Emended Diagnois: Paraclupeine with comparatively low body depth. Surface of opercle with striations. Teeth on jaws and palate very weakly developed. Anterior dorsal scutes slightly longer than broad and ornamented with ridges; posterior ones broader than long and ornamented with tubercles or tubercles plus ridges. Sharp spine extending from median keel of posterior few dorsal scutes pointing posteriorly rather than posterodorsally. Origin of pelvic fin opposite to middle point of base of dorsal fin. Procurrent rays 5 and 6 on each side of base of caudal fin.

Type Species: †Ellipes branneri Jordan, 1910, the only species (based on page priority).

Holotype: CM 5249/1 and CM 5249/2. Note that only CM 5249/1 was designated the type by Schaeffer (1947), and this was also the type specimen chosen by Jordan for †Ellimma branneri. In fact, CM 5249/2 is the counterpart to CM 5249/1, although this was not noticed in previous descriptions. The holotype is a small fish, with a standard length of 38.5 mm (measured from the counterpart CM 5249/2, which is more complete than CM 5249/1).

Additional Material: AMNH 10046–10062; CM 5248/10, 11, 21, 36, 39; 5249/2 (counterpart of 5249/1), 3, 25, 27, 33, 49, 52, 54, 55, 56 (counterpart of 5248/11), 125, 130 (counterpart of 5249/125), 175 (counterpart of 5249/52).

Diagnosis: As for genus.

Horizon and Localities: From Riacho Doce, Sergipe Basin, Brazil (Muribeca Formation, Lower Cretaceous, Aptian–Albian, = “Alagoan” local stage). The age of this unit is well established from palynomorphs and foraminifers (Feijó, 1994).

Comments: The type specimen of †Ellimma riacensis (CM 5248/4) has a standard length of 102.2 mm and is therefore a somewhat larger individual than the type of †E. branneri. It is also much better preserved, and it consequently shows more morphological details than the type of †E. branneri. We were however unable to identify any unequivocally unique features of †Ellimma riacensis, and we consequently agree with Schaeffer (1947) that †Ellimma riacensis is a synonym of †E. branneri. We regard †E. branneri as the type species of the genus on the basis of page priority. Our observations led us to the conclusion that all the †Ellimma specimens studied by Schaeffer indeed represent a single species, but we must also admit that the holotype of †E. riacensis (CM 5248/4) provides some morphological information, including details of the skull roof ornamentation, dorsal scute morphology, and entopterygoid teeth, that are not discernible in the holotype of †E. branneri. There is inevitably some circularity in our inclusion of these features in our description of †E. branneri, but we found no evidence to retain †Ellimma riacensis as a distinct species, and this taxon is regarded here as a subjective synonym of †E. branneri. Thus, we conclude there is only one species of †Ellimma represented in the Riaco Doce material, rather than two sympatric species.

General Shape: The overall outline of †E. branneri resembles that of †Paraclupea chetungensis (Chang and Chou, 1977; Chang and Grande, 1997), with a markedly convex ventral outline and a gently curved dorsal outline rising to an apex at the origin of the dorsal fin (figs. 1A, B, 2A–C). The maximum body depth occurs at the origin of the dorsal fin. The standard length of individuals in our samples varies from 20.8 (CM 5249/125) to 102.2 mm (CM 5248/4); in other words, our largest specimens are five times longer than the smallest. The proportion between the body depth and the standard length is therefore somewhat variable, usually being approximately 41–44% standard length, but rarely 36 or 50%.

Skull Roof: The bones of the braincase in †E. branneri are usually so badly crushed that it is impossible to recognize bone margins or even various recesses, fossae, and bullae within the bones; importantly, it has not been possible to determine whether a recessus lateralis was present (an important clupeiform character absent in primitive clupeomorphs; Grande, 1985a). Few features of the braincase can be described. In the anterior part of the skull roof, the posterior portion of the dermethmoid (seen in CM 5249/49) bears a lateral process on both sides and a pair of processes extending posteriorly above the frontals. In larger specimens the posterior one-third of the frontals, as well as the parietals, are strongly ornamented with ridges radiating from their growth centers. The dorsal part of the supraoccipital (seen in CM 5249/56) is small and triangular and shows indications of ridges radiating from the posteromedian point of the bone anterolaterally. In smaller individuals (e.g., CM 5248/39, standard length approximately 51 mm) the skull roof bones are smooth and without sculpture (fig. 3). Unlike in †Paraclupea chetungensis there is no anterior fontanelle between the frontals. The parietals meet at the midline and are not separated by the supraoccipital (in clupeiforms the supraoccipital separates the parietals; Grande, 1985a). As far as can be determined the supraoccipital crest is small and low (CM 5249/56, 5248/11, and 5248/39). The supraorbital sensory canal is enclosed in a bony ridge and extends from the frontal to the parietal. The supratemporal commissure passes through the parietals.

A comparatively broad and flat ridge runs the length of the posttemporal along its midline, and its broader posterior part is ornamented with ridges in larger individuals. The dorsal (epiotic) limb of the posttemporal is long and narrow anteriorly but becomes broader posteriorly, while the ventral limb of this bone is thin. The lateral line sensory canal runs diagonally through the supracleithrum.

Orbital Region: Faint impressions of three infraorbital bones can be seen in a few specimens, for example, AMNH 10048 (fig. 2A) and 10050. Two detached infraorbitals (perhaps the second and third) are exposed on CM 5249/52 and 5249/175 (part and counterpart; fig. 4). The first two infraorbitals seem to be narrow and elongated, but the third is broader and shorter. The infraorbital sensory canal runs along the orbital margin of all three bones. Parts of the sclerotic ring were seen in several specimens, although no details were evident.

Parasphenoid and Entopterygoid: No teeth are present on the parasphenoid. A basipterygoid process is recognizable on the latex cast of AMNH 10051. In most specimens of †Ellimma branneri no teeth were observed on the entopterygoid, although faint traces of tiny broken tips of teeth are sometimes present (e.g., AMNH 10047, 10048 and CM 5248/4). In contrast, †Ellimmichthys and †Paraclupea have an entopterygoid dentition that is relatively better developed, and the reduced dentition in †Ellimma is a distinguishing feature of the genus within paraclupeids.

