New information on Brindabellaspis stensioi Young, 1980, highlights morphological disparity in Early Devonian placoderms

Acid-prepared specimens of the placoderm Brindabellaspis stensioi (Early Devonian of New South Wales, Australia) revealed placoderm endocranial anatomy in unprecedented detail. Brindabellaspis has become a key taxon in discussions of early gnathostome phylogeny, and the question of placoderm monophyly versus paraphyly. The anterior orientation of the facial nerve and related hyoid arch structures in this taxon resemble fossil osteostracans (jawless vertebrates) rather than other early gnathostomes. New specimens of Brindabellaspis now reveal the previously unknown anterior region of the skull, including an exceptionally elongate premedian bone forming a long rostrum, supported by a thin extension of the postethmo-occipital unit of the braincase. Lateral overlap surfaces indicate an unusual anterior position for the jaws. Digital rendering of a synchrotron radiation scan reveals a uniquely specialized ethmoid commissure sensory canal, doubled back and fused into a midline canal. The visceral surface of the premedian bone has a plexus of perichondral bone canals. An updated skull roof reconstruction of Brindabellaspis adds to the highly variable dermal skull patterns of the probably non-monophyletic ‘acanthothoracids'. The unusual morphology revealed by the new specimens suggests that the earliest known reef fish fauna contained a diverse range of fishes with specialized ecological roles.

Acid-prepared specimens of the placoderm Brindabellaspis stensioi (Early Devonian of New South Wales, Australia) revealed placoderm endocranial anatomy in unprecedented detail. Brindabellaspis has become a key taxon in discussions of early gnathostome phylogeny, and the question of placoderm monophyly versus paraphyly. The anterior orientation of the facial nerve and related hyoid arch structures in this taxon resemble fossil osteostracans (jawless vertebrates) rather than other early gnathostomes. New specimens of Brindabellaspis now reveal the previously unknown anterior region of the skull, including an exceptionally elongate premedian bone forming a long rostrum, supported by a thin extension of the postethmo-occipital unit of the braincase. Lateral overlap surfaces indicate an unusual anterior position for the jaws. Digital rendering of a synchrotron radiation scan reveals a uniquely specialized ethmoid commissure sensory canal, doubled back and fused into a midline canal. The visceral surface of the premedian bone has a plexus of perichondral bone canals. An updated skull roof reconstruction of Brindabellaspis adds to the highly variable dermal skull patterns of the probably non-monophyletic 'acanthothoracids'. The unusual morphology revealed by the new specimens suggests that the earliest known reef fish fauna contained a diverse range of fishes with specialized ecological roles.

Introduction
The Pragian-Emsian Wee Jasper-Lake Burrinjuck limestones preserve the oldest known tropical reef gnathostome assemblage, which includes over 70 species of fossil fishes [1]. The placoderms were the dominant group in this fauna, with at least 45 species assigned to five major placoderm subgroups: arthrodires, 'acanthothoracids', petalichthyids, ptyctodontids and a rhenanid. One of the most significant discoveries from the fauna was the placoderm Brindabellaspis stensioi Young, 1980 [2], which revealed endocranial anatomy in unprecedented detail.
Brindabellaspis has variably been assigned to or compared with the 'acanthothoracids' (see §2), a poorly defined placoderm assemblage sharing characters including dorsal nasal capsules, a premedian plate and a short trunk armour. However, all of these characters are also known in other placoderm groups [3], and the monophyly of 'acanthothoracids' has been questioned [4,5].
The status of 'acanthothoracids' has important implications for the study of gnathostome evolution. 'Acanthothoracid' paraphyly suggests that some of their features represent the ancestral condition for gnathostomes. For example, the structure of the cranial cavity of Brindabellaspis has been compared to jawless vertebrates [5], as has the facial morphology of 'acanthothoracids', with their dorsal nasal capsules and an elongate trabecular region [10,20]. This facial morphology has been considered an intermediate condition between jawless and jawed vertebrate cranial anatomies [10].
A potentially confounding factor for resolving the phylogenetic position of Brindabellaspis is its highly unusual morphology compared with other placoderms. The nasal capsules are situated within the anterior cavity of the orbits, the braincase is unusually deep, and the hyoid arch attachment has an extreme anterior position, suggesting that the jaws of Brindabellaspis (unknown) were situated largely anterior to the orbits [2]. Any new information on the previously unknown anterior part of the skull is therefore important for reassessing the phylogenetic significance of Brindabellaspis.
