Since their first appearance in the fossil record at the beginning of the Devonian period (late Lockhovian, c. 410 Ma), coelacanths stand out among other sarcopterygians (lobe-finned fishes) by a unique set of features, many of them retained by the iconic extant Latimeria ‌Smith, 1939 (L.chalumnae ‌Smith, 1939 from the Eastern African coast, L. ‌menadoensis ‌Pouyaud, Wirjoatmodjo, Rachmatika, Tjakrawidjaja, Hadiaty & Hadie, 1999, and a possible third species (Kadarusman et al. 2020) from Indonesia). Anatomically modern coelacanths (i.e., forms with an elongated preorbital portion of the skull roof, two pairs of parietals, and a trilobed caudal fin; Zhu et al. 2012; Clement et al. 2024), broadly similar to Latimeria, can be traced back to the Early Devonian (Johanson et al. 2006) and include Palaeozoic taxa such asEuporosteus ‌Jaeckel, 1927,Diplocercides ‌Stensiö, 1922,Rhabdoderma ‌Newberry 1856, and Coelacanthus ‌Agassiz, 1839 (Stensiö 1922, 1937; Moy-Thomas & Westoll 1935; Moy-Thomas 1937; Schaumberg 1978; Forey 1981, 1998; Zhu et al. 2012). This apparent evolutionary conservatism of their anatomy, as displayed in Latimeria is one of the most remarkable traits of coelacanths among osteichthyans (bony fishes). Other Devonian species like Miguashaia ‌bureaui ‌Schultze, 1973 (Schultze 1973; Cloutier 1996), Gavinia ‌syntrips ‌Long, 1999 (Long 1999), and Ngamugawi wirngarri Clement, Cloutier, Lee, King, Vanhaesebroucke, Bradshaw, Dutel, Trinajstic & Long, 2024 (Clement et al. 2024) are considered morphologically primitive coelacanths due to the retention of a generalized sarcopterygian arrangement of the skull roof and a heterocercal caudal fin (when found).

The fossil record reveals that the peak of coelacanth diversity occurred during the Early Triassic, with minor peaks in the Late Devonian, Early Carboniferous and Late Jurassic (Cloutier 1991a, b; Forey 1991, 1998; Schultze 2004; Toriño et al. 2021a; Ferrante et al. 2022; Clement et al. 2024; Ferrante & Cavin 2025). However, it is during the Devonian and Carboniferous that the highest morphological disparity is recorded in the Palaeozoic (e.g., Clement et al. 2024), evidenced by forms such as Allenypterus ‌montanus ‌Melton, 1969 (Lund & Lund 1984, 1985; Forey 1998) and Holopterygius ‌nudus ‌Jessen, 1973 (Friedman & Coates 2006). In contrast, post-Carboniferous taxa retained a more stereotypical coelacanth shape despite a higher taxonomical diversity, with the exceptions of the unusual Triassic Rebellatrix ‌divaricerca ‌Wendruff & Wilson, 2012 (Wendruff & Wilson 2012), Foreyia ‌maxkuhni ‌Cavin, Mennecart, Obrist, Costeur & Furrer, 2017 (Cavin et al. 2017), and Rieppelia heinzfurreri Ferrante & Cavin, 2023 (Ferrante & Cavin 2023). These departures from the “typical” coelacanth body-plan challenge the traditional portrayal of coelacanths as morphologically invariant.

The anatomy of the extant Latimeriachalumnaeis among the best-known in vertebrates (e.g., Millot & Anthony 1958, 1965; Millot et al. 1978; Forey 1998; Amemiya et al. 2013; Cupello et al. 2017a, b; Dutel et al. 2019; Manuelli et al. 2023; Datovo & Johnson 2025). However, some important aspects of the anatomical and physiological evolution of coelacanths are still incompletely understood, especially those dealing with key internal structures of the skull such as the braincase (neurocranium) or the visceral skeleton (splanchnocranium). The visceral skeleton of gnathostomes (jawed vertebrates) is constituted of the mandibular arch (i.e., the jaws), followed by the hyoid arch and the subsequent branchial (or gill) arches, ventrally linked to the median basibranchial series, and as such it plays a major role in breathing and feeding (e.g., Nelson 1969; Janvier 1996; Dutel et al. 2013, 2015a). Nevertheless, despite a relatively rich fossil diversity, our knowledge on the morphology and evolution of the hyoid and branchial arches of coelacanths is extremely limited. This is caused by the poor preservation of many small elements forming the arches due to their weakly-ossified and presumed to be mainly cartilaginous nature. As such, the bones easily disarticulate after death and, when preserved, are found scattered and displaced from their original position. Unfortunately, three-dimensionally preserved Palaeozoic coelacanths are rare since many Devonian and Carboniferous coelacanths are usually preserved as fragmentary, flattened or highly compressed specimens, complicating the access to these important, but delicate internal features of the skull.

Here we describe a new three-dimensionally preserved coelacanth from the Late Carboniferous (Pennsylvanian) of Texas, United States, using micro-computed tomography (μCT). It exquisitely preserves the complete series of the hyoid and branchial arches as well as part of the skull roof, cheek, palate, braincase, and pectoral girdle. Through comparisons with other extinct and extant taxa, we reveal and discuss new evolutionary trends, shedding light on the anatomical organization and evolution of the skull of coelacanths.

The specimen (AMNH FF 20686) is preserved in a phosphatic nodule from dark grey shale near the base of the Finis Shale Member, Graham Formation (Pennsylvanian age, early Virgilian corresponding to the early Gzhelian; Fig. 1). The hillside exposure, which contains a diverse fauna and plant remains, is the Lost Creek Lake emergency spillway (approximately 4.0 km northeast of Jacksboro, Texas and 1.5 km south of Texas Highway 59: American Museum of Natural History loc. 5562; Mapes loc. TXV-200: Jacksboro NE 71/2’ quadrangle). The fossil specimen was recovered in a bed of phosphatic nodules and broken shell debris, which is interpreted as a storm deposit in a marine environment. The dark grey shale surrounding the specimen has yielded numerous other nodules containing a diverse calcified fauna, including vertebrates and cephalopods.

The specimen (AMNH FF 20686; Fig. 2A-D) consists of a partial skull of a coelacanth including elements of the skull roof, cheek, hyoid and branchial arches, shoulder girdle, and a partially preserved braincase. Due to its preservation in a nodule, many osseous elements have not been subject to substantial crushing and are still in close articulation. The specimen measures approximately 5 cm in length and 4 cm at its widest width. It exposes certain dermal bones of the cheek and skull roof, evidenced by their tubercular ornamentation, as well as the basisphenoid and both palatoquadrates. Several elements from the hyobranchial skeleton, like the urohyal and both ceratohyals, are exposed as imprints on the lateral and ventral surfaces.

The specimen (AMNH FF 20686; Fig. 2A-D) was scanned at the American Museum of Natural History (AMNH) in 2013 on a 2010 GE phoenix v|tome|x s240 high-­resolution microfocus computed tomography system (General ­Electric, Fairfield, CT, United States), using a diamond target and 0.5 mm copper filter, at 100 kV, 120 µA, producing 1002 images with a voxel size of 0.01299285 mm. Image stacks were reconstructed using GE phoenix datos|x and FIJI software (Schindelin et al. 2012). Data were segmented manually in Mimics Innovation Suite V.18.0 (http://biomedical.materialise.com/mimics; Materialise, Leuven, ­Belgium). Meshes were exported as.ply files and rendered in Blender V.2.80 (http://www.blender.org; Blender Institute, Amsterdam, the Netherlands). Image stacks and .stl files are available in the Appendices 1 and 2.