Jaws: In †Ellimma branneri the articulation of the lower jaw is positioned more or less below the posterior margin of the orbit (figs. 2A–C, 5B). The dentary, angulo-articular, and retroarticular are seen clearly in AMNH 10048 (fig. 5A). A well-developed coronoid process is present on the dentary. The mandibular sensory canal runs above the lower margin of both the dentary and the angulo-articular (fig. 5B). The maxilla is long and extends behind the jaw joint. In most specimens of †Ellimma branneri the jaw teeth are not visible, but a few fine, conical teeth are recognizable on the oral margin of the premaxilla and dentary in the latex peel of AMNH 10057. Very fine serrations are also present on the oral margin in one detached maxilla (CM 5248/11 and CM5249/56, part and counterpart; fig. 6C, D).

Collectively, these observations suggest that the entire jaw dentition of †Ellimma branneri consists only of very small teeth, and is much reduced in comparison with both †Ellimmichthys longicostatus and †Paraclupea chetungensis, where a well-developed dentition is present both on the jaws and the palatal surface of the entopterygoid (Chang and Chou, 1977; Chang and Grande, 1997). The presence of very small entopterygoid teeth in †Ellimma branneri may be related to a microphagous diet, as in many Recent clupeomorphs. †Ellimma branneri possesses two supramaxillae (a primitive teleostean state), of which the posterior one is larger. Both supramaxillae are ornamented by fine shallow grooves on their external surface.

Opercular Series and Hypobranchial Apparatus: A detached but well-preserved opercle with an almost complete margin is present in CM 5249/52 and 175. It is deep and rectangular in shape, with a protruding anteroventral corner. The ventral part of the opercle in most specimens of †Ellimma branneri is ornamented on its lateral surface by striations which radiate ventrally from the area of attachment (fig. 4A, B), but the surface of the opercle in the smallest specimen is smooth, with no striations (CM 5249/125 and 130, standard length 20.8 mm). The subopercle shows the anterior ascending process which is usually covered by the opercle (CM 5249/52; fig. 4A). The preopercle has two arms forming an obtuse angle; its vertical arm is longer than the horizontal one, which contains three or four branches of the preopercular sensory canal (fig. 5B).

The hyomandibular shaft is narrow, with a single head and no anterodorsal process. The quadrate is thick, with the symplectic inserted in a notch above its posteroventral margin, as in teleosts generally (fig. 5A, B). The anterior ceratohyal is rectangular in shape, longer than deep, and contains an elongated oval bericiform foramen (AMNH 10048, 10050; fig. 5A). This opening is traversed by a groove for the hyoidean artery on the outer face of the anterior ceratohyal. Absence of the bericiform foramen is considered to be a synapomorphic character of clupeiforms by Grande (1985a). There are approximately 10 branchiostegal rays (AMNH 10048; fig. 5A). An outline of the urohyal was detected on AMNH 10048, but few details are discernible because the bone lies beneath the branchiostegal rays; it is narrow and triangular, with a long, narrow anterior process.

Vertebral Column and Fins: Two longitudinal ridges are present along the lateral side of each vertebra (fig. 2). The total vertebral count in †Ellimma branneri could only be estimated because the anteriormost centra are covered by the opercle in all the available specimens. We estimate that the vertebral column originally comprised 36–38 centra, of which 20–22 are abdominal (based on the number of pairs of ribs) and 15–16 are caudal. There are eight predorsal bones with thin anterior and posterior bony expansions (fig. 5A, B).

In †Ellimma branneri the dorsal and anal fins are of more or less equal size (figs. 1, 2). The dorsal fin contains 2 unbranched and 14 branched rays, and is supported by 14–16 pterygiophores. In both AMNH 10048 and 10050 the anal fin has 15 rays and 14–15 pterygiophores. The pelvic fin is small and contains about six or seven rays (seven are present in CM 5249/27). Its insertion is approximately opposite to the middle point of the base of the dorsal fin. The pectoral fin is large but does not reach the origin of the pelvic, and has at least 12 fin rays. The cleithrum is large and curved. No postcleithra have been recognized in the available specimens.

Scutes: †Ellimma branneri possesses a complete dorsal scute series (unlike in †Armigatus), extending from the back of the head to the origin of the dorsal fin (figs. 1A, 2A, C, 7). The entire series includes 12–14 scutes, each of which is overlapped anteriorly by the next (AMNH 10048, 10060, and CM 5249/2). The more anterior scutes are generally rounded in shape and are slightly longer than broad (like those illustrated by Grande, 1982a: fig. 23), but those farther posteriorly tend to be broader than long. The last three or four scutes in the dorsal series are expanded laterally and are much broader than those farther anteriorly. All the dorsal scutes in †Ellimma branneri have a median keel, but scute margins are not always well preserved, and it is therefore uncertain whether the keel of every scute was extended as a spine (as in †Ellimmichthys and †Paraclupea; Chang and Grande, 1997). The last few scutes in †Ellimma branneri are extended into a sharp spine pointing posteriorly. This arrangement is slightly different from that seen in †Ellimmichthys and †Paraclupea, where the spine projects from the scute in a more posterodorsal direction. In †E. branneri most of the scutes are ornamented with ridges, which arise laterally from the posterior end of the keel and then turn posteriorly to pass almost parallel to the keel. The ornamentation of the posteriormost scutes is rather different, however, as some are covered with tubercles arranged in rows, and others have both tubercles and ridges (fig. 7A, B). Almost the entire upper surface of each scute is ornamented in †E. branneri; by contrast, in †Paraclupea chetungensis, only the posterior half or third of the scute upper surface is ornamented.

The abdominal scutes in †E. branneri form a continuous series extending from the lower end of the cleithrum to just in front of the origin of the anal fin (fig. 2A–C). There are 27–30 abdominal scutes, of which 8–10 lie behind the origin of the pelvic fin. The abdominal scutes behind the pelvic fin are smaller than in †Ellimmichthys longicostatus (see Jordan, 1910: pl. XI).

Scales: The scales are thin, with fine concentric growth lines, although no vertical circuli were observed like those in †Diplomystus and many other clupeiforms. From the material available it could not be determined whether lateral line scales are present.

