and the classification of placoderm fishes [PDF]

zoological Journal of the Linnean Society (1986), 86: 43-74. With 5 figures

Observations on Ctenurella (Ptyctodontida) and the classification of placoderm fishes PETER L. FOREY Department of Palaeontology, British Museum (Natural History), London S W7 5 B D

AND BRIAN G. GARDINER Sir John Atkins Laboratories, Biology Department, King’s College, Campden Hill Road, London W8 7AH Received June 1984, accepted for publication NoNovember 1984

The structure of the hyoid arch of the ptyctodont Ctenurella is described and discussed with reference to theories ofjaw suspension in placoderm fishes. It is concluded that primitive placoderms had a modified hyoid arch hut that the hyomandibular took no direct role in supporting the jaws. The relationships of ptyctodonts are discussed and it is concluded that they are placoderm fishes. Several different classifications of placoderrn fishes are evaluated and are shown to be weakly based, chiefly because of lack of precise knowledge of character distribution. An attempt is made to produce a classification by using a simple cladistic computer analysis. The result highlights homoplasy in character distribution amongst traditionally recognized placoderni groups.

KEY WORDS:--Ptyctodontida - jaw suspension classification - cladistics - Devonian.

-

ptyctodont relationships

placoderm

~

CONTENTS Introduction . . . . . . . . . . . Visceral arch anatomy and jaw suspension in ptyctodonts Relationships of ptyctodonts . . . . . . . Classification of placoderms . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . References.

. .

.

. . .

. .

.

. . .

. .

.

. . .

. .

.

. . .

.

. .

. . .

.

. .

. . .

.

. .

. . .

.

. .

. . .

43 45 53 57 72 72

INTRODUCTION

The classification of placoderm fishes has received renewed interest during the last decade, catalysed by the works of Denison (1975) and Miles & Young (1977). These authors broke away from an erstwhile mould where individual placoderm subgroups were considered in isolation, and instead dealt with Originally submitted as a paper for Professor K. A. Kermack F.L.S. (rest published in Vol. 82, Nos 1 & 2).

+

0024-4082/86/010043 32 $03.00/0

43

0 1986 The Linnean

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

Society of London

44

1’. L. FOKEY AND B G . GARDINER

placoderm fishes as a whole. Denison’s survey concentrated on the anatomy of the pectoral girdle and concluded that a short shoulder girdle was primitive for placoderms (contra Gross, 1954; Westoll, 1945). O n this basis Denison produced an evolutionary tree suggesting that groups such as Ptyctodon tida, Gemuendinida and Pseudopetalichthyida were primitive placoderm fishcs. Miles & Young (1977) agreed, but went further by acknowledging that, for progress to be made, classifications should be explicit statements open to criticism. Thus, they rephrased Denison’s evolutionary tree as a cladogram, criticized the non-parsimonious consequences, and produced their own classification. The classifications of Denison and Miles & Young agree for the better known placoderms in considering Arthrodira (including Wuttagoonaspis), Antiarcha and Phyllolepis to belong to a group, Dolichothoracomorpha. But differences between these classifications, and between those subsequently produced (Denison, 1978; Young, 1980; Gardiner, 1984a) affect the groupings of less derived placoderms (Petalichthyida, Palaeacanthaspida, ptyctodonts, gemuendinids and pseudopetalichthyids) . As usual the differences stem from weighting of characters or assessment of character states. Denison (1975, 1978, 1983) assumes that the tesserate condition is primitive, hence ptyctodonts (non-tesserate) are cladistically more derived than gemuendinids or pseudo-petalichthyids. Miles & Young (1977), however, considered that the tesserate condition was secondary and that the possession of pelvic claspers by ptyctodonts, a feature shared with chondrichthyans but not known in other placoderms, specifies ptyctodonts as the most primitive placoderms. Ptyctodonts have therefore assumed some importance in discussions of placoderm classification, and this is one of the reasons why we chose to look at ptyctodonts again. The dispute between Denison and Miles & Young reflects a fragility in placoderm classification which is underlined by Young ( 1980). Young proposed three different cladograms of placoderms (Young, 1980: fig. 27). Young regarded these as equally parsimonious alternatives and any preference depended on a choice between assumptions which have questionable empirical support. Young deferred decision and concluded that we need more information in certain character complexes, among which he mentioned (1980: 69) “. . . the relationship between the opercular cartilage, hyomandibula and submarginal plates in different placoderm groups, and the homologisation of their points of articulation to the endocranium relative to the position of the hyomandibular nerve foramen. . .” Since Miles & Young (1977) completed their study of ptyctodonts we have had the opportunity to examine a further acid-prepared specimen of Ctenurella gardineri Miles & Young which, we believe, shows evidence of a hyomandibular. This allows us to clarify some points of anatomy and to comment on some details in Young’s research. T o do this we begin with some new observations on Ctenurella and consider jaw suspension in placoderms. We follow this with an assessment of a still-held belief ((arvig, 1960, 1962, 1980b; Westoll, 1978) that ptyctodonts and Holocephali are immediately related. Finally, we review the various classifications which have been proposed for placoderm fishes, and propose our own. Our initial assumptions are: ( 1 ) that placoderms are a monophyletic group; and (2) that placoderms are the sister-group of Osteichthyes. Monophyly is

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENJRELLA AND PLACODERM FISHES

45

evidenced by a median dorsal plate in the shoulder girdle. O n this basis Stensioella cannot be shown to be a placoderm (see also Miles & Young, 1977: 135) and for most of our discussion we disregard it. Pseudopetalichthyids (Pseudopetalichthys and Paraplesiobatis) are included despite the fact that there are problems with the identification of a median dorsal plate (p. 62 and Westoll, 1967). We can muster no good reason for considering pseudopetalichthyids as placoderms but appeal to convention and belief that the pattern of sensory canals converging on the nuchal area and the large single cheek bone (submarginal) are sufficiently similar to those of undoubted placoderms. The hypothesis that placoderms and osteichthyans are sister-groups has most recently been discussed by Forey (1980) and especially Gardiner (1984a, b). Some of the more important synapomorphies are: neurocranium protected by a series of large dermal plates, some of which have descending laminae; a true dermal shoulder girdle with lateral plates and a denticulated postbranchial lamina; supracoracoid foramen; parasphenoid with teeth and bearing a spiracular groove and pierced by the buccohypophysial canal; urohyal. Most of the specimens used in the preparation of this paper belong to the British Museum (Natural History) and are referred to by register number with prefix P. The remainder belong to the Royal Scottish Museum, Edinburgh and are prefixed with RSM. VISCERAL ARCH ANATOMY AND JAW SUSPENSION IN PTYCTODONTS

I n this section we describe and discuss the hyoid arch of Ctenurella and follow this with a more general discussion of jaw suspension in placoderms. Previous ideas of ptyctodont jaw suspension have been influenced by the holocephalan model. There is general agreement that ptyctodonts are autostylic (endocranio-autostylic, Stensio, 1963), but there is less agreement on whether this is a primary or secondary condition. Resolution of this question depends in part on whether the hyoid arch is modified or whether i t looks like the succeeding branchial arches. Watson (1938: fig. 1) considered that the hyoid arch was unmodified and he identified a series of pharyngo-, epi- and ceratohyal elements closely applied to the posterior margin of the palate and lower jaw of Rhamphodopsis trispinatus Watson. These identifications have been subject to doubt. Miles (1967: 105) was unable to recognize the ceratohyal; Miles & Young (1977: 178) had “no confidence in these determinations”; and Holmgren (1942: fig. 36B) reinterpreted Watson’s hyoid arch elements as labial cartilages. We have examined casts of Watson’s key specimen (the original is a natural mould) and remain unconvinced that the structures are hyoid elements. They appear to be the edges of the palatoquadrate elements pushed through from the opposite side (see also Stensio, 1969: 422). The evidence for a modified arch is based on the supposed occurrence of a hyomandibular in Rhynchodontus eximius Jaekel. The element in question is one of several scattered head bones in the holotype (Jaekel, 1919: fig. 16, Q and has been described by Stensio (1969: 422, fig. 164) who includes it in a restoration of Ctenurella gladbachensis Brvig (Stensio, 1969: fig. 40A). This ‘hyomandibular’ is shown in an upright position stretching between the head of the submarginal and the jaw articulation. Stensio does not believe, however, that this

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

P. L. FOREY AND B. G . GARDINER

46

‘hyomandibular’ is suspensory. Miles ( 197 1 : 197) has questioned Stensio’s identification and prefers to interpret the element as a palatoquadrate bone, possibly the quadrate. We have only examined casts of Jaekel’s material (the originals are natural moulds) and so we are not in a position confidently to confirm or deny Stensio’s determination or description. However, we note that the overall shape of the ‘hyomandibular’ is very similar to that of a rostra1 cartilage of Ctenurella (Fig. l E , F). Furthermore, we cannot find evidence of the foramina said to be for V I I hm by Stensio. Parenthetically, we note that Stensio (1969: fig. 164) restores the hyomandibular nerve both entering and leaving the lateral face of the ‘hyomandibular’ and this is most unusual among vertebrates. In sum, we remain sceptical about the hitherto published evidence for a hyomandibular in p t yctodonts. Descriptions of Ctenurella gladbachensis (Brvig, 1960, 1962) and C. gardineri (Miles & Young, 1977) have included incidental references to the hyoid and branchial skeleton but little attempt has been made to restore constituent parts. Hm a.h

A

\

D

C

-

0

Y

Figure 1. A-E, G , H. Ctenurellu gurdineri Miles & Young, A, Left submarginal in lateral view, P. 61507. B, Right submarginal and attached hyomandibular in mesial view, P. 61507. C, Left ceratohyal in lateral view, P. 50906h. D, Right ceratohyal in medial view, P. 50906h. E, Rostra1 ossification as preserved in P. 57637, orientation unknown. G. Basihyal in ventral view, P. 50906. H, Urohyal in ventral view, P. 50906. F, Rynchodontus exirnius Jaekel, element figured by Jaekel (1919: fig. 16, Q!, and interpreted by Stensio (1959) as a hyomandihular. Drawn from a latex cast, P. 15337. a.h, Open articular head; Hm, hyomandihular; pr, closed process. Scale bars 3 mm.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

