Novitates^yAMERICAN MUSEUM PUBLISHED BY THE AMERICAN MUSEUM OF NATURAL HISTORY CENTRAL PARK WEST AT 79TH STREET NEW YORK, N.Y. 10024 U.S.A. NUMBER 2719 NOVEMBER 27. 1981

DONN E. ROSEN AND LYNNE R. PARENTI Relationships of , and the Groups of Atherinomorph Fishes

Novitates^ T AMERICAN MUSEUM PUBLISHED BY THE AMERICAN MUSEUM OF NATURAL HISTORY CENTRAL PARK WEST AT 79TH STREET, NEW YORK, NY. 10024 Number 2719, pp. 1-25, figs. 1-20 November 27, 1981

Relationships of Oryzias, and the Groups of Atherinomorph Fishes

DONN E. ROSEN1 AND LYNNE R. PARENT!2

ABSTRACT Newly discovered evidence, particularly that by that name, preferring instead to include them pertaining to the gill arch skeleton and hyoid ap- in a general classification of the paratus, indicates that adrianichthyoids (- by a listing convention; and (2) used the ordinal es and their allies) are related more closely to half- term for killifishes, in con- beaks, flyingfishes, , and sauries than formity with a recent monographic revision by to the killifishes with which they have been as- Parent! (1981), and the term (in- sociated for over a century. This discovery was cluding the Adrianichthyoidei and Exocoetoidei) used as an occasion to reevaluate atherinomorph for its coordinate sister group. We find the Ath- interrelationships and the monophyly of the in- erinomorpha to be supported by 10 characters cluded groups. We conclude that atherinoids are uniquely derived among ctenosquamate neote- not presently a definable group, but that killifishes leostean fishes and a subdivision including cy- and the plus and allies are. prinodontiforms and beloniforms to be supported We also support the monophyly of the Atherino- by four characters uniquely derived within the morpha. In our proposed theory of relationships Atherinomorpha. Some or all "atherinoid" fishes we have (1) abandoned use of the term Ather- are thought to be plesiomorphous to that subdi- inoidei to represent the fishes formerly grouped

INTRODUCTION Atherinomorph fishes may be defined as a sen, 1964; Nelson, 1969), rostral cartilage monophyletic group by derived characters of (Alexander, 1967; Parent!, 1981), upper-jaw the egg, embryo (Rosen, 1964; Breder and protrusile mechanism (Alexander, 1967), Rosen, 1966; Foster, 1967), ethmoid ossifi- spermatogonium formation (Grier, Linton, cation (Rosen, 1964), infraorbital bones (Ro- Leatherland, and DeVlaming, 1980; Grier, in

1 Curator, Department of Ichthyology, American Museum of Natural History. 2 Smithsonian Postdoctoral Fellow, Division of Fishes, National Museum of Natural History.

Copyright © American Museum of Natural History 1981 ISSN 0003-0082 / Price $2.50 AMERICAN MUSEUM NOVITATES NO. 2719 press; Grier, Burns, and Flores, in press), MCZ, Museum of Comparative , Cam- nasal capsule (Melinkat and Zeiske, 1979), bridge and at least two features of the dorsal gill SU, Stanford University Collections in the Cali- arch skeleton to be described below. fornia Academy of Sciences, San Francisco The discovery of the gill arch synapomor- UBC, University of British Columbia, Vancouver UMMZ, University of Michigan Museum of Zo- phies of atherinomorphs was an outgrowth ology, Ann Arbor of an earlier discovery of gill arch evidence that the ricefishes, Oryzias, and their close relatives in , , GILL ARCH ANATOMY and Horaichthys, are allied with the flying- The crucial evidence that, for us, prompt- fishes, halfbeaks, needlefishes, and sauries ed the reinvestigation of this group of fishes rather than with the killifishes—a group with concerns the anatomy of the gill arch skele- which they have been continuously associ- ton. Rosen and Greenwood (1976) had noted ated for over a century. We have, therefore, previously that many groups of acanthopter- taken this occasion to review some new and ygians are characterized by the presence of old evidence for interrelationships among the an accessory cartilage in the dorsal gill arch- groups of , an order formally es that connects the epibranchial bone of the established by Rosen (1964) to include ath- first arch with the pharyngobranchial bone of erinoids (silversides and phallostethids), cy- the second arch. They pointed out that this prinodontoids (killifishes and ricefishes), and interarcual cartilage and its connections are exocoetoids (halfbeaks and their relatives) distinctively modified in various groups of and coextensive with the Atherinomorpha of fishes and that synbranchiform fishes, for ex- Rosen (1973). Rosen (1964) had left the in- ample, are uniquely characterized in part by terrelationships of the three suborders un- having the interarcual cartilage ossified. specified and had defined them and their Among atherinomorph fishes there are also subgroups using characters in a manner that a number of unusual features of the dorsal we find, in part, to be unworkable. Accepting and ventral gill arch anatomy, including in- as an initial premise the monophyly of the terarcual cartilages, that specify a set of hi- atherinomorph fishes, as defined above, we erarchical relationships among the various present our analyses of the derived charac- taxa. ters that define component groups and Allis (1903) apparently was the first to re- subgroups. port on the existence in acanthopterygians of a separate cartilage between the first epi- and ABBREVIATIONS AND SYMBOLS second infrapharyngobranchial, but he mis- ANATOMICAL: takenly identified this cartilage as a supra- AC: accessory cartilage pharyngobranchial (see Nelson, 1968, p. C-1,2,3,4: ceratobranchials 1 to 4 137). Later, Allis (1915) recognized the sec- E-1,2,3,4: epibranchials 1 to 4 ondary nature of this element and introduced IAC: interarcual cartilage the term interarcual cartilage for it. Nelson PB-1,2,3,4: pharyngobranchials 1 to 4 (loc. cit.) remarked that it is "common UNC-1: uncinate process of first epibranchial among perciform fishes, e.g., Epinephelus,'" UNC-PB-2: uncinate process of second pharyn- and, indeed, we have found the cartilage to gobranchial be primitively present in the dorsal gill arch UP-4: fourth upper pharyngeal toothplate skeleton of every major group of the Acan- ANATOMICAL SYMBOLS IN FIGURES: thopterygii {sensu Rosen, 1973) in which an open circles: cartilage uncinate process, or its equivalent, is present stippling: bone on the first epibranchial (fig. 1). In published INSTITUTIONAL: illustrations of acanthopterygian dorsal gill AMNH, American Museum of Natural History, arches, however, the cartilage has not al- New York ways been distinguished as a separate ele- 1981 ROSEN AND PARENTI: ORYZIAS