Caudal Skeleton and Tail (fig. 8A, B): The structure of the caudal skeleton in †E. branneri closely resembles that of †Ellimmichthys and †Paraclupea. Collectively, all these taxa have a caudal fin morphology that differs from the clupeiform condition. There are two free ural centra, the first ural centrum is approximately equal in size to the first preural centrum, and there are six hypurals. The first hypural is in close contact with (but not fused to) the first ural centrum, although the second hypural is fused to this centrum. The third hypural is comparatively narrow, not expanded posteriorly, and there is a gap between the second and third hypurals. The parhypural is fused with the first preural centrum, which has a short neural arch. As many as three epurals are present (e.g., AMNH 10060), although in most specimens only one or two are visible. There are three uroneurals, all of which are autogenous. The first uroneural is long and thick, extending to the posterodorsolateral side of the second preural centrum, but the second and third uroneurals are much smaller.

The caudal fin is deeply forked (figs. 1, 2, 8). Its upper lobe contains one unbranched and nine branched rays, and its lower lobe has one unbranched and eight branched rays. The proximal end of the lowermost branched ray from the upper lobe is bifurcated into upper and lower branches, as is the uppermost branched ray from the lower lobe. In both cases, the upper branches are longer, broader, and more flattened than the lower ones. There are five or six procurrent rays on both the upper and lower margins of the peduncle in front of the caudal fin, and one caudal scute is present on the upper margin of the peduncle anterior to the procurrent rays.

Emended Diagnosis: Differs from other genera of †Paraclupeidae in having a high number of dorsal scutes; posterior border of dorsal scutes pectinate; supraoccipital crest very well developed and high; posttemporal with long and slender epiotic branch; entopterygoid with robust teeth along dorsal margin of buccal surface; seven or eight predorsal bones; H₃ much expanded posteriorly. No gap between H₂ and H₃, lowermost ray of upper caudal fin lobe bifurcates proximally, with two branches of more or less equal size, uppermost ray of lower caudal fin lobe slightly bent and enlarged.

Type Species: †Diplomystus dentatus Cope, 1877

Other Included Species: †D. shengliensis Zhang et al., 1985; †D. birdi Woodward, 1895; †D. dubertreti Signeux, 1951.

Holotype: SOF 790001, first illustrated in Zhang et al., 1985: pl. 1, fig. 1 (see figs. 9, 10A, 11, 12A here).

Additional Material: SOF 790002 (fig. 10B), 790003 (fig. 12B).

Horizon and Localities: Top of series 4 to bottom of series 3 of the Shahejie Formation, Middle Eocene; SOF 790001 and 790002 from Bore Lai 38–8, depth from the ground surface 2628m and 2668m, respectively, SOF 790003 from Bore Lai 1–9, depth from the ground surface 2622 m, Kenli county, Shandong Province, East China.

Emended Diagnosis: †Diplomystus with elongate fusiform body, depth and standard length ratio of about 40%; with number of dorsal scutes (41) and dorsal fin rays (1, 13) slightly higher than in †D. dentatus (highest number of dorsal scutes 40 and that of dorsal fin rays 1, 12), but number of ribs (15 pairs) and number of vertebrae (around 42), namely number of abdominal vertebrae (around 17), lower than in the latter (17–18 pairs of ribs and 20 abdominal vertebrae).

Etymology: sheng-li-, transliteration of the Chinese word “victory”, the species named after Sheng-li Oil Field.

General Shape: The only more or less complete specimen of †D. shengliensis is the holotype SOF 790001 (fig. 9), which fortuitously lies entirely within the diameter of a drilling core. This example is nevertheless somewhat distorted in the abdominal region, and its scale cover is missing. Data from this specimen are supplemented by SOF 790002, in which the anterior portion of vertebral column and ribs is better preserved; and SOF 790003, which has a well preserved caudal skeleton and fin rays. Some skeletal features cannot be observed in any of the available material, and our description is therefore incomplete.

The total length of the holotype of †D. shengliensis (SOF 790001, the only complete specimen) is 5.5 cm. The other fragmentary specimens evidently represent larger individuals, although their original size is uncertain. In the holotype, the body is elongate and fusiform, reaching a maximum depth (approximately 40% of standard length) a short way behind the head (fig. 9). The dorsal margin of the body curves upward a short distance behind the head, and then descends gradually to the caudal peduncle, without an abrupt angle at the origin of the dorsal fin. The ventral outline is moderately convex.

Skull Roof: The head can be studied in the holotype and in SOF 790002 (fig. 10). It is not easy to determine the arrangement of the bones along the midline of the skull roof in †D. shengliensis because the specimens are all laterally compressed. The supraoccipital is situated far posteriorly and probably did not separate the parietals, which therefore may have met at the dorsal midline. The external surfaces of the frontals and parietals are generally smooth, apart from ridges containing the sensory canals. The supraorbital sensory canal extends from the frontal into the parietal, where it meets a prominent curved ridge which appears to contain part of the supratemporal commissure. The supraoccipital crest is well-developed, being high and triangular in shape (fig. 10B), but it cannot be determined whether this bone contained part of the supratemporal commissure. The posttemporal has a long and slender epiotic limb (fig. 11).

Orbital Region: Nothing is preserved except part of the sclerotic ring (SOF 790002; fig. 10B).

Parasphenoid and Entopterygoid: No teeth were observed on the parasphenoid, and it is apparently edentulous in †D. shengliensis. A basipterygoid process is present in the holotype SOF 790001 (fig, 10A), although no features are descernible. In this specimen, the left entopterygoid is observed in situ against the parasphenoid margin, while the right entopterygoid has its buccal side turned outward and its posterior end displaced downward. This bone therefore occupies an oblique position, and it was mistakenly identified as the ectopterygoid by Zhang et al. (1985). The buccal surface of the entopterygoid bears numerous teeth, of which the posterodorsal ones adjacent to the parasphenoid are much stouter than the rest.

Jaws: The dentary and premaxilla bear small conical teeth, and the oral margin of the maxilla is finely serrated. There are two supramaxillae, but unlike in †Paraclupea and †Ellimma these bones both have smooth surfaces.

Opercular Series and Hypobranchial Apparatus (fig. 10A, B): The bones of the opercular series are smooth in †D. shengliensis. The vertical branch of the preopercle is much longer than the horizontal branch. In SOF 790001 there is a bone covered with stout teeth, in occlusion with the left entopterygoid; we interpret this as the basibranchial bone and its associated toothplate.

