The Giant Cretaceous Coelacanth (Actinistia, Sarcopterygii) Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, and Its Bearing on Latimerioidei Interrelationships

We present a redescription of Megalocoelacanthus dobiei, a giant fossil coelacanth from Upper Cretaceous strata of North America. Megalocoelacanthus has been previously described on the basis of composite material that consisted of isolated elements. Consequently, many aspects of its anatomy have remained unknown as well as its phylogenetic relationships with other coelacanths. Previous studies have suggested that Megalocoelacanthus is closer to Latimeria and Macropoma than to Mawsonia. However, this assumption was based only on the overall similarity of few anatomical features, rather than on a phylogenetic character analysis. A new, and outstandingly preserved specimen from the Niobrara Formation in Kansas allows the detailed description of the skull of Megalocoelacanthus and elucidation of its phylogenetic relationships with other coelacanths. Although strongly flattened, the skull and jaws are well preserved and show many derived features that are shared with Latimeriidae such as Latimeria, Macropoma and Libys. Notably, the parietonasal shield is narrow and flanked by very large, continuous vacuities forming the supraorbital sensory line canal. Such an unusual morphology is also known in Libys. Some other features of Megalocoelacanthus, such as its large size and the absence of teeth are shared with the mawsoniid genera Mawsonia and Axelrodichthys. Our cladistic analysis supports the sister-group relationship of Megalocoelacanthus and Libys within Latimeriidae. This topology suggests that toothless, large-sized coelacanths evolved independently in both Latimeriidae and Mawsoniidae during the Mesozoic. Based on previous topologies and on ours, we then review the high-level taxonomy of Latimerioidei and propose new systematic phylogenetic definitions.


Introduction
Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994 is a giant marine coelacanth discovered in 1987 [1] in the Upper Cretaceous Bluffown Formation, southeastern USA. With an estimated length of 3.5 m, Megalocoelacanthus is among the largest known coelacanths. Similar dimensions (i.e. more than 2.0 m in total length) are reached by the late Jurassic-mid Cretaceous genus Mawsonia from North Africa and South America [2][3][4][5][6][7], and the Early Cretaceous genus Axelrodichthys from Brazil [5] also attained a large size. Previous phylogenetic analyses [8][9][10][11][12][13][14] supported a sister-group relationship of Mawsonia and Axelrodichthys within Mawsoniidae Schultze, 1993 [15]. This close affinity between toothless, large-sized Mesozoic coelacanths raised the question whether these features could be synapomorphies of a putative clade or have evolved independently in several lineages. The evolution of large sized, toothless coelacanths has been briefly discussed by Schwimmer et al. [1] who suggested that these features evolved independently in Mawsonia and Megalocoelacanthus. The authors have suggested that Megalocoelacanthus is closer to Macropoma and Latimeria than to Mawsonia. However, their assumption was only based on the comparison of meristic data of some anatomical features (table 2 in [1]), but not on a phylogenetic analysis.
Since its discovery, the relationships of Megalocoelacanthus with other coelacanths have never been investigated. Here we present a description of cranial and postcranial skeleton of Megalocoelacanthus based on new and holotype material. A phylogenetic analysis of 39 taxa and 110 characters is performed to clarify the position of Megalocoelacanthus among coelacanths, and its bearings on Mesozoic coelacanth's interrelationships are subsequently discussed. Implications for the coelacanth taxonomy will also be discussed and the application of phylogenetic definitions to coelacanth taxonomy will be proposed based on our novel topology.

Geological context
Megalocoelacanthus remains are known only from the United States. The holotype CCK 88-2-1 consists of cranial and branchials elements [1], and was found in the lower part of the Blufftown Formation in eastern Alabama, that is early Campanian in age (Figure 1). Remains of Megalocoelacanthus were also first reported in five additional regional localities from the southeastern states of Alabama and Georgia [1], along with a single coronoid fragment from New Jersey. Subsequently, unpublished specimens have been collected in coeval strata in Mississippi (Schwimmer, Earl Manning, pers. comm.), and Kansas. All known fossils of Megalocoelacanthus with well-known stratigraphic associations are of late Santonian to mid-Campanian age [1], except the New Jersey specimen which is in a late Campanian-early Maastrichtian deposit. However, the latter fossil is a highly ablated principal coronoid fragment, preserved in a near shore lag deposit, which may have been reworked from older material: its age is thus uncertain.

Material
This redescription of Megalocoelacanthus is mainly based on AMNH FF 20267, which was collected in 2007 in the Niobrara Formation, in the Northern Lane County, Kansas ( Figure 1). It is early Campanian in age and thus approximately coeval with the holotype CCK 88-2-1. The new specimen consists of skull (both ethmosphenoid and otoccipital portions), snout, lower jaws, gular plates, branchial arches, urohyal, hyoid skeleton, and shoulder girdle. Although most of these isolated remains are strongly flattened laterally, they are outstandingly preserved. Significant elements from the holotype specimen CCK 88-2-1 are also included in the description for further comments.

Nomenclatural Acts
The electronic version of this document does not represent a published work according to the International Code of Zoological Nomenclature (ICZN), and hence the nomenclatural acts contained in the electronic version are not available under that Code from the electronic edition. Therefore, a separate edition of this document was produced by a method that assures numerous identical and durable copies, and those copies were simultaneously obtainable (from the publication date noted on the first page of this article) for the purpose of providing a public and permanent scientific record, in accordance with Article 8.1 of the Code. The separate print-only edition is available on request from PLoS by sending a request to PLoS ONE, Public Library of Science, 1160 Battery Street, Suite 100, San Francisco, CA 94111, USA along with a check for $10 (to cover printing and postage) payable to ''Public Library of Science''.
In addition, this published work and the nomenclatural acts it contains have been registered in ZooBank, the proposed online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information The Giant Cretaceous Coelacanth Megalocoelacanthus PLOS ONE | www.plosone.org viewed through any standard web browser by appending the LSID to the prefix ''http://zoobank.org/''. The LSID for this publication is: urn:lsid:zoobank.org:pub:6F452EF9-02A6-4B3F-9E0B-3E4C14BADC80.

Anatomical description
1.1 Dermal bones of the skull roof. Despite strong lateral compression, all elements are very well preserved and allow detailed description. The parietonasal shield (Figures 2, 3, 4) is longer than the postparietal shield (Figures 5,6,7,8) very narrow and straight in lateral view. The intracranial joint is transversely straight. The parietonasal series consists of two pairs of parietals (Pa.a, Pa.p, Figures 2, 3), but the number of nasals cannot be assessed, and only the posteriormost nasal (Na, Figure 2) can be observed. However, its relationships with the elements of the anteriormost part of the ethmosphenoid portion of the skull are very difficult to observe. The bones of the parietonasal series are slender and elongated. The posterior parietals are the largest of the series as in Macropoma, Holophagus [12] and Swenzia [8,16], and their center of ossification is situated posteriorly. The posterior parietals meet the anterior parietals in a transverse indented suture. The median suture between the two paired parietals is straight and only leaves a narrow gap on the midline, probably due to the settling of the bones under lateral constraints during fossilization. The sutures between the different parietonasal elements (posterior parietal/anterior parietal; anterior parietal/ nasal) extend far laterally, and extend into the vacuity of the supraorbital sensory line canal at the level of the contact between the dermal bones ( Figure 2). The supraorbital series can be distinguished by the presence of sutures between to the parietonasal series and the pillars that are crossing the supraorbital sensory line canal (So, pi, so.s.c, Figures 2,3). However their precise number cannot be assessed due to the poor preservation of this area of the skull roof. On both side of the skull, an elbowed process (?v.pr.Pa, Figures 2, 3) extends just below the margin of joint between the posterior pair of parietal and the postparietal. This process could correspond to the pedicel of the ventral process of the parietal, which forms a ridge that slides in the groove on the postparietal.
The lateral rostral (L.r, Figures 2, 3) is a large bone which extends posteroventrally as a relatively short and flattened tube that encloses the infraorbital sensory line canal. There is no evidence of grooves on its dorsal surface, unlike in Diplurus, Macropoma, Latimeria, Laugia [12], Swenzia [8], and Rhabdoderma [17]. The lateral rostral usually separates the anterior nostril from the posterior nostril. Anteriorly, the ventral process of the lateral rostral (v.pr.L.r, Figures 3) is sutured to the lateral ethmoid, and its anterior margin is notched and corresponds to the posterior margin of the opening for the anterior nostril (nos.a, Figure 3) as in Macropoma (figure 6.10 in [12]). On both sides, the posterior margin of the lateral rostral is notched dorsally to the tube enclosing the infraorbital sensory line canal. A space is clearly observable on the left side, between the lateral rostral and the supraorbital series and suggests that the position of the opening of the posterior nostril (nos.p, Figure 3) was similar to that in other coelacanths. On the left side, a conspicuous suture is observable between the lateral rostral and the supraorbital series.