47

Acid-prepared material of C. gardineri includes several well preserved, perichondrally ossified elements clearly belonging to the hyoid and branchial arches but in all specimens these are scattered without clear indication of topographic interrelationship. Material of C. gladbachensis also shows visceral elements which, although poorly preserved, appear to have retained life positions. Therefore it seems worthwhile to combine information from these two sources in an attempt partially to reconstruct the visceral arches, particularly in the light of varying views of ptyctodont jaw suspension. We have identified three elements of the hyoid arch: a hyomandibular, a ceratohyal and a basihyal. T h e basihyal (Fig. 1G) is a small pyramid with an open base orientated posteriorly and which presumably passed into cartilage. It resembles the (cartilaginous) basihyal of Rhinochimaera pacijica as figured by Garman (1904: pl. 12, fig. 3) and also that bone in some teleosts. The ceratohyal (Fig. lC, D) is preserved in three specimens of C. gardineri (P. 57637, P. 50906, P. 61507) where it is considerably larger than any other branchial element. Several specimens of C. gladbachensis show the ceratohyal in position (P. 48253-4, P. 48255-6, P. 53343, P. 53345-6) where it lies close behind the articular and extends posteriorly and slightly dorsally. Brvig (1960: 322) recognized these “certain stout ossifications” as ceratohyals but unfortunately chose to illustrate (1960: pl. 29, fig. 1, Chy?) a probable pectoral basal element (Miles & Young, 1977: 179). Specimens of C. gladbachensis (e.g. P. 53344, Fig. 2) also show that the larger end of the ceratohyal lies anteriorly, and this allows us to orientate the bone in the reconstruction of C. gardineri (Fig. 3). The ceratohyal is slender and about half as long as the submarginal and has single posterior and double anterior heads. I n cross-section it is oval. The ventrolateral surface is constricted where a small ridge is developed. This ridge is roughened and probably represents the surface for the origin of the interhyoideus muscles. The posterior and larger of the two anterior heads is open (cartilage-filled in life) while the smaller anterior head is closed. This smaller head is directed ventrolaterally and appears to lie adjacent to a circular socket on the posteromesial surface of the articular described by Miles & Young (1977: fig. 24E). However, the shapes of the ceratohyal head and the articular socket do not match exactly so it is unlikely that there was a ball-and-socket articulation between them; rather, there was probably a ligament connection. Thus restored, we suggest that the ceratohyal lay wholly behind the mandibular arch and that there was firm connection between the anterior end of the ceratohyal and the articular. The former condition is to be expected in a fish in which the jaws are small and anteriorly placed. The second condition is rare among Recent fishes where the ceratohyal-mandibular ligament originates over the posterior half of the ceratohyal and inserts o n the lower jaw over a large area. An exception is Lepisosteus where the ceratohyal-mandibular ligament inserts in a pit in the retroarticular (Wiley, 1976: fig. 9). I n P. 61507 there is a very small curved ossification with bevelled open ends, lying immediately anterior to the open head of the ceratohyal. It is possible that this is a hypobranchial or hypohyal element. It has not been recognized in any other specimen of either species of Ctenurella. The dorsal part of the hyoid arch is represented as a thin perichondral shell attached to the inside of the head of the submarginal (Fig. 1B). It is well preserved only in one specimen of C. gardineri (P. 61507), but at least one

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

48

P. L. FOKEY AND B. G. GARDINER orb

-

art

I

chy

Figure 2. Ctenurella gladbachenJis (arvig. Drawings of parts of two specimens (upper, P. 53343; lowcr, P. 53344) to show typical preservation relationships between toothplates (hatched), palatoquadrate elements, ceratohyal and submarginal. Endoskeletal elements are stippled. Scale bars 3 mm. art, Articular; aup, autopalatine; C, central; cb, ceratobranchial; chy, ceratohyal; eb, epibranchial; etli, ethmoid ossification; M, marginal; mpt, metapterygoid; Nu, nuchal; orb, orbital ossifications; PNu, paranuchal; P r o , preorbital; PtO, postorbital; qu, quadrate; ros, rostra1 ossifications; SM, submarginal; suffixes I and r, bones of left and right sides.

specimen of C. gladbachensis (P. 53344) also shows a complex head of the submarginal, suggesting that similar conditions existed in that species. The anterior end of this hyoid element is open (cartilage-capped in life) and there is a small process directed anteroventrally and slightly mesially. T h e posterior end is open so it must have continued in cartilage but for how far it is difficult to say. The posterior end is somewhat flattened, which might suggest that the element did not extend much beyond the limit of perichondral ossification. O n the other hand, the flattening may simply reflect a surface contour, real or preservational, in an otherwise long element. I n all, this perichondral structure looks very similar to the hyomandibular of Coccosteus sp. (Miles, 197 1: figs 110, 11 1). That this element in Ctenurella is a hyomandibular, rather than the upper part ofan unmodified branchial arch (cf. Watson, 1938; Miles, 1967), is suggested by

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

qu

I

art

49

\

chy

Figure 3. Restoration of Ctenurella gardineri Miles & Young showing conjectured position of hyoid arch. Abbreviations as in Figs 1, 2.

two observations. First, the head is complex as would be expected in an articulatory hyomandibular. Secondly, it is associated with an anamestic bone correctly identified as a submarginal by Miles & Young (1977). Thus it has the same relations to the dermal covering as in arthrodires (Miles, 1969, 1971) and in primitive actinopterygians (Allis, 1922; Gardiner, 1984a, b). In arthrodires there is a prominent groove on the visceral surface of the submarginal which marks the position of the hyomandibular (Denison, 1958; Miles & Westoll, 1968; Goujet, 1972, 1973; Mark-Kurik, 1973), although the only clear evidence of the hyomandibular is seen in Coccosteus sp. (Miles, 1971), Dicksonosteus (Goujet, 1972, 1975) and possibly Leiosteus concams Gross (Stensio, 1934: pl. 13, fig. 1). Since the submarginal lies horizontally or subhorizontally in C. gludbachensis it is probable that the hyomandibular curved away from the submarginal to meet the ceratohyal and it has been restored in this fashion in Fig. 3. I t is also likely that the ptyctodont hyomandibular was considerably shorter than the overlying submarginal. I n this description we have identified the perichondral element associated with the submarginal as the hyomandibular. But there is an alternative interpretation which demands consideration: that the endoskeletal element is an opercular cartilage. Among placoderms a perichondrally ossified opercular cartilage has been identified in Holonema (Miles, 1971: fig. 36B), Jagorina (Stensio, 1963) and Brindabellaspis (Young, 1980: fig. 17) where, in all cases, it is associated with a dermal bone of the cheek, the submarginal. In Holonema and Brindabellaspis the opercular cartilage is small and irregularly shaped with illdefined margins and spreads over the visceral surface of the dermal covering. It is perforated by many small canals. In Jugorina the opercular cartilage is reputedly large and perforated dorsally by many canals (Stensio, 1963: fig. 78D). In none of these instances is there any trace of a n articular head. 4

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

50

P. L. FOREY AND B. G . GARDINER

Despite this, Young (1980: fig. 8, art. OF) interprets a facet behind the hyomandibular foramen of Brindabellasjis as a point of articulation of the opercular cartilage on the braincase. We disagree with this interpretation because, like Miles (1971: 198), we note that the opercular cartilage of Recent vertebrates never articulates with the braincase. Rather, we consider that the facet in Brindabellaspis received the hyomandibular (unknown in this form) and suggest that the elongate facet lying beneath the hyomandibular foramen (Young, 1980: fig. 8, art. hm) is a basal articulation between the palate and braincase. The hyomandibular of Clenurella, like that of Coccosteus, differs from a n opercular cartilage in that it is compact, has a distinct articulatory head and is not perforated by canals. In the description of the hyomandibular of Ctenurella given above we pointed out that it is closely associated with a dermal bone. It is therefore pertinent to ask if this association holds good for all placoderms and if such an association might help discriminate between a hyomandibular and an opercular cartilage, bearing in mind that in primitive actinopterygians (Pohjterus) the hyomandibular is associated with a dermohyal and the opercular cartilage with an opercular (Gardiner, 198413). Unfortunately, we find that the endoskeletalldermal bone association in placoderms is perplexing. The cheek in the arthrodires Holonema and Coccosteus is constructed in similar fashion with similar mutual relationships between bones. However, in Holonema the dermal submarginal is associated with a n opercular cartilage while in Coccosteus and [Bicksonosteus] (Goujet, 1971: pl. fig. 3), it is associated with the hyomandibular. We do not doubt the homology of this dermal bone between arthrodires (cf. Young, 1980: 46) and so acknowledge that two different endoskeletal elements may be associated with the same dermal bone in arthrodires. In rhenanids (sensu Young, 1980) the identity of the dermal bone associated with the opercular cartilage is open to question. I n Brindabelluspis the dermal bone (which is incompletely preserved in the only specimen showing this area) with attached opercular cartilage has been found in only one specimen (Young, 1980: fig. 17). In Jagorina the associated dermal bone is the only large bone in the cheek, otherwise represented by tesserae. Both dermal bones in these ‘rhenanids’ are ovoid, like the submarginal in Holonema and primitive arthrodires, and are not marked by any canal or pitline. Therefore, there is no reason to doubt homology between these bones, particularly as the submarginal (operculum of Jagorina, extralateral of antiarchs) occupies the same position relative to the skull roof and the eye. I n Jagorina the hyomandibular* lies immediately in front of the submarginal/opercular cartilage. We are therefore forced to accept that, among placoderms in general as among arthrodires in particular, two endoskeletal structures may be associated with the same dermal bone but we know of no placoderm in which this condition exists. It follows that *Goujet (1982) has suggested that the hyomandibular of Jagorina is a hypertrophied postorbital process. Dr Goujrt has recently shown one of us (P.L.F.) unretouched photographs ofjagorina in which it is clear that the so-called hyomandibular of either side is differently preserved. Furthermore there is no clear evidence of an articulatory head on either the left or right ‘hyomandibular’. Instead Goujet’s photographs show a n irregular unossified area between the ‘hyomandibular’ and the main body of the neurocranium. We have not examined original material of Jugorina and must therefore remain undecided about the identification of the ‘hyomandibular’ in Jagorina.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELL.4 AND PLACODERM FISHES