Here, we present descriptions of new specimens that reveal the hitherto unknown rostral region of Brindabellaspis, and show that its morphology was even more unusual than initially interpreted. The pattern of dermal skull roof bones anterior to the orbits is described in detail for the first time, and we apply synchrotron X-ray tomography to investigate a unique specialization of the lateral line sensory system. The highly specialized morphology of Brindabellaspis expands the known morphological disparity of early placoderms, and shows that placoderms filled a more diverse array of ecological niches in the Early Devonian than previously thought. 1944 [23]. Other 'acanthothoracids' described prior to the publication of Brindabellaspis were Kolymaspis [24], Kosoraspis [25], Kimaspis [26], Radotina [27] and Romundina [28]. The order Rhenanida of Broili 1930 [29] originally included Asterosteus, Gemuendina and Jagorina, three flattened ray-like placoderms with enlarged pectoral fins and dorsal eyes and nares.
Young [2] originally placed Brindabellaspis within a broader 'rhenanid' grouping defined by dorsal nasal openings. Included were the Rhenanida sensu stricto of Broili [29], plus the various 'palaeacanthaspid' genera listed above. These were all united by having nares in a mid-dorsal position, compared with the more lateral position in Brindabellaspis. Embryological evidence from living groups was cited to support a ventral position for nasal openings being the primitive condition ( [2], p. 54). Denison [32] grouped the above 'palaeacanthaspids' in the order Acanthothoraci and family Palaeacanthaspidae of Stensiö 1944 [23]. Most subsequent authors have followed this usage, the informal 'acanthothoracid' replacing 'palaeacanthaspid'. Similarly, Denison [32] included the above ray-like placoderms in the order Rhenanida of Broili [29], also generally followed by subsequent authors, the informal 'rhenanid' replacing 'gemuendinid' of previous usage. Denison [32] considered the dorsal nares of rhenanids to be acquired independently of acanthothoracids.
In the same year, White [33] described a new genus Weejasperaspis from the Burrinjuck fish assemblage, placed in its own family within the order Acanthothoraci. Young [2] suggested that Brindabellaspis might be closely related to Weejasperaspis on the evidence of trunk-shield morphology. Long [31] erected a third Australian 'acanthothoracid' genus, Murrindalaspis, also placed in the family Weejasperaspidae [33] on the evidence of two similarities: the ornament, and a crest on the median dorsal plate. As the skull of Murrindalaspis and the median dorsal plate of Brindabellaspis were both unknown, whether one or the other might be closer to Weejasperaspis could not be determined on available evidence [31].
Later genera assigned to 'Acanthothoraci' include Breizosteus Goujet 1980 [34], Connemarraspis Burrow 2006 [35], Hagiangella Dupret et al. 2011 [36] and Arabosteus Olive et al. 2011 [37]. Only Arabosteus is represented by skull and braincase material that can be compared with Brindabellaspis, and it was assigned to the family Palaeacanthaspidae [23] on the basis that this was the only family (but Weejasperaspidae of White [33] was overlooked). Similarly, Early Devonian forms from the Prague Basin have been assigned to the order Acanthothoraci and family Palaeacanthaspidae [38], although previously there have been reservations [33] that Radotina and associated forms belong to 'typical' acanthothoracids.
Burrow ([35], pp. 61-62) modified Denison's [32] diagnosis of the order Acanthothoraci using characters from Goujet & Young [39]. Of various recent publications that describe or analyse 'acanthothoracids', only one provides an updated diagnosis [37]. Apart from three features (deep posterior skull embayment bounded by strongly projecting paranuchals; some skull bones separated or overlain by tesserae; ornamental tubercles commonly stellate) Brindabellaspis conforms to that diagnosis. The ornament of Arabosteus was noted to differ from typical acanthothoracids, and resemble Brindabellaspis, in lacking stellate tuberculation, but its other morphological features were considered to indicate provisional assignment to the family Palaeacanthaspidae, rather than Brindabellaspidae [22]. Thus, the order Brindabellaspida and family Brindabellaspidae at present contains only a single genus and species.