The tip of the snout is preserved in three dimensions, but is isolated from the rest of the skull (Figure 4). The snout is partially fused and consists of a pair of premaxilla, a median rostral, and the right anterior portion of the lateral rostral (Pmx, ros.m, L.r, Figure 4A). It shows no trace of strong distortion: its shape may thus reflect the actual width of the skull roof as in Whiteia [12] and Macropoma (figures 3.15, 3.19A in [12]). The tip of the snout appears to be strongly consolidated due to the tight suture between the premaxilla and the median rostral ( Figure 4A). Although the snout is heavily ossified, it may have been loosely attached to the lateral ethmoid, anterior nasal, and tectal (if present). This condition is also observed in Macropoma lewesiensis (BMNH 4207). The entire surface of the snout is ornamented with coarse rugosities, making it difficult to observe the suture between the bones precisely. The premaxilla is robust, bears no teeth, and is separated at the symphysis from its antimere by a large trapezoidal, median pore of the sensory line canal (m.p.s.c, Figure 4A). Such a condition is also observed in Macropoma (figure 3.20B in [12]) and Latimeria (figure 8 p.s.m.e.m in [18]), but the median pore of the sensory line canal is much larger in Megalocoelacanthus. The ventral half of the premaxilla is very broad and the base of the bone extends medially as a thin flattened surface that closes ventrally the median pore of the sensory line canal ( Figure 4A). The base of the premaxilla meets its antimere in a recess, posterior to the opening of the median pore of the sensory line canal. The anteroventral margin of the premaxilla is slightly folded ventrally, and crenate. Ventrally, the right premaxilla presents a raised area pierced by a pore that is directed posteroventrally. The dorsal lamina of the premaxilla (d.l.Pmx, Figure 4A) is well expanded and forms the lateral margin of the opening of the anterior tube of the rostral organ (a.ros, Figure 4A).
The median rostral (ros.m, Figure 4A, 4B), usually poorly preserved in coelacanths, is here complete and in its natural position. It is cross-shaped, shorter than broad, and shows no indication of strong distortion. It is raised anteroposteriorly, and its surface is ornamented with coarse rugosities. On both side, it shows a curved, posteroventral extension towards the dorsal lamina of the premaxilla. The better-preserved right side of the snout allows the description of the relations between these bones ( Figure 4A). The median rostral reaches the lateral tectal medial to the premaxilla, and forms the dorsal margin of the opening of the anterior tube of the rostral organ. Posteriorly, the median rostral caps the anterior wall of a large and deep ovoid cavity (n.c, Figure 4B). The lateral wall, formed anteriorly by the lateral extension of the median rostral, is then extended by the lateral rostral that curves medially in its posterior portion. The ovoid cavity is paired, but not separated from its antimere by a medial septum. Indeed, it is clearly individualized on either side of the snout by a medial, bell-shaped cavity ( Figure 4B). It is unlikely that this paired cavity corresponds to the rostral organ cavity of Latimeria, because the latter is median, and usually situated more posteriorly along the body axis, posterodorsally to the nasal capsule. Consequently, the anterior portion of the rostral organ lies dorsal to the olfactory capsules and the internasal septum. The very anterior position, the individualization of the lateral cavities, and their openings towards the exterior, suggests that they housed the nasal capsules. As in Latimeria, the cavities are oriented anteromedially. The right nasal cavity is opened by a canal directed laterally, that pierces the lateral rostral (c.nos.a, Figure 4B) and opens ventrally to the notch formed by the contact of the dorsal lamina of the premaxilla and the lateral rostral. This opening is interpreted as being the anterior nostril (nos.a, Figure 4A). Another foramen pierces the anterior wall of the nasal capsule towards the opening of the anterior tube of the rostral organ (f.a.n.c, Figure 4A, 4B). The medial cavity that separates the nasal capsules is partly closed ventrally by the flattened base of the premaxillaries, and opens anteriorly through the median pore of the sensory line canal.
The postparietal shield of the skull is divided medially into two strongly flattened halves (Figures 5,6,7,8). Although it is slightly broken posteriorly and medially, the postparietal shield is shorter than the parietonasal shield. The anterior tip of the postparietal is thick and narrow, leaving no gap between the two halves when they are put in contact. In its anteriormost portion, the postparietal shield is thus as broad as the parietonasal shield. It is most probable that the postparietal shield was strongly vaulted and broad over most of the otoccipital portion as in Macropoma, Libys, Holophagus [12] and Swenzia [8], and narrow anteriorly. The center of ossification of the postparietal is situated very close to the joint margin ( Figures 6, 8), as in Libys [12]. The anterior thickening of the postparietal bears a concave facet (Figure 8), which probably matched the contour of the parietal descending process. The ventral surface of the anterior thickening of the postparietal exhibits a ridge that is oriented anteroposteriorly and flanked by tiny pores. The dorsal surface of the postparietal (Pp, Figures 5, 7) shows well-marked longitudinal grooves, especially in its posterior portion. Similar grooves are also present in Macropoma, Latimeria [12] and Swenzia [8], and were interpreted as the anterior branches of the supratemporal sensory line canal commissure. However, the same interpretation is difficult to make because of the poor preservation of the surface of the skull roof. The anterior portion of the postparietal presents a semi-circular depression (Pp.a.d, Figure 7) as in Swenzia [8]. The suture between the postparietals and supratemporals runs anteroventrally on both sides (Figures 5,7). It extends far ventrolaterally through the third opening of the sensory line canal and terminates at the level of its ventral margin: such a condition is also observable in Libys (figure 3.17 in [12] separation between their dorsal edge and the parietonasal shield. This could be attributed to the lateral compression, or it could suggest that most of the pillars may have been sutured to the parietonasal shield. We support the latter assumption based on the presence of clear suture between the anteriormost pillar and the lateral rostral on both side of the skull. Consequently, we identify the pillars as expansions of the supraorbital series. The condition observed in Megalocoelacanthus is thus very similar to that of the Jurassic genus Libys, where the pillars forming the large vacuity of the supraorbital sensory line canal are interpreted as elements of the supraorbital series (Dutel pers. obs. on BMNH P.3337). A twist in the orientation of the pillars is observable along the anteroposterior axis (Figures 2, 3). The first three pillars are oriented anterodorsally whereas the more posterior ones are oriented posterodorsally. The supratemporal bears one pillar, the posterior parietal four, the anterior parietal two, and the anteriormost ones are borne by the nasals. A vacuity occurs between the edges of adjacent bones of the parietonasal series (i.e. posterior parietal/ anterior parietal; anterior parietal/nasal). The suture between the bones of the parietonasal series extends through the vacuity and is subsequently overlapped by the ventral edge of the cavity ( Figure 2). This suggests that the parietonasal series could have   extended laterally on both sides of the skull, so that it was overlapped by the supraorbital series. The first three vacuities are more dorsal in position than the posterior one. Their shapes are also different along the antero-posterior axis ( Figure 2): the anteriormost vacuities are dorsoventrally flattened while the posteriormost are anteroposteriorly flattened. This suggests that on either side, the branches of the sensory line canal were oriented anteromedially along the ethmosphenoid portion of the skull, meeting dorsomedially behind the lateral rostral to form the antorbital sensory line canal commissure (ant.com.so.s.c, Figure 2). The position of the antorbital commissure is thus at the same level along the shield as in Latimeria [19] and Macropoma (figure 3.18 in [12]).
The course of the sensory line canal along the anteriormost portion of the skull is more difficult to reconstruct. The lateral rostral (L.r, Figures 2,3) shows no pores on its dorsal surface, as in Diplurus, Macropoma, Latimeria, Laugia [12], Swenzia [8] and Rhabdoderma [17]. Consequently, the course of the infraorbital sensory line canal cannot be followed throughout the surface of the bone. Megalocoelacanthus presents, as in Macropoma, Rhabdoderma, and Whiteia, a small median opening between the premaxillae that was interpreted as the median pore of the sensory line canal (m.p.s.c, Figure 4A). Forey [12] suggested that this opening was related to the ethmoid commissure. Following the interpretation of Forey [12], the presence of such a pore in Megalocoelacanthus could suggest the presence of an ethmoid commissure running beneath the bones, as in Latimeria [19] and Macropoma [12], and emitting canals that are directed towards the tip of the snout as in Latimeria.
