising just 56 living species [3]. The distribution of extant species belies past diversity: teleosts  the Cretaceous [4,5]. Indeed, their sister group, the holosteans, outstripped teleosts in terms of diversity and disparity in the earliest part of their history [5], although only eight species survive in the modern. Despite this, the understanding of relationships among early holosteans remains poor, in particular with regard to taxa branching outside of the living radiation. A bare stem is typically recovered in most analyses (e.g. [6][7][8][9]), although a stem holostean identity has recently been suggested for Dapedium [10].
Representing an early experiment in body form and ecological diversity, Dapedium is a modestly diverse (approx. 20 species) radiation of neopterygians known from the Late Triassic (Rhaetian) to mid Jurassic (earliest Aalenian), most notable for its adaptations for durophagy. Members of this genus possess thick, interlocking ganoid scales; dense, pustular ornament; and a small mouth with styliform teeth borne on the stout lower jaw. Other genera closely associated with Dapedium, and typically united into Dapediidae (or Dapediiformes), include Aetholepis [11], Dandya [12], Hemicalypterus [13,14], Heterostrophus [15], Paradapedium [16], Sargodon [12] and Tetragonolepis [16,17], although the dapediid affinity of at least Heterostrophus and Dandya has been called into question [13,16]. Most Dapedium species have a standard length of around 15 -35 cm [18], although dapediids range in size from approximately 8 cm (Dapedium noricum [12]) to approximately 1 m (Sargodon tomicus [12]). Only around five of the 30 or so known species of dapediid are present in the Triassic, and are exclusively represented by isolated teeth [19] or flattened [12,14,20] remains. Consequently, their adaptation for durophagy is thought to have facilitated the group's success in the aftermath of the End-Triassic extinction, a time at which many other durophagous taxa went extinct [21,22].
The equivocal phylogenetic signal of dapediids is in part due to the conflicting datasets used to resolve relationships. The phylogenetically informative endoskeleton [32,52,53], in particular the braincase, is described for only a small number of dapediids (Dapedium punctatum, D. caelatum [18]), and exhaustively in just a single taxon (Dapedium sp. [27,38,40,54,55]). While three-dimensional crania are known in a number of taxa (e.g. Tetragonolepis semicincta [17]; Dapedium granulatum [22], the majority of dapediids are preserved as laterally compressed fossils, with heavy dermal ornament obscuring most of the internal skeleton. In particular, no neurocrania are described from the Triassic, when the group first appeared. Here we present a new, large (approx. 17 cm length) braincase from the Rhaetian (Late Triassic) of the Kö ssen Formation (Schesaplana, Grisons, Switzerland; figure 1). Marine reptiles and pterosaurs have previously been described from this locality [56][57][58]. Actinopterygian remains attributed to Birgeria have been described from a lateral extension of the formation into Austria [59], but no dapediids have previously been identified. This new specimen expands our knowledge of the taxonomic richness of the locality and increases the size range known for dapediids, with implications for the diversification of the group and the timing of their radiation. Furthermore, CT scanning of this well-preserved neurocranium presents an opportunity to revisit the phylogenetic affinities of the group.

Phylogenetic analysis
Our dataset is based on Giles et al. [8], with 25 additional characters both novel and taken from the literature [10,14,33,36,45,46,49]. Eighteen taxa were added to the matrix in order to increase the sampling of dapediids and investigate the relationship of pycnodonts to other neopterygians. Brachydegma caelatum was removed pending a redescription of this taxon. This gives a total of 291 characters and 110 taxa. Full details are given in the electronic supplementary material. An equally weighted parsimony analysis was conducted using a heuristic search in PAUP* v. 4.0a158 [60] with the following settings: 1000 random addition sequences, five trees held at each step, maxtrees set to automatically increase, nchuck ¼ 10 000, chuckscore ¼ 1, tree bisection and reconstruction strategy enabled. Seven characters (c.89, c.91, c.151, c.171, c.241, c.247, c.251) were ordered following Giles et al. [8]. Taxonomic equivalence [61] was assessed using Claddis [62]

Etymology
Generic name describes the marine deposits on the alpine cliff sides of Schesaplana Scopuli-from scopulus (Latin)-meaning a lookout place, cliff or crag, in or under the sea-, piscis (Latin) for fish. In the specific epithet, sax-(Latin) is for 'rock' and -ciput from caput (Latin) for head because the holotype specimen PIMUZ A/I 3026 is a well-ossified neurocranium.