New skull material-Five new Brindabellaspis specimens provide additional evidence on skull morphology. AM F81911 (figures 1a and 2) is a partial skull and braincase, partly acid-etched from limestone, which was the basis for the skull reconstruction of Young [13];  premedian plate with underlying perichondral ossifications, broken off posteriorly at the anterior edge of the orbit and nasal cavity.

Material and methods
The type material of Young [2], and the new specimens listed above, have been partly (AM F81911, ANU V1224) or completely (ANU 49493, V2584, V3247) removed from the limestone matrix by etching in dilute acetic or formic acid, the bone strengthened with Mowital or Paraloid during extraction. ANU V3247 was scanned at the imaging and medical beam of the Australian Synchrotron facility in Melbourne. A total of 26 overlapping 3 mm subscans were imaged, covering an oblique longitudinal strip from the right anterior to the left posterior of the specimen. It was scanned with a monochromatic beam   fig. 1) was based on two specimens with the preorbital part broken away at the same level.
ANU V3247 shows the most complete premedian plate (figure 3), which was greatly elongated, with prominent overlap surfaces (figure 3c, so.os) complete on both sides (although slightly asymmetrical due to distortion). These overlap surfaces have a patchy distribution of tubercles. The anterior end is complete on the right side ( figure 3). The left side has a small broken portion (due to distortion, the left side is stretched slightly forward), which shows that the anterior margin had a small median embayment  dorsal branch was previously unknown. ANU V3247 shows the dorsal branch of each side meeting in an anastomosing plexus just beneath the overlying dermal bone, from which the median perichondral canal arises, to run forward to the anterior end of the premedian plate. The dorsal preorbital canal also gives off one large lateral branch (and many smaller anastomosing branches), the main one connecting to the inner lateral branch running forward. Previously [2], it was suggested that the dorsomesial branch of the preorbital canal may be equivalent to the ophthalmicus lateralis canal of Macropetalichthys [16], and also recently described in Shearsbyaspis [18]. However, because the supraorbital canals in Brindabellaspis terminate at the mid-level of the orbits [2], we presume that the superficial ophthalmic nerve did not continue anteriorly through the preorbital canal. Rather, the preorbital canal may have carried branches of the profundus nerve to the premedian plexus.
Distinct foramina in front of and behind the 'preorbital foramen' of Young [2], where the ventral preorbital canal opens into the lateral preorbital space (preserved on the left side of V3247; figure 3d, f.pro), also lead into larger canals joining the lateral anastomosing network. It is presumed these were branches from the structure contained within the ventral preorbital canal. The outer lateral branch arises from a foramen further forward, just inside the dermal groove beneath the posterior end of the suborbital overlap area (figure 3b,d).
In Brindabellaspis, the endocranium is preserved as a single ossification, although a double perichondral lamina within the endocranium was interpreted as the line of fusion between the rhino-capsular and postethmo-occipital units (Young ([2], fig. 4), lam.sn). This lamina is preserved at the posterior end of ANU V3247 (figure 3c, lam.sn), but it is incompletely preserved anteriorly. The synchrotron microtomography scan suggests that anteriorly the lamina breaks up into various (presumably vascular) canals.
A highly unusual aspect of the endocranium of Brindabellaspis is its pronounced lateral projection outside the lateral edge of the dermal skull roof, from about the level of the hyomandibular nerve foramen forwards. This was not understood during acid preparation of all the new specimens, so this entire region was broken away. In AM F81911, two broken perichondral laminae extending forward beneath the dermal rostrum are all that remains of this region (figure 2). The upper layer may represent the floor of the lateral preorbital space (figure 2, cav.pro). The upper perichondral lamina attaches to the inner dermal bone surface just anterior to the suture crossing the overlap area on the external surface ( figure 2). This attachment may also be preserved as a ridge in ANU V3247 (figure 3d, ridge).
Anterior to this, the premedian plate in Brindabellaspis is supported only by an anterior expansion of the postethmo-occipital unit of the endocranium, its floor preserved as a single perichondral lamina, and the overlying cartilage supporting the premedian plate evidently reduced to about 3 mm thick anteriorly ( figure 2). AM F81911 also shows that the curvature of the ventral surface of the endocranium continued forward beneath the premedian plate, following the curvature of the overlying dermal bone.