In the otoccipital portion, the sensory line canal passes within the postparietal and the supratemporal (Figures 5, 7). The vacuities present on the otoccipital portion of the skull are larger than those of the ethmosphenoid portion and nearly square. The posteriormost vacuity of the otoccipital portion opens throughout the supratemporal. It is triangular and that of the left side is pierced by two foramina that are aligned along the anteroposterior axis ( Figure 7). A slight swelling on the dorsal surface suggests that the otic sensory line canal, lateral sensory line canal and supratemporal commissure (Stt.com, Figures 5, 7) met near the posterior edge of the supratemporal, as in Libys. A few pits lie on the dorsal surface of the postparietal (ot.s.c.m, Figures 5, 7), close to the joint margin. As in Holophagus (figure 3.18 in [12]), Libys (figure 3.17 in [12]), Macropoma (figure 3.21 in [12]) these pits could The basiphenoid is preserved in connection with the whole ethmosphenoid portion of the skull in AMNH FF 20267 (Figures 2, 3), and as an isolated element in the holotype specimen CCK 88-2-1 ( Figure 9). In AMNH FF 20267, the basisphenoid is relatively large compared to the entire ethmosphenoid portion of the skull. The paired processus connectens (pr.con, Figures 2,3,9) are robust, well developed, and posteriorly elongated. The angle between the parasphenoid and the processus connectens is smaller than what can be observed in Latimeria and Macropoma (Figures 2,  3). The surface of the processus connectens is rugose and was probably capped by cartilage to articulate with the otic shelf of the prootic. The area between the processus connectens and the antotic process is marked by a deep groove (gr.j.v, Figures   the bones. Although both basisphenoids are flattened laterally, the one of the holotype specimen CCCK 88-2-1 ( Figure 9) is less so and enables further comparisons with other coelacanths. The basisphenoid of Megalocoelacanthus is higher and narrower than that of Axelrodichthys, Mawsonia and Diplurus [5,20]. The dorsum sella (d.s, Figure 9) is much elongated in Megalocoelacanthus compared with these genera. As in other coelacanths [12], the basisphenoid was certainly pierced medially by the buccohypophyseal canal (?pit.fos, Figure 9C). However, it fails to pierce the ventral surface of the parasphenoid in AMNH FF 20267 ( Figure 11B), contrary to what is observed in basal coelacanths such as Diplocercides, Euporosteus [12], Miguashaia [21], and Styloichthys [22,23]. This condition observed in Megalocoelacanthus seems to be derived among coelacanths and also observed in Axerodichthys (AMNH 14026L), Latimeria (MNHN C24), Macropoma (figure 6.10 in [12]), Undina (BSPG 1870 XIV 508). In posterior and dorsal views, the basisphenoid of the holotype specimen CCK 88-2-1 presents an overlapping surface for the descending process of the parietal (o.v.pr.Pa, Figure 9B). The sphenoid condyles (sph.c, Figure 9A, 9C, 9D) are well separated by the concave posterior margin of the basisphenoid. The condition observed in Megalocoelacanthus is very similar to that of Latimeria [18] and Macropoma (figure 6.12B, D in [12]), rather than to that of Mawsonia or Axelrodichthys where the sphenoid condyles are very close medially and separated by a deep notch (figures 1A, 18A in [5]).
Two halves of the otoccipital portion are preserved, but extremely flattened (Figures 5,6,7,8). The prootic (Pro, Figures 5,6,7,8) is short, ventrally oriented, and shows a very short and narrow otic shelf whose inner surface is covered by tiny rugosities (ot.sh, Figures 6,8). It is enlarged posteriorly and its posterior end is broken on both sides. The inclination of the otic shelf is less prominent than in Macropoma, but this may be due to deformations of the specimen.
The prootic presents two roughened areas. The anterior one (the so-called prefacial eminence) is sutured on the inner side with the postparietal descending process (v.pr.Pp, Figures 6, 8), and the posterior one is sutured with the supratemporal descending process (v.pr.Stt, Figures 5,6,7). This latter suture can be observed only on the lateral part of the right side ( Figure 5). Between these two areas, temporal excavation is marked by a slight concavity on both sides ( Figures 5, 7), and seems to be lined with bones like in Axelrodichthys and Mawsonia [4]. The condition observed in Megalocoelacanthus is thus different from that of Latimeria and Macropoma, where the temporal excavation is cartilaginous. In the inner part of both sides ( Figures 6, 8), the suture between the descending process of the postparietal and the prootic runs anteroventrally from the dorsal edge of the otic shelf to the anterior margin of the otoccipital portion. The path of this suture is unclear on the lateral part of both sides because it is completely overlapped by the postparietal shield on the right side, and only the anteriormost portion of the suture can be observed on the left. However, it seems that the suture is directed posterodorsally on the lateral side. The contact between the prootic and the descending process of the postparietal ( Figure 5) appears to be much more ventral that what can be observed in other coelacanths, but this is probably an artefact due to the strong deformation of this portion of the skull.
The course of the arteries and nerves through the prootic is quite variable among coelacanths [12]. In Macropoma, the lateral surface of the prootic is pierced by several foramina that are lacking in Latimeria: the palatine nerve emerges ventrally to the prefacial eminence, and the orbital artery opens ventral to the contact with the supratemporal descending process. The identification of such foramina is very difficult in Megalocoelacanthus. Most probably, evidence for a foramen could be found on the lateral face of the right moiety of the prootic (?f.o.a, Figure 5). Like in Macropoma and Mawsonia a foramen opens ventrally to the roughened area contacting the supratemporal descending process, which allows us to suggest that it could correspond to the foramen for the orbital artery.
Posterior to the otic shelf, the lateral wall extending posteroventrally is broken on both sides of the skull. The saccular chamber is completely flattened between the postparietal shield and the medial wall that usually separates it from the notochordal canal.
In coelacanths, the base of the otoccipital portion of the neurocranium is poorly ossified and composed of several elements that embed the notochord ventrally. The basioccipital alone is sutured to the posterior wing of the prootic, whereas the anazygal, and the anterior and posterior catazygals occupy the basicranial fenestra and lie free from the rest of the neurocranium. In Latimeria, these elements are firmly attached to the neurocranium by mean of strong ligaments (Dutel pers. obs. on MNHN C24). The anterior and the posterior catazygals are preserved in AMNH FF 20267 ( Figure 10). The anterior catazygal ( Figure 10A) is the largest one, semi-circular in shape, wider than long, with concave lateral margins. The posterior catazygal ( Figure 10B) is much smaller, longer than wide, and bell-shaped in dorsal/ventral views. Its anterior margin is straight whereas the posterior is rounded and narrower. Although these elements are rarely preserved in fossil coelacanths, they are well known in Mawsonia [2,3]. Anterior catazygals referred to this genus are butterfly-shaped whereas the posterior catazygal is semi-lunar in shape [2]. Catazygals are unossified in Axelrodichthys [2]. The catazygals described in Megalocoelacanthus are very different to that of Mawsonia, but rather resemble those of Holophagus (figure 6.9 in [12]) Macropoma (figure 6.10 in [12]) and Latimeria [18].
1.4 Palate. The parasphenoid (Par, Figures 2, 3, 11) is very narrow and deep. As in Latimeria, its anterior half is marked by low and elongated lateral wings (a.w.Par, Figures 2, 3), which are connected to the ventral margin of the lateral ethmoid. In ventral view (Figure 11), the parasphenoid is strongly compressed laterally along its posterior half, and expands on both sides at the level of the anterior half. The anterior portion is ovoid in shape, ventrally concave and covered by tiny villiform teeth (t.Par, Figure 11). In lateral views (Figures 2, 3), the posterior half of the parasphenoid rises steeply, so that its posteriormost dorsal surface reaches the anterior margin of the processus connectens. This condition is unknown in Macropoma, Latimeria (figures 6.1, 6.11 in [12]) and other Mesozoic coelacanths such as Axelrodichthys (Maisey 1986), but is present in Rhabdoderma (figure 6.5 in [12]). The parasphenoid ends posteriorly abruptly below the anterior side of the processus connectens, with a slightly curved ventral margin.
The palatoquadrate ( Figure 11A) consists of the pterygoid anteriorly, the metapterygoid posterodorsally, and the quadrate posteroventrally. The palatoquadrate is triangular in shape, short and very deep. Its general shape is thus proportionally similar to that of Latimeria, Macropoma, and Holophagus [12], whereas Mawsonia and Axelrodichthys possess a longer and shallower palatoquadrate [5]. The anteriormost part of both pterygoids (Pt, Figure 11A) of AMNH FF 20267 as well as that of the holotype specimen CCK 88-2-1 (figure 2F in [1]) is thin and covered with striations. In these specimens as well as in AUMP 3834, and FMNH P27524 (Schwimmer pers. obs.) the anterior termination is thus similar, suggesting that it was poorly ossified or capped by a cartilage layer that was connecting it to the autopalatines which are preserved separately in AMNH FF 20267 ( Figure 11C). The pterygoid (Pt, Figure 11A) is triangular, shallow and short, and presents a ventral swelling (v.sw.Pt, Figure 11A) anterior to the quadrate that is more pronounced than that of Macropoma or Latimeria. The pterygoid forms a narrow and straight edge along the anterior margin of the metapterygoid. On the medial side of the palatoquadrate, the pterygoid also overlaps the metapterygoid and forms its posterior edge. The medial surface of the pterygoid is ornamented with tubercular shagreen.
The metapterygoid (Mpt, Figure 11A) is short but very large compared to that in other coelacanths. As in Latimeria, it is saddleshaped and its dorsal surface was articulating with the antotic process (art.ant.pr, Figure 11A). When considering the entire palatoquadrate, its size is proportionally quite small. Contrary to what can be observed in Mawsonia or Axelrodichthys, the metapterygoid displays a marked ventral recess ventral to the dorsal edge of the pterygoid on the medial side.
