A Microanatomical and Histological Study of the Postcranial Dermal Skeleton in the Devonian Sarcopterygian Eusthenopteron foordi

The fin rays and two types of scales (enlarged and regular) of the Devonian sarcopterygian Eusthenopteron foordi are redescribed using light, scanning and transmission electron microscopy. The fin rays consist of lepidotrichia composed of ossified, jointed and branched segment pairs. The basal segments are cylindrical, but more distal elements are crescentic in section. The distribution of Sharpey's fibres varies along the lepidotrichia. In the proximal segments, lateral bundles form a belt connecting adjacent hemisegments. In the distal segments, thin bundles are restricted to the area facing the fin surface. Both enlarged and regular scales have a similar spatial organisation. They are composed of a superficial highly mineralised layer covering a thick basal plate where the fibrils are distributed in superimposed strata forming a plywood-like structure. Nevertheless, the enlarged and regular scales differ in their shape, in the mineralised tissues of the superficial layer, and in the organisation of the plywood-like structure. The superficial layer of the enlarged scales is composed of parallel-fibered bone covering a deeper layer of woven-fibered bone. The basal plate is made of an orthogonal plywood-like structure. The thin, lamellar, imbricated regular scales display the characteristics of elasmoid scales. The mineralised tissue forming the superficial layer resembles that of extant teleost scales. In the basal plate, the twisted plywood-like structure is composed of closely packed fibrils that are preserved down to the ultrastructural level owing to the persistence of bridges connecting the fibrils. The enlarged and the regular scales of Eusthenopteron foordi do not present superficial odontodes, in contrast to ancestral thick rhomboid scales. The disappearance of enamel/enameloid and dentine may be related to the evolutionary trend towards a lightening of the dermal skeleton that would improve the swimming abilities of the animal. The characteristics of the dermal skeleton of Eusthenopteron foordi support the hypothesis that this process began early in osteichthyans.


Introduction
The Late Devonian (Frasnian) sarcopterygian Eusthenopteron foordi Whiteaves, 1881, a tristichopterid tetrapodomorph (Janvier 1996), is one of the most abundant tetrapodomorphs in the fossil record and has been described in detail (Jarvik 1944(Jarvik , 1965Ørvig 1957;Andrews and Westoll 1970;Schultze 1984;Cloutier 1996;Cote et al. 2002). It documents one of the latest stages in the evolution of this group before much of the dermal scale cover was lost. Detailed anatomical data on Eusthenopteron can be obtained because the abun− dance of the material is sufficient to allow sectioning (An− drews and Westoll 1970;Schultze 1984;Cote et al. 2002), and the good preservation enables histological descriptions. Despite this, histological studies on the dermal skeleton of Eusthenopteron are few (Laurin et al. 2007). To our knowl− edge, histological data on the scale structure of Eustheno− pteron have been published by Goodrich (1907) and by Ørvig (1957), but no paleohistological study concerns the fin rays and the dermal plates, also called hemicylindrical scales, or enlarged scales (Andrews and Westoll 1970). Yet, the der− mal skeleton can reveal interesting clues about the affinities of various vertebrates (e.g., Khemiri et al. 2001;Scheyer and Sander 2009).
The good preservation of Eusthenopteron and its phylo− genetic position prompted us to study the structure of the vari− ous elements of the dermal skeleton using histological tech− niques including scanning and transmission electron micros− copy. Organic molecules are known to fossilise in excep− tional cases, and mineralised tissues can be preserved down to the ultrastructural level (Pawlicki et al. 1966;Zocco and Schwartz 1994;Rimblot et al. 1995;Kemp 2002;Schweitzer et al. 2008). Collagen may be preserved well enough to be sequenced (Service 2009). The new data presented below should facilitate comparison with more primitive rhomboid scales described in other early Sarcopterygii (Goodrich 1907;Ørvig 1957;Schultze 1977) on the one hand, and with the de− rived elasmoid scales of extant Sarcopterygii like Neocerato− dus (Meunier and François 1980), Protopterus (Zylberberg 1988) or Latimeria (Smith et al. 1972;Castanet et al. 1975;Miller 1979;Meunier and Zylberberg 1999;Hadiaty and Rachmatika 2003;Meunier et al. 2008) on the other hand. These data should also help comparisons with other elements forming the dermotrichia, such as the camptotrichia of the fins of extant Dipnoi (Géraudie and Meunier 1984), and may help assess hypotheses about homologies and evolution of various dermal skeletal elements of sarcopterygians.