Ethmoid commissure
The premedian plate bears a median sensory line canal (preserved in AM F81911, ANU V1224, V3247), the last specimen the only one showing its forked anterior end (figure 3c, eth.com). Paired foramina, clearly visible in anterior view on the anterior margin, may indicate continuation of these sensory canals into the soft tissue of the rostrum.
ANU V3247 shows that the median sensory canal is connected to the perichondral plexus on the ventral surface of the plate via two pairs of canals (figure 5b), presumed to carry nerves. They run in a posterodorsal direction from the large midline canal in the perichondral plexus below the premedian plate, and connect to the sensory line at a slight constriction (figure 5b). At this same point, the crosssection of the sensory line changes from being obviously double anteriorly to a single fused sensory line posteriorly (figure 5a). Individual synchrotron microtomography slices at three points (figure 5c) show the transition from a double canal anteriorly to a fused single canal posteriorly. The cross-sectional area of the ethmoid commissure mask in Mimics, plotted along the anterior-posterior axis (figure 5d), shows that the anterior section (with the double canal morphology) has a much larger cross-sectional area than the posterior section (with the single canal morphology). It also clearly shows a constriction at the transition between the 'double' and 'single' sections. unpreserved in the original material. The long rostrum, with extensive overlap areas on each side, is based mainly on ANU V3247. For the post-pineal part of the skull, the new reconstruction generally follows that of Young [2]. This was based on radiographs, radiating striations on the inner surface of the holotype and assumptions about sensory grooves passing through ossification centres of skull bones (right side, figure 4a). However, ornament alignment in AM F81911 and ANU 49493 (figure 1a,c) suggests that the suture behind the pineal opening may be more V-shaped than first reconstructed. This region was previously only known from one abraded specimen (the holotype). The interpretation of bone sutures lateral and posterolateral to the orbits remains very uncertain, and alternatives as previously discussed [2] are shown on the left and right sides of figure 4c. No specimen shows evidence of an anterior suture separating the pineal from the rostral plate, so we interpret a composite rostropineal in this position. Small bones (figure 4b, ao) were previously interpreted anterior to the orbits [13], based on slightly raised areas delimited by notches in the orbital margin in AM F81911 (figure 1a, n). These bones do not have an obvious equivalent in other placoderms, and if present would be apomorphic for Brindabellaspis. However, both sides of ANU 49493 lack this elevated area of ornament. The right side of ANU V1224 shows a slight elevation similar to AM F81911. The new evidence of these specimens thus suggests that the raised ornamented area in AM F81911, variably developed or absent in other examples of Brindabellaspis, does not delimit a separate bone, but is more likely to be equivalent to elevations anterior to the orbits seen in some other placoderms, for example the petalichthyid Shearsbyaspis [18,40].
The left side of ANU 49493 shows traces of a suture crossing the overlap area in front of the orbit, and this suture is clearly visible on both sides of AM F81911 (figure 1a,c). A notch and vertical partition in AM F81911 may represent the posterior end of this suture.
A 'postnasal' element around the anterior margin of the orbit ( figure 4a,c, PN), as previously interpreted [2], would be anterior to the inferred position of nasal openings in Brindabellaspis. Good evidence for a separate postnasal element is provided by the clear overlap area for its posterior suture lateral to the orbit as preserved on the right side of ANU V1678 ([2], pl. 1, fig. 4), and a ridge is preserved in this position in the new specimens ( figure 1a-c). In AM F81911, the infraorbital sensory groove passes onto the suborbital overlap area on both sides (figure 1a), but in ANU V3247 it terminates well behind the overlap ( figure 3). This variation is shown on the left and right sides of the reconstruction (figure 4c).
The postnasal element can be assumed to connect mesially with the unpaired rostropineal plate, and thus including slight notches in the anterodorsal corner of the orbits in AM F81911 (figure 1a, n), previously interpreted to represent a nasal notch ([13], fig. 4h). The left side of AM F81911 suggests a connecting groove to the nasal cavity, which may have carried a nasal tube. This part of the orbital margin was completely unknown in the original material (badly abraded in the holotype; missing in ANU V1678). A distinct process in the left orbit of AM F81911 (less developed on the right side) delineates a separate anterior notch, now considered to be the end of the bone suture described above. The anterodorsal margin of the orbit is otherwise preserved only in ANU 49493 (both sides) and ANU V1224 (right side), where these slight notches are less distinct and variably developed.