The quadrate (Q, Figure 11A) is straight vertically and extends dorsally up to the level of the ventral half of the pterygoid. It finishes dorsally as an open-end prolonged by an anterodorsally curved ridge extending onto the pterygoid and metapterygoid. This suggests the presence of a posterior cartilage joining the dorsal end of the quadrate to the metapterygoid. The double condyle of the quadrate is large and robust, as in Latimeria and Mawsonia.
1.5 Lower jaw, coronoids and gular plates. Both lower jaws are well preserved (Figures 12, 13). The lower jaw is long and shallow throughout, and resembles that of Undina, Holophagus, Latimeria and Macropoma. The dentary (De, Figures 12A, 13A) is long and narrow, and its proportion relative to the total jaw length is close to that seen in Undina and Holophagus, where it reaches about 40% of the total jaw length [12]. The dentary overlaps the angular and possesses a hook-shaped process extending posterodorsally. The distal part of this hook-shaped process is broken, suggesting that it may have been more prominent and elongated, comparable to that of Libys, Undina, Holophagus and Macropoma. The splenial is narrow, and slender. It forms the ventral edge and ventrolateral side of the anterior portion of the jaw, and extends posteriorly to the hook-shaped process of the dentary (Figures 13A). It presents no traces of ornamentation on its surface. Ventrally, the splenial (Spl, Figures 12, 13) forms a hump that winds around the ventral edge of the mentomeckelian (Mm, Figures 12B, 13B). The mentomeckelian forms the anteriormost part of medial side of the jaw (Figures 12B, 13B). It is rectangular in shape, with a posterior finger-like expansion ventrally to the prearticular (Part, Figure 12B). It is inwardly curved, so that its anteriormost portion overlaps the splenial in lateral view ( Figures 12A, 13A), to form the mandibular symphysis.
The angular (Ang, Figures 12A, 13A) is triangular in shape and slightly concave ventrally. It is deeper than in Holophagus, Latimeria, Macropoma, Undina, and Swenzia. Just behind its contact with the dentary, it shows a prominent dorsal, anteriorly bent extension that forms the coronoid process (co.pr.Ang, Figures 12, 13). The dorsal margin of the angular is concave posteriorly, and then runs straight anteriorly up to the blunt process. This condition is very different from that of the latimeriids Latimeria, Macropoma, and Swenzia where the dorsal edge of the angular is convex and regular throughout, and of the mawsoniids Mawsonia and Axelrodichthys where it forms a deep hump. The center of ossification of the angular can be observed at the level of its deepest portion. A long The coronoid series is poorly preserved and few elements of the series are only observed on the medial side of the right lower jaw (Co, Figure 12B). The coronoid posterior to the hook-shaped process of the dentary is elongated and closely associated with the dentary. The surface of the coronoid is covered with shagreen tubercles, and does not bear any enlarged teeth. Anteriorly to the hook-shaped process of the dentary, only a smaller coronoid is preserved. However, the rugous dorsal surface of the dentary suggests that the coronoid series were extending anteriorly up to the level of the mentomeckelian. The left principal coronoid The Giant Cretaceous Coelacanth Megalocoelacanthus ( Figure 14A) is preserved as an isolated element. Its general shape is rounded with a broad base that extends posteriorly as a fingerlike process. The dorsal portion of the principal coronoid is curved and much longer than in Macropoma and Libys where it is developed as a narrow dorsal process [12]. Deep furrows mark the entire length of its ventrolateral edge. Digitations are present on the posterodorsal edge, from which long grooves extend radially. The medial surface of this coronoid is covered by tiny, villiform teeth that form a shagreen area.
The prearticular is preserved only on the right jaw (Part, Figure 12B). It consists in a slender, long and shallow bone that covers most of the medial side of the lower jaw, and like the principal coronoid, is covered with tiny villiform teeth. The articulars ( Figure 14B) are preserved as isolated elements and consist of a single ossification. They possess two concave articulatory facets, one dorsal and one ventral that are anteromedially inclined. The retroarticular (Rart, Figures 12B, 13B) possesses a double articulatory facet that faces the articular. Posteriorly, the retroarticular bears a small facet for the articulation of the symplectic. As in other coelacanths, the posterior portion of the retroarticular may have been capped with cartilage. A ridge interpreted as the insertion point for the articular-hyomandibular ligament is observable on the lateral surface of the retroarticular (i.art-hy.lig, Figure 12). Longitudinal ridges are present on the medial surface of the retroarticular.
The course of the main sensory line canal in Megalocoelacanthus seems to be very similar to that observed in Latimeria and Macropoma. The mandibular sensory line canal is a large canal that opens ventrally through large pores on the surface of the angular and splenial (m.s.c, Figures 12A, 13A). The pores are oriented ventrally on both bones. Four pores are clearly observed on the splenial and five on the angular. The anteriormost pores open on the angular are oriented anteroventrally whereas the posterior ones are oriented posteroventrally. Posteriorly, a marked notch is present at the margin of the angular, lateral and posterior to the articular glenoid. A foramen can be clearly observed in this area on the right mandible of AMNH FF 20267 (f.pop.s.c, Figure 13B) as well as on holotype CCK88-2-1, suggesting that the main mandibular sensory line canal runs down from the preopercular and enters the angular at this level, as in Latimeria [18,19,24,25] and Macropoma [12]. Anteriorly, a long and slender foramen lies along the suture between the splenial and the dentary at the level of the hook-shaped process of the dentary (dp, Figures 12A, 13A). Such a foramen is also present in the same position in Macropoma, Holophagus and Undina [12]. This foramen could correspond in Latimeria to the enlarged pore of the sensory line canal that pierces the splenial just beneath the suture with the dentary. This pore is connected to the main mandibular canal that runs through the angular. In Macropoma and Holophagus, the subopercular branch of the preopercular canal exits posteriorly on the angular immediately beneath the foramen for the external ramus of the mandibular branch of the facial nerve [12]. Such a foramen has not been observed in this area in Megalocoelacanthus. If the subopercular branch of the preopercular is present in Megalocoelacanthus, the posterior orientation of the posteriormost pore in the angular would mark its point of exit (?sop.br, Figures 12A, 13A). This condition would thus be similar to that in Latimeria [18].
Anterior to the enlarged pore of the sensory line canal, another foramen pierces the dentary just above the suture with the splenial  Figures 12A, 13A). As in Latimeria, this foramen may correspond to the exit of the mandibular branch of the trigeminal nerve for innervating the skin of the lower lip. The relative position of the enlarged pore of the sensory line canal and the foramen for the mandibular branch of the trigeminal nerve differs from that of Latimeria. In this genus the enlarged pore of the sensory line canal is ventral to the foramen for the mandibular branch of the trigeminal nerve, whereas both are aligned antero-posteriorly along the dentary-splenial suture in Megalocoelacanthus.
A foramen for the external mandibular ramus of the facial nerve (f.VII.m.ext, Figures 12A, 13A) is present on the lateral surface of the jaw between the retroarticular and the angular. In Latimeria, the internal ramus of the mandibular branch of the facial nerve penetrates the mandible by a foramen situated between the prearticular and the retroarticular. In Megalocoelacanthus a ventral groove can be observed on the anterior margin of the retroarticular, at the level of the articulation facet ( Figures 12B, 13B). The absence of the prearticular on the left lower jaw shows the course of this groove within the mandible (gr.VII.m.int, Figure 13A). Forey [12] also observed such a groove in Macropoma and interpreted it as the mark of the path of the internal mandibular ramus of the trigeminal nerve within the mandible.
Both gular plates are preserved with little deformation (Figures. 14C). Their shape is very similar to those of Latimeria, and their surface is slightly concave dorsally. Their anterolateral edge is slightly swollen and the anterior tips diverge strongly along the posterior midline. The lateral edge of the gular plates is curved in its posterior half, so that the posterior tips meet medially. On the internal surface of the gular plate, a ridge runs along the anteroposterior axis parallel to the lateral edge (i.rd.intermd, Figures. 14C). In Latimeria, this ridge corresponds to the insertion point of the anterior and posterior ramus of intermandibular muscle (Dutel pers. obs. on MNHN C24). Neither their internal nor their dorsal surfaces are ornamented and the gular pit-lines are not observed.
Cheek bones are poorly preserved in AMNH FF 20267. An isolated element presenting a sensory line canal crossed by pillars could be interpreted as a fragment from the lachrymojugal or the postorbital. This suggests that the sensory line canal was opening through cheek bones by large vacuities as on the skull roof. Consequently, the condition in Megalocoelacanthus would have been identical to that of Libys were the large sensory line canal is opening through a large continuous groove crossed by pillars on the lachrymojugal and postorbital (BSPG 1860 XIV 502). Like in this genus and other latimerioids [12], it is probable that cheek bones were well separated from each other in Megalocoelacanthus, explaining the poor preservation of this complex.
Both operculars are preserved ( Figure 15D). They are partly broken posteriorly and ventrally, but seem to be deeper than long as in Latimeria, Holophagus, and Macropoma. Their center of ossification is situated anterodorsally, and both their lateral and medial surfaces are ornamented by isolated tubercles. The anterodorsal edge of the operculars is notched and similar in shape to that of Macropoma and Holophagus [12].