Methods
Light microscopy.-Parts and counterparts were glued back together and prepared from one side using a pneumatic ham− mer, a dental drill, and a mounted needle under a binocular microscope. The lepidotrichia, dermal bones and scales were then embedded in stratyl resin and sectioned for histological examination. The sections were polished down to a thickness of about 80 μm and observed under transmitted natural and polarised light with a Nikon Eclipse E66 POL and with a Zeiss Axiovert 35 equipped with Nomarski Differential In− terference Contrast (DIC).
Scanning Electron Microscopy (SEM).-After soft clean− ing with a 1% sodium hypochlorite solution, the surfaces of the fragments were dried and coated with gold and observed in a Jeol−SEM−35 scanning electron microscope at 20kV.
Transmission Electron Microscopy (TEM).-Small, thin pieces (3-5 mm thick) containing scales in their original, partly overlapping positions were isolated and cleaned with a solution of 2.5% formic acid. The pieces were rinsed in etha− nol, air−dried, and embedded in epon. Polymerisation was per− formed at 60°C. The blocks were sectioned to obtain sections about 10 μm thick and containing the whole scale thickness. These thick sections were submitted to demineralisation pro− cedures for TEM examination. Both surfaces of these sections were covered with a thin layer of epon that was polymerised before demineralisation to prevent the breaking up of the tis− sues that were weakened during demineralisation. Various de− mineralisation procedures were carried out using either 0.1 N hydrochloric acid, 0.1 M citric acid, 5% formic acid, or 5% EDTA (ethylene diamino tetracetic acid). The demineralisa− tion agent was added to the fixative solution composed of 1.5% paraformaldehyde and 1.5% glutaraldehyde in a 0.1 M cacodylate buffer. The demineralisation solution was changed every two days during one month. Demineralisation was stopped when the thick sections became less opaque and more flexible. The sections were then post−fixed in a solution of 2% osmium tetroxide in the same buffer, then dehydrated in a graded series of ethanol solutions, left for 12 hours in a 1/1 eth− anol−epon mixture and embedded in epon. Semi−thin sections (1 μm thick) were stained with a buffered toluidine blue solu− tion (pH 4) and examined by light microscopy using natural transmitted light and DIC to select appropriate areas for TEM examination. Toluidine blue was used only to stain the pre− served organic matrix in order to select areas for TEM obser− vation. This staining does not allow discrimination between collagen and other proteins. Ultrathin sections (100 nm thick) were stained with uranyl acetate and lead citrate and observed in a Philips EM 201 transmission electron microscope with an operating voltage of 80 kV.

Results
Observations carried out on three components of the dermal skeleton (the fin rays and both enlarged and regular scales) show the very good preservation of the Eusthenopteron re− mains at the histological level. Moreover, despite the rough treatment required for demineralisation, the organic residues of the fossilised scales were retained with their original spa− tial arrangement and ultrastructural characteristics after de− mineralisation by formic acid or EDTA. In this study, vertical sections of scales refer to sections perpendicular to the body surface.  Fin rays.-The basal part of the fin rays in Eusthenopteron foordi lie in the vicitnity of the radials, the distal elements of the endoskeleton (Fig. 1A 1 ). The fin rays are composed of slender, elongated elements that lie parallel to each other. Each ray or lepidotrichium is composed of repetitive seg− ments, 1 to 3 mm long, in the proximal part of the fin ray ( Fig.  1A 1 ). More distally, the segments are composed of two paral− lel series of mineralised elements close to each other, one near each surface of the fin (Fig. 1A 2 ). These are called the hemi− segments (Lanzing 1976;Géraudie 1988). Adjacent segments are separated by thin joints (Fig. 1A 1 -A 4 ). Longitudinal sec− tions of fin rays show that the joints are very narrow un− mineralised spaces between adjacent segments (Fig. 1A 3 ). Mineralised, longitudinal fibres form an osseous mantle at the periphery of some joints (Fig. 1A 4 ). In the fin