The phylogenetic analysis was based on a modified version of the discrete matrix from Clement et al. (2024) composed of 268 characters and 90 species. Three coelacanth taxa were added to the original matrix and coded based on the published descriptions: Rieppellia heinzfurreri Ferrante & Cavin, 2023 (Ferrante & Cavin 2023), Graulia branchiodonta Manuelli, Mondéjar Fernández, Dollman, Jakata & Cavin, 2024 (Manuelli et al. 2024), and Aemilia ‌stellata ‌n. gen., n. sp. (this study). The matrix was modified in MorphoBank. Only 59 out of 268 characters could be coded for Aemilia ‌stellata ‌n. gen., n. sp., 45 of which are hyoid and branchial arches characters that are poorly known in most coelacanths. Of these codings, 30 are plesiomorphic conditions, while 29 are apomorphic conditions.

A few codings were modified from the original matrix of Clement et al. (2024). The coding of character 82 for Graulia branchiodonta was based on the CT-scan rendering that shows the presence of the postorbital limb of the lacrimojugal (Manuelli et al. 2024: figs 5, 10A, B); however, the reconstruction of G. ‌branchiodonta (Manuelli et al. 2024: figs 5, 10A, B) does not show a clear postorbital limb. The coding of characters 207(0), 245(1), and 247(0) for Macropomoides ‌orientalis ‌Woodward, 1942 was completed from new observations on specimens from the Oishi Fossils Gallery of the Mizuta Memorial Museum of Tokyo, Japan (RC, pers. observ.). The coding of character 29 for Ngamugawi wirngarri was corrected to correspond to the apomorphic condition in which the postparietal is fused to the tabular (as described in Clement et al. 2024). We added a new apomorphic character-state for character 60 to take into account the condition of the lateral rostral of Rieppelia heinzfurreri: elongated vertically (length/height ratio < 1). Ferrante & Cavin (2023) described a one-to-two ratio between the radials and the lepidotrichia based on the numerous lepidotrichia identified in the caudal fin of R. ‌heinzfurreri. However, after careful examination (JMF) and discussions with C. Ferrante, we consider that the hemilepidotrichia were incorrectly identified as lepidotrichia, and thus the total amount of lepidotrichia (formed of two symmetrical hemilepidotrichia) is half of what was originally identified. Thus, the caudal fin of R. ‌heinzfurreri shows a one-to-one ratio between its radials and the lepidotriochia (character 54; character 106 of Ferrante & Cavin [2023, 2025]). Finally, the definition of character 56 of Clement et al. (2024) (size of sensory pores [Forey 1991: char. 38]) was modified to: pores on ethmosphenoid shield similar to postparietal shield pores (0), pores on ethmosphenoid shield larger than postparietal shield pores (1).

The stratigraphic information (first and last appearance dates, FAD and LAD, respectively) used for the tip-dating analysis of Aemilia ‌stellata ‌n. gen., n. sp. and Graulia branchiodonta (Manuelli et al. 2024) corresponds to the standards used by Clement et al. (2024: SI). The FAD and LAD of Rieppelia heinzfurreri correspond to the error on the dating (242.1 +/- 0.6) provided in figure 1 of Ferrante & Cavin (2023). The outgroup for the analyses is represented by five species of Onychodontida Andrews, 1973 from the Devonian.

We performed a parsimony analysis using TNT 1.6 (Goloboff & Morales 2023) using the New Traditionnal Search with Ratchet and Drift. We added ten initial sequences and set the parameters to find minimum length at least ten times. The analysis produced 34 equally parsimonious trees with a score of 1335. We also performed 1000 bootstrap replications. All the results are available in the Appendix 3.

We performed Bayesian tip-dated phylogenetic analyses using BEAST 2.7.7 (Bouckaert et al. 2019). This approach was particularly suitable given the extensive missing data in Aemilia ‌stellata ‌n. gen., n. sp., which precluded a reliable parsimony analysis but could be appropriately handled within a Bayesian framework. The most appropriate tree prior was the fossilized birth-death (FBD) model, implemented in the FBD package for BEAST 2 (Heath et al. 2014). We also conducted a complementary analysis using the Birth Death Skyline (BDSKY) model (Stadler et al. 2013), which produced similar topologies (see Appendix 3). Both analyses employed an optimized relaxed clock model, and Markov Chain Monte Carlo (MCMC) sampling as implemented in BEAST 2. We used coupled MCMC with four chains, which yielded consistent and converging results. Output log files were inspected using Tracer 1.7.2 (Rambaut et al. 2018), confirming robust parameter estimates with high effective sample sizes (ESS) (see.log files in the Appendix 3). The final tree was summarized using the Maximum Clade Credibility (MCC) approach with median node heights, via TreeAnnotator (from the BEAST 2 package), and visualized using FigTree 1.4.4 (Rambaut 2018).

The new features revealed by Aemilia ‌stellata ‌n. gen., n. sp. broaden our understanding of the organization and evolution of the neurocranium and hyobranchial skeleton of coelacanths, two of their most conspicuous and incompletely known anatomical systems. However, comparisons with other taxa are challenging, due to the scarcity of data concerning the internal structures of the skull, especially in Palaeozoic coelacanths. We will first discuss the new anatomical features of the skull roof, cheek, braincase and pectoral girdle, and their evolutionary implications highlighted by the new material of Aemiliastellata ‌n. gen., n. sp., and then focus exclusively on the hyobranchial skeleton to decipher the assembly and evolution of the hyoid and branchial arches in coelacanths.

The dermal skull roof of coelacanths is divided by the intracranial joint into an anterior parietonasal and a posterior postparietal shields. Latimeria ‌Smith, 1939 is the only living vertebrate to retain this plesiomorphic feature of sarcopterygians (Janvier 1996). The joint between the parietal (or posterior parietal sensuTeng et al. 2019) and postparietal bones can be straight and simple as in Latimeria (Millot & Anthony 1958; Forey 1998) or digitated as in Ngamugawi Clement, Cloutier, Lee, King, Vanhaesebroucke, Bradshaw, Dutel, Trinajstic & Long, 2024 (Clement et al. 2024). The straight condition can be considered plesiomorphic, when compared with early coelacanths (e.g., Miguashaia ‌Schultze, 1973; Cloutier 1996) and other sarcopterygians (e.g., ­Jarvik 1980; Janvier 1996). Unfortunately, in Aemiliastellata ‌n. gen., n. sp. the parietals are unknown and the postparietal shield is too badly preserved to ascertain whether the joint was either simple or digitated (Fig. 5).