1.7 Hyoid arch. The hyoid arch of coelacanths consists of hyomandibular, interhyal, ceratohyal, hypohyal and symplectic. Only one ceratohyal and the left symplectic are preserved here.
The left symplectic ( Figure 15B) is entirely preserved but strongly compressed laterally. Its shape is typical, with an upper part enlarged posteriorly. As in other coelacanths, both ends were probably cartilaginous and its actual length may have been longer.
The ceratohyal ( Figure 15C) has a typical shape for coelacanth. It is laterally compressed, curved posteriorly, and its posteriorly directed elbow-like expansion is hook-shaped like in Axelrodichthys (AMNH 13962 R), Latimeria [18], Macropoma [12] and Mawsonia (AMNH 11758). The ventral expansion is slender and slightly convex, and its posterior portion bears a groove that may have been capped with cartilage, as well as the anterior margin of the bone that finishes in a dead-end. In AMNH FF 20267 as well as in the holotype specimen CCK 88-2-1 (figure 2L in [1]), the ceratohyal is much more curved, and its ventral expansion appears to be shorter than that of Axelrodichthys (FMNH FM11856), Latimeria [18], and Mawsonia (AMNH 11758). The dorsal end and the elbow-like expansion are broken, but they are usually also capped by cartilage.
1.8 Branchial arches, urohyal. The branchial skeleton is poorly known in fossil coelacanths. The branchial skeleton of Megalocoelacanthus consists of a complete urohyal, and a basibranchial associated with tooth plates and ceratobranchials.
The basibranchial (Bb, Figure 16A) is large and rounded in shape. It is fused with three tooth plates (p.t.p.Bb, a.t.p.Bb, Figure 16A). As in Macropoma [12], the basibranchial consists in a central embedded ossification surrounded by a perimeter of cartilage. Its ventral surface is marked by a posterior median pit for the articulation of the urohyal (art.Uhy, Figure 16A), and by paired lateral pits that were probably articulated with the first two ceratobranchials as in Latimeria and Macropoma (art.Cb1, art.Cb2, Figure 16A). Tooth plates associated with the basibranchial consist in an anterior median diamond-shape plate (a.t.p.Bb, Figure 16A) and a pair of large plates, extending symmetrically posteriorly along the midline (p.t.p.Bb, Figure 16A). The ventral surface of the latter displays a paired concavity that is interpreted for the articulation of the ceratohyal (art.Ch, Figure 16A). Tiny villiform teeth cover all the tooth plates. Although basibranchial tooth plates are rarely found in situ, previous studies support that the trend in basibranchial tooth plate evolution in coelacanths was towards a mid-line fusion of many paired plates into larger tooth plates [12,26]. The pattern observed in Megalocoelacanthus is also observed in Latimeria and Undina where the tooth plates associated with the basibranchial consist of large paired plates posteriorly to a smaller median plate, while Diplurus and Axelrodichthys bear, respectively, three and two pairs of large plates [12].
The urohyal ( Figure 16B) is very well preserved and shows no evidence of strong deformation. The general shape of the urohyal is characteristic of what can be observed in other coelacanths. The bifid posterior end is broad and form two semi-lunar wings. They arise at the level of the first third of the bone, making them more prominent than those of Latimeria and Macropoma. The slit between the posterior wings is straight, narrow and prolonged anteriorly by a broad groove. These conditions are thus different from that observed in Axelrodichthys (FMNH FM11856) in which the lateral wing extends straightly and the slit is V-shaped. The slit is more expended anteriorly than that of Macropoma (figure 7.7 in [12]) and Latimeria (figure 7.6 in [12]). The ventral surface of the urohyal is flat, except on its anterior portion where the edges are slightly raised. The anterior end is also bifid with a hump on each side. The dorsal surface of the posterior wings is concave in its anterior portion and flat in its posterior portion. A long and deep septum covered by small grooves extends medially. Although some notable differences are observed, the general shape of the urohyal of Megalocoelacanthus is much more similar to that of the latimeriids Macropoma and Latimeria, than to that of the mawsoniids Mawsonia and Axelrodichthys.
As usual in fossil coelacanths, little remains of the gill arches. Four ceratobranchials are preserved ( Figure 15A), and two of them are almost complete. However, it is difficult to determine their position in the branchial series. Ceratobranchials are curved and compressed laterally. The ventral extremity is rounded and was probably articulated with the basibranchial. The bone narrows dorsally into a sharp dorsal end. Isolated small, sharp, denticles measuring about 1-2 mm are preserved in the matrix at the edge of the ceratobranchial of the holotype specimen CCK 8-22-1 ( Figure 17). These denticles were recognized as branchial teeth by Schwimmer [27] and we here follow this interpretation. When compared to the size of the ceratobranchial, these denticles are significantly smaller than those of Latimeria and Axelrodichthys (FMNH FM11856).
1.9 Postcranial skeleton and scales. Very few elements of the postcranial skeleton are preserved. All axial skeleton and fins are missing, and only the pectoral girdle and some isolated scales are present.
The shoulder girdle ( Figure 18) is narrow and is represented by the cleithrum, the extracleithrum, the clavicle and the scapulocoracoid. The pectoral girdle of coelacanths usually includes one supplementary dermal bone, the anocleithrum, which is not preserved here.
The cleithrum (Figures 18A, B) is compressed laterally and bent anteriorly. The left cleithrum ( Figure 18A) is the best preserved with the extracleithrum sutured on the lateral side, and only lacks its dorsal and ventral tips. It is elbow-shaped posteriorly and presents a ventral and dorsal half very distinct in shape. The latter is slender, and more developed and straight than in Latimeria [18], whereas the former is broad and rounded posteriorly. Contrary to what can be observed in Mawsonia and Axelrodichthys [12], there is no broad medial extension of the cleithrum, and its general shape is more similar to what can be observed in Latimeria or Macropoma.
The most remarkable feature is the very small size of the clavicle ( Figure 18C) compared to that of the cleithrum. As is typical, the clavicle twists medially from the leading edge of the cleithrum. The distal tip of the clavicle presents marked, irregular digitations, and the contact with its antimere was probably made through cartilage as in Latimeria [18]. Dorsally, the clavicle enlarges as a thin, medially concave layer of bone that was overlapping the lateral side of the cleithrum.
Both scapulocoracoids are preserved, but strongly flattened ( Figures 18D, E). Each consists of a single, ossified element with a broad and flat proximal portion articulated with the cleithrum, and a short distal portion bearing the glenoid surface for the first axial mesomere. Although it is partially crushed, the best preserved scapulocoracoid presents a concave glenoid surface, covered with coarse rugosities, and which may have been capped with cartilage in life ( Figure 18E).
The axial skeleton of Megalocoelacanthus is only known from an isolated and well preserved vertebra found on the holotype specimen CCK 8-2-22 ( Figure 19A). It is very similar to that of other coelacanths: the neural arch is forked and co-ossified with a median neural spine. In coelacanths, the anterior neural spines are short, and gradually increase in height posteriorly. Here, the neural spine is bended posteriorly and relatively short compared to those situated posteriorly to the D1 in other coelacanths. Thus, the vertebra preserved here was most probably situated anteriorly along the body axis.
The best-preserved body scale of AMNH FF 20267 is subcircular, and is about 5 cm in diameter ( Figure 19B). Only its overlapped portion is preserved. As in other coelacanths, the exposed portion seems to represent here less than one third of the total surface of the scale. The scale shows concentric ridges and presents no evidence of pore canal system or lateral line system.

Phylogenetic analysis
2.1 Revised matrix with new characters and taxa. The recent phylogenetic analyses investigating the relationships of coelacanths were made by Clément [8], Friedman & Coates [13], Yabumoto [14], Geng et al. [9], and Wendruff & Wilson [28] and are based on the data matrix of Forey [12] with several corrections and additions.  corrected scoring for character 31 ''preopercular absent (0), present (1)'' in the original Forey's matrix. Yabumoto (2008) subsequently changed Clément's scoring for Mawsonia to ''1''. According to Friedman & Coates [13], the state of character 54 ''dentary teeth fused to the dentary (0)'' cannot be assessed for Allenypterus because its mandible is edentulous, and has to be scored as question mark. We here follow these scorings. Character 52 ''sclerotic ossicles absent (0), present (1)'' was scored as ''1'' for Libys by Forey [12]. This character is here scored as ''0'' because the respective holotypes of L. superbus and L. polypterus do not present sclerotic ossicles [29]. Forey [12] and other authors [5,17,21,29] considered the visceral calcified structure present in fossil coelacanths as a calcified swim bladder. However, recent histological studies on the structure of the calcified organ in different specimens of Axelrodichthys and comparison with the fatty organ of the extant coelacanth Latimeria, suggest that this is instead an ossified bladder with a respiratory function rather than a buoyancy function [30]. Considering this interpretation we modified character 107 ''swim bladder not ossified (0), swim bladder ossified (1)'' into ''ossified bladder absent (0), present (1)''. We also reviewed the coding of this character for several taxa: it was coded ''0'' for Polyosteorhynchus, Allenypterus, and ''?'' for Mawsonia [12]. We here code this character ''1'' for these  [12]) and we do not consider it can be coded under the same state of character. Consequently, we integrated an additional state to character 23: ''supraorbital sensory line canal opening through bones as a single large pore (0), bifurcating pores (1), many tiny pores (2), a large and continuous groove crossed by pillars (3)''. Character 50 ''Infraorbital, jugal and preopercular sensory canals opening through many tiny pores (0), opening through a few large pores (1)'' was coded ''1'' for Libys. However, in this genus (BSPG 1870 XIV 502) the infraorbital and preopercular sensory line canal displays the same condition as the supraorbital sensory canal, i.e. a large and continuous groove crossed by pillars. Based on an isolated element of the lachrymojugal, the same condition was probably present in Megalocoelacanthus. We thus propose an additional state for character 50 ''Infraorbital, jugal and preopercular sensory canals opening through many tiny pores (0), opening through a few large pores (1), a large, continuous groove stretched by pillars (2)''.