rays, the axes of the osteocyte lacunae are parallel to the fibres that are oriented along the longitudinal axis of the segments ( Fig. 1A 3 , A 4 ). Cross−sections in the proximal region where the fins are covered with scales ( Fig. 1A 5 , A 6 ) show that the rays are ap− proximately cylindrical at that level. They exhibit concentric layers of cellular bone separated by cementing lines of arrested growth (LAGs). The cross−sections of the fin rays in more dis− tal portions of the fins show that the shape of the segments is modified progressively. First, a furrow appears on the inner surface ( Fig. 1A 6 ). More distally, the two hemisegments whose inner face gradually becomes concave assume a crescentic shape in section (Fig. 1A 7 ). They form a series of opposing concave segment pairs; within each segment, a clear central area appears (Fig. 1A 7 ). These segments are composed of thin concentric layers with cellular lacunae inserted between the layers, as shown by Nomarski transmitted light (Fig. 1A 8 ). The osteocyte lacunae whose axes are oriented along the longitudi− nal axis of the segments appear isodiametric in cross section ( Fig. 1B 1 ). The distribution of the Sharpey's fibres varies con− comitantly with the shape modification of the segments. In the proximal cylindrical segments, abundant Sharpey's fibres ori− ented orthogonally to the lateral surface of the fin ray connect the adjacent hemisegments like a belt, and thin bundles are present in the outer part of the hemisegments (Fig. 1A 6 ). Dis− tally, the bundles are also present but they are restricted to the outer part of the hemisegments (Fig. 1A 8 ), and they become less abundant and thinner (Fig. 1B 1 ). A few longitudinal and radial vascular canals are also present ( Fig. 1B 1 ). The lepido− trichia of Eusthenopteron are forked; each hemisegment gives rise to two more distal hemisegments (Fig. 1B 2 ).
Enlarged scales.-Enlarged scales, the hemicylindrical scales of Andrews and Westoll (1970: text−fig. 17), are located at the proximal part of the preaxial edge of the fins. They are characterised by the presence of tubercles that form rounded elevations of about 0.5-1 mm on the outer surface ( Fig. 2A, B). The holes observed at the surface of the enlarged scales represent the apertures of vascular canals (Fig. 2B). Vertical ground sections show that the enlarged scales are composed of two layers: a thin superficial bony layer whose elevations form the tubercles, and a thick basal plate composed of dense lamellar bone (Fig. 2C). The outer part of the superficial layer (the only one that forms a continuous sheet) is composed of poorly vascularised parallel−fibered bone (Fig. 2C), also called pseudo−lamellar bone (Francillon−Vieillot et al. 1990). Osteo− cyte lacunae inserted among the fibres are spindle−shaped (Fig. 2C 1 ). Bundles of woven−fibered bone located under the layer of parallel−fibered bone form the tubercles (Fig. 2C 1 ). The woven−fibered bone contains isodiametric osteocyte lacu− nae measuring 5-7 μm in diameter. Most of the scale vascu− larisation occurs in this woven−fibered, discontinuous layer (Fig. 2C), although the vascular canals obviously had to cross the superficialmost layer to reach the scale surface (Fig. 2B). Both primary and secondary vascular canals occur (Fig. 2D). The former lack delimiting cement lines (Fig. 2D insert) whereas the latter, whose walls are made of a secondary lamellar bone, are separated from the surrounding primary bone by cementing lines and lack Sharpey's fibres (Fig. 2D). Sharpey's fibres are observed in the entire superficial layer, where their direction is approximately perpendicular to the scale surface. They form loose bundles in the woven−fibered bone (Fig. 2D), and thicker bundles that perpendicularly cross the fibres of the parallel−fibered bone (Fig. 2C 2 ).
The lamellar bone of the basal part of the enlarged scales is made of superimposed strata of aligned fibres parallel to each other in each stratum, but whose direction varies by about 90°from stratum to stratum, thus forming an orthogo− nal plywood−like architecture (Fig. 2C 2 ). The thickness of each stratum is about 6-8 μm. The osteocytes of the basal plate are located in lenticular lacunae about 2.5 μm thick and 12-14 μm in diameter (not illustrated).
