A three‐dimensional placoderm (stem‐group gnathostome) pharyngeal skeleton and its implications for primitive gnathostome pharyngeal architecture

Abstract The pharyngeal skeleton is a key vertebrate anatomical system in debates on the origin of jaws and gnathostome (jawed vertebrate) feeding. Furthermore, it offers considerable potential as a source of phylogenetic data. Well‐preserved examples of pharyngeal skeletons from stem‐group gnathostomes remain poorly known. Here, we describe an articulated, nearly complete pharyngeal skeleton in an Early Devonian placoderm fish, Paraplesiobatis heinrichsi Broili, from Hunsrück Slate of Germany. Using synchrotron light tomography, we resolve and reconstruct the three‐dimensional gill arch architecture of Paraplesiobatis and compare it with other gnathostomes. The preserved pharyngeal skeleton comprises elements of the hyoid arch (probable ceratohyal) and a series of branchial arches. Limited resolution in the tomography scan causes some uncertainty in interpreting the exact number of arches preserved. However, at least four branchial arches are present. The final and penultimate arches are connected as in osteichthyans. A single median basihyal is present as in chondrichthyans. No dorsal (epibranchial or pharyngobranchial) elements are observed. The structure of the pharyngeal skeleton of Paraplesiobatis agrees well with Pseudopetalichthys from the same deposit, allowing an alternative interpretation of the latter taxon. The phylogenetic significance of Paraplesiobatis is considered. A median basihyal is likely an ancestral gnathostome character, probably with some connection to both the hyoid and the first branchial arch pair. Unpaired basibranchial bones may be independently derived in chondrichthyans and osteichthyans.

pharyngeal skeleton character polarities, limiting their phylogenetic usefulness.
The two main divisions of the gnathostome crown group (Osteichthyes and Chondrichthyes) differ significantly in the pattern of topological relationships of their pharyngeal arch segments, particularly in the ventral elements. Both chondrichthyans and osteichthyans possess one or more median basibranchial elements. Chondrichthyan hyoid arches consist of ceratohyals attaching directly to a median hypohyal element (Allis, 1922(Allis, , 1923Carvalho, Bockmann, & de Carvalho, 2013;Garman, 1913;Pradel et al., 2014;Shirai, 1992). By contrast, in osteichthyans, the ceratohyals attach to a median bone (usually termed a basibranchial) via hypohyals, (Allis, 1897(Allis, , 1922Grande & Bemis, 1998;Jarvik, 1954Jarvik, , 1972) the latter of which are usually considered absent in chondrichthyans. In both chondrichthyans and osteichthyans, each branchial arch terminates ventrally and medially by a hypobranchial bone or cartilage, which may connect either at the midline to its antimere or to a ventral basibranchial bone or cartilage. In osteichthyans, the hypobranchials are considered to be "anteriorly directed" and at least one anterior branchial arch pair usually joins the same basibranchial as the hyoid arch. In elasmobranch chondrichthyans, all but the first hypobranchials are (usually) posteriorly directed, and join either at the anatomical mid-line or to a basibranchial copula. The first hypobranchial is often anteriorly directed and connected to the posterolateral angle of the basihyal (see e.g., Garman, 1913).
All of these contrasts are potentially phylogenetically informative variables (Brazeau & Friedman, 2014;Pradel et al., 2014). Although consideration of fossils aids the separation of plesiomorphic and derived conditions (Pradel et al., 2014), considerable uncertainty remains. The lack of an outgroup information from stem-group gnathostomes has inhibited character mapping exercises attempting to reconstruct primitive branchial arch conditions (Pradel et al., 2014).
Fossilized gill skeletons are extremely poorly known from jawless gnathostome outgroups (Conway Morris, & Caron, 2014;Janvier & Arsenault, 2007;Janvier, Desbiens, Willett, & Arsenault, 2006). However, partially ossified examples exist in the placoderms: an assemblage of Paleozoic jaw-bearing stem-group gnathostomes. Unfortunately, these gill skeletons are weakly mineralized and therefore tend to be poorly preserved. Complete, articulated placoderm fossils, particularly with preserved endoskeletons, are extremely rare. However, multiple examples are known from the exceptional Early Devonian Hunsr€ uck Slate Lagerstätte of Germany. Among these are the anatomically and phylogenetically enigmatic "stensioellids," including Stensioella, Pseudopetalichthys, Nessariostoma, and Paraplesiobatis. This assemblage is so morphologically diverse, however, that the group is unlikely to be phylogenetically coherent. Nevertheless, these taxa are known from complete and articulated fossils. Furthermore, these fossils exhibit X-ray contrast (Gross, 1962) and are therefore amenable to computed tomography (CT) investigations.
This article provides further details on the morphology of placoderm gill arches through a synchrotron CT-analysis of Paraplesiobatis heinrichsi Broili (1933). Data from Paraplesiobatis confirm some generalized aspects of placoderm branchial arch anatomy observed in less complete examples. However, it also presents peculiarities that raise questions about the homology of some elements in chondrichthyan and osteichthyan branchial skeletons. A comparative analysis of basal gill arch elements in early and modern gnathostomes is here used to infer some aspects of primitive branchial arch patterns.