2.2 Searching methods. The data matrix (Information S2) was constructed in Mesquite 2.74 [33]. It comprises 39 taxa and 110 anatomical characters including 88 cranial and 22 postcranial anatomical characters (Information S1). Maximum parsimony analyses were carried out using the software PAUP 4.0b10 [34]. A heuristic search was performed using the tree-bisection-reconnection branch swapping algorithm (TBR) with 10,000 random addition-sequence replicates. We ran the analysis with all characters unweighted and multistate characters unordered. Branches with a maximum length of zero were collapsed, so that any branch supported by ambiguous synapomorphies is conserved. Bootstraps values were calculated with this program using heuristic searches and 1,000 bootstrap replicates, with 100 random sequence additions per replicate. Bremer decay indices were calculated by combining PAUP 4.0b10 [34] and TreeRot v.3 [35]. Optimizations of ambiguous states of characters were performed using the software WINCLADA 1.00.08 [36]. The tree is rooted by two outgroups, porolepiforms and actinopterygians.
Two analyses were carried out. The first analysis (Information S3) was run with all the taxa of the data matrix. The strict consensus tree (Figure 21) of the 584 equally parsimonious trees (length = 288; consistency index = 0.4132; retention index = 0.6938) showed two areas of conflict within clade 2 and clade 3. The irresolution of the phylogenetic relationships is due to the instability of Indocoelacanthus (clade 2), and Lualabaea (clade 3). The instability of these taxa has already been noted by Forey [12] and attributed to the high percentage of missing values (Indocoelacanthus, 80%; Lualabaea, 99%) due to incompleteness of the fossil material. A second analysis was then conducted without these two taxa (Information S4). The strict consensus tree is shown Figure 22, and it will be used for presenting the following results. The diagnostic information for the nodes and terminal taxa are presented in Information S5. Three unambiguous synapomorphies support the node [Megalocoelacanthus + Libys]: a supraorbital sensory line canal opening through a large and continuous groove (23 [3], that is a nonhomoplastic synapomorphy), the infraorbital, jugal, and preopercular sensory line canals opening through a large and continuous groove crossed by pillars (50 [2]), and a robust prearticular and principal coronoid, marked with fine striations (68 [1]). Additionally, clade 16 is supported by six ambiguous synapomorphies: the presence of snout bones consolidated (2 [1]), the presence of a preoperculum developed as a posterior tube-like canal-bearing portion and an anterior blade-like portion (39 [1]), the absence of ornamentation upon cheek bones (49[0]), the optic foramen lying , or the sister-group of Garnbergia. The last two hypotheses are corroborated in our topology, suggesting that Libys is more closely related to Latimeria than to Mawsonia. However, these last two topologies suggested that Latimeria is more closely related to Libys than it is to both Holophagus and Undina. This is inconsistent with our results and those of Geng et al. [9] where Libys is the sister group of a clade including Holophagus, Undina, Macropoma, Swenzia and Latimeria.
Our analysis provides new insights in the interrelationships of Latimerioidei. It provides new information on the unsolved relationships between taxa that are well informed, and on the affinities of taxa that were traditionally considered as problematic, and excluded from the analysis because of the presence of many missing data.
The position of Garnbergia was unstable in previous analyses: it was alternatively placed within Mawsoniidae [8,12], as the sistergroup of Libys within Latimeriidae [8], or as the sister group of the least inclusive clade containing Latimeria, Mawsonia and Coelacanthus [10]. In contrast, sister-group relationship between Garnbergia and Latimerioidei is suggested by the present analysis. This hypothesis is supported by a single unambiguous synapomorphy: the presence of 8 to 9 fin rays on the first dorsal fin (96 [1]). This synapomorphy is homoplastic and also represents an apomorphy of Laugia. Within clade 13, this character is well documented and in all taxa where it is known (except Mawsonia, Megalocoelacanthus and Swenzia) possess 8 to 9 fin rays on the first dorsal fin. Nevertheless, this character should be taken with caution: fine rays can be easily lost after the death of the animal, and the number of fine rays is variable between individuals in the extant coelacanth Latimeria.
Clade 13 is supported by a single unambiguous synapomorphy: the presence of denticles on the fin rays of the D1 (98 [1]). This clade was recognized as Latimerioidei in previous studies. The presence of a postparietal descending process (13 [1]) that was previously considered as a synapomorphy of this clade [9,12] is here an ambiguous synapomorphy of clade 13. The ambiguity comes from the lack of information for this character in Garnbergia and Rebellatrix the most closely related taxa of latimerioids.  [14]. Additional synapomorphies included the loss of the suboperculum (32[0]) and of the subopercular branch of the mandibular sensory line canal (60[0]). However, these reversions are actually symplesiomorphies that were misinterpreted because of the lack of resolution in other latimerioids relationships. Indeed, other latimerioids that have a suboperculum (Holophagus, Libys, and Latimeria) and a subopercular branch of the mandibular sensory line canal (Holophagus, Latimeria, Libys, Macropoma and here Megalocoelacanthus) formed a polytomy with Mawsoniidae in Yabumoto's (2008) results. Consequently, the most parsimonious scenario in the phylogenetic analysis of Yabumoto [14] was favouring a reversion in Mawsoniidae instead of the retention of the plesiomorphic condition in this clade, while Holophagus, Libys, Macropoma, and Latimeria would have acquired convergently the apomorphic condition. Thanks to the better resolution the relationships of these taxa, these putative synapomorphies should be there considered as symplesiomorphies.
Within clade 21, the sister-group relationship between Mawsonia and Axelrodichthys is retained, and strongly supported by six unambiguous synapomorphies, including three non-homoplastic ones: the extrascapulars forming part of the skull roof (16 [1]), the presence of a ventral process on the postorbital (41 [1]), and the principal coronoid sutured to the angular (66[1]). The relation-  ships between Lualabaea, Parnaibaia, Chinlea and the clade [Mawsonia + Axelrodichthys] were unresolved in the strict consensus tree from the first analysis ( Figure 21) due to the instability of Lualabaea ( Figure 21B). The first strict consensus tree proposed by Yabumoto (2008) supported a sister-group relationship between Parnabaia and Chinlea based on the presence of two homoplastic synapomorphies, the presence of anterior and posterior parietals of similar size (8[0]), and the presence of an angle on the anterior end of the lachrymojugal (36 [1]). Otherwise, its second consensus tree resulting from successive weighting procedure was supporting Parnaibaia as the sister-group of [Mawsonia + Axelrodichthys]. Our strict consensus tree is inconsistent with these hypotheses and rather supports the sister-group relationship of Chinlea and [Mawsonia + Axelrodichthys] based on three homoplastic synapomorphies: the presence of coarse rugosities on the parietals and postparietals (27[2]), the presence of coarse rugosity on cheek bones (49 [2]) and rugose scales (106 [1]).
Ticinepomis branches at the base of clade 14 ( Figure 22). Its position changed dramatically compared to Cloutier's hypothesis (clade F in [10]) in which it was the sister-group of a clade including Wimania, Axelia, and Coelacanthus. This group appears here to be polyphyletic. Ticinepomis was subsequently excluded by Forey [12] from the analysis because it was raising instability in the intrarelationships of the sister-group of the clade [Coccoderma + Laugia]. As pointed out by Forey [12], the irresolution resulting from the inclusion of Ticinepomis arose from the contradiction in the distribution of its character states rather than to the lack of data. The position of Ticinepomis is here supported by a single unambiguous synapomorphy: the presence of expanded median fin rays (103 [1]). The same position was found in the strict consensus tree of the first analysis we performed ( Figure 21), but the node was only supported by ambiguous synapomorphies. The presence of expanded median fin rays on the D1 (103 [1]) is only found in Ticinepomis, Libys, and Holophagus. Undina, Macropoma, and Latimeria display unexpanded median fin rays (103[0]) and this condition is a synapomorphy of clade 18 interpreted as a reversal toward the ancestral condition of actinistians.