Regular scales.-The flat scales are regularly imbricated and cover the body, including the lateral line organ, with canals that open to the surface through large pores (Fig. 3A). Because of their overlap, only the posterior field of the scale is normally exposed; it is ornamented with thick tubercles while the lateral and anterior fields have thin radial ridges (not illustrated). In vertical sections, the scales appear to be composed of two su− perimposed layers: a superficial layer, and a thick basal plate (Fig. 3B), both of which are composed entirely of bone. Very thin superimposed growth lines were observed in the external layer. They are more obvious within the tubercles, suggesting that throughout ontogeny, a cyclic thickening occurred in the whole superficial layer, but it was more dominant in the tuber− cles (Fig. 3C, D). As in the enlarged scales, the vascularisation is largely confined to the deep part of the tubercles.
The basal plate is composed of series of superimposed plies about 10-12 μm thick (Fig. 3D). Each ply is formed by a layer of thick bundles of closely packed parallel collagen fibres oriented in the same direction. The fibre direction var− ies from one ply to the next by a specific angle of rotation. On a micrograph showing an oblique section of the basal plate, the direction of the fibres in each ply is represented by lines parallel to the fibres (Fig. 3E). The lines of successive odd and even plies form a double system of nested arches (Fig.  3E). This double system of arches corresponds to a twisted  (Giraud et al. 1978). Numerous densely packed ovoid corpuscles, Mandl's corpuscles, are observed in SEM micrographs of the mineralisation front of the basal plate (Fig. 3F, G). Their long axes are oriented in the same di− rection as the mineralised fibrils that form sheets with which they are co−aligned.
Osteocyte lacunae are present in both the superficial layer and the basal plate (Fig. 3B, E, H, I). Flat, elongated, spin− dle−shaped cell lacunae are located between adjacent plies across the whole thickness of the basal plate (Fig. 3E). Their abundance is best observed in vertical ground sections of scales embedded in toto in epon for TEM observations after a demineralisation carried out with formic acid or EDTA (Fig.  3H, I). These sections became transparent enough to be ob− served with a transmitted light microscope even if the pres− ence of opaque dots indicates that the demineralisation was not completely achieved. The osteocyte lacunae are 25 to 35 μm long and about 2.5 μm thick, and their long axis is parallel with the surrounding collagen fibrils. They are equipped with numerous long canaliculi that were occupied by cellular pro− cesses and extended between the plies (Fig. 3H, I).
These sections show that the direction of the fibrils varies, suggesting that the organisation of the fibrils in the ply− wood−like structure was preserved during the demineralisa− tion procedures for TEM observations. Indeed, the semi−thin sections perpendicular to the scale surface show that the gen− eral organisation of the scale is preserved (Fig. 4A-D), even though abundant fractures (Fig. 4B) and bacteria are visible on the surface (Fig. 4D). Staining with toluidine blue indi− cates that organic material was preserved in both layers, even though the scales are not completely demineralised; tissues that remain mineralised are not stained with toluidine blue (Fig. 4A). In section of the external layer, well−stained, thin lines parallel to the surface (Fig. 4C) may represent remains of growth lines, also observed in ground sections in light mi− croscopy (Fig. 3B, C). The basal plate observed with No− marski optics shows a surface (Fig. 4D) covered with ripples parallel to each other. At the ultrastructural level, long, com− pact fibrillary structures are composed of closely packed fi− brils (Fig. 4E). Adjacent fibrils are linked to each other by regularly spaced bridges (Fig. 4E, F). The 100 nm thick fi− brils do not show the specific striation that characterises the fibrils of type I collagen present in osseous tissues of extant vertebrates; however, the regular banding that can occasion− ally be observed in the Eusthenopteron scales may represent a remnant of the original structure (Fig. 4G). Mineral still present in thin sections after demineralisation procedures forms patches associated with the fibrils and is regularly distributed along these fibrils (Fig. 4G).