| Geological context
The Hunsr€ uck Slate is an offshore, muddy facies of Early Devonian age in the Rhenish Mountains of southeastern Germany (Bartels, Briggs, & Brassel, 1998;Schindler, Sutcliffe, Bartels, Poschmann, & Wuttke, 2002), but lacks a formal stratigraphic status despite common usage in the paleontological and geological literature. Bartels et al. (1998) suggest that the Hunsr€ uck Slate is effectively comparable to a group, rather than formation, in lithostratigraphic nomenclature. The exact collection horizon for Paraplesiobatis within the Hunsr€ uck succession is unclear, but it likely derives from the clay-rich, mid-basinal Kaub Formation at the Bundenbach locality (Gross, 1962). An ash layer near the base of the Kaub Formation is radiometrically dated as 407.7 6 0.7 Ma (Kaufmann, Trapp, Mezger, & Weddige, 2005), while the top of the formation at Bundenbach extends into the Nowakia elegans Dacryoconarid Zone (De Baets, Klug, Korn, Bartels, & Poschmann, 2013). The N.
elegans Zone lies within the Polygnathus inversus Conodont Zone, the top of which has been spline-dated to 397.68 6 2.144 Ma (Becker, Gradstein, & Hammer, 2012). Thus the age of Paraplesiobatis can be roughly constrained to between 398 and 408 Ma.

| Synchrotron tomography
The specimen was scanned using synchrotron radiation X-ray microtomography at the I12-JEEP beamline of the Diamond Light Source, Didcot, UK (Drakopoulos et al., 2015). Tomography was performed by acquiring X-ray radiographs at 0.1 deg angular spacing using lambda50.0124 nm (100 keV) monochromatic X-rays, illuminating a 0.9 mm Cadmium Tungstate (CdWO4) scintillator which was imaged by microscope optics onto the detector of PCO4000 CMOS camera (PCO-AG, Germany), with a projected pixel size of 12.6 micron at the sample.
The tomographic 3-d images were reconstructed using filtered back projection and ring suppression (Titarenko, Titarenko, Kyrieleis, Withers, & De Carlo, 2011). The resulting voxel size was 12.7 mm. Because of the size of the specimen, seven overlapping scans were generated using the automated vertical translation stages supporting the tomography rotation stage. These were later "stitched" together in postprocessing.

| Segmentation and virtual modeling
The resultant volume was loaded in Mimics (Materialise Software) and completed in version 18 and some earlier versions. Particular care was taken to avoid the influence of a pair of heavy ring artefacts and the "brightness artefacts" caused by highly dense pyrite crystals (Supporting Information Figure 1). Furthermore, the discontinuous grayscale normalization between each separate scan series comprising the total volume (Supporting Information Figure 1) caused difficulties in selecting consistent threshold values. Therefore, most of the individual bones were formed from multiple masks that were later united using Boolean operations. Surface models used in this study are freely available at https://doi.org/10.6084/m9.figshare.4555462