51

we are unable to use dermal bone association to identify the underlying endoskeletal element. Our interpretation of Ctenurella suggests that the hyoid arch is well developed. In restored life position, the hyomandibular must have articulated with the braincase beneath the eye. From there it swung posteroventrally and articulated with the ceratohyal, the latter reaching forward to end at the level of the jaw articulation. Restored in this manner it resembles the restoration of the hyoid arch of Coccosteus cuspidatus by Miles (1969: fig, 7C) and Miles’ “moderately advanced arthrodire” (197 1: fig. 1 12C) in that the hyomandibular lies free from the palatoquadrate, and the hyomandibular is posteriorly directed. T h e unusual position of the hyomandibular articulation beneath the eye is not known in arthrodires but is seen in Brindabellaspis and is approached in Macropetalichthys. The anterior position of the hyomandibular articulation means that the palatoquadrate, and especially the metapterygoid, is probably more anteriorly placed than is shown in Brvig’s (1960, 1962) restorations. We note that in better preserved specimens of Ctenurella gladbachensis examined here (e.g. P. 53343, Fig. 2) the metapterygoid is found in front of or at the anteroventral corner of the eye. The metapterygoid has been restored in this position in Fig. 3. The remains of at least three branchial arches are also preserved in Clenurella and these probably represent cerato- and epibranchials. Specimens of C. gladbachensis suggest that the branchial arches lay wholly beneath the skull and that at least the first three arches lay medial to the hyoid arch. As in chondrichthyans, the branchial arches of Ctenurella do not bear a posterior groove for the branchial arteries as they do in acanthodians and osteichthyans. We may now broaden this discussion to include placoderm jaw suspension in general. Miles (1971: 201) specified four primitive conditions: ( 1 ) a continuous palatoquadrate which ossified in a single piece; (2) palatoquadrate with a single orbital connection with the neurocranium; (3) palatoquadrate free from the dermal skeleton; and (4)no hyomandibular, jaw suspension entirely through the palatoquadrate (autodiastylic). We agree with point number 3, noting that the derived condition, a tight union between palatoquadrate and suborbital, occurs in arthrodires, antiarchs (where the quadrate is firmly attached to the infraprelateral in Bothriolepis sp.-P. 50898) and Romundina (Brvig, 1975: pl. 2, fig. 5). We are less confident about the remaining three suggestions. I t is probable that the placoderm palatoquadrate primitively ossified from more than one centre. Among placoderms there are three bones in the ptyctodont palatoquadrate (autopalatine, metapterygoid and quadrate) and two in the palatoquadrate of brachythoracids (autopalatine and quadrate) which may or may not have beenjoined by cartilage (Miles, 1969, 1971). Also, it is difficult to imagine how the elongate and complex-shaped palatoquadrate of Jagorina (Stensio, 1969: fig. 155) could have ossified from a single centre. T h e palatoquadrate of acanthodians and osteichthyans also primitively ossified from three centres (Miles, 1973b; Gardiner, 1984b). T o sum this information we suggest that the primitive condition in placoderms is a palatoquadrate ossifying from three centres, with a reduction to two centres in brachythoracids (at least) representing a derived condition, perhaps associated with the union of palatoquadrate and the dermal cheek (Miles, 1969). Interpretation of the relationship between palate and braincase in placoderms

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

52

P. L. FOREY AND B. G. GARDINER

has always proved difficult because there are so few forms in which both palate and braincase are known and because of the differing interpretations of neurocranial facets and the processes of the palatoquadrate. We are not here concerned with terminology of palatoquadrate processes (whether any particular process should be called an ethmoid, orbital, basal or otic process; see Maisey, 1980; Gardiner, 1984b), but rather with establishing whether the palatoquadrate contacts the braincase over one or more areas. According to Stensio’s ( 1969) interpretation of Jagorina, the palatoquadrate does not apparently contact the braincase; an unusual condition. However, if we accept Gross’ (1963) reinterpretation of Jagorina (in which the antorbital process of Stensio is identified as the palatoquadrate) and his interpretation of the closely related Gemuendina (Gross, 1963: fig. 4A) then the palatoquadrate articulates with the braincase in front of the orbit. In ptyctodonts ((arvig, 1960; Miles & Young, 1977) there is a well developed contact between the autopalatine and the ethmoid ossification anterior to the orbit and probably another at the anteroventral corner of the orbit since the posterior limb of the metapterygoid bears complex relief mesially (Miles & Young, 1977: fig. 26C). In other placoderms which are reasonably well known (Brindabellaspis, Romundina, Macropetalichthys, Kosoraspis, Radotina, Kudjanowiaspis, Dicksonosteus and Buchanosteus) there is a facet on the braincase at the level of the nasal capsule or the postnasal wall. Thus, this anterior facet (the orbital articulation of Miles, 1971) is very widespread and probably primitive for placoderms. In some of these placoderms there is another more posterior facet located anterior to the hyomandibular foramen, or beneath the eye in Buchanosteus (Young, 1979: fig. 4). This posterior facet is absent in Dicksonosteus and the relevant area of the braincase is insufficiently known in Kosoraspis and Romundina. Furthermore, in the brachythoracids Tapinosteus and Pholidosteus, Stensio (1963: figs 46A, 75) illustrates a prominent posterior articulation which he likens to the basal process of Recent fishes. Stensio (1963, 1969) has interpreted all these posterior facets as articulation points for the palatoquadrate. Miles (197 1), however, interpreted these facets as the hyomandibular articulation, a view based on Jagorina where the hyomandibular certainly articulated against the braincase anterior to the hyomandibular foramen (according to Stensio (1950: fig. 7) it virtually straddled the hyomandibular foramen). Miles suggested that the additional facet behind the hyomandibular foramen was the site of the pharyngobranchial articulation and Young suggested that it may have received the opercular cartilage in Brindabellaspis. We do not agree with Young (p. 50) and we are sceptical about Miles’ suggestion (Gardiner, l984a). In those placoderms in which there are facets on both sides of the hyomandibular foramen the posterior is either much larger than the anterior (Brindabellaspis, Kudjanoteriaspis, Tapinosteus and Pholidosteus) or about the same size (Macropetalichthys), Pharyngobranchials are unknown in placoderms, but in osteichthyans the neurocranial facets receiving these elements are very small compared to the hyomandibular facet. We also note that Goujet (1975) showed that the hyomandibular facet of some arthrodires is located behind the hyomandibular foramen and in others it is in front (see also Young, 1979; Stensio, 1969). We acknowledge that the data are very incomplete but prefer to regard the posterior facet as for the hyomandibular and the anterior as for the

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

C'TENURELLA AND PLACODERM FISHES

53

palatoquadrate. This is the generalized gnathostome condition and it follows that we believe that the placoderm palatoquadrate primitively articulated with the braincase at two points (see also Maisey (1980) for elasmobranchs and Gardiner (1984b) for osteichthyans). T h e relation between the hyomandibular facet and nerve in Jagorina could be regarded as specialized, as is the apparent lack of any articulation between the braincase and palatoquadrate (Stensio, 1969)". We are tempted to suggest that, in those forms in which a single facet is present, the differing relations to the hyomandibular nerve foramen may have little significance. The fourth suggestion by Miles (1971) is that placoderms were primitively without a hyomandibular (modified epal element) and thus the hyoid arch took no part in jaw suspension. This view contrasts with that of Stensio (1963) who considered that hyostyly, and thus the presence of a hyomandibular, is primitive for placoderms. Miles believes that the hyomandibular of placoderms is nonhomologous with that in chondrichthyans which, in turn, is non-homologous with that of osteichthyans because of supposedly unique relationships, between the head of the hyomandibular and the hyomandibular foramen in each group. We have questioned that uniqueness above, and since we believe that a hyomandibular is present in ptyctodonts, gemuendinids and arthrodires we suggest that it is primitive for placoderms. Whether placoderms are primitively hyostylic is more difficult to decide because of confusion over the terminology of jaw suspension (Maisey, 1980). O u r observations suggest that, as a whole, the hyoid arch is involved in jaw suspension and that there is a hyomandibular. But direct involvement of the hyomandibular, such as is seen in elasmobranchs and actinopterygians, may only be seen in gemuendinids (Goujet, 1982, has questioned even this) among placoderms. Therefore, we adopt a position 'between the competing theories by suggesting that a hyomandibular is primitive for placoderms (Stensio) but that it took no direct part in jaw suspension (Miles) (see also Goujet, 1975: 97). Judged against other gnathostomes we suggest that the non-involvement of the hyomandibular is a derived condition. RELAIYONSHIPS OF PTYCTODONTS

In this paper we have assumed that ptyctodonts are placoderm fishes. T h e purpose of this section is to justify that assertion and we choose to do this by examining the most popular counter-theory. The suggestion that ptyctodonts are either directly ancestral to, or collateral descendants of, holocephalans has had many adherents including Pander ( 1858), Jaekel ( 1903), Dean ( 1909), Holmgren ( 1942), Stensio ( 1948, 1959, 1963), Westoll (1945, 1962, 1978) and Brvig (1960, 1962, 1971). Pander (1858), Jaekel (1903) and Dean (1909) were impressed by the resemblance in gross morphology between the dental plates of ptyctodonts and chimaeroids, whereas Stensio ( 1925, 1934) considered the chimaeroid dentition to be similar to that of placoderms. Stensio also noted seven other similarities between placoderms and holocephalans, including the nature of the joint *Goujet (pers. comm.), like Gross (1963), prefers to interpret the so-called antotic process ofJagorina as the palatoquadrate. If this is accepted then Jqorina would fall into line with all placoderms in having an anterior articulation between the palate and the neurocraniuni.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