Holotype
PIMUZ A/I 3026 (figure 2), a neurocranium and associated dermal elements infilled by a fine carbonate matrix including small (approx. 2-8 mm) bivalves. Collected in 1976 by Dr Heinz Furrer, PIMUZ, the specimen was subsequently prepared mechanically.

Locality and horizon
The specimen was collected as float from the Kö ssen Formation of Schesaplana Mountain, in Canton Grisons/Graubü nden, Switzerland (figure 1), in the lowest part of the Alplihorn Member, Late Triassic [56], where the boundary between the Late Norian and Early Rhaetian is currently undefined (Heinz Furrer 2016, personal communication). Actinopterygians from the same locality include fragmentary elements and teeth from Birgeria [64] and isolated elements in the collections at PIMUZ assigned to Sargodon and Lepidotes. The Austrian portion of the Kö ssen Formation has produced other fish remains [59]. Other taxa from Schesaplana include marine reptiles [56,58] and pterosaurs [57].

Diagnosis
Large neopterygian with a well-ossified neurocranium and the following combination of characters: spiracular canal extending through the postorbital process to open on the ventral surface of the braincase; elongate hyomandibular facet approximately one-fourth of neurocranial length; pronounced median occipital crest, large sinus between lateral cranial canals, parasphenoid keel extending dorsally between orbits.

General
PIMUZ A/I 3026 preserves a partial skull roof and a near-complete neurocranium (17 cm preserved length), missing the nasal capsules and anterior portion of the orbit. There is some distortion, with the right orbit partially crushed and the basiocciput sheared slightly to the right. The skull roofing bones have also collapsed in the region of the anterior dorsal fontanelle. The complete ossification of the braincase and partial obliteration of dermal sutures suggests the cranium belonged to an adult specimen. Sutures between braincase ossifications are not apparent; the neurocranium appears to be ossified as a single unit. Some areas of perichondral bone were weathered away prior to collection, exposing cancellous bone, particularly in the occipital region and along the parasphenoid.

Skull roof
The dermal bones of the skull roof are heavily ornamented with pustular tubules, from 0.5 to 1 mm in diameter, radiating from ossification centres and increasing in size towards element margins (figures 2 and 3). The tubules are particularly large around the midline where the frontals and parietals meet. Such dense ornament means that sutures between bones are typically difficult to trace, although they  can sometimes be followed in CT scans. The frontal is relatively short, accounting for a little under half the length of the skull roof (fr, figure 3d). It is widest anterior to the parietal contact, and tapers anteriorly. There is no pineal foramen, and the frontals appear to be fused at the midline. Only a small fragment of the left nasal is preserved, but an unornamented area at the anteriormost point of the left frontal represents the overlap area for this bone (n.ov, figure 3d). A series of pores marks the passage of the supraorbital sensory canal through the frontal and into the parietal (soc, figure 3d ), and the path of the canal can be followed in CT sections. Very little of the midline suture is preserved between the elongate parietals ( pa, figure 3d ), which have partially collapsed into the anterior dorsal fontanelle (adf, figure 3c). The suture between the rectangular dermopterotic and parietal (dpt, figure 3d) is obscured by ornament externally, but can be traced in CT sections. However, it is unclear whether the dermopterotic contacts the frontal anteriorly. As with the supraorbital canal, the path of the otic sensory canal through the dermopterotic and into the dermosphenotic is indicated by a line of pores (ioc, figure 3d ), with the canal visible in CT slices.