| D E S C RI P T ION
The gill skeleton of KGM 1983/294 is preserved as a segmented network of perichondrally ossified bones (Figures 1 and 2). The skeleton is mostly intact and nearly complete ventral gill basket spanning the entire breadth of the skull. No dorsal (epibranchial or pharyngobranchial) elements are preserved. The gill skeleton comprises a single, median basihyal and set of four to five segmented arches (explained below). The first four arches consist of two segments: a medial segment (that joins its antimere across the ventral midline), and a lateral segment. The fifth arch consists of a single visible segment that connects to the posterolateral facet of the fourth arch, rather than meeting its antimere. There are no additional unpaired median bones preserved. The first arch in the series is preserved only on the right side (however, a small fragment of it may be preserved on the left). This is the most problematic element to describe because the tomographic data is severely afflicted by a number of artifacts. The bone has an unusual morphology, consisting of a broad, roughly oval ventral "blade," and a dorsal flattened rod-like region (Figure 3). This element is described as a single structure here. However, we consider it possible that it is a composite of two separate bones, a ceratohyal and first branchial arch, that were brought into close proximity when the pharynx was disrupted and cannot be resolved in the scan.  (Carr, Johanson, & Ritchie, 2009;Ritchie, 2005). It is notable that in KGM 1983/294 these are completely continuous with the lateral rod-like projection ( Figure 2). It is, therefore, unclear whether these correspond to fused ceratobranchial, hypobranchial, and basibranchial elements.

| Alternative interpretation of Pseudopetalichthys
The most complete placoderm branchial skeleton known belongs to the now lost (Wuttke, 1986) type and only specimen of Pseudopetalichthys problematicus Moy-Thomas. It shares in common with A key difference in the gill skeletons of Pseudopetalichthys and Paraplesiobatis is that the gill skeleton of the former appears segmented between the "basibranchials" and the "ceratobranchials." However, a similar pattern can be generated in Paraplesiobatis by embedding the reconstructed arches in a plane (Figure 4). The differences could, therefore, reflect differences in the degree of ossification or partial burial of the gill skeleton in Pseudopetalichthys.   Gross, 1962) with pharyngeal arch elements highlighted to show correspondence with Paraplesiobatis suggesting that the segmented appearance may be misleading (c). Not to scale pytctodontids (Forey & Gardiner, 1986;Long, 1997;Miles, 1967) and the so-called "stensioellids" Stensioella and Pseudopetalichthys (Gross, 1962), the latter will be treated in a separate subsection below. The remaining "stensioellids" will be treated elsewhere.

| Basihyal
A single median basihyal is common (possibly universal) among placoderms ( Figure 5). This structure is observed clearly in Tapinosteus

| The number and structure of branchial arches in placoderms
In all examples of placoderms, and here further corroborated by KGM 1983/294, the basihyal is followed by a paired series of basibranchiallike ossifications. In Tapinosteus  The branchial bones of KGM 1983/294 consist of arches comprising only two segments: a medial and lateral one. The area corresponding to the "basibranchials" ("hypobranchials" of Stensi€ o) in other placoderms are joined to extended lateral processes that would themselves correspond to either hypobranchials or ceratobranchials. The first arch is of uncertain hyoid or branchial identity and the fifth arch does not have its own "basibranchial" region, but rather joins the preceding (fourth) arch. The "basibranchial" series of Cowralepis numbers four elements (Carr et al., 2009), while there are three pairs in Tapinos teus (Stensi€ o, 1969). However, Stensi€ o reconstructs multiple arches joining these bones based on a chondrichthyan, and more specifically, hexanchiform, interpretive model.
In Gemuendina, Gross (1963) illustrates up to four separate arches.
However, an indistinct fifth arch can be observed in his plate 8, figure   B. Notably, the proximal end of the first arch bears a spatulate expansion (shown both in Gross's illustration and photographic plate). This spatulate expansion is nestled behind the mandibular arch cartilage and could reasonably be interpreted as a ceratohyal. Based on these details, it seems reasonable to conclude that placoderms generally possess a median basibranchial and at least four "basibranchial" elements, corresponding to at least four individual pharyngeal arches. What remains unclear is whether the "basibranchial" elements should, in fact, be interpreted as that, or whether they more realistically correspond to hypobranchials of other gnathostomes.

| Comparison with crown-group gnathostomes 4.3.1 | Similarities with chondrichthyans
The presence of a median basihyal in Paraplesiobatis coupled with no apparent hypohyals compares well with modern chondrichthyans (Figure 5; see e.g., Carvalho et al., 2013). The outline of the basihyal in KGM 1983/294 is similar to that of the Carboniferous holocephalan Debeerius (Grogan & Lund, 2000) in being roughly trapezoidal with anterolateral "notches," possibly accommodating articulation with the hyoid arch ( Figure 5).