An anterior suture to the rostropineal element, separating it from the unpaired premedian plate, is assumed by comparison with other placoderms possessing this bone. There is no indication of this suture in the new specimens, so the posterior extent of the premedian plate remains very uncertain. It can be assumed that the suture was anterior to the nasal notches, and posterior to the median sensory groove (ethmoid commissure), which we assume was confined to the premedian plate, as in other placoderms where present. Given the great rostral elongation, the rostropineal bone is shown somewhat larger than in the original reconstruction, on the assumption that it was also elongated to some extent. A comparison with the elongate rostrum in some arthrodires shows that elongation of the rostral bone also involves the pineal and preorbital bones (Carolowilhelmina [41]), elongate preorbitals only (Rolfosteus [42]) or an elongate postnasal element (Tubonasus [42]).

Comparison with previous interpretations of Brindabellaspis
The new specimens of Brindabellaspis provide information on the anterior part of the skull, demonstrating the presence of an elongate premedian plate, and the dermal bone pattern anterior to the orbits ( figure 4). The most recent previous skull reconstruction of Brindabellaspis came from a review paper [13] based on a preliminary study of AM F81911 for the preorbital region. The five new specimens now indicate a quite different dermal bone pattern, comprising elongate rostropineal and premedian bones in the midline and paired postnasal elements anterolateral to the orbits.
The suborbital plate and palatoquadrate remain unknown in Brindabellaspis, but the available material provides some clues about their configuration. The large overlap areas on either side of the rostrum are interpreted as for the suborbital plate. However, given that the overlap surface is partly covered in tubercles, it is possible that the overlaps were partly covered in tesserae (known from the skulls of various petalichthyid and 'acanthothoracid' placoderms [38,43,44]), and that the suborbital plate did not extend the full length of the premedian plate. The palatoquadrate attachments are unknown, but the cartilage underlying the anterior part of the premedian plate was very thin (figure 2), possibly too thin to support an articulation. It was thicker further posteriorly, just anterior to the lateral preorbital space (figure 2, asterisk), and so a palatoquadrate attachment may have been in this region.
A new lateral braincase restoration shows a single ethmoid articulation (figure 4d, art.eth) in this position, although we note that in Romundina three palatoquadrate attachment areas are present on the endocranium lateral to the premedian plate [10,28], even though this bone is much shorter than in Brindabellaspis.
Three other articular surfaces are known on the lateral braincase surface of Brindabellaspis, the most posterior (figure 4d, art.p) probably for a gill arch, and another (art.op) interpreted as for an opercular cartilage [2]. The hyoid arch articulation (art.hy), as previously interpreted [2], lies anteroventral to the hyomandibular nerve opening, in an unusually anterior position. Gardiner [30] preferred to interpret this as a palatoquadrate articulation, and the more posterior articulation (the opercular cartilage articulation of Young) as the hyomandibular articulation. However, new evidence from early arthrodires [45], and Romundina [17], demonstrates separate opercular and hyomandibular articulations associated with the hyomandibular nerve foramen, and no contact with the palatoquadrate.

Comparisons with other placoderms
As interpreted (figure 4c), the skull roof pattern of Brindabellaspis is unique. The large rostropineal plate is sutured firmly to the rest of the skull roof, which shows unique rostral elongation resulting from its long premedian plate. This places the centre of the orbits in the posterior half of the skull roof length (about 42% of skull length from the posterior margin). Other placoderms with orbits enclosed in the skull roof have orbits in a more anterior position, even when a pronounced rostrum is developed. Thus, in Wuttagoonaspis (which lacks a premedian plate) the orbits are 33-45% of skull length from the anterior margin [46], and in petalichthyids (e.g. Macropetalichthys) this is about 30% (accentuated by the nuchal region being more elongate in petalichthyids than in Wuttagoonaspis or Brindabellaspis). Also unique is the position of the nasal capsules within the orbits, in contrast with other acanthothoracids (as represented by Romundina) where the nasal openings occupy the space between the rostral capsule and the premedian plate [10,17,47].