Occipital region
The occipital region is the deepest part of the braincase, with the basiocciput accounting for half the vertical height of the specimen (figures 4c,d and 5). It is narrower than either the otic or orbitotemporal   regions (figure 6a,b). A 'bulge' on the midline projects dorsally above the skull roof (bul, figures 4d and 5b,d ) and houses a large cavity within the perichondrium of the braincase (cav, figure 7a). The posterior face of the occiput is developed into a broad median occipital crest, punctuated by a large ligamentous pit (lig, figures 4d and 7a). A process lateral to this pit is pierced by small anastomosing branches for the occipital nerve (focn, figures 4d and 5d). Medial to the posterodorsal angle of the braincase is a deep posttemporal fossa ( ptf, figures 4d and 5b,d), which is continuous anteriorly with the fossa bridgei (fb, figure 4d). The foramen magnum is small and triangular, and both its floor and roof are well-ossified (fm, figures 4d and 7a). Two foramina for the occipital arteries open onto the lateral face of the braincase just above the parasphenoid (foca, figures 4d and 5d), with the smaller opening for the occipital nerve dorsal to these (focn, figures 4d and 5d). The notochordal facet is large and rounded (not, figure 4d), and leads into a narrow notochordal canal that pinches out approximately halfway along the length of the basiocciput (not, figures 7a,b and 8b). Ventral to the opening for the notochord is the circular aortic canal, which is enclosed ventrally by the parasphenoid (aort, figures 4d, 6b and 7a). A dorsally directed canal leaves the dorsal margin of the aorta to open on the lateral wall of the braincase above the parasphenoid. There is no indication of either a posterior dorsal fontanelle or otoccipital fissure.

Otic region
The otic region is the widest part of the braincase ( figure 6a,b). Much of its dorsal surface is unossified to accommodate the large anterior dorsal fontanelle, which extends from the posterior margin of the hyoid facet to above the orbits (adf, figures 3c, 4b and 9b ). Lateral to the fontanelle is an excavation for the fossa bridgei (fb, figures 8b and 9), which extends posteriorly to the post-temporal fossa and anteriorly to the  postorbital process. The lateral margin of the fossa bridgei is strongly concave, wrapping around the hyomandibular facet. The medial margin is convex, with multiple small projections giving an irregular edge. A total of five canals connect with the fossa bridgei. Two canals issue from its anteromedial corner. The more ventral, for the spiracular canal, enters the top of the postorbital process and travels along its entire length, opening just posterior to the lateral commissure on the ventral surface of the braincase (spic, figures 5b, 6b and 9e ). On the left side, the lateral commissure is broken, and consequently this canal appears to open midway along its length. The more dorsal of the two anteromedial canals opens into the posterolateral corner of the orbit, just below the orbital roof, and transmitted the otic nerve (otn, figures 4a, 6b,d and 8b). An additional canal leaves more posteriorly from the medial margin and opens into the roof of the hyomandibular facet and may have transmitted the recurrent lateralis branch of the facial nerve (frla, figures 6b and 8b). Two narrow canals exit the anterolateral corner to open on the side of braincase just below the skull roof, and probably housed branches of the otic nerve (otn, figures 5b,d, 8b and 9b,e). Dorsal to the fossa bridgei is a separate small cavity, which also lies within the perichondrium of the braincase. Owing to specimen preservation, the chamber is only fully observable on the right side. The chamber has a medial connection to the orbit, opening into the dorsolateral corner of the posterior wall (soph2, figure 4b). An additional small canal leaves the lateral margin of the chamber to open onto the lateral surface of the braincase dorsal to the otic nerve canals (ca, figure 5b). The articular facet for the hyomandibular is elongate and horizontally oriented (hmf, figures 5b,d and 6b), and a deep dilatator fossa lies dorsal to the hyoid articulation (dil, figures 5b,d and 6b). Ventral to the facet, the lateral wall of the braincase narrows towards the midline, and this face is pierced by a number of canals. The most posterior of these is the lenticular opening for the vagus nerve (X, figures 5b,d and 6b), positioned ventral to the posterior margin of the hyoid facet. A shallow depression between the vagus nerve and hyoid facet is the supratemporal fossa (stf, figures 5d and 6b), delimited anteriorly by a vertical strut. The glossopharyngeal nerve ( probably its supratemporal branch) exits the braincase via a small canal at the base of this strut (IXst, figures 5b,d and 6b), adjacent to an elliptical pit. This pit appears to be at least partly the result of mechanical preparation, as it punches through the perichondrium to reach the saccular chamber on one side, but comparison of both sides shows that the perichondral lining of the endocavity is complete, and there is no vestibular fontanelle. The original morphology of the pit is unclear, but it may represent an articular facet for the first suprapharyngobranchial (sup.ph, figure 5). A shallow groove for the jugular marks the lateral face of the braincase anterior to this facet, and can be traced to a foramen in the lateral commissure ( jug, figures 5b,d and 6b). A large opening on the ventral margin of the jugular groove marks the exit of a stout canal, which originates on the lateral face of the sacculus near its anterior margin. This canal may be the main trunk of the glossopharyngeal nerve (IX, figures 5b,d and 6b). More anteriorly, a large opening for the hyomandibular branch of the facial nerve (VIIfhm, figures 5b,d and 6b) pierces the roof of the jugular groove. The jugular vein and main trunk of the facial nerve continue through the postorbital process into the orbit.