Vertebral Column and Fins: The vertebral column is fairly well preserved in the holotype of †D. shengliensis, but the exact number of abdominal vertebrae cannot be determined because the anterior portion of the vertebral column is distorted (fig. 9). We estimate that there were 42 vertebrae, 25 of which are caudal. In SOF 790002 (where the anterior portion of the vertebral column is better preserved) there are 17 abdominal vertebrae and 15 paired ribs. There are seven predorsal bones on the holotype of †D. shengliensis, with both anterior and posterior bony expansions (fig. 11).

The dorsal fin of †D. shengliensis contains one unbranched and 13 branched fin rays (fig. 9). The anal fin is very long and contains at least 39 fin rays. The pterygophores of both dorsal and anal fins are badly preserved in all the available specimens.

Scutes: The dorsal scute series in the holotype, SOF 790001 is shown in figure 11. As the stereophotographs illustrate, the dorsal scutes of †D. shengliensis are much broader than long, with a prominent median keel and a pectinate posterior margin, but the scute surface is smooth and lacks ornamentation. There are 41dorsal scutes on the holotype of †D. shengliensis, but the number of abdominal scutes is uncertain because of poor preservation. These scutes are quite similar in shape to those of other clupeiforms, and as in †D. dentatus the ventral part of the posterior border is pectinate in the anteriormost few scutes.

Scales: Scale morphology cannot be determined in any of the specimens of †D. shengliensis, but detached scales associated with them resemble those referred to †Knightia bohaiensis by Zhang et al. (1985: fig. 16).

Caudal Skeleton and Tail: Only parts of the caudal skeleton and fin are preserved in the holotype, SOF 790001, but more can be seen in SOF 790003 (fig. 12). There are seven hypurals in †D. shengliensis, of which the first is in contact with the first ural centrum and the second is fused to it. The third hypural fills the entire space between the second and the fourth hypurals, unlike in †Paraclupea, †Ellimma, and †Ellimmichthys, where there is a gap. The parhypural in †D. shengliensis is apparently fused with the first preural. The latter carries a short neural arch protruding backward into a very sharp spine. There are three epurals and three uroneurals in †D. shengliensis, the first of which extends anteriorly to the second preural.

The caudal fin is deeply forked in †D. shengliensis, and the lower lobe is slightly longer than the upper. There are 19 principal fin rays (as in primitive teleosts generally), of which the 2 lateral ones are unbranched. The proximal end of the lowermost ray from the upper lobe bifurcates into two short branches of approximately equal size. The corresponding end of the uppermost ray of the lower lobe is slightly bent and spatulate.

Phylogenetic Comments: Jordan (1910: 23) noted that the herringlike fishes he recognized among fossils collected in 1907 by John Casper Branner from Riaco Doce, Brazil “… have the general traits of †Diplomystus dentatus, with the short anal and fewer vertebrae of †Knightia, while at the same time their squamatiom seems to be different from both …”.

Schaeffer (1947) placed the genus †Ellimma into synonymy with †Knightia, which considerably extended the stratigraphic range of the latter (until then it had only been discovered in Eocene strata of western North America; Jordan, 1907). He also placed †Ellimmichthys into synonymy with †Diplomystus, despite admitting there were similarites between †Ellimma branneri and †Diplomystus in the ovate shape of the dorsal scutes and the absence of scute lateral wings (Cope, 1877), as well as similarities in the size and shape of body scales. Schaeffer (1947) also concurred with Jordan (1910) that the dorsal scutes of †Knightia branneri differ from those of North American †Knightia in being wider and in lacking a posterior median spine.

It is possible that such morphological differences, however slight, might have been given greater emphasis had it been realized that the Brazilian taxa were in fact far older than those from North America. Jordan (1910) commented that the age of the Riacho Doce shales is “… Lower Eocene, possibly but not probably Upper Cretaceous”, and Schaeffer (1947) also considered them to be Eocene. †Ellimmichthys from Bahia was nevertheless considered to be of Upper Cretaceous age by both Jordan (1910) and Schaeffer (1947). In fact, the fossils from both Alagoas and Bahia are Lower Cretaceous in age; †Ellimma from the Riacho Doce shales of Alagoas (Muribeca Formation) dates from the Aptian–Albian (Alagoan local stage), and †Ellimmichthys from the sedimentary sequence in Bahia (Marfim Formation) is even older (late Hauterivian–early Barremian) (for references, see Maisey, 2000).

Grande (1982a, 1982b, 1985a) restored the generic name †Ellimma and removed †E. branneri from the genus †Knightia on morphological grounds: (1) †Ellimma has two supramaxillae, whereas †Knightia has only one; and (2) the dorsal scutes of †Ellimma show a complex pattern of sculpture, whereas in †Knightia they do not. Grande (1982a, 1985a) also resurrected the genus †Ellimmichthys, again on morphological grounds, and thus he essentially restored the systematic distinctions recognized by Jordan (1910, 1913), albeit within a far more comprehensive and refined phylogenetic framework for clupeomorph fishes. In Grande's (1985a) phylogenetic analysis †E. branneri is not closely related to †Ellimmichthys or †Diplomystus, and is placed instead within Clupeidae (i.e., within Clupeiformes, rather than †Ellimmichthyiformes); as such, †Ellimma would be the only clupeine having dorsal scutes.

Unfortunately, there is little evidence in support of that conclusion. Grande's (1985a) single morphological character supporting monophyly of the Clupeidae (presence of two long, rodlike postcleithra; Grande, 1985a: character 23) has not been identified in †E. branneri. Clupeids are nested within the Clupeoidea, whose single synapomorphy (increase in pleural rib to preural vertebrae ratio; Grande, 1985a: character 22) is certainly present in †E. branneri, but is also a feature of †Ellimmichthys (where it was regarded as convergent by Grande, 1985a: 263). †Ellimma branneri also lacks three of the derived characters of clupeoids (fusion of the first uroneural with the first preural centrum, reduction in relative size of the first ural centrum, separation of the parhypural from the first preural centrum; Grande, 1985a: characters 14, 15 17) and is unknown with respect to the fourth (absence of lateral line scales; Grande, 1985a: character 16). Finally, †Ellimma branneri lacks two of the three synapomorphies shared by the Clupeiformes (parietals separated by supraoccipital, beryciform foramen lost; Grande, 1985a: characters 10, 11) and the third (presence of a recessus lateralis, Grande, 1985a: character 9) is unknown.