The snout is generally a poorly preserved region of the skull in fossil actinistians [12]. Among Latimerioidei, it is only observed in Axelrodichthys and Diplurus, which possess a single rostral (3[0]), and in Macropoma, Latimeria and Parnaibaia which possess several median rostrals (3[1]). This condition cannot be assessed for other genera of the clade. The presence of several rostrals (3[1]) is plesiomorphic in actinistians, and lost at most in clade 3, the least inclusive clade containing Hadronector, Lochmocercus and Latimeria. Depending on character optimization, reversion toward the ancestral state occurs independently in Parnaibaia and Latimeria (SLOW optimization), or in Parnaibaia and clade 14 (FAST optimization). The latter hypothesis would support Forey's analysis [12] where the presence of several rostrals is a synapomorphy of Latimeriidae. Then, it would also suggest that Megalocoelacanthus possessed such a condition. Re-examination of Libys specimens, and better-preserved material of Ticinepomis and other latimeriids are needed to better understand the polarity of this character.
Among actinistians, the presence of an anterior branch of the supratemporal commissure (22 [1]) over the postparietal is only observed in Macropoma, Swenzia, and Latimeria. The ambiguity of the polarity of this character arises from the fact that all other genera belonging to clade 14 are scored with a question mark for this character. FAST optimization supposes that the presence of the supratemporal commissure is a synapomorphy of clade 14, whereas SLOW optimization supports it as a synapomorphy of clade 19 [Macropoma [Latimeria + Swenzia]]. This character is unknown in Undina and Ticinepomis [12,37,38]. In Megalocoelacanthus, the posterior portion of the postparietal is covered by grooves directed anteriorly (Figures 5, 7), but the state of preservation of the skull roof does not enable to clearly identify them as anterior branches of the supratemporal commissure. In its sister-group Libys polypterus, the supratemporal commissure does not present anterior branches (figure 3.17 in [12]). The presence of such a feature has been considered in Holophagus gulo based on the presence of longitudinal grooves extending from the extrascapulars (figure 3.18 in [12]). If the presence of anterior branches of the supratemporal commissure is confirmed in Holophagus, our topology would imply that this feature could be a synapomorphy of clade 17.
Characters ( [1]) occurred in clade 20 [Latimeria + Swenzia]. The spiracular is absent in Libys and this condition is inferred for Megalocoelacanthus according to our analysis. There is a subopercular branch of the mandibular sensory line canal (60 [1]) in Megalocoelacanthus, Libys, Holophagus, Macropoma, and Latimeria. This condition is thus a nonhomoplastic synapomorphy of at least clade 15 (SLOW optimization), and is therefore inferred in Undina and Swenzia. Potentially, it could be shared by Ticinepomis (FAST optimization) but the available material is too poorly preserved to assess its condition [38]. The ascending lamina of the parasphenoid is originally absent in actinistians (79[0]). Its presence (79 [1]) is observed in Megalocoelacanthus, Undina, Macropoma, Swenzia, and Latimeria, and is therefore a non-homoplastic synapomorphy of at least clade 15 (SLOW optimization). The presence of the ascending lamina of the parasphenoid is inferred in Ticinepomis under FAST optimization. The ventral swelling of the palatoquadrate (110) is here a newly recognised character. It is only observed in Holophagus, Latimeria, Libys, Macropoma, Megalocoelacanthus, and Undina, and unknown in Ticinepomis and Swenzia. Like previous characters, the presence of this ventrally swelling is either a synapomorphy of   [1]), and the presence of multiple opening for the lateral line on scales (105 [1]). the presence of a hook-shaped dentary (57 [1]) was previously considered as a non-homoplastic synapomorphy of the clade [Whiteia + Latimerioidei] in the constrained analysis of Forey [12]. This is not consistent with our analysis in which this change is convergent for clades 15, 22, and in Whiteia. As noted before, the relationships between Indocoelacanthus, Undina, Holophagus, and the clade [Macropoma [Latimeria + Swenzia]] were unresolved in the strict consensus tree of first analysis ( Figure 21) because of the unstable position of Indocoelacanthus due to its incompleteness. Undina is more closely related to [Macropoma [Latimeria + Swenzia]] than Holophagus based on two unambiguous synapomorphies, the presence of the oral pit line removed from the center of ossification (59 [1]), and the presence of unexpanded median fin rays (103[0]). This is inconsistent with previous hypotheses [9,12,14] that were suggesting a sister-group relationship between Holophagus and Undina. Within clade 15, the sister-group relationship between the clade [Latimeria + Swenzia] and Macropoma is retained in both analyses we performed (Figures 21, 22). This sister-group relationship is supported by three unambiguous synapomorphies: the presence of a preoperculum developed as a posterior tube-like canal-bearing portion and an anterior blade-like portion (39 [1]), the presence of an anterodorsal excavation in the postorbital (40 [1]), that is a feature exclusive to this clade, and the presence of less than eight fin rays on the first dorsal fine (96 [2]). Additionally, this clade is supported by two ambiguous synapomorphies. The presence of snout bones consolidated (2 [1]) is a synapomorphy of clade 19 under FAST optimization that is subsequently lost in Latimeria, or is a convergence in Macropoma and Swenzia under SLOW optimization. The presence of a splenial without ornament (64[0]) is alternatively a synapomorphy of clade 19 (FAST optimization), or a synapomorphy of clade 17 that is lost by convergence in Holophagus and Undina. The sister-taxon relationship between Latimeria and Swenzia is supported by two unambiguous synapomorphies: the absence of pit lines making postparietals (26 [1]), and the presence of a spiracular (30 [1]). These synapomorphies where also supporting this clade in the phylogenetic analysis of Clément [8] and both represent a reversal towards the ancestral condition in actinistians.

Systematic paleontology
Based on the topology proposed by Cloutier [10], Schultze [15] erected the suborder Latimerioidei (node 15 in [10]) comprising the family Mawsoniidae Schultze, 1993 (node I in [10]) and the family Latimeriidae Berg, 1940 (node 16 in [10]). Although diagnoses for each taxa cited above were given by Berg [39] and Forey [12], these taxa remain undefined. The confusion between taxon definition and taxon diagnosis has been discussed by Ghiselin [40] and Rowe [41], who emphasized the difference between these two expressions.
Phylogenetic taxonomy aims to formulate taxonomic definitions based on the phylogenetic relationships and to make a clear distinction between taxonomic diagnosis and taxonomic definition [42][43][44][45][46]. Three classes of definitions have been initially stated in phylogenetic taxonomy: apomorphy-based, stem-based, and nodebased definitions [42,47]. Apomorphy-based definition makes reference to characters, and defines the membership of a taxon as the clade stemming from the ''first ancestor with a particular synapomorphy'' [42,43,45,46]. Apomorphy-based definitions are not used here because they encounter three problems: character ambiguity, variation in characters optimization, and homoplasy [46].
A node-based definition specifies the membership of a taxon by ''the least inclusive clade that contains at least two internal specifiers'', while a stem-based definition specifies the membership of a taxon by ''the most inclusive clade that contains at least one internal specifier'' [46]. Stem-based and node-based definitions thus rely on the use of a definitional component termed as specifiers, i.e species or specimens stated in the phylogenetic definition as a reference point. Both definitional types have at least one internal specifier (anchor within the ingroup defined), but stem-based definition contains optionally an external specifier to define the exclusion group. An optional definitional component termed as ''qualifier'' can also be used to provide conditions on the clade membership. Qualifiers can be species, specimens, or features of species or specimens.
We propose here phylogenetic definitions to the taxa Latimeriidae, Mawsoniidae and Latimerioidei. The genera Latimeria Smith 1938 and Mawsonia Woodward 1907 are given as reference taxa within Latimerioidei because they are well-known and complete, deeply nested, and their position is stable in the successive topologies that have been proposed up to now [8][9][10][11][12][13][14]28]. We thus consider Latimeria chalumnae and Mawsonia gigas reliable enough to preserve the taxonomic content of the taxa we define here.
Our study supports the presence of two major clades of coelacanths (clade 14 and 21) within a more inclusive clade (clade 13) that are retained in the strict consensus trees of both analyses performed (Figures 21, 22). However, their inter-and intrarelationships are still weakly supported. Indeed clade 14 is supported by a single unambiguous synapomorphy in the second analysis (Figure 22), and only by ambiguous synapomorphies in the first analysis we performed (Figure 21), raising a polytomy between Ticinepomis, clade 21 and clade 15 when branches with minimum length of zero are collapsed. Consequently, we decided to give a stem-based definition of Latimeriidae and Mawsoniidae. Moreover, it is worth noting that the family Mawsoniidae as coined by Schultze [15] is a nomen nudum. Actually, the name is invalid in regard with the Article 13.1 of the ICZN [48], because neither diagnostic characters nor definitions are stated in the publication. A diagnosis was subsequently given by Forey [12], but no definition has been stated since the name was coined. Based on previous topology and ours, we propose the following definitions: Latimeriidae Berg, 1940: the most inclusive clade containing Latimeria chalumnae Smith, 1938 but excluding Mawsonia gigas Woodward, 1907.