Orbitotemporal region
The posterior margin of the orbitotemporal region is formed by the lateral commissure and postorbital process ( por, figures 4b, 5b,d and 6b,d). The lateral commissure is reduced to a thin splint laterally, and the extent to which it is covered by the ascending process of the parasphenoid is difficult to discern. The orbital walls are pierced by a complex anastomosing series of canals, some of which are difficult to identify.    Rendering (e,g) and interpretive drawing (f,h) of the basisphenoid and parasphenoid circulatory system in left lateral (e,f ) and ventral (g,h) view. Endocast in grey, canals in red. Scale bar is 50 mm. aa, anterior ampulla; anas, anastomosing canal; aort, dorsal aorta; asc, anterior semicircular canal; bhc, buccohypophyseal canal; can.bsp, unidentified canal in basisphenoid; cc, crus commune; epsa, efferent pseudobranchial; fb, fossa bridgei; foca, occipital artery; focn, occipital nerve; fm, foramen magnum; frla, recurrent lateralis branches of the facial nerve; ha, horizontal ampulla; hsc, horizontal semicircular canal; ica, internal carotid; lcc, lateral cranial canal; mcv, middle cerebral vein; myo, myodome; not, notochord; opa, ophthalmic artery; optl, optic lobe; otn, otic nerve; pa, posterior ampulla; pal, palatine artery; psc, posterior semicircular canal; pv, pituitary vein; sacc, sacculus; spic, spiracular canal; ssu, sinus superior; tel, telencephalon; utr, utriculus; v?, unknown vein; III/Vprof oculomotor nerve and profundus branch of the trigeminal nerve; IV, trochlear nerve; V, trigeminal nerve; VI, abducens nerve; VII, facial nerve; VIIfhm, hyomandibular branch of the facial nerve; IX, glossopharyngeal nerve; IXst, supratemporal branch of the glossopharyngeal nerve; X, vagus nerve. rsos.royalsocietypublishing.org R. Soc. open sci. 5: 180497 contact the visceral surface of the skull roof (soph, figures 4b and 6d). The two canals in the posterolateral corner of the orbit have different points of origin: the more ventral, for the otic nerve (otn, figures 4b, 6d, 8b and 9b,e), connects to the fossa bridgei; and the more dorsal, which may also carry branches of the superficial ophthalmic (soph2, figures 4b and 6d), to the chamber that lies above the fossa bridgei. Of the two canals in the posteromedial corner of the orbit, the more anterior and dorsal carries the trochlear nerve from the optic lobes (IV, figures 6d, 8b and 9b). The more posterior originates from the dorsal part of the midbrain, above and some way posterior to the trochlear nerve, but its identity is unclear (v?, figures 6d, 8b and 9e). Two canals open on the medial wall of the orbit via a single foramen. These canals have a single root on the forebrain on the right of the specimen and a double root on the left, and transmitted the oculomotor and profundus nerves into the orbit (III and Vprof, figures 6d and 8b). The final set of canals, approximately midway up the posterior orbital wall, is centred about the trigeminofacialis chamber (tfc, figure 4b), which is continuous posteroventrally with the jugular canal. Two canals exit from the same point on the anterior face of the utriculus before separating into separate canals. These canals enter the orbit separately, but both are continuous within the trigeminofacialis chamber. The more dorsal transmits the trigeminal nerve (V, figures 4b, 6d, 8b and 9e) and the more ventral the facial nerve (VII, figures 4b, 6d and 8b,f ). Two grooves extend from the dorsal margin of the chamber. The groove on the posterior margin of the orbit is short, with no canal at its dorsal limit. It terminates some way ventral and medial to any of the canals that connect to the skull roof, but may have transmitted the superficial ophthalmic nerves, which entered the orbit via the underlying chamber. The groove on the medial orbital wall is deeper, and a canal at its dorsal margin connects to the cranial cavity via two closely positioned openings. This probably transmitted the internal carotid artery (ica, figures 4b, 6d and 8d). A ventral connection links the trigeminofacialis chamber and the median posterior myodome (myo figures 4b, 6d and 8b,f,h) 4b, 6d and 8b,f,h). There is no obvious opening for the palatine branch of the facial nerve into the orbit. The braincase is broken midway along the orbit, and the skull roof has partially collapsed. A single opening on the midline represents where the braincase is broken through the endocranial cavity of the forebrain (endo, figures 4b and 6d); neither the exit of the optic (II) nor olfactory (I) nerve is preserved. The basisphenoid is well ossified, contributing to the medial walls and floor of the orbit, although the interorbital septum is formed entirely by the parasphenoid ( psp, figures 4b, 5d, 6d and 7a). The basipterygoid process, which has both a dermal and endoskeletal component, is robust. The internal carotid opens into the posterior margin of the basisphenoid via a foramen anterior to the posterior myodome. The subocular shelf is narrow and formed entirely by the parasphenoid.