The ventral surface of the postparietals of Aemiliastellata ‌n. gen., n. sp. is more phylogenetically informative as it displays a reduced descending process represented by a ridge (d.p.Pp; Fig. 5D) as in other Palaeozoic coelacanths (e.g., Coelacanthus ‌Agassiz, 1839, Spermatodus ‌Cope, 1894, and Rhabdoderma ‌Newberry, 1856; Forey 1998; Ferrante & Cavin 2025) compared with the stronger and more developed processes of the Mesozoic and Recent Latimerioidei (e.g., Graulia Manuelli, Mondéjar Fernández, Dollman, Jakata & Cavin, 2024,Macropoma ‌Agassiz, 1835, Latimeria; Forey 1998; Manuelli et al. 2024;). The tabular of Aemiliastellata ‌n. gen., n. sp. (Fig. 5) does not carry a significantly developed descending process, a plesiomorphic condition also known in Diplocercides ‌Stensiö, 1922 (Stensiö 1922, 1937; Forey 1998), but variably found among younger fossil coelacanths (e.g., Mawsonia ‌Woodward, 1907, Axelrodichthys ‌Maisey, 1986,Trachymetopon ‌Hennig, 1951; Forey 1998; Dutel et al. 2015b; Toriño et al. 2021b; Ferrante & Cavin 2025), in which it likely represents a derived loss from well-developed processes in Palaeozoic and early Mesozoic taxa like Rhabdoderma, Laugia, Dobrogeria ‌Cavin & Grădinaru, 2014, and Graulia (Forey 1998; Cavin & Grădinaru 2014; Manuelli et al. 2024).

The pores associated with the otic sensory canal open within the bones lateral to the postparietals, in case of Aemilia ‌stellata ‌n. gen., n. sp. it is the tabular (po.oc; Fig. 5F), a plesiomorphic feature for many sarcopterygians and early coelacanths like Miguashaia ‌(Cloutier 1996) and Diplocercides ‌(Forey 1998). Finally, the presence of pores on the preserved posterior portion of the postparietals in Aemiliastellata ‌n. gen., n. sp. suggests the earliest occurrence of anterior branches of the occipital commissure piercing the postparietal (po.a.occ; Fig. 5B), a character previously known only in Mesozoic and Recent coelacanths (e.g., the LatimerioideiGraulia, Macropoma, Swenzia ‌Clément, 2005, and Latimeria; Forey 1998; Clément 2005; Manuelli et al. 2024).

The posterior margin of the tabular of Aemiliastellata ‌n. gen., n. sp. carries a small shelf projecting posteriorly and receiving the unpreserved lateral extrascapular (po.sf; Fig. 5F), a condition similar to Laugia, Coccoderma ‌Quenstedt, 1858, and Spermatodus (Forey 1998; Westoll 1939). The occurrence of separated tabulars and postparietals is the most common arrangement in coelacanths, with only a few exceptions of tabulars fused to the postparietals (e.g., Ngamugawi,Foreyia ‌Cavin, Mennecart, Obrist, Costeur & Furrer, 2017,Ticinepomis ‌Rieppel, 1980,Hainbergia ‌Schweizer, 1966; Schweizer 1966; Cavin et al. 2017; Ferrante et al. 2023; Clement et al. 2024). The posterior extension of the tabular among the postparietal shield is difficult to reconstruct in Aemiliastellata ‌n. gen., n. sp. as the tabular was not found articulated with the postparietal. The plesiomorphic condition (e.g., Miguashaia and onychodontids; Cloutier 1996; Mondéjar-Fernández 2020) is the posterior margin of the tabular ending anteriorly to the posterior margin of the postparietals (Clement et al. 2024). Tabulars levelled with the posterior margin of the postparietals are seen in many Palaeozoic taxa (e.g., Diplocercides,Hadronector ‌Lund & Lund, 1984, Rhabdoderma; Forey 1998). A tabular extending beyond the posterior margin of the postparietals and the extrascapulars is considered the consequence of the fusion of the lateralmost extrascapular with the tabular (Stensiö 1921; Forey 1998; Ferrante & Cavin 2025). As a result, a junction of the otic canal with the occipital commissure and the main lateral line canal occurs in the tabular and not in the lateral extrascapular. These characters are frequently found correlated (e.g., Dobrogeria,Whiteia ‌Moy-Thomas, 1935, Axelrodichthys, Macropoma, Latimeria; Forey 1998; Cavin & Grădinaru 2014), but differ in Aemiliastellata ‌n. gen., n. sp. and other early coelacanths (e.g., Miguashaia,Ngamugawi; Cloutier 1996; Clement et al. 2024) and other sarcopterygians (e.g., onychodontids; Andrews et al. 2006; Mondéjar-Fernández 2020) in which the tabular is pierced by a single canal and the triple junction of lateral line canals occurs in the lateral extrascapular.

Unfortunately, the cheek of Aemiliastellata ‌n. gen., n. sp. is among the least well-preserved regions of the skull and little information can be obtained from it. All cheek bones are ornamented by rounded to slightly pointed tubercles (Figs 2A, C; 3E, G), similar to other Palaeozoic coelacanths like Miguashaia ‌(Cloutier 1996),Serenichthys ‌Gess & Coates, 2015 (Gess & Coates 2015), Ngamugawi (Clement et al. 2024), and Rhabdoderma (Forey 1981, 1998), but not quite as rounded as those of Latimeria and its close relatives (Millot & Anthony 1958; Forey 1998; Manuelli et al. 2024; Ferrante & Cavin 2025). Both the squamosal (Sq) and postorbital (Po; Figs 2E,G, 3E-G) are relatively large, different from the tube-like bones of various Mesozoic and Recent taxa (e.g., Coccoderma, Coelacanthus, Ticinepomis, Axelrodichthys, Latimeria; Forey 1998; Cavin et al. 2013; Ferrante et al. 2023; Ferrante & Cavin 2025). However, given the preservation of the specimen, it is unclear whether the cheek bones were tightly sutured to each other (representing the plesiomorphic condition known for instance in the Devonian Miguashaia,Gavinia ‌Long, 1999, Serenichthys, Diplocercides, and the Carboniferous Rhabdoderma; Forey 1981; Cloutier 1996; Long 1999; Gess & Coates 2015) or not (as most post-Carboniferous taxa such as the Triassic Sassenia ‌Stensiö, 1921; Forey 1998).

The few palatal elements found in Aemiliastellata ‌n. gen., n. sp. comprise an ectopterygoid (Ectp; Fig. 6L, N, P, R), a bone barely known in fossil coelacanths. In Aemilia ‌stellata ‌n. gen., n. sp., it carries a combination of small pointed teeth and larger fangs posteriorly, a condition different from the solely denticulated ectopterygoid of the Triassic Dobrogeria ‌(Cavin & Grădinaru 2014) and Graulia (Manuelli et al. 2024) and more similar to the combined occurrence of small teeth and fangs (albeit slightly reduced compared with the dermopalatine fangs) of the Cretaceous Macropoma ‌(Forey 1998) and Latimeria ‌(Millot & Anthony 1958).