| Interpretation and reconstruction
Paraplesiobatis specimen KGM 1983/294 preserves at least five pharyngeal arches. Here, we offer two competing interpretations in relation to the arches of other gnathostomes ( Figure 5). The first interpretation is that the first preserved arch is a hyoid arch (comprising a pair of ceratohyals articulating with the basihyal). The subsequent arches would therefore consist of four branchial arches. Alternatively, the model of the first arch is, in fact, a composite of two separate elements that were pushed together when the front of the gill skeleton was dislodged. The ovate, blade-like element would correspond to a displaced ceratohyal, the upper rod-shaped region would be the medial element of branchial arch 1. In this case, the arches correspond to branchial arches only, numbering 1-5. As in osteichthyans, arch 5 articulates with arch 4, but is highly differentiated.

| Phylogenetic distributions
Under the assumption that placoderms (including Paraplesiobatis) are stem-group gnathostomes ( Figure 5), we can make a number of phylogenetic inferences. There are assumed to be monophyletic chondrichthyan and osteichthyan crown groups. Most analyses favor placoderm paraphyly, however, King, Qiao, Lee, Zhu and Long (2016) recently demonstrate problems with this result and recover a majority of placoderms as monophyletic (King et al., 2016) under modified analytical methods. We, therefore, elect to make no prior assumptions about placoderm monophyly. Unless otherwise indicated, we will offer interpretations consistent with both monophyly or paraphyly of placoderms.
A median basihyal is apparently universal among placoderms, it is common to crown and some putative stem-group chondrichthyans. The alternative state, a pair of hypohyals connecting to a median basihyal, appears to be unique to osteichthyans among extant gnathostomes.
When extinct gnathostomes are considered, Acanthodes raises a caveat with respect to the presence or absence of hypohyals in chondrichthyans. This taxon is now increasingly considered a stem-group chondrichthyan (Brazeau & de Winter, 2015;Giles, Friedman, & Brazeau, 2015;Zhu et al., 2013), but it presents a pattern of pharyngeal segmentation with multiple (not necessarily competing) interpretations.
The ceratohyal of Acanthodes, according to Gardiner (1984) is composed of two discrete ossifications. This could be interpreted as either a subdivided ceratohyal or a ceratohyal and a hypohyal. Furthermore, nothing precludes interpretation of the osteichthyan hypohyal as a subdivision of the ceratohyal. However, the heavily subdivided visceral and branchial series of Acanthodes may be apomorphic, as there is no evidence of this type of ossification pattern in other nonacanthodiform acanthodians (Blais, Hermus, & Wilson, 2015;Brazeau, 2012;Burrow, Davidson, den Blaauwen, & Newman, 2015;Hanke, Davis, & Wilson, 2001;Hanke & Wilson, 2010).
The absence of a median basihyal in Ozarcus (Pradel et al., 2014) is anomalous and can be considered a derived state, either of that taxon or symmoriiforms more generally. Paired hypohyals in Ozarcus resemble osteichthyans, but their phylogenetic significance is somewhat more ambiguous ( Figure 5). Hypohyals also observed in the stemholocephalan Debeerius ellefseni (Grogan & Lund, 2000). They may have been gained in prior to the origin of the gnathostome crown and lost (at least) twice in chondrichthyans (minimum three steps). Alternatively, they were gained in symmoriiforms, osteichthyans, and Debeerius (also three steps). A more parsimonious distribution for these structures could be arrived at by placing Ozarcus (along with other symmoriiforms) on the holocephalan stem ( Figure 5), as has been suggested elsewhere on the basis of other lines of evidence (Coates, Gess, Finarelli, Criswell, & Tietjen, 2017;Coates & Sequeira, 2001;Giles et al., 2015).
Median, unpaired basibranchials are absent in placoderms. No examples have been identified in Acanthodes, the only acanthodian for which substantial gill skeleton ossifications are known. The chain of median basibranchials identified by Nelson (1968) are disputed by Miles (1973) and Gardiner (Gardiner, 1984). Computed tomography investigations of two other acanthodian species is currently underway by the authors and RP Dearden (Imperial College London) will help clarify this.
However, provisionally, we argue that the absence of unpaired median mineralisations posterior to the basihyal is primitive and that their origins in osteichthyans and chondrichthyans are separate. Alternatively, the basibranchial of osteichthyans is homologous to the basihyal of chondrichthyans, but has become foreshortened in elasmobranchs.

ACKNOWLEDGMENTS
The authors would like to thank Dr. Angela Nestler-Zapp (Schloßparkmuseum, Bad Kreuznach) for access to the specimen and for allowing it to be transported to the UK for study and Dr.