The unusual skull roof pattern is also shown in the position of the postnasal element, around the anterior margin of the orbit. This would be anterior to the nasal openings of Brindabellaspis, presumably located within the orbits as previously interpreted [2]. In Radotina, a postnasal element has been restored lateral to the nasal opening [44], but mesial to the orbit, whereas in Brindabellaspis the equivalent element is lateral and anterolateral to the orbit. The postnasal element shows notches in its orbital margin in some specimens. If representing nasal openings, this would be comparable to the notched postnasal of brachythoracid arthrodires (e.g. [48]). Possibly the nasal opening was bounded laterally by a dermal process of the sclerotic capsule (unknown for Brindabellaspis), as in the isolated weejasperaspid sclerotic capsule previously described [49]. This structure is also seen in early brachythoracids [45] and other more distantly related placoderm taxa such as antiarchs [50] and Entelognathus [9]. Possibly both incurrent and excurrent nasal openings were dorsal in position for Brindabellaspis, because the floor of the orbit and preorbital space probably occluded any ventral passage.
The braincase of Brindabellaspis may be compared with that of Romundina, and the proportions differ greatly due to the rostral elongation in Brindabellaspis. As previously shown [2], the division of the braincase into separately ossified postethmo-occipital and rhinocapsular units is still evident in Brindabellaspis, with these units fused together and the division represented by a double perichondral lamina (the subnasal lamina; figure 3a,c, lam.sn) within the braincase. The skull roof is also consolidated, showing no trace of division into a separate rostral capsule, and the nasal capsules are in a more lateral position within the orbits. In Romundina, the rhinocapsular unit is separately ossified, and posteriorly placed between the orbits. The postethmo-occipital unit (specifically the trabecular region) extends anteriorly and underlies the premedian plate [10]. In Brindabellaspis, the subnasal lamina evidently terminated adjacent to the posterior end of the premedian plate, and the entire premedian plate is underlain by a thin extension of the postethmo-occipital part of the braincase. Thus, the trabecular region in Brindabellaspis would have an even larger preorbital extension than in Romundina.
Clearly, the morphology of Brindabellaspis is quite different from that of other 'acanthothoracids'. Dermal skull roof patterns were already recorded to be highly variable within that assemblage, sometimes including zones of tesserae between the larger bones [44]. Dermal bone patterns have also been found to vary intraspecifically [37], although similar variation in 'Radotina' has been explained by previously unknown species diversity [38]. This variability in dermal bone pattern might be taken as additional evidence that 'acanthothoracids' may not be monophyletic, although dermal skull roof characters are often difficult to polarize.
Comparing the morphology of Brindabellaspis with that of Romundina, obvious differences can be explained by two possibly related changes: great rostral elongation of the ethmoid region and the consolidation of the rostral capsule with the rest of the skull. In Romundina, the nasal capsules occupy the space between the rostral capsule and the premedian plate. However, in Brindabellaspis, the rostral is firmly fused to the premedian, which would be possible only with a more lateral position for the nasal capsules within the orbits. Similarly the comparatively long premedian plate, the extensive underlying trabecular region and the anterior position of the hyoid arch attachment are all likely to be related: these features have essentially the same relative positions as in Romundina, but the whole ethmoid region is stretched anteriorly in comparison.
Another similarity with Romundina is the plexus of perichondral canals underlying the premedian plate. Possibly such a plexus was a common feature of all placoderms with a premedian plate, with the lack of braincase ossification obscuring its presence in antiarchs. The base of the premedian plate in the antiarch Bothriolepis shows significant porosity ( [51], pl. 57), which could be related to a similar plexus. In both Romundina and Brindabellaspis, the plexus comprises larger canals running forward in the midline,  fig. 2B2). However, it is difficult to say on present evidence whether these similarities could be characters linking all 'acanthothoracids', due to the absence of an equivalent degree of preservation in most other placoderms.

Unique adaptation to the sensory line system
Brindabellaspis has a sensory canal on the midline of the premedian plate. An anteroposteriorly directed sensory canal in this position is, to our knowledge, unknown in any other vertebrate. An anterior transverse ethmoid commissure is, however, present in a wide variety of gnathostomes, including on the premedian plate of various 'acanthothoracid' and antiarch placoderms. The fork at the anterior end of the midline canal in Brindabellaspis suggests that this midline canal represents the ethmoid commissure, which has folded back on itself and fused in the midline.