An open buccohypophysial canal on the posterior portion of the parasphenoid is a plesiomorphic condition among coelacanths, present both in Palaeozoic (e.g., Diplocercides,Euporosteus ‌Jaeckel, 1927, Ngamugawi, Spermatodus) and Mesozoic taxa (e.g., Laugia, ‌ Piveteauaia Lehman, 1952, ­Parnaibaia Yabumoto, 2008; Forey 1998; Clément 1999; Yabumoto 2008; Clement et al. 2024; Ferrante & Cavin 2025). Aemiliastellata ‌n. gen., n. sp. retains an open buccohyphysial canal (bh.c; Fig. 6H, J), situated roughly at the ossification center of the parasphenoid, reminiscent of its position in Diplocercides (Bjerring 1993: fig.3). However, the patchy distribution of this character across coelacanth phylogeny is likely the result of a morphological disparity and the poor record of well-preserved braincases. For instance, in the case of Parnaibaia ‌maranhoensis ‌Yabumoto, 2008 (Yabumoto 2008), it may represent a reversion from a closed condition in the Latimerioidei and its Mesozoic close relatives (Manuelli et al. 2024), whereas in the Carboniferous Rhabdoderma ‌elegans ‌Newberry, 1856 the closed opening may represent an autapomorphic trait (Forey 1998).

The reduction of the ossification and fragmentation of a primitively solidly ossified ethmosphenoid and otico-occipital portions of the neurocranium into separated smaller ossifications is one of the clearest evolutionary trends among coelacanths (Forey 1998; Ferrante & Cavin 2025); similar trends are also occurring in actinopterygians (Patterson 1975; Gardiner 1984). Until now, very few preserved braincases were known in Palaeozoic coelacanths, hampering the precise reconstruction of the evolutionary timing and the modalities of the separation of the components of the ethmosphenoid and otico-occipital units of the braincase. Fully ossified braincases are solely known in Devonian coelacanths like Diplocercides, Euporosteus, and Ngamugawi (Stensiö 1922; Clement et al. 2024) but also in the Triassic Sassenia ‌groenlandica ‌Forey, 1998 (Forey 1998), likely representing an autapomorphic reversal trait found into the early Mesozoic. Aemiliastellata displays the derived condition of an otico-occipital portion of the braincase formed of separate elements, with a clear separation between the parasphenoid, basisphenoid, and prootics (Fig. 9), common in all post-Palaeozoic taxa. This is the earliest occurrence of this feature, prior to the Mesozoic, since the only known braincases from the Carboniferous are those of Rhabdoderma ‌elegans ‌Newberry, 1856 where the otico-occipital portion of the braincase is still ossified as a unit (Forey 1981, 1998).

The prootic is the largest ossification of the otico-occipital portion of the neurocranium in coelacanths. In Aemilia ‌stellata ‌n. gen., n. sp., it is clearly detached from the rest of the braincase (Figs 8; 9) and its postero-lateral margins are unfinished, indicating that the lateral commissure was largely cartilaginous. Due to this, there is no clear evidence for the articulation facets of the hyomandibular, pharyngo- and epibranchials as seen in Latimeriachalumnae ‌(Millot & Anthony 1958). The jugular vein runs along a prominent jugular canal and groove across the prootic of Aemilia ‌stellata ‌n. gen., n. sp. (c.ju; Fig. 8B, F, H), similar to Laugia (Forey 1998: fig. 6.7), a common and plesiomorphic condition of coelacanths, but secondarily absent in certain Mesozoic and Recent taxa (Macropoma, Mawsonia, Sassenia, and Latimeria; Maisey 1986; Forey 1998).

Aemiliastellata ‌n. gen., n. sp. presents a basioccipital separated from the prootic with a complex digitating suture between both bones (Fig. 9), as in many Mesozoic and Recent Latimerioidei (e.g., Graulia, Axelrodichthys, Undina ‌Münster, 1834, Macropoma, Holophagus ‌Egerton, 1861, and Latimeria; Forey 1998; Manuelli et al. 2024; Ferrante & Cavin 2025). This condition was previously unknown in the Palaeozoic, thus currently corresponding to the earliest occurrence of this configuration among coelacanths. This new combination of features displayed in Aemiliastellata ‌n. gen., n. sp. confirms that fully disjointed braincases in which both the ethmosphenoid (i.e., vomers, lateral ethmoids, basisphenoid, parasphenoid) and the otico-occipital (i.e., prootic, basioccipital, supraoccipital, exoccipital) portions are separated in different bones originated in coelacanths no later than the Upper Carboniferous.

It has also been stated that the development of descending processes in the skull roof may be associated with this overall reduction of the ossification of the neurocranium (Forey 1998). For instance, in Latimeria the neurocranium is securely attached to the roofing dermal bones through the descending processes of the parietals (corresponding to the posterior parietals), postparietals, and tabulars (Millot & Anthony 1958; Forey 1998; Manuelli et al. 2023, 2024). Descending processes occur in the ethmosphenoid division of the braincase, carried by the parietals, but only in post-Devonian coelacanths (e.g., Rhabdoderma, Spermatodus, Coelacanthus, Sassenia; Westoll 1939; Schaumberg 1978; Forey 1998; Ferrante & Cavin 2025). In the otico-occipital unit, descending processes appear first in the tabulars already in the Palaeozoic (e.g., the Carboniferous Rhabdoderma and the Permian Coelacanthus and Spermatodus; Forey 1998), but only later in the postparietals during the Mesozoic (e.g., the Triassic Sassenia, Laugia, Dobrogeria, and Piveteauia; Forey 1998; Clément 1999; Cavin & Grădinaru 2014; Ferrante & Cavin 2025). Aemiliastellata ‌n. gen., n. sp. does not display any clear descending processes on the tabular, whereas the postparietals display a narrow a ridge, visible on the left postparietal (d.p.Pp; Fig. 5D), similar to certain Mesozoic coelacanths like Laugia (Forey 1998: fig. 3.9) and Wimania ‌Stensiö, 1921 (Stensiö 1921: fig. 19), which could have fitted in a groove along the prefacial eminence of the prootic. However, in Aemiliastellata ‌n. gen., n. sp. the autapomorphic occurrence in the prootic of a long postero-dorsal process probably articulating with the tabular (ov.Ta?; Fig. 5D, F, J) may constitute an analogous structure to the ventral process of the tabular known in other coelacanths, but it is currently challenging to support this statement.

Attachment surfaces between the palate and the braincase are also quite variable and diagnostic among coelacanths. A basipterygoid process on the ethmosphenoid unit is only known in Diplocercides ‌(Forey 1998). Similarly, a suprapterygoid process occurs only in the Devonian Diplocercides ‌(Stensiö 1922) and Euporosteus ‌(Stensiö 1937), but also in the Triassic Sassenia ‌(Forey 1998). The suprapterygoid process of the ethmoid unit of the braincase has usually been synonymized with the antotic process of the separated basisphenoid in coelacanths (e.g., Stensiö 1932; Jarvik 1954), but the condition in these three latter taxa reveals that these are different processes, with the suprapterygoid process located anterior to the antotic process, which truly constitutes an actinistian synapomorphy (Forey 1998). The suprapterygoid process also disappears in post-Devonian coelacanths, and is no longer present in the Carboniferous as illustrated by Aemiliastellata ‌n. gen., n. sp. and Rhabdoderma, with the antotic process becoming larger and projecting farther laterally as the main articulation surface between the metapterygoid (palate) and the basisphenoid (braincase; Schaeffer & Gregory 1961).