PB-I IAC

PB-3

FIG. 1. Percoid dorsal gill arches. Morone americana (Gmelin) AMNH 26515 same as Centropristis striata (Linnaeus) AMNH 22052. merit (e.g., in Rosen, 1973) and in many in- acanthopterygians such as some "beryci- stances no cartilages of any sort are shown. forms," as in Holocentrus (fig. 2). The con- Primitively among euteleosts the cartilag- dition derived relative to this is to have the inous tip of a short uncinate process near the uncinate process of the first epibranchial and distal end of the first epibranchial directly second pharyngobranchial separated by an contacts the cartilage of an uncinate process intervening interarcual cartilage as in Mo- on the dorsolateral side of the second pha- rone, Centropristis (fig. 1), Caranx (fig. 3A), ryngobranchial [Rosen, 1973, figure 3 (a Monodactylus (fig. 3B), Drepane (fig. 4A), characoid), figure 5 (a salmonid), figure 58 Sphyraena (fig. 4B) and Agonostomus (fig. (an esocoid)]. This of contact between 4C). Among acanthopterygians presently the first two arches persists in primitive neo- classified as "perciforms" or as closely al- teleosts and in plesiomorphous groups of lied with "perciforms" the interarcual car- AMERICAN MUSEUM NOVITATES NO. 2719

tilage is absent only in those groups with some specialized condition of the epi- or pharyngobranchials, as, for example, when the first epibranchial has a very small (No- tothenia, fig. 5A) or no (Xiphister, fig. 5B) uncinate process. It is also primitive for euteleosts (and other main groups of teleosts as well) to have a fourth pharyngobranchial cartilage (Nelson, 1969; Rosen, 1973) and to have each of the four epibranchials approximately equal in size. The derived conditions among acan- thopterygians are to have the fourth pha- ryngobranchial reduced or absent and to have one or more epibranchials specialized in size or shape. In Caranx (fig. 3A), for ex- FIG. 2. Berycoid dorsal gill arches. Holocen- ample, a slender fourth epibranchial articu- trus vexillarius Poey, AMNH 23374. Note that the lates with a relic fourth pharyngobranchial, cartilaginous tip of the uncinate processes on the first epibranchial and second pharyngobranchial whereas the other three epibranchials are ro- come directly together without an intervening car- bust. In Monodactylus (fig. 3B) the articular tilage. Contrast with figures 1, 3, 4. head of the second epibranchial is, by far, the largest epibranchial element. In Drepane

FIG. 3. Percoid dorsal gill arches. A, Caranx mate Cuvier and Valenciennes, AMNH 15206. B, Monodactylus argenteus (Linnaeus), AMNH 30803. 1981 ROSEN AND PARENTI: ORYZIAS

PB-3

UP-4

FIG. 4. Percoid, sphyraenoid, and mugiloid dorsal gill arches. A, Drepane punctata (Linnaeus), AMNH 13922. B, Sphyraena borealis De Kay, AMNH 4339. C, Agonostomus monticola (Bancroft), AMNH 11613. AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 5. Notothenioid and blennioid dorsal gill arches. A, Notothenia cornucola Richardson, AMNH 3606. B, Xiphister atropurpurens (Kittlitz), AMNH 2709.