Thus, none of Grande's (1985a) characters support the inclusion of †Ellimma within the Clupeidae, Clupeoidei, or Clupeiformes, and its inclusion within Clupeoidea is only supported by a single ambiguous character which could be convergent with clupeoids, as in †Ellimmichthys. Grande (1985a) found a single character to unite his †Ellimmichthyidae (lateral expansion of dorsal scute “wings” which give scute a subrectangular shape; Grande, 1985a: character 7; this feature is also shared with †Diplomystus). In †Ellimmichthys and †Paraclupea, all the dorsal scutes are broader than long, as are the posterior dorsal scutes in †Ellimma branneri (although its anterior dorsal scutes are slightly longer than broad); thus, lateral expansion of dorsal scutes remains a synapomorphy of † Paraclupeidae (= †Ellimmichthyidae), including †Ellimma, but the condition is clearly variable. In addition, †Ellimma shares several generalized clupeomorph characters with †Ellimmichthys and †Paraclupea, such as presence of abdominal and dorsal scutes, enclosure of the supratemporal commissure within the parietals, and fusion of the second hypural with the first ural centrum, although these similarities do not help elucidate its relationship within clupeomorphs.

Two derived characters are shared by †Ellimma, †Ellimmichthys, and †Paraclupea and distinguish them collectively from †Diplomystus: strongly sculptured skull roofing bones, with ridges radiating from the growth center, and dorsal scute ornament of prominent ridges. It is concluded from this that †Ellimma branneri (Jordan, 1910) is closely related to †Ellimmichthys longicostatus (Cope, 1886), from the Hauterivian–Barremiam of Brazil, †E. goodi (Eastman, 1912), from the late Aptian of Equatorial Guinea, and †Paraclupea chetungensis Sun, 1956, from the Lower Cretaceous of southeastern China. All these taxa are grouped here in the subfamily †Paraclupeinae.

†Ellimma can be distinguished at generic level from these other taxa by the ornamentation on the last few dorsal scutes, where there are tubercles (or tubercles plus ridges); in †Ellimmichthys and †Paraclupea the ornamentation of all dorsal scutes consists only of ridges. Additionally, the jaw and palatal dentition of †Ellimma is reduced in comparison with †Ellimmichthys and †Paraclupea. The origin of the pelvic fin also differs. In †Ellimma it is opposite to the middle point of the dorsal fin base, whereas in †Ellimmichthys longicostatus it is in advance of the dorsal fin, and in †Ellimmichthys goodi the pelvic fin inserts opposite to the posterior third of the base of the dorsal fin; in †Paraclupea chetungensis the pelvic fin lies behind the origin of the dorsal fin. There are also fewer procurrent rays on each side of the peduncle (approximately five in †Ellimma; in †Ellimmichthys and †Paraclupea there are eight or nine rays on the upper side and four to six on the lower side). These differences suggest that †Ellimma is a distinct genus from †Ellimmichthys and †Paraclupea.

Observed meristic characters of †Ellimma branneri agree more closely with those of the two species of †Ellimmichthys than with †Paraclupea (see table 2) The number of vertebrae, dorsal fin rays, dorsal scutes and abdominal scutes in †Ellimma are respectively, 36–38, ii, 12–14, 12–14, and 27–30. Those in †Ellimmichthys are 35–38, 12–15, 12–13, and 28–32. In †Paraclupea the number is noticeably higher (41, 17–18, 18, and 38–43). Furthermore, in †Ellimma and †Ellimmichthys almost the entire surface of each dorsal scute is ornamented, whereas in †Paraclupea only the posterior third or half is ornamented.

†Ellimna guineensis from Equatorial Guinea, West Africa (apparently misspelled as “†Ellimna”; Gayet, 1989), was referred to Clupeidae and has been considered close to †Diplomystus goodi. Grande (1982a) previously noted that the characters of †”Diplomystus” goodi mentioned by Taverne (1975) are all primitive for clupeomorphs, however, and do not clearly place this taxon in †Diplomystus; he consequently removed it from †Diplomystus and placed it in Clupeomorpha, incertae sedis. Gayet (1989) regarded two characters (presence of two supramaxillae and ornamented dorsal scutes) as diagnostic of the genus †Ellimna, but the first is a primitive teleost feature and the other is present in †Ellimmichthys and †Paraclupea. Chang and Grande (1997) considered that both †Ellimna guineensis and †”Diplomystus” goodi should be placed within †Ellimmichthys.

Based on the original description by Silva Santos and Correa (1985), †Scutatuspinosus itapagipensis displays some features suggesting paraclupeine affinity. Its dorsal scutes are progressively larger and wider caudally (as in †Ellimmichthys, †Paraclupea, and †Ellimma). They are also ornamented with ridges (a paraclupeine character), but lack tubercles (as in †Ellimmichthys and †Paraclupea). It is unknown whether the skull roofing bones are sculptured with ridges, as in paraclupeines. The origin of the pelvic fin is anterior to the dorsal fin in †Scutatuspinosus itapagipensis, as in †Ellimmichthys longicostatus, not beneath it, as in †Ellimma branneri. Unfortunately few of the described features in this taxon are phylogenetically informative, and it would benefit from a revision, but our observations suggest that †Scutatuspinosus is also a paraclupeine. Its caudally-expanded dorsal scute series is an unusual feature and may represent a synapomorphy with †Ellimma. According to Silva Santos and Correa (1985), †Scutatuspinosus itapagipensis has 32 vertebral centra and 12 dorsal fin rays (comparable with paraclupeines other than †Paraclupea itself), 10 dorsal scutes, and 25 ventral scutes (the lowest number of each for any paraclupeine).