Parasphenoid
The parasphenoid extends the full length of the preserved braincase, although a fragment is missing in the middle where the specimen is broken between the midpoint of the basiocciput and the postorbital process. The posterior margin of the parasphenoid is weathered, exposing the basioccipital, but a midline aortic notch appears to have been absent (figures 4 and 6). Although housed entirely within the basiocciput posteriorly, most of the floor of the aortic canal is formed by the parasphenoid (aort, figure 7a). As the parasphenoid is broken midway along the basiocciput, the nature of the bifurcation into the lateral aortae is unknown. A narrow median keel on the ventral surface of the parasphenoid (ke, figures 5b,d and 6b) is flanked by paired shallow grooves (gr, figures 5b and 6b); the depth of the left groove has been exaggerated during specimen preparation. As the dorsal and lateral aortae are enclosed within the braincase, the purpose of these grooves is unclear. The parasphenoid extends up the lateral walls of the braincase to the level of the foramen magnum, effectively cloaking the basiocciput. The ascending processes of the parasphenoid are narrow and strut-like, although it is difficult to assess how far they extend dorsally up the postorbital process. The basipterygoid processes are robust, projecting dorsally as well as laterally (bpt, figures 4b, 5b,d and 6b,d), and the ophthalmic artery pierces the orbital floor posteromedial to the process (opa, figures 4b, 5d and 6d,e). An opening at the anterior limit of the basipterygoid process marks the exit of the palatine nerve and artery from the basisphenoid ( pal, figures 4b, 5b,d and 6b,d,e). A V-shaped ridge is present on the ventral surface of the parasphenoid, and the efferent pseudobranchial artery exits posterior to where the two arms of the ridge converge on the midline (epsa, figures 5d and 6b,e). Anteriorly, the buccohypophyseal canal exits the parapshenoid on the midline (bhc, figure 6b). The internal carotid foramen is not preserved, and presumably entered the parasphenoid through the missing section. A broad parasphenoid keel extends dorsally between the orbits, forming the interorbital septum ( psp, figures 4b, 5d, 6d and 7a). No teeth are present on the parasphenoid.

Endocast
The regions of the endocranial cavity corresponding to the midbrain, hindbrain and bony labyrinth are almost completely preserved, although only a short portion corresponding to the narrow forebrain is present in the specimen (tel, figure 8). As the internal walls of the braincase are sometimes poorly mineralized (figure 7a), contrast between the braincase and infilling sediment is low, with the exact boundaries of some regions of the endocranial cavity difficult to discern.

Midbrain
The area of the optic lobes (midbrain) is large, representing the widest part of the endocast (optl, figure 8b). Much of the dorsal surface is unfinished, indicating the position of the extensive anterior dorsal fontanelle (adf, figure 8b). Two closely associated canals leave from the anterolateral face of the midbrain, entering the orbit via a single (on the right of the specimen) or double (on the left) opening. These probably transmitted the oculomotor and profundus nerves (III/Vprof, figure 8b). The trochlear nerve leaves the cranial cavity posteriorly and dorsally (IV, figures 8b and 9b).