54

P. L. FOREY AND B. G . GARDINER

between the neurocranium and the vertebral column, the presence and nature of the gill cover, the dorsal extension of the pelvic girdle, the location of the branchial arches beneath the neurocranium, the well developed rostral face of the neurocranium, similarities of the arterial system, and the general body shape. However, it was Holmgren (1942: 215-220, 254) who, by restricting Stensio’s ideas to ptyctodonts, first produced a list of additional resemblances between Rhamphodopsis (then the best known ptyctodont) and chimaeroids. These included the short deep palatoquadrate, the first dorsal fin supported by a spine and triangular basal, the structure of the second dorsal fin, the resemblance between the median ridge scales of Rhamphodopsis and the paired placoid scale rows near the midline of the back of Callorhynchus, and the presence of labial cartilages (based on his reinterpretation (1942: fig. 36A) of the structures described by Watson (1938: 400) as hyoid arch elements in Rhamphodopsis; see above). Holmgren (1942: 21 7) also suggested possible homologies between the sensory canal bones of Rhamphodopsis and the crescents around the sensory canals in chimaeroids and between the anamestic bones of Rhamphodopsis and calcifications in the corium of Callorhynchus. Following his description of the ptyctodont Ctenurella, Brvig (1960, 1962) noted a further eight resemblances between ptyctodonts and chimaeroids. These range from paired, upwardly directed rostral processes, a crested synarcual, prepelvic and pelvic claspers, to the anterior position of the mandibular articulation, the (presumed) presence of a small additional anterior pair of upper tooth plates and the shape of the neurocranium and mandible. Stensio ( 1963: 407-409) added that the fenestra endonarina communis opened ventrally in both ptyctodonts and chimaeroids, and suggested that, in arthrodires, the palatoquadrate commissure had fused into the neurocranium as in chimaeroids. Finally, 0rvig (1971) added weight to one of Holmgren’s suggestions by proposing that the laterosensory components in the dermal bones of Ctenurella (in particular the tubular laterosensory bone of the infraorbital canal) correspond to the crescents around the sensory canals in chimaeroids. He further proposed that dilations in the lateral line canals of the head of the ptyctodont Chelyophorus matched those of the lateral lines of the snout region of some Recent chimaeroids. The anterior position of the branchial arches below the neurocranium is seen in holocephalans and many placoderms but not in selachians. O n the other hand the short deep palatoquadrate, the anterior position of the mandibular articulation, the structure of the second dorsal fin and the prepelvic claspers are confined to holocephalans and ptyctodonts. To regard these last similarities as synapomorphies would be to destroy the monophyly of both the placoderms and the elasmobranchs. Moreover, the prepelvic claspers are represented in primitive chimaeroids by a group of enlarged scales (e.g. Metopacanthus, Patterson, 1965: 139) while the structure of the second dorsal fin appears to be primitive for gnathostomes (Patterson, 1965: 2 10). The prepelvic clasper in living chimaeroids, as well as having an armament of scales, is supported by one, two or three separate cartilages. Those in ptyctodonts consist of a pair of dermal spines in Ctenurella (0rvig, 1960: 327) and a pair of enlarged dermal plates in Rhamphodopsis (Miles, 1967: figs 16, 17). Pelvic claspers occur in both holocephalans and selachians where they are supported by cartilage. Those of ptyctodonts appear to lack any endochondral

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

55

support and are therefore here regarded as non-homologous with those of chondrich thyans. The rostral processes in holocephalans are confined to the later members Wurassic-Recent, Patterson, 1965) where the paired lateral cartilages lie ventral to a median cartilage. This arrangement is the reverse of the situation in selachians where the lateral cartilages are dorsal to the median cartilage. Furthermore the lateral cartilages in holocephalans articulate with the neurocranium whereas those in generalized selachians are outgrowths of the neurocranium (Holmgren, 1942: 205). In rays, however, there are paired separate rostral appendices which have been considered by Holmgren (1942: 206) to be the homologues of the lateral rostral cartilages of holocephalans, but from their variable relationships to the antorbital cartilages this seems unlikely. A synarcual appears to have been present in the earliest holocephalans (Patterson, 1965: 194) where it was apparently developed in connection with the anterior dorsal fin. In chimaeroids and myriacanthoids it supports both the fin and the large fin spine. Elsewhere in chondrichthyans a synarcual is confined to rays where it seems to have been developed in relation to the enlarged pectoral fins. Among placoderm taxa the synarcual has a sporadic distribution. In Ctenurella the synarcual is formed by the fusion of the first four neural arches (Miles & Young, 1977) and unlike holocephalans and other placoderms it does not embrace the notochord. It has a well developed dorsal crest and articulates with the median dorsal plate. Posteriorly, it supports the large anterior basal plate of the anterior dorsal fin. In Jagorina (Stensio, 1963: fig. 7) the synarcual is composed of three vertebrae and is separate from and anterior to the median dorsal plate. A synarcual is absent in most arthrodires but in TremaLosteus (Stensio, 1963: fig. 8) it is made up of 11 or 12 fused vertebrae and in Paraleiosteus (Stensio, 1963: fig. 93) and Leiosteus (Stensio, 1963: fig. 102) of some 20 partially fused vertebrae. In each of these three forms there is no dorsal expansion and the synarcual does not articulate with either the median dorsal plate or the dorsal fin base. A short synarcual with paired occipital processes appears to have been present in Pseudopetalichthys and Stensioella. I n view of the different development and the sporadic distribution we conclude that a synarcual developed independently in holocephalans, selachians and several placoderm groups. The gill chamber of placoderms is covered by a large dermal bone in Pseudopetalichthys, ptyctodonts, gemuendinids, palaeacanthaspids, petalichthyids, arthrodires and antiarchs. In gemuendinids and arthrodires the dermal bone may be associated with an opercular cartilage (see above). In holocephalans the gill cover is entirely cartilaginous, and quite unlike that of placoderms. Dental plates are known in ptyctodonts, phyllolepids, arthrodires and antiarchs. In ptyctodonts, phyllolepids and antiarchs there are generally considered to be two pairs, whereas in arthrodires there are three (two in the upper jaw and one in the lower). We note however, that Young’s (1984) excellent description of Bothriolepis sp. from Gogo, Australia, includes the observation that there is no upper dental plate but rather the mental plate, traditionally regarded as the dental plate, is better interpreted as the suborbital. I n chimaeroids and Squaloraja there are three pairs of tooth plates, much as in arthrodires (two upper and one lower), but in myriacanthoids there are two or

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

56

P L. FOREY AND B G. GARDINER

three pairs in the upper jaws and one pair in the lower jaw (plus a median symphysial tooth) while in menaspoids there are from one to three pairs of upper tooth plates and one pair of lower tooth plates (Patterson, 1965). The tooth plates of ptyctodonts have a basal layer of dense bone and are strengthened by tubular dentine, which in Ptyctodus forms the tritural areas. A similar tubular dentine is developed in holocephalans, in rays, some durophagous sharks and dipnoans, but the precise histological details may vary between these groups. The biting surface of ptyctodont tooth plates is further hardened by addition of secondary dentine. T h e tooth plates of arthrodires often bear tooth rows which are constructed of semidentine rather than dentinc (0rvig, 1980a). Like Patterson (1965: 210) we conclude that the dentition of holocephalans has a general resemblance to that of placoderms and not to ptyctodonts in particular. Lund (1977: 218) points out several differences between the tooth plates of Ptyctodus and chimaeroids and, with brief discussion of other points, concludes that ptyctodonts are “too specialized to be even distantly related to any known holocephalian or bradyodont”. The structures in Rhamphodopsis described as labial cartilages by Holmgren (1942) and hyomandibular by Watson (1938: 400) and Miles (1967: 105) are more reasonably interpreted as parts of the palatoquadrate (see above) and we have never found any trace of labial cartilages in Ctenurella. The presumed resemblance between the median ridge scales of Rhamphodopsis and the paired scale rows on the dorsal surface of Callorhynchus is likewise fallacious. As Miles (1967: 114) has pointed out, the dorsal elements in Rhamphodopsis interpreted by Watson (1938) as ridge scales are more sensibly interpreted (in the light of Ctenurella) as radials of the dorsal fin. Furthermore, the comparison by Brvig (1971) of the laterosensory components in the dermal bones of Clenurella and the crescents around the sensory canals in chimaeroids is based on the presumed presence of a tubular bone on the infraorbital canal in Ctenurella (0rvig, 1971: fig. 6A). We have never observed such a bone in any ptyctodont and, further, Patterson (1965) has shown that in menaspoids the sensory canals are associated with modified scales and has argued for a similar explanation for the crescents surrounding the sensory canals in chimaeroids. We agree with his conclusion. The calcifications in the corium of Callorhynchus (Holmgren, 1942: 217) on the other hand are interpreted as relics of the armour of primitive chimaeriforms (Patterson, 1965: 201 ) . Finally, there are two characters which have been used by Stensio (1925, 1963) to link placoderms with holocephalans: the presumed fusion of the palatoquadrate commissure with the neurocranium and the dorsal extension of the pelvic girdle such as is seen in Coccosteus. T h e palatoquadrate commissure is also incorporated into the floor of the neurocranium in actinopterygians (Holmgren, 1943) and dipnoans (Bertmar, 1966) while the pelvic girdle is without a dorsal extension in such holocephalans as Helodus (Patterson, 1965: 211) and in ptyctodonts (0rvig, 1960: fig. 5). We conclude that there is little evidence of direct relationship between holocephalans and ptyctodonts or between holocephalans and placoderms in general. The only ptyctodont character which stands critical examination is the presence of paired, articulated rostra1 processes but since similar processes are confined to later holocephalans we rate them as a convergence.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