The basisphenoid of Aemiliastellata ‌n. gen., n. sp. (Figs 7A-D; 9) lacks its anterior and dorsal margins, but the ventral half is relatively complete and recalls in many aspects that of Spermatodus as described by Westoll (1939). The processus connectens is short and thus it could not have met the parasphenoid (Fig. 9). A separation between the processus connectens and the parasphenoid can be considered a plesiomorphic feature of coelacanths as displayed in Devonian taxa like Diplocercides, Euporosteus, and Ngamugawi (Forey 1998; Clement et al. 2024), as well as in the LatimeriidaeMacropoma and Latimeria (Millot & Anthony 1958; Forey 1998). A contact between the processus connectens and the parasphenoid is mainly known in Mesozoic taxa (e.g., Laugia,Diplurus ‌Newberry, 1878, Dobrogeria, Wimania, Mawsonia,Megalocoelacanthus ‌Schwimmer, Stewart & Williams, 1994, Undina, Axelrodichthys; Maisey 1986; Forey 1998; Dutel et al. 2012; Cavin & Grădinaru 2014; Toriño et al. 2021b; Ferrante & Cavin 2025) but it also occurs in the Carboniferous Rhabdoderma ‌(Forey 1998). The distribution of this character reveals that it is a relatively homoplastic feature, making it difficult to reconstruct its phylogenetic history, especially given the current absence of data from the braincase of many Palaeozoic and early Mesozoic coelacanths.

The pectoral girdle is considered as remarkably conservative in coelacanths, with only minor differences in proportions of the cleithrum, clavicle, and shape of the anocleithrum among taxa (Forey 1998; Schaeffer 1941; Manuelli et al. 2024). However, the size and relationships of the extracleithrum with the other bones of the girdle shows a certain morphological disparity. The extracleithrum is a unique feature of actinistians among osteichthyans, located lateral to the cleithrum, and ventral to the level of the insertion of the pectoral fin on the scapulocoracoid. The extracleithrum is unornamented in Latimeria ‌(Millot & Anthony 1958) but in fossil coelacanths it can be variably ornamented by rounded tubercles (e.g., Miguashaia, Aemilia ‌stellata ‌n. gen., n. sp., Foreyia, Whiteia; Cloutier 1996; Forey 1998; Cavin et al. 2017) or thin ridges (e.g., Allenypterus ‌Melton, 1969, Coelacanthus, Laugia, Macropoma; Forey 1998). In Aemilia ‌stellata ‌n. gen., n. sp., the cleithrum and extracleithrum display a homogenous ornamentation of small tubercles (Fig. 13A, C), similar to the condition of Miguashaia ‌(Cloutier 1996) or Whiteia ‌(Forey 1998). In Rhabdoderma, only the dorsal portion of the cleithrum is ornamented while the extracleithrum is smooth (Forey 1981). However, in some taxa the ornamentation of the cleithrum and extracleithrum is slightly different from that of other dermal bones of the skull (e.g., Allenypterus, Laugia, Macropoma), represented by small ridges in the girdle as opposed to tubercles or larger ridges in the cheek and skull roof (Forey 1998).

The dorsal extension of extracleithrum is also quite variable. It projects more dorsally than the clavicle in Aemilia ‌stellata ‌n. gen., n. sp. (Fig. 13), as in many other taxa, both from the Palaeozoic (e.g., Allenypterus, Caridosuctor ‌Lund & Lund, 1984, Coelacanthus; Moy-Thomas & Westoll 1935; Forey 1998), Mesozoic (e.g., Laugia, Dobrogeria, Macropoma; Forey 1998; Cavin & Grădinaru 2014), and Recent (Latimeria; Millot & Anthony 1958; Forey 1998). In certain cases, the dorsal edge of the extracleithrum is levelled with that of the clavicle (e.g., Rhabdoderma, Mawsonia, Axelrodichthys, ­Holophagus; Forey 1998; Toriño et al. 2021b; Clement et al. 2024), whereas in Rebellatrix, the clavicle extends more dorsally than the extracleithrum (Wendruff & Wilson 2012). Miguashaia shows an interesting intrageneric disparity: in M. ‌bureaui ‌Schultze, 1973 the clavicle is levelled with the extracleithrum (Cloutier 1996) but in M. ‌grossi ‌Forey, Ahlberg, Lukševičs & Zupiņš, 2000 the clavicle extends more dorsally than the extracleithrum (Forey et al. 2000). In Aemilia ‌stellata ‌n. gen., n. sp., the suture of the extracleithrum with the cleithrum is more important than with the clavicle, similar to the condition of Coelacanthus, Macropoma, and Laugia (Stensiö 1932; Moy-Thomas & Westoll 1935; Schaumberg 1978; Forey 1998) but differing from the long contact between the extracleithrum and the clavicle in other taxa like Holophagus, Rebellatrix ‌Wendruff & Wilson, 2012, or Whiteia (Forey 1998; Wendruff & Wilson 2012). It is currently difficult to determine the phylogenetic significance of these features.

The cleithrum and clavicle do not display a postbranchial lamina in Aemilia ‌stellata ‌n. gen., n. sp. (Fig. 13), as in many other coelacanths (e.g., Rhabdoderma, Coelacanthus, Laugia, Axelrodichthys, Macropoma, Latimeria; Stensiö 1932; Moy-Thomas & Westoll 1935; Schaumberg 1978; Forey 1998). The postbranchial lamina is a plesiomorphic feature for gnathostomes (Brazeau & Friedman 2014) as present for instance in Psarolepis ‌romeri ‌Yu, 1998 (Zhu et al. 1999) and the putative early coelacanth Styloichthys ‌changae ‌Zhu & Yu, 2002 (Zhu & Yu 2002; Friedman 2007). Postbranchial laminae are more common in Devonian coelacanths like Miguashaia, Diplocercides, and Ngamugawi (Forey 1998; Cloutier 1996; Clement et al. 2024) but are also known in younger taxa like the Triassic Diplurus (Schaeffer 1948, 1952; Forey 1998) and Yunnancoelacanthus ‌Wen, Zhang, Hu, Benton, Zhou, Tiao, Huang & Chen, 2013 (Wen et al. 2013). The absence of a branchial lamina in post-Devonian coelacanths may indicate a slight shift in the position of the pectoral girdle relative to the opercular, in which the ventral portion of the cleithrum and clavicle no longer projects as anteriorly as in taxa with more consolidated dermal skulls (e.g., Miguashaia, Diplocercides), thus becoming more loosely attached to the opercular series and allowing more space to the water flow to be directed out of the gill chamber.

Finally, the scapulocoracoid of fossil coelacanths is poorly known, usually represented by small elements lying separated from the rest of the pectoral girdle (e.g., Diplocercides, Rhaboderma,Laugia, Graulia, Macropoma, Axelrodichthys; Stensiö 1922, 1932; Millot & Anthony 1958; Forey 1981, 1998; Manuelli et al. 2024). In Aemiliastellata ‌n. gen., n. sp. it is particularly massive, despite being weakly ossified and internally hollow, and it is one of the largest preserved elements of the pectoral girdle, similar to the condition of Latimeria ‌chalumnae ‌(Mansuit et al. 2020). This preserved portion of the scapulocoracoid corresponds to the postero-dorsal articulatory portion, which associated with the internal concave and smooth surfaces of the cleithrum, extracleithrum, and clavicle, suggests that the unmineralized blade of the scapulocoracoid was largely developed, as in Latimeria, and probably in many other fossil coelacanths. This large cartilaginous ventral blade of the scapulocoracoid explains why the small articulatory process is usually the only preserved element in fossil specimens (Mansuit et al. 2020). The relatively large scapulocoracoid of coelacanths differs from many other osteichthyans, where the scapulocoracoid is usually small compared with other components of the pectoral girdle (Janvier 1996), whereas among sarcopterygians large scapulocoracoids also convergently evolved in the lungfish Neoceratodus ‌forsteri ‌Krefft, 1870 (Johanson et al. 2004) and among tetrapodomorphs (Vickaryous & Hall 2006).