(fig. 4A) and Xiphister (fig. 5B) the third 6 to 9). (2) In killifishes the first epibranchial epibranchial is the largest element. has no uncinate process on its shaft, the The dorsal gill arch skeleton of atherino- usually rather long interarcual cartilage is morphs is of a derived acanthopterygian borne instead on an expanded basal epibran- type. The articular head of the fourth epi- chial cartilage and inserts, as in silversides, branchial is very large and forms the main on the bony shaft of the uncinate process of supporting element for the pharyngobranchi- the second pharyngobranchial (fig. 10). (3) In al dentition, a condition that is unusual in some ricefishes (the species of Xenopoeci- forms lacking a fourth pharyngobranchial. lus, fig. 11) and some halfbeaks and flying- The more usual condition is for support of fishes [species of Chriodorus, Arrhamphus, the posterior toothplates to shift from the Parexocoetus, Hirundichthys, and Cypselu- fourth to the third and fourth or mainly the rus (figs. 12 and 13)] the first epibranchial has third epibranchial (Rosen, 1973). The first no uncinate process, has an expanded basal epibranchial and interarcual cartilage also cartilage to which is attached a small inter- are specialized, but in atherinomorphs the arcual cartilage confined to the region be- situation is rather complex. We recognize tween the bases of epibranchials 1 and 2; the four basic conditions: (1) In silversides the first arch has thus entirely lost contact with uncinate process arises at the midpoint or the pharyngobranchial of the second arch. nearer the proximal rather than distal end of (4) In other ricefishes and halfbeaks and in the epibranchial at a sharp angle to the main the sauries and needlefishes examined, the shaft of the bone and the interarcual cartilage anatomical arrangements are just as in con- articulates with the base or shaft rather than dition (3), but there is no interarcual cartilage with the cartilaginous tip of the uncinate pro- present (figs. 14 to 17). We can assume that cess on the second pharyngobranchial (figs. these conditions are transformations of the 1981 ROSEN AND PARENTI: ORYZIAS same character or characters because all of these fishes are united by the two synapo- morphies of the posterior dorsal gill arches (loss of the fourth pharyngobranchial and en- largement of the distal end of the fourth epi- branchial as the main supporting element of the pharyngobranchial dentition) and the eight synapomorphies enumerated at the out- set. Given that assumption of monophyly, we infer the transformation of this character to be from condition (1) to condition (4), rather than the reverse, since the presence of an uncinate process and a narrow proxi- mal end on the first epibranchial are primi- tive for euteleosteans. A consequence of that inference is that in atherinomorphs the un- cinate process has shifted proximally on the first epibranchial, carrying the interarcual cartilage with it, that the uncinate process is represented in killifishes, ricefishes, half- beaks, etc. as part of the enlarged basal car- tilage, that the interarcual cartilage is repre- sented in some ricefishes, halfbeaks, and FIG. 6. "Atherinoid" dorsal gill arches. Tel- flyingfishes by a vestige at the base of the matherina ladigesi Abe, AMNH 35378. Note an- epibranchial, and that its absence in other gle of UNC-1 and point of contact of I AC with PB-2, features found in all unreduced "ather- ricefishes, halfbeaks, needlefishes, and sau- inoid" gill arch skeletons. Contrast with figures ries is not primitive but due to secondary 1, 3, and 4, and compare with figures 7, 8, and 9. loss. Evidence that the uncinate process can occupy a basal position is illustrated by the silverside Melanorhinus microps (fig. 18) and the species of Pseudomugil (fig. 9C). branchials. This means that in these fishes The proximal position of the uncinate pro- the pharyngobranchials are supported main- cess in Melanorhinus is correlated with the ly by the enlarged fourth epibranchial and by absence of an interarcual cartilage; the car- the connective tissue and muscles from the tilage is also absent in some killifishes (Par- basicranium and that the second pharyngo- ent!, 1981, figs. 45, 48a, and 48b). Perhaps branchial is supported by the connective tis- the most interesting independent evidence sue it shares with the large third pharyngo- that the absence of an uncinate process can branchial. Being supported in this way, the be associated with a reduced interarcual car- second pharyngobranchial has a character- tilage with proximal (basal) articulation to istic orientation in which the anterior part of the first epibranchial is the condition found the bone is angled sharply upward toward in Ceratostethus (fig. 19A). In this phallo- the anterior end of the third pharyngobran- stethid, the form of the cartilage closely re- chial. A third derived feature shared by these sembles those found in some adrianichthy- fishes is the very large ventral flange on each oids, hemiramphids, and exocoetoids. of the fifth ceratobranchials (toothed lower Ricefishes and exocoetoids share three pharyngeals) and the close apposition (in other derived features of the gill arch skele- ricefishes) or fusion (in exocoetoids) of the ton. All show a reduction in size of the sec- right and left elements. ond and third epibranchials which no longer Some parts of the gill arch anatomy, there- have any direct contact with the pharyngo- fore, support the monophyly of the Ather- AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 7. "Atherinoid" dorsal gill arches. Rheocles alaotrensis (Pellegrin), AMNH 28127 (position of IAC estimated). B, Melanotaenia maccullochi Ogilby, AMNH 44401. inomorpha (loss of fourth pharyngobranchial 19) (and we have no way of distinguishing and enlargement of fourth epibranchial); oth- the two cases of bone loss as different) and er parts define a group including cyprinodon- the second is somewhat ambiguous because toids, ricefishes, and exocoetoids (absence the difference in size between the first and of uncinate process and expansion of base of the second and third epibranchials is only first epibranchial; two other possible features slight in aplocheiloids (although it is quite are discussed below), and define a subgroup pronounced among apomorph groups of kil- consisting of ricefishes and exocoetoids. The lifishes referred by Parent! to the cyprino- two additional features of the dorsal gill arch- dontoids). es that are consistent with an alignment of killifishes with ricefishes and exocoetoids are the absence in all of a first pharyngobranchial REPRODUCTION AND DEVELOPMENT (as contrasted with its presence in ather- Since the sexual products have previously inoids) and the trend toward size reduction been considered evidence for monophyly of of the second and third epibranchials in the the Atherinomorpha and since no recent plesiomorphous groups referred to by Par- summary of this evidence exists, some com- ent! (1981) as aplocheiloids (fig. 10). The ments are in order. In 1964 Rosen noted that problems with interpretation of these two a "large, spherical, demersal, chorionated features are that the first is a loss-character egg with adhesive filaments occurs in all which is also true of phallostethid fishes (fig. . . . groups." He also recorded that (1) in the 1981 ROSEN AND PARENTI: ORYZIAS

FIG. 8. "Atherinoid" dorsal gill arches. A, Melaniris chagresi (Meek and Hildebrand), UMMZ 179955. B, Bedotia geayi Pellegrin, AMNH 28132. developing embryo of Exocoetus, Oryzias, sistently observed in the eggs of [cyp- the cyprinodontoid Xiphophorus and Meni- rinodontoids and atherinoids] . . . these dia, but not in Sphyraena, and probably not globules are never observed in the eggs of in Mugil and other fishes, the heart is dis- any [exocoetoid] . . . except those of certain placed forward in front of the head on the hemiramphids." Foster did not mention yolk sac instead of developing in the throat which hemiramphids, but his general obser- region and (2) that the consequence of this vations might suggest that exocoetoids have exceptional embryonic cardiac inversion is eggs in some sense different from those of "the complete separation of the afferent and other atherinomorphs. All of these observa- efferent circulation in the pericardial se- tions and opinions were made within a con- rosa, whereas the embryos of fishes with a text of a classification in which ricefishes more usual position of the heart have the af- were considered cyprinodontoids rather than ferent and efferent circulations superim- exocoetoids. But with ricefishes as the ple- posed." Rosen also noted, as did Green- siomorph group of exocoetoids Rugh's wood et al. (1966), that the atherinomorph (1952) description of the Oryzias egg has spe- egg lacks an oil globule, and this was based cial interest: "At oviposition many oil glob- on an earlier statement by Orton (1955). ules may be seen between the yolk and the Foster (1967), however, regarded exocoe- periblast. During early development these toids as reproductively more specialized decrease in number by confluence and merge than atherinoids or cyprinodontoids because into a single large globule at the vegetal "exocoetoids have secondarily lost the con- pole." About the related Horaichthys Kul- spicuous oil globules which are present in the karni (1940) wrote: "The ovum contains a eggs of members of the other two subor- large amount of yolk with a number of glob- ders." Later Foster (1968) wrote that "Al- ules in it. The globules are numerous, small though conspicuous lipid globules are con- and scattered in eggs just removed from the 10 AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 9. "Atherinoid" dorsal gill arches. A, Menidia menidia (Linnaeus), AMNH 35924. B, Qui- richthys stramineus (Whitley), AMNH 20571. C, Pseudomugil novaeguineae Weber, AMNH 20345. 1981 ROSEN AND PARENTI: ORYZIAS 11

FIG. 10. Aplocheiloid cyprinodontiform dorsal gill arches. A, Aplocheilus panchax (Hamilton-Bu- chanan), AMNH 44403. B, Rivulus hard (Boulenger), AMNH 15189. The base of E-l in A is the usual condition of cyprinodontiforms.