57

CLASSIFICATION O F PLACODERMS

As stated in the introduction a number of classifications have been proposed in recent years, and in this section we attempt to evaluate them, and to suggest our own. Five previously published classifications are shown in Fig. 4, the earliest being that of Miles & Young (1977). Those authors had already criticized an earlier classification by Denison ( 1975) by listing nine ‘undesirable consequences’ (Miles & Young, 1977: 128). Since then, Denison (1978) reclassified the placoderms and, in so doing, answered some of those criticisms, and we therefore choose his later classification. We note, however, that Denison ( 1983: 82) most recently concluded a paper dealing with placoderm phylogeny with the following . . . “My analysis of currently available facts does not support relationship between any of the orders of Placodermi except Arthrodira and Antiarchi.” Regrettably, such a conclusion is not very helpful and, while we acknowledge that it is not his latest, we rely on his more explicit contribution (Denison, 1978: fig. 10). These classifications show considerable similarity; in particular, all acknowledge a group Dolichothoracomorpha which includes Arthrodira, Antiarcha, Wuttugoonuspis and Phyllolepis. This group is characterized by the presence of a posterior median ventral (PMV) by Miles & Young (1977) and Denison (1978), a conclusion with which we agree. T h e presence of a PMV in Phyllolepis was questioned by Denison (1978) but has been confirmed by Long (pers. comm.). Young (1980) suggested that the PMV specified a larger group (Dolichothoracomorpha Petalichthyida) but we consider that a P M V has not been satisfactorily demonstrated in petalichthyids (see also Miles & Young, 1977: table 1; Denison, 1978: 37). Miles & Young (1977) and Denison (1978) thought that a posterior ventrolateral (PVL) was restricted to dolichothoracomorphs but we agree with Young (1980) that a PVL specifies a larger group, including petalichthyids. The classification of dolichothoracomorphs is not the primary subject of this discussion, but there are various hypotheses (Young, 1981) and some comments are added at the end of this paper. Another area of agreement between the classifications illustrated in Fig. 4 is that Acanthothoraci and Radotina form a monophyletic group, Palaeacanthaspida (=Acanthothoraci of Denison, 1978). We therefore agree with the arguments of Miles & Young (1977: 135) and recognize this group whose members have a paranuchal which projects posteriorly beyond the skull margin (Young, 1980). We cannot agree with further characters of this group given by Denison (1978: rostrum, two pairs of paranuchals and loss of tesserae in some). I n other areas of the classifications there is little agreement (Fig. 4), a fact which can be traced to three sources: different assumptions about primitive conditions for placoderms, different weighting of prescribed characters, and difficulty of topographic comparison. The differences can be seen by examining Fig. 4 and the accompanying legend. One of the important points to note is the different hierarchical levels a t which the same character is used in the different classifications. For instance, in classifications B and E (Fig. 4) the anterior median ventral plate (AMV) specifies a group Dolicothoracornorpha Petalichthyida, yet both acknowledge that a n AMV also occurs in Ptyctodontida. For classification E we therefore have the situation where one

+

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

59

major group is recognized solely on the basis of a parallelism. A similar situation also exists in classification D (Fig. 4) where a group Palaeacanthaspida Gemuendinida is recognized on the basis of characters also occurring in antiarchs among dolichothoracomorphs. A major difference between classification E and the rest is that Denison assumes that the tesserate condition is primitive. This fixes the root of his tree at Pseudopetalichthyida. Classifications A, B and D assume that ptyctodonts are the most primitive group because they alone among placoderms have pelvic and prepelvic claspers. We have questioned the basis of this assumption (p. 54; and Gardiner, 1984a). One of the consequences of this assumption is that non-ptyctodont placoderms are united by a character of absence (no pelvic claspers). It might also be noted that an absence character (absence of AMV) specifies a group Paiaeacanthaspida Gemuendinida Pseudopetalichthyida in classifications A and D. In other words, Young and Miles & Young have weighted loss characters within their respective classifications, an action justified by Miles & Young (1977: 128) but rejected by Denison (1983: 69). The third area of disagreement between these classifications concerns identification of bones, particularly skull roofing bones. As noted by Denison (1978) and Gardiner (1984a), each recognized order of placoderms has a distinctive skull roofing pattern, although it is generally agreed that there is some commonality of pattern (Denison uses this as a synapomorphy but, unfortunately, does not clearly state what the pattern is). Beyond this, bone homologies have been painstakingly discussed (see, for instance, the complexity of bone nomenclature summarized for ptyctodonts by Westoll, 1962: table 1 ) in support of one classification or another. Nowhere is the dilemma better illustrated than in identification of paranuchal plates (Westoll, 1967; Miles & Young, 1977; Gardiner, 1984a). There is no point in discussing the issue again at length but, in comparing the classifications, it is worth noting the very different identifications which have been made. Denison identifies two paranuchals in petalichthyids and palaeacanthaspids (a distribution he thought to be due to parallelism). Miles & Young (1977: 134) cautiously suggested that two paranuchals could be additionally identified in gemuendinids and Psuedopetalichthys. However, this suggestion was based on the belief that the number of paranuchals could be correlated with the length of the occipital

+

+

+

Figure 4. Diagram to show five recent classificationsof placoderm fishes. A, After Young (1980: fig. 27B); B, after Young (1980: fig. 27C); C, Gardiner (1984a); D, Miles & Young (1977); E, Denison (1978). Against each classification the synapomorphies used by the respective authors are shown, together with their homoplastic occurrences. The synapomorphies are shown to the right of the clade to which they refer. Numbers in parentheses mean that the particular synapomorphy is found in some but not all members of the group. The numbering is consistent with that in Fig. 5. Numbered synapomorphies are as follows: 1, MD; 2, ADL; 3, AL; 4, PL; 5, PDL ( - 5 , absent in Phyllolgis); 6, I L f AVL, two separate bones; 7, simple dermal neck joint; 8, PVL; 9, ginglymoid joint; 10, PMV; 11, sliding joint; 12, Sp (-12, absent in Ctenurella); 13, main sensory canals running parallel to skull margin; 14, AMV separating AVLs; 15, centrals present; 16, projecting paranuchals; 17, palatoquadrate attached to suborbital; 18, prepectoral process; 19, eye completely enclosed in inflexible hones; 20, dorsal nares; 21, tesserae overlying dermal bones; 22, centrals excluded from skull margin; 23, two paranuchals; 24, postmarginal; 25, AMV; 26, AVLs meeting in midline; 27, sensory canals in open grooves; 28, tritural teeth with hypermineralization; 29, two supragnathals; 30, PVLs meet in midline; 31, loss of prepelvic and pelvic claspers; 32, dorsal orbits; 33, premedian (also present in antiarchs); 34, basic placoderm pattern of large cranial hones; 35, nares in middorsal position; 36, synarcual enclosing the notochord; 37, ventral fossa of pectoral girdle closed; 38, elongate occipital region; 39, IL and AVL represented as a single bone.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

60

P. L. FOREY AND B. G GARDINER

region; but at the same time they recognized (Miles & Young, 1977: 133) that no one can agree on what constitutes a long or short occipital region. The purpose of these notes is not to denounce these classifications. All have put forward explicit theories which can be appraised because the author’s assumptions are clearly stated. But a perusal of these classifications shows them all to be highly homoplastic and susceptible to substantial change with a change of weighting and/or assumption. We realize that many of the problems exemplified in the notes above are not unique to placoderm classification but they are severely exacerbated by lack of precise knowledge about many of the taxa (Denison, 1978). For our own attempt we decided to list 30 features for as many of the better known placoderm taxa as possible and to subject this data to a cladistic computer analysis using PAUP version 2.2 (Phylogenetic Analysis using Parsimony, a program prepared by Dr David Swofford), made available through the British Museum (Natural History). Since the purpose is to consider as wide a range of features as possible for each taxon, we could not include forms such as Macropetalichthys, known only by the skull and poorly known trunk fragments, even though they have figured prominently in discussions of placoderm phylogeny. Further, we considered arthrodires and antiarchs each as a single terminal taxon. The monophyly of each is, in our view, well established (Miles & Young, 1977; Young, 1980; Janvier & P’an, 1982). We therefore disagree with Denison’s assertion (1978: fig. 30) that arthrodires are paraphyletic. The data matrix is presented in Table 1 where each feature is recorded as being present (1) or absent (0). These features are presumed derived characters (or character states) and therefore represent our theories justified, in the most part, by in-group and out-group comparisons. For example, projecting paranuchals (character 16) only occur within placoderms and even here have a restricted distribution. O n this basis we score it as ‘1’. Additionally, the PDL plate has a wide distribution within placoderm subgroups but is not found in out-groups (unless we assume homology between the PDL of placoderms and the postcleithrum or anocleithrum of osteichthyans). I n other instances we have ignored the possibility of transformation of one structure into another. For example, some authors (Young, 1979) have regarded the ginglymoid neck joint of most arthrodires to be a phyletic transformation of the sliding joint as seen in actinolepids. Others (e.g. Miles, 1973a) deny such a transformation. In this paper we regard three types of neck joint as separate characters and, because none occurs outside placoderms, all to be derived. ‘Therefore, each is scored as ‘1’. Our justification for this action is that we are initially concerned with recognizing groups, although it is accepted that transformations may be inferred from the classification. Not all features are known for all taxa. I n these instances we have two choices in coding the data. The PAUP program will accept missing data and we have used this facility on two occasions (we do not know if a PDL is present in Brindabellaspis, or if Lunaspis has a dermal neck joint). I n other instances reasonable assumptions of presence or absence can be made. In one instance (character 13-pattern of sensory canals) we are not confident of the derived condition and we have therefore instructed the program to treat this character as ‘unordered data’.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTEMURELLA AND PLACODERM FISHES