The two pairs of nerve canals that enter the ethmoid commissure either side of the constriction may have innervated a neuromast organ. In some living species, neuromast organs occur at constrictions in the sensory canal [52], which may increase sensitivity by amplifying particle motions within the canal [53]. As there are two pairs of nerve canals in Brindabellaspis (shown by ANU V3247), a possible interpretation is that the ethmoid commissure has fused at the point of two neuromasts that would normally be positioned either side of the midline in a transverse ethmoid commissure.
The ethmoid commissure is typically innervated by the buccal nerve in living species [54,55], and in Romundina it was suggested that the buccal nerve entered the lateral ends of a transverse neurovascular web underlying the ethmoid commissure [47]. The buccal nerve is associated with the maxillary branch of the trigeminal, which in Brindabellaspis ran through the preorbital space. It may have entered the perichondral plexus from the anterior to innervate the ethmoid commissure, doubling back to follow the path of the ethmoid commissure itself (electronic supplementary material, figure S1). This would be consistent with the posterodorsal orientation of the nerve canals, which suggests that the nerve fibres within entered the plexus anteriorly.

Ecological role of Brindabellaspis
The unusual morphology of Brindabellaspis indicates a specialized role, but inferences about the biology of Brindabellaspis are somewhat limited without preservation of the jaws. The dorsolaterally positioned eyes of Brindabellaspis are consistent with a benthic niche, and the rostrum formed by the premedian plate may have functioned in the detection of bottom-dwelling prey.
The elaborate plexus of perichondral canals underlying the premedian plate might also be suggestive of a sensory function for the elongate rostral region of Brindabellaspis. Our current hypothesis is that the tip of the rostrum of Brindabellaspis was continued as soft tissue, which could have housed a sensory system. However, the nature of such a sensory system, and the precise functional significance of the perichondral plexus, is unclear. Lungfish also possess a system of perichondral tubes in the cartilage of the snout [56][57][58], although this forms an upwardly branching system rather than a horizontal plexus. The function of the lungfish tubuli has been a matter of debate, with competing hypotheses that they housed nerves, blood vessels or lymphatics [58][59][60][61]. Recently, lungfish rostral tubuli have been suggested to supply nerves to electroreceptors covering the snout [62]. One potential function of the perichondral plexus for Brindabellaspis was to supply nerves to electroreceptors on the snout. A possible analogue is the paddlefish Polyodon, which uses its rostral paddle as an antenna to seek out plankton, aided by a dense array of electroreceptors on the underside [63]. Similarly, the rostrum of shovelnose rays has dense electroreceptors on the ventral surface for prey detection [64]. Without further evidence, this hypothesis remains largely speculative for Brindabellaspis, but possibly the irregular 'paddle-like' lateral expansion of the rostrum (shown by ANU V1224) may be consistent with that idea.
An alternative is that the plexus played no special functional role in Brindabellaspis. In placoderms, such a plexus may be a common feature where endocranial cartilage meets dermal bone over an extended area. For example, the arthrodire Goodradigbeeon, also from Burrinjuck, has an extensive internasal wall between the left and right nasal capsules [33]. The roof of the internasal wall, where the cartilage of the rhinocapsular ossification contacts the overlying rostropineal plate, has a plexus of perichondral canals (electronic supplementary material, figure S2). Extensive canal networks are also found at the interface of dermal and perichondral bone throughout the skull roof in Romundina [17].
Our new information on Brindabellaspis shows that this placoderm, already known to have unusual morphology, was even more specialized than previously thought. There is now good evidence that during the Devonian Period reef ecosystems were, as today, major centres for biodiversity [1,65]. Beyond species diversity, Brindabellaspis provides evidence for disparate body forms in the Taemas-Wee Jasper fossil fish fauna, the earliest known example of a diverse coral reef fish assemblage. Disparate body forms are also known from the Late Devonian Gogo Formation, another highly diverse reef assemblage [65]. At Gogo, the long-snouted lungfish Griphognathus [57] may have filled a similar ecological niche to that of the placoderm Brindabellaspis in the Early Devonian.
Data accessibility. Ply files and a three-dimensional PDF of the digital models are available from the Dryad Digital Repository [66]. Scan data are available via e-researchSA (https://doi.org/10.4226/86/5a05055b47a98).