ovary . . . but in those which . . . develop- bryogenesis among the main groups of ath- ment has . . . proceeded the oil globules are erinomorphs are paralleled by those in tes- large, fewer in number and concentrated at ticular structure. According to Grier, Linton, one pole." Leatherland, and DeVlaming (1980) and Kulkarni {op. cit.) was also the first author Grier, Burns, and Flores (in press) sperma- to notice the similarity of the chorionic fila- togonia are entirely restricted to the distal ments in various kinds of atherinomorph end of the tubule immediately beneath the eggs. He recorded two kinds of filaments, tunica albuginea whereas other groups of te- short ones of mostly uniform length distrib- leosts have the spermatogonia distributed uted over the egg and a tuft of longer ones along the length of the tubule. There is also that entangled the egg on plant material. He a possibility that atherinomorph sperm are compared the egg of Horaichthys with those distinctive (Grier, 1976). of killifishes and needlefishes and then wrote that "with its two types of filaments, appears MONOPHYLY in many respects to be very similar to that of the Philippine Gulaphallus [a phallostethid] "ATHERINOIDS": The modern taxonomic . . . [and that the] egg of Oryzias ... is also concept of "atherinoid" fishes is derived similar, though the shorter filaments . . . ap- from their former inclusion in a larger group, pear to be much smaller and the longer ones Percesoces or Mugiliformes, which con- rolled into a thicker cluster than in Horaich- tained also mullets, barracudas, and some- thys." Earlier, Breder (1932) had illustrated times, threadfins. Their taxonomic definition the structure and relative size of the long and usually amounted to a statement that they short filaments on the egg chorion of Par- are different from barracudas and similar to exocoetus, a flyingfish. mullets or that they lacked the defining char- Derived similarities in the ovum and em- acters of both barracudas and mullets. Jor- 12 AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 11. Adrianichthyoid dorsal gill arches. A, Xenopoecilus sarasinorum (Popta), AMNH 20481. B, Xenopoecilus poptae Weber and de Beaufort, AMNH 20480. dan (1905), for example, identified them as merous, from 45 to 52; first dorsal with from the "most primitive of living Percesoces," 3 to 8 spines; anal with 1 spine." Parts of stating that they are small, slender fishes this definition, consisting entirely of ambig- with a small mouth and feeble teeth, no lat- uous or primitive characters, or statements eral line, and in color translucent green now known to be inaccurate, have been in- sometimes with a broad band of silver or bur- corporated into some subsequent definitions nished black. Jordan and Hubbs (1919), in of the group or have been replaced entirely the first major review of the family, were of by comparable lists. Berg (1940) gave only: the opinion that "the numerous genera of the pelvic bones connected with cleithra by a lig- Atherinidae . . . form a compact and ob- ament; vertebrae 31 to 60; lower and upper viously natural group," but gave no diagnos- ribs present; no intermuscular bones. Gos- tic characters for it. line (1962) wrote: pelvic girdle not supported The earliest attempt at detailed anatomical by postcleithral strut; vertebrae more than definition of "atherinoids" was by Starks 26; eggs usually adhesive; spinous dorsal (1899) who wrote: "Lower limb of post-tem- placed well back on body; pectoral fins high poral attached to opisthotic by ligament; ba- on sides; pelvic fins with a spine and five soft sisphenoid developed; myodome opening to rays; third and fourth upper pharyngeals exterior posteriorly; region about foramen fused; infraorbital canal interrupted. Rosen magnum not produced; superior pharyngeals (1964), in a key, provided this list: lateral line typical in shape, bearing teeth; vertebrae nu- wanting or represented by a series of pits or 1981 ROSEN AND PARENTI: ORYZIAS 13

PB-3

FIG. 12. Exocoetoid dorsal gill arches. A, Parexocoetus brachypterus (Richardson), AMNH 44402; same as Hirundichthys affinis (Giinther), AMNH 22001 and Cypselurus furcatus (Mitchill), AMNH 21810. B, Anhamphus sclerolepis Giinther, AMNH 40002.

scale canals at midside; lower pharyngeal (number of vertebrae, position of spinous bones separate; parietals present; branchi- dorsal fin, development of lateral line) and ostegal rays five to seven; usually with a first the kinds of characters included. It seems dorsal fin of flexible spines above or in ad- fair to conclude that there has been a great vance of anal origin; anal fin usually preced- deal of uncertainty about exactly what it ed by a spine; narial opening paired; pelvic takes to be an "atherinoid" fish. fins abdominal, subabdominal or thoracic in We are unable to diagnose the "ather- position, not modified into a clasping organ; inoids" cladistically. For example, if we first pleural rib on third vertebra; and adduc- judge that atherinomorphs are acanthopter- tor arcus palatini muscle restricted to pos- ygian fishes, then it is primitive for atherino- terior part of orbit. More recently Nelson morphs to have dorsal, anal and pelvic fin (1976) listed a selection of some of the am- spines, about 26 vertebrae, 15 branched cau- biguous and primitive features given above. dal rays, a thoracic or subthoracic pelvic gir- One of the most striking features of this col- dle, and the spinous and soft dorsal rays lection of definitions is the extent to which joined or only narrowly separated. The ath- they differ on the nature of certain characters erinomorphs that most closely approximate 14 AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 13. Exocoetoid dorsal gill arches. Chriodorus atherinoides Goode and Bean, AMNH 20599. A, detail of bases of first two epibranchials. B, dorsal gill arch skeleton. these primitive conditions are the freshwater cyprinodontoids and exocoetoids than with Malagaysian species of the genera Rheocles, the or Malagaysian silversides Rheocloides, and Bedotia. The freshwater (see Myers, 1928). Therefore, an Atherinidae rainbowfishes, or melanotaeniids, of Austra- might only be definable by exclusion of both lia and New Guinea resemble the Malagay- bedotiids and melanotaeniids as in the sug- sian forms but have a more derived condition gested alignments above. of the two dorsals, thereby aligning them Although, as used by previous authors, with non-Malagaysian atherinomorphs. neither the Atherinidae nor the Atherinoidei Likewise other "atherinoids" (the freshwa- can be presently regarded as monophyletic ter pseudomugilids of New Guinea and Aus- groups, the groups referred to as the bedo- tralia, the freshwater phallostethids from the tiids, melanotaeniids, atherinids, pseudo- Philippines, and the worldwide freshwater mugilids, telmatherinids, isonids, and phal- and marine atherinids) have still further de- lostethids can collectively be regarded as rived conditions of the dorsal fins, and of the outgroups for specifying the defining char- pelvics, the extent of spine development, acters and relationships of cyprinodontoids, and number of vertebrae, as well, which adrianichthyoids, and exocoetoids. Allen align at least some of them more closely with (1980, pp. 451, 452) combined the pseudo- 1981 ROSEN AND PARENTI: ORYZIAS 15