The hyobranchial skeleton, a portion of the visceral skeleton (or splanchnocranium) of gnathostomes plays not only an essential role in breathing for aquatic jawed vertebrates, but also assists in the opening of the jaws and the expansion of the bucco-pharyngeal cavity during suction feeding (e.g., Lauder & Reilly 1994; Dutel et al. 2015a). Among piscine gnathostomes, the gills (or hemibranchs) are supported by endoskeletal segmented elements (or arches), lying medially to branchial blood vessels and nerves. The primitive configuration of the gnathostome hyobranchial skeleton consists of the hyoid arch and five branchial (or gill) arches (Nelson 1969; Pradel et al. 2014; Dearden et al. 2019, 2024). A reduction of the number of branchial arches is relatively common among crown gnathostomes (e.g., four branchial arches in lungfishes and tetrapodomorphs; Janvier 1996) while supernumerary gill arches are extremely unusual (one singular exception are hexanchiform sharks with six gill arches, and among them Heptranchias ‌Rafinesque, 1810 with seven gill arches; Janvier 2004). Nevertheless, precise counting of the branchial arches is often difficult in fossil specimens due to their variable ossification degree, which usually decreases in the rearmost arches. Cases of completely preserved and ossified hyoid and branchial arches are rare in the Palaeozoic, not only among coelacanths, but also in gnathostomes as a whole (e.g., Brazeau et al. 2017; Coates et al. 2018; Dearden et al. 2019, 2024; Pradel et al. 2021).

The plesiomorphic configuration of the hyobranchial skeleton of osteichthyans consists of a hyoid arch, divided from ventral to dorsal into the hypohyal, ceratohyal, and hyomandibula, followed by the branchial arches, in turn divided into hypobranchials, ceratobranchials, epibranchials, and pharyngobranchials (infra- and suprapharyngobranchials in certain cases), with the whole system ventrally linked to the median basibranchial series, including an anterior basihyal, a variable number of basibranchials, and a posterior urohyal in certain cases (Nelson 1969; Pradel et al. 2014). One or two small bones may be interconnected between the ceratohyal and the hyomandibula in osteichthyans: the symplectic and the interhyal (Millot & Anthony 1958; Véran 1988; Argyriou et al. 2022). Five branchial arches are plesiomorphic for sarcopterygians (Janvier 1996) with coelacanths (Forey 1998) and porolepiforms (e.g., Laccognathus ‌panderi ‌Gross, 1941; Kanyukin 2006) displaying this condition. On the other hand, early dipnoans (e.g., Griphognathus ‌Gross, 1956,Chirodipterus ‌Gross, 1933) and their close relatives (e.g., Youngolepis ‌praecursor ‌Chang & Yu, 1981) only show four branchial arches (Miles 1977; Cui et al. 2022) and no more than four branchial arches can be confidently identified in tetrapodomorph fishes (e.g., Eusthenopteron ‌foordi ‌Whiteaves, 1881,Medoevia ‌lata ‌Lebedev, 1995,Mandageria ‌fairfaxi ‌Johanson & Ahlberg, 1997; Jarvik 1954; Lebedev 1995; Johanson & Ahlberg 1997) and early tetrapods (Witzmann 2013). The exact number of branchial arches in onychodontids is still unknown (Andrews et al. 2006; Mondéjar-Fernández 2020).

The organization of the coelacanth hyobranchial skeleton is considered unusual among bony fishes. Coelacanths, exemplified by Latimeria, are remarkable in lacking a basihyal (present in actinopterygians and other sarcopterygians) and hypobranchials, resulting in a direct contact of the single basibranchial with the ceratobranchials. A small cartilaginous hypohyal is present in Latimeria ‌chalumnae, but it is unknown in all other coelacanths (Millot & Anthony 1958; Wiley 1979; Forey 1998). Indeed, gill arches in fossil coelacanths are rarely preserved, and when found they are greatly incomplete, fragmentary or displaced from their original position. Among these, the upper portion of the branchial skeleton is poorly known and, in many cases, confusingly illustrated. This is due to the fact that most of our knowledge on the branchial arches come from flattened specimens. Observations on Latimeria also revealed a reduction of ossification (decrease of periosteal thickness) occurring from lateral to medial in the cerato-, epi-, and pharyngobranchial series (Millot & Anthony 1958; Forey 1998; Manuelli et al. 2023, 2024). This pattern may explain why so many elements from the hyobranchial skeleton are often missing or unpreserved in fossil coelacanths. Moreover, the anatomy of many early coelacanths does not necessarily match that of Latimeria, thus complicating the identification of elements. Because of these issues, the gill arches of fossil coelacanths are among the least surveyed skeletal systems of their anatomy and thus few evolutionary trends can be recognized (Forey 1998). Nevertheless, exceptionally preserved fossil specimens like Graulia (Manuelli et al. 2024) and Aemiliastellata ‌n. gen., n. sp. may reveal new and unexpected combinations of characters that could shed light on the organization and evolution of the hyobranchial skeleton in coelacanths.

The new phylogenetical analysis reveals two previously unrecognized radiation episodes during the Late Palaeozoic-Early Mesozoic coelacanth evolutionary history. These radiations correspond to a Permian-Triassic-Jurassic clade to which Aemilia ‌stellata ‌n. gen., n. sp. is the sister group (clade 3; Fig. 15) and a clade including all species of Whiteia and other closely related taxa (clade 4; Fig. 15). The diversification of these clades encompasses the early radiation of the Latimerioidei, especially during the Triassic and Jurassic. Moreover, Aemilia ‌stellata ‌n. gen., n. sp. also constitutes the oldest representative of a larger clade comprising the main diversification episodes of coelacanths starting at the beginning of the Mesozoic. By using a phylogenetic definition, we name this clade Coelacanthiformes (Fig. 15). The recent study by Ferrante & Cavin (2025) was conducted in parallel to ours and in it they list the characters supporting the Coelacanthiformes clade, as well as other actinistian taxa. We will briefly discuss and compare their approach to ours.