As far as †Diplomystus is concerned, although several nominal North American species have been recognized, it has long been suspected that their taxonomic diversity is exaggerated. Woodward (1901) suggested that †D. analis, †D. pectorosus, and †D. theta were synonyms of †D. dentatus, while Eastman (1912) pointed out that species of †Diplomystus from the Green River were almost indistinguishable. In his revision of †Diplomystus, Grande (1982a) also concluded that the only valid species from the Green River shales is †D. dentatus, and that the others are all junior synonyms of that taxon. Two other species were retained by Grande (1982a) in the genus †Diplomystus: †D. birdi and †D. dubertreti from the Upper Cretaceous of Lebanon (Woodward, 1895; Signeux, 1951). Grande (1982a, 1985a) united these three †Diplomystus species by the presence of dorsal scutes with a pectinate posterior border, more numerous dorsal scutes (22–36) than in †Ellimmichthys, and reduction or loss of a median recess in the posterior edge of the scute (present in †Ellimmichthys). Grande (1982a) regarded the two Cretaceous species from Lebanon as closer to each other than to †D. dentatus, because they both have more dorsal fin rays and fewer vertebrae.

Zhang et al. (1985) noted that †D. shengliensis is very similar to †Diplomystus dentatus from the Green River shales. In fact, they are so similar that it is difficult to distinguish them as separate species, despite their widely separated occurrences today on opposite sides of the Pacific. A list of characters comparing †D. shengliensis and †D. dentatus is provided in table 2. Despite the small sample size for †D. shengliensis (with merely three specimens), information is lacking in only 4 out of 32 characters (data are missing for the pelvic fin, the number of abdominal scutes, and scale morphology).

†Diplomystus shengliensis resembles †D. dentatus in 22 of the 28 characters that can be compared. The ratio of depth/standard length is slightly greater in †D. shengliensis (40%) than that in †D. dentatus (35–37%), but this may be a taphonomic rather than morphological distinction related to deformation of the ventral portion of the body in available specimens of †D. shengliensis. There are 41 dorsal scutes in †D. shengligensis, only one more than the maximum number observed in †D. dentatus, although the majority of specimens have fewer (between 33 and 36; Grande, 1982a). There are 13 branched rays in the dorsal fin of †D. shengligensis, again one more than the maximum count in †D. dentatus. An estimated 42 vertebrae are present in †D. shengligensis, only one fewer than the minimum count from †D. dentatus (most individuals have more than 44). Greater apparent differences are found in the number of abdominal vertebrae (17 in †D. shengligensis, 20 in †D. dentatus) and ribs (15 pairs in †D. shengligensis, 17–18 in †D. dentatus. Given the extent to which these features vary in †D. dentatus, and the very small available sample of †D. shengliensis, the meristic distinctions between them are best regarded as provisional. Whether †D. shengliensis is a distinct species, however, the presence of †Diplomystus in China during the Eocene is confirmed by the present observations.

The morphological similarity of †Diplomystus dentatus and †D. shengliensis is further emphasized by comparison with the two Lebanese Cretaceous species †D. birdi and †D. dubertreti, both of which have dorsal scutes like those of †D. dentatus in their adult morphology and presumed ontogenetic development, but which differ from †D. dentatus in having many more dorsal fin rays and far fewer dorsal scutes, abdominal scutes, anal fin rays, vertebrae, and ribs (Grande, 1982a). There is also a gap between the second and third hypurals in the two Cretaceous species, unlike in †D. dentatus and †D. shengliensis. Some other potentially useful systematic features, such as the supraoccipital crest and entopterygoid dentition, are unknown in the Lebanese species. While both species agree in most respects with the generic diagnosis of †Diplomystus presented earlier, their relationships are left unresolved and require further analysis.

According to Greenwood et al. (1966), Patterson (1970), and Nelson (1973), the modern Clupeomorpha are monophyletic and share derived features of the ear-swimbladder connection, the recessus lateralis, the cranial morphology, the caudal skeleton, the abdominal scutes, and the structure of gill arches (see also Nelson, 1967, 1970). With the inclusion of extinct taxa (especially at basal levels), however, monophyly of the Clupeomorpha becomes increasingly tenuous. Forey (1975) conducted a comprehensive analysis of clupeomorph characters while studying †Erichalcis, and paid particular attention to the morphology of the recessus lateralis, the position and size of the dermosphenotic, and the sensory canal arrangement around the recessus. He found that presence of an upper division of the levator arcus palatini muscle in clupeoids can be correlated with presence of a ridge on the frontal and an anterodorsal process on the hyomandibular. Both features are present in †Knightia, leading Forey (1975) to include it in clupeoids, but they are absent in †Denticeps, †Ornategulum, and †Diplomystus brevissimus (Patterson, 1967; now †Armigatus, Grande, 1982a).

The relationships between the two Recent clupeiform suborders (Denticepitoidei, Clupeioidei) and some putative clupeomorph fossils have been investigated by Forey (1975) and Grande (1982a, 1985a), while the problem of clupeomorph relationships generally within teleosts has been discussed by Patterson and Rosen (1977) and Arratia (1996). †Diplomystus, †Armigatus, and †Ellimmichthys all seem to be close extinct relatives of modern clupeomorphs (i.e., Clupeiformes sensu Grande, 1985a; in his phylogeny there is an unresolved trichotomy between †Armigatus, the group †Diplomystus + †Ellimmichthys, and Clupeiformes). There are differences of opinion regarding the inclusion of certain other extinct taxa within clupeomorphs. †Ornategulum was included in clupeomorphs by Forey (1975), Patterson and Rosen (1977), and Grande (1982a), but was excluded by Grande (1985a). †Erichalcis was placed in the Clupeomorpha by Forey (1975) and Grande (1985a), but was regarded as the sister taxon of Esox + salmonids by Arratia (1996). Part of the uncertainty is due to the incomplete preservation of the material, and in the case of †Erichalcis the type series may include more than one taxon (Mark Wilson and Lance Grande, personal commun., 2000). †Santanaclupea shares at least two synapomorphies with clupeiforms, but cannot be placed unambiguously within any of the modern clupeiform groups (Maisey, 1993).

In Grande's (1985a) revision of Recent and fossil clupeomorphs, †Diplomystus and †Ellimmichthys form a monophyletic group (order †Ellimmichthyiformes) supported by a single character (dorsal scutes broader than long or of rectangular shape). Clearly this character also unites †Ellimma branneri with these genera, although it is developed only in the posterior part of the scute series in this form.