FIG. 14. Adrianichthyoid dorsal gill arches. (Herre and Ablan), SU 29564. A, detail of ceratobranchials and epibranchials. The triangular cartilaginous processes on C-2 and C-3 are diagnostic of unreduced adrianichthyoid ceratobranchials. B, dorsal gill arch skeleton. mugilids with the melanotaeniids based on fourth), the lowset pectoral fins with a large, shared specializations of the pelvic region scalelike postcleithrum, a pattern of early and, although his studies were confined to sexual maturation and prolonged embryonic only certain "atherinoid" groups of the Aus- development, and the structure of the dorsal tralia-New Guinea region, we provisionally gill arch skeleton. As described above, the accept that alignment. last character is the elongate interarcual car- CYPRINODONTIFORMS (=CYPRINODON- tilage that joins the expanded base of the first TOIDS OF ROSEN, 1964; GREENWOOD ET AL., epibranchial with the shaft of the second 1966): Cyprinodontiform synapomorphies pharyngobranchial. Only one other group of are discussed at length by Parent! (1981). As teleosts with a similar condition is known to atherinomorphs, their unique defining fea- us, viz., the gobioid fishes in which a long tures are the symmetrical caudal fin endo- cartilage joins the base of the first epibran- skeleton in which the epural symmetrically chial with the tip of the second pharyngo- opposes the parhypural, the unlobed caudal branchial uncinate process. fin, the position of the first pleural rib on the CYPRINODONTIFORMS PLUS BELONI- second vertebra (rather than the third or FORMS: Characters that unite cyprinodonti- 16 AMERICAN MUSEUM NOVITATES NO. 2719

PB-3

FIG. 15. Adrianichthyoid dorsal gill arches. A, Horaichthys setnai Kulkarni, AMNH 36576. B, Weber, UBC. forms with adrianichthyoid and exocoetoid duced autopalatine with posterior articular fishes (=Beloniformes in the present usage) cartilage (Rosen, 1964), and no metaptery- are the expanded base of the first epibran- goid or ectopterygoid (the last two are hom- chial, the size reduction of the second and oplasious with the condition in some or all third epibranchials, the loss of the first pha- cyprinodontoids). ryngobranchial (also true of phallostethids) Exocoetoids are defined by the presence and the absence of a second infraorbital bone of a median lower pharyngeal tooth plate (that is, the infraorbital series is represented (see Rosen, 1964), a ventral platelike process only by the preorbital, or lacrimal and posteriorly on the basioccipital, an elongate dermosphenotic, whereas in atherinoids lower jaw in at least some stage of the life there are three bones present in the infraor- history (Nichols and Breder, 1928), more bital series). than three anterior branchiostegal rays (see BELONIFORMS (=ADRIANICHTHYOIDS AND Rosen, 1964, pp. 239-240), and a single na- EXOCOETOIDS): Adrianichthyoid fishes are rial opening on each side (Burne, 1909). One defined easily by the great expansion of the other feature mentioned frequently (e.g., articular surface of the fourth epibranchial, Hubbs and Wisner, 1979) as an exocoetoid the presence of a complex, branched, car- trait is the low trunk lateral line; however, tilaginous ceratobranchial epiphysis, a re- some freshwater hemiramphids lack a later- 1981 ROSEN AND PARENTI: ORYZIAS 17

FIG. 16. Scomberesocoid and exocoetoid dorsal gill arches. A, Belonion apodion Collette, AMNH 36579. B, Nomorhamphus celebensis Weber and de Beaufort, AMNH 35379, showing fusion of right and left PB-3. alis canal system as do the adrianichthyoids, CLADISTIC SUMMARY OF MAIN nearest allies to the exocoetoids. A study of GROUPS the early development of the lateralis system might resolve the uncertainty about the sig- CHARACTERS: The characters rated here as nificance of this feature. synapomorphies of major atherinomorph A group comprising adrianichthyoids and groups are those of the cladistically plesio- exocoetoids, the beloniforms, is defined by morphous members of each group. The rea- the small or absent (inferred reduction or soning is that derived characters shared only loss, see above) interarcual cartilage, rela- by apomorph groups belonging to different tively very small second and third epibran- lineages or by only one or a few apomorph chials, vertical reorientation of the second species of different groups require numerous pharyngobranchial, presence of large ventral assumptions of character convergence or flanges on the fifth ceratobranchials, only a reversal (homoplasy) to account for the ab- single, ventral hypohyal bone, no interhyal sence of these characters elsewhere. Theo- bone, and the lower caudal fin lobe with ries of relationship such as those incor- more principal rays than the upper lobe (e.g., porating numerous assumptions of homoplasy with formulas of 1,6-7,1; 1,6-6,11; 1,5-6,1; are by definition less parsimonious than 1,5-5,11, etc., but never with more principal those in which the derived characters, pres- rays in the upper lobe as is primitive for all ent in cladistically plesiomorphous species other euteleosts except some catfishes). One or groups, are inferred to have been lost or species of Pseudomugil (P. tenellus) that we gained but once in the ancestor of all apo- have examined also has a 1,6-7,1 caudal ray morph (descendant) members. A simple il- count. lustration of this problem is the occurrence 18 AMERICAN MUSEUM NOVITATES NO. 2719