The term Coelacanthini (or Coelacanthiformes) was first introduced by Huxley (1861) in his classification of fossil fishes to include the only coelacanths known at the time: Undina ‌penicillata ‌Münster, 1834,Macropoma ‌mantelli ‌Agassiz, 1835, andCoelacanthus ‌granulatus ‌Agassiz, 1839. The initial sense of Huxley’s term was later synonymized with Actinistia ‌Cope, 1871 but both names coexisted during much of the 20th century (e.g., Berg 1940; Echols 1963; Lund & Lund 1985; Long 1999). In an attempt of reinstating the term and assigning to it a proper phylogenetic definition, Forey (1998) defined the order Coelacanthiformes as “actinistians in which the otico-occipital portion of the neurocranium is fragmented to separate prootic, opisthotic, basioccipital and supraoccipital bones, with a complex suture between the prootic and the basioccipital, loss of vestibular fontanelle and ­buccohypophysial canal” (Forey 1998: 300), with the Permian Coelacanthus ‌granulatus as its oldest member (Schultze 2004), but without a similarly rank-named sister group (Forey 1998; Ferrante & Cavin 2025). However, Forey’s definition turned out to be problematic as it combined up to four different characters that have been revealed to have evolved independently among coelacanths (e.g., Ferrante & Cavin 2023, 2025; Manuelli et al. 2024; Clement et al. 2024). At the time of publication, these characters occurred simultaneously in the main taxa used by Forey to build his phylogenetic study. New discoveries have highlighted a greater morphological disparity than previously expected, resulting in a patchy distribution of these features, especially the occurrence of a closed buccohypophysial canal. Moreover, the poor record of well-preserved braincases among coelacanths makes it particularly challenging to test the putative combined occurrence of these characters in many fossil specimens.

The main characters defining the order Coelacanthiformes are found in the neurocranium, and concern the separate ossification of the otico-occipital portion and the presence or absence of buccohypophysial canal on the parasphenoid. The separation between the elements forming the otico-­occipital portion of the braincase is indeed a derived character exclusively known in Mesozoic and Recent taxa (e.g., Laugia ‌groenlandica, Wimania ‌sinuosa, Dobrogeria aegyssensis, Graulia branchiodonta, Rieppelia heinzfurreri, Diplurus ‌newarki, Whiteia ‌woodwardi,Axelrodichthys ‌araripensis ‌Maisey, 1986, Mawsonia ‌gigas, Macropoma (M. ‌lewesiensis ‌Agassiz, 1835 and M. ‌precursor ‌Woodward, 1909), Holophagus ‌gulo, Undina ‌penicillata,Trachymetopon ‌liassicum ‌Hennig, 1951,Megalocoelacanthus ‌dobiei ‌Schwimmer, Stewart & Williams, 1994, and Latimeria ‌chalumnae). Aemilia ‌stellata ‌n. gen., n. sp. is the first Palaeozoic coelacanth to display this feature. Similarly, a closed buccohypophysial foramen occurs mainly in Mesozoic coelacanths like the Triassic Axelia ‌robusta, Wimania sinuosa, Graulia branchiodonta, Rieppelia heinzfurreri, Whiteia ‌woodwardi,Luopingcoelacanthus ‌eurylacrimalis ‌Wen, Zhang, Hu, Benton, Zhou, Tiao, Huang & Chen, 2013, the Jurassic Diplurus ‌newarki, the Cretaceous Axelrodichthys (A. ‌araripensis, A. ‌lavocati ‌Tabaste, 1963,A. ‌megadromos ‌Cavin, Valenti & Garcia, 2016), Macropoma (M. ‌lewesiensis and M. ‌precursor), Undina ‌penicillata, Megalocoelacanthus ‌dobiei, and the extant Latimeria ‌chalumnae, but it has also been described in the Carboniferous Polyosteorhynchus ‌simplex ‌Lund & Lund, 1984 and Rhabdoderma (R. ‌elegans,R.? ‌newelli ‌Hibbard, 1933). On the other hand, an open buccohypophysial foramen on the parasphenoid represents the plesiomorphic condition and is thus more common among Palaeozoic taxa (e.g., Styloichthyschangae, Miguashaia ‌bureaui, Euporosteus eifeliensis, Ngamugawi wirngarri, Diplocercides ‌kayseri, Allenypterus ‌montanus, Caridosuctor ‌populosum, Rhabdoderma ‌exiguum, Coelacanthus ‌granulatus), but it also occurs in some Mesozoic taxa (e.g., Laugia ‌groenlandica, Piveteauia ‌madagascarensis, Parnaibaia ‌maranhoensis). The case of intrageneric variation in Rhaboderma is remarkable: in Rhabdoderma ‌exiguum ‌Eastman, 1902 the foramen is open whereas in Rhabdoderma ‌elegans it is closed.

According to Forey’s (1998) initial diagnosis of the group, Aemilia ‌stellata ‌n. gen., n. sp. should not be considered a Coelacanthiformes since the buccohypophyseal foramen is plesiomorphically open (Figs 6G, H; 9B, C). However, Aemilia ‌stellata ‌n. gen., n. sp. also displays the derived conditions of a separate basioccipital and a prootic with a complex suture with the basioccipital (Figs 7Q-T; 8; 9). Aemilia ‌stellata ‌n. gen., n. sp. thus suggests that these features may not be evolutionary linked and thus their combined occurrence constitutes an unreliable argument for the definition of the order Coelacanthiformes as an apomorphy-based clade, sensu Forey (1998).

Recently, Ferrante & Cavin (2025) emended the diagnosis of the Coelacanthiformes as “actinistians characterised by the following unique combination of characters: neurocranium fragmented to separate prootic, opisthotic and basioccipital; premaxilla usually without a dorsal lamina; anterior opening of the rostral organ occurring usually within separated rostral ossicles; posterior margin of the skull roof usually embayed; subopercle usually absent; posterior opening(s) of the rostral organ usually mark as a notch(es); infraorbital sensory canal usually running through centre of the postorbital; pit lines usually not marking cheek bones; ventral margin of the basal plate of the anterior dorsal fin usually smooth.” (Ferrante & Cavin 2025: 130). In their diagnosis, there is no reference to the buccohypophysial canal, as opposed to Forey (1998), and this character is not diagnostic of any node in their tree (Ferrante & Cavin 2025: fig. 73). In our Bayesian analysis, the clade Coelacanthiformes is supported by four characters: 77 (anterior lacrymojugal sensory pore(s) expanded or forming a groove), 103 (absence of a suboperculum), 137 (deepest portion of the angular at approximately midway along the length of the angular, but also present in Caridosuctorpopulosum), and 184 (otico-occipital portion of the braincase separated into prootic and opisthotic).

Based on the homoplastic distribution of some of these characters and the lack of proper knowledge on many fossil coelacanths neurocrania, we propose instead to redefine the order Coelacanthiformes ‌Huxley, 1861 as a node-based taxon consisting of the least inclusive clade comprising the last common ancestor of Coelacanthus ‌granulatus ‌Agassiz, 1839 and Latimeria ‌chalumnae ‌Smith, 1939, and all of its descendants. This definition is congruent with the character-based one proposed by Ferrante & Cavin (2025).

Finally, the Bayesian tip-dating analysis also suggests that the splitting of the main clades of Coelacanthiformes (i.e., Laugiidae+Axeliidae clade, clade 3 including Aemilia ‌stellata ‌n. gen., n. sp., and clade 4+Latimerioidei) may have occurred deep in the Carboniferous, while the vast majority of taxa belonging to these clades date from the beginning of the Mesozoic, especially the Triassic. However, the number of coelacanth taxa known from the Permian is extremely limited (e.g., Coelacanthus ‌granulatus, Spermatodus ‌pustulosus, Changxingia (C. ‌weii ‌Jin, 1997 andC. ‌aspratilis ‌Wang & Liu, 1981), and Youngichthys ‌xinghuainsis ‌Wang & Liu, 1981; Agassiz 1839; Moy-Thomas & Westoll 1935; Westoll 1939; Wang & Liu 1981;Friedman et al. 2007). The unexpected phylogenetic patterns revealed by our new phylogenetic analysis suggest that the diversity of Carboniferous and more particularly Permian coelacanths remains largely undersampled.