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

61

62

P. L. FOREY AND B. G. GARDINER

In these areas-particularly assessment of derived characters, predictions of presence or absence-our method is most obviously open to criticism. Finally, in some instances we disagree with other authors about morphological interpretation and €or this and reasons above some justification of the data is required. (1) Median dorsal ( M D ) : Distribution as in Miles & Young (1977: table 1) except that we do not believe a M D to be present in Pseudopetalichthys (see also Denison, 1978: 22). A M D is here recognized as a plate showing overlap relationships with lateral trunk plates. Among palaeacanthaspids a M D is unknown in Romundina and Radotina. We assume it to be present in Romundina because of the dorsal overlap area on the posterior dorsolateral (PDL) described by Brvig (1975: 59). We assume a M D to be present in Radotina, and in Brindabellaspis on comparative grounds. (2) Anterior dorsolateral ( A D L ) : Distribution as in Miles & Young (1977: table 1 ) except that we consider an ADL to be absent in Pseudopetalichthys (Miles & Young were uncertain). An ADL is unknown in Romundina, Radotina, Palaeacanthaspis or Brindabellaspis, but its presence is inferred from the evidence of overlap area of adjacent plates (Stensio, 1944; Gross, 1959; Brvig, 1975; Young, 1980) and in Brindabellaspis from the facet on the skull (Young, 1980) implying the presence of an ADL with a trochlea. (3) Anterior lateral ( A L ) : Distribution as in Miles & Young (1977: table 1). ( 4 ) Posterior lateral ( P L ) : Distribution as in Miles & Young (1977: table l ) , and in Wuttagoonaspis where it is deduced to be present because of the overlap areas on the PDL (Ritchie, 1973: 63). (5) Posterior dorsolateral ( P D L ) : Distribution as in Miles & Young (1977: table 1). A PDL has not been found in Palaeacanthaspis (presence inferred from overlap areas on MD, Stensio, 1944: fig. 5c) or Radotina. Young (1980: 47, fig. 19) described a possible PDL in Brindabellaspis. It is possible that this is a correct interpretation but the diagram presented by Young suggests that it may be interpreted as the top of the scapulocoracoid. We record it here as missing data (?). (6) Anterior ventrolateral ( A VL) and separate interolateral ( I L ) : Identification of the plates covering the ventral surface of the scapulocoracoid has been discussed on many occasions (Denison, 1983: 80-8 1, for review). Most placoderm groups have a single bone in the position occupied by separate AVL and I L in arthrodires, Phyllolepis and Wuttagoonaspis. Where a single bone is present some authors have assumed a compound origin (Stensio, 1959; Brvig, 1960, 1975). Other consider only one to be present, sometimes called AVL, sometimes IL. In this work we record only those groups in which it has been clearly demonstrated that two separate bones occur. We have not recorded an IL in antiarchs but acknowledge that it may be present as a separate element in Yunnanolepis (Zhang, 1980: fig. 3D). This feature, as expressed here, is equivalent to the distribution of an I L recorded by Miles & Young (1977: table l ) , since they adopted the convention where a single bone is termed an AVL. We disagree with Miles & Young to the extent that we cannot find a convincing example of a separate IL recorded for palaeacanthspids, pseudopetalichthyids or pet alich thyids. (7) Simple dermal neckjoint: By this we mean a dermal neck joint which is not obviously of the sliding-type or of the ginglymoid type (character 9)’ but is instead a simple ‘butting’ joint with reciprocal flat faces. Amongst the taxa

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

63

considered here it is represented in Ctenurella (Miles & Young, 1977) and probably Rhamphodopsis (Miles, 1967: 107). Such a joint was probably more for restricting and bracing head movements than facilitating movement. (8) Posterior ventrolateral ( P V L ) : As in Miles & Young (1977: table 1 ) except that we believe the PVL to be present in Lunaspis (Gross, 1961: pl. 5). (9) Ginglymoid neckjoint: Here we record taxa in which a ginglymoid joint is present. The term ‘ginglymoid’ is used here to mean a joint which is formed by a condyle (on the paranuchal or the ADL), matched by an articulatory fossa, and which clearly facilitated some rotational movement. Our reason for recording a ginglymoid joint in Brindabellaspis is the endocranial articulatory fossa (Young, 1980: fig. 9). Its presence in Phyllolepis is based on the fact that the ADL apparently has an articular process (J. Long, pers. comm. to B.G.G.). It is also present in some specimens of Wuttagoonaspis (A. Ritchie, pers. comm.). I n the data matrix we have recorded missing data for the dermal neck joint in Lunaspis based on the work of Gross (1961) who was unable to identify evidence of such a joint in either of the two species (L. broilii Gross and L. heroldi Broili). However, it should be noted that other petalichthyids such as Macropetalichthys and Wijdeaspis (Stensio, 1969; Young, 1978) do show good evidence of such a joint. We feel, therefore, that the missing data entry under Lunaspis is probably more a reflection of poor preservation than a statement of absence. T h e form of the joint in Macropetalichthys and Wijdeaspis is somewhat problematical and it is not obvious how it worked (Young, 1978: 113). However, it seems to correspond with our ‘ginglymoid joint’ in which some rotational movement is hypothesized. (10) Posterior median ventral ( P M V ) : This plate is found in arthrodires, antiarchs and Phyllolepis Long, pers. comm.) and Wuttagoonaspis (see also p. 57). (11) Sliding dermal neck joint: This feature is normally said to be present in actinolepid arthrodires and Wuttagoonaspis and is recorded so here. The classificatory significance of the sliding joint is disputed, with Miles (1973a) and Miles & Young (1977) suggesting that it specifies a group actinolepids, or actinolepids+ Wuttagoonaspis, while Young (1979, 1980, 1981) suggests that it specifies a more inclusive group (euarthrodires Wuttagoonaspis). The knowledge that Wuttagoonaspis has one species ( W .Jetcheri) with a sliding joint (Ritchie, 1973) and one unnamed species with a ginglymoid joint (A. Ritchie, pers. comm.) adds confusion to any attempt to regard one type of joint as a transformation of the other (Gardiner, 1984a). As a further confusing point we might add that for the palaeacanthaspid genera considered here we cannot be certain whether a dermal neck joint was absent or whether it may have been a sliding joint. (The lack of any glenoid fossa on the skull suggests that a ginglymoid joint is unlikely to have been present, despite the restoration of a trochlea on the ADL of Palaeacanthaspis by Stensio (1944).) In none of the palaeacanthaspid genera considered here is the ADL known and there is no indication of the undersurface of the skull roofing bones of any sliding surface. However, a sliding joint may not leave any mark on the skull (see, for instance, Kujdanowiaspis Stensio, 1944). We record the palaeacanthaspid genera as having neither ginglymoid nor sliding joints but recognize that this may be incorrect. We feel more confident in proposing that a dermal neck joint was absent from gemuendinids and Pseudopetalichthys. In these forms the shoulder girdle is separated from the skull by a considerable space. (12) Spinal: Present in all taxa considered here except Ctenurella, Pseudopetalichthys and, as a secondary feature, a few brachythoracid arthrodires.

u.

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

64

1’. L. FOREY AND B. G . GARDINER

We thus agree with Denison (1983) about the condition in Pseudopetalichthys (cf. Miles & Young, 1977: table 1). We suggest also that a spinal is present as a general condition of antiarchs, a view based on the evidence in Yunnanolepis (Zhang, 1980). (13) Main cephalic sensory canals running parallel to the margin of the skull roof: We have coded this feature with the assumption that a pattern of sensory lines converging to a point above the otic region of the braincase is a primitive gnathostome feature (Gardiner, 1984b). I n arthrodires the central canal and posterior pit-line do converge towards the midline but they do so within the central, not the nuchal. We accept that the canals converge in ptyctodonts (Miles, 1967) on the nuchal plate (terminology of Miles & Young, 1977) although we acknowledge considerable dispute over the homology of the median plate of the skull roof receiving the main lateral line canals. We are not entirely confident that we have made the correct assumption that ‘converging canals’ is primitive, chiefly because we recognize that heterostracans have an essentially parallel pattern of sensory canals. For this reason the program was run with the instruction to consider this character as ‘unordered’, meaning that it considered both the ‘1’ and ‘0’ condition as equiprobable derived conditions. (14) AMV completel3, separaling AVLs and meeting PMV: No comment necessary. (15) Central: We do not believe that central plates can be identified with certainty in Pseudopetalichthys. There is some doubt about the presence of a central in Brindabellaspis (Young, 1980: 15) and the so-called central of petalichthyids (Denison, 1978) is unusual in entering the orbit. The large median bone in the skull roof of Phyllolepis, here called nuchal, is sometimes thought to contain a central. Similarly, the lateral of antiarchs is thought to have ‘captured’ the central. In none of these instances is there any evidence of fusion, ontogenetic or phylogenetic. We have recorded the presence of a central in Wuttagoonaspis even though Ritchie (1973) does not record such a separate element. Ritchie (1973: 64) “feels that there is evidence in Wuttagoonaspis to suggest that there has been fusion of the nuchal and central plates.” He did not cite evidence but D. Goujet (pers. comm.) has observed separate centrals in some specimens, an observation which we accept. (16) Paranuchal projecting posteriorly, producing an embayed skull margin: This feature is used by Young (1980: 65) as a character of palaeacanthaspids; we agree with his thesis. (17) Palatoquadrate j r m h attached to suborbital: We regard this as a functionally significant character (Miles, 1969; Schaeffer, 1975) because it implies a considerable strengthening of the upper jaw and, at the same time, an increasing inflexibility of the jaw mechanism and a change in the position of insertion of the adductor musculature. Unfortunately, the condition of the palatoquadrate is known for few placoderms. For instance recorded absence of character 17 (Table 1 ) in Brindabellaspis, Lunaspis and most palaeacanthaspid genera may mean no more than ignorance of this part of the anatomy. (18) Prepecloral process developed on scapulocoracoid: This feature is absent in ptyctodonts and Pseudopetalichthys but is recorded as present in all other taxa. For many placoderms there is no direct evidence of the presence of a prepectoral process, rather it is inferred from the hollow base of the spinal. It is probable