PB-3

E-3

FIG. 17. Scomberesocoid dorsal gill arches. A, Xenentodon cancila (Hamilton-Buchanan), AMNH 38433. B, Cololabis brevirostris (Peters), AMNH 14133. of fin spines in the Atherinomorpha. If ath- "atherinoids" are their plesiomorph sister erinomorphs are members of the Ctenosqua- group. If, on the other hand, we use the ar- mata (myctophiforms, paracanthopterygi- gument of Rosen (1964) that cyprinodonti- ans, atherinomorphs, and percomorphs, as forms and beloniforms are plesiomorphous treated by Rosen, 1973), then fin spines may to "atherinoids," then spines would have to be regarded as synapomorphous for cteno- have been lost once and then regained by squamates and plesiomorphous for atherino- "atherinoids." In addition to the added as- morphs. Because cyprinodontiforms and be- sumption of regaining fin spines, Rosen's loniforms lack fin spines it is necessary to scheme requires five other homoplasies (in postulate a single loss in their common characters 3 and 14 to 17 as enumerated be- ancestor if one or more of the spine-bearing low). The present scheme is therefore pre- 1981 ROSEN AND PARENTI: ORYZIAS 19 ferred because it generates the fewest ad hoc assumptions about character convergence or reversal (i.e., only in characters 2 and 3). Another example of this problem that is useful to mention here concerns the hyoid bar and branchiostegal rays because a cer- tain pattern of these elements has been said to characterize apomorph groups of eute- leosts (Hubbs, 1919; McAllister, 1968). The primitive condition of these elements is for the ventral margin of the anterior ceratohyal to be entire rather than notched and for the numerous (10 or more) branchiostegal rays to decrease gradually in size anteriorly and to be attached to the lateral face of the hyoid bar. The derived condition, characteristic of most neoteleosts, is for the ventral margin of the anterior ceratohyal to be notched and for the 10 or fewer branchiostegals to be divided into two series: a posterior series of bladelike rays on the lateral face of the hyoid bar pos- terior to the notch in the anterior ceratohyal and an anterior series of hairlike rays at- tached to the ventral edge of the anterior cer- atohyal anterior to the notch. In the most derived condition there are generally no more than four bladelike rays on the lateral face of the bar. Based on this assessment, FIG. 18. "Atherinoid" dorsal gill arches. Me- the hyoid apparatus of some exocoetoids lanorhinus microps (Poey), AMNH 25878. Note (large number of branchiostegals, and ab- basal position of UNC-1 and absence of interar- sence of an anterior ceratohyal notch in be- cual cartilage. lonids and scomberesocids) was assessed as primitive and constituted one of the early rived hyoid bar apparatus of percomorphs, reasons for excluding these fishes and the prescribe two inferences: (1) that the struc- related hemiramphids and exocoetoids from ture and position of the anterior hairlike rays the acanthopterygian assemblage. Once the of adrianichthyoids is a transformed state of adrianichthyoids are included as the sister the condition in cyprinodontiform, "ather- group of the exocoetoids, however, that ear- inoid" and percomorph fishes, and (2) that ly interpretation becomes problematical be- the apparently primitive condition of the cause the species of Oryzias and Xenopoe- hyoid apparatus in some exocoetoids is sec- cilus have only four posterior bladelike rays ondary (i.e., homoplasious). and one or two anterior hairlike rays in se- A summary of the 17 characters we have quence with the bladelike ones and the an- used to establish a hypothesis of atherino- terior ceratohyal notched (Adrianichthys morph relationships follows: and Horaichthys have fewer rays). Accept- A relationship between atherinomorphs ing the synapomorphies that unite cyprino- and the neoteleosts is specified by, among dontiforms and beloniforms (adrianich- other characters: thyoids and exocoetoids), and these with "atherinoids," and observing that "atherin- (1) The four posterior bladelike branchi- oids" and cyprinodontiforms have the de- ostegals inserting laterally on the 20 AMERICAN MUSEUM NOVITATES NO. 2719

FIG. 19. Phallostethid dorsal gill arches. A, Ceratostethus bicornis (Regan), MCZ 47304-5. B, Gu- laphallus mirabilis Herre, SU 38903.

hyoid bar, the anteriormost inserting (4) A large demersal egg with long ad- just posterior to a notch on the ante- hesive and short filaments and many rior ceratohyal. lipid globules that coalesce at the ve- That atherinomorphs are also members of getal pole. a more restrictive group, the Ctenosquama- (5) The complete separation of the em- ta, is specified by, among other characters: bryonic afferent and efferent circula- tions by development of the heart in (2) The presence of dorsal, anal and pel- front of, rather than under, the head. vic fin spines. (6) The spermatogonia forming only at The relationship of the atherinomorphs to the blind end of the tubule near the a still more restrictive grouping, the Perco- tunica albuginea. morpha, is indicated by: (7) The rostral cartilage being decoupled from the premaxilla. (3) The presence in the dorsal gill arch (8) Protrusile upper jaw mechanism with skeleton of an interarcual cartilage crossed palatomaxillary ligaments between the first epibranchial and and with a maxillary ligament to the second pharyngobranchial. cranium. Atherinomorphs are themselves defined as a (9) Dermal and endochondral disclike monophyletic group by: ethmoid ossifications. 1981 ROSEN AND PARENTI: ORYZIAS 21

(10) A hydraulic pump mechanism in the some specimens of Oryzias and in Xenopoe- nasal organ. cilus; (2) absence of most fin spines in some (11) The absence of third, fourth, and fifth "atherinoids" and in most cyprinodonti- infraorbital bones. forms and beloniforms; (3) absence of an an- (12) In the dorsal gill arch skeleton, the terior ceratohyal notch in belonids and scom- uncinate process arising on the prox- beresocids and a large number of size-graded imal half of the first epibranchial, co- branchiostegals in all exocoetoids; (4) the alesced with the base of this epibran- posterior location of the pelvics in some ath- chial, or absent. erinids and cyprinodontiforms, and in belon- (13) The absence of a fourth pharyngo- iforms (also rated, above, as convergent branchial. among these forms). CLASSIFICATIONS: Branching diagrams A subgroup of the atherinomorphs, con- from five sources are compared (Boulenger, sisting of cyprinodontiforms + beloniforms, 1904; Regan, 1910; Gosline, 1963; Rosen, is defined by: 1964; and the proposed scheme) to illustrate (14) The absence of a second infraorbital the relative number of inferred convergent bone. characters (homoplasies) in each (fig. 20). (15) The first epibranchial with an expand- Cladistic representations of relationships of ed base and no separate uncinate pro- "atherinoids," cyprinodontiforms and be- cess. loniforms are based on explicit statements of (16) The absence of a first pharyngobran- relationships in the various sources, or are chial. abstracted from a branching diagram provid- (17) The second and third epibranchials ed by an author. In each case, the perco- noticeably smaller than the first and morphs are included to represent both other fourth. ctenosquamates (for purposes of adding fin spines to the analysis) and other neoteleosts For reasons of parsimony, as explained (for purposes of adding the hyoid apparatus). above, we rate a number of shared features The character state tree proposed here based as convergent or reversed. Convergent char- on 17 characters, as just mentioned, incor- acters include: (1) absence of an interarcual porates only two homoplasies—the mini- cartilage in the atherinid Melanorhinus and mum number possible with these data. When in Phallostethus and present as a small basal these same 17 characters are placed on the cartilage in Ceratostethus (convergent with branches of the cladogram representing Bou- beloniforms); (2) a pelvic spine in male kil- lenger's scheme, 16 homoplasies are gener- lifishes of the Pantanodon and a dor- ated, character 1 being the only uncontra- sal spine in the killifish Jordanella (conver- dicted synapomorphy—and this is the same gent with spines in "atherinoids"); the pelvic as the maximum number of homoplasies that spines in some exocoetoids (Rosen, 1964, p. would be generated by a completely unre- 249) may be a retained primitive condition, solved polychotomy of the four taxa. Re- however; (3) the attachment of the pelvic gir- gan's scheme requires 11 homoplasies, and dle posterior to the fourth rib in some ath- Rosen's, six. Gosline's scheme is similar to erinids, some cyprinodontiforms and in be- Regan's but fails to resolve the relationships loniforms; (4) presence of long premaxillary of cyprinodontiforms and beloniforms in re- ascending processes in some "atherinoids" lation to "atherinoids" and percomorphs and in some cyprinodontiforms; (5) absence (i.e., the first two groups form a trichotomy of an ectopterygoid and metapterygoid in cy- with a branch that includes the last two prinodontoids and adrianichthyoids. Re- groups), and is contradicted by 15 of the 17 versed characters include: (1) more than six characters. These results are not affected by pelvic fin rays in many species of New World the fact that we recognize the "atherinoids" aplocheiloid and a few apomorph cyprino- as constituting six subgroups of unresolved dontoid killifishes (Parent!, 1981) and in interrelationships (bedotiids, melanota- 22 AMERICAN MUSEUM NOVITATES NO. 2719