Hindbrain
There is no clear division between the midbrain and hindbrain, although the portion of the hindbrain positioned behind the crus commune sits far ventral relative to the rest of the endocavity, perhaps indicating that the cerebellum projected into the fourth ventricle. Cerebellar auricles cannot be identified. Several canals exit the cranial cavity in the anterior part of the hindbrain. Dorsally, a stout canal transmits an unidentified vessel into the dorsal part of the orbit (v?, figure 8b). The middle cerebral vein exits the cranial cavity some way ventral to this (mcv, figure 8b), and the internal carotid enters more ventrally still (ica, figure 8b). The trigeminal and facial nerves exit from the angle between the midbrain and utricular recess (V and VII, figure 8b). Posterior to the labyrinth, the vagus nerve leaves the ventral margin of the hindbrain through a large, anteroposteriorly elongate canal (X, figure 8b). The region corresponding to the spinal cord is by far the narrowest portion of the endocast, and from its lateral margins issue a dorsal, bifurcating canal for the occipital nerves (focn, figures 8b and 9b,e ), and a ventral, single canal for the occipital artery (foca, figures 8b and 9b,e ).

Labyrinth
The walls of the bony labyrinth are fairly well ossified, except in the region of the sinus superior and anterior ampulla. The three semicircular canals are narrow, and all are fully enclosed in bone (asc, hsc, psc, figures 8b and 9b,e). Both the anterior and posterior canals are anteroposteriorly short, and the horizontal canal is almost isoclinal in its curvature (figure 9c). The orientation of the anterior ampulla is unclear (aa, figure 8b). The utriculus is dorsoventrally tall, and there is a slight suggestion of division into dorsal and ventral components (utr, figure 8b). A small saccular notch is present (sac, figure 8b). Although incompletely ossified, the sinus superior is long (ssu, figure 8d), and the crus commune projects above the roof of the hindbrain and slightly towards the midline (cc, figures 8d and 9b). The lateral cranial canal is very well developed (lcc, figures 8b and 9b,e ). As well as anterior and posterior connections to the cranial cavity, it has a dorsal extension that almost reaches the underside of the skull roof. At least a single canal, probably for the endolymphatic ducts (end.d, figure 7a), connects the cranial cavity to the median sinus positioned between the lateral cranial canals.

Circulatory system
The buccohypophyseal canal is elongate, leaving the anteriormost point of the posterior myodome and continuing anteroventrally (bhc, figure 8). A narrow canal branches from its lateral margin level with the exit of the internal carotids into the orbit (can.bsp: unidentified canal in the region of the basisphenoid, figure 8f,h). Further anteriorly, the buccohypophyseal canal bends dorsally in an 's' shape, before opening onto the ventral surface of the parasphenoid. The path of the internal carotid artery is incompletely known, largely due to the loss of part of the parasphenoid. It presumably entered the parasphenoid posterior to the basipterygoid process (?ica, figure 6e) and continued via an anterodorsally directed canal. This canal opens onto the orbital floor posteromedial to the basipterygoid processes (ica, figures 6e and 8f,h), and enters the cranial cavity through a foramen in a dorsally directed groove medial to the trigeminofacialis chamber (ica, figure 6d,e). The efferent pseudobranchial pierces the parasphenoid beneath the basipterygoid process (epsa, figures 5d, 6d and 8f,h) and is transmitted through a vertical canal, exiting onto the floor of the orbital as the ophthalmic artery (opa, figures 4b, 5b,d, 6d,e and 8f,h). The internal carotid and efferent pseudobranchial communicate via a short anastomosing canal (anas, figures 6e and 8f ), and are also linked via the parabasal canal ( pbc, figure 6e), which seems to have housed the palatine nerve and artery. The path of the palatine into the orbit is not known, but it appears to have entered the parabasal canal through the same opening as the internal carotid, and opens anteriorly on the lateral margin of the parasphenoid ( pal, figures 4b, 5b,d, 6b,d,e and 8f,h). The internal carotids and efferent pseudobranchials do not anastomose with their antimeres, but a narrow transverse canal connects the two ophthalmic arteries.