Chang and Grande (1997) referred †Paraclupea chetungensis to an equivalent monophyletic group (family Paraclupeidae = Ellimmichthyidae). They concluded that †Paraclupea is more closely related to †Ellimmichthys than to †Diplomystus, and should be included with †Ellimmichthys in the subfamily †Paraclupeinae. Our investigation of †Ellimma branneri from Brazil suggests that it is also a paraclupeine, in which case the ornamentation of dorsal scutes becomes a synapomorphy of the subfamily. †Ellimma, †Ellimmichthys, and †Paraclupea form an unresolved trichotomy in our phylogenetic analysis, but meristic data for †Ellimma and †Ellimmichthys are similar whereas those for †Paraclupea are different. Our conclusions regarding †Diplomystus shengliensis agree with those of Chang and Chow (1978) and Zhang et al. (1985) that it is most closely related to †D. dentatus from North America.

Our analysis (fig. 13) supports the main conclusions reached by previous workers, especially those of Grande (1985a), although some aspects of our results require critical comment. First, our morphological evidence that the †Ellimmichthyiformes form a monophyletic group is not as strong as we would wish. Second, some extinct “double-armored” clupeomorph taxa (e.g., †Scutatospinosus, †Ezkutuberezi carmeni) were excluded from our analysis, because we were unable to examine material or retrieve sufficient data from the published descriptions; their inclusion in future phylogenetic analyses is clearly desirable and may provide a useful test of the conclusions reached here. Third, although the trichotomy formed by †Armigatus, †Ellimmichthyiformes, and Clupeiformes in Grande's (1985a) cladogram seems to be resolved here, our grouping of †Armigatus and †Diplomystus is actually supported by ambiguous characters (3, 9, 10, and 26 in our character list). Even the most convincing synapomorphy (a posteriorly expanded third hypural, leaving no gap between hypurals 2 and 4) also occurs in some clupeiforms not included in our analysis. For all these reasons the phylogenetic hypothesis presented here is considered tentative, as it reflects current uncertainties concerning paraclupeine relationships.

Two interesting paleobiogeographical puzzles involving China have emerged from this work. First, an apparent sister-group relationship has emerged between †Paraclupea and Early Cretaceous paraclupeines of western Gondwanan. Second, the presence of very similar †Diplomystus fossils in China and North America during the Eocene is curious. According to our analysis, paraclupeines and †Diplomystus are closely related (as members of the family †Paraclupeidae), but their respective distribution patterns are fundamentally different and may be the result of quite unrelated biogeographic histories. In both cases the fossil record is undoubtedly incomplete, and the known distributions of these forms may be so strongly biased as to make them biogeographically unintelligible. We restrict our remarks to a few salient points that may be worthy of further investigation.

Biogeographic Observations: All the genera included here within the †Paraclupeinae (†Ellimmichthys, †Ellimma, †Paraclupea, and perhaps †Scutatospinosus) are Early Cretaceous in age. †Ellimmichthys and †Ellimma both occur in Brazil, but †Ellimmichthys is also known from Africa and Mexico. The two better known species of †Ellimmichthys are †E. longicostatus (Hauterivian–Barremian, Marfim Formation, Recôncavo Basin, Brazil) and †E. goodi (late Aptian–Albian, West Africa). The genus †Ellimma contains three nominal species, all from Brazil: †E. branneri and †E. riacensis (considered synonymous; Schaeffer, 1947; Grande, 1985a) from the late Aptian (Muribeca Formation, Sergipe Basin; Jordan, 1910); and †E. cruzi from the Aptian–early Albian (Cabo Formation, Cabo Basin; Silva Santos, 1990; previously identified as †Ellimmichthys by Costa et al., 1979). The undescribed Albian species of †Ellimmichthys from Mexico mentioned by Chang and Grande (1997) represents a western Tethyan (Caribbean) rather than Gondwanan occurrence, and is the only marine paraclupeine record. Another double-armored clupeomorph, †Scutatospinosus itapagipensis, occurs along with †Ellimma branneri in the Marfim Formation of Brazil (Silva Santos and Correa, 1985).

Today there is little doubt that the margins of Brazil and West Africa were contiguous during the Early Cretaceous (Pitman et al., 1993). The earliest record of †Ellimmichthys (and consequently the minimum postulated age of divergence between †Ellimma and †Ellimmichthys) is Hauterivian–Barremian, which predates the formation of an equatorial seaway by a considerable margin, although †Ellimmichthys goodi from Africa and all records of †Ellimma are Aptian or younger (approximately contemporary with permanent emplacement of the seaway in the late Aptian; Maisey, 2000). All Gondwanan paraclupeines are from strata deposited within rift basins that were involved in the final tectonic separation of Africa and South America. At that time these basins were intracontinental, but subsequent rifting, crustal extension, and drifting transformed them into parts of the new continental margins. Their distribution pattern differs from those of some other Early Cretaceous fishes from western Gondwana, such as †Mawsonia and †Calamopleurus, which occur more widely across interior localities of Africa and Brazil (including nonrift settings), although these occurrences probably also represent former regions of endemism which were disrupted by formation of the Equatorial Atlantic seaway (Maisey, 2000).

The habitats and history of Gondwanan paraclupeine fishes were undoubtedly affected by the early (predrift) phases of these momentous tectonic processes. The distribution of Gondwanan †Ellimmichthys (and †Scutatuspinosus?) may reflect vicariant isolation of lacustrine populations along the tectonically active Recôncavo-Tucano-Jatobá and Gabon-Sergipe-Alagoas (GSA) rift trends (Maisey, 2000), and †Ellimma may be an endemic South American genus which perhaps evolved as a result of vicariant isolation within lakes of the GSA trend (Sergipe and Cabo basins). Albian occurrences of †Ellimmichthys are restricted to freshwater deposits within African marginal rift basins, but have not been documented from the marine strata overlying them. The genus has nevertheless been identified in marine sediments from Mexico, suggesting that a salt-tolerant form reached western (Caribbean) Tethys by the Albian–Cenomanian.

The distribution of †Paraclupea is a biogeographic conundrum (Chang and Chow, 1986; Chang and Chen, 2000). †Paraclupea is not known from any Gondwanan locality; †P. chetungensis is found in freshwater deposits near the western Pacific coast of China, but an Early Cretaceous freshwater paraclupeid from Japan (described as †Diplomystus; Uyeno, 1979; Uyeno and Yabumoto, 1980; Yabumoto, 1994) is rather similar to the Chinese form and may extend the range of †Paraclupea. These discoveries considerably expand the distribution of paraclupeines beyond western Gondwana and nearby western Tethys, far into eastern Asia. At this stage we can only speculate about the overall paraclupeine distribution pattern in the Cretaceous.