Discussion
Fin rays.-Our data on the morphology and structure of the fin rays of Eusthenopteron foordi support previous interpreta− tions (Goodrich 1904) that this taxon possesses true lepido− trichia composed of ossified, jointed and branched segment pairs ( Fig. 1A 5 , A 8 ), whereas extant dipnoans have unseg− mented, cylindrical camptotrichia (Géraudie andMeunier 1982, 1984) that do not form paired structures, unlike the lepidotrichia of Eusthenopteron. In Neoceratodus, mineralisa− tion is present along the whole length of the camptotrichia, but it is interrupted in some places, which increases flexibility. The superficial, subepidermal area is mineralised while the deep "dermal one" is not. Dipterus had true lepidotrichia (Goodrich 1904) but reduction of mineralisation was observed in more Recent dipnoids, such as Scaumenacia (Géraudie and Meunier 1984). These camptotrichia must represent an apo− morphic condition (Géraudie and Meunier 1984), as lepido− trichia were presumably present in most, if not all, Palaeozoic sarcopterygians (Janvier 1996). Lepidotrichia and campto− trichia may only share a deep homology, to the extent that they are both derived from a same morphogenetic system.
In the proximal parts of the fins that are covered with en− larged or regular scales, the basalmost segments of the lepido− trichia are circular in cross−section and are made of concentric layers of cellular bone. These basal segments do not appear much more elongated than the distal ones (Goodrich 1904;Andrews and Westoll 1970;Jeffery 2001). In that respect, Eusthenopteron foordi differs from rhizontids, whose elon− gated basal segments extend to the endoskeleton of the pecto− ral fin (Andrews 1985;Davis et al. 2001Davis et al. , 2004Jeffery 2001;Johanson et al. 2005). Distally, the hemisegments gradually acquire a crescentic cross−section. The paired hemisegments align to enclose a central area. The presence of articulations between successive segments of the fin rays in Eustheno− pteron suggests that they could bend when the fins moved.
We have identified Sharpey's fibres, whose distribution changes along the segments. In the proximal cylindrical seg− ments, Sharpey's fibres form strong bundles connecting the proximal elements to each other like a reinforcing belt. These Sharpey's fibres most probably correspond to the interlepido− trichial ligaments described in the fin rays of extant teleosts (Beccera et al. 1983). Distally, thin bundles of Sharpey's fibres are restricted to the external part of each hemisegment. Such thin bundles perpendicular to the surface of the lepido− trichia cross the dermis to reach the dermo−epithelial mem− brane in extant teleosts. The distribution of Sharpey's fibres could be related to the organisation of the lepidotrichia in a fan configuration in the paired fins. Surprisingly, in the lepidotrichia of extant teleosts, fibrils of type I collagen have a narrower diameter (about 30 nm) than those of the surrounding dermis of the lepidotrichia (Landis and Géraudie 1990). Similarly in the external layer of the elasmoid scales of teleosts, thin type I collagen fibrils (about 30 nm) differ from the thick type I collagen fibrils (about 100 nm) of the basal plate (Zylberberg and Nicolas 1982). Moreover, in the lepidotrichia and in the scale external layer, the crystals are not oriented by the thin collagen fibrils as in a regular osseous tissue. These ultrastructural data support the hypothesis formu− lated by Jarvik (1959) and Andrews and Westoll (1970) that lepidotrichia and scales (where enamel/enameloid, dentine and bone are identified) are somewhat differently−shaped manifes− tations of the same morphogenetic system (Schaeffer 1977;Zylberberg et al. 1992).
Enlarged and regular scales.-In this part, we will first consider data concerning the structure of the enlarged and regular scales and then the evolutionary trend to reduction of the scales.
Our histological and ultrastructural investigations carried out on these two dermal elements provide new data on the structure of the basal plate. In these scales, Mandl's corpuscles were identified only in the regular scales (Fig. 3F); we did not observe them in enlarged scales. Indeed, these mineralised globules located ahead of the mineralisation front were first considered as a characteristic of the basal plate of elasmoid scales of teleosts (Schönbörner et al. 1981). However, in subsequent studies, Mandl's corpuscles were described in elasmoid scales of other taxa, such as the holostean Amia (Meunier 1980;Meunier and Poplin 1995) and the dipnoan Protopterus (Meunier and François 1980;Zylberberg 1988;Meunier and Poplin 1995). They were also observed in the early stage of formation of ganoid scales in polypterids (Daget et al. 2001). Moreover, similar corpuscles have been described in the basal plate of the osteoderms of a reptile, the squamate Chalcides viridanus (Zylberberg et al. 1992). The basal plates of the elasmoid scales in the Gymnophiona (caecilians) do not have such mineralised globules ahead of the mineralisation front (Zylberberg et al. 1980;Zylberberg and Wake 1990). Functional and/or evolutionary significance of such mineral− ised corpuscles that precede the mineralisation front in the basal plate of the integumental skeletal elements sheathing the body of various osteichthyans remains unsolved.