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

65

that most antiarchs lack the process but, accepting the classification proposed by Young (1981) and Janvier & P’an (1982), in which Yunnanolepis with a hollow spinal (Zhang, 1980: fig. 4B) is the plesiomorphic sister-taxon, absence of the prepectoral process in most antiarchs is probably secondary. ( 1 9 ) Eye completely enclosed within inflexible bones: We have recorded this feature as absent in arthrodires but acknowledge that some genera (e.g. Holonema) do show this feature. ( 2 0 ) Dorsal nares: We have considered Brindabellaspis to show this feature recognizing that, in opening into the anterior edge of the orbit (Young, 1980), the condition is dissimilar to that in antiarchs, gemuendinids and palaecanthaspids. We have made an assumption that Kosoraspis has dorsal nares: the anterior half of the skull is, however, very poorly known. ( 2 1 ) Tesserae overlying dermal plates: The tesserate condition of some Lunaspis, gemuendinids and placoderms (some palaeacanthaspids, Pseudopetalichthys) has been discussed at length (summaries in Westoll, 1967; Miles & Young, 1977; Denison, 1983; Gardiner, 1984a). There is still no consensus about whether the tesserate condition is primary or secondary. We adopt the view that there are at least two conditions of tessera development: a primary type possibly exemplified by Pseudopetalichthys; and a secondary type exemplified in other tesserate placoderms. Distinction between the two depends on the fact that in the secondary type tesserae at least partly overlie deeper-lying dermal bone (Patterson, 1977). Such seems to be the case in gemuendinids, Radotina (Westoll, 1967), Kosoraspis and probably Lunaspis. ( 2 2 ) Centrals excluded from posterior skull margin: I n placoderms there are two obvious patterns with respect to the dominant pair of bones, usually called centrals, which cover the otic region of the braincase and which lie medial to the main cephalic lateral line. Arthrodires exemplify one pattern, where these bones are excluded from the posterior margin of the skull because the nuchal meets the paired paranuchals. The palaeacanthaspid genera dealt with here (Romundina, Radotina, Palaeacanthaspis, Kosoraspis) and gemuendinids exemplify another pattern where the centrals form part of the posterior skull margin. For those forms (Brindabellaspis, Lunaspis) where the identification of centrals is questionable (character 15) or centrals do not generally exist as separate elements (antiarchs, Phyllolepis and Wuttagoonaspis), we consider these to show the ‘arthrodire-type’ on the grounds that the nuchal meets the paired paranuchals. ( 2 3 ) T w o paranuchals: As mentioned above (p. 59) the number of paranuchals has posed problems of identification (Westoll, 1967; Miles & Young, 1977; Denison, 1983; Gardiner, 1984a). T h e single paranuchal of arthrodires carries the main lateral line and bears the origin of the posterior pitline/canal. In Lunaspis and Brindabellaspis there are two bones in series which bear these structures and we regard it as reasonable to consider these as two paranuchals. But in other forms in which two paranuchals have been identified at one time or another (palaeacanthaspids, gemuendinids and Pseudopetalichthys) identification is difficult because in some (Radotina kosorensis, Kosoraspis, gemuendinids and Pseudopetalichthys) there are either several small plates or tesserae while in others the bone margins are by no means clear. I t is possible that Romundina shows two paranuchals but, if this is accepted, we would have to deny the existence of a marginal. (24) Postmarginal: No comment necessary. 5

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

66

P. L. FOREY AND B. G. GARDINER

(25) Anterior median ventral ( A M V ) : Distribution as in Miles & Young (1977: table 1) except that we are sceptical about the report of an AMV in Rhamphodupsis (Watson, 1938). We have examined many specimens of Rhamphodupsis, including some excellently preserved and prepared specimens (R.trispinahs Watson, RSM 1859.33.607; R. threiplandi Watson, RSM 1962.26.38B) and have never seen a convincing AMV. Rather, the AVL platcs of either side seem to meet in the midline over their length. However, in Ctenurella gardinera a large AMV is present (Miles & Young, 1977: fig. 31). (26) Anterior uentrolaterals meeting in midline suture: No comment necessary. (27) Sensory canals in open grooves: For this feature we have listed those taxa in which part or all of the sensory canal system lay in open grooves (some placoderms show both closed canals and open grooves). Under this character we consider only the true lateral line system which, in recent fishes, is developed as neuromasts set within a liquid-filled tube. This is distinct from pit-lines where the sensory organs are not interconnected by a tube. I n those fishes in which the sensory canals pass through bones (cf. chondrichthyans and acanthodians where they pass between scales) the widespread condition amongst primitive members of the major fish groups is for the canals to be enclosed in bony tubes, opening to the surface by pores. The open groove is therefore assumed to be the derived condition. (28) Tritural teeth: This feature is restricted to ptyctodonts. There are arthrodires with grinding dentition but they show neither the unusual occlusal pattern, nor the histology of true tritural teeth. (29) Two supragnalhals: This feature is restricted to arthrodires. Parenthetically, we note that while many brachythoracid arthrodires are known to have anterior and posterior supragnathals there is no known actinolepid or phlyctaeniid arthrodire in which two supragnathals have been found in position. The presence of two supragnathals in these latter forms is inferred from the shapes and positions of individual plates in different genera. (30) P V L s meeting in midline suture: No comment necessary. The result of our computer analysis is shown in Fig. 5. It should be noted that this particular algorithm produces a rooted tree and therefore requires an ancestor to be specified. We decided to root the tree at Pseudopetalichthys because, as can be seen from the data matrix (Table 1)) this taxon shows the greatest number of absences. (We had earlier tried a Wagner unrooted algorithm using the algorithm in Felsenstein’s taxonomy package made available at the BM(NH) and then drew the resulting diagram using (a) Pseudopetalichthys, and (b) ptyctodonts as the root. Placing ptcytodonts as the root demanded assumptions of several more character reversals.) We present the results in their purest form (Fig. 5). T h e tree required 51 steps, with some individual characters used more than once, sometimes as reversals, sometimes as parellisms, sometimes both. The geometry of the tree is revealing to the extent that Lunaspis is placed as the sister-group to dolichothoracomorphs, with Brindabellaspis as the sister-group to dolichothoracomorphs Lunaspis. There is one trichotomy involving palaeacanthaspid genera. The characters fall into several categories. Those numbered 1, 2, 4, 7, 8, 9, 10, 11, 14, 16, 28, 29 and 30 appear once as changes from presumed primitive to derived state (0-1) and represent synapomorphies. Characters 14 and 29

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

Dolichothoracomorpha

-;q; -26 -17

-6

Palaeacanthaspida

1

17

-5

-24 -2 7

2,-11

4

17 6

T

I

-12

21

8 -20 25 26 30

I -

L9

-13 -15 19 22 23 24

T 7

21

16

-13 28

d 5

18 20 27 I

I

I

1 2 12 15

i 13

I Figure 5. Results of PAUP (version 2.2) computer analysis based on data presented in Table 1 and in which the tree has been rooted at Pseudopetalichthys (length of tree= 5 1, consistency index =0.577, F=21.411). The distribution of characters is shown with a minus sign preceding those characters which change state from present -+ absent. See text for explanation.

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

P. L. FOKEY AND B. G. GAKDINER

68

would be synapomorphies of arthrodires. T h e analysis also implied the prediction that Lunupsis has a ginglymoid joint (this was entered as missing data). This seems reasonable judged in the light of what is known about other joint-is ranked as a petalichthyids (see p. 63). Character 1 1-sliding synapomorphy of Wultagoonapsis arthrodires. But most arthrodires have a ginglymoid joint and therefore the sliding joint should more correctly be cited as a character of Wuttagoonaspis actinolopid arthrodires. We have already hinted at one problem of Wuttagoonaspis and joint morphology (p. 63). Another problem is that the alternative hypothesis-that arthrodires form a monophyletic group exclusive of Wuttagoonaspis-is supported by two unique characters, 14 and 29. One character (3-AL) appears once, as a loss character within antiarchs. This means that an AL is primitive for placoderms but that it has been secondarily lost in one group, antiarchs (we d o not care to argue whether the AL has been lost or fused with neighbouring plates). The remainder of the characters have been used more than once and these account for 57y0 of the total. This figure underlines the homoplastic distribution of the data and therefore requires some explanation. Characters appearing more than once fall into several categories, The first of these categories consists of those characters which occur twice and which behave differently; that is, reading up the tree, those which are gained and then subsequently lost. These include characters numbered 5, 6, 12, 18, 19, 23, 24, 26 and 27. Some of these characters are acquired very high in the hierarchy and then lost by a single terminal taxon. Thus, character 5 (PDL) group including palaeacanthaspids Brindabellaspis Lunaspecifies a spis+ dolichothoracomorphs but does not appear in Phyllolepis. It seems reasonable to regard this as a secondary reversal and thus accept the acquisition of a PDL as a synapomorphy. T h e same arguments can be applied to characters 12, 18 and 27. The last mentioned is interesting. Character 27-sensory canals within open grooves-shows a reversal in Lunaspis (sensory canals completely enclosed as in ptyctodonts). We regard this reversal as secondary and cite as support the fact that some of the sensory canal system of Notopetalichthys is developed as open grooves (White, 1952). We therefore feel reasonably confident in proposing character 27 as a synapomorphy of all placoderms minus ptyctodonts and Pseudopetalichlfys. Ritchie ( 1973: 68) also regarded the petalichthyid condition as specialized within placoderms. Character 6-- two bones ( I L and AVL) covering the ventral surface of the scapulocoracoidappears at the node embracing dolichothoracomorphs but is recorded as being subsequently lost in one subgroup (antiarchs). We have already mentioned (p. 62) that it may be present as a plesiomorphic condition of antiarchs and if this should prove correct then it may well be regarded as a synapomorphy of dolichothoracomorphs. Denison (1983: 81) has already questioned the validity of the Dolichothoracomorpha because it is supported by a single character (PVL), which is probably present also in Lunaspis. We therefore accept Denison’s criticism but suggest that the group may be recognized on the basis of a PMV and, probably, two separate plates (IL and AVL) on the ventral surface of the scapulocoracid. Character 26-AVLs meeting in the mid-line-is acquired at the petalichthyid dolichothoracomorph level but subsequently recorded as lost in