Body size
The neurocranium of Dapedium is inclined at approximately 508 from the horizontal in articulated specimens, and body size typically ranges from around 15 to 35 cm [18]. Skull length accounts for approximately 18% of body length in Dapedium, but accounts for a much smaller proportion (approx. 10%) in the largest known dapediids. Body length of a large Sargondon tomicus is estimated about 1 m [12], with absolute head length of a medium-sized specimen approximately 6 cm and of the large specimen approximately 10 cm [65].
The Although represented only by an isolated neurocranium, this specimen bears conspicuous similarities to dapediids, in particular the braincase described for Dapedium sp. [27,55]: short aortic canal, parasphenoid 'wings' cloaking the basiocciput, median occipital crest punctuated by a large ligamentous pit, parasphenoid extending dorsally between the orbits, fossa bridgei continuous posteriorly with the post-temporal fossa, median sinus between the lateral cranial canals, canals for endolymphatic ducts connecting the cranial cavity to the median sinus and absence of differentiated braincase ossifications. Similar likenesses are found with the endocast of Dapedium sp. [40], including a large lateral cranial canal extending dorsal to the endocranial roof, narrow semicircular canals and a lenticular utriculus. The majority of actinopterygian material from Schesaplana has been identified as either Birgeria or Lepidotes [66] although unpublished elements in the collections at PIMUZ have been referred to the giant dapediid taxon Sargodon (A. Latimer 2016, personal observation), known from the Late Triassic (Norian) of Italy [12]. The limited anatomical overlap between Sargodon and the braincase described here makes comparisons difficult: Sargodon is known from articulated but laterally compressed fossils, with no part of the braincase, and only small portions of the skull roof, described. Differences are evident in some of the few comparable regions, for example the parietals and dermopterotic of Sargodon are quadrate, and those of the specimen described here are markedly rectangular. Furthermore, the cranium of even the largest Sargodon specimen known is only around 10 cm in length. By contrast, the braincase described here is approximately 17 cm, and is incomplete anterior to the orbit, suggesting a total cranial length of at least 20 cm. Consequently, we feel justified in erecting a new genus and species for this specimen: Scopulipiscis saxciput. It is possible that undescribed cranial elements at PIMUZ belong to Scopulipiscis saxciput rather than Sargodon, as their identification was most probably based on size.
In the strict consensus tree of our phylogenetic analysis (figure 10), Scopulipiscis saxciput is resolved in a polytomy with all included Dapedium species, Heterostrophus phillipsi, Sargodon tomicus, Dandya ovalis, Hemicalypterus weiri, Paradapedium egertoni, Tetragonolepis oldhami and Tetragonolepis semicincta. This clade is supported by seven homoplastic characters: c.58, one or two suborbitals; c.156, braincase ossifications not differentiated; c.247, complete series of dorsal ridge scales; c.258, caudal fin not forked; c.269, suborbitals extend ventral to orbit; c.275, hem-like median fins; c.276, long-based anal fin. The monophyly of Dapedium is not upheld, and the genus is recovered as paraphyletic in both the Adams consensus tree and agreement subtree (electronic supplementary material, figure S1). Additional characters found in dapediid taxa to the exclusion of immediate outgroups, but which cannot be optimized due to uncertainty over the primitive condition in the clade, include: c.270, coronoids contribute to lateral dentition field; c.273, dorsal extension of parasphenoid between orbits; c.274, ligament pit on posterior face of braincase; c.278, ossified centra absent.
The failure to resolve Dapedium as monophyletic is hardly surprising, given that placement of material in this genus traditionally largely relied on the presence of a deep-bodied morphology, and differentiation of species on characters that could covary in individual specimens [12]. Doubts over the monophyly of the genus have been raised before [28], and it is possible that detailed investigation of Sinemurian material from Dorset may find further objections. Only four dapediid taxa are excluded in the agreement subtree (electronic supplementary material, figure S1a), suggesting that only some of the uncertainty stems from conflicting endoskeletal and dermal character sets. The monophyly of dapediids, in contrast, seems well supported, albeit with a different set of characters to those identified in previous phylogenetic analyses [14,37].