A : "atherinoids" B : Beloniformes C i Cyprinodontiformes P : Percomorpha

Character distribution Character distribution BOULENGER 2 - loss in B REGAN 3 - loss in B 1904 3 - loss in B 1910 4-13 - losses in P 4-13 - losses in P 11 , 14-17 - gains in C or B 16 conflicts

Character distribution Character distribution GOSL1NE 3 - ambiguous ROSEN 2 - gain in A or P 1963 4-13 - ambiguous 1964 (1) 3 - loss in B 14-17 - ambiguous 14-17 - ambiguous 15 conflicts 6 conflicts

4-13 4-13 Character distribution Character distribution ROSEN 2 - gain in A or P PROPOSED 2 - ambiguous 1964 (2) 3 - gain in P 14-17 - losses in A 2 conflicts 6 conflicts

FIG. 20. Distribution of 17 apomorph characters (black dots) in six theories of relationship of four taxa. Numbers to left or right of dots in each diagram represent numbered characters in synapomorphy scheme in text. The character distributions show the most parsimonious interpretations of character conflict with cladistic structure. Thus, an inference of character loss or independent gain minimizes the number of character changes. Ambiguous characters are those involving two of the three branches in an unresolved trichotomy or those requiring the same number of assumptions about character loss or gain. The theories of Boulenger, Regan, and Gosline are implied by their classifications of these and 1981 ROSEN AND PARENTI: ORYZIAS 23 eniids, atherinids, telmatherinids, isonids, Superfamily Scomberesocoidea and phallostethids), that Rosen (1964) treat- Family Belonidae ed "atherinoids" as a definable taxon, or Family Scomberesocidae that other authors considered the "atheri- noids" to be part of a larger group containing ACKNOWLEDGMENTS also mugilids and sphyraenids. We thank Drs. Gareth Nelson for com- We conclude that since our cladogram of ments on the typescript and William Esch- relationships represents the most parsimo- meyer (California Academy of Sciences), nious arrangement of taxa based on the 17 William Fink (Museum of Comparative Zo- characters employed, and represents what ology), Norman Wilimovsky (University of we believe to be the present state of knowl- British Columbia), and Reeve Bailey (Uni- edge about atherinomorph interrelation- versity of Michigan, Museum of Zoology) for ships, that cladogram should be used as a lending specimens in their care. basis for a revised classification of ather- inomorph fishes. A classification derived from the proposed LITERATURE CITED scheme in figure 20, and following conven- Alexander, R. McN. tion with respect to exocoetoids, is: 1967. Mechanisms of the jaws in some ather- iniform fishes. Jour. Zool., London, Series Atherinomorpha vol. 151, pp. 233-255. Division I Allen, G. R. Family Atherinidae 1980. A generic classification of the rainbow- Family Bedotiidae fishes (family Melanotaeniidae). Rec. Family Isonidae Western Australian Mus., vol. 8, no. 3, Family Melanotaeniidae (includ- pp. 449-490. Allis, E. P., Jr. ing Pseudomugilidae) 1903. The skull, and the cranial and first spinal Family Phallostethidae (including muscles and nerves in Scomber scom- Neostethidae) ber. Jour. Morph., vol. 18, pp. 45-328. Family Telmatherinidae 1915. The homologies of the hyomandibula of Division II the gnathostome fishes. Ibid., vol. 26, Order Cyprinodontiformes (see Parent!, pp. 563-624. 1981) Berg, L. S. Order Beloniformes 1940. Classification of fishes both Recent and Suborder Adrianichthyoidei fossil. Reprint, English and Russian, Family Adrianichthyidae (includ- 1947. Ann Arbor, Michigan, J. W. Ed- wards, 517 pp. ing Horaichthyidae and Oryzi- Boulenger, G. A. idae) 1904. A synopsis of the suborders and families Suborder Exocoetoidei of teleostean fishes. Ann. Mag. Nat. Superfamily Exocoetoidea Hist., ser. 7, vol. 13, pp. 161-190. Family Hemiramphidae Breder, C. M., Jr. Family Exocoetidae 1932. On the habits and development of cer-

other taxa. Gosline's is abstracted from his published branching diagram of numerous taxa. The diagram labeled Rosen (1964) (1) is implied by his classification with the addition of the percomorphs as an outgroup; that labeled Rosen (1964) (2) is implied by statements in his text. Some taxonomic equivalents in the literature are: (1) Cyprinodontiformes (=Cyprinodontoidei, part; Cyprinodontes, part; Microcy- prini, part; Haplomi, part). (2) Beloniformes (=Phary ngognathi malacopterygii, part; Synentognathi, part). (3) "atherinoids" (=Atherinoidei; Mugiliformes, part; Percesoces, part). (4) Percomorpha (=Per- comorphi, part; Acanthopterygii, part). 24 AMERICAN MUSEUM NOVITATES NO. 2719