+

+

+

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

+

CTENURELLA AND PLACODERM FISHES

69

arthrodires. This ‘loss’ can be attributed to the very large AMV which has a broad area of contact with a PMV, so separating the AVL of either side. T h e behaviour of character 23-two paranuchals-is unusual. It appears and disappears within two nodes straddling Lunaspis and Brindabellaspis. An alternative explanation, and equally parsimonious solution, is that two paranuchals were acquired in parallel by Lunaspis and Brindabellaspis. In view of the difficulty in identifying paranuchals (p. 59) it seems unwise to try to explain the distribution of this character. Character 24-postmarginal plate-appears in Brindabellaspis and dolichothoracomorphs. T h e tree suggests that Lunaspis has lost the postmarginal. I t is, however, possible that a postmarginal is restricted to dolichothoracomorphs, and that the presence of a postmarginal in Brindabellaspis should be questioned. Clearly, this is a feature worth reinvestigation (see Young, 1980: 15-16). The second group of homoplastic characters are parallelisms or convergent features. Characters 21, 22 and 25 fall into this category, with the first appearing on three separate occasions. T h e last two appear twice. Tesserae (21) appear in Lunaspis, some palaeacanthaspid genera and in rhenanids. We remain convinced that such structures represent the derived condition and acknowledge multiple origin. Character 22-centrals excluded from the skull marginappears at the node leading to most placoderms (Brindabellaspis and cladistically more derived relatives) and, exceptionally, in Rhamphodopsis. We cannot explain such a distribution but note that the skull roof pattern of Rhamphodopsis is exceptional among ptycotodonts in showing the extra bone (identified as a nuchal) in the posterior margin of the skull. Character 25-AMV-appears as a convergence between Lunaspis dolichothoracomorphs and Ctenurella. What is perhaps surprising is the absence of an AMV in intervening taxa. We rated an AMV as an apomorphy within placoderms ( a traditional viewpoint). But it is possible this was a n incorrect decision because, if the Placodermi are the sistergroup to osteichthyans (p. 45), then one might expect a ventral median bone in the dermal shoulder girdle to be primitive and to correspond to the interclavicle. Therefore, the absence of this bone may be more significant. I n some osteichthyans the interclavicle has sunk beneath the surface (Patterson, 1977) as a derived condition. It is possible this has been its fate in those placoderms ‘lacking’ an AMV. The remaining characters (13, 15, 17, 20) are those which appear three times and behave differently. Character 13---main cephalic sensory canal running parallel to skull margin-was entered in the program unordered and its performance has therefore been determined by the parsimonious distribution of the other characters. The distribution of this character within the tree implies that parallel sensory canals are primitive for placoderms, contrary to our a pviori assumption, and that canals converging on a nuchal plate has arisen twice within the group with one secondary reversal in arthrodires. T h e highly incongruent distribution of this character suggests that it was ill-formulated. The behaviour of character 15-centrals present-suggests that the central of arthrodires Wuttagoonaspis is non-homologous with that of rhenanids, palaeacanthaspids and ptyctodonts. This is a conclusion we find difficult to accept. We have already mentioned problems associated with the identification of centrals in Lunaspis, Brindabellaspis and Pseudopetalichthys (p. 64), and this difficulty adds a confusing element. The alternative groupings of taxa definitely known not to

+

+

5*

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

P. L. FOREY AND B. G. GARDINER

70

have centrals (antiarchs, Phzllolepis) is contradicted by two characters which suggest that antiarchs are the sister-group of arthrodires Wuttagoonaspis: 4-PDL and 17-palatoquadrate attached to the suborbital. We have no confidence in the latter (see below) suggesting, therefore, that we should denote a trichotomy at this point (arthrodires Wuttagoonaspis, antiarchs, Ph$lolepi.c). To resolve this, we choose to weigh in favour of the hypothesis as shown in the tree and accept the PDL as a more clearly identifiable character. Not only are there problems with identifying centrals but we also acknowledge that centrals may be present in some specimens of Wuttagoonaspis but not others. Character 17-palatoquadrate firmly attached to the suborbital-is also highly incongruent but this may simply be a reflection of inadequate knowledge. Character 20-dorsal nares- has an incongruent distribution which suggests that the dorsal nares of antiarchs are not homologous with those in gemuendinids, palaeacanthaspids and Brindabellaspis (although here the nares are presumed to open into the orbit rather than on the dorsal surface of the head). We therefore accept the spirit of Denison’s arguments (1983: 75-76) for non-homology even if we disagree with some of his specific points. Because of this distribution we find if difficult to accept use of this character to specify a group gemuendinids palaeacanthaspids (Fig. 4D). The shape of our tree is, in large part, similar to that proposed by Denison (1978). The grouping of Lunaspis, as representative of petalichthyids, with dolichothoracomorphs seems to be well founded and we cannot agree with the concept of Petalichthyomorpha (Westoll, 1967; Miles & Young, 1977). The grouping of Lunaspis with dolichothoracomorphs may be further supported by the fact that the skull roof of members of this group is distinctively broad at the level of the otic region, as opposed to the parallel-sided profile of cladistically less derived groups. Within the palaeacanthaspids the tree shows a trichotomy between Radotina Kosoraspis, Palaeacanthaspis and Romundina. Interrelationships between these genera, and others (Kolymaspis, Kirnaspis) regarded as palaeacanthaspids, are a particularly difficult problem. The grouping Radotina Kosoraspis is very weakly based (tesserae, seen elsewhere), although it does agree with the ideas of Young (1980). Young (1980: 67) discusses the problem of interrelationships of palaeacanthaspid genera in more detail and proposes a more highly resolved classification. We would agree with many of the points he makes but would also agree with him that knowledge of these genera is very patchy, which prevents comparison of all features in all genera. One area of our tree (Fig. 5) with which we are less than confident is the theory of relationships proposed between arthrodires, Wuttagoonaspis and antiarchs. Our tree shows it as ((arthrodires, Wuttagoonaspis) antiarchs). But the only evidence we can muster for the arthrodire Wuttagoonaspis pairing is the sliding joint. An alternative theory-( (arthrodires, antiarchs) Wuttagoonaspis)was proposed by Young (1981: fig. 4E) and has recently been elaborated by the same author (Young, 1984: fig. 12). We feel that we should comment on the weakness of our own theory and also point out some problems with Young’s theory. The alternative theories are similar to the extent that they recognize a group (arthrodires, Wuttagoonaspis, antiarchs) based on the possession of a PL plate. Young (1984) also adds a pectoral fenestra as a synapomorphy at this hierarchical level.

+

+

+

+

+

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

CTENURELLA AND PLACODERM FISHES

71

O u r theory that Wuttagoonaspis is the collateral descendant of arthrodires is only supported by the common possession of the ‘sliding’ type of dermal neck joint. But this should, more correctly, group Wuttagoonaspis with Actinolepidae within arthrodires. Our preferred grouping therefore leads to two alternative consequences. Either arthrodires are paraphyletic, with actinolepids more closely related to Wuttagoonaspis than to other arthrodires. O r the sliding joint is primitive for a group arthrodires Wuttagoonaspis and the ginglymoid joint of most arthrodires is a unique derived character, non-homologous with that in antiarchs. We find the first alternative unacceptable because arthrodires (including actinolepids) do seem to form a monophyletic group, recognized on the possession of two supragnathals (notwithstanding the problem of identifying both elements in some arthrodires; p. 66 and Young, 1984: 656) and the very large AMV which has a broad area of contact with the PMV. We therefore agree with Young (1984) that arthrodires are monophyletic and exclude Wuttagoonaspis. T h e second alternative is possible (Young, 1984: fig. 12), and indeed Young ranks the sliding joint much higher in the hierarchy. But our confidence in this character is shaken by the knowledge that an as yet unnamed species of Wuttagoonaspis has a ginglymoid joint (p. 51). At the very least, therefore, the morphology of the dermal neck joint still requires more examination. The relationship between the pertinent taxa in our tree would therefore be better expressed as a trichotomy (arthrodires, Wuttagoonaspis, antiarchs) . Young’s hypothesis-( (arthrodires, antiarchs) Wuttagoonaspis)- is supported by three characters at the node arthrodires+antiarchs. These are: a long obstantic margin producing prominent posterolateral corners on the skull roof; an elongate submarginal plate; and a prelateral and infraprelateral, one or both of which are homologous with the postsuborbital (Young, 1984). O u r observations suggest that all of these characters can be found outside the group arthrodires and antiarchs. A long obstantic margin producing prominent posterolateral corners is surely also seen in petalichthyids (Denison, 1978: fig. 25) and possibly also phyllolepids. T h e shape of the submarginal plate varies considerably within arthrodires. There are forms which have an elongate straplike submarginal (e.g. Coccosteus) which is matched by ptyctodonts among nonarthrodire placoderms. Conversely, some arthrodires (e.g. Bryantolepis, Holonerna) have submarginal plates of very similar shape to those in the petalichthyid Lunaspis. We acknowledge that the shape of the submarginal jextralateral of antiarchs) does vary, but not in a fashion congruent with either hypothesis preferred above. Finally, the presence of a postsuborbital ( = prelateral and/or infraprelateral) is not confined to arthrodires and antiarchs. A postsuborbital is found in Wuttagoonaspis (Ritchie, 1973: fig. 4; pl. 4, fig. l a ) and may also be present in Romuizdina (0rvig, 1975: 53), although it must be admitted that the evidence for the latter occurrence is circumstantial. I n sum, we do not think that either hypothesis of relationship involving arthrodires, antiarchs and Wuttagoonaspis is particularly successful in explaining the data. It is possible that knowledge of the condition of the supragnathals in Wuttagoonaspis may help resolve a trichotomy which we feel is all present data allow. In the introduction we dismissed Stensioella from consideration. We did this because we could not identify a n y placoderm synapomorphy in Stensioella, or find

+

Downloaded from https://academic.oup.com/zoolinnean/article-abstract/86/1/43/2648673 by guest on 07 December 2017

72

P. L. FOREY AND B. G. GARDINER

+

any feature that would justify a grouping placoderms Stensioella. Stensioella is a very poorly known genus (Gross, 1962) from a single Lower Devonian locality. We can only associate Stensioella with placoderms on phenetic grounds; that is on the basis of what it has, and does not have. Stensioella shows a dermal shoulder girdle with a well developed postbranchial wall bearing denticles, a feature of placoderms and osteichthyans, but it does not show any of the synapomorphies which specify osteichthyan subgroups. I t resembles ptyctodonts in having ceratohyals behind the jaws and in possessing enlarged scales in front of the pelvic fins. ACKNOWLEDGEMENTS

We wish to thank Ms Alison Longbottom who prepared X-radiographs of several specimens of Ctenurella gardineri used in this paper. We also acknowledge the time and expertise offered by Dr Chris Humphries and M r Alan Paterson who steered us through the computer program. The above are members of the British Museum (Natural History). Finally, we thank Dr Daniel Goujet ( M u s h m national d’Histoire Naturelle, Paris) for reading the manuscript and his ensuing helpful criticism. REFERENCES

ALLIS, E. P., 1922. The cranial anatomy of Polypterus, with special reference to Polypterus bichir.