Placement of dapediids
Dapediids (along with Hulettia) form a clade of stem holosteans, sharing with other holosteans the following five homoplastic characters: c.89, two infradentaries; c.112, anterodorsal process of suboperculum; c.135, anterodorsal myodome single; c.173, parasphenoid with multifid anterior margin; c.178, aortic notch in parasphenoid absent. While dapediids have been recovered as total group holosteans before, they are more typically resolved as the sister group to ginglymodians [36,37]. A stem holostean position was recovered by Ló pez-Arbarello & Sferco [10] on the basis of two homoplastic characters (four or more supraorbitals and a presupracleithrum), neither of which are optimized as supporting this relationship in the current analysis. A position deep on the holostean stem goes some way to explaining rsos.royalsocietypublishing.org R. Soc. open sci. 5: 180497 the peculiar combination of apparent holostean and teleost features that have influenced past hypotheses of phylogenetic relationships.
The position of Hulettia as sister taxon to dapediids, with the support of three homoplastic characters (c.247, basal fulcra on dorsal fin; c.249, scutes anterior to anal fin; c.284, preoperculum shorter than operculum), is unexpected. We note that Hulettia shows large numbers of character reversals (for the full list of character optimizations see electronic supplementary material, Supplementary Data), suggesting that this position may not be robust. However, Hulettia has previously been associated with Dapedium [32]. Macrepistius is resolved as sister to all other holosteans, supported by one unambiguous (c.10, olfactory nerve pierces premaxilla) and five homoplastic characters (c.14, single median dermal bone capping snout; c.50, tube-like canal bearing anterior arm of antorbital; c.61, circumorbital ring of supraorbitals and infraorbitals closes the orbit; v.107, paired vomer; c.161, sphenotic with small dermal component). The holostean crown is united by just three homoplastic characters (c.175, parasphenoid teeth small; c.176, parasphenoid not pierced by internal carotid; c.272, parasphenoid wings around basiocciput absent), indicating that features previously identified as uniting the living radiations may in fact be more widely distributed across the total group.

Placement of pycnodonts
Pycnodonts are recovered as stem neopterygians in our analysis. The monophyly of pycnodonts is robustly supported by one unambiguous (c.282, prearticular symphysis) and thirteen homoplastic characters. Discoserra is recovered as the immediate outgroup of pycnodonts, with a clade comprising Bobasatrania þ Ebenequa falling as the sister taxon. While this broader clade is supported by five homoplastic characters, we do not regard it as robust: at least some of the characters optimized as supporting this group (e.g. c.276, long-based anal fin) probably reflect anatomical similarities related to a deep-bodied morphology rather than a close phylogenetic relationship. Furthermore, Platysomus is resolved as sister taxon to an unlikely clade comprising cladistians and chondrosteans. These incongruent placements highlight that further work is needed to resolve the relationships of other radiations of deep-bodied Palaeozoic fishes.
A limited understanding of the internal skeleton of pycnodonts-and those taxa recovered as closely related-may also be a contributing factor. Despite an extensive fossil record stretching from the Late Triassic to the Eocene, few pycnodont neurocrania are described: the most comprehensive description of an early occurring pycnodont braincase is based on a specimen that now appears to be lost (Mesturus sp. [67]). Derived pycnodont neurocrania are known from well-preserved Cretaceous specimens of Neoproscinetes penlavi and Iemanja palma from Brazil, but are poorly ossified [50,68]. The earliest described pycnodonts are named from articulated but two-dimensional remains from the Norian of Italy [69], with isolated dental elements found in Norian and Rhaetian deposits of Belgium and Luxembourg [70]. As such, we consider our results relating to pycnodonts as preliminary: only a small proportion of the vast diversity of pycnodonts are included, and only one for which the braincase is known in detail. In particular, detailed description of primitive pycnodonts (for example, material from the Early Norian of Italy mentioned as possessing lateral gulars [71]), as well as three-dimensional neurocranial material, may help clarify the position of the group.

Dapediid diversity
Ray-finned fishes with an adaptation to durophagy were well represented in the Triassic [21,72], although dapediids are thought to have made up a fairly small proportion of this diversity. Although many durophages were wiped out during the End-Triassic extinction, dapediids survived and went on to radiate in the Jurassic, presumably filling vacated ecospace [73]. The discovery of a new, exceptionally large [74] dapediid in the Late Triassic indicates that dapediids may have been more diverse before the End-Triassic extinction than previously thought. Although dental elements are not preserved for Scopulipiscis saxciput, similarities in braincase anatomy to Dapedium seem likely to perform a functional role in strengthening the skull during durophagy, including the pronounced midline occipital crest, the deep insertion point for a longitudinal ligament, and the extension of the parasphenoid between the orbits and around the basioccipital. Many of these specializations are seen in other durophagous actinopterygians, including Carboniferous eurynotiforms [75] and Eocene gymnodonts [76].