The ultrastructural preservation of the twisted plywood− like structure of the basal plate (Fig. 4E), despite the cracks and the presence of bacteria, may have been facilitated by the high density of the fibrils within the plies. The twisted ply− wood−like structure of the regular scales is more compact than the orthogonal plywood−like structure of the enlarged scales. In the plies where the fibrils are closely packed, abun− dant bridges connecting the fibrils are preserved, even if these bridges may have formed during fossilisation and may not correspond to the bridges composed of glycosamino− glycans and glycoproteins found in fresh osseous tissues (Fawcett 1994;Vogel 1994). The fibrils show an occasional striation evoking that of the type I collagen of osseous tis− sues. Their thickness (100 nm in diameter) is frequently ob− served in the fibrils of the basal plate in elasmoid scales of extant teleosts (Zylberberg et al. 1992). Studies dealing with the ultrastructural preservation of collagen in fossil bone showed a loss of striation characteristic of bone type I colla− gen fibrils, and of the ability of these fibrils to be stained (Pawlicki et al. 1966;Doberenz and Wikoff 1967). The pro− cesses that affect bone preservation in fossils are complex because bone consists of interrelated mineral and organic components whose degradation depends on intrinsic factors, such as composition and structure, and extrinsic factors, such as the conditions of burial and post−depositional history (Schweitzer et al. 2008).
The enlarged and regular scales differ owing to their shape, the type of mineralised tissues forming the superficial layer, the organisation of the plywood−like structure of the basal plate (either orthogonal for enlarged scales or twisted in regular scales; see below), and its mineralisation.
In enlarged scales, the superficial layer is proportionately thicker than that of regular scales and its outer surface does not show exposed or overlapped fields with different orna− mentations like the regular scales. This layer is composed of outer parallel−fibered bone covering woven−fibered bone, both being vascular bone. The basal plate is composed of a largely avascular orthogonal plywood−like structure forming lamellar bone apparently devoid of Mandl's corpuscles; the mineralisation front is regular.

Comparisons
The present study supports the hypothesis that only the regular scales of Eusthenopteron are elasmoid scales since they show characteristics of such scales as defined by Bertin (1944): they are thin lamellar and imbricated plates composed of a thin mineralised and ornamented outer layer overlying a thick basal layer with a plywood−like structure; nevertheless, we consider enlarged scales and regular scales to be homologous.
The superficial layer of the scales of Eusthenopteron re− sembles that in the scales of extant dipnoan taxa that also lack odontodes (Meunier and François 1980;Zylberberg 1988), especially Neoceratodus (Meunier and François 1980). The mineralised tissue of the superficial layer also has similarities to that of elasmoid scales of extant teleosts (Schönbörner et al. 1979;Zylberberg and Nicolas 1982). The basal plate is composed of thick collagen fibrils organised in regular su− perimposed plies forming a twisted plywood−like structure, the isopedine (Francillon et al. 1990).
The enlarged and regular scales of Eusthenopteron foordi lack several tissues and structures that were primitively pres− ent in the earliest sarcopterygians, such as the early Middle to Late Devonian dipnomorph Glyptolepis and the Middle Devo− nian tetrapodomoprh Osteolepis, and that persisted in some more recent tetrapodomorphs, such as the Early Permian Ecto− steorhachis (Thomson 1975). The latter had cosmoid scales comprised of three layers (Ørvig 1951, 1968Sire et al. 2009). The most superficial layer was formed of cosmine, a special arrangement of odontodes and a pore canal system (presum− ably representing mostly vascular spaces); the odontodes are formed of orthodentine, with a superficial layer of enamel or enameloid. These sometimes formed odontocomplexes, rather than completely distinct odontodes. Beneath the cosmine, a thick layer of spongy bone was typically present (Thomson 1975), and was underlain by a much less densely vascularised layer of lamellar bone similar to the basal layer of the enlarged and regular scales in Eusthenopteron. The scales (both en− larged and regular) of Eusthenopteron lack the cosmine, the enamel (or enameloid) and the dentine, although bony tuber− cles occur. These tubercles may have a hydrodynamic func− tion (Burdak 1986). Eusthenopteron also lacks the layer of spongy, densely vascularised bone found in older, more basal sarcopterygians (e.g., Glyptolepis, Osteolepis), or even the more recent (Early Permian) tetrapodomorph Ectosteorhachis (Thomson 1975).