The dichotomy between Latimeriidae and Mawsoniidae is traditionally recognized in the phylogenies, stating the relationship ''Latimerioidei = Latimeriidae + Mawsoniidae''. In order to handle the taxa Latimerioidei in the case of relocation of species, clades, or changes in the apomorphies, we use a node-stem triplet (NST). The NST is the only phylogenetic definition that preserves the taxonomic content at a dichotomy by combining a node-based definition of a taxon with two stem-based definitions of more inclusive taxa [49]. Criteria of diversity, morphology and tradition are recommended for establishing a NST for a certain taxa [49]. With the NST the traditional equivalence statement ''Latimerioidei = Latimeriidae + Mawsoniidae'' is anchored and will always be stable in the case of a taxon branching basally to the clade 14 or to clade 21. The NTS we propose consists in a node-based definition for Latimerioidei composed of two stem-based taxa, Latimeriidae and Mawsoniidae. Using Latimeria chalumnae and Mawsonia gigas as nested specifiers, the phylogenetic definition of Latimerioidei is proposed as follows: Latimerioidei Schultze, 1993 Megalocoelacanthus was previously included in the family Coelacanthidae Agassiz, 1844 by Schwimmer et al. [1]. However, the family Coelacanthidae as defined by Agassiz [50] should not be longer recognized. It is actually inconsistent with the topology obtained here and in previous studies because it does not represent a monophyletic group [8][9][10][11][12][13][14]28]. Based on the hypothesis of relatedness obtained here and using the new definitions we stated, we thus propose the following taxonomy for Megalocoelacanthus as well as a new diagnosis: OSTEICHTHYES Huxley, 1880 [51]. SARCOPTERYGII Romer, 1955 [52].
ACTINISTIA Cope, 1891 [53]. LATIMERIOIDEI Schultze, 1993 [15]  Diagnosis (revised). Parietonasal shield narrow, and longer than the posparietal shield. Supraorbital sensory line canal opening through a large groove crossed by slender pillars. Ventral descending processes present on the supratemporal and on the parietal. Basisphenoid very deep dorsoventrally. Palatoquadrate deeper than long, with distinct swelling extending from the ventral pterygoid margin immediately anterior to the quadrate. Mandibles relatively elongate posterior to the articular; articular sutured to the angular. Medial surfaces of the prearticular, palatoquadrate, and coronoid covered with tubercular shagreen. No marginal teeth on the mandible. Coronoid large, with subcircular ventral margin. Lateral surface of angular bears large pores of the sensory line canal and very faint longitudinal grooves on its posterior portion. Gular and operculum lack external ornamentation. Operculum subrhomboidal, with sharply angled anteroventral margin. Gular plates diverge strongly along posterior midline.

Discussion and Conclusions
Since its first description, Megalocoelacanthus was previously related to the latimeriids Latimeria and Macropoma based on few meristic data [1]. Our phylogenetic analysis of Megalocoelacanthus supports the sister-group relationship with Libys, an Upper Jurassic genus from Bavaria, Germany. Although it is significantly smaller in size, Libys shares many features with Megalocoelacanthus. Notably, this genus also possesses a supraorbital sensory line canal that opens through a large, continuous groove stretched by pillars on either side of the parietonasal shield. When specimens of Libys are observed under binocular microscope (Dutel pers. obs. on BSPG XIV 501b, BSPG 1870 XIV 502, BMNH P3337), one can distinguish a suture between the base of adjacent pillars. It is thus probable that the same condition is present in Megalocoelacanthus, but because of the poor preservation, the suture between the supraorbitals cannot be seen. Despite the size difference, the lower jaw of Megalocoelacanthus and Libys are virtually identical, i.e. a slender and elongated mandible opened by large pores for the mandibular sensory line canal. These two genera also share the absence of a suture between the parasphenoid and the basisphenoid, which suggests fusion of the two bones, the presence of a narrow and unornamented parietonasal shield, and the palatoquadrate being very deep and short in length with a ventral swelling on the palate. Very few elements from the postcranial skeleton are present on the specimen AMNH FF 20267. However, the pectoral girdle which is preserved is virtually identical in shape and proportion in Megalocoelacanthus and Libys: the cleithrum is slender and elongated, and the clavicle is relatively very small compared to the whole pectoral girdle.
A very interesting aspect of Megalocoelacanthus lies in the shape and the proportion of its skull. In genera such as Axelrodichthys and Mawsonia where the skull is shallow and elongated, the basisphenoid has a short dorsum sellae, with laterally well-expanded wings, as well as a palatoquadrate longer than high [5]. In contrast, the well preserved basisphenoid of the holotype specimen CCK 88-2-1 is deeper and narrower than in these genera. The palatoquadrate of Megalocoelacanthus is deeper than long, and appears to be proportionally shorter in length than those of the latimeriids Latimeria, Macropoma, Holophagus and of the mawsoniids Mawsonia, Axelrodichthys and Parnaibaia. The shoulder girdle also appears to be much deeper and more slender than that of the mawsoniids Diplurus, Mawsonia and Axelrodichthys. Megalocoelacanthus clearly shows features of Latimeriidae, but the proportions of the elements of the skull and shoulder girdle are closer to what can be observed in Libys than in other genera of this clade.
In all the latimerioids examined, the parietals are much wider than the supraorbitals and the parietonasal shield thus represents most of the width of skull roof. Megalocoelacanthus and Libys are also very unique in that their parietonasal shield is considerably narrower compared to other coelacanths, and the skull is mainly roofed by the supraorbitals that are lying alongside the parietonasal shield. However, we cannot determine whether the bony surface within the vacuities of the supraorbital sensory line canal is a lateral extension of the parietonasal shield that is overlapped by the supratemporal. Although the skull of AMNH FF 20267 is strongly flattened laterally, the skull roof was most probably well-vaulted. In any case, it is clear that the top of the skull was much narrower than the buccal floor. When the jaws and the gular plates of AMNH FF 20267 are assembled, the minimum width of the buccal floor is much wider than the skull roof width. Taken together, these elements enable us to depict Megalocoelacanthus as a large-sized coelacanth with a short, laterally compressed and very deep skull, with a bell-like shape in transverse section. This is thus quite different from the more ovoid transverse section of the skull of the extant coelacanth Latimeria chalumnae.
Coelacanths have been depicted as a conservative group that experienced little anatomical change during their evolution. However, this widespread idea has been challenged at several times by the discovery of Paleozoic coelacanths such as Allenypterus [17,54], Holopterygius [13] and Miguashaia [21], whose morphologies differ significantly from that of Latimeria. It now appears that coelacanths experienced a wide range of morphologies and ecologies very early in their evolutionary history. Based on geometric morphometric analysis Friedman & Coates [13] showed that the highest morphological disparity in coelacanths was reached by the Middle Devonian, but dropped in post-Carboniferous forms despite a significant increase in the taxonomic diversity. Indeed, Mesozoic coelacanths actually fit to the ''Latimeria bodyplan'' and anatomical variations in Mesozoic coelacanths seem to lie in variations of proportion of the skeletal elements rather than in radical morphological shifts.
The evolution of Mesozoic coelacanths also appears to now be marked by a significant increase in body size in at least two lineages during the Cretaceous. Nevertheless, the skull morphology of these forms is far from being homogenous and clearly falls into two morphotypes: long, shallow, wide skulls in the mawsoniids Mawsonia and Axelrodichthys and short, deep, narrow skulls in the latimeriids such as in Latimeria, Macropoma, which is amplified in Libys and Megalocoelacanthus. It is worth noting that these variations may focus considerable interest in the intracranial joint kinetic, and thereby in the feeding strategy of these large-sized taxa. Based on a better understanding of the intracranial joint kinetic in Latimeria, this question will deserve further investigation in the future.
The paleoenvironments represented by the occurrences of Megalocoelacanthus differ widely. The marine chalks of western Kansas and western Alabama represent offshore, deep-water marine environments, whereas the detrital sediments from eastern Alabama, Georgia and New Jersey (including the holotype occurrence) represent near shore, shallow marginal marine to estuarine environments. Given the range of occurrences, this suggests that Megalocoelacanthus was eurytopic and favored both marine and brackish waters. The records of Megalocoelacanthus [55] collectively suggest that it was a fairly common fish, but it is not frequently recognized in fossil assemblages and its distribution may still be underestimated. Further study will have to be carried on in order to determine the potential ecological niches that these common large, toothless, coelacanths were occupying in the Western Interior Seaway during the Late Cretaceous.
By comparison with Libys and Latimeria, the length of AMNH FF 20267 is estimated to range between 2.30 m and 3 m. However, a large isolated principal coronoid found near the holotype site in eastern Alabama extrapolates the maximum length of Megalocoelacanthus to ,4.5 m [56]. Similar dimensions were previously known only through the genus Mawsonia from the Lower Cretaceous of Brazil, Morocco and Niger, which largest specimens are estimated to range between 3.5 and 6.3 m [2,57,58], and the poorly known genus Trachymetopon from the lower Toarcian of the Posidonia Shale of Germany [59]. Mawsonia was certainly nonmarine and restricted to continental and estuarine environments [2,60], whereas Trachymetopon was clearly marine. Although the phylogenetic position of Trachymetopon has still to be elucidated, our study suggests that large coelacanths evolved in at least two different lineages during the Mesozoic, in both marine and nonmarine environments.

Supporting Information
Information S1 Character list.