Coelacanth evolution has traditionally been described as conservative (e.g., Cloutier 1991a, b; Forey 1998; Clement et al. 2024; Ferrante & Cavin 2025) but not necessarily due to a simple lack of variation (which is now well established by new and sometimes odd morphologies; e.g., Friedman & Coates 2006; Wendruff & Wilson 2012; Cavin et al. 2017; Ferrante & Cavin 2023) but rather because these few limited, punctual variations span more than 400 million years of evolution, making coelacanths one of the most fascinating groups of vertebrates. Clement et al. (2024) pointed out that the early Palaeozoic (especially the Devonian) saw a rapid burst of morphological evolution and that major morphological innovations ceased to occur from the late Mesozoic, followed by a still unexistant fossil record of Cenozoic coelacanths.

It has been proposed that one of the key factors explaining the survival of extant coelacanths after the Cretaceous is their adaptation to moderate oceanic depths (Cupello et al. 2019a) and an incredibly slow metabolism that allows Latimeria ‌chalumnae to live past one hundred years, despite an impressively late reach of sexual maturity at around fifty years (Mahé et al. 2021). An explanation to this slow metabolism lies in its extremely reduced proportion of oxygen absorbed by the gills (Hughes 1972, 1976, 1980, 1998). However, fossil coelacanths are also known to posses a lung to support bimodal respiration along with the gills, which allowed them to thrive in both fresh and shallow to moderate deep marine waters (Cupello et al. 2017a, b, 2019a, b). The lung was functional and surrounded by bony plates in many Palaeozoic and Mesozoic taxa (e.g., Rhabdoderma, Coelacanthus,Laugia,Graulia, Macropoma, Mawsonia, Axelrodichthys; Forey 1998; Brito et al. 2010; Cupello et al. 2017b; Manuelli et al. 2024), whereas in the extant Latimeriachalumnaethe lung is present but vestigial and non-­functional (Millot & Anthony 1965; Cupello et al. 2017b), and the gas exchange is made exclusively through the gills. Unfortunately, physiology does not fossilize so it is currently not possible to evaluate the evolution of oxygen intake in coelacanths. However, the skeleton supporting the gills, namely the hyobranchial skeleton, appears not to be affected by these metabolic changes as evidenced by the rather stable morphology and organization of the hyoid and branchial arches across coelacanth evolution, especially the elements carrying hemibranchs (i.e., the hyomandibula and ceratobranchials). Nevertheless, the dorsal portion of the branchial arches, comprising the epibranchial and pharyngobranchial series appears to have been more plastic, but unfortunately its components are among the least well-preserved elements of the hyobranchial skeleton in fossil coelacanths.

The new Carboniferous coelacanth Aemilia ‌stellata ‌n. gen., n. sp. fills an important gap in our knowledge of the morphological diversity of coelacanths during the Late Palaeozoic. Aemilia ‌stellata ‌n. gen., n. sp. preserves almost the entire hyoid and branchial arches in partial articulation. Other exceptionally preserved coelacanths also display a rather complete hyobranchial skeleton (e.g., Ticinepomis, ‌ Ngamugawi, Graulia; Cavin et al. 2013; Clement et al. 2024; Manuelli et al. 2024; Table 1), however Aemilia ‌stellata ‌n. gen., n. sp. is remarkable in that it uncovers elements never before described in a fossil coelacanth. Among these, the epibranchial and pharyngobranchial series have been revealed in great detail, allowing comparisons with only a few extinct forms (e.g., Laugia,Dobrogeria,Graulia) besides the extant Latimeria. The new anatomical information supports the idea that the hyobranchial skeleton of post-Carboniferous coelacanths may be considered derived relative to the plesiomorphic condition displayed in Aemilia ‌stellata ‌n. gen., n. sp., and that the hyoid and branchial arches of younger and Recent taxa like Latimeria are the evolutionary result of a morphological simplification and a reduction in ossification, especially in the epibranchial and pharyngobranchial series. Further comparative data on other Palaeozoic coelacanths is greatly desired in order to support this statement.

A new phylogenetic analysis with the addition of new taxa to the previously published matrix by Clement et al. (2024) has reconstructed Aemilia ‌stellata ‌n. gen., n. sp. as the oldest Coelacanthiformes (sensu our new phylogenetic definition of the order) and the earliest representative of a major post-­Palaeozoic radiation of coelacanths. This new middle Palaeozoic species bridges the gap between the Devonian and post-Carboniferous clades of coelacanths, shedding light on a relatively uncharted period of their evolutionary history and revealing that the fossil record of Carboniferous and especially Permian coelacanths is still greatly unexplored.

New taxa like Aemilia ‌stellata ‌n. gen., n. sp. unveil novel and unexpected combinations of features that reconfigure our understanding of the systematics and anatomical evolution of coelacanths, especially focusing on the neurocranial and hyobranchial skeleton. We predict that future studies on coelacanth anatomy will heavily focus on the yet untapped diversity of these skull systems as tomographic studies applied to exceptionally preserved fossils continue developing. We hope that the new data such as the ones presented here will establish a firm comparative basis for future descriptions of extinct coelacanths and will continue shedding light on some of the lesser-known episodes of the long evolutionary history of these iconic fishes.

We warmly thank Philippe Janvier (Muséum national d’Histoire naturelle, Paris), Per Ahlberg (Uppsala University), John Long and Alice Clement (Flinders University, Adelaide), Luigi Manuelli, Christophe Ferrante, and Lionel Cavin (Muséum d’Histoire naturelle, Genève) for insights and inspiring discussions on coelacanth anatomy and evolution. Dieter Uhl (Senckenberg Research Institute and Natural History Museum) is thanked for providing essential hardware and ressources. Florent Goussard (Muséum national d’Histoire naturelle, Paris) provided software and technical assistance, and Marc Herbin (Muséum national d’Histoire naturelle, Paris) granted access to the coelacanth collections at the MNHN under his care. Richard Dearden (University of Birmingham) shared precious advice and guidance on the use of Blender. JMF was supported by the Louis Gentil-Jacques Bourcart prize of the French Academy of Sciences. We acknowledge the work of Lionel Cavin and an anonymous reviewer and thank them for their constructive remarks.

Images stacks (CT scan microtomography) of Aemilia ‌stellata n.gen, n.sp (AMNHN FF 20686). Available at: https://doi.org/10.17602/M2/M859331

3D mesh (.stl files) of Aemilia ‌stellata n.gen, n.sp (AMNHN FF 20686). Available at: https://doi.org/10.17602/M2/L859329

Supplementary information: 1, Coelacanth phylogeny including Aemilia ‌stellata ‌n.gen., n.sp. Character matrix, parsimony and Bayesian (BDSKY and FBD) analyses; 2, table of fossil coelacanths with preserved hyoid, branchial and basibranchial elements; 3, Coelacanth silhouettes and supplementary references. Available at: https://doi.org/10.7934/P6353