Eusthenopteron shares the loss of the cosmine (enamel/ enameloid and dentine) with stegocephalians. However, the loss of the spongy bone layer is probably convergent with sev− eral extant tetrapods, such as gymnophionans (Zylberberg et al. 1980;Zylberberg and Wake 1990), that typically lack spongy bone in their scales, or lepidosaurs, that lack spongy bone in their osteoderms   fig. 10). Spongy bone is found in some extant tetrapod osteoderms, such as in crocodilians and mammals   fig. 10). More importantly, the Late Permian temno− spondyl Australerpeton cosgriffi retained a spongy bone layer in its ventral scales (Dias and Richter 2002). There is currently a debate about the affinities of temnospondyls. The traditional viewpoint, still upheld by some authors (e.g., Ruta and Coates 2007), is that temnospondyls are stem−amphibians, but some recent studies suggest that they represent stem−tetrapods (Laurin 1998;Marjanović and Laurin 2009). Under both hy− potheses, the condition found in Permian temnospondyls is more likely to be primitive than that found in extant gymno− phionans. The lack of a vascular layer in the scales of our spec− imens contrasts with the current schematic depiction of its scales as having a densely vascularised middle layer (Vicka− ryous and Sire 2009: fig. 10). The extent to which our findings apply to other tristichopterids will need to be determined by further histological investigations.
This argument of a convergent loss of the vascular layer between Eusthenopteron and gymnophionans rests on the acceptance of the homology between genuine scales and gastralia of early stegocephalians. The only histological study of sections of the ventral dermal ossifications of early stego− cephalians suggests that these are indeed genuine scales, rather than osteoderms, based especially on their taxonomic distribution (Dias and Richter 2002: 477;Witzmann 2007). Another supporting argument is their overlapping pattern and their lack of dermal sculpturing. Thus, these gastralia may represent transformed scales, rather than new structures.

Conclusions
Our new data on the dermal skeleton of Eusthenopteron are further evidence of a pervasive evolutionary trend towards a lightening of the dermal skeleton in osteichthyans and other vertebrates (Zylberberg et al. 1992;Sire et al. 2009;Vickaryous and Sire 2009). A similar trend is also obvious in the geologically most recent osteostracans (jawless ver− tebrates), which were slightly older (Middle Devonian) than Eusthenopteron. In osteostracans, the reduction in thickness or complexity of the dermal skeleton probably ap− peared convergently at least three times in the most recent lineages: once in Afanassiaspis porata Laurin 1999, 2001a), once in Yvonaspis, and once in Balticaspis latvica (Otto and Laurin 2001b).
This trend is more advanced in the regular than in the en− larged scales of Eusthenopteron. The lightening probably improves swimming abilities (Burdak 1986) without impair− ing various other functions of the integument, especially pro− tection and hydrodynamics. The ornamentation of the scale surface probably improved hydrodynamics of the animal by regularising water flow in the limiting layer at the surface of the skin, experimentally demonstrated on two extant species (one chondrichthyan and one teleostean) by Burdak (1986). Eusthenopteron is considered to be an ambush predator (Clack 2002), so it needed relatively high acceleration abili− ties to seize its preys. Both the ornamentation (reducing hy− drodynamic drag) and lightening of the scales (reducing iner− tia) may have contributed to improving swimming perfor− mance. The lesser degree of lightening of the enlarged scales may be explained by biomechanical constraints: they may have been subject to substantial mechanical stress, because of their position near the base of the paired fins.