There is little geologic evidence to support an Early Cretaceous nonmarine paleogeographic connection between the eastern Asiatic margin and western Gondwana that would account for the observed distribution of paraclupeine fishes. We could plead for more widespread distribution and earlier occurrences of freshwater paraclupeine than is presently realized, especially since the †Diplomystus + †Armigatus clade represents the sister group to paraclupeines and is presumably of equal antiquity (in which case there should be relatives of that clade in the Barremian–Hauterivian). Another possibility is that paraclupeines may have marine origins and their distribution may reflect a largely unknown marine history. Unfortunately, both scenarios involve sweeping ad hoc assumptions, for example that the remains of a widespread fauna are still to be found elsewhere, or that their traces in other areas have been obliterated by later geological events.

†Diplomystus shengliensis from Bohai Gulf and †D. dentatus from Green River are morphologically so similar that there is little doubt they represent closely related sister species. Interestingly, besides †Diplomystus, four other genera of fishes (†Eohiodon, †Knightia, †Amyzon, and Esox) supposedly are shared by the Green River and Bohai Gulf faunas (Chang and Chow, 1986; Chang and Zhou, 1993, 2002; Chang et al., 2001), and several other lineages from the Bohai Gulf deposits are also represented in the Green River ichthyofaunas (e.g., “Dasyatidae”, Amiidae). As pointed out by Grande and Bemis (1998: 337), the Chinese amiid is still inadequately known and cannot yet be assigned either to Amia or †Cyclurus, although it is almost certainly an amiine. †Diplomystus, †Eohiodon, †Knightia, †Amyzon and Esox have an Eocene transpacific distribution pattern, but amiines and dasyatids are more broadly distributed.

Interestingly, similar faunal distribution patterns have been noticed among Eocene terrestrial vertebrates of Asia and North America. These have been used to suggest that a land bridge existed between these areas in the vicinity of the Bering Straits, permitting mammals and other tetrapods to pass freely between these regions (Russell and Zhai, 1987). McKenna (1975, 1980, 1983, 1984) has also proposed that arctic connections between the continents have played a significant role in such dispersals.

Grande (1985b, 1989, 1994) noted this repeating pattern of area relationship in the early Tertiary between western North America and east Asia, based on the phylogenetic relationships of teleosts from Green River and the west Pacific region (including China, Indonesia, and Australia). He characterized this distribution as a “transpacific” pattern that could be explained by the Pacifica hypothesis (Grande, 1994). While we do not question these observations, note that much of the similarity in ichthyofaunas is restricted to the Eocene fossil assemblages of Bohai Gulf and Green River, whereas other areas on the western side of the Pacific share far fewer taxa with Green River (known examples include osteoglossids of Australia and Indonesia; osteoglossids and catostomids of South China; amiines and catostomids of inland North China; amiines, hiodontids and catostomids of East Khazakstan).

More recent geophysical and geochemical studies (some still in progress) have raised a startling and radically different alternative to the Pacifica hypothesis. There is mounting evidence that, for a brief period (approximately 2 million years, during the late Paleocene and early Eocene), the Arctic Ocean was almost completely landlocked, with a terrestrial corridor between North America, Iceland, and Europe and another through Beringia, and a single outlet to the sea via the Turgai Strait (McKenna, 1998). With several major river systems draining northward into the Arctic, net runoff/evaporation may have caused dramatic lowering in salinity, sufficient perhaps to allow freshwater fishes to move unimpeded by a saltwater barrier between Asia and North America.

Such a rapid dispersal event could certainly account for the sister-group relationship between †Diplomystus from China and North America. It would also offer a partial explanation for some Eocene fish distribution patterns, and perhaps even the holarctic distribution of some modern fishes (e.g., paddlefishes). †Eohiodon falcatus occurs in the Green River Formation, and hiodontids are represented in eastern China and East Khazakstan (Central Asia). Eocene catostomids also occur in both these areas and are widely spread in the vast plains area of East Asia, from Mongolia in the north to Guangdong Province at the very south of China (Chang et al., 2001). The osteoglossid †Phareodus occurs in the Green River Formation, as well as in Australia (†Phareodus queenslandicus; Hills, 1934; Li, 1994). Eocene osteoglossids are found in Sichuan Province (†Sinoglossus, Su, 1986) and Hubei provinces, South China (Song, in prep.) and Eocene or Oligocene of Indonesia (†Musperia radiata; Sanders, 1934). Eocene hiodontids, catostomids, and osteoglossids are also reported from British Columbia (Wilson, 1977). The wide boreal distribution of amiines during the Eocene (including North America, northern China, East Khazakstan, and Europe) is remarkably similar to the pan-Arctic distribution of many riparian and terrestrial vertebrates during the Late Cretaceous and early Cenozoic (Estes and Hutchison, 1980). Freshwater Arctic connections between the northern continents may therefore have played a significant role in amiine dispersal.

On the other hand, in the vast area between the Bohai Gulf and Green River occurrences there are several other Eocene freshwater deposits with abundant fossil fishes, yet these ichthyofaunas do not yet include dasyatids, †Diplomystus, or clupeids (e.g., middle Eocene of Washington State, U.S., and British Columbia, Canada; inland areas of South China and East Khazakstan; Wilson, 1977, 1978; Tang, 1959; Liu et al., 1962; Wang et al., 1981; Sytchevskaya, 1986). In addition, some taxa occur in one region but not the other (e.g., Eocene lepisosteids in Green River and western Canada but not Asia; Eocene cyprinids in Bohai Gulf, South China, and East Khazakstan but not in North America or Europe). Such absences from the fossil record may reflect a preservational bias against certain taxa in one area or the other, or it could indicate a real absence, suggesting that some taxa did not participate in any holarctic dispersal event during the late Paleocene and early Eocene.

Quite probably, there is no single factor responsible for the Eocene “transpacific” distribution pattern. A broad connection between Asia and North America in the Bering Strait area could have played a role, as could temporary desalination of the Arctic Ocean. Both hypotheses provide equally plausible biogeographic alternatives to the Pacifica model and are certainly more in accordance with geological data than the latter. At present the only common link between the Early Cretaceous distribution of paraclupeines and the Eocene distribution of †Diplomystus is China, for only there have both taxa been discovered.