tain Atlantic Synentognathi. Papers Hubbs, C. L., and R. L. Wisner Tortugas Lab., vol. 28, publ. no. 1, pp. 1979. Revision of the sauries (Pisces, Scom- 1-35. beresocidae) with descriptions of two Breder, C. M., Jr., and D. E. Rosen new genera and one new species. Fish- 1966. Modes of reproduction in fishes. New ery Bull., vol. 77, no. 3, pp. 521-566. York, Nat. Hist. Press, 941 pp. Jordan, D. S. Burne, R. H. 1905. A guide to the study of fishes. New 1909. The anatomy of the olfactory organ of York, Henry Holt and Co., 2 vols., 624 teleostean fishes. Proc. Zool. Soc. Lon- and 599 pp. don, May-December, pp. 610-633. Jordan, D. S., and C. L. Hubbs Foster, N. R. 1919. A monographic review of the family of 1967. Trends in the evolution of reproductive Atherinidae or silversides. Studies Ich- behavior in killifishes. Studies Trop. thyol., Stanford Univ. Publ., Univ. Oceanogr., no. 5, pp. 549-566. Sen, 87 pp. 1968. Utility of egg and larval characters in Kulkarni, C. V. the classification of atheriniform fishes 1940. On the systematic position, structural (abstract). Amer. Zool., vol. 8, no. 4, modifications, bionomics and develop- abstr. 388. ment of a remarkable new family of cy- Gosline, W. A. prinodont fishes from the province of 1962. Systematic position and relationships of Bombay. Rec. Indian Mus., vol. 42, pt. the percesocine fishes. Pacific Sci., vol. 2, pp. 379-423. 16 April, pp. 207-217. McAllister, D. E. 1963. Considerations regarding the relation- 1968. The evolution of branchiostegals and ships of the percopsiform, cyprinodon- associated opercular, gular, and hyoid tiform, and gadiform fishes. Occas. Pa- bones and the classification of teleo- pers Mus. Zool., Univ. Michigan, no. stome fishes, living and fossil. Bull. 629, pp. 1-38. Natl. Mus. Canada, no. 221, 239 pp. Greenwood, P. H., D. E. Rosen, S. H. Weitzman, Melinkat, R., and E. Zeiske and G. S. Myers 1979. Functional morphology of ventilation of 1966. Phyletic studies of teleostean fishes, the olfactory organ in Bedotia geayi with a provisional classification of living Pellegrin 1909 (Teleostei, Atherinidae). forms. Bull. Amer. Mus. Nat. Hist., Zool. Anz., Jena, vol. 203, nos. 5/6, pp. vol. 131, art. 4, pp. 339-456. 354_468. Myers, G. S. Grier, H. J. 1928. The systematic position of the phallo- 1976. Sperm development in the teleost Ory- stethid fishes, with diagnosis of a new zias latipes. Cell. Tiss. Res., vol. 168, genus from Siam. Amer. Mus. Novi- pp. 419-431. tates, no. 295, 12 pp. In press. Cellular organization of the testis Nelson, G. J. and spermatogenesis in fishes. Amer. 1968. Gill-arch structure in Acanthodes. In Zool. 0rvig, T. (ed.), Current problems of Grier, H. J., J. R. Burns, and J. A. Flores lower vertebrate phylogeny. Proc. In press. Testis structure in three species of Fourth Nobel Symp. June 1967 Swedish teleosts with tubular gonopodia: Ana- Mus. Nat. Hist. Stockholm, Almquist bleps anableps, Anableps dowi, and and Wiksell, pp. 129-143. Jenynsia lineata. Copeia. 1969. Gill arches and the phylogeny of fishes, Grier, H. J., J. R. Linton, J. F. Leatherland, and with notes on the classification of ver- V. L. DeVlaming tebrates. Bull. Amer. Mus. Nat. Hist., 1980. Structural evidence for two different vol. 141, art. 4, pp. 475-552. testicular types in teleost fishes. Amer. Nelson, J. S. Jour. Anat., vol. 159, pp. 331-345. 1976. Fishes of the world. New York, John Hubbs, C. L. Wiley, 416 pp. 1919. A comparative study of the bones form- Nichols, J. T., and C. M. Breder, Jr. ing the opercular series of fishes. Jour. 1928. An annotated list of the Synentognathi. Morph., vol. 33, no. 1, pp. 61-71. Zoologica, vol. 8, no. 7, pp. 423-448. 1981 ROSEN AND PARENTI: ORYZIAS 25

Orton, G. L. 1973. Interrelationships of higher euteleos- 1955. Separation of eggs of synentognath and tean fishes. In Greenwood, P. H., R. S. allotriognath fishes in early embryonic Miles, and C. Patterson (eds.), Interre- stages. California Fish and Game, vol. lationships of fishes. London, Academic 41, no. l,pp. 103-105. Press, pp. 397-513. Parent!, L. R. Rosen, D. E., and P. H. Greenwood 1981. A phylogenetic and biogeographic anal- 1976. A fourth neotropical species of synbran- ysis of cyprinodontiform fishes (Teleos- chid eel and the phylogeny and system- tei: Atherinomorpha). Bull. Amer. Mus. atics of synbranchiform fishes. Bull. Nat. Hist., vol. 168, art. 4, pp. 335-557. Amer. Mus. Nat. Hist., vol. 157, art. 1, pp. 1-70. Regan, C. T. Rugh, R. 1910. Notes on the classification of the te- 1952. Experimental embryology. A manual of leostean fishes. Proc. Seventh Internatl. techniques and procedures. Revised Zool. Congr., Cambridge, Massachu- ed., second printing. Minneapolis, Bur- setts, pp. 1-16. gess Publ. Co., 480 pp. Rosen, D. E. Starks, E. C. 1964. The relationships and taxonomic posi- 1899. The osteological characters of the fishes tion of the halfbeaks, killifishes, silver- of the suborder Percesoces. Proc. U.S. sides, and their relatives. Bull. Amer. Natl. Mus., vol. 22, no. 1179, pp. 1-10. Mus. Nat. Hist., vol. 127, art. 5, pp. 217-268.